Release 0.7.0.

Direct3D 12 GPU Profiling

.appveyor.yml

@@ -2,18 +2,26 @@ version: '{build}'
 platform:
   - x64
 image:
-  - Visual Studio 2017
-  - Ubuntu
-before_build:
-  - cmd: cd profiler\build\win32
-  - cmd: nuget restore
-  - cmd: cd ..\..\..
+  - Visual Studio 2019
+  - Ubuntu1804
+install:
+  - cmd: cd c:\tools\vcpkg
+  - cmd: git pull
+  - cmd: bootstrap-vcpkg.bat
+  - cmd: vcpkg install freetype glfw3 capstone[core,arm,arm64,x86] --triplet x64-windows-static
+  - cmd: vcpkg integrate install
+  - cmd: cd %APPVEYOR_BUILD_FOLDER%
 build_script:
   - cmd: msbuild .\update\build\win32\update.vcxproj
   - cmd: msbuild .\profiler\build\win32\Tracy.vcxproj
   - cmd: msbuild .\capture\build\win32\capture.vcxproj
-  - sh: sudo apt-get update && sudo apt-get -y install libglfw3-dev libgtk2.0-dev
+  - cmd: msbuild .\library\win32\TracyProfiler.vcxproj /property:Configuration=Release
+  - sh: sudo apt-get update && sudo apt-get -y install libglfw3-dev libgtk2.0-dev libcapstone-dev
   - sh: make -C update/build/unix debug release
   - sh: make -C profiler/build/unix debug release
   - sh: make -C capture/build/unix debug release
+  - sh: make -C library/unix debug release
+  - sh: make -C test
+  - sh: make -C test clean
+  - sh: make -C test TRACYFLAGS=-DTRACY_ON_DEMAND
 test: off

.github/FUNDING.yml

@@ -0,0 +1 @@
+github: wolfpld

.github/workflows/gcc.yml

@@ -0,0 +1,30 @@
+name: gcc
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v2
+    - name: Install libraries
+      run: sudo apt-get update && sudo apt-get -y install libglfw3-dev libgtk2.0-dev libcapstone-dev
+    - name: Profiler GUI
+      run: make -C profiler/build/unix debug release
+    - name: Update utility
+      run: make -C update/build/unix debug release
+    - name: Capture utility
+      run: make -C capture/build/unix debug release
+    - name: Library
+      run: make -C library/unix debug release
+    - name: Test application
+      run: |
+        make -C test
+        make -C test clean
+        make -C test TRACYFLAGS=-DTRACY_ON_DEMAND

.github/workflows/latex.yml

@@ -0,0 +1,28 @@
+name: Manual
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v2
+    - name: Fix stupidity
+      run: |
+        cp AUTHORS AUTHORS.
+        cp LICENSE LICENSE.
+    - name: Compile LaTeX
+      uses: xu-cheng/latex-action@v2
+      with:
+        working_directory: manual
+        root_file: tracy.tex
+    - uses: actions/upload-artifact@v2
+      with:
+        name: manual
+        path: manual/tracy.pdf

.github/workflows/msvc.yml

@@ -0,0 +1,47 @@
+name: MSVC
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+
+    runs-on: windows-2019
+
+    steps:
+    - uses: actions/checkout@v2
+    - uses: microsoft/setup-msbuild@v1.0.0
+    - name: Integrate vcpkg
+      run: vcpkg integrate install
+    - name: Build vcpkg libraries
+      run: vcpkg install freetype glfw3 capstone[arm,arm64,x86] --triplet x64-windows-static
+    - name: Profiler GUI Debug
+      run: msbuild .\profiler\build\win32\Tracy.vcxproj /property:Configuration=Debug /property:Platform=x64
+    - name: Profiler GUI Release
+      run: msbuild .\profiler\build\win32\Tracy.vcxproj /property:Configuration=Release /property:Platform=x64
+    - name: Update utility Debug
+      run: msbuild .\update\build\win32\update.vcxproj /property:Configuration=Debug /property:Platform=x64
+    - name: Update utility Release
+      run: msbuild .\update\build\win32\update.vcxproj /property:Configuration=Release /property:Platform=x64
+    - name: Capture utility Debug
+      run: msbuild .\capture\build\win32\capture.vcxproj /property:Configuration=Debug /property:Platform=x64
+    - name: Capture utility Release
+      run: msbuild .\capture\build\win32\capture.vcxproj /property:Configuration=Release /property:Platform=x64
+    - name: Library
+      run: msbuild .\library\win32\TracyProfiler.vcxproj /property:Configuration=Release /property:Platform=x64
+    - name: Package binaries
+      run: |
+        mkdir bin
+        mkdir bin\dev
+        copy profiler\build\win32\x64\Release\Tracy.exe bin
+        copy update\build\win32\x64\Release\update.exe bin
+        copy capture\build\win32\x64\Release\capture.exe bin
+        copy library\win32\x64\Release\TracyProfiler.dll bin\dev
+        copy library\win32\x64\Release\TracyProfiler.lib bin\dev
+        7z a Tracy.7z bin
+    - uses: actions/upload-artifact@v2
+      with:
+        path: Tracy.7z

.gitignore

@@ -12,10 +12,18 @@ imgui.ini
 test/tracy_test
 test/tracy_test.exe
 */build/unix/*-*
-manual/tracy.aux
-manual/tracy.log
-manual/tracy.out
-manual/tracy.pdf
-manual/tracy.synctex.gz
-manual/tracy.toc
+manual/t*.aux
+manual/t*.log
+manual/t*.out
+manual/t*.pdf
+manual/t*.synctex.gz
+manual/t*.toc
+manual/t*.bbl
+manual/t*.blg
 profiler/build/win32/packages
+profiler/build/win32/Tracy.aps
+# include the vcpkg install script but not the files it produces
+vcpkg/*
+!vcpkg/install_vcpkg_dependencies.bat
+.deps/
+.dirstamp

AUTHORS

@@ -1,7 +1,11 @@
 Bartosz Taudul <wolf.pld@gmail.com>
-Kamil Klimek <kamil.klimek@sharkbits.com>
-Bartosz Szreder <zgredder@gmail.com>
-Arvid Gerstmann <dev@arvid-g.de>
-Rokas Kupstys <rokups@zoho.com>
-Till Rathmann <till.rathmann@gmx.de>
-Sherief Farouk <sherief.personal@gmail.com>
+Kamil Klimek <kamil.klimek@sharkbits.com>           (initial find zone implementation)
+Bartosz Szreder <zgredder@gmail.com>                (view/worker split)
+Arvid Gerstmann <dev@arvid-g.de>                    (compatibility fixes)
+Rokas Kupstys <rokups@zoho.com>                     (compatibility fixes, initial CI work, MingW support)
+Till Rathmann <till.rathmann@gmx.de>                (DLL support)
+Sherief Farouk <sherief.personal@gmail.com>         (compatibility fixes)
+Dedmen Miller <dedmen@dedmen.de>                    (find zone bug fixes, improvements)
+Michał Cichoń <michcic@gmail.com>                   (OSX call stack decoding backport)
+Thales Sabino <thales@codeplay.com>                 (OpenCL support)
+Andrew Depke <andrewdepke@gmail.com>                (Direct3D 12 support)

FAQ.md

@@ -1,77 +0,0 @@
-# A quick tracy FAQ
-
-### I already use VTune/perf/Very Sleepy/callgrind/MSVC profiler.
-
-These are statistical profilers, which can be used to find hot spots in the code. This is very useful, but it won't show you the underlying reason for semi-random frame stutter that may occur every couple of seconds.
-
-### You can use Telemetry for that.
-
-Telemetry license costs about 8000 $ per year. Tracy is open source software. Telemetry doesn't have Lua bindings.
-
-### You can use the free Brofiler. Crytek does use it, so it has to be good.
-
-After a cursory look at the Brofiler code I can tell that the timer resolution there is at 300 ns. Tracy can achieve 5 ns timer resolution. Brofiler event logging infrastructure seems to be over-engineered. Brofiler can't track lock contention, nor does it have Lua bindings.
-
-### So tracy is supposedly faster?
-
-My measurements show that logging a single zone with tracy takes only 15 ns. In theory, if the program was doing nothing else, tracy should be able to log 66 million zones per second.
-
-### Bullshit, RAD is advertising that they are able only to log about a million zones, over the network nevertheless: "Capture over a million timing zones per second in real-time!"
-
-Tracy can perform network transfer of 15 million zones per second. Should the client and server be on separate machines, this number will be even higher, but you will need more than a gigabit link to achieve the maximum throughput. [Click here for a video of a max-throughput capture.](https://www.youtube.com/watch?v=DSMIHShKGAc)
-
-### Can I connect to my application at any time and start profiling at this moment?
-
-By default no, all events are registered from the beginning of program execution and are waiting in a queue. There's a separate on-demand mode, enabled by using a `TRACY_ON_DEMAND` macro.
-
-### Am I seeing correctly that the profiler allocates one gigabyte of memory per second?
-
-Only in extreme cases. Normal usage has much lower memory pressure.
-
-### Why do you do magic with the static initialization order? Everyone says that's a bad practice.
-
-It allows tracking construction of static objects and memory allocations performed before main() is entered.
-
-### There's no support for consoles.
-
-Welp. But there's mobile support.
-
-### I do need console support.
-
-The code is open. Write your own, then send a patch.
-
-### I don't believe you can capture a zone in 15 ns. Show me the code!
-
-Following is the annotated assembly code (generated from C++ sources) that's responsible for logging start of the zone:
-
-```
-call        qword ptr [__imp_GetCurrentThreadId]
-mov         r14d,eax
-mov         qword ptr [rsp+0F0h],r14        // save thread id for later use
-mov         r12d,10h
-mov         rax,qword ptr gs:[58h]          // TLS
-mov         r15,qword ptr [rax]             // queue address
-mov         rdi,qword ptr [r12+r15]         // data address
-mov         rbp,qword ptr [rdi+20h]         // buffer counter
-mov         rbx,rbp
-and         ebx,7Fh                         // 128 item buffer
-jne         Application::InnerLoop+66h --+
-mov         rdx,rbp                      |
-mov         rcx,rdi                      |
-call        enqueue_begin_alloc          |  // reclaim/alloc next buffer
-shl         rbx,5  <---------------------+  // buffer items are 32 bytes
-add         rbx,qword ptr [rdi+40h]
-mov         byte ptr [rbx],4                // queue item type
-rdtscp
-mov         dword ptr [rbx+19h],ecx         // cpu id
-shl         rdx,20h
-or          rax,rdx                         // 64 bit timestamp
-mov         qword ptr [rbx+1],rax
-mov         qword ptr [rbx+9],r14           // thread id
-lea         rax,[__tracy_source_location]   // static struct address
-mov         qword ptr [rbx+11h],rax
-lea         rax,[rbp+1]                     // increment buffer counter
-mov         qword ptr [rdi+20h],rax
-```
-
-There's also a second code block, for the end of the zone. It's similar, but a bit smaller, as it can use some of the variables that were retrieved above.

LICENSE

@@ -1,7 +1,7 @@
-Tracy Profiler (https://bitbucket.org/wolfpld/tracy) is licensed under the
+Tracy Profiler (https://github.com/wolfpld/tracy) is licensed under the
 3-clause BSD license.
 
-Copyright (c) 2017-2018, Bartosz Taudul <wolf.pld@gmail.com>
+Copyright (c) 2017-2020, Bartosz Taudul <wolf.pld@gmail.com>
 All rights reserved.
 
 Redistribution and use in source and binary forms, with or without

NEWS

@@ -6,9 +6,261 @@ Note: Release numbers are nothing more than numbers. There are some
 "missing" versions due to trace file changes during development. This is not
 a mistake.
 
-v0.5 (xxxx-xx-xx)
+v0.7 (2020-06-11)
 -----------------
 
+This is the last release which will be able to load pre-v0.6 traces. Use the
+update utility to convert your old traces now!
+
+- chrome:tracing importer now imports zone metadata from "args" key.
+- Added display of statistical mode to find zone menu.
+- Automatic stack sampling is now available on windows.
+- Properly handle tracing on long-running systems.
+- Message list entries can now show associated frame image.
+- Call stack window will now display module names.
+- Symbol location in call stack window may now also display symbol address.
+- Statistics menu can now be used to display call stack sampling data or
+  list available symbols.
+- All call paths leading to the sampled instruction in a call stack can be
+  now displayed.
+- Frame image compression ratio (lossless in-memory compression, not taking
+  into account DXT compression) is displayed in playback window.
+- Allow reconnection straight from the discard data dialog.
+- Added ability to set custom names for locks.
+- Improved handling of network ports.
+- Added time percentage display to instrumentation statistics.
+- Display of ghost zones (generated from automated call stack sampling).
+- Notify when empty labels display is enabled.
+- Small fragments of executable code will be now sent from client to server.
+- Added notification about query backlog.
+- Fixed performance problem with query backlog.
+- Display number of in-flight queries, in addition to query backlog.
+- Improved failure reports.
+- The capture utility will connect to localhost by default.
+- Added optional support for QPC timer on windows.
+- Complete rewrite of source file viewer. It is now 100% reliable when going
+  to a source location.
+- Symbol source view was added.
+  - Extension of source file viewer.
+  - Can display source file, assembly view, or both at the same time.
+  - May include display of statistical profiling data.
+  - Ability to switch between source files which were used to build the
+    symbol.
+  - Ability to switch between inlined functions which are incorporated into
+    the symbol.
+  - Graphical representation of control flow in program.
+  - Display of micro-architectural data for each assembly instruction.
+  - Tracking register dependencies between assembly instructions.
+  - Disassembly may be saved to a file, in order to be processed by external
+    tools.
+- If the default listening port is occupied, profiler will now try listening
+  on other ports.
+- Added possibility to perform source file names substitution.
+- Profiler windows can be now docked.
+- CPU usage tooltip now displays a list of running threads.
+- Added possibility to filter discovered clients list.
+- Source files are now cached during capture.
+- Profiler will now display a popup when application crashes.
+- Added ability to send simple integral values as extra payload for zones.
+- Per-frame zone times on the frames plot can now display self time.
+- Ability to bind only on localhost interface.
+- OpenCL profiling.
+- Direct3D 12 profiling.
+
+v0.6.3 (2020-02-13)
+-------------------
+
+- Fixed performance issues with loading saved traces on Ryzen CPUs.
+- Profiler window contents are now properly updated during window resize.
+- Improved tid to pid mapping on windows.
+- Zero length and unfinished zones are no longer taken into account for
+  statistics.
+- Build files for shared library are now available (experimental).
+- GPU zones now also have "active" parameter.
+- Further reduction of memory usage and on-disk trace size.
+- Replaced ska::flat_hash_map with robin-hood-hashing.
+- Speed-up rendering of long lists of items.
+- Exact event time is displayed in some places in the UI.
+- Memory allocation lists can now be sorted.
+- Added display of trace file compression ratio.
+- Optional Zstd compression of trace files.
+- Frame images are now internally compressed using Zstd (instead of LZ4).
+- Fix display of continuous frame set tooltips.
+
+v0.6.2 (2019-12-30)
+-------------------
+
+- Improved call stack decoding on OSX.
+- Collection of CPU topology data.
+- C API now supports allocated source locations.
+- Added chrome:tracing importer.
+- Allow merging of ZoneText() strings.
+- Time distribution can now show both exclusive and inclusive times.
+- Display proper value of selection time in find zone menu.
+- Implemented limiting find zone search to a specified time range.
+- Highlight hovered zone from find zone menu zone list on the histogram.
+- Allow copying user data directory location to the clipboard.
+
+v0.6.1 (2019-11-28)
+-------------------
+
+- Dropped support for pre-v0.5 traces.
+- Improve BSD support.
+- GPU zone CPU thread highlight will now highlight whole thread, not only
+  the thread name.
+- Added CPU thread highlight for CPU data items.
+- Client parameters may be now set from the server.
+- Minor UI fixes.
+
+v0.6 (2019-11-17)
+-----------------
+
+This is the last release which will be able to load pre-v0.5 traces. Use the
+update utility to convert your old traces now!
+
+- Dropped support for pre-v0.4 traces.
+- Major memory usage decrease.
+- Significant network bandwidth decrease.
+- Implemented context switch capture on selected platforms.
+  - Zone timings in various UI places can now take into account only the
+    time when the thread was executing.
+  - Zone information window can now display regions in which thread was
+    suspended by the operating system.
+  - CPUs on which the zone was running are enumerated.
+  - Thread activity regions can be graphed on the timeline.
+- API breakage: SetThreadName() now only works on current thread.
+- Fixed thread name retrieval after thread is destroyed.
+- Added number of CPU cores to host info.
+- Limited number of possible source locations to 64K.
+- Limited supported capture length to 1.6 days.
+- CPU cores are now displayed on the timeline.
+  - Thread execution workload is displayed, including threads from external
+    programs.
+  - Thread migrations across CPU cores can be graphed.
+  - System-wide workload distribution is now plotted on the timeline.
+- Added "CPU data" window showing programs competing for CPU during the
+  capture.
+- Switched to using native thread identifiers (relatively small numbers), as
+  opposed to pthreads identifiers, which in reality were pointers.
+- Improved thread name discovery if context switch capture is enabled.
+- Per-trace state is now preserved between profiling sessions:
+  - Timeline view position.
+  - Item categories draw/hide settings.
+- Timeline zones will be highlighted using a different color, when a
+  matching time range is selected on histogram.
+- Per-frame zone times are now displayed on the frames plot when a zone is
+  selected in the find zone menu.
+- Zone color is now displayed in zone information window.
+- Zone colors can now be determined basing on depth and thread or source
+  location.
+- Thread colors are displayed across the profiler application.
+- Frame times can be now compared.
+- Expose more lock handling functionality.
+- Network port can be now specified by the user.
+- Proper handling of multithreaded Vulkan code.
+- Added extreme compression level in update utility.
+- Added time distribution data in the zone information window.
+- Trace file name is now displayed in trace information window.
+- Annotations can be now added to the timeline.
+- Server now performs network data retrieval and decompression on a dedicated
+  thread.
+- Added examples of Tracy integration.
+- Allow grouping of zones in the find zone menu by zone parent or with no
+  grouping.
+- Zone list in the statistics window can be now filtered.
+- Implemented configuration of plots.
+- Messages can now collect call stacks.
+
+v0.5 (2019-08-10)
+-----------------
+
+This is the last release which will be able to load pre-v0.4 traces. Use the
+update utility to convert your old traces now!
+
+- Major decrease of trace dump file size.
+- Major optimizations across the board.
+- Vcpkg is now used for library management on Windows.
+- Display dump file size change in the update utility.
+- Added notification area.
+  - Display trace loading time.
+  - Display background processing tasks after trace is loaded.
+  - Display trace save notification.
+  - Show crash icon, if there was a crash.
+- Added C API.
+- Profiling session may now gracefully terminate, due to incorrect
+  instrumentation. A popup with termination reason will be displayed.
+- Call stack improvements.
+  - Call stack frames now have a proper source file and file line
+    information on Linux.
+  - Single call stack frame may now have multiple entries, representing
+    inlined function calls.
+  - Call stack grouping in the find zone menu now has a special display
+    mode.
+  - Call stack memory allocations tree improvements:
+    - Add top-down variant to complement the previously available bottom-up
+      one.
+    - Add ability to group tree nodes by function name.
+    - Allow restricting tree to display only active allocations.
+  - Added support for Lua call stack capture.
+- Self time of zones may be now displayed in the find zone menu.
+- Added ability to disconnect from a client.
+- Find zone groups can now be sorted by mean time per call.
+- Zones displayed in the find zone menu can be now grouped by order of
+  appearance, execution time or name.
+- Time is now displayed without trailing fractional zeros (e.g. "2.5 ms"
+  instead of "2.50 ms").
+- Child zones displayed in zone info window can be now grouped by source
+  location.
+- Selected or hovered lock is now highlighted on the timeline.
+- Locks are now grouped into single and multithreaded (contended and
+  uncontended) in the options menu locks list.
+- On broken platforms the profiler can now be initialized as needed (and
+  possible), taking a performance and functionality hit.
+- User experience improvements in the graphical profiler.
+  - Thread position and height is now animated, to eliminate flickering that
+    was happening when depth of displayed zones was changing.
+  - Zooming in/out using the mouse wheel is now animated.
+  - Plot range adjustment is now animated.
+  - Various other UI improvements.
+- System CPU usage is now being monitored.
+- Threads that have nothing to display in the current view are now hidden by
+  default.
+- Dimmed-out the timeline outside the profiling area.
+- Source file view can now be opened also from statistics menu.
+- Display standard deviation in find zone and compare traces menus.
+- Display zone messages in zone information window.
+- Display order of threads can be changed in the options menu.
+- Prevent deadlocks by querying socket send buffer size.
+- Frame set statistics can be now limited to frames visible on the screen.
+- Messages can be now colored.
+- Zone selection in compare traces menu can be now linked to the other
+  trace.
+- Added support for frame image (screen shot) storage.
+- Implemented ability to cut off outliers on histograms.
+- Zone or frame that is currently hovered by the mouse cursor will be
+  highlighted on the histogram.
+- Server now displays available clients in the local network.
+- Source code whitespace visibility can now be enabled or disabled.
+- Profiler will now check if proper timer readings can be performed on
+  x86/x64.
+- Application can now log app-specific information, similarly to how the
+  host info reports system information.
+- Message list will automatically scroll down to the most recent message.
+  - Feature will disable when the list is scrolled by user.
+  - To re-enable, scroll to the bottom of the list.
+- Message list can be now filtered.
+- A notification popup will be displayed during trace cleanup.
+- Source file view won't be available if a source file is newer than the
+  capture.
+- Added ability to set custom trace descriptions.
+- Added frame time target lines.
+- FPS counts are now displayed next to frame times.
+- GPU drift value can be now automatically measured.
+- Connection window is now a popup hidden under a dedicated button.
+
+v0.4.1 (2018-12-30)
+-------------------
+
 - Active frame set can be now switched by clicking on a frame set on the
   timeline.
 - Add ability to go to a specified frame.

README.md

@@ -1,69 +1,22 @@
 # Tracy Profiler
 
-[![Build status](https://ci.appveyor.com/api/projects/status/968a88arq06gm3el/branch/master?svg=true)](https://ci.appveyor.com/project/wolfpld/tracy/branch/master)
+[![Sponsor](.github/sponsor.png)](https://github.com/sponsors/wolfpld/)
 
-Tracy is a real time, nanosecond resolution frame profiler that can be used for remote or embedded telemetry of your application. It can profile CPU (C++, Lua), GPU (OpenGL, Vulkan) and memory. It also can display locks held by threads and their interactions with each other.
+### A real time, nanosecond resolution, remote telemetry, hybrid frame and sampling profiler for games and other applications.
+
+Tracy supports profiling CPU (C, C++11, Lua), GPU (OpenGL, Vulkan, OpenCL, Direct3D 12), memory, locks, context switches, per-frame screenshots and more.
+
+For usage **and build process** instructions, consult the user manual [at the following address](https://github.com/wolfpld/tracy/releases).
 
 ![](doc/profiler.png)
 
-Tracy requires compiler support for C++11, Thread Local Storage and a way to workaround static initialization order fiasco. There are no other requirements. The following platforms are confirmed to be working (this is not a complete list):
+![](doc/profiler2.png)
 
-- Windows (x86, x64)
-- Linux (x86, x64, ARM, ARM64)
-- Android (ARM, x86)
-- FreeBSD (x64)
-- Cygwin (x64)
-- WSL (x64)
-- OSX (x64)
-
-The following compilers are supported:
-
-- MSVC
-- gcc
-- clang
+[Changelog](NEWS)
 
 [Introduction to Tracy Profiler v0.2](https://www.youtube.com/watch?v=fB5B46lbapc)  
 [New features in Tracy Profiler v0.3](https://www.youtube.com/watch?v=3SXpDpDh2Uo)  
-[New features in Tracy Profiler v0.4](https://www.youtube.com/watch?v=eAkgkaO8B9o)
-
-[A quick FAQ.](FAQ.md)  
-[List of changes.](NEWS)
-
-### High-level overview
-
-![](doc/design.svg)
-
-Tracy is split into client and server side. The client side collects events using a high-efficiency queue and awaits for an incoming connection. The server part connects to client and receives collected data from the client, which is then reconstructed into a viewable timeline. The transfer is performed using a TCP connection.
-
-### Performance impact
-
-To check how much slowdown is introduced by using Tracy, I have profiled [etcpak](https://bitbucket.org/wolfpld/etcpak), which is the fastest ETC texture compression utility there is. I used an 8192×8192 test image as input data and instrumented everything down to the 4×4 pixel block compression function (that's 4 million blocks to compress). It should be noted that Tracy needs to calibrate its internal timers at each run. This introduces a delay of 115 ms (on my machine), which is negligible when doing lengthy profiling runs, but it skews the results of etcpak timing. The following times have this delay subtracted, to give focus on zone collection impact, which is the thing that really matters here.
-
-|                        Scenario                       |  Zones  | Clean run | Profiling run | Difference |
-|-------------------------------------------------------|---------|-----------|---------------|------------|
-| Compression of an image to ETC1 format                | 4194568 |    0.94 s |       1.003 s |   +0.063 s |
-| Compression of an image to ETC2 format, with mip-maps | 5592822 |   1.034 s |       1.119 s |   +0.085 s |
-
-In both scenarios the per-zone time cost is at ~15 ns. This is in line with the measured 8 ns single event collection time (each zone has to report start and end event).
-
-## Usage instructions
-
-The user manual for Tracy is available [at the following address](https://bitbucket.org/wolfpld/tracy/downloads/tracy.pdf). It provides information about the integration process, required code markup and so on.
-
-## Features
-
-#### Marking locks
-
-![](doc/locks.png)
-
-#### Plotting data
-
-![](doc/plot.png)
-
-#### Message log
-
-![](doc/messages.png)
-
-#### Approximation of capture cost
-
-![](doc/cost.png)
+[New features in Tracy Profiler v0.4](https://www.youtube.com/watch?v=eAkgkaO8B9o)  
+[New features in Tracy Profiler v0.5](https://www.youtube.com/watch?v=P6E7qLMmzTQ)  
+[New features in Tracy Profiler v0.6](https://www.youtube.com/watch?v=uJkrFgriuOo)  
+[New features in Tracy Profiler v0.7](https://www.youtube.com/watch?v=_hU7vw00MZ4)

TODO

@@ -0,0 +1,7 @@
+"Would be nice to have" list for 1.0 release:
+=============================================
+
+* Pack queue items tightly in the queues.
+* Use level-of-detail system for plots.
+* Use per-thread lock data structures.
+* Use DTrace for BSD/OSX context switch capture.

Tracy.hpp

@@ -17,13 +17,19 @@
 #define ZoneScopedNC(x,y)
 
 #define ZoneText(x,y)
+#define ZoneTextV(x,y,z)
 #define ZoneName(x,y)
+#define ZoneNameV(x,y,z)
+#define ZoneValue(x)
+#define ZoneValueV(x,y)
 
 #define FrameMark
 #define FrameMarkNamed(x)
 #define FrameMarkStart(x)
 #define FrameMarkEnd(x)
 
+#define FrameImage(x,y,z,w,a)
+
 #define TracyLockable( type, varname ) type varname;
 #define TracyLockableN( type, varname, desc ) type varname;
 #define TracySharedLockable( type, varname ) type varname;
@@ -31,11 +37,16 @@
 #define LockableBase( type ) type
 #define SharedLockableBase( type ) type
 #define LockMark(x) (void)x;
+#define LockableName(x,y,z);
 
 #define TracyPlot(x,y)
+#define TracyPlotConfig(x,y)
 
 #define TracyMessage(x,y)
 #define TracyMessageL(x)
+#define TracyMessageC(x,y,z)
+#define TracyMessageLC(x,y)
+#define TracyAppInfo(x,y)
 
 #define TracyAlloc(x,y)
 #define TracyFree(x)
@@ -53,6 +64,14 @@
 #define TracyAllocS(x,y,z)
 #define TracyFreeS(x,y)
 
+#define TracyMessageS(x,y,z)
+#define TracyMessageLS(x,y)
+#define TracyMessageCS(x,y,z,w)
+#define TracyMessageLCS(x,y,z)
+
+#define TracyParameterRegister(x)
+#define TracyParameterSetup(x,y,z,w)
+
 #else
 
 #include "client/TracyLock.hpp"
@@ -77,13 +96,19 @@
 #define ZoneScopedNC( name, color ) ZoneNamedNC( ___tracy_scoped_zone, name, color, true )
 
 #define ZoneText( txt, size ) ___tracy_scoped_zone.Text( txt, size );
+#define ZoneTextV( varname, txt, size ) varname.Text( txt, size );
 #define ZoneName( txt, size ) ___tracy_scoped_zone.Name( txt, size );
+#define ZoneNameV( varname, txt, size ) varname.Name( txt, size );
+#define ZoneValue( value ) ___tracy_scoped_zone.Value( value );
+#define ZoneValueV( varname, value ) varname.Value( value );
 
-#define FrameMark tracy::Profiler::SendFrameMark();
-#define FrameMarkNamed( name ) tracy::Profiler::SendFrameMark( name, tracy::QueueType::FrameMarkMsg );
+#define FrameMark tracy::Profiler::SendFrameMark( nullptr );
+#define FrameMarkNamed( name ) tracy::Profiler::SendFrameMark( name );
 #define FrameMarkStart( name ) tracy::Profiler::SendFrameMark( name, tracy::QueueType::FrameMarkMsgStart );
 #define FrameMarkEnd( name ) tracy::Profiler::SendFrameMark( name, tracy::QueueType::FrameMarkMsgEnd );
 
+#define FrameImage( image, width, height, offset, flip ) tracy::Profiler::SendFrameImage( image, width, height, offset, flip );
+
 #define TracyLockable( type, varname ) tracy::Lockable<type> varname { [] () -> const tracy::SourceLocationData* { static const tracy::SourceLocationData srcloc { nullptr, #type " " #varname, __FILE__, __LINE__, 0 }; return &srcloc; }() };
 #define TracyLockableN( type, varname, desc ) tracy::Lockable<type> varname { [] () -> const tracy::SourceLocationData* { static const tracy::SourceLocationData srcloc { nullptr, desc, __FILE__, __LINE__, 0 }; return &srcloc; }() };
 #define TracySharedLockable( type, varname ) tracy::SharedLockable<type> varname { [] () -> const tracy::SourceLocationData* { static const tracy::SourceLocationData srcloc { nullptr, #type " " #varname, __FILE__, __LINE__, 0 }; return &srcloc; }() };
@@ -91,16 +116,27 @@
 #define LockableBase( type ) tracy::Lockable<type>
 #define SharedLockableBase( type ) tracy::SharedLockable<type>
 #define LockMark( varname ) static const tracy::SourceLocationData __tracy_lock_location_##varname { nullptr, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; varname.Mark( &__tracy_lock_location_##varname );
+#define LockableName( varname, txt, size ) varname.CustomName( txt, size );
 
 #define TracyPlot( name, val ) tracy::Profiler::PlotData( name, val );
+#define TracyPlotConfig( name, type ) tracy::Profiler::ConfigurePlot( name, type );
 
-#define TracyMessage( txt, size ) tracy::Profiler::Message( txt, size );
-#define TracyMessageL( txt ) tracy::Profiler::Message( txt );
+#define TracyAppInfo( txt, size ) tracy::Profiler::MessageAppInfo( txt, size );
 
 #if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+#  define TracyMessage( txt, size ) tracy::Profiler::Message( txt, size, TRACY_CALLSTACK );
+#  define TracyMessageL( txt ) tracy::Profiler::Message( txt, TRACY_CALLSTACK );
+#  define TracyMessageC( txt, size, color ) tracy::Profiler::MessageColor( txt, size, color, TRACY_CALLSTACK );
+#  define TracyMessageLC( txt, color ) tracy::Profiler::MessageColor( txt, color, TRACY_CALLSTACK );
+
 #  define TracyAlloc( ptr, size ) tracy::Profiler::MemAllocCallstack( ptr, size, TRACY_CALLSTACK );
 #  define TracyFree( ptr ) tracy::Profiler::MemFreeCallstack( ptr, TRACY_CALLSTACK );
 #else
+#  define TracyMessage( txt, size ) tracy::Profiler::Message( txt, size, 0 );
+#  define TracyMessageL( txt ) tracy::Profiler::Message( txt, 0 );
+#  define TracyMessageC( txt, size, color ) tracy::Profiler::MessageColor( txt, size, color, 0 );
+#  define TracyMessageLC( txt, color ) tracy::Profiler::MessageColor( txt, color, 0 );
+
 #  define TracyAlloc( ptr, size ) tracy::Profiler::MemAlloc( ptr, size );
 #  define TracyFree( ptr ) tracy::Profiler::MemFree( ptr );
 #endif
@@ -114,10 +150,15 @@
 #  define ZoneScopedS( depth ) ZoneNamedS( ___tracy_scoped_zone, depth, true )
 #  define ZoneScopedNS( name, depth ) ZoneNamedNS( ___tracy_scoped_zone, name, depth, true )
 #  define ZoneScopedCS( color, depth ) ZoneNamedCS( ___tracy_scoped_zone, color, depth, true )
-#  define ZoneScopedNCS( name, color, depth ) ZoneNamedNCS( ___tracy_scoped_zone, name, color depth, true )
+#  define ZoneScopedNCS( name, color, depth ) ZoneNamedNCS( ___tracy_scoped_zone, name, color, depth, true )
 
 #  define TracyAllocS( ptr, size, depth ) tracy::Profiler::MemAllocCallstack( ptr, size, depth );
 #  define TracyFreeS( ptr, depth ) tracy::Profiler::MemFreeCallstack( ptr, depth );
+
+#  define TracyMessageS( txt, size, depth ) tracy::Profiler::Message( txt, size, depth );
+#  define TracyMessageLS( txt, depth ) tracy::Profiler::Message( txt, depth );
+#  define TracyMessageCS( txt, size, color, depth ) tracy::Profiler::MessageColor( txt, size, color, depth );
+#  define TracyMessageLCS( txt, color, depth ) tracy::Profiler::MessageColor( txt, color, depth );
 #else
 #  define ZoneNamedS( varname, depth, active ) ZoneNamed( varname, active )
 #  define ZoneNamedNS( varname, name, depth, active ) ZoneNamedN( varname, name, active )
@@ -131,8 +172,16 @@
 
 #  define TracyAllocS( ptr, size, depth ) TracyAlloc( ptr, size )
 #  define TracyFreeS( ptr, depth ) TracyFree( ptr )
+
+#  define TracyMessageS( txt, size, depth ) TracyMessage( txt, size )
+#  define TracyMessageLS( txt, depth ) TracyMessageL( txt )
+#  define TracyMessageCS( txt, size, color, depth ) TracyMessageC( txt, size, color )
+#  define TracyMessageLCS( txt, color, depth ) TracyMessageLC( txt, color )
 #endif
 
+#define TracyParameterRegister( cb ) tracy::Profiler::ParameterRegister( cb );
+#define TracyParameterSetup( idx, name, isBool, val ) tracy::Profiler::ParameterSetup( idx, name, isBool, val );
+
 #endif
 
 #endif

TracyC.h

@@ -0,0 +1,197 @@
+#ifndef __TRACYC_HPP__
+#define __TRACYC_HPP__
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include "client/TracyCallstack.h"
+#include "common/TracyApi.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#ifndef TRACY_ENABLE
+
+typedef const void* TracyCZoneCtx;
+
+#define TracyCZone(c,x)
+#define TracyCZoneN(c,x,y)
+#define TracyCZoneC(c,x,y)
+#define TracyCZoneNC(c,x,y,z)
+#define TracyCZoneEnd(c)
+#define TracyCZoneText(c,x,y)
+#define TracyCZoneName(c,x,y)
+#define TracyCZoneValue(c,x)
+
+#define TracyCAlloc(x,y)
+#define TracyCFree(x)
+
+#define TracyCFrameMark
+#define TracyCFrameMarkNamed(x)
+#define TracyCFrameMarkStart(x)
+#define TracyCFrameMarkEnd(x)
+#define TracyCFrameImage(x,y,z,w,a)
+
+#define TracyCPlot(x,y)
+#define TracyCMessage(x,y)
+#define TracyCMessageL(x)
+#define TracyCMessageC(x,y,z)
+#define TracyCMessageLC(x,y)
+#define TracyCAppInfo(x,y)
+
+#define TracyCZoneS(x,y,z)
+#define TracyCZoneNS(x,y,z,w)
+#define TracyCZoneCS(x,y,z,w)
+#define TracyCZoneNCS(x,y,z,w,a)
+
+#define TracyCAllocS(x,y,z)
+#define TracyCFreeS(x,y)
+
+#define TracyCMessageS(x,y,z)
+#define TracyCMessageLS(x,y)
+#define TracyCMessageCS(x,y,z,w)
+#define TracyCMessageLCS(x,y,z)
+
+#else
+
+#ifndef TracyConcat
+#  define TracyConcat(x,y) TracyConcatIndirect(x,y)
+#endif
+#ifndef TracyConcatIndirect
+#  define TracyConcatIndirect(x,y) x##y
+#endif
+
+struct ___tracy_source_location_data
+{
+    const char* name;
+    const char* function;
+    const char* file;
+    uint32_t line;
+    uint32_t color;
+};
+
+struct ___tracy_c_zone_context
+{
+    uint32_t id;
+    int active;
+};
+
+// Some containers don't support storing const types.
+// This struct, as visible to user, is immutable, so treat it as if const was declared here.
+typedef /*const*/ struct ___tracy_c_zone_context TracyCZoneCtx;
+
+TRACY_API uint64_t ___tracy_alloc_srcloc( uint32_t line, const char* source, const char* function );
+TRACY_API uint64_t ___tracy_alloc_srcloc_name( uint32_t line, const char* source, const char* function, const char* name, size_t nameSz );
+
+TRACY_API TracyCZoneCtx ___tracy_emit_zone_begin( const struct ___tracy_source_location_data* srcloc, int active );
+TRACY_API TracyCZoneCtx ___tracy_emit_zone_begin_callstack( const struct ___tracy_source_location_data* srcloc, int depth, int active );
+TRACY_API TracyCZoneCtx ___tracy_emit_zone_begin_alloc( uint64_t srcloc, int active );
+TRACY_API TracyCZoneCtx ___tracy_emit_zone_begin_alloc_callstack( uint64_t srcloc, int depth, int active );
+TRACY_API void ___tracy_emit_zone_end( TracyCZoneCtx ctx );
+TRACY_API void ___tracy_emit_zone_text( TracyCZoneCtx ctx, const char* txt, size_t size );
+TRACY_API void ___tracy_emit_zone_name( TracyCZoneCtx ctx, const char* txt, size_t size );
+TRACY_API void ___tracy_emit_zone_value( TracyCZoneCtx ctx, uint64_t value );
+
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+#  define TracyCZone( ctx, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { NULL, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), TRACY_CALLSTACK, active );
+#  define TracyCZoneN( ctx, name, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), TRACY_CALLSTACK, active );
+#  define TracyCZoneC( ctx, color, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { NULL, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), TRACY_CALLSTACK, active );
+#  define TracyCZoneNC( ctx, name, color, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), TRACY_CALLSTACK, active );
+#else
+#  define TracyCZone( ctx, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { NULL, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin( &TracyConcat(__tracy_source_location,__LINE__), active );
+#  define TracyCZoneN( ctx, name, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin( &TracyConcat(__tracy_source_location,__LINE__), active );
+#  define TracyCZoneC( ctx, color, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { NULL, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin( &TracyConcat(__tracy_source_location,__LINE__), active );
+#  define TracyCZoneNC( ctx, name, color, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin( &TracyConcat(__tracy_source_location,__LINE__), active );
+#endif
+
+#define TracyCZoneEnd( ctx ) ___tracy_emit_zone_end( ctx );
+
+#define TracyCZoneText( ctx, txt, size ) ___tracy_emit_zone_text( ctx, txt, size );
+#define TracyCZoneName( ctx, txt, size ) ___tracy_emit_zone_name( ctx, txt, size );
+#define TracyCZoneValue( ctx, value ) ___tracy_emit_zone_value( ctx, value );
+
+
+TRACY_API void ___tracy_emit_memory_alloc( const void* ptr, size_t size );
+TRACY_API void ___tracy_emit_memory_alloc_callstack( const void* ptr, size_t size, int depth );
+TRACY_API void ___tracy_emit_memory_free( const void* ptr );
+TRACY_API void ___tracy_emit_memory_free_callstack( const void* ptr, int depth );
+
+TRACY_API void ___tracy_emit_message( const char* txt, size_t size, int callstack );
+TRACY_API void ___tracy_emit_messageL( const char* txt, int callstack );
+TRACY_API void ___tracy_emit_messageC( const char* txt, size_t size, uint32_t color, int callstack );
+TRACY_API void ___tracy_emit_messageLC( const char* txt, uint32_t color, int callstack );
+
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+#  define TracyCAlloc( ptr, size ) ___tracy_emit_memory_alloc_callstack( ptr, size, TRACY_CALLSTACK )
+#  define TracyCFree( ptr ) ___tracy_emit_memory_free_callstack( ptr, TRACY_CALLSTACK )
+
+#  define TracyCMessage( txt, size ) ___tracy_emit_message( txt, size, TRACY_CALLSTACK );
+#  define TracyCMessageL( txt ) ___tracy_emit_messageL( txt, TRACY_CALLSTACK );
+#  define TracyCMessageC( txt, size, color ) ___tracy_emit_messageC( txt, size, color, TRACY_CALLSTACK );
+#  define TracyCMessageLC( txt, color ) ___tracy_emit_messageLC( txt, color, TRACY_CALLSTACK );
+#else
+#  define TracyCAlloc( ptr, size ) ___tracy_emit_memory_alloc( ptr, size );
+#  define TracyCFree( ptr ) ___tracy_emit_memory_free( ptr );
+
+#  define TracyCMessage( txt, size ) ___tracy_emit_message( txt, size, 0 );
+#  define TracyCMessageL( txt ) ___tracy_emit_messageL( txt, 0 );
+#  define TracyCMessageC( txt, size, color ) ___tracy_emit_messageC( txt, size, color, 0 );
+#  define TracyCMessageLC( txt, color ) ___tracy_emit_messageLC( txt, color, 0 );
+#endif
+
+
+TRACY_API void ___tracy_emit_frame_mark( const char* name );
+TRACY_API void ___tracy_emit_frame_mark_start( const char* name );
+TRACY_API void ___tracy_emit_frame_mark_end( const char* name );
+TRACY_API void ___tracy_emit_frame_image( const void* image, uint16_t w, uint16_t h, uint8_t offset, int flip );
+
+#define TracyCFrameMark ___tracy_emit_frame_mark( 0 );
+#define TracyCFrameMarkNamed( name ) ___tracy_emit_frame_mark( name );
+#define TracyCFrameMarkStart( name ) ___tracy_emit_frame_mark_start( name );
+#define TracyCFrameMarkEnd( name ) ___tracy_emit_frame_mark_end( name );
+#define TracyCFrameImage( image, width, height, offset, flip ) ___tracy_emit_frame_image( image, width, height, offset, flip );
+
+
+TRACY_API void ___tracy_emit_plot( const char* name, double val );
+TRACY_API void ___tracy_emit_message_appinfo( const char* txt, size_t size );
+
+#define TracyCPlot( name, val ) ___tracy_emit_plot( name, val );
+#define TracyCAppInfo( txt, color ) ___tracy_emit_message_appinfo( txt, color );
+
+
+#ifdef TRACY_HAS_CALLSTACK
+#  define TracyCZoneS( ctx, depth, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { NULL, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), depth, active );
+#  define TracyCZoneNS( ctx, name, depth, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), depth, active );
+#  define TracyCZoneCS( ctx, color, depth, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { NULL, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), depth, active );
+#  define TracyCZoneNCS( ctx, name, color, depth, active ) static const struct ___tracy_source_location_data TracyConcat(__tracy_source_location,__LINE__) = { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; TracyCZoneCtx ctx = ___tracy_emit_zone_begin_callstack( &TracyConcat(__tracy_source_location,__LINE__), depth, active );
+
+#  define TracyCAllocS( ptr, size, depth ) ___tracy_emit_memory_alloc_callstack( ptr, size, depth )
+#  define TracyCFreeS( ptr, depth ) ___tracy_emit_memory_free_callstack( ptr, depth )
+
+#  define TracyCMessageS( txt, size, depth ) ___tracy_emit_message( txt, size, depth );
+#  define TracyCMessageLS( txt, depth ) ___tracy_emit_messageL( txt, depth );
+#  define TracyCMessageCS( txt, size, color, depth ) ___tracy_emit_messageC( txt, size, color, depth );
+#  define TracyCMessageLCS( txt, color, depth ) ___tracy_emit_messageLC( txt, color, depth );
+#else
+#  define TracyCZoneS( ctx, depth, active ) TracyCZone( ctx, active )
+#  define TracyCZoneNS( ctx, name, depth, active ) TracyCZoneN( ctx, name, active )
+#  define TracyCZoneCS( ctx, color, depth, active ) TracyCZoneC( ctx, color, active )
+#  define TracyCZoneNCS( ctx, name, color, depth, active ) TracyCZoneNC( ctx, name, color, active )
+
+#  define TracyCAllocS( ptr, size, depth ) TracyCAlloc( ptr, size )
+#  define TracyCFreeS( ptr, depth ) TracyCFree( ptr )
+
+#  define TracyCMessageS( txt, size, depth ) TracyCMessage( txt, size )
+#  define TracyCMessageLS( txt, depth ) TracyCMessageL( txt )
+#  define TracyCMessageCS( txt, size, color, depth ) TracyCMessageC( txt, size, color )
+#  define TracyCMessageLCS( txt, color, depth ) TracyCMessageLC( txt, color )
+#endif
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

TracyClient.cpp

@@ -18,8 +18,26 @@
 #include "common/tracy_lz4.cpp"
 #include "client/TracyProfiler.cpp"
 #include "client/TracyCallstack.cpp"
+#include "client/TracySysTime.cpp"
+#include "client/TracySysTrace.cpp"
 #include "common/TracySocket.cpp"
 #include "client/tracy_rpmalloc.cpp"
+#include "client/TracyDxt1.cpp"
+
+#if TRACY_HAS_CALLSTACK == 2 || TRACY_HAS_CALLSTACK == 3 || TRACY_HAS_CALLSTACK == 4 || TRACY_HAS_CALLSTACK == 6
+#  include "libbacktrace/alloc.cpp"
+#  include "libbacktrace/dwarf.cpp"
+#  include "libbacktrace/fileline.cpp"
+#  include "libbacktrace/mmapio.cpp"
+#  include "libbacktrace/posix.cpp"
+#  include "libbacktrace/sort.cpp"
+#  include "libbacktrace/state.cpp"
+#  if TRACY_HAS_CALLSTACK == 4
+#    include "libbacktrace/macho.cpp"
+#  else
+#    include "libbacktrace/elf.cpp"
+#  endif
+#endif
 
 #ifdef _MSC_VER
 #  pragma comment(lib, "ws2_32.lib")

TracyClientDLL.cpp

@@ -1,73 +0,0 @@
-//
-//          Tracy profiler
-//         ----------------
-//
-// On multi-DLL projects compile and
-// link with this source file (and none
-// other) in the executable and in
-// DLLs / shared objects that link to
-// the main DLL.
-//
-
-// Define TRACY_ENABLE to enable profiler.
-
-#include "common/TracySystem.cpp"
-
-#ifdef TRACY_ENABLE
-
-#include "client/TracyProfiler.hpp"
-#include "client/concurrentqueue.h"
-
-#include "common/TracyQueue.hpp"
-
-namespace tracy
-{
-#ifdef _MSC_VER
-#  define DLL_IMPORT __declspec(dllimport)
-#else
-#  define DLL_IMPORT
-#endif
-
-    DLL_IMPORT moodycamel::ConcurrentQueue<QueueItem>::ExplicitProducer* get_token();
-    DLL_IMPORT void*(*get_rpmalloc())(size_t size);
-    DLL_IMPORT void(*get_rpfree())(void* ptr);
-    DLL_IMPORT Profiler& get_profiler();
-
-#if defined TRACY_HW_TIMER && __ARM_ARCH >= 6
-    DLL_IMPORT int64_t(*get_GetTimeImpl())();
-
-    int64_t(*GetTimeImpl)() = get_GetTimeImpl();
-#endif
-
-#ifdef TRACY_COLLECT_THREAD_NAMES
-    DLL_IMPORT std::atomic<ThreadNameData*>& get_threadNameData();
-    DLL_IMPORT void(*get_rpmalloc_thread_initialize())();
-
-    std::atomic<ThreadNameData*>& s_threadNameData = get_threadNameData();
-    void(*rpmalloc_thread_initialize_fpt)() = get_rpmalloc_thread_initialize();
-
-    void rpmalloc_thread_initialize(void)
-    {
-        rpmalloc_thread_initialize_fpt();
-    }
-#endif
-
-    static void*(*rpmalloc_fpt)(size_t size) = get_rpmalloc();
-    static void(*rpfree_fpt)(void* ptr) = get_rpfree();
-
-    RPMALLOC_RESTRICT void* rpmalloc(size_t size)
-    {
-        return rpmalloc_fpt(size);
-    }
-
-    void rpfree(void* ptr)
-    {
-        rpfree_fpt(ptr);
-    }
-
-    Profiler& s_profiler = get_profiler();
-
-    thread_local ProducerWrapper s_token { get_token() };
-}
-
-#endif

TracyD3D12.hpp

@@ -0,0 +1,351 @@
+#ifndef __TRACYD3D12_HPP__
+#define __TRACYD3D12_HPP__
+
+#ifndef TRACY_ENABLE
+
+#define TracyD3D12Context(device, queue) nullptr
+#define TracyD3D12Destroy(ctx)
+
+#define TracyD3D12NamedZone(ctx, varname, cmdList, name, active)
+#define TracyD3D12NamedZoneC(ctx, varname, cmdList, name, color, active)
+#define TracyD3D12Zone(ctx, cmdList, name)
+#define TracyD3D12ZoneC(ctx, cmdList, name, color)
+
+#define TracyD3D12Collect(ctx)
+
+namespace tracy
+{
+	class D3D12ZoneScope {};
+}
+
+using TracyD3D12Ctx = void*;
+
+#else
+
+#include "Tracy.hpp"
+#include "client/TracyProfiler.hpp"
+
+#include <cstdlib>
+#include <cassert>
+#include <d3d12.h>
+#include <dxgi.h>
+#include <wrl/client.h>
+#include <queue>
+
+namespace tracy
+{
+
+	struct D3D12QueryPayload
+	{
+		uint32_t m_queryIdStart = 0;
+		uint32_t m_queryCount = 0;
+	};
+
+	// Command queue context.
+	class D3D12QueueCtx
+	{
+		friend class D3D12ZoneScope;
+
+		static constexpr uint32_t MaxQueries = 64 * 1024;  // Queries are begin and end markers, so we can store half as many total time durations. Must be even!
+
+		bool m_initialized = false;
+
+		ID3D12Device* m_device;
+		ID3D12CommandQueue* m_queue;
+		uint8_t m_context;
+		Microsoft::WRL::ComPtr<ID3D12QueryHeap> m_queryHeap;
+		Microsoft::WRL::ComPtr<ID3D12Resource> m_readbackBuffer;
+		
+		// In-progress payload.
+		uint32_t m_queryLimit = MaxQueries;
+		uint32_t m_queryCounter = 0;
+		uint32_t m_previousQueryCounter = 0;
+
+		uint32_t m_activePayload = 0;
+		Microsoft::WRL::ComPtr<ID3D12Fence> m_payloadFence;
+		std::queue<D3D12QueryPayload> m_payloadQueue;
+
+	public:
+		D3D12QueueCtx(ID3D12Device* device, ID3D12CommandQueue* queue)
+			: m_device(device)
+			, m_queue(queue)
+			, m_context(GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed))
+		{
+			// Verify we support timestamp queries on this queue.
+
+			if (queue->GetDesc().Type == D3D12_COMMAND_LIST_TYPE_COPY)
+			{
+				D3D12_FEATURE_DATA_D3D12_OPTIONS3 featureData{};
+
+				if (FAILED(device->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS3, &featureData, sizeof(featureData))))
+				{
+					assert(false && "Platform does not support profiling of copy queues.");
+				}
+			}
+
+			uint64_t timestampFrequency;
+
+			if (FAILED(queue->GetTimestampFrequency(&timestampFrequency)))
+			{
+				assert(false && "Failed to get timestamp frequency.");
+			}
+
+			uint64_t cpuTimestamp;
+			uint64_t gpuTimestamp;
+
+			if (FAILED(queue->GetClockCalibration(&gpuTimestamp, &cpuTimestamp)))
+			{
+				assert(false && "Failed to get queue clock calibration.");
+			}
+
+			cpuTimestamp = Profiler::GetTime();
+
+			D3D12_QUERY_HEAP_DESC heapDesc{};
+			heapDesc.Type = queue->GetDesc().Type == D3D12_COMMAND_LIST_TYPE_COPY ? D3D12_QUERY_HEAP_TYPE_COPY_QUEUE_TIMESTAMP : D3D12_QUERY_HEAP_TYPE_TIMESTAMP;
+			heapDesc.Count = m_queryLimit;
+			heapDesc.NodeMask = 0;  // #TODO: Support multiple adapters.
+
+			while (FAILED(device->CreateQueryHeap(&heapDesc, IID_PPV_ARGS(&m_queryHeap))))
+			{
+				m_queryLimit /= 2;
+				heapDesc.Count = m_queryLimit;
+			}
+
+			// Create a readback buffer, which will be used as a destination for the query data.
+
+			D3D12_RESOURCE_DESC readbackBufferDesc{};
+			readbackBufferDesc.Alignment = 0;
+			readbackBufferDesc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
+			readbackBufferDesc.Width = m_queryLimit * sizeof(uint64_t);
+			readbackBufferDesc.Height = 1;
+			readbackBufferDesc.DepthOrArraySize = 1;
+			readbackBufferDesc.Format = DXGI_FORMAT_UNKNOWN;
+			readbackBufferDesc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;  // Buffers are always row major.
+			readbackBufferDesc.MipLevels = 1;
+			readbackBufferDesc.SampleDesc.Count = 1;
+			readbackBufferDesc.SampleDesc.Quality = 0;
+			readbackBufferDesc.Flags = D3D12_RESOURCE_FLAG_NONE;
+
+			D3D12_HEAP_PROPERTIES readbackHeapProps{};
+			readbackHeapProps.Type = D3D12_HEAP_TYPE_READBACK;
+			readbackHeapProps.CPUPageProperty = D3D12_CPU_PAGE_PROPERTY_UNKNOWN;
+			readbackHeapProps.MemoryPoolPreference = D3D12_MEMORY_POOL_UNKNOWN;
+			readbackHeapProps.CreationNodeMask = 0;
+			readbackHeapProps.VisibleNodeMask = 0;  // #TODO: Support multiple adapters.
+
+			if (FAILED(device->CreateCommittedResource(&readbackHeapProps, D3D12_HEAP_FLAG_NONE, &readbackBufferDesc, D3D12_RESOURCE_STATE_COPY_DEST, nullptr, IID_PPV_ARGS(&m_readbackBuffer))))
+			{
+				assert(false && "Failed to create query readback buffer.");
+			}
+
+			if (FAILED(device->CreateFence(0, D3D12_FENCE_FLAG_NONE, IID_PPV_ARGS(&m_payloadFence))))
+			{
+				assert(false && "Failed to create payload fence.");
+			}
+
+			auto* item = Profiler::QueueSerial();
+			MemWrite(&item->hdr.type, QueueType::GpuNewContext);
+			MemWrite(&item->gpuNewContext.cpuTime, cpuTimestamp);
+			MemWrite(&item->gpuNewContext.gpuTime, gpuTimestamp);
+			memset(&item->gpuNewContext.thread, 0, sizeof(item->gpuNewContext.thread));
+			MemWrite(&item->gpuNewContext.period, 1E+09f / static_cast<float>(timestampFrequency));
+			MemWrite(&item->gpuNewContext.context, m_context);
+			MemWrite(&item->gpuNewContext.accuracyBits, uint8_t{ 0 });
+			MemWrite(&item->gpuNewContext.type, GpuContextType::Direct3D12);
+
+#ifdef TRACY_ON_DEMAND
+			GetProfiler().DeferItem(*item);
+#endif
+
+			Profiler::QueueSerialFinish();
+
+			m_initialized = true;
+		}
+
+		void NewFrame()
+		{
+			m_payloadQueue.emplace(D3D12QueryPayload{ m_previousQueryCounter, m_queryCounter });
+			m_previousQueryCounter += m_queryCounter;
+			m_queryCounter = 0;
+
+			if (m_previousQueryCounter >= m_queryLimit)
+			{
+				m_previousQueryCounter -= m_queryLimit;
+			}
+
+			m_queue->Signal(m_payloadFence.Get(), ++m_activePayload);
+		}
+
+		void Collect()
+		{
+			ZoneScopedC(Color::Red4);
+
+#ifdef TRACY_ON_DEMAND
+			if (!GetProfiler().IsConnected())
+			{
+				m_queryCounter = 0;
+
+				return;
+			}
+#endif
+
+			// Find out what payloads are available.
+			const auto newestReadyPayload = m_payloadFence->GetCompletedValue();
+			const auto payloadCount = m_payloadQueue.size() - (m_activePayload - newestReadyPayload);
+
+			if (!payloadCount)
+			{
+				return;  // No payloads are available yet, exit out.
+			}
+
+			D3D12_RANGE mapRange{ 0, m_queryLimit * sizeof(uint64_t) };
+
+			// Map the readback buffer so we can fetch the query data from the GPU.
+			void* readbackBufferMapping = nullptr;
+
+			if (FAILED(m_readbackBuffer->Map(0, &mapRange, &readbackBufferMapping)))
+			{
+				assert(false && "Failed to map readback buffer.");
+			}
+
+			auto* timestampData = static_cast<uint64_t*>(readbackBufferMapping);
+
+			for (uint32_t i = 0; i < payloadCount; ++i)
+			{
+				const auto& payload = m_payloadQueue.front();
+
+				for (uint32_t j = 0; j < payload.m_queryCount; ++j)
+				{
+					const auto counter = (payload.m_queryIdStart + j) % m_queryLimit;
+					const auto timestamp = timestampData[counter];
+					const auto queryId = counter;
+
+					auto* item = Profiler::QueueSerial();
+					MemWrite(&item->hdr.type, QueueType::GpuTime);
+					MemWrite(&item->gpuTime.gpuTime, timestamp);
+					MemWrite(&item->gpuTime.queryId, static_cast<uint16_t>(queryId));
+					MemWrite(&item->gpuTime.context, m_context);
+
+					Profiler::QueueSerialFinish();
+				}
+
+				m_payloadQueue.pop();
+			}
+
+			m_readbackBuffer->Unmap(0, nullptr);
+		}
+
+	private:
+		tracy_force_inline uint32_t NextQueryId()
+		{
+			assert(m_queryCounter < m_queryLimit && "Submitted too many GPU queries! Consider increasing MaxQueries.");
+
+			const uint32_t id = (m_previousQueryCounter + m_queryCounter) % m_queryLimit;
+			m_queryCounter += 2;  // Allocate space for a begin and end query.
+
+			return id;
+		}
+
+		tracy_force_inline uint8_t GetId() const
+		{
+			return m_context;
+		}
+	};
+
+	class D3D12ZoneScope
+	{
+		const bool m_active;
+		D3D12QueueCtx* m_ctx = nullptr;
+		ID3D12GraphicsCommandList* m_cmdList = nullptr;
+		uint32_t m_queryId = 0;  // Used for tracking in nested zones.
+
+	public:
+		tracy_force_inline D3D12ZoneScope(D3D12QueueCtx* ctx, ID3D12GraphicsCommandList* cmdList, const SourceLocationData* srcLocation, bool active)
+#ifdef TRACY_ON_DEMAND
+			: m_active(active && GetProfiler().IsConnected())
+#else
+			: m_active(active)
+#endif
+		{
+			if (!m_active) return;
+
+			m_ctx = ctx;
+			m_cmdList = cmdList;
+
+			m_queryId = ctx->NextQueryId();
+			cmdList->EndQuery(ctx->m_queryHeap.Get(), D3D12_QUERY_TYPE_TIMESTAMP, m_queryId);
+
+			auto* item = Profiler::QueueSerial();
+#if defined(TRACY_HAS_CALLSTACK) && defined(TRACY_CALLSTACK)
+			MemWrite(&item->hdr.type, QueueType::GpuZoneBeginCallstackSerial);
+#else
+			MemWrite(&item->hdr.type, QueueType::GpuZoneBeginSerial);
+#endif
+			MemWrite(&item->gpuZoneBegin.cpuTime, Profiler::GetTime());
+			MemWrite(&item->gpuZoneBegin.srcloc, reinterpret_cast<uint64_t>(srcLocation));
+			MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
+			MemWrite(&item->gpuZoneBegin.queryId, static_cast<uint16_t>(m_queryId));
+			MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+
+			Profiler::QueueSerialFinish();
+
+#if defined(TRACY_HAS_CALLSTACK) && defined(TRACY_CALLSTACK)
+			GetProfiler().SendCallstack(TRACY_CALLSTACK);
+#endif
+		}
+
+		tracy_force_inline ~D3D12ZoneScope()
+		{
+			if (!m_active) return;
+
+			const auto queryId = m_queryId + 1;  // Our end query slot is immediately after the begin slot.
+			m_cmdList->EndQuery(m_ctx->m_queryHeap.Get(), D3D12_QUERY_TYPE_TIMESTAMP, queryId);
+
+			auto* item = Profiler::QueueSerial();
+			MemWrite(&item->hdr.type, QueueType::GpuZoneEndSerial);
+			MemWrite(&item->gpuZoneEnd.cpuTime, Profiler::GetTime());
+			MemWrite(&item->gpuZoneEnd.thread, GetThreadHandle());
+			MemWrite(&item->gpuZoneEnd.queryId, static_cast<uint16_t>(queryId));
+			MemWrite(&item->gpuZoneEnd.context, m_ctx->GetId());
+
+			Profiler::QueueSerialFinish();
+
+			m_cmdList->ResolveQueryData(m_ctx->m_queryHeap.Get(), D3D12_QUERY_TYPE_TIMESTAMP, m_queryId, 2, m_ctx->m_readbackBuffer.Get(), m_queryId * sizeof(uint64_t));
+		}
+	};
+
+	static inline D3D12QueueCtx* CreateD3D12Context(ID3D12Device* device, ID3D12CommandQueue* queue)
+	{
+		InitRPMallocThread();
+
+		auto* ctx = static_cast<D3D12QueueCtx*>(tracy_malloc(sizeof(D3D12QueueCtx)));
+		new (ctx) D3D12QueueCtx{ device, queue };
+
+		return ctx;
+	}
+
+	static inline void DestroyD3D12Context(D3D12QueueCtx* ctx)
+	{
+		ctx->~D3D12QueueCtx();
+		tracy_free(ctx);
+	}
+
+}
+
+using TracyD3D12Ctx = tracy::D3D12QueueCtx*;
+
+#define TracyD3D12Context(device, queue) tracy::CreateD3D12Context(device, queue);
+#define TracyD3D12Destroy(ctx) tracy::DestroyD3D12Context(ctx);
+
+#define TracyD3D12NewFrame(ctx) ctx->NewFrame();
+
+#define TracyD3D12NamedZone(ctx, varname, cmdList, name, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location, __LINE__) { name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, 0 }; tracy::D3D12ZoneScope varname{ ctx, cmdList, &TracyConcat(__tracy_gpu_source_location, __LINE__), active };
+#define TracyD3D12NamedZoneC(ctx, varname, cmdList, name, color, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location, __LINE__) { name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, color }; tracy::D3D12ZoneScope varname{ ctx, cmdList, &TracyConcat(__tracy_gpu_source_location, __LINE__), active };
+#define TracyD3D12Zone(ctx, cmdList, name) TracyD3D12NamedZone(ctx, ___tracy_gpu_zone, cmdList, name, true)
+#define TracyD3D12ZoneC(ctx, cmdList, name, color) TracyD3D12NamedZoneC(ctx, ___tracy_gpu_zone, cmdList, name, color, true)
+
+#define TracyD3D12Collect(ctx) ctx->Collect();
+
+#endif
+
+#endif

TracyLua.hpp

@@ -23,6 +23,10 @@ static inline void LuaRegister( lua_State* L )
     lua_pushcfunction( L, detail::noop );
     lua_setfield( L, -2, "ZoneBeginN" );
     lua_pushcfunction( L, detail::noop );
+    lua_setfield( L, -2, "ZoneBeginS" );
+    lua_pushcfunction( L, detail::noop );
+    lua_setfield( L, -2, "ZoneBeginNS" );
+    lua_pushcfunction( L, detail::noop );
     lua_setfield( L, -2, "ZoneEnd" );
     lua_pushcfunction( L, detail::noop );
     lua_setfield( L, -2, "ZoneText" );
@@ -81,6 +85,18 @@ static inline void LuaRemove( char* script )
                     memset( script, ' ', end - script );
                     script = end;
                 }
+                else if( strncmp( script + 10, "BeginS(", 7 ) == 0 )
+                {
+                    auto end = FindEnd( script + 17 );
+                    memset( script, ' ', end - script );
+                    script = end;
+                }
+                else if( strncmp( script + 10, "BeginNS(", 8 ) == 0 )
+                {
+                    auto end = FindEnd( script + 18 );
+                    memset( script, ' ', end - script );
+                    script = end;
+                }
                 else
                 {
                     script += 10;
@@ -112,6 +128,7 @@ static inline void LuaRemove( char* script )
 
 #include "common/TracyColor.hpp"
 #include "common/TracyAlign.hpp"
+#include "common/TracyForceInline.hpp"
 #include "common/TracySystem.hpp"
 #include "client/TracyProfiler.hpp"
 
@@ -119,161 +136,190 @@ namespace tracy
 {
 
 #ifdef TRACY_ON_DEMAND
-extern thread_local LuaZoneState s_luaZoneState;
+TRACY_API LuaZoneState& GetLuaZoneState();
 #endif
 
 namespace detail
 {
 
-static inline int LuaZoneBegin( lua_State* L )
+#ifdef TRACY_HAS_CALLSTACK
+static tracy_force_inline void SendLuaCallstack( lua_State* L, uint32_t depth )
+{
+    assert( depth <= 64 );
+    lua_Debug dbg[64];
+    const char* func[64];
+    uint32_t fsz[64];
+    uint32_t ssz[64];
+    uint32_t spaceNeeded = 4;     // cnt
+
+    uint32_t cnt;
+    for( cnt=0; cnt<depth; cnt++ )
+    {
+        if( lua_getstack( L, cnt+1, dbg+cnt ) == 0 ) break;
+        lua_getinfo( L, "Snl", dbg+cnt );
+        func[cnt] = dbg[cnt].name ? dbg[cnt].name : dbg[cnt].short_src;
+        fsz[cnt] = uint32_t( strlen( func[cnt] ) );
+        ssz[cnt] = uint32_t( strlen( dbg[cnt].source ) );
+        spaceNeeded += fsz[cnt] + ssz[cnt];
+    }
+    spaceNeeded += cnt * ( 4 + 4 + 4 );     // source line, function string length, source string length
+
+    auto ptr = (char*)tracy_malloc( spaceNeeded + 4 );
+    auto dst = ptr;
+    memcpy( dst, &spaceNeeded, 4 ); dst += 4;
+    memcpy( dst, &cnt, 4 ); dst += 4;
+    for( uint32_t i=0; i<cnt; i++ )
+    {
+        const uint32_t line = dbg[i].currentline;
+        memcpy( dst, &line, 4 ); dst += 4;
+        memcpy( dst, fsz+i, 4 ); dst += 4;
+        memcpy( dst, func[i], fsz[i] ); dst += fsz[i];
+        memcpy( dst, ssz+i, 4 ); dst += 4;
+        memcpy( dst, dbg[i].source, ssz[i] ), dst += ssz[i];
+    }
+    assert( dst - ptr == spaceNeeded + 4 );
+
+    TracyLfqPrepare( QueueType::CallstackAlloc );
+    MemWrite( &item->callstackAlloc.ptr, (uint64_t)ptr );
+    MemWrite( &item->callstackAlloc.nativePtr, (uint64_t)Callstack( depth ) );
+    TracyLfqCommit;
+}
+
+static inline int LuaZoneBeginS( lua_State* L )
 {
 #ifdef TRACY_ON_DEMAND
-    const auto zoneCnt = s_luaZoneState.counter++;
-    if( zoneCnt != 0 && !s_luaZoneState.active ) return 0;
-    s_luaZoneState.active = s_profiler.IsConnected();
-    if( !s_luaZoneState.active ) return 0;
+    const auto zoneCnt = GetLuaZoneState().counter++;
+    if( zoneCnt != 0 && !GetLuaZoneState().active ) return 0;
+    GetLuaZoneState().active = GetProfiler().IsConnected();
+    if( !GetLuaZoneState().active ) return 0;
 #endif
 
-    const uint32_t color = Color::DeepSkyBlue3;
-
+    TracyLfqPrepare( QueueType::ZoneBeginAllocSrcLocCallstack );
     lua_Debug dbg;
     lua_getstack( L, 1, &dbg );
     lua_getinfo( L, "Snl", &dbg );
+    const auto srcloc = Profiler::AllocSourceLocation( dbg.currentline, dbg.source, dbg.name ? dbg.name : dbg.short_src );
+    MemWrite( &item->zoneBegin.time, Profiler::GetTime() );
+    MemWrite( &item->zoneBegin.srcloc, srcloc );
+    TracyLfqCommit;
 
-    const uint32_t line = dbg.currentline;
-    const auto func = dbg.name ? dbg.name : dbg.short_src;
-    const auto fsz = strlen( func );
-    const auto ssz = strlen( dbg.source );
-
-    // Data layout:
-    //  4b  payload size
-    //  4b  color
-    //  4b  source line
-    //  fsz function name
-    //  1b  null terminator
-    //  ssz source file name
-    //  1b  null terminator
-    const uint32_t sz = uint32_t( 4 + 4 + 4 + fsz + 1 + ssz + 1 );
-    auto ptr = (char*)tracy_malloc( sz );
-    memcpy( ptr, &sz, 4 );
-    memcpy( ptr + 4, &color, 4 );
-    memcpy( ptr + 8, &line, 4 );
-    memcpy( ptr + 12, func, fsz+1 );
-    memcpy( ptr + 12 + fsz + 1, dbg.source, ssz + 1 );
-
-    Magic magic;
-    auto& token = s_token.ptr;
-    auto& tail = token->get_tail_index();
-    auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneBeginAllocSrcLoc );
-#ifdef TRACY_RDTSCP_OPT
-    MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
+#ifdef TRACY_CALLSTACK
+    const uint32_t depth = TRACY_CALLSTACK;
 #else
-    uint32_t cpu;
-    MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
-    MemWrite( &item->zoneBegin.cpu, cpu );
+    const auto depth = uint32_t( lua_tointeger( L, 1 ) );
 #endif
-    MemWrite( &item->zoneBegin.thread, GetThreadHandle() );
-    MemWrite( &item->zoneBegin.srcloc, (uint64_t)ptr );
-    tail.store( magic + 1, std::memory_order_release );
+    SendLuaCallstack( L, depth );
+
     return 0;
 }
 
+static inline int LuaZoneBeginNS( lua_State* L )
+{
+#ifdef TRACY_ON_DEMAND
+    const auto zoneCnt = GetLuaZoneState().counter++;
+    if( zoneCnt != 0 && !GetLuaZoneState().active ) return 0;
+    GetLuaZoneState().active = GetProfiler().IsConnected();
+    if( !GetLuaZoneState().active ) return 0;
+#endif
+
+    TracyLfqPrepare( QueueType::ZoneBeginAllocSrcLocCallstack );
+    lua_Debug dbg;
+    lua_getstack( L, 1, &dbg );
+    lua_getinfo( L, "Snl", &dbg );
+    size_t nsz;
+    const auto name = lua_tolstring( L, 1, &nsz );
+    const auto srcloc = Profiler::AllocSourceLocation( dbg.currentline, dbg.source, dbg.name ? dbg.name : dbg.short_src, name, nsz );
+    MemWrite( &item->zoneBegin.time, Profiler::GetTime() );
+    MemWrite( &item->zoneBegin.srcloc, srcloc );
+    TracyLfqCommit;
+
+#ifdef TRACY_CALLSTACK
+    const uint32_t depth = TRACY_CALLSTACK;
+#else
+    const auto depth = uint32_t( lua_tointeger( L, 2 ) );
+#endif
+    SendLuaCallstack( L, depth );
+
+    return 0;
+}
+#endif
+
+static inline int LuaZoneBegin( lua_State* L )
+{
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+    return LuaZoneBeginS( L );
+#else
+#ifdef TRACY_ON_DEMAND
+    const auto zoneCnt = GetLuaZoneState().counter++;
+    if( zoneCnt != 0 && !GetLuaZoneState().active ) return 0;
+    GetLuaZoneState().active = GetProfiler().IsConnected();
+    if( !GetLuaZoneState().active ) return 0;
+#endif
+
+    TracyLfqPrepare( QueueType::ZoneBeginAllocSrcLoc );
+    lua_Debug dbg;
+    lua_getstack( L, 1, &dbg );
+    lua_getinfo( L, "Snl", &dbg );
+    const auto srcloc = Profiler::AllocSourceLocation( dbg.currentline, dbg.source, dbg.name ? dbg.name : dbg.short_src );
+    MemWrite( &item->zoneBegin.time, Profiler::GetTime() );
+    MemWrite( &item->zoneBegin.srcloc, srcloc );
+    TracyLfqCommit;
+    return 0;
+#endif
+}
+
 static inline int LuaZoneBeginN( lua_State* L )
 {
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+    return LuaZoneBeginNS( L );
+#else
 #ifdef TRACY_ON_DEMAND
-    const auto zoneCnt = s_luaZoneState.counter++;
-    if( zoneCnt != 0 && !s_luaZoneState.active ) return 0;
-    s_luaZoneState.active = s_profiler.IsConnected();
-    if( !s_luaZoneState.active ) return 0;
+    const auto zoneCnt = GetLuaZoneState().counter++;
+    if( zoneCnt != 0 && !GetLuaZoneState().active ) return 0;
+    GetLuaZoneState().active = GetProfiler().IsConnected();
+    if( !GetLuaZoneState().active ) return 0;
 #endif
 
-    const uint32_t color = Color::DeepSkyBlue3;
-
+    TracyLfqPrepare( QueueType::ZoneBeginAllocSrcLoc );
     lua_Debug dbg;
     lua_getstack( L, 1, &dbg );
     lua_getinfo( L, "Snl", &dbg );
-
-    const uint32_t line = dbg.currentline;
-    const auto func = dbg.name ? dbg.name : dbg.short_src;
     size_t nsz;
     const auto name = lua_tolstring( L, 1, &nsz );
-    const auto fsz = strlen( func );
-    const auto ssz = strlen( dbg.source );
-
-    // Data layout:
-    //  4b  payload size
-    //  4b  color
-    //  4b  source line
-    //  fsz function name
-    //  1b  null terminator
-    //  ssz source file name
-    //  1b  null terminator
-    //  nsz zone name
-    const uint32_t sz = uint32_t( 4 + 4 + 4 + fsz + 1 + ssz + 1 + nsz );
-    auto ptr = (char*)tracy_malloc( sz );
-    memcpy( ptr, &sz, 4 );
-    memcpy( ptr + 4, &color, 4 );
-    memcpy( ptr + 8, &line, 4 );
-    memcpy( ptr + 12, func, fsz+1 );
-    memcpy( ptr + 12 + fsz + 1, dbg.source, ssz + 1 );
-    memcpy( ptr + 12 + fsz + 1 + ssz + 1, name, nsz );
-
-    Magic magic;
-    auto& token = s_token.ptr;
-    auto& tail = token->get_tail_index();
-    auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneBeginAllocSrcLoc );
-#ifdef TRACY_RDTSCP_OPT
-    MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
-#else
-    uint32_t cpu;
-    MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
-    MemWrite( &item->zoneBegin.cpu, cpu );
-#endif
-    MemWrite( &item->zoneBegin.thread, GetThreadHandle() );
-    MemWrite( &item->zoneBegin.srcloc, (uint64_t)ptr );
-    tail.store( magic + 1, std::memory_order_release );
+    const auto srcloc = Profiler::AllocSourceLocation( dbg.currentline, dbg.source, dbg.name ? dbg.name : dbg.short_src, name, nsz );
+    MemWrite( &item->zoneBegin.time, Profiler::GetTime() );
+    MemWrite( &item->zoneBegin.srcloc, srcloc );
+    TracyLfqCommit;
     return 0;
+#endif
 }
 
 static inline int LuaZoneEnd( lua_State* L )
 {
 #ifdef TRACY_ON_DEMAND
-    assert( s_luaZoneState.counter != 0 );
-    s_luaZoneState.counter--;
-    if( !s_luaZoneState.active ) return 0;
-    if( !s_profiler.IsConnected() )
+    assert( GetLuaZoneState().counter != 0 );
+    GetLuaZoneState().counter--;
+    if( !GetLuaZoneState().active ) return 0;
+    if( !GetProfiler().IsConnected() )
     {
-        s_luaZoneState.active = false;
+        GetLuaZoneState().active = false;
         return 0;
     }
 #endif
 
-    Magic magic;
-    auto& token = s_token.ptr;
-    auto& tail = token->get_tail_index();
-    auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneEnd );
-#ifdef TRACY_RDTSCP_OPT
-    MemWrite( &item->zoneEnd.time, Profiler::GetTime( item->zoneEnd.cpu ) );
-#else
-    uint32_t cpu;
-    MemWrite( &item->zoneEnd.time, Profiler::GetTime( cpu ) );
-    MemWrite( &item->zoneEnd.cpu, cpu );
-#endif
-    MemWrite( &item->zoneEnd.thread, GetThreadHandle() );
-    tail.store( magic + 1, std::memory_order_release );
+    TracyLfqPrepare( QueueType::ZoneEnd );
+    MemWrite( &item->zoneEnd.time, Profiler::GetTime() );
+    TracyLfqCommit;
     return 0;
 }
 
 static inline int LuaZoneText( lua_State* L )
 {
 #ifdef TRACY_ON_DEMAND
-    if( !s_luaZoneState.active ) return 0;
-    if( !s_profiler.IsConnected() )
+    if( !GetLuaZoneState().active ) return 0;
+    if( !GetProfiler().IsConnected() )
     {
-        s_luaZoneState.active = false;
+        GetLuaZoneState().active = false;
         return 0;
     }
 #endif
@@ -281,27 +327,22 @@ static inline int LuaZoneText( lua_State* L )
     auto txt = lua_tostring( L, 1 );
     const auto size = strlen( txt );
 
-    Magic magic;
-    auto& token = s_token.ptr;
     auto ptr = (char*)tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
     ptr[size] = '\0';
-    auto& tail = token->get_tail_index();
-    auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneText );
-    MemWrite( &item->zoneText.thread, GetThreadHandle() );
+    TracyLfqPrepare( QueueType::ZoneText );
     MemWrite( &item->zoneText.text, (uint64_t)ptr );
-    tail.store( magic + 1, std::memory_order_release );
+    TracyLfqCommit;
     return 0;
 }
 
 static inline int LuaZoneName( lua_State* L )
 {
 #ifdef TRACY_ON_DEMAND
-    if( !s_luaZoneState.active ) return 0;
-    if( !s_profiler.IsConnected() )
+    if( !GetLuaZoneState().active ) return 0;
+    if( !GetProfiler().IsConnected() )
     {
-        s_luaZoneState.active = false;
+        GetLuaZoneState().active = false;
         return 0;
     }
 #endif
@@ -309,41 +350,31 @@ static inline int LuaZoneName( lua_State* L )
     auto txt = lua_tostring( L, 1 );
     const auto size = strlen( txt );
 
-    Magic magic;
-    auto& token = s_token.ptr;
     auto ptr = (char*)tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
     ptr[size] = '\0';
-    auto& tail = token->get_tail_index();
-    auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::ZoneName );
-    MemWrite( &item->zoneText.thread, GetThreadHandle() );
+    TracyLfqPrepare( QueueType::ZoneName );
     MemWrite( &item->zoneText.text, (uint64_t)ptr );
-    tail.store( magic + 1, std::memory_order_release );
+    TracyLfqCommit;
     return 0;
 }
 
 static inline int LuaMessage( lua_State* L )
 {
 #ifdef TRACY_ON_DEMAND
-    if( !s_profiler.IsConnected() ) return 0;
+    if( !GetProfiler().IsConnected() ) return 0;
 #endif
 
     auto txt = lua_tostring( L, 1 );
     const auto size = strlen( txt );
 
-    Magic magic;
-    auto& token = s_token.ptr;
+    TracyLfqPrepare( QueueType::Message );
     auto ptr = (char*)tracy_malloc( size+1 );
     memcpy( ptr, txt, size );
     ptr[size] = '\0';
-    auto& tail = token->get_tail_index();
-    auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-    MemWrite( &item->hdr.type, QueueType::Message );
     MemWrite( &item->message.time, Profiler::GetTime() );
-    MemWrite( &item->message.thread, GetThreadHandle() );
     MemWrite( &item->message.text, (uint64_t)ptr );
-    tail.store( magic + 1, std::memory_order_release );
+    TracyLfqCommit;
     return 0;
 }
 
@@ -356,6 +387,17 @@ static inline void LuaRegister( lua_State* L )
     lua_setfield( L, -2, "ZoneBegin" );
     lua_pushcfunction( L, detail::LuaZoneBeginN );
     lua_setfield( L, -2, "ZoneBeginN" );
+#ifdef TRACY_HAS_CALLSTACK
+    lua_pushcfunction( L, detail::LuaZoneBeginS );
+    lua_setfield( L, -2, "ZoneBeginS" );
+    lua_pushcfunction( L, detail::LuaZoneBeginNS );
+    lua_setfield( L, -2, "ZoneBeginNS" );
+#else
+    lua_pushcfunction( L, detail::LuaZoneBegin );
+    lua_setfield( L, -2, "ZoneBeginS" );
+    lua_pushcfunction( L, detail::LuaZoneBeginN );
+    lua_setfield( L, -2, "ZoneBeginNS" );
+#endif
     lua_pushcfunction( L, detail::LuaZoneEnd );
     lua_setfield( L, -2, "ZoneEnd" );
     lua_pushcfunction( L, detail::LuaZoneText );

TracyOpenCL.hpp

@@ -0,0 +1,333 @@
+#ifndef __TRACYOPENCL_HPP__
+#define __TRACYOPENCL_HPP__
+
+#if !defined TRACY_ENABLE
+
+#define TracyCLContext(x, y) nullptr
+#define TracyCLDestroy(x)
+#define TracyCLNamedZone(c, x, y, z, w)
+#define TracyCLNamedZoneC(c, x, y, z, w, a)
+#define TracyCLZone(c, x, y)
+#define TracyCLZoneC(c, x, y, z)
+#define TracyCLCollect(c)
+
+#define TracyCLNamedZoneS(c, x, y, z, w, a)
+#define TracyCLNamedZoneCS(c, x, y, z, w, v, a)
+#define TracyCLZoneS(c, x, y, z)
+#define TracyCLZoneCS(c, x, y, z, w)
+
+namespace tracy
+{
+    class OpenCLCtxScope {};
+}
+
+using TracyCLCtx = void*;
+
+#else
+
+#include <CL/cl.h>
+
+#include <atomic>
+#include <cassert>
+
+#include "Tracy.hpp"
+#include "client/TracyCallstack.hpp"
+#include "client/TracyProfiler.hpp"
+#include "common/TracyAlloc.hpp"
+
+namespace tracy {
+
+    enum class EventPhase : uint8_t
+    {
+        Begin,
+        End
+    };
+
+    struct EventInfo
+    {
+        cl_event event;
+        EventPhase phase;
+    };
+
+    class OpenCLCtx
+    {
+    public:
+        enum { QueryCount = 64 * 1024 };
+
+        OpenCLCtx(cl_context context, cl_device_id device)
+            : m_contextId(GetGpuCtxCounter().fetch_add(1, std::memory_order_relaxed))
+            , m_head(0)
+            , m_tail(0)
+        {
+            assert(m_contextId != 255);
+
+            m_hostStartTime = Profiler::GetTime();
+            m_deviceStartTime = GetDeviceTimestamp(context, device);
+
+            auto item = Profiler::QueueSerial();
+            MemWrite(&item->hdr.type, QueueType::GpuNewContext);
+            MemWrite(&item->gpuNewContext.cpuTime, m_hostStartTime);
+            MemWrite(&item->gpuNewContext.gpuTime, m_hostStartTime);
+            memset(&item->gpuNewContext.thread, 0, sizeof(item->gpuNewContext.thread));
+            MemWrite(&item->gpuNewContext.period, 1.0f);
+            MemWrite(&item->gpuNewContext.type, GpuContextType::OpenCL);
+            MemWrite(&item->gpuNewContext.context, (uint8_t) m_contextId);
+            MemWrite(&item->gpuNewContext.accuracyBits, (uint8_t)0);
+#ifdef TRACY_ON_DEMAND
+            GetProfiler().DeferItem(*item);
+#endif
+            Profiler::QueueSerialFinish();
+        }
+
+        void Collect()
+        {
+            ZoneScopedC(Color::Red4);
+
+            if (m_tail == m_head) return;
+
+#ifdef TRACY_ON_DEMAND
+            if (!GetProfiler().IsConnected())
+            {
+                m_head = m_tail = 0;
+            }
+#endif
+
+            while (m_tail != m_head)
+            {
+                EventInfo eventInfo = m_query[m_tail];
+                cl_event event = eventInfo.event;
+                cl_int eventStatus;
+                cl_int err = clGetEventInfo(event, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &eventStatus, nullptr);
+                assert(err == CL_SUCCESS);
+                if (eventStatus != CL_COMPLETE) return;
+
+                cl_int eventInfoQuery = (eventInfo.phase == EventPhase::Begin)
+                    ? CL_PROFILING_COMMAND_START
+                    : CL_PROFILING_COMMAND_END;
+
+                cl_ulong eventTimeStamp = 0;
+                err = clGetEventProfilingInfo(event, eventInfoQuery, sizeof(cl_ulong), &eventTimeStamp, nullptr);
+                assert(err == CL_SUCCESS);
+                assert(eventTimeStamp != 0);
+
+                auto item = Profiler::QueueSerial();
+                MemWrite(&item->hdr.type, QueueType::GpuTime);
+                MemWrite(&item->gpuTime.gpuTime, TimestampOffset(eventTimeStamp));
+                MemWrite(&item->gpuTime.queryId, (uint16_t)m_tail);
+                MemWrite(&item->gpuTime.context, m_contextId);
+                Profiler::QueueSerialFinish();
+
+                if (eventInfo.phase == EventPhase::End)
+                {
+                    // Done with the event, so release it
+                    assert(clReleaseEvent(event) == CL_SUCCESS);
+                }
+
+                m_tail = (m_tail + 1) % QueryCount;
+            }
+        }
+
+        tracy_force_inline uint8_t GetId() const
+        {
+            return m_contextId;
+        }
+
+        tracy_force_inline unsigned int NextQueryId(EventInfo eventInfo)
+        {
+            const auto id = m_head;
+            m_head = (m_head + 1) % QueryCount;
+            assert(m_head != m_tail);
+            m_query[id] = eventInfo;
+            return id;
+        }
+
+        tracy_force_inline EventInfo& GetQuery(unsigned int id)
+        {
+            assert(id < QueryCount);
+            return m_query[id];
+        }
+
+    private:
+        tracy_force_inline int64_t GetHostStartTime() const
+        {
+            return m_hostStartTime;
+        }
+
+        tracy_force_inline int64_t GetDeviceStartTime() const
+        {
+            return m_deviceStartTime;
+        }
+
+        tracy_force_inline int64_t TimestampOffset(int64_t deviceTimestamp) const
+        {
+            return m_hostStartTime + (deviceTimestamp - m_deviceStartTime);
+        }
+
+        tracy_force_inline int64_t GetDeviceTimestamp(cl_context context, cl_device_id device) const
+        {
+            cl_ulong deviceTimestamp = 0;
+            cl_int err = CL_SUCCESS;
+            cl_command_queue queue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
+            assert(err == CL_SUCCESS);
+            uint32_t dummyValue = 42;
+            cl_mem dummyBuffer = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(uint32_t), nullptr, &err);
+            assert(err == CL_SUCCESS);
+            cl_event writeBufferEvent;
+            err = clEnqueueWriteBuffer(queue, dummyBuffer, CL_TRUE, 0, sizeof(uint32_t), &dummyValue, 0, nullptr, &writeBufferEvent);
+            assert(err == CL_SUCCESS);
+            err = clWaitForEvents(1, &writeBufferEvent);
+            assert(err == CL_SUCCESS);
+            cl_int eventStatus;
+            err = clGetEventInfo(writeBufferEvent, CL_EVENT_COMMAND_EXECUTION_STATUS, sizeof(cl_int), &eventStatus, nullptr);
+            assert(err == CL_SUCCESS);
+            assert(eventStatus == CL_COMPLETE);
+            err = clGetEventProfilingInfo(writeBufferEvent, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &deviceTimestamp, nullptr);
+            assert(err == CL_SUCCESS);
+            err = clReleaseEvent(writeBufferEvent);
+            assert(err == CL_SUCCESS);
+            err = clReleaseMemObject(dummyBuffer);
+            assert(err == CL_SUCCESS);
+            err = clReleaseCommandQueue(queue);
+            assert(err == CL_SUCCESS);
+
+            return (int64_t)deviceTimestamp;
+        }
+
+        unsigned int m_contextId;
+
+        EventInfo m_query[QueryCount];
+        unsigned int m_head;
+        unsigned int m_tail;
+
+        int64_t m_hostStartTime;
+        int64_t m_deviceStartTime;
+    };
+
+    class OpenCLCtxScope {
+    public:
+        tracy_force_inline OpenCLCtxScope(OpenCLCtx* ctx, const SourceLocationData* srcLoc, bool is_active)
+#ifdef TRACY_ON_DEMAND
+            : m_active(is_active&& GetProfiler().IsConnected())
+#else
+            : m_active(is_active)
+#endif
+            , m_ctx(ctx)
+            , m_event(nullptr)
+        {
+            if (!m_active) return;
+
+            m_beginQueryId = ctx->NextQueryId(EventInfo{ nullptr, EventPhase::Begin });
+
+            auto item = Profiler::QueueSerial();
+            MemWrite(&item->hdr.type, QueueType::GpuZoneBeginSerial);
+            MemWrite(&item->gpuZoneBegin.cpuTime, Profiler::GetTime());
+            MemWrite(&item->gpuZoneBegin.srcloc, (uint64_t)srcLoc);
+            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
+            MemWrite(&item->gpuZoneBegin.queryId, (uint16_t)m_beginQueryId);
+            MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+            Profiler::QueueSerialFinish();
+        }
+
+        tracy_force_inline OpenCLCtxScope(OpenCLCtx* ctx, const SourceLocationData* srcLoc, int depth, bool is_active)
+#ifdef TRACY_ON_DEMAND
+            : m_active(is_active&& GetProfiler().IsConnected())
+#else
+            : m_active(is_active)
+#endif
+            , m_ctx(ctx)
+            , m_event(nullptr)
+        {
+            if (!m_active) return;
+
+            m_beginQueryId = ctx->NextQueryId(EventInfo{ nullptr, EventPhase::Begin });
+
+            auto item = Profiler::QueueSerial();
+            MemWrite(&item->hdr.type, QueueType::GpuZoneBeginCallstackSerial);
+            MemWrite(&item->gpuZoneBegin.cpuTime, Profiler::GetTime());
+            MemWrite(&item->gpuZoneBegin.srcloc, (uint64_t)srcLoc);
+            MemWrite(&item->gpuZoneBegin.thread, GetThreadHandle());
+            MemWrite(&item->gpuZoneBegin.queryId, (uint16_t)m_beginQueryId);
+            MemWrite(&item->gpuZoneBegin.context, ctx->GetId());
+            Profiler::QueueSerialFinish();
+
+            GetProfiler().SendCallstack(depth);
+        }
+
+        tracy_force_inline void SetEvent(cl_event event)
+        {
+            m_event = event;
+            assert(clRetainEvent(m_event) == CL_SUCCESS);
+            m_ctx->GetQuery(m_beginQueryId).event = m_event;
+        }
+
+        tracy_force_inline ~OpenCLCtxScope()
+        {
+            const auto queryId = m_ctx->NextQueryId(EventInfo{ m_event, EventPhase::End });
+
+            auto item = Profiler::QueueSerial();
+            MemWrite(&item->hdr.type, QueueType::GpuZoneEndSerial);
+            MemWrite(&item->gpuZoneEnd.cpuTime, Profiler::GetTime());
+            MemWrite(&item->gpuZoneEnd.thread, GetThreadHandle());
+            MemWrite(&item->gpuZoneEnd.queryId, (uint16_t)queryId);
+            MemWrite(&item->gpuZoneEnd.context, m_ctx->GetId());
+            Profiler::QueueSerialFinish();
+        }
+
+        const bool m_active;
+        OpenCLCtx* m_ctx;
+        cl_event m_event;
+        unsigned int m_beginQueryId;
+    };
+
+    static inline OpenCLCtx* CreateCLContext(cl_context context, cl_device_id device)
+    {
+        InitRPMallocThread();
+        auto ctx = (OpenCLCtx*)tracy_malloc(sizeof(OpenCLCtx));
+        new (ctx) OpenCLCtx(context, device);
+        return ctx;
+    }
+
+    static inline void DestroyCLContext(OpenCLCtx* ctx)
+    {
+        ctx->~OpenCLCtx();
+        tracy_free(ctx);
+    }
+
+}  // namespace tracy
+
+using TracyCLCtx = tracy::OpenCLCtx*;
+
+#define TracyCLContext(context, device) tracy::CreateCLContext(context, device);
+#define TracyCLDestroy(ctx) tracy::DestroyCLContext(ctx);
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+#   define TracyCLNamedZone(ctx, varname, name, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, 0 }; tracy::OpenCLCtxScope varname(ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), TRACY_CALLSTACK, active );
+#   define TracyCLNamedZoneC(ctx, varname, name, color, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, color }; tracy::OpenCLCtxScope varname(ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), TRACY_CALLSTACK, active );
+#   define TracyCLZone(ctx, name) TracyCLNamedZoneS(ctx, __tracy_gpu_zone, name, TRACY_CALLSTACK, true)
+#   define TracyCLZoneC(ctx, name, color) TracyCLNamedZoneCS(ctx, __tracy_gpu_zone, name, color, TRACY_CALLSTACK, true)
+#else
+#   define TracyCLNamedZone(ctx, varname, name, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__){ name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, 0 }; tracy::OpenCLCtxScope varname(ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), active);
+#   define TracyCLNamedZoneC(ctx, varname, name, color, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__){ name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, color }; tracy::OpenCLCtxScope varname(ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), active);
+#   define TracyCLZone(ctx, name) TracyCLNamedZone(ctx, __tracy_gpu_zone, name, true)
+#   define TracyCLZoneC(ctx, name, color) TracyCLNamedZoneC(ctx, __tracy_gpu_zone, name, color, true )
+#endif
+
+#ifdef TRACY_HAS_CALLSTACK
+#   define TracyCLNamedZoneS(ctx, varname, name, depth, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__){ name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, 0 }; tracy::OpenCLCtxScope varname(ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), depth, active);
+#   define TracyCLNamedZoneCS(ctx, varname, name, color, depth, active) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__){ name, __FUNCTION__, __FILE__, (uint32_t)__LINE__, color }; tracy::OpenCLCtxScope varname(ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), depth, active);
+#   define TracyCLZoneS(ctx, name, depth) TracyCLNamedZoneS(ctx, __tracy_gpu_zone, name, depth, true)
+#   define TracyCLZoneCS(ctx, name, color, depth) TracyCLNamedZoneCS(ctx, __tracy_gpu_zone, name, color, depth, true)
+#else
+#define TracyCLNamedZoneS(ctx, varname, name, depth, active) TracyCLNamedZone(ctx, varname, name, active)
+#define TracyCLNamedZoneCS(ctx, varname, name, color, depth, active) TracyCLNamedZoneC(ctx, varname, name, color, active)
+#define TracyCLZoneS(ctx, name, depth) TracyCLZone(ctx, name)
+#define TracyCLZoneCS(ctx, name, color, depth) TracyCLZoneC(ctx, name, color)
+#endif
+
+#define TracyCLNamedZoneSetEvent(varname, event) varname.SetEvent(event)
+#define TracyCLZoneSetEvent(event) __tracy_gpu_zone.SetEvent(event)
+
+#define TracyCLCollect(ctx) ctx->Collect()
+
+#endif
+
+#endif

TracyOpenGL.hpp

@@ -1,22 +1,35 @@
 #ifndef __TRACYOPENGL_HPP__
 #define __TRACYOPENGL_HPP__
 
-// Include this file after you include OpenGL 3.2 headers.
+#if !defined GL_TIMESTAMP && !defined GL_TIMESTAMP_EXT
+#  error "You must include OpenGL 3.2 headers before including TracyOpenGL.hpp"
+#endif
 
 #if !defined TRACY_ENABLE || defined __APPLE__
 
 #define TracyGpuContext
-#define TracyGpuNamedZone(x,y)
-#define TracyGpuNamedZoneC(x,y,z)
+#define TracyGpuNamedZone(x,y,z)
+#define TracyGpuNamedZoneC(x,y,z,w)
 #define TracyGpuZone(x)
 #define TracyGpuZoneC(x,y)
 #define TracyGpuCollect
 
-#define TracyGpuNamedZoneS(x,y,z)
-#define TracyGpuNamedZoneCS(x,y,z,w)
+#define TracyGpuNamedZoneS(x,y,z,w)
+#define TracyGpuNamedZoneCS(x,y,z,w,a)
 #define TracyGpuZoneS(x,y)
 #define TracyGpuZoneCS(x,y,z)
 
+namespace tracy
+{
+struct SourceLocationData;
+class GpuCtxScope
+{
+public:
+    GpuCtxScope( const SourceLocationData*, bool ) {}
+    GpuCtxScope( const SourceLocationData*, int, bool ) {}
+};
+}
+
 #else
 
 #include <atomic>
@@ -29,28 +42,36 @@
 #include "common/TracyAlign.hpp"
 #include "common/TracyAlloc.hpp"
 
-#define TracyGpuContext tracy::s_gpuCtx.ptr = (tracy::GpuCtx*)tracy::tracy_malloc( sizeof( tracy::GpuCtx ) ); new(tracy::s_gpuCtx.ptr) tracy::GpuCtx;
-#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
-#  define TracyGpuNamedZone( varname, name ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), TRACY_CALLSTACK );
-#  define TracyGpuNamedZoneC( varname, name, color ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), TRACY_CALLSTACK );
-#  define TracyGpuZone( name ) TracyGpuNamedZoneS( ___tracy_gpu_zone, name, TRACY_CALLSTACK )
-#  define TracyGpuZoneC( name, color ) TracyGpuNamedZoneCS( ___tracy_gpu_zone, name, color, TRACY_CALLSTACK )
-#else
-#  define TracyGpuNamedZone( varname, name ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__) );
-#  define TracyGpuNamedZoneC( varname, name, color ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__) );
-#  define TracyGpuZone( name ) TracyGpuNamedZone( ___tracy_gpu_zone, name )
-#  define TracyGpuZoneC( name, color ) TracyGpuNamedZoneC( ___tracy_gpu_zone, name, color )
+#if !defined GL_TIMESTAMP && defined GL_TIMESTAMP_EXT
+#  define GL_TIMESTAMP GL_TIMESTAMP_EXT
+#  define GL_QUERY_COUNTER_BITS GL_QUERY_COUNTER_BITS_EXT
+#  define glGetQueryObjectiv glGetQueryObjectivEXT
+#  define glGetQueryObjectui64v glGetQueryObjectui64vEXT
+#  define glQueryCounter glQueryCounterEXT
 #endif
-#define TracyGpuCollect tracy::s_gpuCtx.ptr->Collect();
+
+#define TracyGpuContext tracy::InitRPMallocThread(); tracy::GetGpuCtx().ptr = (tracy::GpuCtx*)tracy::tracy_malloc( sizeof( tracy::GpuCtx ) ); new(tracy::GetGpuCtx().ptr) tracy::GpuCtx;
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+#  define TracyGpuNamedZone( varname, name, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), TRACY_CALLSTACK, active );
+#  define TracyGpuNamedZoneC( varname, name, color, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), TRACY_CALLSTACK, active );
+#  define TracyGpuZone( name ) TracyGpuNamedZoneS( ___tracy_gpu_zone, name, TRACY_CALLSTACK, true )
+#  define TracyGpuZoneC( name, color ) TracyGpuNamedZoneCS( ___tracy_gpu_zone, name, color, TRACY_CALLSTACK, true )
+#else
+#  define TracyGpuNamedZone( varname, name, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), active );
+#  define TracyGpuNamedZoneC( varname, name, color, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), active );
+#  define TracyGpuZone( name ) TracyGpuNamedZone( ___tracy_gpu_zone, name, true )
+#  define TracyGpuZoneC( name, color ) TracyGpuNamedZoneC( ___tracy_gpu_zone, name, color, true )
+#endif
+#define TracyGpuCollect tracy::GetGpuCtx().ptr->Collect();
 
 #ifdef TRACY_HAS_CALLSTACK
-#  define TracyGpuNamedZoneS( varname, name, depth ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), depth );
-#  define TracyGpuNamedZoneCS( varname, name, color, depth ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), depth );
-#  define TracyGpuZoneS( name, depth ) TracyGpuNamedZoneS( ___tracy_gpu_zone, name, depth )
-#  define TracyGpuZoneCS( name, color, depth ) TracyGpuNamedZoneCS( ___tracy_gpu_zone, name, color, depth )
+#  define TracyGpuNamedZoneS( varname, name, depth, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), depth, active );
+#  define TracyGpuNamedZoneCS( varname, name, color, depth, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::GpuCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), depth, active );
+#  define TracyGpuZoneS( name, depth ) TracyGpuNamedZoneS( ___tracy_gpu_zone, name, depth, true )
+#  define TracyGpuZoneCS( name, color, depth ) TracyGpuNamedZoneCS( ___tracy_gpu_zone, name, color, depth, true )
 #else
-#  define TracyGpuNamedZoneS( varname, name, depth ) TracyGpuNamedZone( varname, name )
-#  define TracyGpuNamedZoneCS( varname, name, color, depth ) TracyGpuNamedZoneC( varname, name, color )
+#  define TracyGpuNamedZoneS( varname, name, depth, active ) TracyGpuNamedZone( varname, name, active )
+#  define TracyGpuNamedZoneCS( varname, name, color, depth, active ) TracyGpuNamedZoneC( varname, name, color, active )
 #  define TracyGpuZoneS( name, depth ) TracyGpuZone( name )
 #  define TracyGpuZoneCS( name, color, depth ) TracyGpuZoneC( name, color )
 #endif
@@ -58,8 +79,6 @@
 namespace tracy
 {
 
-extern std::atomic<uint8_t> s_gpuCtxCounter;
-
 class GpuCtx
 {
     friend class GpuCtxScope;
@@ -68,7 +87,7 @@ class GpuCtx
 
 public:
     GpuCtx()
-        : m_context( s_gpuCtxCounter.fetch_add( 1, std::memory_order_relaxed ) )
+        : m_context( GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed ) )
         , m_head( 0 )
         , m_tail( 0 )
     {
@@ -84,23 +103,21 @@ public:
         glGetQueryiv( GL_TIMESTAMP, GL_QUERY_COUNTER_BITS, &bits );
 
         const float period = 1.f;
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuNewContext );
+        const auto thread = GetThreadHandle();
+        TracyLfqPrepare( QueueType::GpuNewContext );
         MemWrite( &item->gpuNewContext.cpuTime, tcpu );
         MemWrite( &item->gpuNewContext.gpuTime, tgpu );
-        MemWrite( &item->gpuNewContext.thread, GetThreadHandle() );
+        MemWrite( &item->gpuNewContext.thread, thread );
         MemWrite( &item->gpuNewContext.period, period );
         MemWrite( &item->gpuNewContext.context, m_context );
         MemWrite( &item->gpuNewContext.accuracyBits, (uint8_t)bits );
+        MemWrite( &item->gpuNewContext.type, GpuContextType::OpenGl );
 
 #ifdef TRACY_ON_DEMAND
-        s_profiler.DeferItem( *item );
+        GetProfiler().DeferItem( *item );
 #endif
 
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
     void Collect()
@@ -110,49 +127,28 @@ public:
         if( m_tail == m_head ) return;
 
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() )
+        if( !GetProfiler().IsConnected() )
         {
             m_head = m_tail = 0;
             return;
         }
 #endif
 
-        auto start = m_tail;
-        auto end = m_head + QueryCount;
-        auto cnt = ( end - start ) % QueryCount;
-        while( cnt > 1 )
+        while( m_tail != m_head )
         {
-            auto mid = start + cnt / 2;
             GLint available;
-            glGetQueryObjectiv( m_query[mid % QueryCount], GL_QUERY_RESULT_AVAILABLE, &available );
-            if( available )
-            {
-                start = mid;
-            }
-            else
-            {
-                end = mid;
-            }
-            cnt = ( end - start ) % QueryCount;
-        }
+            glGetQueryObjectiv( m_query[m_tail], GL_QUERY_RESULT_AVAILABLE, &available );
+            if( !available ) return;
 
-        start %= QueryCount;
-
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-
-        while( m_tail != start )
-        {
             uint64_t time;
             glGetQueryObjectui64v( m_query[m_tail], GL_QUERY_RESULT, &time );
 
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::GpuTime );
+            TracyLfqPrepare( QueueType::GpuTime );
             MemWrite( &item->gpuTime.gpuTime, (int64_t)time );
             MemWrite( &item->gpuTime.queryId, (uint16_t)m_tail );
             MemWrite( &item->gpuTime.context, m_context );
-            tail.store( magic + 1, std::memory_order_release );
+            TracyLfqCommit;
+
             m_tail = ( m_tail + 1 ) % QueryCount;
         }
     }
@@ -183,86 +179,71 @@ private:
     unsigned int m_tail;
 };
 
-extern thread_local GpuCtxWrapper s_gpuCtx;
-
 class GpuCtxScope
 {
 public:
-    tracy_force_inline GpuCtxScope( const SourceLocationData* srcloc )
+    tracy_force_inline GpuCtxScope( const SourceLocationData* srcloc, bool is_active )
 #ifdef TRACY_ON_DEMAND
-        : m_active( s_profiler.IsConnected() )
+        : m_active( is_active && GetProfiler().IsConnected() )
+#else
+        : m_active( is_active )
 #endif
     {
-#ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
-#endif
-        const auto queryId = s_gpuCtx.ptr->NextQueryId();
-        glQueryCounter( s_gpuCtx.ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuZoneBegin );
+        const auto queryId = GetGpuCtx().ptr->NextQueryId();
+        glQueryCounter( GetGpuCtx().ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
+
+        TracyLfqPrepare( QueueType::GpuZoneBegin );
         MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
         MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
         memset( &item->gpuZoneBegin.thread, 0, sizeof( item->gpuZoneBegin.thread ) );
         MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, s_gpuCtx.ptr->GetId() );
-        tail.store( magic + 1, std::memory_order_release );
+        MemWrite( &item->gpuZoneBegin.context, GetGpuCtx().ptr->GetId() );
+        TracyLfqCommit;
     }
 
-    tracy_force_inline GpuCtxScope( const SourceLocationData* srcloc, int depth )
+    tracy_force_inline GpuCtxScope( const SourceLocationData* srcloc, int depth, bool is_active )
 #ifdef TRACY_ON_DEMAND
-        : m_active( s_profiler.IsConnected() )
+        : m_active( is_active && GetProfiler().IsConnected() )
+#else
+        : m_active( is_active )
 #endif
     {
-#ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
-#endif
-        const auto queryId = s_gpuCtx.ptr->NextQueryId();
-        glQueryCounter( s_gpuCtx.ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
+
+        const auto queryId = GetGpuCtx().ptr->NextQueryId();
+        glQueryCounter( GetGpuCtx().ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
 
         const auto thread = GetThreadHandle();
-
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuZoneBeginCallstack );
+        TracyLfqPrepare( QueueType::GpuZoneBeginCallstack );
         MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
         MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
         MemWrite( &item->gpuZoneBegin.thread, thread );
         MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneBegin.context, s_gpuCtx.ptr->GetId() );
-        tail.store( magic + 1, std::memory_order_release );
+        MemWrite( &item->gpuZoneBegin.context, GetGpuCtx().ptr->GetId() );
+        TracyLfqCommit;
 
-        s_profiler.SendCallstack( depth, thread );
+        GetProfiler().SendCallstack( depth );
     }
 
     tracy_force_inline ~GpuCtxScope()
     {
-#ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
-#endif
-        const auto queryId = s_gpuCtx.ptr->NextQueryId();
-        glQueryCounter( s_gpuCtx.ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuZoneEnd );
+        const auto queryId = GetGpuCtx().ptr->NextQueryId();
+        glQueryCounter( GetGpuCtx().ptr->TranslateOpenGlQueryId( queryId ), GL_TIMESTAMP );
+
+        TracyLfqPrepare( QueueType::GpuZoneEnd );
         MemWrite( &item->gpuZoneEnd.cpuTime, Profiler::GetTime() );
+        memset( &item->gpuZoneEnd.thread, 0, sizeof( item->gpuZoneEnd.thread ) );
         MemWrite( &item->gpuZoneEnd.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneEnd.context, s_gpuCtx.ptr->GetId() );
-        tail.store( magic + 1, std::memory_order_release );
+        MemWrite( &item->gpuZoneEnd.context, GetGpuCtx().ptr->GetId() );
+        TracyLfqCommit;
     }
 
 private:
-#ifdef TRACY_ON_DEMAND
     const bool m_active;
-#endif
 };
 
 }

TracyVulkan.hpp

@@ -3,18 +3,25 @@
 
 #if !defined TRACY_ENABLE
 
-#define TracyVkContext(x,y,z,w)
-#define TracyVkDestroy
-#define TracyVkNamedZone(x,y,z)
-#define TracyVkNamedZoneC(x,y,z,w)
-#define TracyVkZone(x,y)
-#define TracyVkZoneC(x,y,z)
-#define TracyVkCollect(x)
+#define TracyVkContext(x,y,z,w) nullptr
+#define TracyVkDestroy(x)
+#define TracyVkNamedZone(c,x,y,z,w)
+#define TracyVkNamedZoneC(c,x,y,z,w,a)
+#define TracyVkZone(c,x,y)
+#define TracyVkZoneC(c,x,y,z)
+#define TracyVkCollect(c,x)
 
-#define TracyVkNamedZoneS(x,y,z,w)
-#define TracyVkNamedZoneCS(x,y,z,w,v)
-#define TracyVkZoneS(x,y,z)
-#define TracyVkZoneCS(x,y,z,w)
+#define TracyVkNamedZoneS(c,x,y,z,w,a)
+#define TracyVkNamedZoneCS(c,x,y,z,w,v,a)
+#define TracyVkZoneS(c,x,y,z)
+#define TracyVkZoneCS(c,x,y,z,w)
+
+namespace tracy
+{
+class VkCtxScope {};
+}
+
+using TracyVkCtx = void*;
 
 #else
 
@@ -25,38 +32,9 @@
 #include "client/TracyProfiler.hpp"
 #include "client/TracyCallstack.hpp"
 
-#define TracyVkContext( physdev, device, queue, cmdbuf ) tracy::s_vkCtx.ptr = (tracy::VkCtx*)tracy::tracy_malloc( sizeof( tracy::VkCtx ) ); new(tracy::s_vkCtx.ptr) tracy::VkCtx( physdev, device, queue, cmdbuf );
-#define TracyVkDestroy tracy::s_vkCtx.ptr->~VkCtx(); tracy::tracy_free( tracy::s_vkCtx.ptr ); tracy::s_vkCtx.ptr = nullptr;
-#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
-#  define TracyVkNamedZone( varname, cmdbuf, name ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::VkCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, TRACY_CALLSTACK );
-#  define TracyVkNamedZoneC( varname, cmdbuf, name, color ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::VkCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, TRACY_CALLSTACK );
-#  define TracyVkZone( cmdbuf, name ) TracyVkNamedZoneS( ___tracy_gpu_zone, cmdbuf, name, TRACY_CALLSTACK )
-#  define TracyVkZoneC( cmdbuf, name, color ) TracyVkNamedZoneCS( ___tracy_gpu_zone, cmdbuf, name, color, TRACY_CALLSTACK )
-#else
-#  define TracyVkNamedZone( varname, cmdbuf, name ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::VkCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf );
-#  define TracyVkNamedZoneC( varname, cmdbuf, name, color ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::VkCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf );
-#  define TracyVkZone( cmdbuf, name ) TracyVkNamedZone( ___tracy_gpu_zone, cmdbuf, name )
-#  define TracyVkZoneC( cmdbuf, name, color ) TracyVkNamedZoneC( ___tracy_gpu_zone, cmdbuf, name, color )
-#endif
-#define TracyVkCollect( cmdbuf ) tracy::s_vkCtx.ptr->Collect( cmdbuf );
-
-#ifdef TRACY_HAS_CALLSTACK
-#  define TracyVkNamedZoneS( varname, cmdbuf, name, depth ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::VkCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, depth );
-#  define TracyVkNamedZoneCS( varname, cmdbuf, name, color, depth ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::VkCtxScope varname( &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, depth );
-#  define TracyVkZoneS( cmdbuf, name, depth ) TracyVkNamedZoneS( ___tracy_gpu_zone, cmdbuf, name, depth )
-#  define TracyVkZoneCS( cmdbuf, name, color, depth ) TracyVkNamedZoneCS( ___tracy_gpu_zone, cmdbuf, name, color, depth )
-#else
-#  define TracyVkNamedZoneS( varname, cmdbuf, name, depth ) TracyVkNamedZone( varname, cmdbuf, name )
-#  define TracyVkNamedZoneCS( varname, cmdbuf, name, color, depth ) TracyVkNamedZoneC( varname, cmdbuf, name, color )
-#  define TracyVkZoneS( cmdbuf, name, depth ) TracyVkZone( cmdbuf, name )
-#  define TracyVkZoneCS( cmdbuf, name, color, depth ) TracyVkZoneC( cmdbuf, name, color )
-#endif
-
 namespace tracy
 {
 
-extern std::atomic<uint8_t> s_gpuCtxCounter;
-
 class VkCtx
 {
     friend class VkCtxScope;
@@ -66,11 +44,11 @@ class VkCtx
 public:
     VkCtx( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf )
         : m_device( device )
-        , m_queue( queue )
-        , m_context( s_gpuCtxCounter.fetch_add( 1, std::memory_order_relaxed ) )
+        , m_context( GetGpuCtxCounter().fetch_add( 1, std::memory_order_relaxed ) )
         , m_head( 0 )
         , m_tail( 0 )
         , m_oldCnt( 0 )
+        , m_queryCount( QueryCount )
     {
         assert( m_context != 255 );
 
@@ -80,9 +58,13 @@ public:
 
         VkQueryPoolCreateInfo poolInfo = {};
         poolInfo.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO;
-        poolInfo.queryCount = QueryCount;
+        poolInfo.queryCount = m_queryCount;
         poolInfo.queryType = VK_QUERY_TYPE_TIMESTAMP;
-        vkCreateQueryPool( device, &poolInfo, nullptr, &m_query );
+        while( vkCreateQueryPool( device, &poolInfo, nullptr, &m_query ) != VK_SUCCESS )
+        {
+            m_queryCount /= 2;
+            poolInfo.queryCount = m_queryCount;
+        }
 
         VkCommandBufferBeginInfo beginInfo = {};
         beginInfo.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
@@ -94,7 +76,7 @@ public:
         submitInfo.pCommandBuffers = &cmdbuf;
 
         vkBeginCommandBuffer( cmdbuf, &beginInfo );
-        vkCmdResetQueryPool( cmdbuf, m_query, 0, QueryCount );
+        vkCmdResetQueryPool( cmdbuf, m_query, 0, m_queryCount );
         vkEndCommandBuffer( cmdbuf );
         vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE );
         vkQueueWaitIdle( queue );
@@ -115,10 +97,7 @@ public:
         vkQueueSubmit( queue, 1, &submitInfo, VK_NULL_HANDLE );
         vkQueueWaitIdle( queue );
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+        auto item = Profiler::QueueSerial();
         MemWrite( &item->hdr.type, QueueType::GpuNewContext );
         MemWrite( &item->gpuNewContext.cpuTime, tcpu );
         MemWrite( &item->gpuNewContext.gpuTime, tgpu );
@@ -126,16 +105,19 @@ public:
         MemWrite( &item->gpuNewContext.period, period );
         MemWrite( &item->gpuNewContext.context, m_context );
         MemWrite( &item->gpuNewContext.accuracyBits, uint8_t( 0 ) );
+        MemWrite( &item->gpuNewContext.type, GpuContextType::Vulkan );
 
 #ifdef TRACY_ON_DEMAND
-        s_profiler.DeferItem( *item );
+        GetProfiler().DeferItem( *item );
 #endif
+        Profiler::QueueSerialFinish();
 
-        tail.store( magic + 1, std::memory_order_release );
+        m_res = (int64_t*)tracy_malloc( sizeof( int64_t ) * m_queryCount );
     }
 
     ~VkCtx()
     {
+        tracy_free( m_res );
         vkDestroyQueryPool( m_device, m_query, nullptr );
     }
 
@@ -146,9 +128,9 @@ public:
         if( m_tail == m_head ) return;
 
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() )
+        if( !GetProfiler().IsConnected() )
         {
-            vkCmdResetQueryPool( cmdbuf, m_query, 0, QueryCount );
+            vkCmdResetQueryPool( cmdbuf, m_query, 0, m_queryCount );
             m_head = m_tail = 0;
             return;
         }
@@ -162,41 +144,36 @@ public:
         }
         else
         {
-            cnt = m_head < m_tail ? QueryCount - m_tail : m_head - m_tail;
+            cnt = m_head < m_tail ? m_queryCount - m_tail : m_head - m_tail;
         }
 
-        int64_t res[QueryCount];
-        if( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( res ), res, sizeof( *res ), VK_QUERY_RESULT_64_BIT ) == VK_NOT_READY )
+        if( vkGetQueryPoolResults( m_device, m_query, m_tail, cnt, sizeof( int64_t ) * m_queryCount, m_res, sizeof( int64_t ), VK_QUERY_RESULT_64_BIT ) == VK_NOT_READY )
         {
             m_oldCnt = cnt;
             return;
         }
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-
         for( unsigned int idx=0; idx<cnt; idx++ )
         {
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+            auto item = Profiler::QueueSerial();
             MemWrite( &item->hdr.type, QueueType::GpuTime );
-            MemWrite( &item->gpuTime.gpuTime, res[idx] );
+            MemWrite( &item->gpuTime.gpuTime, m_res[idx] );
             MemWrite( &item->gpuTime.queryId, uint16_t( m_tail + idx ) );
             MemWrite( &item->gpuTime.context, m_context );
-            tail.store( magic + 1, std::memory_order_release );
+            Profiler::QueueSerialFinish();
         }
 
         vkCmdResetQueryPool( cmdbuf, m_query, m_tail, cnt );
 
         m_tail += cnt;
-        if( m_tail == QueryCount ) m_tail = 0;
+        if( m_tail == m_queryCount ) m_tail = 0;
     }
 
 private:
     tracy_force_inline unsigned int NextQueryId()
     {
         const auto id = m_head;
-        m_head = ( m_head + 1 ) % QueryCount;
+        m_head = ( m_head + 1 ) % m_queryCount;
         assert( m_head != m_tail );
         return id;
     }
@@ -207,106 +184,138 @@ private:
     }
 
     VkDevice m_device;
-    VkQueue m_queue;
     VkQueryPool m_query;
     uint8_t m_context;
 
     unsigned int m_head;
     unsigned int m_tail;
     unsigned int m_oldCnt;
-};
+    unsigned int m_queryCount;
 
-extern VkCtxWrapper s_vkCtx;
+    int64_t* m_res;
+};
 
 class VkCtxScope
 {
 public:
-    tracy_force_inline VkCtxScope( const SourceLocationData* srcloc, VkCommandBuffer cmdbuf )
-        : m_cmdbuf( cmdbuf )
+    tracy_force_inline VkCtxScope( VkCtx* ctx, const SourceLocationData* srcloc, VkCommandBuffer cmdbuf, bool is_active )
 #ifdef TRACY_ON_DEMAND
-        , m_active( s_profiler.IsConnected() )
+        : m_active( is_active && GetProfiler().IsConnected() )
+#else
+        : m_active( is_active )
 #endif
     {
-#ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
-#endif
-        auto ctx = s_vkCtx.ptr;
-        const auto queryId = ctx->NextQueryId();
-        vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, ctx->m_query, queryId );
+        m_cmdbuf = cmdbuf;
+        m_ctx = ctx;
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuZoneBegin );
+        const auto queryId = ctx->NextQueryId();
+        vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId );
+
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::GpuZoneBeginSerial );
         MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
         MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
         MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
         MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
         MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
-        tail.store( magic + 1, std::memory_order_release );
+        Profiler::QueueSerialFinish();
     }
 
-    tracy_force_inline VkCtxScope( const SourceLocationData* srcloc, VkCommandBuffer cmdbuf, int depth )
-        : m_cmdbuf( cmdbuf )
+    tracy_force_inline VkCtxScope( VkCtx* ctx, const SourceLocationData* srcloc, VkCommandBuffer cmdbuf, int depth, bool is_active )
 #ifdef TRACY_ON_DEMAND
-        , m_active( s_profiler.IsConnected() )
+        : m_active( is_active && GetProfiler().IsConnected() )
+#else
+        : m_active( is_active )
 #endif
     {
-#ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
-#endif
-        const auto thread = GetThreadHandle();
+        m_cmdbuf = cmdbuf;
+        m_ctx = ctx;
 
-        auto ctx = s_vkCtx.ptr;
         const auto queryId = ctx->NextQueryId();
-        vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, ctx->m_query, queryId );
+        vkCmdWriteTimestamp( cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, ctx->m_query, queryId );
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuZoneBeginCallstack );
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::GpuZoneBeginCallstackSerial );
         MemWrite( &item->gpuZoneBegin.cpuTime, Profiler::GetTime() );
         MemWrite( &item->gpuZoneBegin.srcloc, (uint64_t)srcloc );
-        MemWrite( &item->gpuZoneBegin.thread, thread );
+        MemWrite( &item->gpuZoneBegin.thread, GetThreadHandle() );
         MemWrite( &item->gpuZoneBegin.queryId, uint16_t( queryId ) );
         MemWrite( &item->gpuZoneBegin.context, ctx->GetId() );
-        tail.store( magic + 1, std::memory_order_release );
+        Profiler::QueueSerialFinish();
 
-        s_profiler.SendCallstack( depth, thread );
+        GetProfiler().SendCallstack( depth );
     }
 
     tracy_force_inline ~VkCtxScope()
     {
-#ifdef TRACY_ON_DEMAND
         if( !m_active ) return;
-#endif
-        auto ctx = s_vkCtx.ptr;
-        const auto queryId = ctx->NextQueryId();
-        vkCmdWriteTimestamp( m_cmdbuf, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, ctx->m_query, queryId );
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::GpuZoneEnd );
+        const auto queryId = m_ctx->NextQueryId();
+        vkCmdWriteTimestamp( m_cmdbuf, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, m_ctx->m_query, queryId );
+
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::GpuZoneEndSerial );
         MemWrite( &item->gpuZoneEnd.cpuTime, Profiler::GetTime() );
+        MemWrite( &item->gpuZoneEnd.thread, GetThreadHandle() );
         MemWrite( &item->gpuZoneEnd.queryId, uint16_t( queryId ) );
-        MemWrite( &item->gpuZoneEnd.context, ctx->GetId() );
-        tail.store( magic + 1, std::memory_order_release );
+        MemWrite( &item->gpuZoneEnd.context, m_ctx->GetId() );
+        Profiler::QueueSerialFinish();
     }
 
 private:
-    VkCommandBuffer m_cmdbuf;
-
-#ifdef TRACY_ON_DEMAND
     const bool m_active;
-#endif
+
+    VkCommandBuffer m_cmdbuf;
+    VkCtx* m_ctx;
 };
 
+static inline VkCtx* CreateVkContext( VkPhysicalDevice physdev, VkDevice device, VkQueue queue, VkCommandBuffer cmdbuf )
+{
+    InitRPMallocThread();
+    auto ctx = (VkCtx*)tracy_malloc( sizeof( VkCtx ) );
+    new(ctx) VkCtx( physdev, device, queue, cmdbuf );
+    return ctx;
+}
+
+static inline void DestroyVkContext( VkCtx* ctx )
+{
+    ctx->~VkCtx();
+    tracy_free( ctx );
+}
+
 }
 
+using TracyVkCtx = tracy::VkCtx*;
+
+#define TracyVkContext( physdev, device, queue, cmdbuf ) tracy::CreateVkContext( physdev, device, queue, cmdbuf );
+#define TracyVkDestroy( ctx ) tracy::DestroyVkContext( ctx );
+#if defined TRACY_HAS_CALLSTACK && defined TRACY_CALLSTACK
+#  define TracyVkNamedZone( ctx, varname, cmdbuf, name, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::VkCtxScope varname( ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, TRACY_CALLSTACK, active );
+#  define TracyVkNamedZoneC( ctx, varname, cmdbuf, name, color, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::VkCtxScope varname( ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, TRACY_CALLSTACK, active );
+#  define TracyVkZone( ctx, cmdbuf, name ) TracyVkNamedZoneS( ctx, ___tracy_gpu_zone, cmdbuf, name, TRACY_CALLSTACK, true )
+#  define TracyVkZoneC( ctx, cmdbuf, name, color ) TracyVkNamedZoneCS( ctx, ___tracy_gpu_zone, cmdbuf, name, color, TRACY_CALLSTACK, true )
+#else
+#  define TracyVkNamedZone( ctx, varname, cmdbuf, name, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::VkCtxScope varname( ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, active );
+#  define TracyVkNamedZoneC( ctx, varname, cmdbuf, name, color, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::VkCtxScope varname( ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, active );
+#  define TracyVkZone( ctx, cmdbuf, name ) TracyVkNamedZone( ctx, ___tracy_gpu_zone, cmdbuf, name, true )
+#  define TracyVkZoneC( ctx, cmdbuf, name, color ) TracyVkNamedZoneC( ctx, ___tracy_gpu_zone, cmdbuf, name, color, true )
+#endif
+#define TracyVkCollect( ctx, cmdbuf ) ctx->Collect( cmdbuf );
+
+#ifdef TRACY_HAS_CALLSTACK
+#  define TracyVkNamedZoneS( ctx, varname, cmdbuf, name, depth, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, 0 }; tracy::VkCtxScope varname( ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, depth, active );
+#  define TracyVkNamedZoneCS( ctx, varname, cmdbuf, name, color, depth, active ) static const tracy::SourceLocationData TracyConcat(__tracy_gpu_source_location,__LINE__) { name, __FUNCTION__,  __FILE__, (uint32_t)__LINE__, color }; tracy::VkCtxScope varname( ctx, &TracyConcat(__tracy_gpu_source_location,__LINE__), cmdbuf, depth, active );
+#  define TracyVkZoneS( ctx, cmdbuf, name, depth ) TracyVkNamedZoneS( ctx, ___tracy_gpu_zone, cmdbuf, name, depth, true )
+#  define TracyVkZoneCS( ctx, cmdbuf, name, color, depth ) TracyVkNamedZoneCS( ctx, ___tracy_gpu_zone, cmdbuf, name, color, depth, true )
+#else
+#  define TracyVkNamedZoneS( ctx, varname, cmdbuf, name, depth, active ) TracyVkNamedZone( ctx, varname, cmdbuf, name, active )
+#  define TracyVkNamedZoneCS( ctx, varname, cmdbuf, name, color, depth, active ) TracyVkNamedZoneC( ctx, varname, cmdbuf, name, color, active )
+#  define TracyVkZoneS( ctx, cmdbuf, name, depth ) TracyVkZone( ctx, cmdbuf, name )
+#  define TracyVkZoneCS( ctx, cmdbuf, name, color, depth ) TracyVkZoneC( ctx, cmdbuf, name, color )
+#endif
+
 #endif
 
 #endif

capture/build/unix/build.mk

@@ -1,8 +1,8 @@
 CFLAGS +=
 CXXFLAGS := $(CFLAGS) -std=gnu++17
 DEFINES += -DTRACY_NO_STATISTICS
-INCLUDES :=
-LIBS := -lpthread
+INCLUDES := $(shell pkg-config --cflags capstone)
+LIBS := $(shell pkg-config --libs capstone) -lpthread
 PROJECT := capture
 IMAGE := $(PROJECT)-$(BUILD)
 
@@ -14,6 +14,11 @@ BASE2 := $(shell egrep 'ClCompile.*c"' ../win32/$(PROJECT).vcxproj | sed -e 's/.
 SRC := $(filter-out $(FILTER),$(BASE))
 SRC2 := $(filter-out $(FILTER),$(BASE2))
 
+TBB := $(shell ld -ltbb -o /dev/null 2>/dev/null; echo $$?)
+ifeq ($(TBB),0)
+	LIBS += -ltbb
+endif
+
 OBJDIRBASE := obj/$(BUILD)
 OBJDIR := $(OBJDIRBASE)/o/o/o
 
@@ -46,7 +51,7 @@ $(IMAGE): $(OBJ) $(OBJ2)
 	$(CXX) $(CXXFLAGS) $(DEFINES) $(OBJ) $(OBJ2) $(LIBS) -o $@
 
 ifneq "$(MAKECMDGOALS)" "clean"
--include $(addprefix $(OBJDIR)/,$(SRC:.cpp=.d)) %(addprefix $(OBJDIR)/,$(SRC2:.c=.d))
+-include $(addprefix $(OBJDIR)/,$(SRC:.cpp=.d)) $(addprefix $(OBJDIR)/,$(SRC2:.c=.d))
 endif
 
 clean:

capture/build/unix/release.mk

@@ -1,11 +1,7 @@
 ARCH := $(shell uname -m)
 
-CFLAGS := -O3 -s -fomit-frame-pointer
+CFLAGS := -O3 -s -march=native
 DEFINES := -DNDEBUG
 BUILD := release
 
-ifeq ($(ARCH),x86_64)
-CFLAGS += -msse4.1
-endif
-
 include build.mk

capture/build/win32/capture.vcxproj

@@ -22,32 +22,33 @@
     <VCProjectVersion>15.0</VCProjectVersion>
     <ProjectGuid>{447D58BF-94CD-4469-BB90-549C05D03E00}</ProjectGuid>
     <RootNamespace>capture</RootNamespace>
-    <WindowsTargetPlatformVersion>10.0.16299.0</WindowsTargetPlatformVersion>
+    <WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
+    <VcpkgTriplet>x64-windows-static</VcpkgTriplet>
   </PropertyGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
     <ConfigurationType>Application</ConfigurationType>
     <UseDebugLibraries>true</UseDebugLibraries>
-    <PlatformToolset>v141</PlatformToolset>
+    <PlatformToolset>v142</PlatformToolset>
     <CharacterSet>MultiByte</CharacterSet>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
     <ConfigurationType>Application</ConfigurationType>
     <UseDebugLibraries>false</UseDebugLibraries>
-    <PlatformToolset>v141</PlatformToolset>
+    <PlatformToolset>v142</PlatformToolset>
     <WholeProgramOptimization>true</WholeProgramOptimization>
     <CharacterSet>MultiByte</CharacterSet>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
     <ConfigurationType>Application</ConfigurationType>
     <UseDebugLibraries>true</UseDebugLibraries>
-    <PlatformToolset>v141</PlatformToolset>
+    <PlatformToolset>v142</PlatformToolset>
     <CharacterSet>MultiByte</CharacterSet>
   </PropertyGroup>
   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
     <ConfigurationType>Application</ConfigurationType>
     <UseDebugLibraries>false</UseDebugLibraries>
-    <PlatformToolset>v141</PlatformToolset>
+    <PlatformToolset>v142</PlatformToolset>
     <WholeProgramOptimization>true</WholeProgramOptimization>
     <CharacterSet>MultiByte</CharacterSet>
   </PropertyGroup>
@@ -88,10 +89,12 @@
       <PreprocessorDefinitions>TRACY_NO_STATISTICS;_CRT_SECURE_NO_DEPRECATE;_CRT_NONSTDC_NO_DEPRECATE;WIN32_LEAN_AND_MEAN;NOMINMAX;_USE_MATH_DEFINES;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <EnableEnhancedInstructionSet>AdvancedVectorExtensions2</EnableEnhancedInstructionSet>
       <LanguageStandard>stdcpplatest</LanguageStandard>
+      <AdditionalIncludeDirectories>..\..\..\vcpkg\vcpkg\installed\x64-windows-static\include</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
-      <AdditionalDependencies>ws2_32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>ws2_32.lib;capstone.lib;%(AdditionalDependencies)</AdditionalDependencies>
       <SubSystem>Console</SubSystem>
+      <AdditionalLibraryDirectories>..\..\..\vcpkg\vcpkg\installed\x64-windows-static\debug\lib</AdditionalLibraryDirectories>
     </Link>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
@@ -120,12 +123,14 @@
       <PreprocessorDefinitions>TRACY_NO_STATISTICS;NDEBUG;_CRT_SECURE_NO_DEPRECATE;_CRT_NONSTDC_NO_DEPRECATE;WIN32_LEAN_AND_MEAN;NOMINMAX;_USE_MATH_DEFINES;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <EnableEnhancedInstructionSet>AdvancedVectorExtensions2</EnableEnhancedInstructionSet>
       <LanguageStandard>stdcpplatest</LanguageStandard>
+      <AdditionalIncludeDirectories>..\..\..\vcpkg\vcpkg\installed\x64-windows-static\include</AdditionalIncludeDirectories>
     </ClCompile>
     <Link>
       <EnableCOMDATFolding>true</EnableCOMDATFolding>
       <OptimizeReferences>true</OptimizeReferences>
-      <AdditionalDependencies>ws2_32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+      <AdditionalDependencies>ws2_32.lib;capstone.lib;%(AdditionalDependencies)</AdditionalDependencies>
       <SubSystem>Console</SubSystem>
+      <AdditionalLibraryDirectories>..\..\..\vcpkg\vcpkg\installed\x64-windows-static\lib</AdditionalLibraryDirectories>
     </Link>
   </ItemDefinitionGroup>
   <ItemGroup>
@@ -134,7 +139,37 @@
     <ClCompile Include="..\..\..\common\tracy_lz4.cpp" />
     <ClCompile Include="..\..\..\common\tracy_lz4hc.cpp" />
     <ClCompile Include="..\..\..\server\TracyMemory.cpp" />
+    <ClCompile Include="..\..\..\server\TracyMmap.cpp" />
+    <ClCompile Include="..\..\..\server\TracyPrint.cpp" />
+    <ClCompile Include="..\..\..\server\TracyTaskDispatch.cpp" />
+    <ClCompile Include="..\..\..\server\TracyTextureCompression.cpp" />
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp" />
     <ClCompile Include="..\..\..\server\TracyWorker.cpp" />
+    <ClCompile Include="..\..\..\zstd\debug.c" />
+    <ClCompile Include="..\..\..\zstd\entropy_common.c" />
+    <ClCompile Include="..\..\..\zstd\error_private.c" />
+    <ClCompile Include="..\..\..\zstd\fse_compress.c" />
+    <ClCompile Include="..\..\..\zstd\fse_decompress.c" />
+    <ClCompile Include="..\..\..\zstd\hist.c" />
+    <ClCompile Include="..\..\..\zstd\huf_compress.c" />
+    <ClCompile Include="..\..\..\zstd\huf_decompress.c" />
+    <ClCompile Include="..\..\..\zstd\pool.c" />
+    <ClCompile Include="..\..\..\zstd\threading.c" />
+    <ClCompile Include="..\..\..\zstd\xxhash.c" />
+    <ClCompile Include="..\..\..\zstd\zstdmt_compress.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_common.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_compress.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_compress_literals.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_compress_sequences.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_compress_superblock.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_ddict.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_decompress.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_decompress_block.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_double_fast.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_fast.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_lazy.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_ldm.c" />
+    <ClCompile Include="..\..\..\zstd\zstd_opt.c" />
     <ClCompile Include="..\..\src\capture.cpp" />
     <ClCompile Include="..\..\src\getopt.c" />
   </ItemGroup>
@@ -147,19 +182,51 @@
     <ClInclude Include="..\..\..\common\TracyQueue.hpp" />
     <ClInclude Include="..\..\..\common\TracySocket.hpp" />
     <ClInclude Include="..\..\..\common\TracySystem.hpp" />
-    <ClInclude Include="..\..\..\common\tracy_benaphore.h" />
     <ClInclude Include="..\..\..\common\tracy_lz4.hpp" />
     <ClInclude Include="..\..\..\common\tracy_lz4hc.hpp" />
-    <ClInclude Include="..\..\..\common\tracy_sema.h" />
     <ClInclude Include="..\..\..\server\TracyCharUtil.hpp" />
     <ClInclude Include="..\..\..\server\TracyEvent.hpp" />
+    <ClInclude Include="..\..\..\server\TracyFileRead.hpp" />
     <ClInclude Include="..\..\..\server\TracyFileWrite.hpp" />
     <ClInclude Include="..\..\..\server\TracyMemory.hpp" />
+    <ClInclude Include="..\..\..\server\TracyMmap.hpp" />
     <ClInclude Include="..\..\..\server\TracyPopcnt.hpp" />
+    <ClInclude Include="..\..\..\server\TracyPrint.hpp" />
     <ClInclude Include="..\..\..\server\TracySlab.hpp" />
+    <ClInclude Include="..\..\..\server\TracyTaskDispatch.hpp" />
+    <ClInclude Include="..\..\..\server\TracyTextureCompression.hpp" />
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp" />
     <ClInclude Include="..\..\..\server\TracyVector.hpp" />
     <ClInclude Include="..\..\..\server\TracyWorker.hpp" />
-    <ClInclude Include="..\..\..\server\tracy_flat_hash_map.hpp" />
+    <ClInclude Include="..\..\..\zstd\bitstream.h" />
+    <ClInclude Include="..\..\..\zstd\compiler.h" />
+    <ClInclude Include="..\..\..\zstd\cpu.h" />
+    <ClInclude Include="..\..\..\zstd\debug.h" />
+    <ClInclude Include="..\..\..\zstd\error_private.h" />
+    <ClInclude Include="..\..\..\zstd\fse.h" />
+    <ClInclude Include="..\..\..\zstd\hist.h" />
+    <ClInclude Include="..\..\..\zstd\huf.h" />
+    <ClInclude Include="..\..\..\zstd\mem.h" />
+    <ClInclude Include="..\..\..\zstd\pool.h" />
+    <ClInclude Include="..\..\..\zstd\threading.h" />
+    <ClInclude Include="..\..\..\zstd\xxhash.h" />
+    <ClInclude Include="..\..\..\zstd\zstd.h" />
+    <ClInclude Include="..\..\..\zstd\zstdmt_compress.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_compress_internal.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_compress_literals.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_compress_sequences.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_compress_superblock.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_cwksp.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_ddict.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_decompress_block.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_decompress_internal.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_double_fast.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_errors.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_fast.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_internal.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_lazy.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_ldm.h" />
+    <ClInclude Include="..\..\..\zstd\zstd_opt.h" />
     <ClInclude Include="..\..\src\getopt.h" />
   </ItemGroup>
   <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />

capture/build/win32/capture.vcxproj.filters

@@ -10,6 +10,9 @@
     <Filter Include="common">
       <UniqueIdentifier>{e39d3623-47cd-4752-8da9-3ea324f964c1}</UniqueIdentifier>
     </Filter>
+    <Filter Include="zstd">
+      <UniqueIdentifier>{043ecb94-f240-4986-94b0-bc5bbd415a82}</UniqueIdentifier>
+    </Filter>
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="..\..\..\common\tracy_lz4.cpp">
@@ -36,6 +39,96 @@
     <ClCompile Include="..\..\..\common\tracy_lz4hc.cpp">
       <Filter>common</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyPrint.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyThreadCompress.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyTaskDispatch.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\debug.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\entropy_common.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\error_private.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\fse_compress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\fse_decompress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\hist.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\huf_compress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\huf_decompress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\pool.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\threading.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\xxhash.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_common.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_compress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_compress_literals.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_compress_sequences.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_compress_superblock.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_ddict.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_decompress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_decompress_block.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_double_fast.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_fast.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_lazy.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_ldm.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstd_opt.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\zstd\zstdmt_compress.c">
+      <Filter>zstd</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyMmap.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\server\TracyTextureCompression.cpp">
+      <Filter>server</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\..\common\tracy_lz4.hpp">
@@ -62,9 +155,6 @@
     <ClInclude Include="..\..\..\common\TracySystem.hpp">
       <Filter>common</Filter>
     </ClInclude>
-    <ClInclude Include="..\..\..\server\tracy_flat_hash_map.hpp">
-      <Filter>server</Filter>
-    </ClInclude>
     <ClInclude Include="..\..\..\server\TracyCharUtil.hpp">
       <Filter>server</Filter>
     </ClInclude>
@@ -95,14 +185,113 @@
     <ClInclude Include="..\..\..\common\TracyAlign.hpp">
       <Filter>common</Filter>
     </ClInclude>
-    <ClInclude Include="..\..\..\common\tracy_benaphore.h">
-      <Filter>common</Filter>
-    </ClInclude>
-    <ClInclude Include="..\..\..\common\tracy_sema.h">
-      <Filter>common</Filter>
-    </ClInclude>
     <ClInclude Include="..\..\..\common\tracy_lz4hc.hpp">
       <Filter>common</Filter>
     </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyPrint.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyThreadCompress.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyTaskDispatch.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\bitstream.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\compiler.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\cpu.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\debug.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\error_private.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\fse.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\hist.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\huf.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\mem.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\pool.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\threading.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\xxhash.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_compress_internal.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_compress_literals.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_compress_sequences.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_compress_superblock.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_cwksp.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_ddict.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_decompress_block.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_decompress_internal.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_double_fast.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_errors.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_fast.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_internal.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_lazy.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_ldm.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstd_opt.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\zstd\zstdmt_compress.h">
+      <Filter>zstd</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyFileRead.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyMmap.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
+    <ClInclude Include="..\..\..\server\TracyTextureCompression.hpp">
+      <Filter>server</Filter>
+    </ClInclude>
   </ItemGroup>
 </Project>

capture/src/capture.cpp

@@ -5,6 +5,7 @@
 #include <chrono>
 #include <inttypes.h>
 #include <mutex>
+#include <signal.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -12,94 +13,21 @@
 #include "../../common/TracyProtocol.hpp"
 #include "../../server/TracyFileWrite.hpp"
 #include "../../server/TracyMemory.hpp"
+#include "../../server/TracyPrint.hpp"
 #include "../../server/TracyWorker.hpp"
 #include "getopt.h"
 
-static const char* TimeToString( int64_t ns )
+
+bool disconnect = false;
+
+void SigInt( int )
 {
-    enum { Pool = 8 };
-    static char bufpool[Pool][64];
-    static int bufsel = 0;
-    char* buf = bufpool[bufsel];
-    bufsel = ( bufsel + 1 ) % Pool;
-
-    const char* sign = "";
-    if( ns < 0 )
-    {
-        sign = "-";
-        ns = -ns;
-    }
-
-    if( ns < 1000 )
-    {
-        sprintf( buf, "%s%" PRIi64 " ns", sign, ns );
-    }
-    else if( ns < 1000ll * 1000 )
-    {
-        sprintf( buf, "%s%.2f us", sign, ns / 1000. );
-    }
-    else if( ns < 1000ll * 1000 * 1000 )
-    {
-        sprintf( buf, "%s%.2f ms", sign, ns / ( 1000. * 1000. ) );
-    }
-    else if( ns < 1000ll * 1000 * 1000 * 60 )
-    {
-        sprintf( buf, "%s%.2f s", sign, ns / ( 1000. * 1000. * 1000. ) );
-    }
-    else
-    {
-        const auto m = int64_t( ns / ( 1000ll * 1000 * 1000 * 60 ) );
-        const auto s = int64_t( ns - m * ( 1000ll * 1000 * 1000 * 60 ) );
-        sprintf( buf, "%s%" PRIi64 ":%04.1f", sign, m, s / ( 1000. * 1000. * 1000. ) );
-    }
-    return buf;
+    disconnect = true;
 }
 
-static const char* RealToString( double val, bool separator )
-{
-    enum { Pool = 8 };
-    static char bufpool[Pool][64];
-    static int bufsel = 0;
-    char* buf = bufpool[bufsel];
-    bufsel = ( bufsel + 1 ) % Pool;
-
-    sprintf( buf, "%f", val );
-    auto ptr = buf;
-    if( *ptr == '-' ) ptr++;
-
-    const auto vbegin = ptr;
-
-    if( separator )
-    {
-        while( *ptr != '\0' && *ptr != ',' && *ptr != '.' ) ptr++;
-        auto end = ptr;
-        while( *end != '\0' ) end++;
-        auto sz = end - ptr;
-
-        while( ptr - vbegin > 3 )
-        {
-            ptr -= 3;
-            memmove( ptr+1, ptr, sz );
-            *ptr = ',';
-            sz += 4;
-        }
-    }
-
-    while( *ptr != '\0' && *ptr != ',' && *ptr != '.' ) ptr++;
-
-    if( *ptr == '\0' ) return buf;
-    while( *ptr != '\0' ) ptr++;
-    ptr--;
-    while( *ptr == '0' && *ptr != ',' && *ptr != '.' ) ptr--;
-    if( *ptr != '.' && *ptr != ',' ) ptr++;
-    *ptr = '\0';
-    return buf;
-}
-
-
 void Usage()
 {
-    printf( "Usage: capture -a address -o output.tracy\n" );
+    printf( "Usage: capture -o output.tracy [-a address] [-p port]\n" );
     exit( 1 );
 }
 
@@ -113,11 +41,12 @@ int main( int argc, char** argv )
     }
 #endif
 
-    const char* address = nullptr;
+    const char* address = "localhost";
     const char* output = nullptr;
+    int port = 8086;
 
     int c;
-    while( ( c = getopt( argc, argv, "a:o:" ) ) != -1 )
+    while( ( c = getopt( argc, argv, "a:o:p:" ) ) != -1 )
     {
         switch( c )
         {
@@ -127,6 +56,9 @@ int main( int argc, char** argv )
         case 'o':
             output = optarg;
             break;
+        case 'p':
+            port = atoi( optarg );
+            break;
         default:
             Usage();
             break;
@@ -135,13 +67,12 @@ int main( int argc, char** argv )
 
     if( !address || !output ) Usage();
 
-    printf( "Connecting to %s...", address );
+    printf( "Connecting to %s:%i...", address, port );
     fflush( stdout );
-    tracy::Worker worker( address );
-    for(;;)
+    tracy::Worker worker( address, port );
+    while( !worker.IsConnected() )
     {
         const auto handshake = worker.GetHandshakeStatus();
-        if( handshake == tracy::HandshakeWelcome ) break;
         if( handshake == tracy::HandshakeProtocolMismatch )
         {
             printf( "\nThe client you are trying to connect to uses incompatible protocol version.\nMake sure you are using the same Tracy version on both client and server.\n" );
@@ -152,17 +83,39 @@ int main( int argc, char** argv )
             printf( "\nThe client you are trying to connect to is no longer able to sent profiling data,\nbecause another server was already connected to it.\nYou can do the following:\n\n  1. Restart the client application.\n  2. Rebuild the client application with on-demand mode enabled.\n" );
             return 2;
         }
+        if( handshake == tracy::HandshakeDropped )
+        {
+            printf( "\nThe client you are trying to connect to has disconnected during the initial\nconnection handshake. Please check your network configuration.\n" );
+            return 3;
+        }
     }
     while( !worker.HasData() ) std::this_thread::sleep_for( std::chrono::milliseconds( 100 ) );
-    printf( "\nQueue delay: %s\nTimer resolution: %s\n", TimeToString( worker.GetDelay() ), TimeToString( worker.GetResolution() ) );
+    printf( "\nQueue delay: %s\nTimer resolution: %s\n", tracy::TimeToString( worker.GetDelay() ), tracy::TimeToString( worker.GetResolution() ) );
+
+#ifdef _WIN32
+    signal( SIGINT, SigInt );
+#else
+    struct sigaction sigint, oldsigint;
+    memset( &sigint, 0, sizeof( sigint ) );
+    sigint.sa_handler = SigInt;
+    sigaction( SIGINT, &sigint, &oldsigint );
+#endif
 
     auto& lock = worker.GetMbpsDataLock();
 
+    const auto t0 = std::chrono::high_resolution_clock::now();
     while( worker.IsConnected() )
     {
+        if( disconnect )
+        {
+            worker.Disconnect();
+            disconnect = false;
+        }
+
         lock.lock();
         const auto mbps = worker.GetMbpsData().back();
         const auto compRatio = worker.GetCompRatio();
+        const auto netTotal = worker.GetDataTransferred();
         lock.unlock();
 
         if( mbps < 0.1f )
@@ -173,19 +126,36 @@ int main( int argc, char** argv )
         {
             printf( "\33[2K\r\033[36;1m%7.2f Mbps", mbps );
         }
-        printf( " \033[0m /\033[36;1m%5.1f%% \033[0m=\033[33;1m%7.2f Mbps \033[0m| Mem: \033[31;1m%.2f MB\033[0m | \033[33mTime: %s\033[0m", compRatio * 100.f, mbps / compRatio, tracy::memUsage.load( std::memory_order_relaxed ) / ( 1024.f * 1024.f ), TimeToString( worker.GetLastTime() - worker.GetTimeBegin() ) );
+        printf( " \033[0m /\033[36;1m%5.1f%% \033[0m=\033[33;1m%7.2f Mbps \033[0m| \033[33mNet: \033[32m%s \033[0m| \033[33mMem: \033[31;1m%s\033[0m | \033[33mTime: %s\033[0m",
+            compRatio * 100.f,
+            mbps / compRatio,
+            tracy::MemSizeToString( netTotal ),
+            tracy::MemSizeToString( tracy::memUsage ),
+            tracy::TimeToString( worker.GetLastTime() ) );
         fflush( stdout );
 
         std::this_thread::sleep_for( std::chrono::milliseconds( 100 ) );
     }
+    const auto t1 = std::chrono::high_resolution_clock::now();
 
-    printf( "\nFrames: %" PRIu64 "\nTime span: %s\nZones: %s\nSaving trace...", worker.GetFrameCount( *worker.GetFramesBase() ), TimeToString( worker.GetLastTime() - worker.GetTimeBegin() ), RealToString( worker.GetZoneCount(), true ) );
+    const auto& failure = worker.GetFailureType();
+    if( failure != tracy::Worker::Failure::None )
+    {
+        printf( "\n\033[31;1mInstrumentation failure: %s\033[0m", tracy::Worker::GetFailureString( failure ) );
+    }
+
+    printf( "\nFrames: %" PRIu64 "\nTime span: %s\nZones: %s\nElapsed time: %s\nSaving trace...",
+        worker.GetFrameCount( *worker.GetFramesBase() ), tracy::TimeToString( worker.GetLastTime() ), tracy::RealToString( worker.GetZoneCount() ),
+        tracy::TimeToString( std::chrono::duration_cast<std::chrono::nanoseconds>( t1 - t0 ).count() ) );
     fflush( stdout );
     auto f = std::unique_ptr<tracy::FileWrite>( tracy::FileWrite::Open( output ) );
     if( f )
     {
         worker.Write( *f );
         printf( " \033[32;1mdone!\033[0m\n" );
+        f->Finish();
+        const auto stats = f->GetCompressionStatistics();
+        printf( "Trace size %s (%.2f%% ratio)\n", tracy::MemSizeToString( stats.second ), 100.f * stats.second / stats.first );
     }
     else
     {

client/TracyArmCpuTable.hpp

@@ -0,0 +1,328 @@
+#ifdef _MSC_VER
+#  pragma warning(disable:4996)
+#endif
+
+namespace tracy
+{
+
+static const char* DecodeArmImplementer( uint32_t v )
+{
+    static char buf[16];
+    switch( v )
+    {
+    case 0x41: return "ARM";
+    case 0x42: return "Broadcom";
+    case 0x43: return "Cavium";
+    case 0x44: return "DEC";
+    case 0x46: return "Fujitsu";
+    case 0x48: return "HiSilicon";
+    case 0x4d: return "Motorola";
+    case 0x4e: return "Nvidia";
+    case 0x50: return "Applied Micro";
+    case 0x51: return "Qualcomm";
+    case 0x53: return "Samsung";
+    case 0x54: return "Texas Instruments";
+    case 0x56: return "Marvell";
+    case 0x61: return "Apple";
+    case 0x66: return "Faraday";
+    case 0x68: return "HXT";
+    case 0x69: return "Intel";
+    default: break;
+    }
+    sprintf( buf, "0x%x", v );
+    return buf;
+}
+
+static const char* DecodeArmPart( uint32_t impl, uint32_t part )
+{
+    static char buf[16];
+    switch( impl )
+    {
+    case 0x41:
+        switch( part )
+        {
+        case 0x810: return "810";
+        case 0x920: return "920";
+        case 0x922: return "922";
+        case 0x926: return "926";
+        case 0x940: return "940";
+        case 0x946: return "946";
+        case 0x966: return "966";
+        case 0xa20: return "1020";
+        case 0xa22: return "1022";
+        case 0xa26: return "1026";
+        case 0xb02: return "11 MPCore";
+        case 0xb36: return "1136";
+        case 0xb56: return "1156";
+        case 0xb76: return "1176";
+        case 0xc05: return " Cortex-A5";
+        case 0xc07: return " Cortex-A7";
+        case 0xc08: return " Cortex-A8";
+        case 0xc09: return " Cortex-A9";
+        case 0xc0c: return " Cortex-A12";
+        case 0xc0d: return " Rockchip RK3288";
+        case 0xc0f: return " Cortex-A15";
+        case 0xc0e: return " Cortex-A17";
+        case 0xc14: return " Cortex-R4";
+        case 0xc15: return " Cortex-R5";
+        case 0xc17: return " Cortex-R7";
+        case 0xc18: return " Cortex-R8";
+        case 0xc20: return " Cortex-M0";
+        case 0xc21: return " Cortex-M1";
+        case 0xc23: return " Cortex-M3";
+        case 0xc24: return " Cortex-M4";
+        case 0xc27: return " Cortex-M7";
+        case 0xc60: return " Cortex-M0+";
+        case 0xd00: return " AArch64 simulator";
+        case 0xd01: return " Cortex-A32";
+        case 0xd03: return " Cortex-A53";
+        case 0xd04: return " Cortex-A35";
+        case 0xd05: return " Cortex-A55";
+        case 0xd06: return " Cortex-A65";
+        case 0xd07: return " Cortex-A57";
+        case 0xd08: return " Cortex-A72";
+        case 0xd09: return " Cortex-A73";
+        case 0xd0a: return " Cortex-A75";
+        case 0xd0b: return " Cortex-A76";
+        case 0xd0c: return " Neoverse N1";
+        case 0xd0d: return " Cortex-A77";
+        case 0xd0e: return " Cortex-A76AE";
+        case 0xd0f: return " AEMv8";
+        case 0xd13: return " Cortex-R52";
+        case 0xd20: return " Cortex-M23";
+        case 0xd21: return " Cortex-M33";
+        case 0xd4a: return " Neoverse E1";
+        default: break;
+        }
+    case 0x42:
+        switch( part )
+        {
+        case 0xf: return " Brahma B15";
+        case 0x100: return " Brahma B53";
+        case 0x516: return " ThunderX2";
+        default: break;
+        }
+    case 0x43:
+        switch( part )
+        {
+        case 0xa0: return " ThunderX";
+        case 0xa1: return " ThunderX 88XX";
+        case 0xa2: return " ThunderX 81XX";
+        case 0xa3: return " ThunderX 83XX";
+        case 0xaf: return " ThunderX2 99xx";
+        default: break;
+        }
+    case 0x44:
+        switch( part )
+        {
+        case 0xa10: return " SA110";
+        case 0xa11: return " SA1100";
+        default: break;
+        }
+    case 0x46:
+        switch( part )
+        {
+        case 0x1: return " A64FX";
+        default: break;
+        }
+    case 0x48:
+        switch( part )
+        {
+        case 0xd01: return " TSV100";
+        case 0xd40: return " Kirin 980";
+        default: break;
+        }
+    case 0x4e:
+        switch( part )
+        {
+        case 0x0: return " Denver";
+        case 0x3: return " Denver 2";
+        case 0x4: return " Carmel";
+        default: break;
+        }
+    case 0x50:
+        switch( part )
+        {
+        case 0x0: return " X-Gene";
+        default: break;
+        }
+    case 0x51:
+        switch( part )
+        {
+        case 0xf: return " Scorpion";
+        case 0x2d: return " Scorpion";
+        case 0x4d: return " Krait";
+        case 0x6f: return " Krait";
+        case 0x200: return " Kryo";
+        case 0x201: return " Kryo Silver (Snapdragon 821)";
+        case 0x205: return " Kryo Gold";
+        case 0x211: return " Kryo Silver (Snapdragon 820)";
+        case 0x800: return " Kryo 260 / 280 Gold";
+        case 0x801: return " Kryo 260 / 280 Silver";
+        case 0x802: return " Kryo 385 Gold";
+        case 0x803: return " Kryo 385 Silver";
+        case 0x804: return " Kryo 485 Gold";
+        case 0xc00: return " Falkor";
+        case 0xc01: return " Saphira";
+        default: break;
+        }
+    case 0x53:
+        switch( part )
+        {
+        case 0x1: return " Exynos M1/M2";
+        case 0x2: return " Exynos M3";
+        default: break;
+        }
+    case 0x56:
+        switch( part )
+        {
+        case 0x131: return " Feroceon 88FR131";
+        case 0x581: return " PJ4 / PJ4B";
+        case 0x584: return " PJ4B-MP / PJ4C";
+        default: break;
+        }
+    case 0x61:
+        switch( part )
+        {
+        case 0x1: return " Cyclone";
+        case 0x2: return " Typhoon";
+        case 0x3: return " Typhoon/Capri";
+        case 0x4: return " Twister";
+        case 0x5: return " Twister/Elba/Malta";
+        case 0x6: return " Hurricane";
+        case 0x7: return " Hurricane/Myst";
+        default: break;
+        }
+    case 0x66:
+        switch( part )
+        {
+        case 0x526: return " FA526";
+        case 0x626: return " FA626";
+        default: break;
+        }
+    case 0x68:
+        switch( part )
+        {
+        case 0x0: return " Phecda";
+        default: break;
+        }
+    default: break;
+    }
+    sprintf( buf, " 0x%x", part );
+    return buf;
+}
+
+static const char* DecodeIosDevice( const char* id )
+{
+    static const char* DeviceTable[] = {
+        "i386", "32-bit simulator",
+        "x86_64", "64-bit simulator",
+        "iPhone1,1", "iPhone",
+        "iPhone1,2", "iPhone 3G",
+        "iPhone2,1", "iPhone 3GS",
+        "iPhone3,1", "iPhone 4 (GSM)",
+        "iPhone3,2", "iPhone 4 (GSM)",
+        "iPhone3,3", "iPhone 4 (CDMA)",
+        "iPhone4,1", "iPhone 4S",
+        "iPhone5,1", "iPhone 5 (A1428)",
+        "iPhone5,2", "iPhone 5 (A1429)",
+        "iPhone5,3", "iPhone 5c (A1456/A1532)",
+        "iPhone5,4", "iPhone 5c (A1507/A1516/1526/A1529)",
+        "iPhone6,1", "iPhone 5s (A1433/A1533)",
+        "iPhone6,2", "iPhone 5s (A1457/A1518/A1528/A1530)",
+        "iPhone7,1", "iPhone 6 Plus",
+        "iPhone7,2", "iPhone 6",
+        "iPhone8,1", "iPhone 6S",
+        "iPhone8,2", "iPhone 6S Plus",
+        "iPhone8,4", "iPhone SE",
+        "iPhone9,1", "iPhone 7 (CDMA)",
+        "iPhone9,2", "iPhone 7 Plus (CDMA)",
+        "iPhone9,3", "iPhone 7 (GSM)",
+        "iPhone9,4", "iPhone 7 Plus (GSM)",
+        "iPhone10,1", "iPhone 8 (CDMA)",
+        "iPhone10,2", "iPhone 8 Plus (CDMA)",
+        "iPhone10,3", "iPhone X (CDMA)",
+        "iPhone10,4", "iPhone 8 (GSM)",
+        "iPhone10,5", "iPhone 8 Plus (GSM)",
+        "iPhone10,6", "iPhone X (GSM)",
+        "iPhone11,2", "iPhone XS",
+        "iPhone11,4", "iPhone XS Max",
+        "iPhone11,6", "iPhone XS Max China",
+        "iPhone11,8", "iPhone XR",
+        "iPhone12,1", "iPhone 11",
+        "iPhone12,3", "iPhone 11 Pro",
+        "iPhone12,5", "iPhone 11 Pro Max",
+        "iPad1,1", "iPad (A1219/A1337)",
+        "iPad2,1", "iPad 2 (A1395)",
+        "iPad2,2", "iPad 2 (A1396)",
+        "iPad2,3", "iPad 2 (A1397)",
+        "iPad2,4", "iPad 2 (A1395)",
+        "iPad2,5", "iPad Mini (A1432)",
+        "iPad2,6", "iPad Mini (A1454)",
+        "iPad2,7", "iPad Mini (A1455)",
+        "iPad3,1", "iPad 3 (A1416)",
+        "iPad3,2", "iPad 3 (A1403)",
+        "iPad3,3", "iPad 3 (A1430)",
+        "iPad3,4", "iPad 4 (A1458)",
+        "iPad3,5", "iPad 4 (A1459)",
+        "iPad3,6", "iPad 4 (A1460)",
+        "iPad4,1", "iPad Air (A1474)",
+        "iPad4,2", "iPad Air (A1475)",
+        "iPad4,3", "iPad Air (A1476)",
+        "iPad4,4", "iPad Mini 2 (A1489)",
+        "iPad4,5", "iPad Mini 2 (A1490)",
+        "iPad4,6", "iPad Mini 2 (A1491)",
+        "iPad4,7", "iPad Mini 3 (A1599)",
+        "iPad4,8", "iPad Mini 3 (A1600)",
+        "iPad4,9", "iPad Mini 3 (A1601)",
+        "iPad5,1", "iPad Mini 4 (A1538)",
+        "iPad5,2", "iPad Mini 4 (A1550)",
+        "iPad5,3", "iPad Air 2 (A1566)",
+        "iPad5,4", "iPad Air 2 (A1567)",
+        "iPad6,3", "iPad Pro 9.7\" (A1673)",
+        "iPad6,4", "iPad Pro 9.7\" (A1674)",
+        "iPad6,5", "iPad Pro 9.7\" (A1675)",
+        "iPad6,7", "iPad Pro 12.9\" (A1584)",
+        "iPad6,8", "iPad Pro 12.9\" (A1652)",
+        "iPad6,11", "iPad 5th gen (A1822)",
+        "iPad6,12", "iPad 5th gen (A1823)",
+        "iPad7,1", "iPad Pro 12.9\" 2nd gen (A1670)",
+        "iPad7,2", "iPad Pro 12.9\" 2nd gen (A1671/A1821)",
+        "iPad7,3", "iPad Pro 10.5\" (A1701)",
+        "iPad7,4", "iPad Pro 10.5\" (A1709)",
+        "iPad7,5", "iPad 6th gen (A1893)",
+        "iPad7,6", "iPad 6th gen (A1954)",
+        "iPad7,11", "iPad 7th gen 10.2\" (Wifi)",
+        "iPad7,12", "iPad 7th gen 10.2\" (Wifi+Cellular)",
+        "iPad8,1", "iPad Pro 11\" (A1980)",
+        "iPad8,2", "iPad Pro 11\" (A1980)",
+        "iPad8,3", "iPad Pro 11\" (A1934/A1979/A2013)",
+        "iPad8,4", "iPad Pro 11\" (A1934/A1979/A2013)",
+        "iPad8,5", "iPad Pro 12.9\" 3rd gen (A1876)",
+        "iPad8,6", "iPad Pro 12.9\" 3rd gen (A1876)",
+        "iPad8,7", "iPad Pro 12.9\" 3rd gen (A1895/A1983/A2014)",
+        "iPad8,8", "iPad Pro 12.9\" 3rd gen (A1895/A1983/A2014)",
+        "iPad11,1", "iPad Mini 5th gen (A2133)",
+        "iPad11,2", "iPad Mini 5th gen (A2124/A2125/A2126)",
+        "iPad11,3", "iPad Air 3rd gen (A2152)",
+        "iPad11,4", "iPad Air 3rd gen (A2123/A2153/A2154)",
+        "iPod1,1", "iPod Touch",
+        "iPod2,1", "iPod Touch 2nd gen",
+        "iPod3,1", "iPod Touch 3rd gen",
+        "iPod4,1", "iPod Touch 4th gen",
+        "iPod5,1", "iPod Touch 5th gen",
+        "iPod7,1", "iPod Touch 6th gen",
+        "iPod9,1", "iPod Touch 7th gen",
+        nullptr
+    };
+
+    auto ptr = DeviceTable;
+    while( *ptr )
+    {
+        if( strcmp( ptr[0], id ) == 0 ) return ptr[1];
+        ptr += 2;
+    }
+    return id;
+}
+
+}

client/TracyCallstack.cpp

@@ -1,10 +1,18 @@
+#include <new>
 #include <stdio.h>
+#include <string.h>
 #include "TracyCallstack.hpp"
+#include "TracyFastVector.hpp"
+#include "../common/TracyAlloc.hpp"
 
 #ifdef TRACY_HAS_CALLSTACK
 
 #if TRACY_HAS_CALLSTACK == 1
+#  ifndef NOMINMAX
+#    define NOMINMAX
+#  endif
 #  include <windows.h>
+#  include <psapi.h>
 #  ifdef _MSC_VER
 #    pragma warning( push )
 #    pragma warning( disable : 4091 )
@@ -13,7 +21,11 @@
 #  ifdef _MSC_VER
 #    pragma warning( pop )
 #  endif
-#elif TRACY_HAS_CALLSTACK >= 2
+#elif TRACY_HAS_CALLSTACK == 2 || TRACY_HAS_CALLSTACK == 3 || TRACY_HAS_CALLSTACK == 4 || TRACY_HAS_CALLSTACK == 6
+#  include "../libbacktrace/backtrace.hpp"
+#  include <dlfcn.h>
+#  include <cxxabi.h>
+#elif TRACY_HAS_CALLSTACK == 5
 #  include <dlfcn.h>
 #  include <cxxabi.h>
 #endif
@@ -21,233 +33,705 @@
 namespace tracy
 {
 
+static inline char* CopyString( const char* src, size_t sz )
+{
+    assert( strlen( src ) == sz );
+    auto dst = (char*)tracy_malloc( sz + 1 );
+    memcpy( dst, src, sz );
+    dst[sz] = '\0';
+    return dst;
+}
+
+static inline char* CopyString( const char* src )
+{
+    const auto sz = strlen( src );
+    auto dst = (char*)tracy_malloc( sz + 1 );
+    memcpy( dst, src, sz );
+    dst[sz] = '\0';
+    return dst;
+}
+
+
 #if TRACY_HAS_CALLSTACK == 1
 
-extern "C" { t_RtlWalkFrameChain RtlWalkFrameChain = 0; }
+enum { MaxCbTrace = 16 };
+enum { MaxNameSize = 8*1024 };
+
+int cb_num;
+CallstackEntry cb_data[MaxCbTrace];
+
+extern "C"
+{
+    typedef unsigned long (__stdcall *t_RtlWalkFrameChain)( void**, unsigned long, unsigned long );
+    t_RtlWalkFrameChain RtlWalkFrameChain = 0;
+}
+
+#if defined __MINGW32__ && API_VERSION_NUMBER < 12
+extern "C" {
+// Actual required API_VERSION_NUMBER is unknown because it is undocumented. These functions are not present in at least v11.
+DWORD IMAGEAPI SymAddrIncludeInlineTrace(HANDLE hProcess, DWORD64 Address);
+BOOL IMAGEAPI SymQueryInlineTrace(HANDLE hProcess, DWORD64 StartAddress, DWORD StartContext, DWORD64 StartRetAddress,
+    DWORD64 CurAddress, LPDWORD CurContext, LPDWORD CurFrameIndex);
+BOOL IMAGEAPI SymFromInlineContext(HANDLE hProcess, DWORD64 Address, ULONG InlineContext, PDWORD64 Displacement,
+    PSYMBOL_INFO Symbol);
+BOOL IMAGEAPI SymGetLineFromInlineContext(HANDLE hProcess, DWORD64 qwAddr, ULONG InlineContext,
+    DWORD64 qwModuleBaseAddress, PDWORD pdwDisplacement, PIMAGEHLP_LINE64 Line64);
+};
+#endif
+
+#ifndef __CYGWIN__
+struct ModuleCache
+{
+    uint64_t start;
+    uint64_t end;
+    char* name;
+};
+
+static FastVector<ModuleCache>* s_modCache;
+#endif
 
 void InitCallstack()
 {
-#ifdef UNICODE
-    RtlWalkFrameChain = (t_RtlWalkFrameChain)GetProcAddress( GetModuleHandle( L"ntdll.dll" ), "RtlWalkFrameChain" );
-#else
-    RtlWalkFrameChain = (t_RtlWalkFrameChain)GetProcAddress( GetModuleHandle( "ntdll.dll" ), "RtlWalkFrameChain" );
-#endif
+    RtlWalkFrameChain = (t_RtlWalkFrameChain)GetProcAddress( GetModuleHandleA( "ntdll.dll" ), "RtlWalkFrameChain" );
+
     SymInitialize( GetCurrentProcess(), nullptr, true );
     SymSetOptions( SYMOPT_LOAD_LINES );
+
+#ifndef __CYGWIN__
+    HMODULE mod[1024];
+    DWORD needed;
+    HANDLE proc = GetCurrentProcess();
+
+    s_modCache = (FastVector<ModuleCache>*)tracy_malloc( sizeof( FastVector<ModuleCache> ) );
+    new(s_modCache) FastVector<ModuleCache>( 512 );
+
+    if( EnumProcessModules( proc, mod, sizeof( mod ), &needed ) != 0 )
+    {
+        const auto sz = needed / sizeof( HMODULE );
+        for( size_t i=0; i<sz; i++ )
+        {
+            MODULEINFO info;
+            if( GetModuleInformation( proc, mod[i], &info, sizeof( info ) ) != 0 )
+            {
+                const auto base = uint64_t( info.lpBaseOfDll );
+                char name[1024];
+                const auto res = GetModuleFileNameA( mod[i], name, 1021 );
+                if( res > 0 )
+                {
+                    auto ptr = name + res;
+                    while( ptr > name && *ptr != '\\' && *ptr != '/' ) ptr--;
+                    if( ptr > name ) ptr++;
+                    const auto namelen = name + res - ptr;
+                    auto cache = s_modCache->push_next();
+                    cache->start = base;
+                    cache->end = base + info.SizeOfImage;
+                    cache->name = (char*)tracy_malloc( namelen+3 );
+                    cache->name[0] = '[';
+                    memcpy( cache->name+1, ptr, namelen );
+                    cache->name[namelen+1] = ']';
+                    cache->name[namelen+2] = '\0';
+                }
+            }
+        }
+    }
+#endif
 }
 
-CallstackEntry DecodeCallstackPtr( uint64_t ptr )
+TRACY_API uintptr_t* CallTrace( int depth )
 {
-    CallstackEntry ret;
+    auto trace = (uintptr_t*)tracy_malloc( ( 1 + depth ) * sizeof( uintptr_t ) );
+    const auto num = RtlWalkFrameChain( (void**)( trace + 1 ), depth, 0 );
+    *trace = num;
+    return trace;
+}
 
+const char* DecodeCallstackPtrFast( uint64_t ptr )
+{
+    static char ret[MaxNameSize];
     const auto proc = GetCurrentProcess();
 
-    char buf[sizeof( SYMBOL_INFO ) + 1024];
+    char buf[sizeof( SYMBOL_INFO ) + MaxNameSize];
     auto si = (SYMBOL_INFO*)buf;
     si->SizeOfStruct = sizeof( SYMBOL_INFO );
-    si->MaxNameLen = 1024;
+    si->MaxNameLen = MaxNameSize;
 
     if( SymFromAddr( proc, ptr, nullptr, si ) == 0 )
     {
-        memcpy( si->Name, "[unknown]", 10 );
-        si->NameLen = 9;
+        *ret = '\0';
+    }
+    else
+    {
+        memcpy( ret, si->Name, si->NameLen );
+        ret[si->NameLen] = '\0';
+    }
+    return ret;
+}
+
+static const char* GetModuleName( uint64_t addr )
+{
+    if( ( addr & 0x8000000000000000 ) != 0 ) return "[kernel]";
+
+#ifndef __CYGWIN__
+    for( auto& v : *s_modCache )
+    {
+        if( addr >= v.start && addr < v.end )
+        {
+            return v.name;
+        }
     }
 
-    auto name = (char*)tracy_malloc( si->NameLen + 1 );
-    memcpy( name, si->Name, si->NameLen );
-    name[si->NameLen] = '\0';
+    HMODULE mod[1024];
+    DWORD needed;
+    HANDLE proc = GetCurrentProcess();
 
-    ret.name = name;
+    if( EnumProcessModules( proc, mod, sizeof( mod ), &needed ) != 0 )
+    {
+        const auto sz = needed / sizeof( HMODULE );
+        for( size_t i=0; i<sz; i++ )
+        {
+            MODULEINFO info;
+            if( GetModuleInformation( proc, mod[i], &info, sizeof( info ) ) != 0 )
+            {
+                const auto base = uint64_t( info.lpBaseOfDll );
+                if( addr >= base && addr < base + info.SizeOfImage )
+                {
+                    char name[1024];
+                    const auto res = GetModuleFileNameA( mod[i], name, 1021 );
+                    if( res > 0 )
+                    {
+                        auto ptr = name + res;
+                        while( ptr > name && *ptr != '\\' && *ptr != '/' ) ptr--;
+                        if( ptr > name ) ptr++;
+                        const auto namelen = name + res - ptr;
+                        auto cache = s_modCache->push_next();
+                        cache->start = base;
+                        cache->end = base + info.SizeOfImage;
+                        cache->name = (char*)tracy_malloc( namelen+3 );
+                        cache->name[0] = '[';
+                        memcpy( cache->name+1, ptr, namelen );
+                        cache->name[namelen+1] = ']';
+                        cache->name[namelen+2] = '\0';
+                        return cache->name;
+                    }
+                }
+            }
+        }
+    }
+#endif
 
-    const char* filename;
+    return "[unknown]";
+}
+
+SymbolData DecodeSymbolAddress( uint64_t ptr )
+{
+    SymbolData sym;
     IMAGEHLP_LINE64 line;
     DWORD displacement = 0;
-    line.SizeOfStruct = sizeof( IMAGEHLP_LINE64 );
-    if( SymGetLineFromAddr64( proc, ptr, &displacement, &line ) == 0 )
+    line.SizeOfStruct = sizeof(IMAGEHLP_LINE64);
+    if( SymGetLineFromAddr64( GetCurrentProcess(), ptr, &displacement, &line ) == 0 )
     {
-        filename = "[unknown]";
-        ret.line = 0;
+        sym.file = "[unknown]";
+        sym.line = 0;
     }
     else
     {
-        filename = line.FileName;
-        ret.line = line.LineNumber;
+        sym.file = line.FileName;
+        sym.line = line.LineNumber;
     }
-
-    const auto fsz = strlen( filename );
-    auto file = (char*)tracy_malloc( fsz + 1 );
-    memcpy( file, filename, fsz );
-    file[fsz] = '\0';
-
-    ret.file = file;
-
-    return ret;
+    sym.needFree = false;
+    return sym;
 }
 
-#elif TRACY_HAS_CALLSTACK == 2
-
-CallstackEntry DecodeCallstackPtr( uint64_t ptr )
+SymbolData DecodeCodeAddress( uint64_t ptr )
 {
-    CallstackEntry ret;
-    ret.line = 0;
+    SymbolData sym;
+    const auto proc = GetCurrentProcess();
+    bool done = false;
 
-    char* demangled = nullptr;
-    const char* symname = nullptr;
-    const char* symloc = nullptr;
-    auto vptr = (void*)ptr;
-    char** sym = nullptr;
-    ptrdiff_t symoff = 0;
+    IMAGEHLP_LINE64 line;
+    DWORD displacement = 0;
+    line.SizeOfStruct = sizeof(IMAGEHLP_LINE64);
 
-    Dl_info dlinfo;
-    if( dladdr( vptr, &dlinfo ) )
+#ifndef __CYGWIN__
+    DWORD inlineNum = SymAddrIncludeInlineTrace( proc, ptr );
+    DWORD ctx = 0;
+    DWORD idx;
+    BOOL doInline = FALSE;
+    if( inlineNum != 0 ) doInline = SymQueryInlineTrace( proc, ptr, 0, ptr, ptr, &ctx, &idx );
+    if( doInline )
     {
-        symloc = dlinfo.dli_fname;
-        symname = dlinfo.dli_sname;
-        symoff = (char*)ptr - (char*)dlinfo.dli_saddr;
-
-        if( symname && symname[0] == '_' )
+        if( SymGetLineFromInlineContext( proc, ptr, ctx, 0, &displacement, &line ) != 0 )
         {
-            size_t len = 0;
-            int status;
-            demangled = abi::__cxa_demangle( symname, nullptr, &len, &status );
-            if( status == 0 )
-            {
-                symname = demangled;
-            }
+            sym.file = line.FileName;
+            sym.line = line.LineNumber;
+            done = true;
         }
     }
-
-    if( !symname )
+#endif
+    if( !done )
     {
-        symname = "[unknown]";
-    }
-    if( !symloc )
-    {
-        symloc = "[unknown]";
-    }
-
-    if( symoff == 0 )
-    {
-        const auto namelen = strlen( symname );
-        auto name = (char*)tracy_malloc( namelen + 1 );
-        memcpy( name, symname, namelen );
-        name[namelen] = '\0';
-        ret.name = name;
-    }
-    else
-    {
-        char buf[32];
-        sprintf( buf, " + %td", symoff );
-        const auto offlen = strlen( buf );
-        const auto namelen = strlen( symname );
-        auto name = (char*)tracy_malloc( namelen + offlen + 1 );
-        memcpy( name, symname, namelen );
-        memcpy( name + namelen, buf, offlen );
-        name[namelen + offlen] = '\0';
-        ret.name = name;
-    }
-
-    char buf[32];
-    sprintf( buf, " [%p]", (void*)ptr );
-    const auto addrlen = strlen( buf );
-    const auto loclen = strlen( symloc );
-    auto loc = (char*)tracy_malloc( loclen + addrlen + 1 );
-    memcpy( loc, symloc, loclen );
-    memcpy( loc + loclen, buf, addrlen );
-    loc[loclen + addrlen] = '\0';
-    ret.file = loc;
-
-    if( sym ) free( sym );
-    if( demangled ) free( demangled );
-
-    return ret;
-}
-
-#elif TRACY_HAS_CALLSTACK == 3
-
-CallstackEntry DecodeCallstackPtr( uint64_t ptr )
-{
-    CallstackEntry ret;
-    ret.line = 0;
-
-    char* demangled = nullptr;
-    const char* symname = nullptr;
-    const char* symloc = nullptr;
-    auto vptr = (void*)ptr;
-    char** sym = nullptr;
-    ptrdiff_t symoff = 0;
-
-    Dl_info dlinfo;
-    if( dladdr( vptr, &dlinfo ) )
-    {
-        symloc = dlinfo.dli_fname;
-        symname = dlinfo.dli_sname;
-        symoff = (char*)ptr - (char*)dlinfo.dli_saddr;
-
-        if( symname && symname[0] == '_' )
+        if( SymGetLineFromAddr64( proc, ptr, &displacement, &line ) == 0 )
         {
-            size_t len = 0;
-            int status;
-            demangled = abi::__cxa_demangle( symname, nullptr, &len, &status );
-            if( status == 0 )
-            {
-                symname = demangled;
-            }
-        }
-    }
-
-    if( !symname )
-    {
-        sym = backtrace_symbols( &vptr, 1 );
-        if( !sym )
-        {
-            symname = "[unknown]";
+            sym.file = "[unknown]";
+            sym.line = 0;
         }
         else
         {
-            symname = *sym;
+            sym.file = line.FileName;
+            sym.line = line.LineNumber;
         }
     }
-    if( !symloc )
+    sym.needFree = false;
+    return sym;
+}
+
+CallstackEntryData DecodeCallstackPtr( uint64_t ptr )
+{
+    int write;
+    const auto proc = GetCurrentProcess();
+#ifndef __CYGWIN__
+    DWORD inlineNum = SymAddrIncludeInlineTrace( proc, ptr );
+    if( inlineNum > MaxCbTrace - 1 ) inlineNum = MaxCbTrace - 1;
+    DWORD ctx = 0;
+    DWORD idx;
+    BOOL doInline = FALSE;
+    if( inlineNum != 0 ) doInline = SymQueryInlineTrace( proc, ptr, 0, ptr, ptr, &ctx, &idx );
+    if( doInline )
     {
-        symloc = "[unknown]";
+        write = inlineNum;
+        cb_num = 1 + inlineNum;
     }
+    else
+#endif
+    {
+        write = 0;
+        cb_num = 1;
+    }
+
+    char buf[sizeof( SYMBOL_INFO ) + MaxNameSize];
+    auto si = (SYMBOL_INFO*)buf;
+    si->SizeOfStruct = sizeof( SYMBOL_INFO );
+    si->MaxNameLen = MaxNameSize;
+
+    const auto moduleName = GetModuleName( ptr );
+    const auto symValid = SymFromAddr( proc, ptr, nullptr, si ) != 0;
+
+    IMAGEHLP_LINE64 line;
+    DWORD displacement = 0;
+    line.SizeOfStruct = sizeof(IMAGEHLP_LINE64);
+
+    {
+        const char* filename;
+        if( SymGetLineFromAddr64( proc, ptr, &displacement, &line ) == 0 )
+        {
+            filename = "[unknown]";
+            cb_data[write].line = 0;
+        }
+        else
+        {
+            filename = line.FileName;
+            cb_data[write].line = line.LineNumber;
+        }
+
+        cb_data[write].name = symValid ? CopyString( si->Name, si->NameLen ) : CopyString( moduleName );
+        cb_data[write].file = CopyString( filename );
+        if( symValid )
+        {
+            cb_data[write].symLen = si->Size;
+            cb_data[write].symAddr = si->Address;
+        }
+        else
+        {
+            cb_data[write].symLen = 0;
+            cb_data[write].symAddr = 0;
+        }
+    }
+
+#ifndef __CYGWIN__
+    if( doInline )
+    {
+        for( DWORD i=0; i<inlineNum; i++ )
+        {
+            auto& cb = cb_data[i];
+            const auto symInlineValid = SymFromInlineContext( proc, ptr, ctx, nullptr, si ) != 0;
+            const char* filename;
+            if( SymGetLineFromInlineContext( proc, ptr, ctx, 0, &displacement, &line ) == 0 )
+            {
+                filename = "[unknown]";
+                cb.line = 0;
+            }
+            else
+            {
+                filename = line.FileName;
+                cb.line = line.LineNumber;
+            }
+
+            cb.name = symInlineValid ? CopyString( si->Name, si->NameLen ) : CopyString( moduleName );
+            cb.file = CopyString( filename );
+            if( symInlineValid )
+            {
+                cb.symLen = si->Size;
+                cb.symAddr = si->Address;
+            }
+            else
+            {
+                cb.symLen = 0;
+                cb.symAddr = 0;
+            }
+
+            ctx++;
+        }
+    }
+#endif
+
+    return { cb_data, uint8_t( cb_num ), moduleName };
+}
+
+#elif TRACY_HAS_CALLSTACK == 2 || TRACY_HAS_CALLSTACK == 3 || TRACY_HAS_CALLSTACK == 4 || TRACY_HAS_CALLSTACK == 6
+
+enum { MaxCbTrace = 16 };
+
+struct backtrace_state* cb_bts;
+int cb_num;
+CallstackEntry cb_data[MaxCbTrace];
+int cb_fixup;
+
+void InitCallstack()
+{
+    cb_bts = backtrace_create_state( nullptr, 0, nullptr, nullptr );
+}
+
+static int FastCallstackDataCb( void* data, uintptr_t pc, uintptr_t lowaddr, const char* fn, int lineno, const char* function )
+{
+    if( function )
+    {
+        strcpy( (char*)data, function );
+    }
+    else
+    {
+        const char* symname = nullptr;
+        auto vptr = (void*)pc;
+        Dl_info dlinfo;
+        if( dladdr( vptr, &dlinfo ) )
+        {
+            symname = dlinfo.dli_sname;
+        }
+        if( symname )
+        {
+            strcpy( (char*)data, symname );
+        }
+        else
+        {
+            *(char*)data = '\0';
+        }
+    }
+    return 1;
+}
+
+static void FastCallstackErrorCb( void* data, const char* /*msg*/, int /*errnum*/ )
+{
+    *(char*)data = '\0';
+}
+
+const char* DecodeCallstackPtrFast( uint64_t ptr )
+{
+    static char ret[1024];
+    backtrace_pcinfo( cb_bts, ptr, FastCallstackDataCb, FastCallstackErrorCb, ret );
+    return ret;
+}
+
+static int SymbolAddressDataCb( void* data, uintptr_t pc, uintptr_t lowaddr, const char* fn, int lineno, const char* function )
+{
+    auto& sym = *(SymbolData*)data;
+    if( !fn )
+    {
+        const char* symloc = nullptr;
+        Dl_info dlinfo;
+        if( dladdr( (void*)pc, &dlinfo ) ) symloc = dlinfo.dli_fname;
+        if( !symloc ) symloc = "[unknown]";
+        sym.file = symloc;
+        sym.line = 0;
+        sym.needFree = false;
+    }
+    else
+    {
+        sym.file = CopyString( fn );
+        sym.line = lineno;
+        sym.needFree = true;
+    }
+
+    return 1;
+}
+
+static void SymbolAddressErrorCb( void* data, const char* /*msg*/, int /*errnum*/ )
+{
+    auto& sym = *(SymbolData*)data;
+    sym.file = "[unknown]";
+    sym.line = 0;
+    sym.needFree = false;
+}
+
+SymbolData DecodeSymbolAddress( uint64_t ptr )
+{
+    SymbolData sym;
+    backtrace_pcinfo( cb_bts, ptr, SymbolAddressDataCb, SymbolAddressErrorCb, &sym );
+    return sym;
+}
+
+SymbolData DecodeCodeAddress( uint64_t ptr )
+{
+    return DecodeSymbolAddress( ptr );
+}
+
+static int CallstackDataCb( void* /*data*/, uintptr_t pc, uintptr_t lowaddr, const char* fn, int lineno, const char* function )
+{
+    enum { DemangleBufLen = 64*1024 };
+    char demangled[DemangleBufLen];
+
+    cb_data[cb_num].symLen = 0;
+    cb_data[cb_num].symAddr = (uint64_t)lowaddr;
+
+    if( !fn && !function )
+    {
+        const char* symname = nullptr;
+        const char* symloc = nullptr;
+        auto vptr = (void*)pc;
+        ptrdiff_t symoff = 0;
+
+        Dl_info dlinfo;
+        if( dladdr( vptr, &dlinfo ) )
+        {
+            symloc = dlinfo.dli_fname;
+            symname = dlinfo.dli_sname;
+            symoff = (char*)pc - (char*)dlinfo.dli_saddr;
+
+            if( symname && symname[0] == '_' )
+            {
+                size_t len = DemangleBufLen;
+                int status;
+                abi::__cxa_demangle( symname, demangled, &len, &status );
+                if( status == 0 )
+                {
+                    symname = demangled;
+                }
+            }
+        }
+
+        if( !symname ) symname = "[unknown]";
+        if( !symloc ) symloc = "[unknown]";
+
+        if( symoff == 0 )
+        {
+            cb_data[cb_num].name = CopyString( symname );
+        }
+        else
+        {
+            char buf[32];
+            const auto offlen = sprintf( buf, " + %td", symoff );
+            const auto namelen = strlen( symname );
+            auto name = (char*)tracy_malloc( namelen + offlen + 1 );
+            memcpy( name, symname, namelen );
+            memcpy( name + namelen, buf, offlen );
+            name[namelen + offlen] = '\0';
+            cb_data[cb_num].name = name;
+        }
+
+        char buf[32];
+        const auto addrlen = sprintf( buf, " [%p]", (void*)pc );
+        const auto loclen = strlen( symloc );
+        auto loc = (char*)tracy_malloc( loclen + addrlen + 1 );
+        memcpy( loc, symloc, loclen );
+        memcpy( loc + loclen, buf, addrlen );
+        loc[loclen + addrlen] = '\0';
+        cb_data[cb_num].file = loc;
+
+        cb_data[cb_num].line = 0;
+    }
+    else
+    {
+        if( !fn ) fn = "[unknown]";
+        if( !function )
+        {
+            function = "[unknown]";
+        }
+        else
+        {
+            if( function[0] == '_' )
+            {
+                size_t len = DemangleBufLen;
+                int status;
+                abi::__cxa_demangle( function, demangled, &len, &status );
+                if( status == 0 )
+                {
+                    function = demangled;
+                }
+            }
+        }
+
+        cb_data[cb_num].name = CopyString( function );
+        cb_data[cb_num].file = CopyString( fn );
+        cb_data[cb_num].line = lineno;
+    }
+
+    if( ++cb_num >= MaxCbTrace )
+    {
+        return 1;
+    }
+    else
+    {
+        return 0;
+    }
+}
+
+static void CallstackErrorCb( void* /*data*/, const char* /*msg*/, int /*errnum*/ )
+{
+    for( int i=0; i<cb_num; i++ )
+    {
+        tracy_free( (void*)cb_data[i].name );
+        tracy_free( (void*)cb_data[i].file );
+    }
+
+    cb_data[0].name = CopyString( "[error]" );
+    cb_data[0].file = CopyString( "[error]" );
+    cb_data[0].line = 0;
+
+    cb_num = 1;
+}
+
+void SymInfoCallback( void* /*data*/, uintptr_t pc, const char* symname, uintptr_t symval, uintptr_t symsize )
+{
+    cb_data[cb_num-1].symLen = (uint32_t)symsize;
+    cb_data[cb_num-1].symAddr = (uint64_t)symval;
+}
+
+void SymInfoError( void* /*data*/, const char* /*msg*/, int /*errnum*/ )
+{
+    cb_data[cb_num-1].symLen = 0;
+    cb_data[cb_num-1].symAddr = 0;
+}
+
+CallstackEntryData DecodeCallstackPtr( uint64_t ptr )
+{
+    cb_num = 0;
+    backtrace_pcinfo( cb_bts, ptr, CallstackDataCb, CallstackErrorCb, nullptr );
+    assert( cb_num > 0 );
+
+    backtrace_syminfo( cb_bts, ptr, SymInfoCallback, SymInfoError, nullptr );
+
+    const char* symloc = nullptr;
+    Dl_info dlinfo;
+    if( dladdr( (void*)ptr, &dlinfo ) ) symloc = dlinfo.dli_fname;
+
+    return { cb_data, uint8_t( cb_num ), symloc ? symloc : "[unknown]" };
+}
+
+#elif TRACY_HAS_CALLSTACK == 5
+
+void InitCallstack()
+{
+}
+
+const char* DecodeCallstackPtrFast( uint64_t ptr )
+{
+    static char ret[1024];
+    auto vptr = (void*)ptr;
+    const char* symname = nullptr;
+    Dl_info dlinfo;
+    if( dladdr( vptr, &dlinfo ) && dlinfo.dli_sname )
+    {
+        symname = dlinfo.dli_sname;
+    }
+    if( symname )
+    {
+        strcpy( ret, symname );
+    }
+    else
+    {
+        *ret = '\0';
+    }
+    return ret;
+}
+
+SymbolData DecodeSymbolAddress( uint64_t ptr )
+{
+    const char* symloc = nullptr;
+    Dl_info dlinfo;
+    if( dladdr( (void*)ptr, &dlinfo ) ) symloc = dlinfo.dli_fname;
+    if( !symloc ) symloc = "[unknown]";
+    return SymbolData { symloc, 0, false };
+}
+
+SymbolData DecodeCodeAddress( uint64_t ptr )
+{
+    return DecodeSymbolAddress( ptr );
+}
+
+CallstackEntryData DecodeCallstackPtr( uint64_t ptr )
+{
+    static CallstackEntry cb;
+    cb.line = 0;
+
+    char* demangled = nullptr;
+    const char* symname = nullptr;
+    const char* symloc = nullptr;
+    auto vptr = (void*)ptr;
+    ptrdiff_t symoff = 0;
+    void* symaddr = nullptr;
+
+    Dl_info dlinfo;
+    if( dladdr( vptr, &dlinfo ) )
+    {
+        symloc = dlinfo.dli_fname;
+        symname = dlinfo.dli_sname;
+        symoff = (char*)ptr - (char*)dlinfo.dli_saddr;
+        symaddr = dlinfo.dli_saddr;
+
+        if( symname && symname[0] == '_' )
+        {
+            size_t len = 0;
+            int status;
+            demangled = abi::__cxa_demangle( symname, nullptr, &len, &status );
+            if( status == 0 )
+            {
+                symname = demangled;
+            }
+        }
+    }
+
+    if( !symname ) symname = "[unknown]";
+    if( !symloc ) symloc = "[unknown]";
 
     if( symoff == 0 )
     {
-        const auto namelen = strlen( symname );
-        auto name = (char*)tracy_malloc( namelen + 1 );
-        memcpy( name, symname, namelen );
-        name[namelen] = '\0';
-        ret.name = name;
+        cb.name = CopyString( symname );
     }
     else
     {
         char buf[32];
-        sprintf( buf, " + %td", symoff );
-        const auto offlen = strlen( buf );
+        const auto offlen = sprintf( buf, " + %td", symoff );
         const auto namelen = strlen( symname );
         auto name = (char*)tracy_malloc( namelen + offlen + 1 );
         memcpy( name, symname, namelen );
         memcpy( name + namelen, buf, offlen );
         name[namelen + offlen] = '\0';
-        ret.name = name;
+        cb.name = name;
     }
 
     char buf[32];
-    sprintf( buf, " [%p]", (void*)ptr );
-    const auto addrlen = strlen( buf );
+    const auto addrlen = sprintf( buf, " [%p]", (void*)ptr );
     const auto loclen = strlen( symloc );
     auto loc = (char*)tracy_malloc( loclen + addrlen + 1 );
     memcpy( loc, symloc, loclen );
     memcpy( loc + loclen, buf, addrlen );
     loc[loclen + addrlen] = '\0';
-    ret.file = loc;
+    cb.file = loc;
+
+    cb.symLen = 0;
+    cb.symAddr = (uint64_t)symaddr;
 
-    if( sym ) free( sym );
     if( demangled ) free( demangled );
 
-    return ret;
+    return { &cb, 1, symloc };
 }
 
 #endif

client/TracyCallstack.h

@@ -0,0 +1,28 @@
+#ifndef __TRACYCALLSTACK_H__
+#define __TRACYCALLSTACK_H__
+
+#if !defined _WIN32 && !defined __CYGWIN__
+#  include <sys/param.h>
+#endif
+
+#if defined _WIN32 || defined __CYGWIN__
+#  define TRACY_HAS_CALLSTACK 1
+#elif defined __ANDROID__
+#  if !defined __arm__ || __ANDROID_API__ >= 21
+#    define TRACY_HAS_CALLSTACK 2
+#  else
+#    define TRACY_HAS_CALLSTACK 5
+#  endif
+#elif defined __linux
+#  if defined _GNU_SOURCE && defined __GLIBC__
+#    define TRACY_HAS_CALLSTACK 3
+#  else
+#    define TRACY_HAS_CALLSTACK 2
+#  endif
+#elif defined __APPLE__
+#  define TRACY_HAS_CALLSTACK 4
+#elif defined BSD
+#  define TRACY_HAS_CALLSTACK 6
+#endif
+
+#endif

client/TracyCallstack.hpp

@@ -1,24 +1,13 @@
 #ifndef __TRACYCALLSTACK_HPP__
 #define __TRACYCALLSTACK_HPP__
 
-#if defined _WIN32 || defined __CYGWIN__
-#  define TRACY_HAS_CALLSTACK 1
-extern "C"
-{
-    typedef unsigned long (__stdcall *t_RtlWalkFrameChain)( void**, unsigned long, unsigned long );
-    extern t_RtlWalkFrameChain RtlWalkFrameChain;
-}
-#elif defined __ANDROID__
-#  define TRACY_HAS_CALLSTACK 2
+#include "../common/TracyApi.h"
+#include "TracyCallstack.h"
+
+#if TRACY_HAS_CALLSTACK == 2 || TRACY_HAS_CALLSTACK == 5
 #  include <unwind.h>
-#elif defined __linux
-#  if defined _GNU_SOURCE && defined __GLIBC__
-#    define TRACY_HAS_CALLSTACK 3
-#    include <execinfo.h>
-#  else
-#    define TRACY_HAS_CALLSTACK 2
-#    include <unwind.h>
-#  endif
+#elif TRACY_HAS_CALLSTACK >= 3
+#  include <execinfo.h>
 #endif
 
 
@@ -26,7 +15,6 @@ extern "C"
 
 #include <assert.h>
 #include <stdint.h>
-#include <string.h>
 
 #include "../common/TracyAlloc.hpp"
 #include "../common/TracyForceInline.hpp"
@@ -34,33 +22,46 @@ extern "C"
 namespace tracy
 {
 
+struct SymbolData
+{
+    const char* file;
+    uint32_t line;
+    bool needFree;
+};
+
 struct CallstackEntry
 {
     const char* name;
     const char* file;
     uint32_t line;
+    uint32_t symLen;
+    uint64_t symAddr;
 };
 
-CallstackEntry DecodeCallstackPtr( uint64_t ptr );
+struct CallstackEntryData
+{
+    const CallstackEntry* data;
+    uint8_t size;
+    const char* imageName;
+};
+
+SymbolData DecodeSymbolAddress( uint64_t ptr );
+SymbolData DecodeCodeAddress( uint64_t ptr );
+const char* DecodeCallstackPtrFast( uint64_t ptr );
+CallstackEntryData DecodeCallstackPtr( uint64_t ptr );
+void InitCallstack();
 
 #if TRACY_HAS_CALLSTACK == 1
 
-void InitCallstack();
+TRACY_API uintptr_t* CallTrace( int depth );
 
 static tracy_force_inline void* Callstack( int depth )
 {
     assert( depth >= 1 && depth < 63 );
-
-    auto trace = (uintptr_t*)tracy_malloc( ( 1 + depth ) * sizeof( uintptr_t ) );
-    const auto num = RtlWalkFrameChain( (void**)( trace + 1 ), depth, 0 );
-    *trace = num;
-
-    return trace;
+    return CallTrace( depth );
 }
 
-#elif TRACY_HAS_CALLSTACK == 2
-
-static tracy_force_inline void InitCallstack() {}
+#elif TRACY_HAS_CALLSTACK == 2 || TRACY_HAS_CALLSTACK == 5
 
 struct BacktraceState
 {
@@ -93,9 +94,7 @@ static tracy_force_inline void* Callstack( int depth )
     return trace;
 }
 
-#elif TRACY_HAS_CALLSTACK == 3
-
-static tracy_force_inline void InitCallstack() {}
+#elif TRACY_HAS_CALLSTACK == 3 || TRACY_HAS_CALLSTACK == 4 || TRACY_HAS_CALLSTACK == 6
 
 static tracy_force_inline void* Callstack( int depth )
 {

client/TracyDxt1.cpp

@@ -0,0 +1,629 @@
+#include "TracyDxt1.hpp"
+#include "../common/TracyForceInline.hpp"
+
+#include <assert.h>
+#include <stdint.h>
+#include <string.h>
+
+#ifdef __ARM_NEON
+#  include <arm_neon.h>
+#endif
+
+#if defined __AVX__ && !defined __SSE4_1__
+#  define __SSE4_1__
+#endif
+
+#if defined __SSE4_1__ || defined __AVX2__
+#  ifdef _MSC_VER
+#    include <intrin.h>
+#  else
+#    include <x86intrin.h>
+#    ifndef _mm256_cvtsi256_si32
+#      define _mm256_cvtsi256_si32( v ) ( _mm_cvtsi128_si32( _mm256_castsi256_si128( v ) ) )
+#    endif
+#  endif
+#endif
+
+namespace tracy
+{
+
+static inline uint16_t to565( uint8_t r, uint8_t g, uint8_t b )
+{
+    return ( ( r & 0xF8 ) << 8 ) | ( ( g & 0xFC ) << 3 ) | ( b >> 3 );
+}
+
+static inline uint16_t to565( uint32_t c )
+{
+    return
+        ( ( c & 0xF80000 ) >> 19 ) |
+        ( ( c & 0x00FC00 ) >> 5 ) |
+        ( ( c & 0x0000F8 ) << 8 );
+}
+
+static const uint16_t DivTable[255*3+1] = {
+    0xffff, 0xffff, 0xffff, 0xffff, 0xcccc, 0xaaaa, 0x9249, 0x8000, 0x71c7, 0x6666, 0x5d17, 0x5555, 0x4ec4, 0x4924, 0x4444, 0x4000,
+    0x3c3c, 0x38e3, 0x35e5, 0x3333, 0x30c3, 0x2e8b, 0x2c85, 0x2aaa, 0x28f5, 0x2762, 0x25ed, 0x2492, 0x234f, 0x2222, 0x2108, 0x2000,
+    0x1f07, 0x1e1e, 0x1d41, 0x1c71, 0x1bac, 0x1af2, 0x1a41, 0x1999, 0x18f9, 0x1861, 0x17d0, 0x1745, 0x16c1, 0x1642, 0x15c9, 0x1555,
+    0x14e5, 0x147a, 0x1414, 0x13b1, 0x1352, 0x12f6, 0x129e, 0x1249, 0x11f7, 0x11a7, 0x115b, 0x1111, 0x10c9, 0x1084, 0x1041, 0x1000,
+    0x0fc0, 0x0f83, 0x0f48, 0x0f0f, 0x0ed7, 0x0ea0, 0x0e6c, 0x0e38, 0x0e07, 0x0dd6, 0x0da7, 0x0d79, 0x0d4c, 0x0d20, 0x0cf6, 0x0ccc,
+    0x0ca4, 0x0c7c, 0x0c56, 0x0c30, 0x0c0c, 0x0be8, 0x0bc5, 0x0ba2, 0x0b81, 0x0b60, 0x0b40, 0x0b21, 0x0b02, 0x0ae4, 0x0ac7, 0x0aaa,
+    0x0a8e, 0x0a72, 0x0a57, 0x0a3d, 0x0a23, 0x0a0a, 0x09f1, 0x09d8, 0x09c0, 0x09a9, 0x0991, 0x097b, 0x0964, 0x094f, 0x0939, 0x0924,
+    0x090f, 0x08fb, 0x08e7, 0x08d3, 0x08c0, 0x08ad, 0x089a, 0x0888, 0x0876, 0x0864, 0x0853, 0x0842, 0x0831, 0x0820, 0x0810, 0x0800,
+    0x07f0, 0x07e0, 0x07d1, 0x07c1, 0x07b3, 0x07a4, 0x0795, 0x0787, 0x0779, 0x076b, 0x075d, 0x0750, 0x0743, 0x0736, 0x0729, 0x071c,
+    0x070f, 0x0703, 0x06f7, 0x06eb, 0x06df, 0x06d3, 0x06c8, 0x06bc, 0x06b1, 0x06a6, 0x069b, 0x0690, 0x0685, 0x067b, 0x0670, 0x0666,
+    0x065c, 0x0652, 0x0648, 0x063e, 0x0634, 0x062b, 0x0621, 0x0618, 0x060f, 0x0606, 0x05fd, 0x05f4, 0x05eb, 0x05e2, 0x05d9, 0x05d1,
+    0x05c9, 0x05c0, 0x05b8, 0x05b0, 0x05a8, 0x05a0, 0x0598, 0x0590, 0x0588, 0x0581, 0x0579, 0x0572, 0x056b, 0x0563, 0x055c, 0x0555,
+    0x054e, 0x0547, 0x0540, 0x0539, 0x0532, 0x052b, 0x0525, 0x051e, 0x0518, 0x0511, 0x050b, 0x0505, 0x04fe, 0x04f8, 0x04f2, 0x04ec,
+    0x04e6, 0x04e0, 0x04da, 0x04d4, 0x04ce, 0x04c8, 0x04c3, 0x04bd, 0x04b8, 0x04b2, 0x04ad, 0x04a7, 0x04a2, 0x049c, 0x0497, 0x0492,
+    0x048d, 0x0487, 0x0482, 0x047d, 0x0478, 0x0473, 0x046e, 0x0469, 0x0465, 0x0460, 0x045b, 0x0456, 0x0452, 0x044d, 0x0448, 0x0444,
+    0x043f, 0x043b, 0x0436, 0x0432, 0x042d, 0x0429, 0x0425, 0x0421, 0x041c, 0x0418, 0x0414, 0x0410, 0x040c, 0x0408, 0x0404, 0x0400,
+    0x03fc, 0x03f8, 0x03f4, 0x03f0, 0x03ec, 0x03e8, 0x03e4, 0x03e0, 0x03dd, 0x03d9, 0x03d5, 0x03d2, 0x03ce, 0x03ca, 0x03c7, 0x03c3,
+    0x03c0, 0x03bc, 0x03b9, 0x03b5, 0x03b2, 0x03ae, 0x03ab, 0x03a8, 0x03a4, 0x03a1, 0x039e, 0x039b, 0x0397, 0x0394, 0x0391, 0x038e,
+    0x038b, 0x0387, 0x0384, 0x0381, 0x037e, 0x037b, 0x0378, 0x0375, 0x0372, 0x036f, 0x036c, 0x0369, 0x0366, 0x0364, 0x0361, 0x035e,
+    0x035b, 0x0358, 0x0355, 0x0353, 0x0350, 0x034d, 0x034a, 0x0348, 0x0345, 0x0342, 0x0340, 0x033d, 0x033a, 0x0338, 0x0335, 0x0333,
+    0x0330, 0x032e, 0x032b, 0x0329, 0x0326, 0x0324, 0x0321, 0x031f, 0x031c, 0x031a, 0x0317, 0x0315, 0x0313, 0x0310, 0x030e, 0x030c,
+    0x0309, 0x0307, 0x0305, 0x0303, 0x0300, 0x02fe, 0x02fc, 0x02fa, 0x02f7, 0x02f5, 0x02f3, 0x02f1, 0x02ef, 0x02ec, 0x02ea, 0x02e8,
+    0x02e6, 0x02e4, 0x02e2, 0x02e0, 0x02de, 0x02dc, 0x02da, 0x02d8, 0x02d6, 0x02d4, 0x02d2, 0x02d0, 0x02ce, 0x02cc, 0x02ca, 0x02c8,
+    0x02c6, 0x02c4, 0x02c2, 0x02c0, 0x02be, 0x02bc, 0x02bb, 0x02b9, 0x02b7, 0x02b5, 0x02b3, 0x02b1, 0x02b0, 0x02ae, 0x02ac, 0x02aa,
+    0x02a8, 0x02a7, 0x02a5, 0x02a3, 0x02a1, 0x02a0, 0x029e, 0x029c, 0x029b, 0x0299, 0x0297, 0x0295, 0x0294, 0x0292, 0x0291, 0x028f,
+    0x028d, 0x028c, 0x028a, 0x0288, 0x0287, 0x0285, 0x0284, 0x0282, 0x0280, 0x027f, 0x027d, 0x027c, 0x027a, 0x0279, 0x0277, 0x0276,
+    0x0274, 0x0273, 0x0271, 0x0270, 0x026e, 0x026d, 0x026b, 0x026a, 0x0268, 0x0267, 0x0265, 0x0264, 0x0263, 0x0261, 0x0260, 0x025e,
+    0x025d, 0x025c, 0x025a, 0x0259, 0x0257, 0x0256, 0x0255, 0x0253, 0x0252, 0x0251, 0x024f, 0x024e, 0x024d, 0x024b, 0x024a, 0x0249,
+    0x0247, 0x0246, 0x0245, 0x0243, 0x0242, 0x0241, 0x0240, 0x023e, 0x023d, 0x023c, 0x023b, 0x0239, 0x0238, 0x0237, 0x0236, 0x0234,
+    0x0233, 0x0232, 0x0231, 0x0230, 0x022e, 0x022d, 0x022c, 0x022b, 0x022a, 0x0229, 0x0227, 0x0226, 0x0225, 0x0224, 0x0223, 0x0222,
+    0x0220, 0x021f, 0x021e, 0x021d, 0x021c, 0x021b, 0x021a, 0x0219, 0x0218, 0x0216, 0x0215, 0x0214, 0x0213, 0x0212, 0x0211, 0x0210,
+    0x020f, 0x020e, 0x020d, 0x020c, 0x020b, 0x020a, 0x0209, 0x0208, 0x0207, 0x0206, 0x0205, 0x0204, 0x0203, 0x0202, 0x0201, 0x0200,
+    0x01ff, 0x01fe, 0x01fd, 0x01fc, 0x01fb, 0x01fa, 0x01f9, 0x01f8, 0x01f7, 0x01f6, 0x01f5, 0x01f4, 0x01f3, 0x01f2, 0x01f1, 0x01f0,
+    0x01ef, 0x01ee, 0x01ed, 0x01ec, 0x01eb, 0x01ea, 0x01e9, 0x01e9, 0x01e8, 0x01e7, 0x01e6, 0x01e5, 0x01e4, 0x01e3, 0x01e2, 0x01e1,
+    0x01e0, 0x01e0, 0x01df, 0x01de, 0x01dd, 0x01dc, 0x01db, 0x01da, 0x01da, 0x01d9, 0x01d8, 0x01d7, 0x01d6, 0x01d5, 0x01d4, 0x01d4,
+    0x01d3, 0x01d2, 0x01d1, 0x01d0, 0x01cf, 0x01cf, 0x01ce, 0x01cd, 0x01cc, 0x01cb, 0x01cb, 0x01ca, 0x01c9, 0x01c8, 0x01c7, 0x01c7,
+    0x01c6, 0x01c5, 0x01c4, 0x01c3, 0x01c3, 0x01c2, 0x01c1, 0x01c0, 0x01c0, 0x01bf, 0x01be, 0x01bd, 0x01bd, 0x01bc, 0x01bb, 0x01ba,
+    0x01ba, 0x01b9, 0x01b8, 0x01b7, 0x01b7, 0x01b6, 0x01b5, 0x01b4, 0x01b4, 0x01b3, 0x01b2, 0x01b2, 0x01b1, 0x01b0, 0x01af, 0x01af,
+    0x01ae, 0x01ad, 0x01ad, 0x01ac, 0x01ab, 0x01aa, 0x01aa, 0x01a9, 0x01a8, 0x01a8, 0x01a7, 0x01a6, 0x01a6, 0x01a5, 0x01a4, 0x01a4,
+    0x01a3, 0x01a2, 0x01a2, 0x01a1, 0x01a0, 0x01a0, 0x019f, 0x019e, 0x019e, 0x019d, 0x019c, 0x019c, 0x019b, 0x019a, 0x019a, 0x0199,
+    0x0198, 0x0198, 0x0197, 0x0197, 0x0196, 0x0195, 0x0195, 0x0194, 0x0193, 0x0193, 0x0192, 0x0192, 0x0191, 0x0190, 0x0190, 0x018f,
+    0x018f, 0x018e, 0x018d, 0x018d, 0x018c, 0x018b, 0x018b, 0x018a, 0x018a, 0x0189, 0x0189, 0x0188, 0x0187, 0x0187, 0x0186, 0x0186,
+    0x0185, 0x0184, 0x0184, 0x0183, 0x0183, 0x0182, 0x0182, 0x0181, 0x0180, 0x0180, 0x017f, 0x017f, 0x017e, 0x017e, 0x017d, 0x017d,
+    0x017c, 0x017b, 0x017b, 0x017a, 0x017a, 0x0179, 0x0179, 0x0178, 0x0178, 0x0177, 0x0177, 0x0176, 0x0175, 0x0175, 0x0174, 0x0174,
+    0x0173, 0x0173, 0x0172, 0x0172, 0x0171, 0x0171, 0x0170, 0x0170, 0x016f, 0x016f, 0x016e, 0x016e, 0x016d, 0x016d, 0x016c, 0x016c,
+    0x016b, 0x016b, 0x016a, 0x016a, 0x0169, 0x0169, 0x0168, 0x0168, 0x0167, 0x0167, 0x0166, 0x0166, 0x0165, 0x0165, 0x0164, 0x0164,
+    0x0163, 0x0163, 0x0162, 0x0162, 0x0161, 0x0161, 0x0160, 0x0160, 0x015f, 0x015f, 0x015e, 0x015e, 0x015d, 0x015d, 0x015d, 0x015c,
+    0x015c, 0x015b, 0x015b, 0x015a, 0x015a, 0x0159, 0x0159, 0x0158, 0x0158, 0x0158, 0x0157, 0x0157, 0x0156, 0x0156
+};
+static const uint16_t DivTableNEON[255*3+1] = {
+    0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000, 0x0000,
+    0x0000, 0x1c71, 0x1af2, 0x1999, 0x1861, 0x1745, 0x1642, 0x1555, 0x147a, 0x13b1, 0x12f6, 0x1249, 0x11a7, 0x1111, 0x1084, 0x1000,
+    0x0f83, 0x0f0f, 0x0ea0, 0x0e38, 0x0dd6, 0x0d79, 0x0d20, 0x0ccc, 0x0c7c, 0x0c30, 0x0be8, 0x0ba2, 0x0b60, 0x0b21, 0x0ae4, 0x0aaa,
+    0x0a72, 0x0a3d, 0x0a0a, 0x09d8, 0x09a9, 0x097b, 0x094f, 0x0924, 0x08fb, 0x08d3, 0x08ad, 0x0888, 0x0864, 0x0842, 0x0820, 0x0800,
+    0x07e0, 0x07c1, 0x07a4, 0x0787, 0x076b, 0x0750, 0x0736, 0x071c, 0x0703, 0x06eb, 0x06d3, 0x06bc, 0x06a6, 0x0690, 0x067b, 0x0666,
+    0x0652, 0x063e, 0x062b, 0x0618, 0x0606, 0x05f4, 0x05e2, 0x05d1, 0x05c0, 0x05b0, 0x05a0, 0x0590, 0x0581, 0x0572, 0x0563, 0x0555,
+    0x0547, 0x0539, 0x052b, 0x051e, 0x0511, 0x0505, 0x04f8, 0x04ec, 0x04e0, 0x04d4, 0x04c8, 0x04bd, 0x04b2, 0x04a7, 0x049c, 0x0492,
+    0x0487, 0x047d, 0x0473, 0x0469, 0x0460, 0x0456, 0x044d, 0x0444, 0x043b, 0x0432, 0x0429, 0x0421, 0x0418, 0x0410, 0x0408, 0x0400,
+    0x03f8, 0x03f0, 0x03e8, 0x03e0, 0x03d9, 0x03d2, 0x03ca, 0x03c3, 0x03bc, 0x03b5, 0x03ae, 0x03a8, 0x03a1, 0x039b, 0x0394, 0x038e,
+    0x0387, 0x0381, 0x037b, 0x0375, 0x036f, 0x0369, 0x0364, 0x035e, 0x0358, 0x0353, 0x034d, 0x0348, 0x0342, 0x033d, 0x0338, 0x0333,
+    0x032e, 0x0329, 0x0324, 0x031f, 0x031a, 0x0315, 0x0310, 0x030c, 0x0307, 0x0303, 0x02fe, 0x02fa, 0x02f5, 0x02f1, 0x02ec, 0x02e8,
+    0x02e4, 0x02e0, 0x02dc, 0x02d8, 0x02d4, 0x02d0, 0x02cc, 0x02c8, 0x02c4, 0x02c0, 0x02bc, 0x02b9, 0x02b5, 0x02b1, 0x02ae, 0x02aa,
+    0x02a7, 0x02a3, 0x02a0, 0x029c, 0x0299, 0x0295, 0x0292, 0x028f, 0x028c, 0x0288, 0x0285, 0x0282, 0x027f, 0x027c, 0x0279, 0x0276,
+    0x0273, 0x0270, 0x026d, 0x026a, 0x0267, 0x0264, 0x0261, 0x025e, 0x025c, 0x0259, 0x0256, 0x0253, 0x0251, 0x024e, 0x024b, 0x0249,
+    0x0246, 0x0243, 0x0241, 0x023e, 0x023c, 0x0239, 0x0237, 0x0234, 0x0232, 0x0230, 0x022d, 0x022b, 0x0229, 0x0226, 0x0224, 0x0222,
+    0x021f, 0x021d, 0x021b, 0x0219, 0x0216, 0x0214, 0x0212, 0x0210, 0x020e, 0x020c, 0x020a, 0x0208, 0x0206, 0x0204, 0x0202, 0x0200,
+    0x01fe, 0x01fc, 0x01fa, 0x01f8, 0x01f6, 0x01f4, 0x01f2, 0x01f0, 0x01ee, 0x01ec, 0x01ea, 0x01e9, 0x01e7, 0x01e5, 0x01e3, 0x01e1,
+    0x01e0, 0x01de, 0x01dc, 0x01da, 0x01d9, 0x01d7, 0x01d5, 0x01d4, 0x01d2, 0x01d0, 0x01cf, 0x01cd, 0x01cb, 0x01ca, 0x01c8, 0x01c7,
+    0x01c5, 0x01c3, 0x01c2, 0x01c0, 0x01bf, 0x01bd, 0x01bc, 0x01ba, 0x01b9, 0x01b7, 0x01b6, 0x01b4, 0x01b3, 0x01b2, 0x01b0, 0x01af,
+    0x01ad, 0x01ac, 0x01aa, 0x01a9, 0x01a8, 0x01a6, 0x01a5, 0x01a4, 0x01a2, 0x01a1, 0x01a0, 0x019e, 0x019d, 0x019c, 0x019a, 0x0199,
+    0x0198, 0x0197, 0x0195, 0x0194, 0x0193, 0x0192, 0x0190, 0x018f, 0x018e, 0x018d, 0x018b, 0x018a, 0x0189, 0x0188, 0x0187, 0x0186,
+    0x0184, 0x0183, 0x0182, 0x0181, 0x0180, 0x017f, 0x017e, 0x017d, 0x017b, 0x017a, 0x0179, 0x0178, 0x0177, 0x0176, 0x0175, 0x0174,
+    0x0173, 0x0172, 0x0171, 0x0170, 0x016f, 0x016e, 0x016d, 0x016c, 0x016b, 0x016a, 0x0169, 0x0168, 0x0167, 0x0166, 0x0165, 0x0164,
+    0x0163, 0x0162, 0x0161, 0x0160, 0x015f, 0x015e, 0x015d, 0x015c, 0x015b, 0x015a, 0x0159, 0x0158, 0x0158, 0x0157, 0x0156, 0x0155,
+    0x0154, 0x0153, 0x0152, 0x0151, 0x0150, 0x0150, 0x014f, 0x014e, 0x014d, 0x014c, 0x014b, 0x014a, 0x014a, 0x0149, 0x0148, 0x0147,
+    0x0146, 0x0146, 0x0145, 0x0144, 0x0143, 0x0142, 0x0142, 0x0141, 0x0140, 0x013f, 0x013e, 0x013e, 0x013d, 0x013c, 0x013b, 0x013b,
+    0x013a, 0x0139, 0x0138, 0x0138, 0x0137, 0x0136, 0x0135, 0x0135, 0x0134, 0x0133, 0x0132, 0x0132, 0x0131, 0x0130, 0x0130, 0x012f,
+    0x012e, 0x012e, 0x012d, 0x012c, 0x012b, 0x012b, 0x012a, 0x0129, 0x0129, 0x0128, 0x0127, 0x0127, 0x0126, 0x0125, 0x0125, 0x0124,
+    0x0123, 0x0123, 0x0122, 0x0121, 0x0121, 0x0120, 0x0120, 0x011f, 0x011e, 0x011e, 0x011d, 0x011c, 0x011c, 0x011b, 0x011b, 0x011a,
+    0x0119, 0x0119, 0x0118, 0x0118, 0x0117, 0x0116, 0x0116, 0x0115, 0x0115, 0x0114, 0x0113, 0x0113, 0x0112, 0x0112, 0x0111, 0x0111,
+    0x0110, 0x010f, 0x010f, 0x010e, 0x010e, 0x010d, 0x010d, 0x010c, 0x010c, 0x010b, 0x010a, 0x010a, 0x0109, 0x0109, 0x0108, 0x0108,
+    0x0107, 0x0107, 0x0106, 0x0106, 0x0105, 0x0105, 0x0104, 0x0104, 0x0103, 0x0103, 0x0102, 0x0102, 0x0101, 0x0101, 0x0100, 0x0100,
+    0x00ff, 0x00ff, 0x00fe, 0x00fe, 0x00fd, 0x00fd, 0x00fc, 0x00fc, 0x00fb, 0x00fb, 0x00fa, 0x00fa, 0x00f9, 0x00f9, 0x00f8, 0x00f8,
+    0x00f7, 0x00f7, 0x00f6, 0x00f6, 0x00f5, 0x00f5, 0x00f4, 0x00f4, 0x00f4, 0x00f3, 0x00f3, 0x00f2, 0x00f2, 0x00f1, 0x00f1, 0x00f0,
+    0x00f0, 0x00f0, 0x00ef, 0x00ef, 0x00ee, 0x00ee, 0x00ed, 0x00ed, 0x00ed, 0x00ec, 0x00ec, 0x00eb, 0x00eb, 0x00ea, 0x00ea, 0x00ea,
+    0x00e9, 0x00e9, 0x00e8, 0x00e8, 0x00e7, 0x00e7, 0x00e7, 0x00e6, 0x00e6, 0x00e5, 0x00e5, 0x00e5, 0x00e4, 0x00e4, 0x00e3, 0x00e3,
+    0x00e3, 0x00e2, 0x00e2, 0x00e1, 0x00e1, 0x00e1, 0x00e0, 0x00e0, 0x00e0, 0x00df, 0x00df, 0x00de, 0x00de, 0x00de, 0x00dd, 0x00dd,
+    0x00dd, 0x00dc, 0x00dc, 0x00db, 0x00db, 0x00db, 0x00da, 0x00da, 0x00da, 0x00d9, 0x00d9, 0x00d9, 0x00d8, 0x00d8, 0x00d7, 0x00d7,
+    0x00d7, 0x00d6, 0x00d6, 0x00d6, 0x00d5, 0x00d5, 0x00d5, 0x00d4, 0x00d4, 0x00d4, 0x00d3, 0x00d3, 0x00d3, 0x00d2, 0x00d2, 0x00d2,
+    0x00d1, 0x00d1, 0x00d1, 0x00d0, 0x00d0, 0x00d0, 0x00cf, 0x00cf, 0x00cf, 0x00ce, 0x00ce, 0x00ce, 0x00cd, 0x00cd, 0x00cd, 0x00cc,
+    0x00cc, 0x00cc, 0x00cb, 0x00cb, 0x00cb, 0x00ca, 0x00ca, 0x00ca, 0x00c9, 0x00c9, 0x00c9, 0x00c9, 0x00c8, 0x00c8, 0x00c8, 0x00c7,
+    0x00c7, 0x00c7, 0x00c6, 0x00c6, 0x00c6, 0x00c5, 0x00c5, 0x00c5, 0x00c5, 0x00c4, 0x00c4, 0x00c4, 0x00c3, 0x00c3, 0x00c3, 0x00c3,
+    0x00c2, 0x00c2, 0x00c2, 0x00c1, 0x00c1, 0x00c1, 0x00c1, 0x00c0, 0x00c0, 0x00c0, 0x00bf, 0x00bf, 0x00bf, 0x00bf, 0x00be, 0x00be,
+    0x00be, 0x00bd, 0x00bd, 0x00bd, 0x00bd, 0x00bc, 0x00bc, 0x00bc, 0x00bc, 0x00bb, 0x00bb, 0x00bb, 0x00ba, 0x00ba, 0x00ba, 0x00ba,
+    0x00b9, 0x00b9, 0x00b9, 0x00b9, 0x00b8, 0x00b8, 0x00b8, 0x00b8, 0x00b7, 0x00b7, 0x00b7, 0x00b7, 0x00b6, 0x00b6, 0x00b6, 0x00b6,
+    0x00b5, 0x00b5, 0x00b5, 0x00b5, 0x00b4, 0x00b4, 0x00b4, 0x00b4, 0x00b3, 0x00b3, 0x00b3, 0x00b3, 0x00b2, 0x00b2, 0x00b2, 0x00b2,
+    0x00b1, 0x00b1, 0x00b1, 0x00b1, 0x00b0, 0x00b0, 0x00b0, 0x00b0, 0x00af, 0x00af, 0x00af, 0x00af, 0x00ae, 0x00ae, 0x00ae, 0x00ae,
+    0x00ae, 0x00ad, 0x00ad, 0x00ad, 0x00ad, 0x00ac, 0x00ac, 0x00ac, 0x00ac, 0x00ac, 0x00ab, 0x00ab, 0x00ab, 0x00ab,
+};
+
+
+static tracy_force_inline uint64_t ProcessRGB( const uint8_t* src )
+{
+#ifdef __SSE4_1__
+    __m128i px0 = _mm_loadu_si128(((__m128i*)src) + 0);
+    __m128i px1 = _mm_loadu_si128(((__m128i*)src) + 1);
+    __m128i px2 = _mm_loadu_si128(((__m128i*)src) + 2);
+    __m128i px3 = _mm_loadu_si128(((__m128i*)src) + 3);
+
+    __m128i smask = _mm_set1_epi32( 0xF8FCF8 );
+    __m128i sd0 = _mm_and_si128( px0, smask );
+    __m128i sd1 = _mm_and_si128( px1, smask );
+    __m128i sd2 = _mm_and_si128( px2, smask );
+    __m128i sd3 = _mm_and_si128( px3, smask );
+
+    __m128i sc = _mm_shuffle_epi32(sd0, _MM_SHUFFLE(0, 0, 0, 0));
+
+    __m128i sc0 = _mm_cmpeq_epi8(sd0, sc);
+    __m128i sc1 = _mm_cmpeq_epi8(sd1, sc);
+    __m128i sc2 = _mm_cmpeq_epi8(sd2, sc);
+    __m128i sc3 = _mm_cmpeq_epi8(sd3, sc);
+
+    __m128i sm0 = _mm_and_si128(sc0, sc1);
+    __m128i sm1 = _mm_and_si128(sc2, sc3);
+    __m128i sm = _mm_and_si128(sm0, sm1);
+
+    if( _mm_testc_si128(sm, _mm_set1_epi32(-1)) )
+    {
+        return uint64_t( to565( src[0], src[1], src[2] ) ) << 16;
+    }
+
+    __m128i min0 = _mm_min_epu8( px0, px1 );
+    __m128i min1 = _mm_min_epu8( px2, px3 );
+    __m128i min2 = _mm_min_epu8( min0, min1 );
+
+    __m128i max0 = _mm_max_epu8( px0, px1 );
+    __m128i max1 = _mm_max_epu8( px2, px3 );
+    __m128i max2 = _mm_max_epu8( max0, max1 );
+
+    __m128i min3 = _mm_shuffle_epi32( min2, _MM_SHUFFLE( 2, 3, 0, 1 ) );
+    __m128i max3 = _mm_shuffle_epi32( max2, _MM_SHUFFLE( 2, 3, 0, 1 ) );
+    __m128i min4 = _mm_min_epu8( min2, min3 );
+    __m128i max4 = _mm_max_epu8( max2, max3 );
+
+    __m128i min5 = _mm_shuffle_epi32( min4, _MM_SHUFFLE( 0, 0, 2, 2 ) );
+    __m128i max5 = _mm_shuffle_epi32( max4, _MM_SHUFFLE( 0, 0, 2, 2 ) );
+    __m128i rmin = _mm_min_epu8( min4, min5 );
+    __m128i rmax = _mm_max_epu8( max4, max5 );
+
+    __m128i range1 = _mm_subs_epu8( rmax, rmin );
+    __m128i range2 = _mm_sad_epu8( rmax, rmin );
+
+    uint32_t vrange = _mm_cvtsi128_si32( range2 ) >> 1;
+    __m128i range = _mm_set1_epi16( DivTable[vrange] );
+
+    __m128i inset1 = _mm_srli_epi16( range1, 4 );
+    __m128i inset = _mm_and_si128( inset1, _mm_set1_epi8( 0xF ) );
+    __m128i min = _mm_adds_epu8( rmin, inset );
+    __m128i max = _mm_subs_epu8( rmax, inset );
+
+    __m128i c0 = _mm_subs_epu8( px0, rmin );
+    __m128i c1 = _mm_subs_epu8( px1, rmin );
+    __m128i c2 = _mm_subs_epu8( px2, rmin );
+    __m128i c3 = _mm_subs_epu8( px3, rmin );
+
+    __m128i is0 = _mm_maddubs_epi16( c0, _mm_set1_epi8( 1 ) );
+    __m128i is1 = _mm_maddubs_epi16( c1, _mm_set1_epi8( 1 ) );
+    __m128i is2 = _mm_maddubs_epi16( c2, _mm_set1_epi8( 1 ) );
+    __m128i is3 = _mm_maddubs_epi16( c3, _mm_set1_epi8( 1 ) );
+
+    __m128i s0 = _mm_hadd_epi16( is0, is1 );
+    __m128i s1 = _mm_hadd_epi16( is2, is3 );
+
+    __m128i m0 = _mm_mulhi_epu16( s0, range );
+    __m128i m1 = _mm_mulhi_epu16( s1, range );
+
+    __m128i p0 = _mm_packus_epi16( m0, m1 );
+
+    __m128i p1 = _mm_or_si128( _mm_srai_epi32( p0, 6 ), _mm_srai_epi32( p0, 12 ) );
+    __m128i p2 = _mm_or_si128( _mm_srai_epi32( p0, 18 ), p0 );
+    __m128i p3 = _mm_or_si128( p1, p2 );
+    __m128i p =_mm_shuffle_epi8( p3, _mm_set1_epi32( 0x0C080400 ) );
+
+    uint32_t vmin = _mm_cvtsi128_si32( min );
+    uint32_t vmax = _mm_cvtsi128_si32( max );
+    uint32_t vp = _mm_cvtsi128_si32( p );
+
+    return uint64_t( ( uint64_t( to565( vmin ) ) << 16 ) | to565( vmax ) | ( uint64_t( vp ) << 32 ) );
+#elif defined __ARM_NEON
+#  ifdef __aarch64__
+    uint8x16x4_t px = vld4q_u8( src );
+
+    uint8x16_t lr = px.val[0];
+    uint8x16_t lg = px.val[1];
+    uint8x16_t lb = px.val[2];
+
+    uint8_t rmaxr = vmaxvq_u8( lr );
+    uint8_t rmaxg = vmaxvq_u8( lg );
+    uint8_t rmaxb = vmaxvq_u8( lb );
+
+    uint8_t rminr = vminvq_u8( lr );
+    uint8_t rming = vminvq_u8( lg );
+    uint8_t rminb = vminvq_u8( lb );
+
+    int rr = rmaxr - rminr;
+    int rg = rmaxg - rming;
+    int rb = rmaxb - rminb;
+
+    int vrange1 = rr + rg + rb;
+    uint16_t vrange2 = DivTableNEON[vrange1];
+
+    uint8_t insetr = rr >> 4;
+    uint8_t insetg = rg >> 4;
+    uint8_t insetb = rb >> 4;
+
+    uint8_t minr = rminr + insetr;
+    uint8_t ming = rming + insetg;
+    uint8_t minb = rminb + insetb;
+
+    uint8_t maxr = rmaxr - insetr;
+    uint8_t maxg = rmaxg - insetg;
+    uint8_t maxb = rmaxb - insetb;
+
+    uint8x16_t cr = vsubq_u8( lr, vdupq_n_u8( rminr ) );
+    uint8x16_t cg = vsubq_u8( lg, vdupq_n_u8( rming ) );
+    uint8x16_t cb = vsubq_u8( lb, vdupq_n_u8( rminb ) );
+
+    uint16x8_t is0l = vaddl_u8( vget_low_u8( cr ), vget_low_u8( cg ) );
+    uint16x8_t is0h = vaddl_u8( vget_high_u8( cr ), vget_high_u8( cg ) );
+    uint16x8_t is1l = vaddw_u8( is0l, vget_low_u8( cb ) );
+    uint16x8_t is1h = vaddw_u8( is0h, vget_high_u8( cb ) );
+
+    int16x8_t range = vdupq_n_s16( vrange2 );
+    uint16x8_t m0 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( is1l ), range ) );
+    uint16x8_t m1 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( is1h ), range ) );
+
+    uint8x8_t p00 = vmovn_u16( m0 );
+    uint8x8_t p01 = vmovn_u16( m1 );
+    uint8x16_t p0 = vcombine_u8( p00, p01 );
+
+    uint32x4_t p1 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 6 ), vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 12 ) );
+    uint32x4_t p2 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 18 ), vreinterpretq_u32_u8( p0 ) );
+    uint32x4_t p3 = vaddq_u32( p1, p2 );
+
+    uint16x4x2_t p4 = vuzp_u16( vget_low_u16( vreinterpretq_u16_u32( p3 ) ), vget_high_u16( vreinterpretq_u16_u32( p3 ) ) );
+    uint8x8x2_t p = vuzp_u8( vreinterpret_u8_u16( p4.val[0] ), vreinterpret_u8_u16( p4.val[0] ) );
+
+    uint32_t vp;
+    vst1_lane_u32( &vp, vreinterpret_u32_u8( p.val[0] ), 0 );
+
+    return uint64_t( ( uint64_t( to565( minr, ming, minb ) ) << 16 ) | to565( maxr, maxg, maxb ) | ( uint64_t( vp ) << 32 ) );
+#  else
+    uint32x4_t px0 = vld1q_u32( (uint32_t*)src );
+    uint32x4_t px1 = vld1q_u32( (uint32_t*)src + 4 );
+    uint32x4_t px2 = vld1q_u32( (uint32_t*)src + 8 );
+    uint32x4_t px3 = vld1q_u32( (uint32_t*)src + 12 );
+
+    uint32x4_t smask = vdupq_n_u32( 0xF8FCF8 );
+    uint32x4_t sd0 = vandq_u32( smask, px0 );
+    uint32x4_t sd1 = vandq_u32( smask, px1 );
+    uint32x4_t sd2 = vandq_u32( smask, px2 );
+    uint32x4_t sd3 = vandq_u32( smask, px3 );
+
+    uint32x4_t sc = vdupq_n_u32( sd0[0] );
+
+    uint32x4_t sc0 = vceqq_u32( sd0, sc );
+    uint32x4_t sc1 = vceqq_u32( sd1, sc );
+    uint32x4_t sc2 = vceqq_u32( sd2, sc );
+    uint32x4_t sc3 = vceqq_u32( sd3, sc );
+
+    uint32x4_t sm0 = vandq_u32( sc0, sc1 );
+    uint32x4_t sm1 = vandq_u32( sc2, sc3 );
+    int64x2_t sm = vreinterpretq_s64_u32( vandq_u32( sm0, sm1 ) );
+
+    if( sm[0] == -1 && sm[1] == -1 )
+    {
+        return uint64_t( to565( src[0], src[1], src[2] ) ) << 16;
+    }
+
+    uint32x4_t mask = vdupq_n_u32( 0xFFFFFF );
+    uint8x16_t l0 = vreinterpretq_u8_u32( vandq_u32( mask, px0 ) );
+    uint8x16_t l1 = vreinterpretq_u8_u32( vandq_u32( mask, px1 ) );
+    uint8x16_t l2 = vreinterpretq_u8_u32( vandq_u32( mask, px2 ) );
+    uint8x16_t l3 = vreinterpretq_u8_u32( vandq_u32( mask, px3 ) );
+
+    uint8x16_t min0 = vminq_u8( l0, l1 );
+    uint8x16_t min1 = vminq_u8( l2, l3 );
+    uint8x16_t min2 = vminq_u8( min0, min1 );
+
+    uint8x16_t max0 = vmaxq_u8( l0, l1 );
+    uint8x16_t max1 = vmaxq_u8( l2, l3 );
+    uint8x16_t max2 = vmaxq_u8( max0, max1 );
+
+    uint8x16_t min3 = vreinterpretq_u8_u32( vrev64q_u32( vreinterpretq_u32_u8( min2 ) ) );
+    uint8x16_t max3 = vreinterpretq_u8_u32( vrev64q_u32( vreinterpretq_u32_u8( max2 ) ) );
+
+    uint8x16_t min4 = vminq_u8( min2, min3 );
+    uint8x16_t max4 = vmaxq_u8( max2, max3 );
+
+    uint8x16_t min5 = vcombine_u8( vget_high_u8( min4 ), vget_low_u8( min4 ) );
+    uint8x16_t max5 = vcombine_u8( vget_high_u8( max4 ), vget_low_u8( max4 ) );
+
+    uint8x16_t rmin = vminq_u8( min4, min5 );
+    uint8x16_t rmax = vmaxq_u8( max4, max5 );
+
+    uint8x16_t range1 = vsubq_u8( rmax, rmin );
+    uint8x8_t range2 = vget_low_u8( range1 );
+    uint8x8x2_t range3 = vzip_u8( range2, vdup_n_u8( 0 ) );
+    uint16x4_t range4 = vreinterpret_u16_u8( range3.val[0] );
+
+    uint16_t vrange1;
+    uint16x4_t range5 = vpadd_u16( range4, range4 );
+    uint16x4_t range6 = vpadd_u16( range5, range5 );
+    vst1_lane_u16( &vrange1, range6, 0 );
+
+    uint32_t vrange2 = ( 2 << 16 ) / uint32_t( vrange1 + 1 );
+    uint16x8_t range = vdupq_n_u16( vrange2 );
+
+    uint8x16_t inset = vshrq_n_u8( range1, 4 );
+    uint8x16_t min = vaddq_u8( rmin, inset );
+    uint8x16_t max = vsubq_u8( rmax, inset );
+
+    uint8x16_t c0 = vsubq_u8( l0, rmin );
+    uint8x16_t c1 = vsubq_u8( l1, rmin );
+    uint8x16_t c2 = vsubq_u8( l2, rmin );
+    uint8x16_t c3 = vsubq_u8( l3, rmin );
+
+    uint16x8_t is0 = vpaddlq_u8( c0 );
+    uint16x8_t is1 = vpaddlq_u8( c1 );
+    uint16x8_t is2 = vpaddlq_u8( c2 );
+    uint16x8_t is3 = vpaddlq_u8( c3 );
+
+    uint16x4_t is4 = vpadd_u16( vget_low_u16( is0 ), vget_high_u16( is0 ) );
+    uint16x4_t is5 = vpadd_u16( vget_low_u16( is1 ), vget_high_u16( is1 ) );
+    uint16x4_t is6 = vpadd_u16( vget_low_u16( is2 ), vget_high_u16( is2 ) );
+    uint16x4_t is7 = vpadd_u16( vget_low_u16( is3 ), vget_high_u16( is3 ) );
+
+    uint16x8_t s0 = vcombine_u16( is4, is5 );
+    uint16x8_t s1 = vcombine_u16( is6, is7 );
+
+    uint16x8_t m0 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( s0 ), vreinterpretq_s16_u16( range ) ) );
+    uint16x8_t m1 = vreinterpretq_u16_s16( vqdmulhq_s16( vreinterpretq_s16_u16( s1 ), vreinterpretq_s16_u16( range ) ) );
+
+    uint8x8_t p00 = vmovn_u16( m0 );
+    uint8x8_t p01 = vmovn_u16( m1 );
+    uint8x16_t p0 = vcombine_u8( p00, p01 );
+
+    uint32x4_t p1 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 6 ), vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 12 ) );
+    uint32x4_t p2 = vaddq_u32( vshrq_n_u32( vreinterpretq_u32_u8( p0 ), 18 ), vreinterpretq_u32_u8( p0 ) );
+    uint32x4_t p3 = vaddq_u32( p1, p2 );
+
+    uint16x4x2_t p4 = vuzp_u16( vget_low_u16( vreinterpretq_u16_u32( p3 ) ), vget_high_u16( vreinterpretq_u16_u32( p3 ) ) );
+    uint8x8x2_t p = vuzp_u8( vreinterpret_u8_u16( p4.val[0] ), vreinterpret_u8_u16( p4.val[0] ) );
+
+    uint32_t vmin, vmax, vp;
+    vst1q_lane_u32( &vmin, vreinterpretq_u32_u8( min ), 0 );
+    vst1q_lane_u32( &vmax, vreinterpretq_u32_u8( max ), 0 );
+    vst1_lane_u32( &vp, vreinterpret_u32_u8( p.val[0] ), 0 );
+
+    return uint64_t( ( uint64_t( to565( vmin ) ) << 16 ) | to565( vmax ) | ( uint64_t( vp ) << 32 ) );
+#  endif
+#else
+    uint32_t ref;
+    memcpy( &ref, src, 4 );
+    uint32_t refMask = ref & 0xF8FCF8;
+    auto stmp = src + 4;
+    for( int i=1; i<16; i++ )
+    {
+        uint32_t px;
+        memcpy( &px, stmp, 4 );
+        if( ( px & 0xF8FCF8 ) != refMask ) break;
+        stmp += 4;
+    }
+    if( stmp == src + 64 )
+    {
+        return uint64_t( to565( ref ) ) << 16;
+    }
+
+    uint8_t min[3] = { src[0], src[1], src[2] };
+    uint8_t max[3] = { src[0], src[1], src[2] };
+    auto tmp = src + 4;
+    for( int i=1; i<16; i++ )
+    {
+        for( int j=0; j<3; j++ )
+        {
+            if( tmp[j] < min[j] ) min[j] = tmp[j];
+            else if( tmp[j] > max[j] ) max[j] = tmp[j];
+        }
+        tmp += 4;
+    }
+
+    const uint32_t range = DivTable[max[0] - min[0] + max[1] - min[1] + max[2] - min[2]];
+    const uint32_t rmin = min[0] + min[1] + min[2];
+    for( int i=0; i<3; i++ )
+    {
+        const uint8_t inset = ( max[i] - min[i] ) >> 4;
+        min[i] += inset;
+        max[i] -= inset;
+    }
+
+    uint32_t data = 0;
+    for( int i=0; i<16; i++ )
+    {
+        const uint32_t c = src[0] + src[1] + src[2] - rmin;
+        const uint8_t idx = ( c * range ) >> 16;
+        data |= idx << (i*2);
+        src += 4;
+    }
+
+    return uint64_t( ( uint64_t( to565( min[0], min[1], min[2] ) ) << 16 ) | to565( max[0], max[1], max[2] ) | ( uint64_t( data ) << 32 ) );
+#endif
+}
+
+#ifdef __AVX2__
+static tracy_force_inline void ProcessRGB_AVX( const uint8_t* src, char*& dst )
+{
+    __m256i px0 = _mm256_loadu_si256(((__m256i*)src) + 0);
+    __m256i px1 = _mm256_loadu_si256(((__m256i*)src) + 1);
+    __m256i px2 = _mm256_loadu_si256(((__m256i*)src) + 2);
+    __m256i px3 = _mm256_loadu_si256(((__m256i*)src) + 3);
+
+    __m256i smask = _mm256_set1_epi32( 0xF8FCF8 );
+    __m256i sd0 = _mm256_and_si256( px0, smask );
+    __m256i sd1 = _mm256_and_si256( px1, smask );
+    __m256i sd2 = _mm256_and_si256( px2, smask );
+    __m256i sd3 = _mm256_and_si256( px3, smask );
+
+    __m256i sc = _mm256_shuffle_epi32(sd0, _MM_SHUFFLE(0, 0, 0, 0));
+
+    __m256i sc0 = _mm256_cmpeq_epi8( sd0, sc );
+    __m256i sc1 = _mm256_cmpeq_epi8( sd1, sc );
+    __m256i sc2 = _mm256_cmpeq_epi8( sd2, sc );
+    __m256i sc3 = _mm256_cmpeq_epi8( sd3, sc );
+
+    __m256i sm0 = _mm256_and_si256( sc0, sc1 );
+    __m256i sm1 = _mm256_and_si256( sc2, sc3 );
+    __m256i sm = _mm256_and_si256( sm0, sm1 );
+
+    const int64_t solid0 = 1 - _mm_testc_si128( _mm256_castsi256_si128( sm ), _mm_set1_epi32( -1 ) );
+    const int64_t solid1 = 1 - _mm_testc_si128( _mm256_extracti128_si256( sm, 1 ), _mm_set1_epi32( -1 ) );
+
+    if( solid0 + solid1 == 0 )
+    {
+        const auto c0 = uint64_t( to565( src[0], src[1], src[2] ) ) << 16;
+        const auto c1 = uint64_t( to565( src[16], src[17], src[18] ) ) << 16;
+        memcpy( dst, &c0, 8 );
+        memcpy( dst+8, &c1, 8 );
+        dst += 16;
+        return;
+    }
+
+    __m256i min0 = _mm256_min_epu8( px0, px1 );
+    __m256i min1 = _mm256_min_epu8( px2, px3 );
+    __m256i min2 = _mm256_min_epu8( min0, min1 );
+
+    __m256i max0 = _mm256_max_epu8( px0, px1 );
+    __m256i max1 = _mm256_max_epu8( px2, px3 );
+    __m256i max2 = _mm256_max_epu8( max0, max1 );
+
+    __m256i min3 = _mm256_shuffle_epi32( min2, _MM_SHUFFLE( 2, 3, 0, 1 ) );
+    __m256i max3 = _mm256_shuffle_epi32( max2, _MM_SHUFFLE( 2, 3, 0, 1 ) );
+    __m256i min4 = _mm256_min_epu8( min2, min3 );
+    __m256i max4 = _mm256_max_epu8( max2, max3 );
+
+    __m256i min5 = _mm256_shuffle_epi32( min4, _MM_SHUFFLE( 0, 0, 2, 2 ) );
+    __m256i max5 = _mm256_shuffle_epi32( max4, _MM_SHUFFLE( 0, 0, 2, 2 ) );
+    __m256i rmin = _mm256_min_epu8( min4, min5 );
+    __m256i rmax = _mm256_max_epu8( max4, max5 );
+
+    __m256i range1 = _mm256_subs_epu8( rmax, rmin );
+    __m256i range2 = _mm256_sad_epu8( rmax, rmin );
+
+    uint16_t vrange0 = DivTable[_mm256_cvtsi256_si32( range2 ) >> 1];
+    uint16_t vrange1 = DivTable[_mm256_extract_epi16( range2, 8 ) >> 1];
+    __m256i range00 = _mm256_set1_epi16( vrange0 );
+    __m256i range = _mm256_inserti128_si256( range00, _mm_set1_epi16( vrange1 ), 1 );
+
+    __m256i inset1 = _mm256_srli_epi16( range1, 4 );
+    __m256i inset = _mm256_and_si256( inset1, _mm256_set1_epi8( 0xF ) );
+    __m256i min = _mm256_adds_epu8( rmin, inset );
+    __m256i max = _mm256_subs_epu8( rmax, inset );
+
+    __m256i c0 = _mm256_subs_epu8( px0, rmin );
+    __m256i c1 = _mm256_subs_epu8( px1, rmin );
+    __m256i c2 = _mm256_subs_epu8( px2, rmin );
+    __m256i c3 = _mm256_subs_epu8( px3, rmin );
+
+    __m256i is0 = _mm256_maddubs_epi16( c0, _mm256_set1_epi8( 1 ) );
+    __m256i is1 = _mm256_maddubs_epi16( c1, _mm256_set1_epi8( 1 ) );
+    __m256i is2 = _mm256_maddubs_epi16( c2, _mm256_set1_epi8( 1 ) );
+    __m256i is3 = _mm256_maddubs_epi16( c3, _mm256_set1_epi8( 1 ) );
+
+    __m256i s0 = _mm256_hadd_epi16( is0, is1 );
+    __m256i s1 = _mm256_hadd_epi16( is2, is3 );
+
+    __m256i m0 = _mm256_mulhi_epu16( s0, range );
+    __m256i m1 = _mm256_mulhi_epu16( s1, range );
+
+    __m256i p0 = _mm256_packus_epi16( m0, m1 );
+
+    __m256i p1 = _mm256_or_si256( _mm256_srai_epi32( p0, 6 ), _mm256_srai_epi32( p0, 12 ) );
+    __m256i p2 = _mm256_or_si256( _mm256_srai_epi32( p0, 18 ), p0 );
+    __m256i p3 = _mm256_or_si256( p1, p2 );
+    __m256i p =_mm256_shuffle_epi8( p3, _mm256_set1_epi32( 0x0C080400 ) );
+
+    __m256i mm0 = _mm256_unpacklo_epi8( _mm256_setzero_si256(), min );
+    __m256i mm1 = _mm256_unpacklo_epi8( _mm256_setzero_si256(), max );
+    __m256i mm2 = _mm256_unpacklo_epi64( mm1, mm0 );
+    __m256i mmr = _mm256_slli_epi64( _mm256_srli_epi64( mm2, 11 ), 11 );
+    __m256i mmg = _mm256_slli_epi64( _mm256_srli_epi64( mm2, 26 ), 5 );
+    __m256i mmb = _mm256_srli_epi64( _mm256_slli_epi64( mm2, 16 ), 59 );
+    __m256i mm3 = _mm256_or_si256( mmr, mmg );
+    __m256i mm4 = _mm256_or_si256( mm3, mmb );
+    __m256i mm5 = _mm256_shuffle_epi8( mm4, _mm256_set1_epi32( 0x09080100 ) );
+
+    __m256i d0 = _mm256_unpacklo_epi32( mm5, p );
+    __m256i d1 = _mm256_permute4x64_epi64( d0, _MM_SHUFFLE( 3, 2, 2, 0 ) );
+    __m128i d2 = _mm256_castsi256_si128( d1 );
+
+    __m128i mask = _mm_set_epi64x( 0xFFFF0000 | -solid1, 0xFFFF0000 | -solid0 );
+    __m128i d3 = _mm_and_si128( d2, mask );
+    _mm_storeu_si128( (__m128i*)dst, d3 );
+    dst += 16;
+}
+#endif
+
+void CompressImageDxt1( const char* src, char* dst, int w, int h )
+{
+    assert( (w % 4) == 0 && (h % 4) == 0 );
+
+#ifdef __AVX2__
+    if( w%8 == 0 )
+    {
+        uint32_t buf[8*4];
+        int i = 0;
+
+        auto blocks = w * h / 32;
+        do
+        {
+            auto tmp = (char*)buf;
+            memcpy( tmp,        src,          8*4 );
+            memcpy( tmp + 8*4,  src + w * 4,  8*4 );
+            memcpy( tmp + 16*4, src + w * 8,  8*4 );
+            memcpy( tmp + 24*4, src + w * 12, 8*4 );
+            src += 8*4;
+            if( ++i == w/8 )
+            {
+                src += w * 3 * 4;
+                i = 0;
+            }
+
+            ProcessRGB_AVX( (uint8_t*)buf, dst );
+        }
+        while( --blocks );
+    }
+    else
+#endif
+    {
+        uint32_t buf[4*4];
+        int i = 0;
+
+        auto ptr = dst;
+        auto blocks = w * h / 16;
+        do
+        {
+            auto tmp = (char*)buf;
+            memcpy( tmp,        src,          4*4 );
+            memcpy( tmp + 4*4,  src + w * 4,  4*4 );
+            memcpy( tmp + 8*4,  src + w * 8,  4*4 );
+            memcpy( tmp + 12*4, src + w * 12, 4*4 );
+            src += 4*4;
+            if( ++i == w/4 )
+            {
+                src += w * 3 * 4;
+                i = 0;
+            }
+
+            const auto c = ProcessRGB( (uint8_t*)buf );
+            memcpy( ptr, &c, sizeof( uint64_t ) );
+            ptr += sizeof( uint64_t );
+        }
+        while( --blocks );
+    }
+}
+
+}

client/TracyDxt1.hpp

@@ -0,0 +1,11 @@
+#ifndef __TRACYDXT1_HPP__
+#define __TRACYDXT1_HPP__
+
+namespace tracy
+{
+
+void CompressImageDxt1( const char* src, char* dst, int w, int h );
+
+}
+
+#endif

client/TracyFastVector.hpp

@@ -1,6 +1,7 @@
 #ifndef __TRACYFASTVECTOR_HPP__
 #define __TRACYFASTVECTOR_HPP__
 
+#include <assert.h>
 #include <stddef.h>
 
 #include "../common/TracyAlloc.hpp"
@@ -21,6 +22,7 @@ public:
         , m_write( m_ptr )
         , m_end( m_ptr + capacity )
     {
+        assert( capacity != 0 );
     }
 
     FastVector( const FastVector& ) = delete;

client/TracyLock.hpp

@@ -11,207 +11,224 @@
 namespace tracy
 {
 
-extern std::atomic<uint32_t> s_lockCounter;
+class LockableCtx
+{
+public:
+    tracy_force_inline LockableCtx( const SourceLocationData* srcloc )
+        : m_id( GetLockCounter().fetch_add( 1, std::memory_order_relaxed ) )
+#ifdef TRACY_ON_DEMAND
+        , m_lockCount( 0 )
+        , m_active( false )
+#endif
+    {
+        assert( m_id != std::numeric_limits<uint32_t>::max() );
+
+        TracyLfqPrepare( QueueType::LockAnnounce );
+        MemWrite( &item->lockAnnounce.id, m_id );
+        MemWrite( &item->lockAnnounce.time, Profiler::GetTime() );
+        MemWrite( &item->lockAnnounce.lckloc, (uint64_t)srcloc );
+        MemWrite( &item->lockAnnounce.type, LockType::Lockable );
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+        TracyLfqCommit;
+    }
+
+    LockableCtx( const LockableCtx& ) = delete;
+    LockableCtx& operator=( const LockableCtx& ) = delete;
+
+    tracy_force_inline ~LockableCtx()
+    {
+        TracyLfqPrepare( QueueType::LockTerminate );
+        MemWrite( &item->lockTerminate.id, m_id );
+        MemWrite( &item->lockTerminate.time, Profiler::GetTime() );
+        MemWrite( &item->lockTerminate.type, LockType::Lockable );
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+        TracyLfqCommit;
+    }
+
+    tracy_force_inline bool BeforeLock()
+    {
+#ifdef TRACY_ON_DEMAND
+        bool queue = false;
+        const auto locks = m_lockCount.fetch_add( 1, std::memory_order_relaxed );
+        const auto active = m_active.load( std::memory_order_relaxed );
+        if( locks == 0 || active )
+        {
+            const bool connected = GetProfiler().IsConnected();
+            if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
+            if( connected ) queue = true;
+        }
+        if( !queue ) return false;
+#endif
+
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockWait );
+        MemWrite( &item->lockWait.thread, GetThreadHandle() );
+        MemWrite( &item->lockWait.id, m_id );
+        MemWrite( &item->lockWait.time, Profiler::GetTime() );
+        MemWrite( &item->lockWait.type, LockType::Lockable );
+        Profiler::QueueSerialFinish();
+        return true;
+    }
+
+    tracy_force_inline void AfterLock()
+    {
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockObtain );
+        MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+        MemWrite( &item->lockObtain.id, m_id );
+        MemWrite( &item->lockObtain.time, Profiler::GetTime() );
+        Profiler::QueueSerialFinish();
+    }
+
+    tracy_force_inline void AfterUnlock()
+    {
+#ifdef TRACY_ON_DEMAND
+        m_lockCount.fetch_sub( 1, std::memory_order_relaxed );
+        if( !m_active.load( std::memory_order_relaxed ) ) return;
+        if( !GetProfiler().IsConnected() )
+        {
+            m_active.store( false, std::memory_order_relaxed );
+            return;
+        }
+#endif
+
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockRelease );
+        MemWrite( &item->lockRelease.thread, GetThreadHandle() );
+        MemWrite( &item->lockRelease.id, m_id );
+        MemWrite( &item->lockRelease.time, Profiler::GetTime() );
+        Profiler::QueueSerialFinish();
+    }
+
+    tracy_force_inline void AfterTryLock( bool acquired )
+    {
+#ifdef TRACY_ON_DEMAND
+        if( !acquired ) return;
+
+        bool queue = false;
+        const auto locks = m_lockCount.fetch_add( 1, std::memory_order_relaxed );
+        const auto active = m_active.load( std::memory_order_relaxed );
+        if( locks == 0 || active )
+        {
+            const bool connected = GetProfiler().IsConnected();
+            if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
+            if( connected ) queue = true;
+        }
+        if( !queue ) return;
+#endif
+
+        if( acquired )
+        {
+            auto item = Profiler::QueueSerial();
+            MemWrite( &item->hdr.type, QueueType::LockObtain );
+            MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+            MemWrite( &item->lockObtain.id, m_id );
+            MemWrite( &item->lockObtain.time, Profiler::GetTime() );
+            Profiler::QueueSerialFinish();
+        }
+    }
+
+    tracy_force_inline void Mark( const SourceLocationData* srcloc )
+    {
+#ifdef TRACY_ON_DEMAND
+        const auto active = m_active.load( std::memory_order_relaxed );
+        if( !active ) return;
+        const auto connected = GetProfiler().IsConnected();
+        if( !connected )
+        {
+            if( active ) m_active.store( false, std::memory_order_relaxed );
+            return;
+        }
+#endif
+
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockMark );
+        MemWrite( &item->lockMark.thread, GetThreadHandle() );
+        MemWrite( &item->lockMark.id, m_id );
+        MemWrite( &item->lockMark.srcloc, (uint64_t)srcloc );
+        Profiler::QueueSerialFinish();
+    }
+
+    tracy_force_inline void CustomName( const char* name, size_t size )
+    {
+        auto ptr = (char*)tracy_malloc( size+1 );
+        memcpy( ptr, name, size );
+        ptr[size] = '\0';
+        TracyLfqPrepare( QueueType::LockName );
+        MemWrite( &item->lockName.id, m_id );
+        MemWrite( &item->lockName.name, (uint64_t)ptr );
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+        TracyLfqCommit;
+    }
+
+private:
+    uint32_t m_id;
+
+#ifdef TRACY_ON_DEMAND
+    std::atomic<uint32_t> m_lockCount;
+    std::atomic<bool> m_active;
+#endif
+};
 
 template<class T>
 class Lockable
 {
 public:
     tracy_force_inline Lockable( const SourceLocationData* srcloc )
-        : m_id( s_lockCounter.fetch_add( 1, std::memory_order_relaxed ) )
-#ifdef TRACY_ON_DEMAND
-        , m_lockCount( 0 )
-        , m_active( false )
-#endif
+        : m_ctx( srcloc )
     {
-        assert( m_id != std::numeric_limits<uint32_t>::max() );
-
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::LockAnnounce );
-        MemWrite( &item->lockAnnounce.id, m_id );
-        MemWrite( &item->lockAnnounce.time, Profiler::GetTime() );
-        MemWrite( &item->lockAnnounce.lckloc, (uint64_t)srcloc );
-        MemWrite( &item->lockAnnounce.type, LockType::Lockable );
-
-#ifdef TRACY_ON_DEMAND
-        s_profiler.DeferItem( *item );
-#endif
-
-        tail.store( magic + 1, std::memory_order_release );
     }
 
     Lockable( const Lockable& ) = delete;
     Lockable& operator=( const Lockable& ) = delete;
 
-    ~Lockable()
-    {
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::LockTerminate );
-        MemWrite( &item->lockTerminate.id, m_id );
-        MemWrite( &item->lockTerminate.time, Profiler::GetTime() );
-        MemWrite( &item->lockTerminate.type, LockType::Lockable );
-
-#ifdef TRACY_ON_DEMAND
-        s_profiler.DeferItem( *item );
-#endif
-
-        tail.store( magic + 1, std::memory_order_release );
-    }
-
     tracy_force_inline void lock()
     {
-#ifdef TRACY_ON_DEMAND
-        bool queue = false;
-        const auto locks = m_lockCount.fetch_add( 1, std::memory_order_relaxed );
-        const auto active = m_active.load( std::memory_order_relaxed );
-        if( locks == 0 || active )
-        {
-            const bool connected = s_profiler.IsConnected();
-            if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
-            if( connected ) queue = true;
-        }
-        if( !queue )
-        {
-            m_lockable.lock();
-            return;
-        }
-#endif
-        const auto thread = GetThreadHandle();
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockWait );
-            MemWrite( &item->lockWait.id, m_id );
-            MemWrite( &item->lockWait.thread, thread );
-            MemWrite( &item->lockWait.time, Profiler::GetTime() );
-            MemWrite( &item->lockWait.type, LockType::Lockable );
-            tail.store( magic + 1, std::memory_order_release );
-        }
-
+        const auto runAfter = m_ctx.BeforeLock();
         m_lockable.lock();
-
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockObtain );
-            MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, thread );
-            MemWrite( &item->lockObtain.time, Profiler::GetTime() );
-            tail.store( magic + 1, std::memory_order_release );
-        }
+        if( runAfter ) m_ctx.AfterLock();
     }
 
     tracy_force_inline void unlock()
     {
         m_lockable.unlock();
-
-#ifdef TRACY_ON_DEMAND
-        m_lockCount.fetch_sub( 1, std::memory_order_relaxed );
-        if( !m_active.load( std::memory_order_relaxed ) ) return;
-        if( !s_profiler.IsConnected() )
-        {
-            m_active.store( false, std::memory_order_relaxed );
-            return;
-        }
-#endif
-
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::LockRelease );
-        MemWrite( &item->lockRelease.id, m_id );
-        MemWrite( &item->lockRelease.thread, GetThreadHandle() );
-        MemWrite( &item->lockRelease.time, Profiler::GetTime() );
-        tail.store( magic + 1, std::memory_order_release );
+        m_ctx.AfterUnlock();
     }
 
     tracy_force_inline bool try_lock()
     {
-        const auto ret = m_lockable.try_lock();
-
-#ifdef TRACY_ON_DEMAND
-        if( !ret ) return ret;
-
-        bool queue = false;
-        const auto locks = m_lockCount.fetch_add( 1, std::memory_order_relaxed );
-        const auto active = m_active.load( std::memory_order_relaxed );
-        if( locks == 0 || active )
-        {
-            const bool connected = s_profiler.IsConnected();
-            if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
-            if( connected ) queue = true;
-        }
-        if( !queue ) return ret;
-#endif
-
-        if( ret )
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockObtain );
-            MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, GetThreadHandle() );
-            MemWrite( &item->lockObtain.time, Profiler::GetTime() );
-            tail.store( magic + 1, std::memory_order_release );
-        }
-
-        return ret;
+        const auto acquired = m_lockable.try_lock();
+        m_ctx.AfterTryLock( acquired );
+        return acquired;
     }
 
     tracy_force_inline void Mark( const SourceLocationData* srcloc )
     {
-#ifdef TRACY_ON_DEMAND
-        const auto active = m_active.load( std::memory_order_relaxed );
-        if( !active ) return;
-        const auto connected = s_profiler.IsConnected();
-        if( !connected )
-        {
-            if( active ) m_active.store( false, std::memory_order_relaxed );
-            return;
-        }
-#endif
+        m_ctx.Mark( srcloc );
+    }
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::LockMark );
-        MemWrite( &item->lockMark.id, m_id );
-        MemWrite( &item->lockMark.thread, GetThreadHandle() );
-        MemWrite( &item->lockMark.srcloc, (uint64_t)srcloc );
-        tail.store( magic + 1, std::memory_order_release );
+    tracy_force_inline void CustomName( const char* name, size_t size )
+    {
+        m_ctx.CustomName( name, size );
     }
 
 private:
     T m_lockable;
-    uint32_t m_id;
-
-#ifdef TRACY_ON_DEMAND
-    std::atomic<uint32_t> m_lockCount;
-    std::atomic<bool> m_active;
-#endif
+    LockableCtx m_ctx;
 };
 
 
-template<class T>
-class SharedLockable
+class SharedLockableCtx
 {
 public:
-    tracy_force_inline SharedLockable( const SourceLocationData* srcloc )
-        : m_id( s_lockCounter.fetch_add( 1, std::memory_order_relaxed ) )
+    tracy_force_inline SharedLockableCtx( const SourceLocationData* srcloc )
+        : m_id( GetLockCounter().fetch_add( 1, std::memory_order_relaxed ) )
 #ifdef TRACY_ON_DEMAND
         , m_lockCount( 0 )
         , m_active( false )
@@ -219,45 +236,37 @@ public:
     {
         assert( m_id != std::numeric_limits<uint32_t>::max() );
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::LockAnnounce );
+        TracyLfqPrepare( QueueType::LockAnnounce );
         MemWrite( &item->lockAnnounce.id, m_id );
         MemWrite( &item->lockAnnounce.time, Profiler::GetTime() );
         MemWrite( &item->lockAnnounce.lckloc, (uint64_t)srcloc );
         MemWrite( &item->lockAnnounce.type, LockType::SharedLockable );
 
 #ifdef TRACY_ON_DEMAND
-        s_profiler.DeferItem( *item );
+        GetProfiler().DeferItem( *item );
 #endif
 
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
-    SharedLockable( const SharedLockable& ) = delete;
-    SharedLockable& operator=( const SharedLockable& ) = delete;
+    SharedLockableCtx( const SharedLockableCtx& ) = delete;
+    SharedLockableCtx& operator=( const SharedLockableCtx& ) = delete;
 
-    ~SharedLockable()
+    tracy_force_inline ~SharedLockableCtx()
     {
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::LockTerminate );
+        TracyLfqPrepare( QueueType::LockTerminate );
         MemWrite( &item->lockTerminate.id, m_id );
         MemWrite( &item->lockTerminate.time, Profiler::GetTime() );
         MemWrite( &item->lockTerminate.type, LockType::SharedLockable );
 
 #ifdef TRACY_ON_DEMAND
-        s_profiler.DeferItem( *item );
+        GetProfiler().DeferItem( *item );
 #endif
 
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
-    tracy_force_inline void lock()
+    tracy_force_inline bool BeforeLock()
     {
 #ifdef TRACY_ON_DEMAND
         bool queue = false;
@@ -265,106 +274,82 @@ public:
         const auto active = m_active.load( std::memory_order_relaxed );
         if( locks == 0 || active )
         {
-            const bool connected = s_profiler.IsConnected();
+            const bool connected = GetProfiler().IsConnected();
             if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
             if( connected ) queue = true;
         }
-        if( !queue )
-        {
-            m_lockable.lock();
-            return;
-        }
+        if( !queue ) return false;
 #endif
-        const auto thread = GetThreadHandle();
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockWait );
-            MemWrite( &item->lockWait.id, m_id );
-            MemWrite( &item->lockWait.thread, thread );
-            MemWrite( &item->lockWait.time, Profiler::GetTime() );
-            MemWrite( &item->lockWait.type, LockType::SharedLockable );
-            tail.store( magic + 1, std::memory_order_release );
-        }
 
-        m_lockable.lock();
-
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockObtain );
-            MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, thread );
-            MemWrite( &item->lockObtain.time, Profiler::GetTime() );
-            tail.store( magic + 1, std::memory_order_release );
-        }
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockWait );
+        MemWrite( &item->lockWait.thread, GetThreadHandle() );
+        MemWrite( &item->lockWait.id, m_id );
+        MemWrite( &item->lockWait.time, Profiler::GetTime() );
+        MemWrite( &item->lockWait.type, LockType::SharedLockable );
+        Profiler::QueueSerialFinish();
+        return true;
     }
 
-    tracy_force_inline void unlock()
+    tracy_force_inline void AfterLock()
     {
-        m_lockable.unlock();
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockObtain );
+        MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+        MemWrite( &item->lockObtain.id, m_id );
+        MemWrite( &item->lockObtain.time, Profiler::GetTime() );
+        Profiler::QueueSerialFinish();
+    }
 
+    tracy_force_inline void AfterUnlock()
+    {
 #ifdef TRACY_ON_DEMAND
         m_lockCount.fetch_sub( 1, std::memory_order_relaxed );
         if( !m_active.load( std::memory_order_relaxed ) ) return;
-        if( !s_profiler.IsConnected() )
+        if( !GetProfiler().IsConnected() )
         {
             m_active.store( false, std::memory_order_relaxed );
             return;
         }
 #endif
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+        auto item = Profiler::QueueSerial();
         MemWrite( &item->hdr.type, QueueType::LockRelease );
-        MemWrite( &item->lockRelease.id, m_id );
         MemWrite( &item->lockRelease.thread, GetThreadHandle() );
+        MemWrite( &item->lockRelease.id, m_id );
         MemWrite( &item->lockRelease.time, Profiler::GetTime() );
-        tail.store( magic + 1, std::memory_order_release );
+        Profiler::QueueSerialFinish();
     }
 
-    tracy_force_inline bool try_lock()
+    tracy_force_inline void AfterTryLock( bool acquired )
     {
-        const auto ret = m_lockable.try_lock();
-
 #ifdef TRACY_ON_DEMAND
-        if( !ret ) return ret;
+        if( !acquired ) return;
 
         bool queue = false;
         const auto locks = m_lockCount.fetch_add( 1, std::memory_order_relaxed );
         const auto active = m_active.load( std::memory_order_relaxed );
         if( locks == 0 || active )
         {
-            const bool connected = s_profiler.IsConnected();
+            const bool connected = GetProfiler().IsConnected();
             if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
             if( connected ) queue = true;
         }
-        if( !queue ) return ret;
+        if( !queue ) return;
 #endif
 
-        if( ret )
+        if( acquired )
         {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+            auto item = Profiler::QueueSerial();
             MemWrite( &item->hdr.type, QueueType::LockObtain );
-            MemWrite( &item->lockObtain.id, m_id );
             MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+            MemWrite( &item->lockObtain.id, m_id );
             MemWrite( &item->lockObtain.time, Profiler::GetTime() );
-            tail.store( magic + 1, std::memory_order_release );
+            Profiler::QueueSerialFinish();
         }
-
-        return ret;
     }
 
-    tracy_force_inline void lock_shared()
+    tracy_force_inline bool BeforeLockShared()
     {
 #ifdef TRACY_ON_DEMAND
         bool queue = false;
@@ -372,103 +357,79 @@ public:
         const auto active = m_active.load( std::memory_order_relaxed );
         if( locks == 0 || active )
         {
-            const bool connected = s_profiler.IsConnected();
+            const bool connected = GetProfiler().IsConnected();
             if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
             if( connected ) queue = true;
         }
-        if( !queue )
-        {
-            m_lockable.lock_shared();
-            return;
-        }
+        if( !queue ) return false;
 #endif
-        const auto thread = GetThreadHandle();
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockSharedWait );
-            MemWrite( &item->lockWait.id, m_id );
-            MemWrite( &item->lockWait.thread, thread );
-            MemWrite( &item->lockWait.time, Profiler::GetTime() );
-            MemWrite( &item->lockWait.type, LockType::SharedLockable );
-            tail.store( magic + 1, std::memory_order_release );
-        }
 
-        m_lockable.lock_shared();
-
-        {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-            MemWrite( &item->hdr.type, QueueType::LockSharedObtain );
-            MemWrite( &item->lockObtain.id, m_id );
-            MemWrite( &item->lockObtain.thread, thread );
-            MemWrite( &item->lockObtain.time, Profiler::GetTime() );
-            tail.store( magic + 1, std::memory_order_release );
-        }
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockSharedWait );
+        MemWrite( &item->lockWait.thread, GetThreadHandle() );
+        MemWrite( &item->lockWait.id, m_id );
+        MemWrite( &item->lockWait.time, Profiler::GetTime() );
+        MemWrite( &item->lockWait.type, LockType::SharedLockable );
+        Profiler::QueueSerialFinish();
+        return true;
     }
 
-    tracy_force_inline void unlock_shared()
+    tracy_force_inline void AfterLockShared()
     {
-        m_lockable.unlock_shared();
+        auto item = Profiler::QueueSerial();
+        MemWrite( &item->hdr.type, QueueType::LockSharedObtain );
+        MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+        MemWrite( &item->lockObtain.id, m_id );
+        MemWrite( &item->lockObtain.time, Profiler::GetTime() );
+        Profiler::QueueSerialFinish();
+    }
 
+    tracy_force_inline void AfterUnlockShared()
+    {
 #ifdef TRACY_ON_DEMAND
         m_lockCount.fetch_sub( 1, std::memory_order_relaxed );
         if( !m_active.load( std::memory_order_relaxed ) ) return;
-        if( !s_profiler.IsConnected() )
+        if( !GetProfiler().IsConnected() )
         {
             m_active.store( false, std::memory_order_relaxed );
             return;
         }
 #endif
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+        auto item = Profiler::QueueSerial();
         MemWrite( &item->hdr.type, QueueType::LockSharedRelease );
-        MemWrite( &item->lockRelease.id, m_id );
         MemWrite( &item->lockRelease.thread, GetThreadHandle() );
+        MemWrite( &item->lockRelease.id, m_id );
         MemWrite( &item->lockRelease.time, Profiler::GetTime() );
-        tail.store( magic + 1, std::memory_order_release );
+        Profiler::QueueSerialFinish();
     }
 
-    tracy_force_inline bool try_lock_shared()
+    tracy_force_inline void AfterTryLockShared( bool acquired )
     {
-        const auto ret = m_lockable.try_lock_shared();
-
 #ifdef TRACY_ON_DEMAND
-        if( !ret ) return ret;
+        if( !acquired ) return;
 
         bool queue = false;
         const auto locks = m_lockCount.fetch_add( 1, std::memory_order_relaxed );
         const auto active = m_active.load( std::memory_order_relaxed );
         if( locks == 0 || active )
         {
-            const bool connected = s_profiler.IsConnected();
+            const bool connected = GetProfiler().IsConnected();
             if( active != connected ) m_active.store( connected, std::memory_order_relaxed );
             if( connected ) queue = true;
         }
-        if( !queue ) return ret;
+        if( !queue ) return;
 #endif
 
-        if( ret )
+        if( acquired )
         {
-            Magic magic;
-            auto& token = s_token.ptr;
-            auto& tail = token->get_tail_index();
-            auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+            auto item = Profiler::QueueSerial();
             MemWrite( &item->hdr.type, QueueType::LockSharedObtain );
-            MemWrite( &item->lockObtain.id, m_id );
             MemWrite( &item->lockObtain.thread, GetThreadHandle() );
+            MemWrite( &item->lockObtain.id, m_id );
             MemWrite( &item->lockObtain.time, Profiler::GetTime() );
-            tail.store( magic + 1, std::memory_order_release );
+            Profiler::QueueSerialFinish();
         }
-
-        return ret;
     }
 
     tracy_force_inline void Mark( const SourceLocationData* srcloc )
@@ -476,7 +437,7 @@ public:
 #ifdef TRACY_ON_DEMAND
         const auto active = m_active.load( std::memory_order_relaxed );
         if( !active ) return;
-        const auto connected = s_profiler.IsConnected();
+        const auto connected = GetProfiler().IsConnected();
         if( !connected )
         {
             if( active ) m_active.store( false, std::memory_order_relaxed );
@@ -484,19 +445,29 @@ public:
         }
 #endif
 
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+        auto item = Profiler::QueueSerial();
         MemWrite( &item->hdr.type, QueueType::LockMark );
-        MemWrite( &item->lockMark.id, m_id );
         MemWrite( &item->lockMark.thread, GetThreadHandle() );
+        MemWrite( &item->lockMark.id, m_id );
         MemWrite( &item->lockMark.srcloc, (uint64_t)srcloc );
-        tail.store( magic + 1, std::memory_order_release );
+        Profiler::QueueSerialFinish();
+    }
+
+    tracy_force_inline void CustomName( const char* name, size_t size )
+    {
+        auto ptr = (char*)tracy_malloc( size+1 );
+        memcpy( ptr, name, size );
+        ptr[size] = '\0';
+        TracyLfqPrepare( QueueType::LockName );
+        MemWrite( &item->lockName.id, m_id );
+        MemWrite( &item->lockName.name, (uint64_t)ptr );
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+        TracyLfqCommit;
     }
 
 private:
-    T m_lockable;
     uint32_t m_id;
 
 #ifdef TRACY_ON_DEMAND
@@ -505,7 +476,74 @@ private:
 #endif
 };
 
+template<class T>
+class SharedLockable
+{
+public:
+    tracy_force_inline SharedLockable( const SourceLocationData* srcloc )
+        : m_ctx( srcloc )
+    {
+    }
 
+    SharedLockable( const SharedLockable& ) = delete;
+    SharedLockable& operator=( const SharedLockable& ) = delete;
+
+    tracy_force_inline void lock()
+    {
+        const auto runAfter = m_ctx.BeforeLock();
+        m_lockable.lock();
+        if( runAfter ) m_ctx.AfterLock();
+    }
+
+    tracy_force_inline void unlock()
+    {
+        m_lockable.unlock();
+        m_ctx.AfterUnlock();
+    }
+
+    tracy_force_inline bool try_lock()
+    {
+        const auto acquired = m_lockable.try_lock();
+        m_ctx.AfterTryLock( acquired );
+        return acquired;
+    }
+
+    tracy_force_inline void lock_shared()
+    {
+        const auto runAfter = m_ctx.BeforeLockShared();
+        m_lockable.lock_shared();
+        if( runAfter ) m_ctx.AfterLockShared();
+    }
+
+    tracy_force_inline void unlock_shared()
+    {
+        m_lockable.unlock_shared();
+        m_ctx.AfterUnlockShared();
+    }
+
+    tracy_force_inline bool try_lock_shared()
+    {
+        const auto acquired = m_lockable.try_lock_shared();
+        m_ctx.AfterTryLockShared( acquired );
+        return acquired;
+    }
+
+    tracy_force_inline void Mark( const SourceLocationData* srcloc )
+    {
+        m_ctx.Mark( srcloc );
+    }
+
+    tracy_force_inline void CustomName( const char* name, size_t size )
+    {
+        m_ctx.CustomName( name, size );
+    }
+
+private:
+    T m_lockable;
+    SharedLockableCtx m_ctx;
 };
 
+
+}
+
 #endif

client/TracyProfiler.hpp

@@ -3,40 +3,62 @@
 
 #include <assert.h>
 #include <atomic>
-#include <chrono>
 #include <stdint.h>
 #include <string.h>
 
-#include "concurrentqueue.h"
+#include "tracy_concurrentqueue.h"
 #include "TracyCallstack.hpp"
+#include "TracySysTime.hpp"
 #include "TracyFastVector.hpp"
-#include "../common/tracy_lz4.hpp"
 #include "../common/TracyQueue.hpp"
 #include "../common/TracyAlign.hpp"
 #include "../common/TracyAlloc.hpp"
 #include "../common/TracyMutex.hpp"
-#include "../common/TracySystem.hpp"
+#include "../common/TracyProtocol.hpp"
 
-#if defined _MSC_VER || defined __CYGWIN__
+#if defined _WIN32 || defined __CYGWIN__
 #  include <intrin.h>
 #endif
-
-#if defined _MSC_VER || defined __CYGWIN__ || ( ( defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64 ) && !defined __ANDROID__ ) || __ARM_ARCH >= 6
-#  define TRACY_HW_TIMER
-#  if defined _MSC_VER || defined __CYGWIN__
-     // Enable optimization for MSVC __rdtscp() intrin, saving one LHS of a cpu value on the stack.
-     // This comes at the cost of an unaligned memory write.
-#    define TRACY_RDTSCP_OPT
-#  endif
+#ifdef __APPLE__
+#  include <TargetConditionals.h>
+#  include <mach/mach_time.h>
 #endif
 
-#define TracyConcat(x,y) TracyConcatIndirect(x,y)
-#define TracyConcatIndirect(x,y) x##y
+#if defined _WIN32 || defined __CYGWIN__ || ( ( defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64 ) && !defined __ANDROID__ ) || __ARM_ARCH >= 6
+#  define TRACY_HW_TIMER
+#endif
+
+#if !defined TRACY_HW_TIMER || ( defined __ARM_ARCH && __ARM_ARCH >= 6 && !defined CLOCK_MONOTONIC_RAW )
+  #include <chrono>
+#endif
+
+#ifndef TracyConcat
+#  define TracyConcat(x,y) TracyConcatIndirect(x,y)
+#endif
+#ifndef TracyConcatIndirect
+#  define TracyConcatIndirect(x,y) x##y
+#endif
 
 namespace tracy
 {
 
+class GpuCtx;
+class Profiler;
 class Socket;
+class UdpBroadcast;
+
+struct GpuCtxWrapper
+{
+    GpuCtx* ptr;
+};
+
+TRACY_API moodycamel::ConcurrentQueue<QueueItem>::ExplicitProducer* GetToken();
+TRACY_API Profiler& GetProfiler();
+TRACY_API std::atomic<uint32_t>& GetLockCounter();
+TRACY_API std::atomic<uint8_t>& GetGpuCtxCounter();
+TRACY_API GpuCtxWrapper& GetGpuCtx();
+TRACY_API uint64_t GetThreadHandle();
+TRACY_API void InitRPMallocThread();
 
 struct SourceLocationData
 {
@@ -47,25 +69,6 @@ struct SourceLocationData
     uint32_t color;
 };
 
-struct ProducerWrapper
-{
-    tracy::moodycamel::ConcurrentQueue<QueueItem>::ExplicitProducer* ptr;
-};
-
-extern thread_local ProducerWrapper s_token;
-
-class GpuCtx;
-struct GpuCtxWrapper
-{
-    GpuCtx* ptr;
-};
-
-class VkCtx;
-struct VkCtxWrapper
-{
-    VkCtx* ptr;
-};
-
 #ifdef TRACY_ON_DEMAND
 struct LuaZoneState
 {
@@ -74,217 +77,317 @@ struct LuaZoneState
 };
 #endif
 
-using Magic = tracy::moodycamel::ConcurrentQueueDefaultTraits::index_t;
 
-#if __ARM_ARCH >= 6
-extern int64_t (*GetTimeImpl)();
-#endif
+#define TracyLfqPrepare( _type ) \
+    moodycamel::ConcurrentQueueDefaultTraits::index_t __magic; \
+    auto __token = GetToken(); \
+    auto& __tail = __token->get_tail_index(); \
+    auto item = __token->enqueue_begin( __magic ); \
+    MemWrite( &item->hdr.type, _type );
 
-class Profiler;
-extern Profiler& s_profiler;
+#define TracyLfqCommit \
+    __tail.store( __magic + 1, std::memory_order_release );
+
+#define TracyLfqPrepareC( _type ) \
+    tracy::moodycamel::ConcurrentQueueDefaultTraits::index_t __magic; \
+    auto __token = tracy::GetToken(); \
+    auto& __tail = __token->get_tail_index(); \
+    auto item = __token->enqueue_begin( __magic ); \
+    tracy::MemWrite( &item->hdr.type, _type );
+
+#define TracyLfqCommitC \
+    __tail.store( __magic + 1, std::memory_order_release );
+
+
+typedef void(*ParameterCallback)( uint32_t idx, int32_t val );
 
 class Profiler
 {
+    struct FrameImageQueueItem
+    {
+        void* image;
+        uint64_t frame;
+        uint16_t w;
+        uint16_t h;
+        uint8_t offset;
+        bool flip;
+    };
+
 public:
     Profiler();
     ~Profiler();
 
-    static tracy_force_inline int64_t GetTime( uint32_t& cpu )
-    {
-#ifdef TRACY_HW_TIMER
-#  if __ARM_ARCH >= 6
-        cpu = 0xFFFFFFFF;
-        return GetTimeImpl();
-#  elif defined _MSC_VER || defined __CYGWIN__
-        const auto t = int64_t( __rdtscp( &cpu ) );
-        return t;
-#  elif defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64
-        uint32_t eax, edx;
-        asm volatile ( "rdtscp" : "=a" (eax), "=d" (edx), "=c" (cpu) :: );
-        return ( uint64_t( edx ) << 32 ) + uint64_t( eax );
-#  endif
-#else
-        cpu = 0xFFFFFFFF;
-        return std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now().time_since_epoch() ).count();
-#endif
-    }
-
     static tracy_force_inline int64_t GetTime()
     {
 #ifdef TRACY_HW_TIMER
-#  if __ARM_ARCH >= 6
-        return GetTimeImpl();
-#  elif defined _MSC_VER || defined __CYGWIN__
-        unsigned int dontcare;
-        const auto t = int64_t( __rdtscp( &dontcare ) );
-        return t;
-#  elif defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64
+#  if defined TARGET_OS_IOS && TARGET_OS_IOS == 1
+        return mach_absolute_time();
+#  elif defined __ARM_ARCH && __ARM_ARCH >= 6
+#    ifdef CLOCK_MONOTONIC_RAW
+        struct timespec ts;
+        clock_gettime( CLOCK_MONOTONIC_RAW, &ts );
+        return int64_t( ts.tv_sec ) * 1000000000ll + int64_t( ts.tv_nsec );
+#    else
+        return std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now().time_since_epoch() ).count();
+#    endif
+#  elif defined _WIN32 || defined __CYGWIN__
+#    ifdef TRACY_TIMER_QPC
+        return GetTimeQpc();
+#    else
+        return int64_t( __rdtsc() );
+#    endif
+#  elif defined __i386 || defined _M_IX86
         uint32_t eax, edx;
-        asm volatile ( "rdtscp" : "=a" (eax), "=d" (edx) :: "%ecx" );
+        asm volatile ( "rdtsc" : "=a" (eax), "=d" (edx) );
         return ( uint64_t( edx ) << 32 ) + uint64_t( eax );
+#  elif defined __x86_64__ || defined _M_X64
+        uint64_t rax, rdx;
+        asm volatile ( "rdtsc" : "=a" (rax), "=d" (rdx) );
+        return ( rdx << 32 ) + rax;
 #  endif
 #else
         return std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::high_resolution_clock::now().time_since_epoch() ).count();
 #endif
     }
 
-    static tracy_force_inline void SendFrameMark()
+    tracy_force_inline uint32_t GetNextZoneId()
     {
+        return m_zoneId.fetch_add( 1, std::memory_order_relaxed );
+    }
+
+    static tracy_force_inline QueueItem* QueueSerial()
+    {
+        auto& p = GetProfiler();
+        p.m_serialLock.lock();
+        return p.m_serialQueue.prepare_next();
+    }
+
+    static tracy_force_inline void QueueSerialFinish()
+    {
+        auto& p = GetProfiler();
+        p.m_serialQueue.commit_next();
+        p.m_serialLock.unlock();
+    }
+
+    static tracy_force_inline void SendFrameMark( const char* name )
+    {
+        if( !name ) GetProfiler().m_frameCount.fetch_add( 1, std::memory_order_relaxed );
 #ifdef TRACY_ON_DEMAND
-        s_profiler.m_frameCount.fetch_add( 1, std::memory_order_relaxed );
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::FrameMarkMsg );
+        TracyLfqPrepare( QueueType::FrameMarkMsg );
         MemWrite( &item->frameMark.time, GetTime() );
-        MemWrite( &item->frameMark.name, uint64_t( 0 ) );
-        tail.store( magic + 1, std::memory_order_release );
+        MemWrite( &item->frameMark.name, uint64_t( name ) );
+        TracyLfqCommit;
     }
 
     static tracy_force_inline void SendFrameMark( const char* name, QueueType type )
     {
-        assert( type == QueueType::FrameMarkMsg || type == QueueType::FrameMarkMsgStart || type == QueueType::FrameMarkMsgEnd );
+        assert( type == QueueType::FrameMarkMsgStart || type == QueueType::FrameMarkMsgEnd );
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
+        auto item = QueueSerial();
         MemWrite( &item->hdr.type, type );
         MemWrite( &item->frameMark.time, GetTime() );
         MemWrite( &item->frameMark.name, uint64_t( name ) );
-        tail.store( magic + 1, std::memory_order_release );
+        QueueSerialFinish();
+    }
+
+    static tracy_force_inline void SendFrameImage( const void* image, uint16_t w, uint16_t h, uint8_t offset, bool flip )
+    {
+        auto& profiler = GetProfiler();
+#ifdef TRACY_ON_DEMAND
+        if( !profiler.IsConnected() ) return;
+#endif
+        const auto sz = size_t( w ) * size_t( h ) * 4;
+        auto ptr = (char*)tracy_malloc( sz );
+        memcpy( ptr, image, sz );
+
+        profiler.m_fiLock.lock();
+        auto fi = profiler.m_fiQueue.prepare_next();
+        fi->image = ptr;
+        fi->frame = profiler.m_frameCount.load( std::memory_order_relaxed ) - offset;
+        fi->w = w;
+        fi->h = h;
+        fi->flip = flip;
+        profiler.m_fiQueue.commit_next();
+        profiler.m_fiLock.unlock();
     }
 
     static tracy_force_inline void PlotData( const char* name, int64_t val )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::PlotData );
+        TracyLfqPrepare( QueueType::PlotData );
         MemWrite( &item->plotData.name, (uint64_t)name );
         MemWrite( &item->plotData.time, GetTime() );
         MemWrite( &item->plotData.type, PlotDataType::Int );
         MemWrite( &item->plotData.data.i, val );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
     static tracy_force_inline void PlotData( const char* name, float val )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::PlotData );
+        TracyLfqPrepare( QueueType::PlotData );
         MemWrite( &item->plotData.name, (uint64_t)name );
         MemWrite( &item->plotData.time, GetTime() );
         MemWrite( &item->plotData.type, PlotDataType::Float );
         MemWrite( &item->plotData.data.f, val );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
     static tracy_force_inline void PlotData( const char* name, double val )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::PlotData );
+        TracyLfqPrepare( QueueType::PlotData );
         MemWrite( &item->plotData.name, (uint64_t)name );
         MemWrite( &item->plotData.time, GetTime() );
         MemWrite( &item->plotData.type, PlotDataType::Double );
         MemWrite( &item->plotData.data.d, val );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
-    static tracy_force_inline void Message( const char* txt, size_t size )
+    static tracy_force_inline void ConfigurePlot( const char* name, PlotFormatType type )
+    {
+        TracyLfqPrepare( QueueType::PlotConfig );
+        MemWrite( &item->plotConfig.name, (uint64_t)name );
+        MemWrite( &item->plotConfig.type, (uint8_t)type );
+
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+
+        TracyLfqCommit;
+    }
+
+    static tracy_force_inline void Message( const char* txt, size_t size, int callstack )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
+        TracyLfqPrepare( callstack == 0 ? QueueType::Message : QueueType::MessageCallstack );
         auto ptr = (char*)tracy_malloc( size+1 );
         memcpy( ptr, txt, size );
         ptr[size] = '\0';
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::Message );
         MemWrite( &item->message.time, GetTime() );
-        MemWrite( &item->message.thread, GetThreadHandle() );
         MemWrite( &item->message.text, (uint64_t)ptr );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
+
+        if( callstack != 0 ) tracy::GetProfiler().SendCallstack( callstack );
     }
 
-    static tracy_force_inline void Message( const char* txt )
+    static tracy_force_inline void Message( const char* txt, int callstack )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::MessageLiteral );
+        TracyLfqPrepare( callstack == 0 ? QueueType::MessageLiteral : QueueType::MessageLiteralCallstack );
         MemWrite( &item->message.time, GetTime() );
-        MemWrite( &item->message.thread, GetThreadHandle() );
         MemWrite( &item->message.text, (uint64_t)txt );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
+
+        if( callstack != 0 ) tracy::GetProfiler().SendCallstack( callstack );
+    }
+
+    static tracy_force_inline void MessageColor( const char* txt, size_t size, uint32_t color, int callstack )
+    {
+#ifdef TRACY_ON_DEMAND
+        if( !GetProfiler().IsConnected() ) return;
+#endif
+        TracyLfqPrepare( callstack == 0 ? QueueType::MessageColor : QueueType::MessageColorCallstack );
+        auto ptr = (char*)tracy_malloc( size+1 );
+        memcpy( ptr, txt, size );
+        ptr[size] = '\0';
+        MemWrite( &item->messageColor.time, GetTime() );
+        MemWrite( &item->messageColor.text, (uint64_t)ptr );
+        MemWrite( &item->messageColor.r, uint8_t( ( color       ) & 0xFF ) );
+        MemWrite( &item->messageColor.g, uint8_t( ( color >> 8  ) & 0xFF ) );
+        MemWrite( &item->messageColor.b, uint8_t( ( color >> 16 ) & 0xFF ) );
+        TracyLfqCommit;
+
+        if( callstack != 0 ) tracy::GetProfiler().SendCallstack( callstack );
+    }
+
+    static tracy_force_inline void MessageColor( const char* txt, uint32_t color, int callstack )
+    {
+#ifdef TRACY_ON_DEMAND
+        if( !GetProfiler().IsConnected() ) return;
+#endif
+        TracyLfqPrepare( callstack == 0 ? QueueType::MessageLiteralColor : QueueType::MessageLiteralColorCallstack );
+        MemWrite( &item->messageColor.time, GetTime() );
+        MemWrite( &item->messageColor.text, (uint64_t)txt );
+        MemWrite( &item->messageColor.r, uint8_t( ( color       ) & 0xFF ) );
+        MemWrite( &item->messageColor.g, uint8_t( ( color >> 8  ) & 0xFF ) );
+        MemWrite( &item->messageColor.b, uint8_t( ( color >> 16 ) & 0xFF ) );
+        TracyLfqCommit;
+
+        if( callstack != 0 ) tracy::GetProfiler().SendCallstack( callstack );
+    }
+
+    static tracy_force_inline void MessageAppInfo( const char* txt, size_t size )
+    {
+        InitRPMallocThread();
+        auto ptr = (char*)tracy_malloc( size+1 );
+        memcpy( ptr, txt, size );
+        ptr[size] = '\0';
+        TracyLfqPrepare( QueueType::MessageAppInfo );
+        MemWrite( &item->message.time, GetTime() );
+        MemWrite( &item->message.text, (uint64_t)ptr );
+
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+
+        TracyLfqCommit;
     }
 
     static tracy_force_inline void MemAlloc( const void* ptr, size_t size )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
         const auto thread = GetThreadHandle();
 
-        s_profiler.m_serialLock.lock();
+        GetProfiler().m_serialLock.lock();
         SendMemAlloc( QueueType::MemAlloc, thread, ptr, size );
-        s_profiler.m_serialLock.unlock();
+        GetProfiler().m_serialLock.unlock();
     }
 
     static tracy_force_inline void MemFree( const void* ptr )
     {
 #ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !GetProfiler().IsConnected() ) return;
 #endif
         const auto thread = GetThreadHandle();
 
-        s_profiler.m_serialLock.lock();
+        GetProfiler().m_serialLock.lock();
         SendMemFree( QueueType::MemFree, thread, ptr );
-        s_profiler.m_serialLock.unlock();
+        GetProfiler().m_serialLock.unlock();
     }
 
     static tracy_force_inline void MemAllocCallstack( const void* ptr, size_t size, int depth )
     {
 #ifdef TRACY_HAS_CALLSTACK
+        auto& profiler = GetProfiler();
 #  ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !profiler.IsConnected() ) return;
 #  endif
         const auto thread = GetThreadHandle();
 
-        rpmalloc_thread_initialize();
+        InitRPMallocThread();
         auto callstack = Callstack( depth );
 
-        s_profiler.m_serialLock.lock();
+        profiler.m_serialLock.lock();
         SendMemAlloc( QueueType::MemAllocCallstack, thread, ptr, size );
         SendCallstackMemory( callstack );
-        s_profiler.m_serialLock.unlock();
+        profiler.m_serialLock.unlock();
 #else
         MemAlloc( ptr, size );
 #endif
@@ -293,46 +396,64 @@ public:
     static tracy_force_inline void MemFreeCallstack( const void* ptr, int depth )
     {
 #ifdef TRACY_HAS_CALLSTACK
+        auto& profiler = GetProfiler();
 #  ifdef TRACY_ON_DEMAND
-        if( !s_profiler.IsConnected() ) return;
+        if( !profiler.IsConnected() ) return;
 #  endif
         const auto thread = GetThreadHandle();
 
-        rpmalloc_thread_initialize();
+        InitRPMallocThread();
         auto callstack = Callstack( depth );
 
-        s_profiler.m_serialLock.lock();
+        profiler.m_serialLock.lock();
         SendMemFree( QueueType::MemFreeCallstack, thread, ptr );
         SendCallstackMemory( callstack );
-        s_profiler.m_serialLock.unlock();
+        profiler.m_serialLock.unlock();
 #else
         MemFree( ptr );
 #endif
     }
 
-    static tracy_force_inline void SendCallstack( int depth, uint64_t thread )
+    static tracy_force_inline void SendCallstack( int depth )
     {
 #ifdef TRACY_HAS_CALLSTACK
         auto ptr = Callstack( depth );
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::Callstack );
-        MemWrite( &item->callstack.ptr, ptr );
-        MemWrite( &item->callstack.thread, thread );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqPrepare( QueueType::Callstack );
+        MemWrite( &item->callstack.ptr, (uint64_t)ptr );
+        TracyLfqCommit;
 #endif
     }
 
-    void SendCallstack( int depth, uint64_t thread, const char* skipBefore );
+    static tracy_force_inline void ParameterRegister( ParameterCallback cb ) { GetProfiler().m_paramCallback = cb; }
+    static tracy_force_inline void ParameterSetup( uint32_t idx, const char* name, bool isBool, int32_t val )
+    {
+        TracyLfqPrepare( QueueType::ParamSetup );
+        tracy::MemWrite( &item->paramSetup.idx, idx );
+        tracy::MemWrite( &item->paramSetup.name, (uint64_t)name );
+        tracy::MemWrite( &item->paramSetup.isBool, (uint8_t)isBool );
+        tracy::MemWrite( &item->paramSetup.val, val );
+
+#ifdef TRACY_ON_DEMAND
+        GetProfiler().DeferItem( *item );
+#endif
+
+        TracyLfqCommit;
+    }
+
+    void SendCallstack( int depth, const char* skipBefore );
+    static void CutCallstack( void* callstack, const char* skipBefore );
 
     static bool ShouldExit();
 
 #ifdef TRACY_ON_DEMAND
-    tracy_force_inline bool IsConnected()
+    tracy_force_inline bool IsConnected() const
     {
-        return m_isConnected.load( std::memory_order_relaxed );
+        return m_isConnected.load( std::memory_order_acquire );
+    }
+
+    tracy_force_inline uint64_t ConnectionId() const
+    {
+        return m_connectionId.load( std::memory_order_acquire );
     }
 
     tracy_force_inline void DeferItem( const QueueItem& item )
@@ -347,18 +468,81 @@ public:
     void RequestShutdown() { m_shutdown.store( true, std::memory_order_relaxed ); m_shutdownManual.store( true, std::memory_order_relaxed ); }
     bool HasShutdownFinished() const { return m_shutdownFinished.load( std::memory_order_relaxed ); }
 
+    void SendString( uint64_t ptr, const char* str, QueueType type );
+
+
+    // Allocated source location data layout:
+    //  4b  payload size
+    //  4b  color
+    //  4b  source line
+    //  fsz function name
+    //  1b  null terminator
+    //  ssz source file name
+    //  1b  null terminator
+    //  nsz zone name (optional)
+
+    static tracy_force_inline uint64_t AllocSourceLocation( uint32_t line, const char* source, const char* function )
+    {
+        const auto fsz = strlen( function );
+        const auto ssz = strlen( source );
+        const uint32_t sz = uint32_t( 4 + 4 + 4 + fsz + 1 + ssz + 1 );
+        auto ptr = (char*)tracy_malloc( sz );
+        memcpy( ptr, &sz, 4 );
+        memset( ptr + 4, 0, 4 );
+        memcpy( ptr + 8, &line, 4 );
+        memcpy( ptr + 12, function, fsz+1 );
+        memcpy( ptr + 12 + fsz + 1, source, ssz + 1 );
+        return uint64_t( ptr );
+    }
+
+    static tracy_force_inline uint64_t AllocSourceLocation( uint32_t line, const char* source, const char* function, const char* name, size_t nameSz )
+    {
+        const auto fsz = strlen( function );
+        const auto ssz = strlen( source );
+        const uint32_t sz = uint32_t( 4 + 4 + 4 + fsz + 1 + ssz + 1 + nameSz );
+        auto ptr = (char*)tracy_malloc( sz );
+        memcpy( ptr, &sz, 4 );
+        memset( ptr + 4, 0, 4 );
+        memcpy( ptr + 8, &line, 4 );
+        memcpy( ptr + 12, function, fsz+1 );
+        memcpy( ptr + 12 + fsz + 1, source, ssz + 1 );
+        memcpy( ptr + 12 + fsz + 1 + ssz + 1, name, nameSz );
+        return uint64_t( ptr );
+    }
+
 private:
-    enum DequeueStatus { Success, ConnectionLost, QueueEmpty };
+    enum class DequeueStatus { DataDequeued, ConnectionLost, QueueEmpty };
 
     static void LaunchWorker( void* ptr ) { ((Profiler*)ptr)->Worker(); }
     void Worker();
 
+    static void LaunchCompressWorker( void* ptr ) { ((Profiler*)ptr)->CompressWorker(); }
+    void CompressWorker();
+
     void ClearQueues( tracy::moodycamel::ConsumerToken& token );
+    void ClearSerial();
     DequeueStatus Dequeue( tracy::moodycamel::ConsumerToken& token );
+    DequeueStatus DequeueContextSwitches( tracy::moodycamel::ConsumerToken& token, int64_t& timeStop );
     DequeueStatus DequeueSerial();
-    bool AppendData( const void* data, size_t len );
     bool CommitData();
-    bool NeedDataSize( size_t len );
+
+    tracy_force_inline bool AppendData( const void* data, size_t len )
+    {
+        const auto ret = NeedDataSize( len );
+        AppendDataUnsafe( data, len );
+        return ret;
+    }
+
+    tracy_force_inline bool NeedDataSize( size_t len )
+    {
+        assert( len <= TargetFrameSize );
+        bool ret = true;
+        if( m_bufferOffset - m_bufferStart + len > TargetFrameSize )
+        {
+            ret = CommitData();
+        }
+        return ret;
+    }
 
     tracy_force_inline void AppendDataUnsafe( const void* data, size_t len )
     {
@@ -367,24 +551,32 @@ private:
     }
 
     bool SendData( const char* data, size_t len );
-    void SendString( uint64_t ptr, const char* str, QueueType type );
+    void SendLongString( uint64_t ptr, const char* str, size_t len, QueueType type );
     void SendSourceLocation( uint64_t ptr );
     void SendSourceLocationPayload( uint64_t ptr );
     void SendCallstackPayload( uint64_t ptr );
+    void SendCallstackPayload64( uint64_t ptr );
+    void SendCallstackAlloc( uint64_t ptr );
     void SendCallstackFrame( uint64_t ptr );
+    void SendCodeLocation( uint64_t ptr );
 
     bool HandleServerQuery();
+    void HandleDisconnect();
+    void HandleParameter( uint64_t payload );
+    void HandleSymbolQuery( uint64_t symbol );
+    void HandleSymbolCodeQuery( uint64_t symbol, uint32_t size );
 
     void CalibrateTimer();
     void CalibrateDelay();
+    void ReportTopology();
 
     static tracy_force_inline void SendCallstackMemory( void* ptr )
     {
 #ifdef TRACY_HAS_CALLSTACK
-        auto item = s_profiler.m_serialQueue.prepare_next();
+        auto item = GetProfiler().m_serialQueue.prepare_next();
         MemWrite( &item->hdr.type, QueueType::CallstackMemory );
         MemWrite( &item->callstackMemory.ptr, (uint64_t)ptr );
-        s_profiler.m_serialQueue.commit_next();
+        GetProfiler().m_serialQueue.commit_next();
 #endif
     }
 
@@ -392,7 +584,7 @@ private:
     {
         assert( type == QueueType::MemAlloc || type == QueueType::MemAllocCallstack );
 
-        auto item = s_profiler.m_serialQueue.prepare_next();
+        auto item = GetProfiler().m_serialQueue.prepare_next();
         MemWrite( &item->hdr.type, type );
         MemWrite( &item->memAlloc.time, GetTime() );
         MemWrite( &item->memAlloc.thread, thread );
@@ -405,23 +597,28 @@ private:
         else
         {
             assert( sizeof( size ) == 8 );
-            memcpy( &item->memAlloc.size, &size, 6 );
+            memcpy( &item->memAlloc.size, &size, 4 );
+            memcpy( ((char*)&item->memAlloc.size)+4, ((char*)&size)+4, 2 );
         }
-        s_profiler.m_serialQueue.commit_next();
+        GetProfiler().m_serialQueue.commit_next();
     }
 
     static tracy_force_inline void SendMemFree( QueueType type, const uint64_t thread, const void* ptr )
     {
         assert( type == QueueType::MemFree || type == QueueType::MemFreeCallstack );
 
-        auto item = s_profiler.m_serialQueue.prepare_next();
+        auto item = GetProfiler().m_serialQueue.prepare_next();
         MemWrite( &item->hdr.type, type );
         MemWrite( &item->memFree.time, GetTime() );
         MemWrite( &item->memFree.thread, thread );
         MemWrite( &item->memFree.ptr, (uint64_t)ptr );
-        s_profiler.m_serialQueue.commit_next();
+        GetProfiler().m_serialQueue.commit_next();
     }
 
+#if ( defined _WIN32 || defined __CYGWIN__ ) && defined TRACY_TIMER_QPC
+    static int64_t GetTimeQpc();
+#endif
+
     double m_timerMul;
     uint64_t m_resolution;
     uint64_t m_delay;
@@ -432,28 +629,52 @@ private:
     std::atomic<bool> m_shutdownManual;
     std::atomic<bool> m_shutdownFinished;
     Socket* m_sock;
+    UdpBroadcast* m_broadcast;
     bool m_noExit;
+    uint32_t m_userPort;
+    std::atomic<uint32_t> m_zoneId;
+    int64_t m_samplingPeriod;
 
-    LZ4_stream_t* m_stream;
+    uint64_t m_threadCtx;
+    int64_t m_refTimeThread;
+    int64_t m_refTimeSerial;
+    int64_t m_refTimeCtx;
+    int64_t m_refTimeGpu;
+
+    void* m_stream;     // LZ4_stream_t*
     char* m_buffer;
     int m_bufferOffset;
     int m_bufferStart;
 
-    QueueItem* m_itemBuf;
     char* m_lz4Buf;
 
     FastVector<QueueItem> m_serialQueue, m_serialDequeue;
     TracyMutex m_serialLock;
 
+    FastVector<FrameImageQueueItem> m_fiQueue, m_fiDequeue;
+    TracyMutex m_fiLock;
+
+    std::atomic<uint64_t> m_frameCount;
 #ifdef TRACY_ON_DEMAND
     std::atomic<bool> m_isConnected;
-    std::atomic<uint64_t> m_frameCount;
+    std::atomic<uint64_t> m_connectionId;
 
     TracyMutex m_deferredLock;
     FastVector<QueueItem> m_deferredQueue;
 #endif
+
+#ifdef TRACY_HAS_SYSTIME
+    void ProcessSysTime();
+
+    SysTime m_sysTime;
+    uint64_t m_sysTimeLast = 0;
+#else
+    void ProcessSysTime() {}
+#endif
+
+    ParameterCallback m_paramCallback;
 };
 
-};
+}
 
 #endif

client/TracyScoped.hpp

@@ -17,114 +17,96 @@ class ScopedZone
 public:
     tracy_force_inline ScopedZone( const SourceLocationData* srcloc, bool is_active = true )
 #ifdef TRACY_ON_DEMAND
-        : m_active( s_profiler.IsConnected() )
+        : m_active( is_active && GetProfiler().IsConnected() )
 #else
         : m_active( is_active )
 #endif
     {
         if( !m_active ) return;
-        const auto thread = GetThreadHandle();
-        m_thread = thread;
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::ZoneBegin );
-#ifdef TRACY_RDTSCP_OPT
-        MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
-#else
-        uint32_t cpu;
-        MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
-        MemWrite( &item->zoneBegin.cpu, cpu );
+#ifdef TRACY_ON_DEMAND
+        m_connectionId = GetProfiler().ConnectionId();
 #endif
-        MemWrite( &item->zoneBegin.thread, thread );
+        TracyLfqPrepare( QueueType::ZoneBegin );
+        MemWrite( &item->zoneBegin.time, Profiler::GetTime() );
         MemWrite( &item->zoneBegin.srcloc, (uint64_t)srcloc );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
     tracy_force_inline ScopedZone( const SourceLocationData* srcloc, int depth, bool is_active = true )
 #ifdef TRACY_ON_DEMAND
-        : m_active( s_profiler.IsConnected() )
+        : m_active( is_active && GetProfiler().IsConnected() )
 #else
         : m_active( is_active )
 #endif
     {
         if( !m_active ) return;
-        const auto thread = GetThreadHandle();
-        m_thread = thread;
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::ZoneBeginCallstack );
-#ifdef TRACY_RDTSCP_OPT
-        MemWrite( &item->zoneBegin.time, Profiler::GetTime( item->zoneBegin.cpu ) );
-#else
-        uint32_t cpu;
-        MemWrite( &item->zoneBegin.time, Profiler::GetTime( cpu ) );
-        MemWrite( &item->zoneBegin.cpu, cpu );
+#ifdef TRACY_ON_DEMAND
+        m_connectionId = GetProfiler().ConnectionId();
 #endif
-        MemWrite( &item->zoneBegin.thread, thread );
+        TracyLfqPrepare( QueueType::ZoneBeginCallstack );
+        MemWrite( &item->zoneBegin.time, Profiler::GetTime() );
         MemWrite( &item->zoneBegin.srcloc, (uint64_t)srcloc );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
 
-        s_profiler.SendCallstack( depth, thread );
+        GetProfiler().SendCallstack( depth );
     }
 
     tracy_force_inline ~ScopedZone()
     {
         if( !m_active ) return;
-        Magic magic;
-        auto& token = s_token.ptr;
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::ZoneEnd );
-#ifdef TRACY_RDTSCP_OPT
-        MemWrite( &item->zoneEnd.time, Profiler::GetTime( item->zoneEnd.cpu ) );
-#else
-        uint32_t cpu;
-        MemWrite( &item->zoneEnd.time, Profiler::GetTime( cpu ) );
-        MemWrite( &item->zoneEnd.cpu, cpu );
+#ifdef TRACY_ON_DEMAND
+        if( GetProfiler().ConnectionId() != m_connectionId ) return;
 #endif
-        MemWrite( &item->zoneEnd.thread, m_thread );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqPrepare( QueueType::ZoneEnd );
+        MemWrite( &item->zoneEnd.time, Profiler::GetTime() );
+        TracyLfqCommit;
     }
 
     tracy_force_inline void Text( const char* txt, size_t size )
     {
         if( !m_active ) return;
-        Magic magic;
-        auto& token = s_token.ptr;
+#ifdef TRACY_ON_DEMAND
+        if( GetProfiler().ConnectionId() != m_connectionId ) return;
+#endif
         auto ptr = (char*)tracy_malloc( size+1 );
         memcpy( ptr, txt, size );
         ptr[size] = '\0';
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::ZoneText );
-        MemWrite( &item->zoneText.thread, m_thread );
+        TracyLfqPrepare( QueueType::ZoneText );
         MemWrite( &item->zoneText.text, (uint64_t)ptr );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
     }
 
     tracy_force_inline void Name( const char* txt, size_t size )
     {
         if( !m_active ) return;
-        Magic magic;
-        auto& token = s_token.ptr;
+#ifdef TRACY_ON_DEMAND
+        if( GetProfiler().ConnectionId() != m_connectionId ) return;
+#endif
         auto ptr = (char*)tracy_malloc( size+1 );
         memcpy( ptr, txt, size );
         ptr[size] = '\0';
-        auto& tail = token->get_tail_index();
-        auto item = token->enqueue_begin<tracy::moodycamel::CanAlloc>( magic );
-        MemWrite( &item->hdr.type, QueueType::ZoneName );
-        MemWrite( &item->zoneText.thread, m_thread );
+        TracyLfqPrepare( QueueType::ZoneName );
         MemWrite( &item->zoneText.text, (uint64_t)ptr );
-        tail.store( magic + 1, std::memory_order_release );
+        TracyLfqCommit;
+    }
+
+    tracy_force_inline void Value( uint64_t value )
+    {
+        if( !m_active ) return;
+#ifdef TRACY_ON_DEMAND
+        if( GetProfiler().ConnectionId() != m_connectionId ) return;
+#endif
+        TracyLfqPrepare( QueueType::ZoneValue );
+        MemWrite( &item->zoneValue.value, value );
+        TracyLfqCommit;
     }
 
 private:
-    uint64_t m_thread;
     const bool m_active;
+
+#ifdef TRACY_ON_DEMAND
+    uint64_t m_connectionId;
+#endif
 };
 
 }

client/TracySysTime.cpp

@@ -0,0 +1,108 @@
+#include "TracySysTime.hpp"
+
+#ifdef TRACY_HAS_SYSTIME
+
+#  if defined _WIN32 || defined __CYGWIN__
+#    include <windows.h>
+#  elif defined __linux__
+#    include <stdio.h>
+#    include <inttypes.h>
+#  elif defined __APPLE__
+#    include <mach/mach_host.h>
+#    include <mach/host_info.h>
+#  elif defined BSD
+#    include <sys/types.h>
+#    include <sys/sysctl.h>
+#  endif
+
+namespace tracy
+{
+
+#  if defined _WIN32 || defined __CYGWIN__
+
+static inline uint64_t ConvertTime( const FILETIME& t )
+{
+    return ( uint64_t( t.dwHighDateTime ) << 32 ) | uint64_t( t.dwLowDateTime );
+}
+
+void SysTime::ReadTimes()
+{
+    FILETIME idleTime;
+    FILETIME kernelTime;
+    FILETIME userTime;
+
+    GetSystemTimes( &idleTime, &kernelTime, &userTime );
+
+    idle = ConvertTime( idleTime );
+    const auto kernel = ConvertTime( kernelTime );
+    const auto user = ConvertTime( userTime );
+    used = kernel + user;
+}
+
+#  elif defined __linux__
+
+void SysTime::ReadTimes()
+{
+    uint64_t user, nice, system;
+    FILE* f = fopen( "/proc/stat", "r" );
+    if( f )
+    {
+        int read = fscanf( f, "cpu %" PRIu64 " %" PRIu64 " %" PRIu64" %" PRIu64, &user, &nice, &system, &idle );
+        fclose( f );
+        if (read == 4)
+        {
+            used = user + nice + system;
+        }
+    }
+}
+
+#  elif defined __APPLE__
+
+void SysTime::ReadTimes()
+{
+    host_cpu_load_info_data_t info;
+    mach_msg_type_number_t cnt = HOST_CPU_LOAD_INFO_COUNT;
+    host_statistics( mach_host_self(), HOST_CPU_LOAD_INFO, reinterpret_cast<host_info_t>( &info ), &cnt ); 
+    used = info.cpu_ticks[CPU_STATE_USER] + info.cpu_ticks[CPU_STATE_NICE] + info.cpu_ticks[CPU_STATE_SYSTEM];
+    idle = info.cpu_ticks[CPU_STATE_IDLE];
+}
+
+#  elif defined BSD
+
+void SysTime::ReadTimes()
+{
+    u_long data[5];
+    size_t sz = sizeof( data );
+    sysctlbyname( "kern.cp_time", &data, &sz, nullptr, 0 );
+    used = data[0] + data[1] + data[2] + data[3];
+    idle = data[4];
+}
+
+#endif
+
+SysTime::SysTime()
+{
+    ReadTimes();
+}
+
+float SysTime::Get()
+{
+    const auto oldUsed = used;
+    const auto oldIdle = idle;
+
+    ReadTimes();
+
+    const auto diffIdle = idle - oldIdle;
+    const auto diffUsed = used - oldUsed;
+
+#if defined _WIN32 || defined __CYGWIN__
+    return diffUsed == 0 ? -1 : ( diffUsed - diffIdle ) * 100.f / diffUsed;
+#elif defined __linux__ || defined __APPLE__ || defined BSD
+    const auto total = diffUsed + diffIdle;
+    return total == 0 ? -1 : diffUsed * 100.f / total;
+#endif
+}
+
+}
+
+#endif

client/TracySysTime.hpp

@@ -0,0 +1,36 @@
+#ifndef __TRACYSYSTIME_HPP__
+#define __TRACYSYSTIME_HPP__
+
+#if defined _WIN32 || defined __CYGWIN__ || defined __linux__ || defined __APPLE__
+#  define TRACY_HAS_SYSTIME
+#else
+#  include <sys/param.h>
+#endif
+
+#ifdef BSD
+#  define TRACY_HAS_SYSTIME
+#endif
+
+#ifdef TRACY_HAS_SYSTIME
+
+#include <stdint.h>
+
+namespace tracy
+{
+
+class SysTime
+{
+public:
+    SysTime();
+    float Get();
+
+    void ReadTimes();
+
+private:
+    uint64_t idle, used;
+};
+
+}
+#endif
+
+#endif

client/TracySysTrace.cpp

@@ -0,0 +1,954 @@
+#include "TracySysTrace.hpp"
+
+#ifdef TRACY_HAS_SYSTEM_TRACING
+
+#  if defined _WIN32 || defined __CYGWIN__
+
+#    ifndef NOMINMAX
+#      define NOMINMAX
+#    endif
+
+#    define INITGUID
+#    include <assert.h>
+#    include <string.h>
+#    include <windows.h>
+#    include <dbghelp.h>
+#    include <evntrace.h>
+#    include <evntcons.h>
+#    include <psapi.h>
+#    include <winternl.h>
+
+#    include "../common/TracyAlloc.hpp"
+#    include "../common/TracySystem.hpp"
+#    include "TracyProfiler.hpp"
+
+namespace tracy
+{
+
+DEFINE_GUID ( /* ce1dbfb4-137e-4da6-87b0-3f59aa102cbc */
+    PerfInfoGuid,
+    0xce1dbfb4,
+    0x137e,
+    0x4da6,
+    0x87, 0xb0, 0x3f, 0x59, 0xaa, 0x10, 0x2c, 0xbc
+);
+
+static TRACEHANDLE s_traceHandle;
+static TRACEHANDLE s_traceHandle2;
+static EVENT_TRACE_PROPERTIES* s_prop;
+static DWORD s_pid;
+
+struct CSwitch
+{
+    uint32_t    newThreadId;
+    uint32_t    oldThreadId;
+    int8_t      newThreadPriority;
+    int8_t      oldThreadPriority;
+    uint8_t     previousCState;
+    int8_t      spareByte;
+    int8_t      oldThreadWaitReason;
+    int8_t      oldThreadWaitMode;
+    int8_t      oldThreadState;
+    int8_t      oldThreadWaitIdealProcessor;
+    uint32_t    newThreadWaitTime;
+    uint32_t    reserved;
+};
+
+struct ReadyThread
+{
+    uint32_t    threadId;
+    int8_t      adjustReason;
+    int8_t      adjustIncrement;
+    int8_t      flag;
+    int8_t      reserverd;
+};
+
+struct ThreadTrace
+{
+    uint32_t processId;
+    uint32_t threadId;
+    uint32_t stackBase;
+    uint32_t stackLimit;
+    uint32_t userStackBase;
+    uint32_t userStackLimit;
+    uint32_t startAddr;
+    uint32_t win32StartAddr;
+    uint32_t tebBase;
+    uint32_t subProcessTag;
+};
+
+struct StackWalkEvent
+{
+    uint64_t eventTimeStamp;
+    uint32_t stackProcess;
+    uint32_t stackThread;
+    uint64_t stack[192];
+};
+
+#ifdef __CYGWIN__
+extern "C" typedef DWORD (WINAPI *t_GetProcessIdOfThread)( HANDLE );
+extern "C" typedef DWORD (WINAPI *t_GetProcessImageFileNameA)( HANDLE, LPSTR, DWORD );
+extern "C" ULONG WMIAPI TraceSetInformation(TRACEHANDLE SessionHandle, TRACE_INFO_CLASS InformationClass, PVOID TraceInformation, ULONG InformationLength);
+t_GetProcessIdOfThread GetProcessIdOfThread = (t_GetProcessIdOfThread)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "GetProcessIdOfThread" );
+t_GetProcessImageFileNameA GetProcessImageFileNameA = (t_GetProcessImageFileNameA)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "K32GetProcessImageFileNameA" );
+#endif
+
+extern "C" typedef NTSTATUS (WINAPI *t_NtQueryInformationThread)( HANDLE, THREADINFOCLASS, PVOID, ULONG, PULONG );
+extern "C" typedef BOOL (WINAPI *t_EnumProcessModules)( HANDLE, HMODULE*, DWORD, LPDWORD );
+extern "C" typedef BOOL (WINAPI *t_GetModuleInformation)( HANDLE, HMODULE, LPMODULEINFO, DWORD );
+extern "C" typedef DWORD (WINAPI *t_GetModuleBaseNameA)( HANDLE, HMODULE, LPSTR, DWORD );
+extern "C" typedef HRESULT (WINAPI *t_GetThreadDescription)( HANDLE, PWSTR* );
+
+t_NtQueryInformationThread NtQueryInformationThread = (t_NtQueryInformationThread)GetProcAddress( GetModuleHandleA( "ntdll.dll" ), "NtQueryInformationThread" );
+t_EnumProcessModules _EnumProcessModules = (t_EnumProcessModules)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "K32EnumProcessModules" );
+t_GetModuleInformation _GetModuleInformation = (t_GetModuleInformation)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "K32GetModuleInformation" );
+t_GetModuleBaseNameA _GetModuleBaseNameA = (t_GetModuleBaseNameA)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "K32GetModuleBaseNameA" );
+
+static t_GetThreadDescription _GetThreadDescription = 0;
+
+
+void WINAPI EventRecordCallback( PEVENT_RECORD record )
+{
+#ifdef TRACY_ON_DEMAND
+    if( !GetProfiler().IsConnected() ) return;
+#endif
+
+    const auto& hdr = record->EventHeader;
+    switch( hdr.ProviderId.Data1 )
+    {
+    case 0x3d6fa8d1:    // Thread Guid
+        if( hdr.EventDescriptor.Opcode == 36 )
+        {
+            const auto cswitch = (const CSwitch*)record->UserData;
+
+            TracyLfqPrepare( QueueType::ContextSwitch );
+            MemWrite( &item->contextSwitch.time, hdr.TimeStamp.QuadPart );
+            memcpy( &item->contextSwitch.oldThread, &cswitch->oldThreadId, sizeof( cswitch->oldThreadId ) );
+            memcpy( &item->contextSwitch.newThread, &cswitch->newThreadId, sizeof( cswitch->newThreadId ) );
+            memset( ((char*)&item->contextSwitch.oldThread)+4, 0, 4 );
+            memset( ((char*)&item->contextSwitch.newThread)+4, 0, 4 );
+            MemWrite( &item->contextSwitch.cpu, record->BufferContext.ProcessorNumber );
+            MemWrite( &item->contextSwitch.reason, cswitch->oldThreadWaitReason );
+            MemWrite( &item->contextSwitch.state, cswitch->oldThreadState );
+            TracyLfqCommit;
+        }
+        else if( hdr.EventDescriptor.Opcode == 50 )
+        {
+            const auto rt = (const ReadyThread*)record->UserData;
+
+            TracyLfqPrepare( QueueType::ThreadWakeup );
+            MemWrite( &item->threadWakeup.time, hdr.TimeStamp.QuadPart );
+            memcpy( &item->threadWakeup.thread, &rt->threadId, sizeof( rt->threadId ) );
+            memset( ((char*)&item->threadWakeup.thread)+4, 0, 4 );
+            TracyLfqCommit;
+        }
+        else if( hdr.EventDescriptor.Opcode == 1 || hdr.EventDescriptor.Opcode == 3 )
+        {
+            const auto tt = (const ThreadTrace*)record->UserData;
+
+            uint64_t tid = tt->threadId;
+            if( tid == 0 ) return;
+            uint64_t pid = tt->processId;
+            TracyLfqPrepare( QueueType::TidToPid );
+            MemWrite( &item->tidToPid.tid, tid );
+            MemWrite( &item->tidToPid.pid, pid );
+            TracyLfqCommit;
+        }
+        break;
+    case 0xdef2fe46:    // StackWalk Guid
+        if( hdr.EventDescriptor.Opcode == 32 )
+        {
+            const auto sw = (const StackWalkEvent*)record->UserData;
+            if( sw->stackProcess == s_pid && ( sw->stack[0] & 0x8000000000000000 ) == 0 )
+            {
+                const uint64_t sz = ( record->UserDataLength - 16 ) / 8;
+                if( sz > 0 )
+                {
+                    auto trace = (uint64_t*)tracy_malloc( ( 1 + sz ) * sizeof( uint64_t ) );
+                    memcpy( trace, &sz, sizeof( uint64_t ) );
+                    memcpy( trace+1, sw->stack, sizeof( uint64_t ) * sz );
+                    TracyLfqPrepare( QueueType::CallstackSample );
+                    MemWrite( &item->callstackSample.time, sw->eventTimeStamp );
+                    MemWrite( &item->callstackSample.thread, (uint64_t)sw->stackThread );
+                    MemWrite( &item->callstackSample.ptr, (uint64_t)trace );
+                    TracyLfqCommit;
+                }
+            }
+        }
+        break;
+    default:
+        break;
+    }
+}
+
+bool SysTraceStart( int64_t& samplingPeriod )
+{
+    if( !_GetThreadDescription ) _GetThreadDescription = (t_GetThreadDescription)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "GetThreadDescription" );
+
+    s_pid = GetCurrentProcessId();
+
+#if defined _WIN64
+    constexpr bool isOs64Bit = true;
+#else
+    BOOL _iswow64;
+    IsWow64Process( GetCurrentProcess(), &_iswow64 );
+    const bool isOs64Bit = _iswow64;
+#endif
+
+    TOKEN_PRIVILEGES priv = {};
+    priv.PrivilegeCount = 1;
+    priv.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
+    if( LookupPrivilegeValue( nullptr, SE_SYSTEM_PROFILE_NAME, &priv.Privileges[0].Luid ) == 0 ) return false;
+
+    HANDLE pt;
+    if( OpenProcessToken( GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES, &pt ) == 0 ) return false;
+    const auto adjust = AdjustTokenPrivileges( pt, FALSE, &priv, 0, nullptr, nullptr );
+    CloseHandle( pt );
+    if( adjust == 0 ) return false;
+    const auto status = GetLastError();
+    if( status != ERROR_SUCCESS ) return false;
+
+    if( isOs64Bit )
+    {
+        TRACE_PROFILE_INTERVAL interval = {};
+        interval.Interval = 1250;   // 8 kHz
+        const auto intervalStatus = TraceSetInformation( 0, TraceSampledProfileIntervalInfo, &interval, sizeof( interval ) );
+        if( intervalStatus != ERROR_SUCCESS ) return false;
+        samplingPeriod = 125*1000;
+    }
+
+    const auto psz = sizeof( EVENT_TRACE_PROPERTIES ) + sizeof( KERNEL_LOGGER_NAME );
+    s_prop = (EVENT_TRACE_PROPERTIES*)tracy_malloc( psz );
+    memset( s_prop, 0, sizeof( EVENT_TRACE_PROPERTIES ) );
+    ULONG flags = EVENT_TRACE_FLAG_CSWITCH | EVENT_TRACE_FLAG_DISPATCHER | EVENT_TRACE_FLAG_THREAD;
+    if( isOs64Bit ) flags |= EVENT_TRACE_FLAG_PROFILE;
+    s_prop->EnableFlags = flags;
+    s_prop->LogFileMode = EVENT_TRACE_REAL_TIME_MODE;
+    s_prop->Wnode.BufferSize = psz;
+    s_prop->Wnode.Flags = WNODE_FLAG_TRACED_GUID;
+#ifdef TRACY_TIMER_QPC
+    s_prop->Wnode.ClientContext = 1;
+#else
+    s_prop->Wnode.ClientContext = 3;
+#endif
+    s_prop->Wnode.Guid = SystemTraceControlGuid;
+    s_prop->BufferSize = 1024;
+    s_prop->LoggerNameOffset = sizeof( EVENT_TRACE_PROPERTIES );
+    memcpy( ((char*)s_prop) + sizeof( EVENT_TRACE_PROPERTIES ), KERNEL_LOGGER_NAME, sizeof( KERNEL_LOGGER_NAME ) );
+
+    auto backup = tracy_malloc( psz );
+    memcpy( backup, s_prop, psz );
+
+    const auto controlStatus = ControlTrace( 0, KERNEL_LOGGER_NAME, s_prop, EVENT_TRACE_CONTROL_STOP );
+    if( controlStatus != ERROR_SUCCESS && controlStatus != ERROR_WMI_INSTANCE_NOT_FOUND )
+    {
+        tracy_free( s_prop );
+        return false;
+    }
+
+    memcpy( s_prop, backup, psz );
+    tracy_free( backup );
+
+    const auto startStatus = StartTrace( &s_traceHandle, KERNEL_LOGGER_NAME, s_prop );
+    if( startStatus != ERROR_SUCCESS )
+    {
+        tracy_free( s_prop );
+        return false;
+    }
+
+    if( isOs64Bit )
+    {
+        CLASSIC_EVENT_ID stackId;
+        stackId.EventGuid = PerfInfoGuid;
+        stackId.Type = 46;
+        const auto stackStatus = TraceSetInformation( s_traceHandle, TraceStackTracingInfo, &stackId, sizeof( stackId ) );
+        if( stackStatus != ERROR_SUCCESS )
+        {
+            tracy_free( s_prop );
+            return false;
+        }
+    }
+
+#ifdef UNICODE
+    WCHAR KernelLoggerName[sizeof( KERNEL_LOGGER_NAME )];
+#else
+    char KernelLoggerName[sizeof( KERNEL_LOGGER_NAME )];
+#endif
+    memcpy( KernelLoggerName, KERNEL_LOGGER_NAME, sizeof( KERNEL_LOGGER_NAME ) );
+    EVENT_TRACE_LOGFILE log = {};
+    log.LoggerName = KernelLoggerName;
+    log.ProcessTraceMode = PROCESS_TRACE_MODE_REAL_TIME | PROCESS_TRACE_MODE_EVENT_RECORD | PROCESS_TRACE_MODE_RAW_TIMESTAMP;
+    log.EventRecordCallback = EventRecordCallback;
+
+    s_traceHandle2 = OpenTrace( &log );
+    if( s_traceHandle2 == (TRACEHANDLE)INVALID_HANDLE_VALUE )
+    {
+        CloseTrace( s_traceHandle );
+        tracy_free( s_prop );
+        return false;
+    }
+
+    return true;
+}
+
+void SysTraceStop()
+{
+    CloseTrace( s_traceHandle2 );
+    CloseTrace( s_traceHandle );
+}
+
+void SysTraceWorker( void* ptr )
+{
+    SetThreadName( "Tracy SysTrace" );
+    ProcessTrace( &s_traceHandle2, 1, 0, 0 );
+    ControlTrace( 0, KERNEL_LOGGER_NAME, s_prop, EVENT_TRACE_CONTROL_STOP );
+    tracy_free( s_prop );
+}
+
+void SysTraceSendExternalName( uint64_t thread )
+{
+    bool threadSent = false;
+    auto hnd = OpenThread( THREAD_QUERY_INFORMATION, FALSE, DWORD( thread ) );
+    if( hnd == 0 )
+    {
+        hnd = OpenThread( THREAD_QUERY_LIMITED_INFORMATION, FALSE, DWORD( thread ) );
+    }
+    if( hnd != 0 )
+    {
+        PWSTR tmp;
+        _GetThreadDescription( hnd, &tmp );
+        char buf[256];
+        if( tmp )
+        {
+            auto ret = wcstombs( buf, tmp, 256 );
+            if( ret != 0 )
+            {
+                GetProfiler().SendString( thread, buf, QueueType::ExternalThreadName );
+                threadSent = true;
+            }
+        }
+        const auto pid = GetProcessIdOfThread( hnd );
+        if( !threadSent && NtQueryInformationThread && _EnumProcessModules && _GetModuleInformation && _GetModuleBaseNameA )
+        {
+            void* ptr;
+            ULONG retlen;
+            auto status = NtQueryInformationThread( hnd, (THREADINFOCLASS)9 /*ThreadQuerySetWin32StartAddress*/, &ptr, sizeof( &ptr ), &retlen );
+            if( status == 0 )
+            {
+                const auto phnd = OpenProcess( PROCESS_QUERY_INFORMATION | PROCESS_VM_READ, FALSE, pid );
+                if( phnd != INVALID_HANDLE_VALUE )
+                {
+                    HMODULE modules[1024];
+                    DWORD needed;
+                    if( _EnumProcessModules( phnd, modules, 1024 * sizeof( HMODULE ), &needed ) != 0 )
+                    {
+                        const auto sz = std::min( DWORD( needed / sizeof( HMODULE ) ), DWORD( 1024 ) );
+                        for( DWORD i=0; i<sz; i++ )
+                        {
+                            MODULEINFO info;
+                            if( _GetModuleInformation( phnd, modules[i], &info, sizeof( info ) ) != 0 )
+                            {
+                                if( (uint64_t)ptr >= (uint64_t)info.lpBaseOfDll && (uint64_t)ptr <= (uint64_t)info.lpBaseOfDll + (uint64_t)info.SizeOfImage )
+                                {
+                                    char buf2[1024];
+                                    if( _GetModuleBaseNameA( phnd, modules[i], buf2, 1024 ) != 0 )
+                                    {
+                                        GetProfiler().SendString( thread, buf2, QueueType::ExternalThreadName );
+                                        threadSent = true;
+                                    }
+                                }
+                            }
+                        }
+                    }
+                    CloseHandle( phnd );
+                }
+            }
+        }
+        CloseHandle( hnd );
+        if( !threadSent )
+        {
+            GetProfiler().SendString( thread, "???", QueueType::ExternalThreadName );
+            threadSent = true;
+        }
+        if( pid != 0 )
+        {
+            {
+                uint64_t _pid = pid;
+                TracyLfqPrepare( QueueType::TidToPid );
+                MemWrite( &item->tidToPid.tid, thread );
+                MemWrite( &item->tidToPid.pid, _pid );
+                TracyLfqCommit;
+            }
+            if( pid == 4 )
+            {
+                GetProfiler().SendString( thread, "System", QueueType::ExternalName );
+                return;
+            }
+            else
+            {
+                const auto phnd = OpenProcess( PROCESS_QUERY_LIMITED_INFORMATION, FALSE, pid );
+                if( phnd != INVALID_HANDLE_VALUE )
+                {
+                    char buf2[1024];
+                    const auto sz = GetProcessImageFileNameA( phnd, buf2, 1024 );
+                    CloseHandle( phnd );
+                    if( sz != 0 )
+                    {
+                        auto ptr = buf2 + sz - 1;
+                        while( ptr > buf2 && *ptr != '\\' ) ptr--;
+                        if( *ptr == '\\' ) ptr++;
+                        GetProfiler().SendString( thread, ptr, QueueType::ExternalName );
+                        return;
+                    }
+                }
+            }
+        }
+    }
+
+    if( !threadSent )
+    {
+        GetProfiler().SendString( thread, "???", QueueType::ExternalThreadName );
+    }
+    GetProfiler().SendString( thread, "???", QueueType::ExternalName );
+}
+
+}
+
+#  elif defined __linux__
+
+#    include <sys/types.h>
+#    include <sys/stat.h>
+#    include <sys/wait.h>
+#    include <fcntl.h>
+#    include <inttypes.h>
+#    include <limits>
+#    include <poll.h>
+#    include <stdio.h>
+#    include <stdlib.h>
+#    include <string.h>
+#    include <unistd.h>
+#    include <atomic>
+
+#    include "TracyProfiler.hpp"
+
+#    ifdef __ANDROID__
+#      include "TracySysTracePayload.hpp"
+#    endif
+
+namespace tracy
+{
+
+static const char BasePath[] = "/sys/kernel/debug/tracing/";
+static const char TracingOn[] = "tracing_on";
+static const char CurrentTracer[] = "current_tracer";
+static const char TraceOptions[] = "trace_options";
+static const char TraceClock[] = "trace_clock";
+static const char SchedSwitch[] = "events/sched/sched_switch/enable";
+static const char SchedWakeup[] = "events/sched/sched_wakeup/enable";
+static const char BufferSizeKb[] = "buffer_size_kb";
+static const char TracePipe[] = "trace_pipe";
+
+static std::atomic<bool> traceActive { false };
+
+#ifdef __ANDROID__
+static bool TraceWrite( const char* path, size_t psz, const char* val, size_t vsz )
+{
+    char tmp[256];
+    sprintf( tmp, "su -c 'echo \"%s\" > %s%s'", val, BasePath, path );
+    return system( tmp ) == 0;
+}
+#else
+static bool TraceWrite( const char* path, size_t psz, const char* val, size_t vsz )
+{
+    char tmp[256];
+    memcpy( tmp, BasePath, sizeof( BasePath ) - 1 );
+    memcpy( tmp + sizeof( BasePath ) - 1, path, psz );
+
+    int fd = open( tmp, O_WRONLY );
+    if( fd < 0 ) return false;
+
+    for(;;)
+    {
+        ssize_t cnt = write( fd, val, vsz );
+        if( cnt == (ssize_t)vsz )
+        {
+            close( fd );
+            return true;
+        }
+        if( cnt < 0 )
+        {
+            close( fd );
+            return false;
+        }
+        vsz -= cnt;
+        val += cnt;
+    }
+}
+#endif
+
+#ifdef __ANDROID__
+void SysTraceInjectPayload()
+{
+    int pipefd[2];
+    if( pipe( pipefd ) == 0 )
+    {
+        const auto pid = fork();
+        if( pid == 0 )
+        {
+            // child
+            close( pipefd[1] );
+            if( dup2( pipefd[0], STDIN_FILENO ) >= 0 )
+            {
+                close( pipefd[0] );
+                execlp( "su", "su", "-c", "cat > /data/tracy_systrace", (char*)nullptr );
+                exit( 1 );
+            }
+        }
+        else if( pid > 0 )
+        {
+            // parent
+            close( pipefd[0] );
+
+#ifdef __aarch64__
+            write( pipefd[1], tracy_systrace_aarch64_data, tracy_systrace_aarch64_size );
+#else
+            write( pipefd[1], tracy_systrace_armv7_data, tracy_systrace_armv7_size );
+#endif
+            close( pipefd[1] );
+            waitpid( pid, nullptr, 0 );
+
+            system( "su -c 'chmod 700 /data/tracy_systrace'" );
+        }
+    }
+}
+#endif
+
+bool SysTraceStart( int64_t& samplingPeriod )
+{
+    if( !TraceWrite( TracingOn, sizeof( TracingOn ), "0", 2 ) ) return false;
+    if( !TraceWrite( CurrentTracer, sizeof( CurrentTracer ), "nop", 4 ) ) return false;
+    TraceWrite( TraceOptions, sizeof( TraceOptions ), "norecord-cmd", 13 );
+    TraceWrite( TraceOptions, sizeof( TraceOptions ), "norecord-tgid", 14 );
+    TraceWrite( TraceOptions, sizeof( TraceOptions ), "noirq-info", 11 );
+    TraceWrite( TraceOptions, sizeof( TraceOptions ), "noannotate", 11 );
+#if defined TRACY_HW_TIMER && ( defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64 )
+    if( !TraceWrite( TraceClock, sizeof( TraceClock ), "x86-tsc", 8 ) ) return false;
+#elif __ARM_ARCH >= 6
+    if( !TraceWrite( TraceClock, sizeof( TraceClock ), "mono_raw", 9 ) ) return false;
+#endif
+    if( !TraceWrite( SchedSwitch, sizeof( SchedSwitch ), "1", 2 ) ) return false;
+    if( !TraceWrite( SchedWakeup, sizeof( SchedWakeup ), "1", 2 ) ) return false;
+    if( !TraceWrite( BufferSizeKb, sizeof( BufferSizeKb ), "512", 4 ) ) return false;
+
+#if defined __ANDROID__ && ( defined __aarch64__ || defined __ARM_ARCH )
+    SysTraceInjectPayload();
+#endif
+
+    if( !TraceWrite( TracingOn, sizeof( TracingOn ), "1", 2 ) ) return false;
+    traceActive.store( true, std::memory_order_relaxed );
+
+    return true;
+}
+
+void SysTraceStop()
+{
+    TraceWrite( TracingOn, sizeof( TracingOn ), "0", 2 );
+    traceActive.store( false, std::memory_order_relaxed );
+}
+
+static uint64_t ReadNumber( const char*& ptr )
+{
+    uint64_t val = 0;
+    for(;;)
+    {
+        if( *ptr >= '0' && *ptr <= '9' )
+        {
+            val = val * 10 + ( *ptr - '0' );
+            ptr++;
+        }
+        else
+        {
+            return val;
+        }
+    }
+}
+
+static uint8_t ReadState( char state )
+{
+    switch( state )
+    {
+    case 'D': return 101;
+    case 'I': return 102;
+    case 'R': return 103;
+    case 'S': return 104;
+    case 'T': return 105;
+    case 't': return 106;
+    case 'W': return 107;
+    case 'X': return 108;
+    case 'Z': return 109;
+    default: return 100;
+    }
+}
+
+#if defined __ANDROID__ && defined __ANDROID_API__ && __ANDROID_API__ < 18
+/*-
+ * Copyright (c) 2011 The NetBSD Foundation, Inc.
+ * All rights reserved.
+ *
+ * This code is derived from software contributed to The NetBSD Foundation
+ * by Christos Zoulas.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
+ * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
+ * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+ * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
+ * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ */
+
+ssize_t getdelim(char **buf, size_t *bufsiz, int delimiter, FILE *fp)
+{
+	char *ptr, *eptr;
+
+	if (*buf == NULL || *bufsiz == 0) {
+		*bufsiz = BUFSIZ;
+		if ((*buf = (char*)malloc(*bufsiz)) == NULL)
+			return -1;
+	}
+
+	for (ptr = *buf, eptr = *buf + *bufsiz;;) {
+		int c = fgetc(fp);
+		if (c == -1) {
+			if (feof(fp))
+				return ptr == *buf ? -1 : ptr - *buf;
+			else
+				return -1;
+		}
+		*ptr++ = c;
+		if (c == delimiter) {
+			*ptr = '\0';
+			return ptr - *buf;
+		}
+		if (ptr + 2 >= eptr) {
+			char *nbuf;
+			size_t nbufsiz = *bufsiz * 2;
+			ssize_t d = ptr - *buf;
+			if ((nbuf = (char*)realloc(*buf, nbufsiz)) == NULL)
+				return -1;
+			*buf = nbuf;
+			*bufsiz = nbufsiz;
+			eptr = nbuf + nbufsiz;
+			ptr = nbuf + d;
+		}
+	}
+}
+
+ssize_t getline(char **buf, size_t *bufsiz, FILE *fp)
+{
+	return getdelim(buf, bufsiz, '\n', fp);
+}
+#endif
+
+static void HandleTraceLine( const char* line )
+{
+    line += 23;
+    while( *line != '[' ) line++;
+    line++;
+    const auto cpu = (uint8_t)ReadNumber( line );
+    line++;      // ']'
+    while( *line == ' ' ) line++;
+
+#if defined TRACY_HW_TIMER && ( defined __i386 || defined _M_IX86 || defined __x86_64__ || defined _M_X64 )
+    const auto time = ReadNumber( line );
+#elif __ARM_ARCH >= 6
+    const auto ts = ReadNumber( line );
+    line++;      // '.'
+    const auto tus = ReadNumber( line );
+    const auto time = ts * 1000000000ll + tus * 1000ll;
+#endif
+
+    line += 2;   // ': '
+    if( memcmp( line, "sched_switch", 12 ) == 0 )
+    {
+        line += 14;
+
+        while( memcmp( line, "prev_pid", 8 ) != 0 ) line++;
+        line += 9;
+
+        const auto oldPid = ReadNumber( line );
+        line++;
+
+        while( memcmp( line, "prev_state", 10 ) != 0 ) line++;
+        line += 11;
+
+        const auto oldState = (uint8_t)ReadState( *line );
+        line += 5;
+
+        while( memcmp( line, "next_pid", 8 ) != 0 ) line++;
+        line += 9;
+
+        const auto newPid = ReadNumber( line );
+
+        uint8_t reason = 100;
+
+        TracyLfqPrepare( QueueType::ContextSwitch );
+        MemWrite( &item->contextSwitch.time, time );
+        MemWrite( &item->contextSwitch.oldThread, oldPid );
+        MemWrite( &item->contextSwitch.newThread, newPid );
+        MemWrite( &item->contextSwitch.cpu, cpu );
+        MemWrite( &item->contextSwitch.reason, reason );
+        MemWrite( &item->contextSwitch.state, oldState );
+        TracyLfqCommit;
+    }
+    else if( memcmp( line, "sched_wakeup", 12 ) == 0 )
+    {
+        line += 14;
+
+        while( memcmp( line, "pid", 3 ) != 0 ) line++;
+        line += 4;
+
+        const auto pid = ReadNumber( line );
+
+        TracyLfqPrepare( QueueType::ThreadWakeup );
+        MemWrite( &item->threadWakeup.time, time );
+        MemWrite( &item->threadWakeup.thread, pid );
+        TracyLfqCommit;
+    }
+}
+
+#ifdef __ANDROID__
+static void ProcessTraceLines( int fd )
+{
+    // Linux pipe buffer is 64KB, additional 1KB is for unfinished lines
+    char* buf = (char*)tracy_malloc( (64+1)*1024 );
+    char* line = buf;
+
+    for(;;)
+    {
+        if( !traceActive.load( std::memory_order_relaxed ) ) break;
+
+        const auto rd = read( fd, line, 64*1024 );
+        if( rd <= 0 ) break;
+
+#ifdef TRACY_ON_DEMAND
+        if( !GetProfiler().IsConnected() )
+        {
+            if( rd < 64*1024 )
+            {
+                assert( line[rd-1] == '\n' );
+                line = buf;
+                std::this_thread::sleep_for( std::chrono::milliseconds( 10 ) );
+            }
+            else
+            {
+                const auto end = line + rd;
+                line = end - 1;
+                while( line > buf && *line != '\n' ) line--;
+                if( line > buf )
+                {
+                    line++;
+                    const auto lsz = end - line;
+                    memmove( buf, line, lsz );
+                    line = buf + lsz;
+                }
+            }
+            continue;
+        }
+#endif
+
+        const auto end = line + rd;
+        line = buf;
+        for(;;)
+        {
+            auto next = line;
+            while( next < end && *next != '\n' ) next++;
+            next++;
+            if( next >= end )
+            {
+                const auto lsz = end - line;
+                memmove( buf, line, lsz );
+                line = buf + lsz;
+                break;
+            }
+
+            HandleTraceLine( line );
+            line = next;
+        }
+        if( rd < 64*1024 )
+        {
+            std::this_thread::sleep_for( std::chrono::milliseconds( 10 ) );
+        }
+    }
+
+    tracy_free( buf );
+}
+
+void SysTraceWorker( void* ptr )
+{
+    SetThreadName( "Tracy SysTrace" );
+    int pipefd[2];
+    if( pipe( pipefd ) == 0 )
+    {
+        const auto pid = fork();
+        if( pid == 0 )
+        {
+            // child
+            close( pipefd[0] );
+            dup2( pipefd[1], STDERR_FILENO );
+            if( dup2( pipefd[1], STDOUT_FILENO ) >= 0 )
+            {
+                close( pipefd[1] );
+#if defined __ANDROID__ && ( defined __aarch64__ || defined __ARM_ARCH )
+                execlp( "su", "su", "-c", "/data/tracy_systrace", (char*)nullptr );
+#endif
+                execlp( "su", "su", "-c", "cat /sys/kernel/debug/tracing/trace_pipe", (char*)nullptr );
+                exit( 1 );
+            }
+        }
+        else if( pid > 0 )
+        {
+            // parent
+            close( pipefd[1] );
+            ProcessTraceLines( pipefd[0] );
+            close( pipefd[0] );
+        }
+    }
+}
+#else
+static void ProcessTraceLines( int fd )
+{
+    char* buf = (char*)tracy_malloc( 64*1024 );
+
+    struct pollfd pfd;
+    pfd.fd = fd;
+    pfd.events = POLLIN | POLLERR;
+
+    for(;;)
+    {
+        while( poll( &pfd, 1, 0 ) <= 0 )
+        {
+            if( !traceActive.load( std::memory_order_relaxed ) ) break;
+            std::this_thread::sleep_for( std::chrono::milliseconds( 10 ) );
+        }
+
+        const auto rd = read( fd, buf, 64*1024 );
+        if( rd <= 0 ) break;
+
+#ifdef TRACY_ON_DEMAND
+        if( !GetProfiler().IsConnected() ) continue;
+#endif
+
+        auto line = buf;
+        const auto end = buf + rd;
+        for(;;)
+        {
+            auto next = line;
+            while( next < end && *next != '\n' ) next++;
+            if( next == end ) break;
+            assert( *next == '\n' );
+            next++;
+
+            HandleTraceLine( line );
+            line = next;
+        }
+    }
+
+    tracy_free( buf );
+}
+
+void SysTraceWorker( void* ptr )
+{
+    SetThreadName( "Tracy SysTrace" );
+    char tmp[256];
+    memcpy( tmp, BasePath, sizeof( BasePath ) - 1 );
+    memcpy( tmp + sizeof( BasePath ) - 1, TracePipe, sizeof( TracePipe ) );
+
+    int fd = open( tmp, O_RDONLY );
+    if( fd < 0 ) return;
+    ProcessTraceLines( fd );
+    close( fd );
+}
+#endif
+
+void SysTraceSendExternalName( uint64_t thread )
+{
+    FILE* f;
+    char fn[256];
+    sprintf( fn, "/proc/%" PRIu64 "/comm", thread );
+    f = fopen( fn, "rb" );
+    if( f )
+    {
+        char buf[256];
+        const auto sz = fread( buf, 1, 256, f );
+        if( sz > 0 && buf[sz-1] == '\n' ) buf[sz-1] = '\0';
+        GetProfiler().SendString( thread, buf, QueueType::ExternalThreadName );
+        fclose( f );
+    }
+    else
+    {
+        GetProfiler().SendString( thread, "???", QueueType::ExternalThreadName );
+    }
+
+    sprintf( fn, "/proc/%" PRIu64 "/status", thread );
+    f = fopen( fn, "rb" );
+    if( f )
+    {
+        int pid = -1;
+        size_t lsz = 1024;
+        auto line = (char*)malloc( lsz );
+        for(;;)
+        {
+            auto rd = getline( &line, &lsz, f );
+            if( rd <= 0 ) break;
+            if( memcmp( "Tgid:\t", line, 6 ) == 0 )
+            {
+                pid = atoi( line + 6 );
+                break;
+            }
+        }
+        free( line );
+        fclose( f );
+        if( pid >= 0 )
+        {
+            {
+                uint64_t _pid = pid;
+                TracyLfqPrepare( QueueType::TidToPid );
+                MemWrite( &item->tidToPid.tid, thread );
+                MemWrite( &item->tidToPid.pid, _pid );
+                TracyLfqCommit;
+            }
+            sprintf( fn, "/proc/%i/comm", pid );
+            f = fopen( fn, "rb" );
+            if( f )
+            {
+                char buf[256];
+                const auto sz = fread( buf, 1, 256, f );
+                if( sz > 0 && buf[sz-1] == '\n' ) buf[sz-1] = '\0';
+                GetProfiler().SendString( thread, buf, QueueType::ExternalName );
+                fclose( f );
+                return;
+            }
+        }
+    }
+    GetProfiler().SendString( thread, "???", QueueType::ExternalName );
+}
+
+}
+
+#  endif
+
+#endif

client/TracySysTrace.hpp

@@ -0,0 +1,25 @@
+#ifndef __TRACYSYSTRACE_HPP__
+#define __TRACYSYSTRACE_HPP__
+
+#if !defined TRACY_NO_SYSTEM_TRACING && ( defined _WIN32 || defined __CYGWIN__ || defined __linux__ )
+#  define TRACY_HAS_SYSTEM_TRACING
+#endif
+
+#ifdef TRACY_HAS_SYSTEM_TRACING
+
+#include <stdint.h>
+
+namespace tracy
+{
+
+bool SysTraceStart( int64_t& samplingPeriod );
+void SysTraceStop();
+void SysTraceWorker( void* ptr );
+
+void SysTraceSendExternalName( uint64_t thread );
+
+}
+
+#endif
+
+#endif

client/TracySysTracePayload.hpp

@@ -0,0 +1,80 @@
+// File: '/home/wolf/desktop/tracy_systrace.armv7' (1210 bytes)`
+// File: '/home/wolf/desktop/tracy_systrace.aarch64' (1650 bytes)
+
+// Exported using binary_to_compressed_c.cpp
+
+namespace tracy
+{
+
+static const unsigned int tracy_systrace_armv7_size = 1210;
+static const unsigned int tracy_systrace_armv7_data[1212/4] =
+{
+    0x464c457f, 0x00010101, 0x00000000, 0x00000000, 0x00280003, 0x00000001, 0x00000208, 0x00000034, 0x00000000, 0x05000200, 0x00200034, 0x00280007, 
+    0x00000000, 0x00000006, 0x00000034, 0x00000034, 0x00000034, 0x000000e0, 0x000000e0, 0x00000004, 0x00000004, 0x00000003, 0x00000114, 0x00000114, 
+    0x00000114, 0x00000013, 0x00000013, 0x00000004, 0x00000001, 0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x000003ed, 0x000003ed, 0x00000005, 
+    0x00001000, 0x00000001, 0x000003ed, 0x000013ed, 0x000013ed, 0x000000cd, 0x000000cf, 0x00000006, 0x00001000, 0x00000002, 0x000003f0, 0x000013f0, 
+    0x000013f0, 0x000000b8, 0x000000b8, 0x00000006, 0x00000004, 0x6474e551, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000006, 
+    0x00000010, 0x70000001, 0x00000394, 0x00000394, 0x00000394, 0x00000008, 0x00000008, 0x00000004, 0x00000004, 0x7379732f, 0x2f6d6574, 0x2f6e6962, 
+    0x6b6e696c, 0x00007265, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000001, 0x00000000, 0x00000000, 0x00000012, 0x00000016, 0x00000000, 
+    0x00000000, 0x00000012, 0x6f6c6400, 0x006e6570, 0x4342494c, 0x62696c00, 0x732e6c64, 0x6c64006f, 0x006d7973, 0x00000001, 0x00000003, 0x00000001, 
+    0x00000000, 0x00000000, 0x00000000, 0x00000001, 0x00000003, 0x00000002, 0x00000000, 0x00000000, 0x00000001, 0x00020000, 0x00000002, 0x00010001, 
+    0x0000000d, 0x00000010, 0x00000000, 0x00050d63, 0x00020000, 0x00000008, 0x00000000, 0x000014b4, 0x00000116, 0x000014b8, 0x00000216, 0xe52de004, 
+    0xe59fe004, 0xe08fe00e, 0xe5bef008, 0x000012bc, 0xe28fc600, 0xe28cca01, 0xe5bcf2bc, 0xe28fc600, 0xe28cca01, 0xe5bcf2b4, 0xe92d4ff0, 0xe28db01c, 
+    0xe24dd01c, 0xe24dd801, 0xe59f0154, 0xe3a01001, 0xe08f0000, 0xebfffff1, 0xe59f1148, 0xe1a07000, 0xe08f1001, 0xebfffff0, 0xe59f113c, 0xe1a09000, 
+    0xe1a00007, 0xe08f1001, 0xebffffeb, 0xe59f112c, 0xe1a04000, 0xe1a00007, 0xe08f1001, 0xebffffe6, 0xe59f111c, 0xe1a05000, 0xe1a00007, 0xe08f1001, 
+    0xebffffe1, 0xe59f110c, 0xe1a06000, 0xe1a00007, 0xe08f1001, 0xebffffdc, 0xe58d0004, 0xe1a00007, 0xe59f10f4, 0xe08f1001, 0xebffffd7, 0xe1a0a000, 
+    0xe59f00e8, 0xe3a01000, 0xe3a08000, 0xe08f0000, 0xe12fff39, 0xe1a07000, 0xe3700001, 0xca000001, 0xe3a00000, 0xe12fff34, 0xe3a00009, 0xe58d4000, 
+    0xe1cd01b4, 0xe3090680, 0xe3400098, 0xe28d4010, 0xe58d000c, 0xe28d9018, 0xe58d8008, 0xe28d8008, 0xe58d7010, 0xea000003, 0xe1a02000, 0xe3a00001, 
+    0xe1a01009, 0xe12fff3a, 0xe1a00004, 0xe3a01001, 0xe3a02000, 0xe12fff35, 0xe3500000, 0xca000008, 0xe1a00008, 0xe3a01000, 0xe12fff36, 0xe1a00004, 
+    0xe3a01001, 0xe3a02000, 0xe12fff35, 0xe3500001, 0xbafffff6, 0xe59d3004, 0xe1a00007, 0xe1a01009, 0xe3a02801, 0xe12fff33, 0xe3500001, 0xaaffffe5, 
+    0xe59d1000, 0xe3a00000, 0xe12fff31, 0xe24bd01c, 0xe8bd8ff0, 0x00000174, 0x0000016c, 0x0000015d, 0x0000014e, 0x0000013f, 0x00000135, 0x00000126, 
+    0x00000114, 0x7ffffe74, 0x00000001, 0x6362696c, 0x006f732e, 0x6e65706f, 0x69786500, 0x6f700074, 0x6e006c6c, 0x736f6e61, 0x7065656c, 0x61657200, 
+    0x72770064, 0x00657469, 0x7379732f, 0x72656b2f, 0x2f6c656e, 0x75626564, 0x72742f67, 0x6e696361, 0x72742f67, 0x5f656361, 0x65706970, 0x00000000, 
+    0x00000003, 0x000014a8, 0x00000002, 0x00000010, 0x00000017, 0x000001cc, 0x00000014, 0x00000011, 0x00000015, 0x00000000, 0x00000006, 0x00000128, 
+    0x0000000b, 0x00000010, 0x00000005, 0x00000158, 0x0000000a, 0x0000001c, 0x6ffffef5, 0x00000174, 0x00000004, 0x0000018c, 0x00000001, 0x0000000d, 
+    0x0000001e, 0x00000008, 0x6ffffffb, 0x00000001, 0x6ffffff0, 0x000001a4, 0x6ffffffe, 0x000001ac, 0x6fffffff, 0x00000001, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x000001dc, 0x000001dc, 
+};
+
+static const unsigned int tracy_systrace_aarch64_size = 1650;
+static const unsigned int tracy_systrace_aarch64_data[1652/4] =
+{
+    0x464c457f, 0x00010102, 0x00000000, 0x00000000, 0x00b70003, 0x00000001, 0x00000300, 0x00000000, 0x00000040, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x00380040, 0x00400006, 0x00000000, 0x00000006, 0x00000005, 0x00000040, 0x00000000, 0x00000040, 0x00000000, 0x00000040, 0x00000000, 
+    0x00000150, 0x00000000, 0x00000150, 0x00000000, 0x00000008, 0x00000000, 0x00000003, 0x00000004, 0x00000190, 0x00000000, 0x00000190, 0x00000000, 
+    0x00000190, 0x00000000, 0x00000015, 0x00000000, 0x00000015, 0x00000000, 0x00000001, 0x00000000, 0x00000001, 0x00000005, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x000004d1, 0x00000000, 0x000004d1, 0x00000000, 0x00010000, 0x00000000, 0x00000001, 0x00000006, 
+    0x000004d8, 0x00000000, 0x000104d8, 0x00000000, 0x000104d8, 0x00000000, 0x0000019a, 0x00000000, 0x000001a0, 0x00000000, 0x00010000, 0x00000000, 
+    0x00000002, 0x00000006, 0x000004d8, 0x00000000, 0x000104d8, 0x00000000, 0x000104d8, 0x00000000, 0x00000170, 0x00000000, 0x00000170, 0x00000000, 
+    0x00000008, 0x00000000, 0x6474e551, 0x00000006, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x00000010, 0x00000000, 0x7379732f, 0x2f6d6574, 0x2f6e6962, 0x6b6e696c, 0x34367265, 0x00000000, 0x00000001, 0x00000004, 
+    0x00000003, 0x00000000, 0x00000000, 0x00000000, 0x00000002, 0x00000000, 0x00000001, 0x00000001, 0x00000001, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x000a0003, 0x00000300, 0x00000000, 
+    0x00000000, 0x00000000, 0x0000000a, 0x00000012, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000011, 0x00000012, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x62696c00, 0x732e6c64, 0x6c64006f, 0x6e65706f, 0x736c6400, 0x4c006d79, 0x00434249, 0x00000000, 0x00020002, 0x00000000, 
+    0x00010001, 0x00000001, 0x00000010, 0x00000000, 0x00050d63, 0x00020000, 0x00000017, 0x00000000, 0x00010668, 0x00000000, 0x00000402, 0x00000002, 
+    0x00000000, 0x00000000, 0x00010670, 0x00000000, 0x00000402, 0x00000003, 0x00000000, 0x00000000, 0xa9bf7bf0, 0x90000090, 0xf9433211, 0x91198210, 
+    0xd61f0220, 0xd503201f, 0xd503201f, 0xd503201f, 0x90000090, 0xf9433611, 0x9119a210, 0xd61f0220, 0x90000090, 0xf9433a11, 0x9119c210, 0xd61f0220, 
+    0xf81b0ffc, 0xa9015ff8, 0xa90257f6, 0xa9034ff4, 0xa9047bfd, 0x910103fd, 0xd14043ff, 0xd10043ff, 0x90000000, 0x91120000, 0x320003e1, 0x97ffffed, 
+    0x90000001, 0x91122021, 0xaa0003f7, 0x97ffffed, 0x90000001, 0xaa0003f8, 0x91123421, 0xaa1703e0, 0x97ffffe8, 0x90000001, 0xaa0003f3, 0x91124821, 
+    0xaa1703e0, 0x97ffffe3, 0x90000001, 0xaa0003f4, 0x91125c21, 0xaa1703e0, 0x97ffffde, 0x90000001, 0xaa0003f5, 0x91128421, 0xaa1703e0, 0x97ffffd9, 
+    0x90000001, 0xaa0003f6, 0x91129821, 0xaa1703e0, 0x97ffffd4, 0xaa0003f7, 0x90000000, 0x9112b000, 0x2a1f03e1, 0xd63f0300, 0x2a0003f8, 0x36f80060, 
+    0x2a1f03e0, 0xd63f0260, 0x90000008, 0x3dc11d00, 0x52800128, 0xb81c83b8, 0x781cc3a8, 0x3d8003e0, 0x14000005, 0x93407c02, 0x320003e0, 0x910043e1, 
+    0xd63f02e0, 0xd100e3a0, 0x320003e1, 0x2a1f03e2, 0xd63f0280, 0x7100001f, 0x5400014c, 0x910003e0, 0xaa1f03e1, 0xd63f02a0, 0xd100e3a0, 0x320003e1, 
+    0x2a1f03e2, 0xd63f0280, 0x7100041f, 0x54ffff0b, 0x910043e1, 0x321003e2, 0x2a1803e0, 0xd63f02c0, 0x7100041f, 0x54fffd0a, 0x2a1f03e0, 0xd63f0260, 
+    0x914043ff, 0x910043ff, 0xa9447bfd, 0xa9434ff4, 0xa94257f6, 0xa9415ff8, 0xf84507fc, 0xd65f03c0, 0x00000000, 0x00000000, 0x00989680, 0x00000000, 
+    0x6362696c, 0x006f732e, 0x6e65706f, 0x69786500, 0x6f700074, 0x6e006c6c, 0x736f6e61, 0x7065656c, 0x61657200, 0x72770064, 0x00657469, 0x7379732f, 
+    0x72656b2f, 0x2f6c656e, 0x75626564, 0x72742f67, 0x6e696361, 0x72742f67, 0x5f656361, 0x65706970, 0x00000000, 0x00000000, 0x00000001, 0x00000000, 
+    0x00000001, 0x00000000, 0x00000004, 0x00000000, 0x000001a8, 0x00000000, 0x6ffffef5, 0x00000000, 0x000001c8, 0x00000000, 0x00000005, 0x00000000, 
+    0x00000248, 0x00000000, 0x00000006, 0x00000000, 0x000001e8, 0x00000000, 0x0000000a, 0x00000000, 0x0000001c, 0x00000000, 0x0000000b, 0x00000000, 
+    0x00000018, 0x00000000, 0x00000015, 0x00000000, 0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00010650, 0x00000000, 0x00000002, 0x00000000, 
+    0x00000030, 0x00000000, 0x00000014, 0x00000000, 0x00000007, 0x00000000, 0x00000017, 0x00000000, 0x00000290, 0x00000000, 0x0000001e, 0x00000000, 
+    0x00000008, 0x00000000, 0x6ffffffb, 0x00000000, 0x00000001, 0x00000000, 0x6ffffffe, 0x00000000, 0x00000270, 0x00000000, 0x6fffffff, 0x00000000, 
+    0x00000001, 0x00000000, 0x6ffffff0, 0x00000000, 0x00000264, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x000104d8, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 
+    0x00000000, 0x00000000, 0x000002c0, 0x00000000, 0x000002c0, 
+};
+
+}

client/TracyThread.hpp

@@ -1,7 +1,7 @@
 #ifndef __TRACYTHREAD_HPP__
 #define __TRACYTHREAD_HPP__
 
-#ifdef _MSC_VER
+#if defined _WIN32 || defined __CYGWIN__
 #  include <windows.h>
 #else
 #  include <pthread.h>
@@ -10,7 +10,7 @@
 namespace tracy
 {
 
-#ifdef _MSC_VER
+#if defined _WIN32 || defined __CYGWIN__
 
 class Thread
 {

client/tracy_rpmalloc.hpp

@@ -1,9 +1,9 @@
-/* rpmalloc.h  -  Memory allocator  -  Public Domain  -  2016 Mattias Jansson / Rampant Pixels
+/* rpmalloc.h  -  Memory allocator  -  Public Domain  -  2016 Mattias Jansson
  *
  * This library provides a cross-platform lock free thread caching malloc implementation in C11.
  * The latest source code is always available at
  *
- * https://github.com/rampantpixels/rpmalloc
+ * https://github.com/mjansson/rpmalloc
  *
  * This library is put in the public domain; you can redistribute it and/or modify it without any restrictions.
  *
@@ -12,51 +12,113 @@
 #pragma once
 
 #include <stddef.h>
+#include "../common/TracyApi.h"
 
 namespace tracy
 {
 
 #if defined(__clang__) || defined(__GNUC__)
-# define RPMALLOC_ATTRIBUTE __attribute__((__malloc__))
-# define RPMALLOC_RESTRICT
+# define RPMALLOC_EXPORT __attribute__((visibility("default")))
+# define RPMALLOC_ALLOCATOR 
+# define RPMALLOC_ATTRIB_MALLOC __attribute__((__malloc__))
+# if defined(__clang_major__) && (__clang_major__ < 4)
+# define RPMALLOC_ATTRIB_ALLOC_SIZE(size)
+# define RPMALLOC_ATTRIB_ALLOC_SIZE2(count, size)
+# else
+# define RPMALLOC_ATTRIB_ALLOC_SIZE(size) __attribute__((alloc_size(size)))
+# define RPMALLOC_ATTRIB_ALLOC_SIZE2(count, size)  __attribute__((alloc_size(count, size)))
+# endif
 # define RPMALLOC_CDECL
 #elif defined(_MSC_VER)
-# define RPMALLOC_ATTRIBUTE
-# define RPMALLOC_RESTRICT __declspec(restrict)
+# define RPMALLOC_EXPORT
+# define RPMALLOC_ALLOCATOR __declspec(allocator) __declspec(restrict)
+# define RPMALLOC_ATTRIB_MALLOC
+# define RPMALLOC_ATTRIB_ALLOC_SIZE(size)
+# define RPMALLOC_ATTRIB_ALLOC_SIZE2(count,size)
 # define RPMALLOC_CDECL __cdecl
 #else
-# define RPMALLOC_ATTRIBUTE
-# define RPMALLOC_RESTRICT
+# define RPMALLOC_EXPORT
+# define RPMALLOC_ALLOCATOR
+# define RPMALLOC_ATTRIB_MALLOC
+# define RPMALLOC_ATTRIB_ALLOC_SIZE(size)
+# define RPMALLOC_ATTRIB_ALLOC_SIZE2(count,size)
 # define RPMALLOC_CDECL
 #endif
 
+//! Define RPMALLOC_CONFIGURABLE to enable configuring sizes
+#ifndef RPMALLOC_CONFIGURABLE
+#define RPMALLOC_CONFIGURABLE 0
+#endif
+
 //! Flag to rpaligned_realloc to not preserve content in reallocation
 #define RPMALLOC_NO_PRESERVE    1
 
 typedef struct rpmalloc_global_statistics_t {
-	//! Current amount of virtual memory mapped (only if ENABLE_STATISTICS=1)
+	//! Current amount of virtual memory mapped, all of which might not have been committed (only if ENABLE_STATISTICS=1)
 	size_t mapped;
-	//! Current amount of memory in global caches for small and medium sizes (<64KiB)
+	//! Peak amount of virtual memory mapped, all of which might not have been committed (only if ENABLE_STATISTICS=1)
+	size_t mapped_peak;
+	//! Current amount of memory in global caches for small and medium sizes (<32KiB)
 	size_t cached;
-	//! Total amount of memory mapped (only if ENABLE_STATISTICS=1)
+	//! Current amount of memory allocated in huge allocations, i.e larger than LARGE_SIZE_LIMIT which is 2MiB by default (only if ENABLE_STATISTICS=1)
+	size_t huge_alloc;
+	//! Peak amount of memory allocated in huge allocations, i.e larger than LARGE_SIZE_LIMIT which is 2MiB by default (only if ENABLE_STATISTICS=1)
+	size_t huge_alloc_peak;
+	//! Total amount of memory mapped since initialization (only if ENABLE_STATISTICS=1)
 	size_t mapped_total;
-	//! Total amount of memory unmapped (only if ENABLE_STATISTICS=1)
+	//! Total amount of memory unmapped since initialization  (only if ENABLE_STATISTICS=1)
 	size_t unmapped_total;
 } rpmalloc_global_statistics_t;
 
 typedef struct rpmalloc_thread_statistics_t {
-	//! Current number of bytes available for allocation from active spans
-	size_t active;
-	//! Current number of bytes available in thread size class caches
+	//! Current number of bytes available in thread size class caches for small and medium sizes (<32KiB)
 	size_t sizecache;
-	//! Current number of bytes available in thread span caches
+	//! Current number of bytes available in thread span caches for small and medium sizes (<32KiB)
 	size_t spancache;
-	//! Current number of bytes in pending deferred deallocations
-	size_t deferred;
-	//! Total number of bytes transitioned from thread cache to global cache
+	//! Total number of bytes transitioned from thread cache to global cache (only if ENABLE_STATISTICS=1)
 	size_t thread_to_global;
-	//! Total number of bytes transitioned from global cache to thread cache
+	//! Total number of bytes transitioned from global cache to thread cache (only if ENABLE_STATISTICS=1)
 	size_t global_to_thread;
+	//! Per span count statistics (only if ENABLE_STATISTICS=1)
+	struct {
+		//! Currently used number of spans
+		size_t current;
+		//! High water mark of spans used
+		size_t peak;
+		//! Number of spans transitioned to global cache
+		size_t to_global;
+		//! Number of spans transitioned from global cache
+		size_t from_global;
+		//! Number of spans transitioned to thread cache
+		size_t to_cache;
+		//! Number of spans transitioned from thread cache
+		size_t from_cache;
+		//! Number of spans transitioned to reserved state
+		size_t to_reserved;
+		//! Number of spans transitioned from reserved state
+		size_t from_reserved;
+		//! Number of raw memory map calls (not hitting the reserve spans but resulting in actual OS mmap calls)
+		size_t map_calls;
+	} span_use[32];
+	//! Per size class statistics (only if ENABLE_STATISTICS=1)
+	struct {
+		//! Current number of allocations
+		size_t alloc_current;
+		//! Peak number of allocations
+		size_t alloc_peak;
+		//! Total number of allocations
+		size_t alloc_total;
+		//! Total number of frees
+		size_t free_total;
+		//! Number of spans transitioned to cache
+		size_t spans_to_cache;
+		//! Number of spans transitioned from cache
+		size_t spans_from_cache;
+		//! Number of spans transitioned from reserved state
+		size_t spans_from_reserved;
+		//! Number of raw memory map calls (not hitting the reserve spans but resulting in actual OS mmap calls)
+		size_t map_calls;
+	} size_use[128];
 } rpmalloc_thread_statistics_t;
 
 typedef struct rpmalloc_config_t {
@@ -67,85 +129,133 @@ typedef struct rpmalloc_config_t {
 	//  actual start of the memory region due to this alignment. The alignment offset
 	//  will be passed to the memory unmap function. The alignment offset MUST NOT be
 	//  larger than 65535 (storable in an uint16_t), if it is you must use natural
-	//  alignment to shift it into 16 bits.
+	//  alignment to shift it into 16 bits. If you set a memory_map function, you
+	//  must also set a memory_unmap function or else the default implementation will
+	//  be used for both.
 	void* (*memory_map)(size_t size, size_t* offset);
 	//! Unmap the memory pages starting at address and spanning the given number of bytes.
-	//  If release is set to 1, the unmap is for an entire span range as returned by
-	//  a previous call to memory_map and that the entire range should be released.
-	//  If release is set to 0, the unmap is a partial decommit of a subset of the mapped
-	//  memory range.
-	void (*memory_unmap)(void* address, size_t size, size_t offset, int release);
-	//! Size of memory pages. The page size MUST be a power of two in [512,16384] range
-	//  (2^9 to 2^14) unless 0 - set to 0 to use system page size. All memory mapping
+	//  If release is set to non-zero, the unmap is for an entire span range as returned by
+	//  a previous call to memory_map and that the entire range should be released. The
+	//  release argument holds the size of the entire span range. If release is set to 0,
+	//  the unmap is a partial decommit of a subset of the mapped memory range.
+	//  If you set a memory_unmap function, you must also set a memory_map function or
+	//  else the default implementation will be used for both.
+	void (*memory_unmap)(void* address, size_t size, size_t offset, size_t release);
+	//! Size of memory pages. The page size MUST be a power of two. All memory mapping
 	//  requests to memory_map will be made with size set to a multiple of the page size.
+	//  Used if RPMALLOC_CONFIGURABLE is defined to 1, otherwise system page size is used.
 	size_t page_size;
-	//! Size of a span of memory pages. MUST be a multiple of page size, and in [4096,262144]
-	//  range (unless 0 - set to 0 to use the default span size).
+	//! Size of a span of memory blocks. MUST be a power of two, and in [4096,262144]
+	//  range (unless 0 - set to 0 to use the default span size). Used if RPMALLOC_CONFIGURABLE
+	//  is defined to 1.
 	size_t span_size;
 	//! Number of spans to map at each request to map new virtual memory blocks. This can
 	//  be used to minimize the system call overhead at the cost of virtual memory address
 	//  space. The extra mapped pages will not be written until actually used, so physical
-	//  committed memory should not be affected in the default implementation.
+	//  committed memory should not be affected in the default implementation. Will be
+	//  aligned to a multiple of spans that match memory page size in case of huge pages.
 	size_t span_map_count;
-	//! Debug callback if memory guards are enabled. Called if a memory overwrite is detected
-	void (*memory_overwrite)(void* address);
+	//! Enable use of large/huge pages. If this flag is set to non-zero and page size is
+	//  zero, the allocator will try to enable huge pages and auto detect the configuration.
+	//  If this is set to non-zero and page_size is also non-zero, the allocator will
+	//  assume huge pages have been configured and enabled prior to initializing the
+	//  allocator.
+	//  For Windows, see https://docs.microsoft.com/en-us/windows/desktop/memory/large-page-support
+	//  For Linux, see https://www.kernel.org/doc/Documentation/vm/hugetlbpage.txt
+	int enable_huge_pages;
 } rpmalloc_config_t;
 
-extern int
+//! Initialize allocator with default configuration
+RPMALLOC_EXPORT int
 rpmalloc_initialize(void);
 
-extern int
+//! Initialize allocator with given configuration
+RPMALLOC_EXPORT int
 rpmalloc_initialize_config(const rpmalloc_config_t* config);
 
-extern const rpmalloc_config_t*
+//! Get allocator configuration
+RPMALLOC_EXPORT const rpmalloc_config_t*
 rpmalloc_config(void);
 
-extern void
+//! Finalize allocator
+RPMALLOC_EXPORT void
 rpmalloc_finalize(void);
 
-extern void
+//! Initialize allocator for calling thread
+RPMALLOC_EXPORT void
 rpmalloc_thread_initialize(void);
 
-extern void
+//! Finalize allocator for calling thread
+RPMALLOC_EXPORT void
 rpmalloc_thread_finalize(void);
 
-extern void
+//! Perform deferred deallocations pending for the calling thread heap
+RPMALLOC_EXPORT void
 rpmalloc_thread_collect(void);
 
-extern int
+//! Query if allocator is initialized for calling thread
+RPMALLOC_EXPORT int
 rpmalloc_is_thread_initialized(void);
 
-extern void
+//! Get per-thread statistics
+RPMALLOC_EXPORT void
 rpmalloc_thread_statistics(rpmalloc_thread_statistics_t* stats);
 
-extern void
+//! Get global statistics
+RPMALLOC_EXPORT void
 rpmalloc_global_statistics(rpmalloc_global_statistics_t* stats);
 
-extern RPMALLOC_RESTRICT void*
-rpmalloc(size_t size) RPMALLOC_ATTRIBUTE;
+//! Dump all statistics in human readable format to file (should be a FILE*)
+RPMALLOC_EXPORT void
+rpmalloc_dump_statistics(void* file);
 
-extern void
+//! Allocate a memory block of at least the given size
+TRACY_API RPMALLOC_ALLOCATOR void*
+rpmalloc(size_t size) RPMALLOC_ATTRIB_MALLOC RPMALLOC_ATTRIB_ALLOC_SIZE(1);
+
+//! Free the given memory block
+TRACY_API void
 rpfree(void* ptr);
 
-extern RPMALLOC_RESTRICT void*
-rpcalloc(size_t num, size_t size) RPMALLOC_ATTRIBUTE;
+//! Allocate a memory block of at least the given size and zero initialize it
+RPMALLOC_EXPORT RPMALLOC_ALLOCATOR void*
+rpcalloc(size_t num, size_t size) RPMALLOC_ATTRIB_MALLOC RPMALLOC_ATTRIB_ALLOC_SIZE2(1, 2);
 
-extern void*
-rprealloc(void* ptr, size_t size);
+//! Reallocate the given block to at least the given size
+RPMALLOC_EXPORT RPMALLOC_ALLOCATOR void*
+rprealloc(void* ptr, size_t size) RPMALLOC_ATTRIB_MALLOC RPMALLOC_ATTRIB_ALLOC_SIZE(2);
 
-extern void*
-rpaligned_realloc(void* ptr, size_t alignment, size_t size, size_t oldsize, unsigned int flags);
+//! Reallocate the given block to at least the given size and alignment,
+//  with optional control flags (see RPMALLOC_NO_PRESERVE).
+//  Alignment must be a power of two and a multiple of sizeof(void*),
+//  and should ideally be less than memory page size. A caveat of rpmalloc
+//  internals is that this must also be strictly less than the span size (default 64KiB)
+RPMALLOC_EXPORT RPMALLOC_ALLOCATOR void*
+rpaligned_realloc(void* ptr, size_t alignment, size_t size, size_t oldsize, unsigned int flags) RPMALLOC_ATTRIB_MALLOC RPMALLOC_ATTRIB_ALLOC_SIZE(3);
 
-extern RPMALLOC_RESTRICT void*
-rpaligned_alloc(size_t alignment, size_t size) RPMALLOC_ATTRIBUTE;
+//! Allocate a memory block of at least the given size and alignment.
+//  Alignment must be a power of two and a multiple of sizeof(void*),
+//  and should ideally be less than memory page size. A caveat of rpmalloc
+//  internals is that this must also be strictly less than the span size (default 64KiB)
+RPMALLOC_EXPORT RPMALLOC_ALLOCATOR void*
+rpaligned_alloc(size_t alignment, size_t size) RPMALLOC_ATTRIB_MALLOC RPMALLOC_ATTRIB_ALLOC_SIZE(2);
 
-extern RPMALLOC_RESTRICT void*
-rpmemalign(size_t alignment, size_t size) RPMALLOC_ATTRIBUTE;
+//! Allocate a memory block of at least the given size and alignment.
+//  Alignment must be a power of two and a multiple of sizeof(void*),
+//  and should ideally be less than memory page size. A caveat of rpmalloc
+//  internals is that this must also be strictly less than the span size (default 64KiB)
+RPMALLOC_EXPORT RPMALLOC_ALLOCATOR void*
+rpmemalign(size_t alignment, size_t size) RPMALLOC_ATTRIB_MALLOC RPMALLOC_ATTRIB_ALLOC_SIZE(2);
 
-extern int
+//! Allocate a memory block of at least the given size and alignment.
+//  Alignment must be a power of two and a multiple of sizeof(void*),
+//  and should ideally be less than memory page size. A caveat of rpmalloc
+//  internals is that this must also be strictly less than the span size (default 64KiB)
+RPMALLOC_EXPORT int
 rpposix_memalign(void **memptr, size_t alignment, size_t size);
 
-extern size_t
+//! Query the usable size of the given memory block (from given pointer to the end of block)
+RPMALLOC_EXPORT size_t
 rpmalloc_usable_size(void* ptr);
 
 }

common/TracyAlloc.hpp

@@ -1,7 +1,7 @@
 #ifndef __TRACYALLOC_HPP__
 #define __TRACYALLOC_HPP__
 
-#include <cstdlib>
+#include <stdlib.h>
 
 #ifdef TRACY_ENABLE
 #  include "../client/tracy_rpmalloc.hpp"

common/TracyApi.h

@@ -0,0 +1,16 @@
+#ifndef __TRACYAPI_H__
+#define __TRACYAPI_H__
+
+#if defined _WIN32 || defined __CYGWIN__
+#  if defined TRACY_EXPORTS
+#    define TRACY_API __declspec(dllexport)
+#  elif defined TRACY_IMPORTS
+#    define TRACY_API __declspec(dllimport)
+#  else
+#    define TRACY_API
+#  endif
+#else
+#  define TRACY_API __attribute__((visibility("default")))
+#endif
+
+#endif    // __TRACYAPI_H__

common/TracyMutex.hpp

@@ -10,15 +10,6 @@ namespace tracy
 using TracyMutex = std::shared_mutex;
 }
 
-#elif defined __CYGWIN__
-
-#include "tracy_benaphore.h"
-
-namespace tracy
-{
-using TracyMutex = NonRecursiveBenaphore;
-}
-
 #else
 
 #include <mutex>

common/TracyProtocol.hpp

@@ -4,31 +4,21 @@
 #include <limits>
 #include <stdint.h>
 
-#include "../common/tracy_lz4.hpp"
-
 namespace tracy
 {
 
-enum : uint32_t { ProtocolVersion = 1 };
+constexpr unsigned Lz4CompressBound( unsigned isize ) { return isize + ( isize / 255 ) + 16; }
+
+enum : uint32_t { ProtocolVersion = 35 };
+enum : uint32_t { BroadcastVersion = 1 };
 
 using lz4sz_t = uint32_t;
 
 enum { TargetFrameSize = 256 * 1024 };
-enum { LZ4Size = LZ4_COMPRESSBOUND( TargetFrameSize ) };
+enum { LZ4Size = Lz4CompressBound( TargetFrameSize ) };
 static_assert( LZ4Size <= std::numeric_limits<lz4sz_t>::max(), "LZ4Size greater than lz4sz_t" );
 static_assert( TargetFrameSize * 2 >= 64 * 1024, "Not enough space for LZ4 stream buffer" );
 
-enum ServerQuery : uint8_t
-{
-    ServerQueryTerminate,
-    ServerQueryString,
-    ServerQueryThreadString,
-    ServerQuerySourceLocation,
-    ServerQueryPlotName,
-    ServerQueryCallstackFrame,
-    ServerQueryFrameName,
-};
-
 enum { HandshakeShibbolethSize = 8 };
 static const char HandshakeShibboleth[HandshakeShibbolethSize] = { 'T', 'r', 'a', 'c', 'y', 'P', 'r', 'f' };
 
@@ -37,7 +27,8 @@ enum HandshakeStatus : uint8_t
     HandshakePending,
     HandshakeWelcome,
     HandshakeProtocolMismatch,
-    HandshakeNotAvailable
+    HandshakeNotAvailable,
+    HandshakeDropped
 };
 
 enum { WelcomeMessageProgramNameSize = 64 };
@@ -45,6 +36,44 @@ enum { WelcomeMessageHostInfoSize = 1024 };
 
 #pragma pack( 1 )
 
+// Must increase left query space after handling!
+enum ServerQuery : uint8_t
+{
+    ServerQueryTerminate,
+    ServerQueryString,
+    ServerQueryThreadString,
+    ServerQuerySourceLocation,
+    ServerQueryPlotName,
+    ServerQueryCallstackFrame,
+    ServerQueryFrameName,
+    ServerQueryDisconnect,
+    ServerQueryExternalName,
+    ServerQueryParameter,
+    ServerQuerySymbol,
+    ServerQuerySymbolCode,
+    ServerQueryCodeLocation
+};
+
+struct ServerQueryPacket
+{
+    ServerQuery type;
+    uint64_t ptr;
+    uint32_t extra;
+};
+
+enum { ServerQueryPacketSize = sizeof( ServerQueryPacket ) };
+
+
+enum CpuArchitecture : uint8_t
+{
+    CpuArchUnknown,
+    CpuArchX86,
+    CpuArchX64,
+    CpuArchArm32,
+    CpuArchArm64
+};
+
+
 struct WelcomeMessage
 {
     double timerMul;
@@ -53,7 +82,13 @@ struct WelcomeMessage
     uint64_t delay;
     uint64_t resolution;
     uint64_t epoch;
+    uint64_t pid;
+    int64_t samplingPeriod;
     uint8_t onDemand;
+    uint8_t isApple;
+    uint8_t cpuArch;
+    char cpuManufacturer[12];
+    uint32_t cpuId;
     char programName[WelcomeMessageProgramNameSize];
     char hostInfo[WelcomeMessageHostInfoSize];
 };
@@ -64,10 +99,23 @@ enum { WelcomeMessageSize = sizeof( WelcomeMessage ) };
 struct OnDemandPayloadMessage
 {
     uint64_t frames;
+    uint64_t currentTime;
 };
 
 enum { OnDemandPayloadMessageSize = sizeof( OnDemandPayloadMessage ) };
 
+
+struct BroadcastMessage
+{
+    uint32_t broadcastVersion;
+    uint32_t protocolVersion;
+    uint32_t listenPort;
+    uint32_t activeTime;        // in seconds
+    char programName[WelcomeMessageProgramNameSize];
+};
+
+enum { BroadcastMessageSize = sizeof( BroadcastMessage ) };
+
 #pragma pack()
 
 }

common/TracyQueue.hpp

@@ -11,66 +11,122 @@ enum class QueueType : uint8_t
     ZoneText,
     ZoneName,
     Message,
+    MessageColor,
+    MessageCallstack,
+    MessageColorCallstack,
+    MessageAppInfo,
     ZoneBeginAllocSrcLoc,
+    ZoneBeginAllocSrcLocLean,
+    ZoneBeginAllocSrcLocCallstack,
+    ZoneBeginAllocSrcLocCallstackLean,
     CallstackMemory,
+    CallstackMemoryLean,
     Callstack,
-    Terminate,
-    KeepAlive,
-    Crash,
-    CrashReport,
+    CallstackLean,
+    CallstackAlloc,
+    CallstackAllocLean,
+    CallstackSample,
+    CallstackSampleLean,
+    FrameImage,
+    FrameImageLean,
     ZoneBegin,
     ZoneBeginCallstack,
     ZoneEnd,
-    FrameMarkMsg,
-    FrameMarkMsgStart,
-    FrameMarkMsgEnd,
-    SourceLocation,
-    LockAnnounce,
-    LockTerminate,
     LockWait,
     LockObtain,
     LockRelease,
     LockSharedWait,
     LockSharedObtain,
     LockSharedRelease,
-    LockMark,
-    PlotData,
-    MessageLiteral,
-    GpuNewContext,
-    GpuZoneBegin,
-    GpuZoneBeginCallstack,
-    GpuZoneEnd,
-    GpuTime,
+    LockName,
     MemAlloc,
     MemFree,
     MemAllocCallstack,
     MemFreeCallstack,
+    GpuZoneBegin,
+    GpuZoneBeginCallstack,
+    GpuZoneEnd,
+    GpuZoneBeginSerial,
+    GpuZoneBeginCallstackSerial,
+    GpuZoneEndSerial,
+    PlotData,
+    ContextSwitch,
+    ThreadWakeup,
+    GpuTime,
+    Terminate,
+    KeepAlive,
+    ThreadContext,
+    Crash,
+    CrashReport,
+    ZoneValidation,
+    ZoneValue,
+    FrameMarkMsg,
+    FrameMarkMsgStart,
+    FrameMarkMsgEnd,
+    SourceLocation,
+    LockAnnounce,
+    LockTerminate,
+    LockMark,
+    MessageLiteral,
+    MessageLiteralColor,
+    MessageLiteralCallstack,
+    MessageLiteralColorCallstack,
+    GpuNewContext,
+    CallstackFrameSize,
     CallstackFrame,
+    SymbolInformation,
+    CodeInformation,
+    SysTimeReport,
+    TidToPid,
+    PlotConfig,
+    ParamSetup,
+    ParamPingback,
+    CpuTopology,
     StringData,
     ThreadName,
     CustomStringData,
     PlotName,
     SourceLocationPayload,
     CallstackPayload,
+    CallstackAllocPayload,
     FrameName,
+    FrameImageData,
+    ExternalName,
+    ExternalThreadName,
+    SymbolCode,
     NUM_TYPES
 };
 
 #pragma pack( 1 )
 
-struct QueueZoneBegin
+struct QueueThreadContext
+{
+    uint64_t thread;
+};
+
+struct QueueZoneBeginLean
 {
     int64_t time;
-    uint64_t thread;
+};
+
+struct QueueZoneBegin : public QueueZoneBeginLean
+{
     uint64_t srcloc;    // ptr
-    uint32_t cpu;
 };
 
 struct QueueZoneEnd
 {
     int64_t time;
-    uint64_t thread;
-    uint32_t cpu;
+};
+
+struct QueueZoneValidation
+{
+    uint32_t id;
+};
+
+struct QueueZoneValue
+{
+    uint64_t value;
 };
 
 struct QueueStringTransfer
@@ -84,6 +140,19 @@ struct QueueFrameMark
     uint64_t name;      // ptr
 };
 
+struct QueueFrameImageLean
+{
+    uint64_t frame;
+    uint16_t w;
+    uint16_t h;
+    uint8_t flip;
+};
+
+struct QueueFrameImage : public QueueFrameImageLean
+{
+    uint64_t image;     // ptr
+};
+
 struct QueueSourceLocation
 {
     uint64_t name;
@@ -97,7 +166,6 @@ struct QueueSourceLocation
 
 struct QueueZoneText
 {
-    uint64_t thread;
     uint64_t text;      // ptr
 };
 
@@ -124,33 +192,39 @@ struct QueueLockTerminate
 
 struct QueueLockWait
 {
+    uint64_t thread;
     uint32_t id;
     int64_t time;
-    uint64_t thread;
     LockType type;
 };
 
 struct QueueLockObtain
 {
+    uint64_t thread;
     uint32_t id;
     int64_t time;
-    uint64_t thread;
 };
 
 struct QueueLockRelease
 {
+    uint64_t thread;
     uint32_t id;
     int64_t time;
-    uint64_t thread;
 };
 
 struct QueueLockMark
 {
-    uint32_t id;
     uint64_t thread;
+    uint32_t id;
     uint64_t srcloc;    // ptr
 };
 
+struct QueueLockName
+{
+    uint32_t id;
+    uint64_t name;      // ptr
+};
+
 enum class PlotDataType : uint8_t
 {
     Float,
@@ -174,10 +248,26 @@ struct QueuePlotData
 struct QueueMessage
 {
     int64_t time;
-    uint64_t thread;
     uint64_t text;      // ptr
 };
 
+struct QueueMessageColor : public QueueMessage
+{
+    uint8_t r;
+    uint8_t g;
+    uint8_t b;
+};
+
+// Don't change order, only add new entries at the end, this is also used on trace dumps!
+enum class GpuContextType : uint8_t
+{
+    Invalid,
+    OpenGl,
+    Vulkan,
+    OpenCL,
+    Direct3D12
+};
+
 struct QueueGpuNewContext
 {
     int64_t cpuTime;
@@ -186,6 +276,7 @@ struct QueueGpuNewContext
     float period;
     uint8_t context;
     uint8_t accuracyBits;
+    GpuContextType type;
 };
 
 struct QueueGpuZoneBegin
@@ -200,6 +291,7 @@ struct QueueGpuZoneBegin
 struct QueueGpuZoneEnd
 {
     int64_t cpuTime;
+    uint64_t thread;
     uint16_t queryId;
     uint8_t context;
 };
@@ -234,24 +326,117 @@ struct QueueCallstackMemory
 struct QueueCallstack
 {
     uint64_t ptr;
+};
+
+struct QueueCallstackAlloc
+{
+    uint64_t ptr;
+    uint64_t nativePtr;
+};
+
+struct QueueCallstackSampleLean
+{
+    int64_t time;
     uint64_t thread;
 };
 
+struct QueueCallstackSample : public QueueCallstackSampleLean
+{
+    uint64_t ptr;
+};
+
+struct QueueCallstackFrameSize
+{
+    uint64_t ptr;
+    uint8_t size;
+    uint64_t imageName;
+};
+
 struct QueueCallstackFrame
 {
-    uint64_t ptr;
     uint64_t name;
     uint64_t file;
     uint32_t line;
+    uint64_t symAddr;
+    char symLen[3];
+};
+
+struct QueueSymbolInformation
+{
+    uint64_t file;
+    uint32_t line;
+    uint64_t symAddr;
+};
+
+struct QueueCodeInformation
+{
+    uint64_t ptr;
+    uint64_t file;
+    uint32_t line;
 };
 
 struct QueueCrashReport
 {
     int64_t time;
-    uint64_t thread;
     uint64_t text;      // ptr
 };
 
+struct QueueSysTime
+{
+    int64_t time;
+    float sysTime;
+};
+
+struct QueueContextSwitch
+{
+    int64_t time;
+    uint64_t oldThread;
+    uint64_t newThread;
+    uint8_t cpu;
+    uint8_t reason;
+    uint8_t state;
+};
+
+struct QueueThreadWakeup
+{
+    int64_t time;
+    uint64_t thread;
+};
+
+struct QueueTidToPid
+{
+    uint64_t tid;
+    uint64_t pid;
+};
+
+enum class PlotFormatType : uint8_t
+{
+    Number,
+    Memory,
+    Percentage
+};
+
+struct QueuePlotConfig
+{
+    uint64_t name;      // ptr
+    uint8_t type;
+};
+
+struct QueueParamSetup
+{
+    uint32_t idx;
+    uint64_t name;      // ptr
+    uint8_t isBool;
+    int32_t val;
+};
+
+struct QueueCpuTopology
+{
+    uint32_t package;
+    uint32_t core;
+    uint32_t thread;
+};
+
 struct QueueHeader
 {
     union
@@ -266,10 +451,16 @@ struct QueueItem
     QueueHeader hdr;
     union
     {
+        QueueThreadContext threadCtx;
         QueueZoneBegin zoneBegin;
+        QueueZoneBeginLean zoneBeginLean;
         QueueZoneEnd zoneEnd;
+        QueueZoneValidation zoneValidation;
+        QueueZoneValue zoneValue;
         QueueStringTransfer stringTransfer;
         QueueFrameMark frameMark;
+        QueueFrameImage frameImage;
+        QueueFrameImage frameImageLean;
         QueueSourceLocation srcloc;
         QueueZoneText zoneText;
         QueueLockAnnounce lockAnnounce;
@@ -278,8 +469,10 @@ struct QueueItem
         QueueLockObtain lockObtain;
         QueueLockRelease lockRelease;
         QueueLockMark lockMark;
+        QueueLockName lockName;
         QueuePlotData plotData;
         QueueMessage message;
+        QueueMessageColor messageColor;
         QueueGpuNewContext gpuNewContext;
         QueueGpuZoneBegin gpuZoneBegin;
         QueueGpuZoneEnd gpuZoneEnd;
@@ -288,55 +481,104 @@ struct QueueItem
         QueueMemFree memFree;
         QueueCallstackMemory callstackMemory;
         QueueCallstack callstack;
+        QueueCallstackAlloc callstackAlloc;
+        QueueCallstackSample callstackSample;
+        QueueCallstackSampleLean callstackSampleLean;
+        QueueCallstackFrameSize callstackFrameSize;
         QueueCallstackFrame callstackFrame;
+        QueueSymbolInformation symbolInformation;
+        QueueCodeInformation codeInformation;
         QueueCrashReport crashReport;
+        QueueSysTime sysTime;
+        QueueContextSwitch contextSwitch;
+        QueueThreadWakeup threadWakeup;
+        QueueTidToPid tidToPid;
+        QueuePlotConfig plotConfig;
+        QueueParamSetup paramSetup;
+        QueueCpuTopology cpuTopology;
     };
 };
-
 #pragma pack()
 
+
 enum { QueueItemSize = sizeof( QueueItem ) };
 
-static const size_t QueueDataSize[] = {
+static constexpr size_t QueueDataSize[] = {
     sizeof( QueueHeader ) + sizeof( QueueZoneText ),
     sizeof( QueueHeader ) + sizeof( QueueZoneText ),        // zone name
     sizeof( QueueHeader ) + sizeof( QueueMessage ),
-    sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // allocated source location
-    sizeof( QueueHeader ) + sizeof( QueueCallstackMemory ),
-    sizeof( QueueHeader ) + sizeof( QueueCallstack ),
-    // above items must be first
-    sizeof( QueueHeader ),                                  // terminate
-    sizeof( QueueHeader ),                                  // keep alive
-    sizeof( QueueHeader ),                                  // crash
-    sizeof( QueueHeader ) + sizeof( QueueCrashReport ),
+    sizeof( QueueHeader ) + sizeof( QueueMessageColor ),
+    sizeof( QueueHeader ) + sizeof( QueueMessage ),         // callstack
+    sizeof( QueueHeader ) + sizeof( QueueMessageColor ),    // callstack
+    sizeof( QueueHeader ) + sizeof( QueueMessage ),         // app info
+    sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // allocated source location, not for network transfer
+    sizeof( QueueHeader ) + sizeof( QueueZoneBeginLean ),   // lean allocated source location
+    sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // allocated source location, callstack, not for network transfer
+    sizeof( QueueHeader ) + sizeof( QueueZoneBeginLean ),   // lean allocated source location, callstack
+    sizeof( QueueHeader ) + sizeof( QueueCallstackMemory ), // not for network transfer
+    sizeof( QueueHeader ),                                  // lean callstack memory
+    sizeof( QueueHeader ) + sizeof( QueueCallstack ),       // not for network transfer
+    sizeof( QueueHeader ),                                  // lean callstack
+    sizeof( QueueHeader ) + sizeof( QueueCallstackAlloc ),  // not for network transfer
+    sizeof( QueueHeader ),                                  // lean callstack alloc
+    sizeof( QueueHeader ) + sizeof( QueueCallstackSample ), // not for network transfer
+    sizeof( QueueHeader ) + sizeof( QueueCallstackSampleLean ),
+    sizeof( QueueHeader ) + sizeof( QueueFrameImage ),      // not for network transfer
+    sizeof( QueueHeader ) + sizeof( QueueFrameImageLean ),
     sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),
     sizeof( QueueHeader ) + sizeof( QueueZoneBegin ),       // callstack
     sizeof( QueueHeader ) + sizeof( QueueZoneEnd ),
+    sizeof( QueueHeader ) + sizeof( QueueLockWait ),
+    sizeof( QueueHeader ) + sizeof( QueueLockObtain ),
+    sizeof( QueueHeader ) + sizeof( QueueLockRelease ),
+    sizeof( QueueHeader ) + sizeof( QueueLockWait ),        // shared
+    sizeof( QueueHeader ) + sizeof( QueueLockObtain ),      // shared
+    sizeof( QueueHeader ) + sizeof( QueueLockRelease ),     // shared
+    sizeof( QueueHeader ) + sizeof( QueueLockName ),
+    sizeof( QueueHeader ) + sizeof( QueueMemAlloc ),
+    sizeof( QueueHeader ) + sizeof( QueueMemFree ),
+    sizeof( QueueHeader ) + sizeof( QueueMemAlloc ),        // callstack
+    sizeof( QueueHeader ) + sizeof( QueueMemFree ),         // callstack
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneBegin ),
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneBegin ),    // callstack
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneEnd ),
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneBegin ),    // serial
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneBegin ),    // serial, callstack
+    sizeof( QueueHeader ) + sizeof( QueueGpuZoneEnd ),      // serial
+    sizeof( QueueHeader ) + sizeof( QueuePlotData ),
+    sizeof( QueueHeader ) + sizeof( QueueContextSwitch ),
+    sizeof( QueueHeader ) + sizeof( QueueThreadWakeup ),
+    sizeof( QueueHeader ) + sizeof( QueueGpuTime ),
+    // above items must be first
+    sizeof( QueueHeader ),                                  // terminate
+    sizeof( QueueHeader ),                                  // keep alive
+    sizeof( QueueHeader ) + sizeof( QueueThreadContext ),
+    sizeof( QueueHeader ),                                  // crash
+    sizeof( QueueHeader ) + sizeof( QueueCrashReport ),
+    sizeof( QueueHeader ) + sizeof( QueueZoneValidation ),
+    sizeof( QueueHeader ) + sizeof( QueueZoneValue ),
     sizeof( QueueHeader ) + sizeof( QueueFrameMark ),       // continuous frames
     sizeof( QueueHeader ) + sizeof( QueueFrameMark ),       // start
     sizeof( QueueHeader ) + sizeof( QueueFrameMark ),       // end
     sizeof( QueueHeader ) + sizeof( QueueSourceLocation ),
     sizeof( QueueHeader ) + sizeof( QueueLockAnnounce ),
     sizeof( QueueHeader ) + sizeof( QueueLockTerminate ),
-    sizeof( QueueHeader ) + sizeof( QueueLockWait ),
-    sizeof( QueueHeader ) + sizeof( QueueLockObtain ),
-    sizeof( QueueHeader ) + sizeof( QueueLockRelease ),
-    sizeof( QueueHeader ) + sizeof( QueueLockWait ),
-    sizeof( QueueHeader ) + sizeof( QueueLockObtain ),
-    sizeof( QueueHeader ) + sizeof( QueueLockRelease ),
     sizeof( QueueHeader ) + sizeof( QueueLockMark ),
-    sizeof( QueueHeader ) + sizeof( QueuePlotData ),
     sizeof( QueueHeader ) + sizeof( QueueMessage ),         // literal
+    sizeof( QueueHeader ) + sizeof( QueueMessageColor ),    // literal
+    sizeof( QueueHeader ) + sizeof( QueueMessage ),         // literal, callstack
+    sizeof( QueueHeader ) + sizeof( QueueMessageColor ),    // literal, callstack
     sizeof( QueueHeader ) + sizeof( QueueGpuNewContext ),
-    sizeof( QueueHeader ) + sizeof( QueueGpuZoneBegin ),
-    sizeof( QueueHeader ) + sizeof( QueueGpuZoneBegin ),    // callstack
-    sizeof( QueueHeader ) + sizeof( QueueGpuZoneEnd ),
-    sizeof( QueueHeader ) + sizeof( QueueGpuTime ),
-    sizeof( QueueHeader ) + sizeof( QueueMemAlloc ),
-    sizeof( QueueHeader ) + sizeof( QueueMemFree ),
-    sizeof( QueueHeader ) + sizeof( QueueMemAlloc ),        // callstack
-    sizeof( QueueHeader ) + sizeof( QueueMemFree ),         // callstack
+    sizeof( QueueHeader ) + sizeof( QueueCallstackFrameSize ),
     sizeof( QueueHeader ) + sizeof( QueueCallstackFrame ),
+    sizeof( QueueHeader ) + sizeof( QueueSymbolInformation ),
+    sizeof( QueueHeader ) + sizeof( QueueCodeInformation ),
+    sizeof( QueueHeader ) + sizeof( QueueSysTime ),
+    sizeof( QueueHeader ) + sizeof( QueueTidToPid ),
+    sizeof( QueueHeader ) + sizeof( QueuePlotConfig ),
+    sizeof( QueueHeader ) + sizeof( QueueParamSetup ),
+    sizeof( QueueHeader ),                                  // param pingback
+    sizeof( QueueHeader ) + sizeof( QueueCpuTopology ),
     // keep all QueueStringTransfer below
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // string data
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // thread name
@@ -344,7 +586,12 @@ static const size_t QueueDataSize[] = {
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // plot name
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // allocated source location payload
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // callstack payload
+    sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // callstack alloc payload
     sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // frame name
+    sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // frame image data
+    sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // external name
+    sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // external thread name
+    sizeof( QueueHeader ) + sizeof( QueueStringTransfer ),  // symbol code
 };
 
 static_assert( QueueItemSize == 32, "Queue item size not 32 bytes" );
@@ -352,6 +599,6 @@ static_assert( sizeof( QueueDataSize ) / sizeof( size_t ) == (uint8_t)QueueType:
 static_assert( sizeof( void* ) <= sizeof( uint64_t ), "Pointer size > 8 bytes" );
 static_assert( sizeof( void* ) == sizeof( uintptr_t ), "Pointer size != uintptr_t" );
 
-};
+}
 
 #endif

common/TracySocket.cpp

@@ -1,4 +1,3 @@
-#include <algorithm>
 #include <assert.h>
 #include <new>
 #include <stdio.h>
@@ -9,15 +8,27 @@
 #include "TracyAlloc.hpp"
 #include "TracySocket.hpp"
 
-#ifdef _MSC_VER
+#ifdef _WIN32
+#  ifndef NOMINMAX
+#    define NOMINMAX
+#  endif
 #  include <winsock2.h>
 #  include <ws2tcpip.h>
-#  pragma warning(disable:4244)
-#  pragma warning(disable:4267)
+#  ifdef _MSC_VER
+#    pragma warning(disable:4244)
+#    pragma warning(disable:4267)
+#  endif
+#  define poll WSAPoll
 #else
+#  include <arpa/inet.h>
 #  include <sys/socket.h>
+#  include <sys/param.h>
+#  include <errno.h>
+#  include <fcntl.h>
+#  include <netinet/in.h>
 #  include <netdb.h>
 #  include <unistd.h>
+#  include <poll.h>
 #endif
 
 #ifndef MSG_NOSIGNAL
@@ -27,13 +38,13 @@
 namespace tracy
 {
 
-#ifdef _MSC_VER
+#ifdef _WIN32
 typedef SOCKET socket_t;
 #else
 typedef int socket_t;
 #endif
 
-#ifdef _MSC_VER
+#ifdef _WIN32
 struct __wsinit
 {
     __wsinit()
@@ -53,37 +64,89 @@ void InitWinSock()
 }
 #endif
 
+
+enum { BufSize = 128 * 1024 };
+
 Socket::Socket()
-    : m_sock( -1 )
-    , m_buf( (char*)tracy_malloc( BufSize ) )
+    : m_buf( (char*)tracy_malloc( BufSize ) )
     , m_bufPtr( nullptr )
+    , m_sock( -1 )
     , m_bufLeft( 0 )
+    , m_ptr( nullptr )
 {
-#ifdef _MSC_VER
+#ifdef _WIN32
     InitWinSock();
 #endif
 }
 
 Socket::Socket( int sock )
-    : m_sock( sock )
-    , m_buf( (char*)tracy_malloc( BufSize ) )
+    : m_buf( (char*)tracy_malloc( BufSize ) )
     , m_bufPtr( nullptr )
+    , m_sock( sock )
     , m_bufLeft( 0 )
+    , m_ptr( nullptr )
 {
 }
 
 Socket::~Socket()
 {
     tracy_free( m_buf );
-    if( m_sock != -1 )
+    if( m_sock.load( std::memory_order_relaxed ) != -1 )
     {
         Close();
     }
+    if( m_ptr )
+    {
+        freeaddrinfo( m_res );
+#ifdef _WIN32
+        closesocket( m_connSock );
+#else
+        close( m_connSock );
+#endif
+    }
 }
 
-bool Socket::Connect( const char* addr, const char* port )
+bool Socket::Connect( const char* addr, int port )
 {
-    assert( m_sock == -1 );
+    assert( !IsValid() );
+
+    if( m_ptr )
+    {
+        const auto c = connect( m_connSock, m_ptr->ai_addr, m_ptr->ai_addrlen );
+        assert( c == -1 );
+#if defined _WIN32 || defined __CYGWIN__
+        const auto err = WSAGetLastError();
+        if( err == WSAEALREADY || err == WSAEINPROGRESS ) return false;
+        if( err != WSAEISCONN )
+        {
+            freeaddrinfo( m_res );
+            closesocket( m_connSock );
+            m_ptr = nullptr;
+            return false;
+        }
+#else
+        if( errno == EALREADY || errno == EINPROGRESS ) return false;
+        if( errno != EISCONN )
+        {
+            freeaddrinfo( m_res );
+            close( m_connSock );
+            m_ptr = nullptr;
+            return false;
+        }
+#endif
+
+#if defined _WIN32 || defined __CYGWIN__
+        u_long nonblocking = 0;
+        ioctlsocket( m_connSock, FIONBIO, &nonblocking );
+#else
+        int flags = fcntl( m_connSock, F_GETFL, 0 );
+        fcntl( m_connSock, F_SETFL, flags & ~O_NONBLOCK );
+#endif
+        m_sock.store( m_connSock, std::memory_order_relaxed );
+        freeaddrinfo( m_res );
+        m_ptr = nullptr;
+        return true;
+    }
 
     struct addrinfo hints;
     struct addrinfo *res, *ptr;
@@ -92,52 +155,87 @@ bool Socket::Connect( const char* addr, const char* port )
     hints.ai_family = AF_UNSPEC;
     hints.ai_socktype = SOCK_STREAM;
 
-    if( getaddrinfo( addr, port, &hints, &res ) != 0 ) return false;
+    char portbuf[32];
+    sprintf( portbuf, "%i", port );
+
+    if( getaddrinfo( addr, portbuf, &hints, &res ) != 0 ) return false;
     int sock = 0;
     for( ptr = res; ptr; ptr = ptr->ai_next )
     {
         if( ( sock = socket( ptr->ai_family, ptr->ai_socktype, ptr->ai_protocol ) ) == -1 ) continue;
 #if defined __APPLE__
         int val = 1;
-        setsockopt( m_sock, SOL_SOCKET, SO_NOSIGPIPE, &val, sizeof( val ) );
+        setsockopt( sock, SOL_SOCKET, SO_NOSIGPIPE, &val, sizeof( val ) );
 #endif
-        if( connect( sock, ptr->ai_addr, ptr->ai_addrlen ) == -1 )
-        {
-#ifdef _MSC_VER
-            closesocket( sock );
+#if defined _WIN32 || defined __CYGWIN__
+        u_long nonblocking = 1;
+        ioctlsocket( sock, FIONBIO, &nonblocking );
 #else
-            close( sock );
+        int flags = fcntl( sock, F_GETFL, 0 );
+        fcntl( sock, F_SETFL, flags | O_NONBLOCK );
 #endif
-            continue;
+        if( connect( sock, ptr->ai_addr, ptr->ai_addrlen ) == 0 )
+        {
+            break;
         }
-        break;
+        else
+        {
+#if defined _WIN32 || defined __CYGWIN__
+            const auto err = WSAGetLastError();
+            if( err != WSAEWOULDBLOCK )
+            {
+                closesocket( sock );
+                continue;
+            }
+#else
+            if( errno != EINPROGRESS )
+            {
+                close( sock );
+                continue;
+            }
+#endif
+        }
+        m_res = res;
+        m_ptr = ptr;
+        m_connSock = sock;
+        return false;
     }
     freeaddrinfo( res );
     if( !ptr ) return false;
 
-    m_sock = sock;
+#if defined _WIN32 || defined __CYGWIN__
+    u_long nonblocking = 0;
+    ioctlsocket( sock, FIONBIO, &nonblocking );
+#else
+    int flags = fcntl( sock, F_GETFL, 0 );
+    fcntl( sock, F_SETFL, flags & ~O_NONBLOCK );
+#endif
+
+    m_sock.store( sock, std::memory_order_relaxed );
     return true;
 }
 
 void Socket::Close()
 {
-    assert( m_sock != -1 );
-#ifdef _MSC_VER
-    closesocket( m_sock );
+    const auto sock = m_sock.load( std::memory_order_relaxed );
+    assert( sock != -1 );
+#ifdef _WIN32
+    closesocket( sock );
 #else
-    close( m_sock );
+    close( sock );
 #endif
-    m_sock = -1;
+    m_sock.store( -1, std::memory_order_relaxed );
 }
 
 int Socket::Send( const void* _buf, int len )
 {
+    const auto sock = m_sock.load( std::memory_order_relaxed );
     auto buf = (const char*)_buf;
-    assert( m_sock != -1 );
+    assert( sock != -1 );
     auto start = buf;
     while( len > 0 )
     {
-        auto ret = send( m_sock, buf, len, MSG_NOSIGNAL );
+        auto ret = send( sock, buf, len, MSG_NOSIGNAL );
         if( ret == -1 ) return -1;
         len -= ret;
         buf += ret;
@@ -145,7 +243,21 @@ int Socket::Send( const void* _buf, int len )
     return int( buf - start );
 }
 
-int Socket::RecvBuffered( void* buf, int len, const timeval* tv )
+int Socket::GetSendBufSize()
+{
+    const auto sock = m_sock.load( std::memory_order_relaxed );
+    int bufSize;
+#if defined _WIN32 || defined __CYGWIN__
+    int sz = sizeof( bufSize );
+    getsockopt( sock, SOL_SOCKET, SO_SNDBUF, (char*)&bufSize, &sz );
+#else
+    socklen_t sz = sizeof( bufSize );
+    getsockopt( sock, SOL_SOCKET, SO_SNDBUF, &bufSize, &sz );
+#endif
+    return bufSize;
+}
+
+int Socket::RecvBuffered( void* buf, int len, int timeout )
 {
     if( len <= m_bufLeft )
     {
@@ -163,35 +275,30 @@ int Socket::RecvBuffered( void* buf, int len, const timeval* tv )
         return ret;
     }
 
-    if( len >= BufSize ) return Recv( buf, len, tv );
+    if( len >= BufSize ) return Recv( buf, len, timeout );
 
-    m_bufLeft = Recv( m_buf, BufSize, tv );
+    m_bufLeft = Recv( m_buf, BufSize, timeout );
     if( m_bufLeft <= 0 ) return m_bufLeft;
 
-    const auto sz = std::min( len, m_bufLeft );
+    const auto sz = len < m_bufLeft ? len : m_bufLeft;
     memcpy( buf, m_buf, sz );
     m_bufPtr = m_buf + sz;
     m_bufLeft -= sz;
     return sz;
 }
 
-int Socket::Recv( void* _buf, int len, const timeval* tv )
+int Socket::Recv( void* _buf, int len, int timeout )
 {
+    const auto sock = m_sock.load( std::memory_order_relaxed );
     auto buf = (char*)_buf;
 
-    fd_set fds;
-    FD_ZERO( &fds );
-    FD_SET( static_cast<socket_t>(m_sock), &fds );
+    struct pollfd fd;
+    fd.fd = (socket_t)sock;
+    fd.events = POLLIN;
 
-#ifndef _WIN32
-    timeval _tv = *tv;
-    select( m_sock+1, &fds, nullptr, nullptr, &_tv );
-#else
-    select( m_sock+1, &fds, nullptr, nullptr, tv );
-#endif
-    if( FD_ISSET( m_sock, &fds ) )
+    if( poll( &fd, 1, timeout ) > 0 )
     {
-        return recv( m_sock, buf, len, 0 );
+        return recv( sock, buf, len, 0 );
     }
     else
     {
@@ -199,42 +306,45 @@ int Socket::Recv( void* _buf, int len, const timeval* tv )
     }
 }
 
-bool Socket::Read( void* _buf, int len, const timeval* tv, std::function<bool()> exitCb )
+bool Socket::Read( void* buf, int len, int timeout )
 {
-    auto buf = (char*)_buf;
-
+    auto cbuf = (char*)buf;
     while( len > 0 )
     {
-        if( exitCb() ) return false;
-        const auto sz = RecvBuffered( buf, len, tv );
-        switch( sz )
-        {
-        case 0:
-            return false;
-        case -1:
-#ifdef _WIN32
-        {
-            auto err = WSAGetLastError();
-            if( err == WSAECONNABORTED || err == WSAECONNRESET ) return false;
-        }
-#endif
-            break;
-        default:
-            len -= sz;
-            buf += sz;
-            break;
-        }
+        if( !ReadImpl( cbuf, len, timeout ) ) return false;
     }
-
     return true;
 }
 
-bool Socket::ReadRaw( void* _buf, int len, const timeval* tv )
+bool Socket::ReadImpl( char*& buf, int& len, int timeout )
+{
+    const auto sz = RecvBuffered( buf, len, timeout );
+    switch( sz )
+    {
+    case 0:
+        return false;
+    case -1:
+#ifdef _WIN32
+    {
+        auto err = WSAGetLastError();
+        if( err == WSAECONNABORTED || err == WSAECONNRESET ) return false;
+    }
+#endif
+    break;
+    default:
+        len -= sz;
+        buf += sz;
+        break;
+    }
+    return true;
+}
+
+bool Socket::ReadRaw( void* _buf, int len, int timeout )
 {
     auto buf = (char*)_buf;
     while( len > 0 )
     {
-        const auto sz = Recv( buf, len, tv );
+        const auto sz = Recv( buf, len, timeout );
         if( sz <= 0 ) return false;
         len -= sz;
         buf += sz;
@@ -244,32 +354,36 @@ bool Socket::ReadRaw( void* _buf, int len, const timeval* tv )
 
 bool Socket::HasData()
 {
+    const auto sock = m_sock.load( std::memory_order_relaxed );
     if( m_bufLeft > 0 ) return true;
 
-    struct timeval tv;
-    memset( &tv, 0, sizeof( tv ) );
+    struct pollfd fd;
+    fd.fd = (socket_t)sock;
+    fd.events = POLLIN;
 
-    fd_set fds;
-    FD_ZERO( &fds );
-    FD_SET( static_cast<socket_t>(m_sock), &fds );
+    return poll( &fd, 1, 0 ) > 0;
+}
 
-    return select( m_sock+1, &fds, nullptr, nullptr, &tv ) > 0;
+bool Socket::IsValid() const
+{
+    return m_sock.load( std::memory_order_relaxed ) >= 0;
 }
 
 
 ListenSocket::ListenSocket()
     : m_sock( -1 )
 {
-#ifdef _MSC_VER
+#ifdef _WIN32
     InitWinSock();
 #endif
 }
 
 ListenSocket::~ListenSocket()
 {
+    if( m_sock != -1 ) Close();
 }
 
-bool ListenSocket::Listen( const char* port, int backlog )
+bool ListenSocket::Listen( int port, int backlog )
 {
     assert( m_sock == -1 );
 
@@ -279,20 +393,40 @@ bool ListenSocket::Listen( const char* port, int backlog )
     memset( &hints, 0, sizeof( hints ) );
     hints.ai_family = AF_INET6;
     hints.ai_socktype = SOCK_STREAM;
+#ifndef TRACY_ONLY_LOCALHOST
     hints.ai_flags = AI_PASSIVE;
+#endif
 
-    if( getaddrinfo( nullptr, port, &hints, &res ) != 0 ) return false;
+    char portbuf[32];
+    sprintf( portbuf, "%i", port );
+
+    if( getaddrinfo( nullptr, portbuf, &hints, &res ) != 0 ) return false;
 
     m_sock = socket( res->ai_family, res->ai_socktype, res->ai_protocol );
-#if defined _MSC_VER || defined __CYGWIN__
+    if (m_sock == -1)
+    {
+        // IPV6 protocol may not be available/is disabled. Try to create a socket
+        // with the IPV4 protocol
+        hints.ai_family = AF_INET;
+        if( getaddrinfo( nullptr, portbuf, &hints, &res ) != 0 ) return false;
+        m_sock = socket( res->ai_family, res->ai_socktype, res->ai_protocol );
+        if( m_sock == -1 ) return false;
+    }
+#if defined _WIN32 || defined __CYGWIN__
     unsigned long val = 0;
     setsockopt( m_sock, IPPROTO_IPV6, IPV6_V6ONLY, (const char*)&val, sizeof( val ) );
+#elif defined BSD
+    int val = 0;
+    setsockopt( m_sock, IPPROTO_IPV6, IPV6_V6ONLY, (const char*)&val, sizeof( val ) );
+    val = 1;
+    setsockopt( m_sock, SOL_SOCKET, SO_REUSEADDR, &val, sizeof( val ) );
 #else
     int val = 1;
     setsockopt( m_sock, SOL_SOCKET, SO_REUSEADDR, &val, sizeof( val ) );
 #endif
-    if( bind( m_sock, res->ai_addr, res->ai_addrlen ) == -1 ) return false;
-    if( listen( m_sock, backlog ) == -1 ) return false;
+    if( bind( m_sock, res->ai_addr, res->ai_addrlen ) == -1 ) { freeaddrinfo( res ); Close(); return false; }
+    if( listen( m_sock, backlog ) == -1 ) { freeaddrinfo( res ); Close(); return false; }
+    freeaddrinfo( res );
     return true;
 }
 
@@ -301,32 +435,23 @@ Socket* ListenSocket::Accept()
     struct sockaddr_storage remote;
     socklen_t sz = sizeof( remote );
 
-    struct timeval tv;
-    tv.tv_sec = 0;
-    tv.tv_usec = 10000;
+    struct pollfd fd;
+    fd.fd = (socket_t)m_sock;
+    fd.events = POLLIN;
 
-    fd_set fds;
-    FD_ZERO( &fds );
-    FD_SET( static_cast<socket_t>(m_sock), &fds );
-
-    select( m_sock+1, &fds, nullptr, nullptr, &tv );
-    if( FD_ISSET( m_sock, &fds ) )
+    if( poll( &fd, 1, 10 ) > 0 )
     {
         int sock = accept( m_sock, (sockaddr*)&remote, &sz);
+        if( sock == -1 ) return nullptr;
+
 #if defined __APPLE__
         int val = 1;
         setsockopt( sock, SOL_SOCKET, SO_NOSIGPIPE, &val, sizeof( val ) );
 #endif
-        if( sock == -1 )
-        {
-            return nullptr;
-        }
-        else
-        {
-            auto ptr = (Socket*)tracy_malloc( sizeof( Socket ) );
-            new(ptr) Socket( sock );
-            return ptr;
-        }
+
+        auto ptr = (Socket*)tracy_malloc( sizeof( Socket ) );
+        new(ptr) Socket( sock );
+        return ptr;
     }
     else
     {
@@ -337,7 +462,7 @@ Socket* ListenSocket::Accept()
 void ListenSocket::Close()
 {
     assert( m_sock != -1 );
-#ifdef _MSC_VER
+#ifdef _WIN32
     closesocket( m_sock );
 #else
     close( m_sock );
@@ -345,4 +470,200 @@ void ListenSocket::Close()
     m_sock = -1;
 }
 
+UdpBroadcast::UdpBroadcast()
+    : m_sock( -1 )
+{
+#ifdef _WIN32
+    InitWinSock();
+#endif
+}
+
+UdpBroadcast::~UdpBroadcast()
+{
+    if( m_sock != -1 ) Close();
+}
+
+bool UdpBroadcast::Open( const char* addr, int port )
+{
+    assert( m_sock == -1 );
+
+    struct addrinfo hints;
+    struct addrinfo *res, *ptr;
+
+    memset( &hints, 0, sizeof( hints ) );
+    hints.ai_family = AF_INET;
+    hints.ai_socktype = SOCK_DGRAM;
+
+    char portbuf[32];
+    sprintf( portbuf, "%i", port );
+
+    if( getaddrinfo( addr, portbuf, &hints, &res ) != 0 ) return false;
+    int sock = 0;
+    for( ptr = res; ptr; ptr = ptr->ai_next )
+    {
+        if( ( sock = socket( ptr->ai_family, ptr->ai_socktype, ptr->ai_protocol ) ) == -1 ) continue;
+#if defined __APPLE__
+        int val = 1;
+        setsockopt( sock, SOL_SOCKET, SO_NOSIGPIPE, &val, sizeof( val ) );
+#endif
+#if defined _WIN32 || defined __CYGWIN__
+        unsigned long broadcast = 1;
+        if( setsockopt( sock, SOL_SOCKET, SO_BROADCAST, (const char*)&broadcast, sizeof( broadcast ) ) == -1 )
+#else
+        int broadcast = 1;
+        if( setsockopt( sock, SOL_SOCKET, SO_BROADCAST, &broadcast, sizeof( broadcast ) ) == -1 )
+#endif
+        {
+#ifdef _WIN32
+            closesocket( sock );
+#else
+            close( sock );
+#endif
+            continue;
+        }
+        break;
+    }
+    freeaddrinfo( res );
+    if( !ptr ) return false;
+
+    m_sock = sock;
+    return true;
+}
+
+void UdpBroadcast::Close()
+{
+    assert( m_sock != -1 );
+#ifdef _WIN32
+    closesocket( m_sock );
+#else
+    close( m_sock );
+#endif
+    m_sock = -1;
+}
+
+int UdpBroadcast::Send( int port, const void* data, int len )
+{
+    assert( m_sock != -1 );
+    struct sockaddr_in addr;
+    addr.sin_family = AF_INET;
+    addr.sin_port = htons( port );
+    addr.sin_addr.s_addr = INADDR_BROADCAST;
+    return sendto( m_sock, (const char*)data, len, MSG_NOSIGNAL, (sockaddr*)&addr, sizeof( addr ) );
+}
+
+IpAddress::IpAddress()
+    : m_number( 0 )
+{
+    *m_text = '\0';
+}
+
+IpAddress::~IpAddress()
+{
+}
+
+void IpAddress::Set( const struct sockaddr& addr )
+{
+#if __MINGW32__
+    auto ai = (struct sockaddr_in*)&addr;
+#else
+    auto ai = (const struct sockaddr_in*)&addr;
+#endif
+    inet_ntop( AF_INET, &ai->sin_addr, m_text, 17 );
+    m_number = ai->sin_addr.s_addr;
+}
+
+UdpListen::UdpListen()
+    : m_sock( -1 )
+{
+#ifdef _WIN32
+    InitWinSock();
+#endif
+}
+
+UdpListen::~UdpListen()
+{
+    if( m_sock != -1 ) Close();
+}
+
+bool UdpListen::Listen( int port )
+{
+    assert( m_sock == -1 );
+
+    int sock;
+    if( ( sock = socket( AF_INET, SOCK_DGRAM, 0 ) ) == -1 ) return false;
+
+#if defined __APPLE__
+    int val = 1;
+    setsockopt( sock, SOL_SOCKET, SO_NOSIGPIPE, &val, sizeof( val ) );
+#endif
+#if defined _WIN32 || defined __CYGWIN__
+    unsigned long reuse = 1;
+    setsockopt( m_sock, SOL_SOCKET, SO_REUSEADDR, (const char*)&reuse, sizeof( reuse ) );
+#else
+    int reuse = 1;
+    setsockopt( m_sock, SOL_SOCKET, SO_REUSEADDR, &reuse, sizeof( reuse ) );
+#endif
+#if defined _WIN32 || defined __CYGWIN__
+    unsigned long broadcast = 1;
+    if( setsockopt( sock, SOL_SOCKET, SO_BROADCAST, (const char*)&broadcast, sizeof( broadcast ) ) == -1 )
+#else
+    int broadcast = 1;
+    if( setsockopt( sock, SOL_SOCKET, SO_BROADCAST, &broadcast, sizeof( broadcast ) ) == -1 )
+#endif
+    {
+#ifdef _WIN32
+        closesocket( sock );
+#else
+        close( sock );
+#endif
+        return false;
+    }
+
+    struct sockaddr_in addr;
+    addr.sin_family = AF_INET;
+    addr.sin_port = htons( port );
+    addr.sin_addr.s_addr = INADDR_ANY;
+
+    if( bind( sock, (sockaddr*)&addr, sizeof( addr ) ) == -1 )
+    {
+#ifdef _WIN32
+        closesocket( sock );
+#else
+        close( sock );
+#endif
+        return false;
+    }
+
+    m_sock = sock;
+    return true;
+}
+
+void UdpListen::Close()
+{
+    assert( m_sock != -1 );
+#ifdef _WIN32
+    closesocket( m_sock );
+#else
+    close( m_sock );
+#endif
+    m_sock = -1;
+}
+
+const char* UdpListen::Read( size_t& len, IpAddress& addr )
+{
+    static char buf[2048];
+
+    struct pollfd fd;
+    fd.fd = (socket_t)m_sock;
+    fd.events = POLLIN;
+    if( poll( &fd, 1, 10 ) <= 0 ) return nullptr;
+
+    sockaddr sa;
+    socklen_t salen = sizeof( struct sockaddr );
+    len = (size_t)recvfrom( m_sock, buf, 2048, 0, &sa, &salen );
+    addr.Set( sa );
+
+    return buf;
+}
+
 }

common/TracySocket.hpp

@@ -1,34 +1,51 @@
 #ifndef __TRACYSOCKET_HPP__
 #define __TRACYSOCKET_HPP__
 
-#include <functional>
+#include <atomic>
+#include <stdint.h>
 
-struct timeval;
+#include "TracyForceInline.hpp"
+
+struct addrinfo;
+struct sockaddr;
 
 namespace tracy
 {
 
-#ifdef _MSC_VER
+#ifdef _WIN32
 void InitWinSock();
 #endif
 
 class Socket
 {
-    enum { BufSize = 128 * 1024 };
-
 public:
     Socket();
     Socket( int sock );
     ~Socket();
 
-    bool Connect( const char* addr, const char* port );
+    bool Connect( const char* addr, int port );
     void Close();
 
     int Send( const void* buf, int len );
+    int GetSendBufSize();
 
-    bool Read( void* buf, int len, const timeval* tv, std::function<bool()> exitCb );
-    bool ReadRaw( void* buf, int len, const timeval* tv );
+    bool Read( void* buf, int len, int timeout );
+
+    template<typename ShouldExit>
+    bool Read( void* buf, int len, int timeout, ShouldExit exitCb )
+    {
+        auto cbuf = (char*)buf;
+        while( len > 0 )
+        {
+            if( exitCb() ) return false;
+            if( !ReadImpl( cbuf, len, timeout ) ) return false;
+        }
+        return true;
+    }
+
+    bool ReadRaw( void* buf, int len, int timeout );
     bool HasData();
+    bool IsValid() const;
 
     Socket( const Socket& ) = delete;
     Socket( Socket&& ) = delete;
@@ -36,14 +53,19 @@ public:
     Socket& operator=( Socket&& ) = delete;
 
 private:
-    int RecvBuffered( void* buf, int len, const timeval* tv );
-    int Recv( void* buf, int len, const timeval* tv );
+    int RecvBuffered( void* buf, int len, int timeout );
+    int Recv( void* buf, int len, int timeout );
 
-    int m_sock;
+    bool ReadImpl( char*& buf, int& len, int timeout );
 
     char* m_buf;
     char* m_bufPtr;
+    std::atomic<int> m_sock;
     int m_bufLeft;
+
+    struct addrinfo *m_res;
+    struct addrinfo *m_ptr;
+    int m_connSock;
 };
 
 class ListenSocket
@@ -52,7 +74,7 @@ public:
     ListenSocket();
     ~ListenSocket();
 
-    bool Listen( const char* port, int backlog );
+    bool Listen( int port, int backlog );
     Socket* Accept();
     void Close();
 
@@ -65,6 +87,67 @@ private:
     int m_sock;
 };
 
+class UdpBroadcast
+{
+public:
+    UdpBroadcast();
+    ~UdpBroadcast();
+
+    bool Open( const char* addr, int port );
+    void Close();
+
+    int Send( int port, const void* data, int len );
+
+    UdpBroadcast( const UdpBroadcast& ) = delete;
+    UdpBroadcast( UdpBroadcast&& ) = delete;
+    UdpBroadcast& operator=( const UdpBroadcast& ) = delete;
+    UdpBroadcast& operator=( UdpBroadcast&& ) = delete;
+
+private:
+    int m_sock;
+};
+
+class IpAddress
+{
+public:
+    IpAddress();
+    ~IpAddress();
+
+    void Set( const struct sockaddr& addr );
+
+    uint32_t GetNumber() const { return m_number; }
+    const char* GetText() const { return m_text; }
+
+    IpAddress( const IpAddress& ) = delete;
+    IpAddress( IpAddress&& ) = delete;
+    IpAddress& operator=( const IpAddress& ) = delete;
+    IpAddress& operator=( IpAddress&& ) = delete;
+
+private:
+    uint32_t m_number;
+    char m_text[17];
+};
+
+class UdpListen
+{
+public:
+    UdpListen();
+    ~UdpListen();
+
+    bool Listen( int port );
+    void Close();
+
+    const char* Read( size_t& len, IpAddress& addr );
+
+    UdpListen( const UdpListen& ) = delete;
+    UdpListen( UdpListen&& ) = delete;
+    UdpListen& operator=( const UdpListen& ) = delete;
+    UdpListen& operator=( UdpListen&& ) = delete;
+
+private:
+    int m_sock;
+};
+
 }
 
 #endif

common/TracySystem.cpp

@@ -6,7 +6,10 @@
 #  define NOMINMAX
 # endif
 #endif
-#ifdef _WIN32
+#ifdef _MSC_VER
+#  pragma warning(disable:4996)
+#endif
+#if defined _WIN32 || defined __CYGWIN__
 #  include <windows.h>
 #else
 #  include <pthread.h>
@@ -15,18 +18,33 @@
 #endif
 
 #ifdef __linux__
-#   ifndef __ANDROID__
-#       include <syscall.h>
-#   endif
-#   include <fcntl.h>
+#  ifdef __ANDROID__
+#    include <sys/types.h>
+#  else
+#    include <sys/syscall.h>
+#  endif
+#  include <fcntl.h>
+#elif defined __FreeBSD__
+#  include <sys/thr.h>
+#elif defined __NetBSD__ || defined __DragonFly__
+#  include <sys/lwp.h>
 #endif
 
+#ifdef __MINGW32__
+#  define __STDC_FORMAT_MACROS
+#endif
 #include <inttypes.h>
 #include <stdio.h>
+#include <stdlib.h>
 
 #include "TracySystem.hpp"
 
-#ifdef TRACY_COLLECT_THREAD_NAMES
+#if defined _WIN32 || defined __CYGWIN__
+extern "C" typedef HRESULT (WINAPI *t_SetThreadDescription)( HANDLE, PCWSTR );
+extern "C" typedef HRESULT (WINAPI *t_GetThreadDescription)( HANDLE, PWSTR* );
+#endif
+
+#ifdef TRACY_ENABLE
 #  include <atomic>
 #  include "TracyAlloc.hpp"
 #endif
@@ -34,98 +52,129 @@
 namespace tracy
 {
 
-#ifdef TRACY_COLLECT_THREAD_NAMES
+namespace detail
+{
+
+TRACY_API uint64_t GetThreadHandleImpl()
+{
+#if defined _WIN32 || defined __CYGWIN__
+    static_assert( sizeof( decltype( GetCurrentThreadId() ) ) <= sizeof( uint64_t ), "Thread handle too big to fit in protocol" );
+    return uint64_t( GetCurrentThreadId() );
+#elif defined __APPLE__
+    uint64_t id;
+    pthread_threadid_np( pthread_self(), &id );
+    return id;
+#elif defined __ANDROID__
+    return (uint64_t)gettid();
+#elif defined __linux__
+    return (uint64_t)syscall( SYS_gettid );
+#elif defined __FreeBSD__
+    long id;
+    thr_self( &id );
+    return id;
+#elif defined __NetBSD__
+    return _lwp_self();
+#elif defined __DragonFly__
+    return lwp_gettid();
+#elif defined __OpenBSD__
+    return getthrid();
+#else
+    static_assert( sizeof( decltype( pthread_self() ) ) <= sizeof( uint64_t ), "Thread handle too big to fit in protocol" );
+    return uint64_t( pthread_self() );
+#endif
+
+}
+
+}
+
+#ifdef TRACY_ENABLE
 struct ThreadNameData
 {
     uint64_t id;
     const char* name;
     ThreadNameData* next;
 };
-extern std::atomic<ThreadNameData*>& s_threadNameData;
+std::atomic<ThreadNameData*>& GetThreadNameData();
+TRACY_API void InitRPMallocThread();
 #endif
 
-void SetThreadName( std::thread& thread, const char* name )
+TRACY_API void SetThreadName( const char* name )
 {
-    SetThreadName( thread.native_handle(), name );
-}
-
-void SetThreadName( std::thread::native_handle_type handle, const char* name )
-{
-#ifdef _WIN32
-#  if defined NTDDI_WIN10_RS2 && NTDDI_VERSION >= NTDDI_WIN10_RS2
-    wchar_t buf[256];
-    mbstowcs( buf, name, 256 );
-    SetThreadDescription( static_cast<HANDLE>( handle ), buf );
-#  else
-    const DWORD MS_VC_EXCEPTION=0x406D1388;
-#    pragma pack( push, 8 )
-    struct THREADNAME_INFO
+#if defined _WIN32 || defined __CYGWIN__
+    static auto _SetThreadDescription = (t_SetThreadDescription)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "SetThreadDescription" );
+    if( _SetThreadDescription )
     {
-        DWORD dwType;
-        LPCSTR szName;
-        DWORD dwThreadID;
-        DWORD dwFlags;
-    };
+        wchar_t buf[256];
+        mbstowcs( buf, name, 256 );
+        _SetThreadDescription( GetCurrentThread(), buf );
+    }
+    else
+    {
+#  if defined _MSC_VER
+        const DWORD MS_VC_EXCEPTION=0x406D1388;
+#    pragma pack( push, 8 )
+        struct THREADNAME_INFO
+        {
+            DWORD dwType;
+            LPCSTR szName;
+            DWORD dwThreadID;
+            DWORD dwFlags;
+        };
 #    pragma pack(pop)
 
-    DWORD ThreadId = GetThreadId( static_cast<HANDLE>( handle ) );
-    THREADNAME_INFO info;
-    info.dwType = 0x1000;
-    info.szName = name;
-    info.dwThreadID = ThreadId;
-    info.dwFlags = 0;
+        DWORD ThreadId = GetCurrentThreadId();
+        THREADNAME_INFO info;
+        info.dwType = 0x1000;
+        info.szName = name;
+        info.dwThreadID = ThreadId;
+        info.dwFlags = 0;
 
-    __try
-    {
-        RaiseException( MS_VC_EXCEPTION, 0, sizeof(info)/sizeof(ULONG_PTR), (ULONG_PTR*)&info );
-    }
-    __except(EXCEPTION_EXECUTE_HANDLER)
-    {
-    }
+        __try
+        {
+            RaiseException( MS_VC_EXCEPTION, 0, sizeof(info)/sizeof(ULONG_PTR), (ULONG_PTR*)&info );
+        }
+        __except(EXCEPTION_EXECUTE_HANDLER)
+        {
+        }
 #  endif
-#elif defined _GNU_SOURCE && !defined __EMSCRIPTEN__
+    }
+#elif defined _GNU_SOURCE && !defined __EMSCRIPTEN__ && !defined __CYGWIN__
     {
         const auto sz = strlen( name );
         if( sz <= 15 )
         {
-            pthread_setname_np( handle, name );
+            pthread_setname_np( pthread_self(), name );
         }
         else
         {
             char buf[16];
             memcpy( buf, name, 15 );
             buf[15] = '\0';
-            pthread_setname_np( handle, buf );
+            pthread_setname_np( pthread_self(), buf );
         }
     }
 #endif
-#ifdef TRACY_COLLECT_THREAD_NAMES
+#ifdef TRACY_ENABLE
     {
-        rpmalloc_thread_initialize();
+        InitRPMallocThread();
         const auto sz = strlen( name );
         char* buf = (char*)tracy_malloc( sz+1 );
         memcpy( buf, name, sz );
-        buf[sz+1] = '\0';
+        buf[sz] = '\0';
         auto data = (ThreadNameData*)tracy_malloc( sizeof( ThreadNameData ) );
-#  ifdef _WIN32
-        data->id = GetThreadId( static_cast<HANDLE>( handle ) );
-#  elif defined __APPLE__
-        pthread_threadid_np( handle, &data->id );
-#  else
-        data->id = (uint64_t)handle;
-#  endif
+        data->id = detail::GetThreadHandleImpl();
         data->name = buf;
-        data->next = s_threadNameData.load( std::memory_order_relaxed );
-        while( !s_threadNameData.compare_exchange_weak( data->next, data, std::memory_order_release, std::memory_order_relaxed ) ) {}
+        data->next = GetThreadNameData().load( std::memory_order_relaxed );
+        while( !GetThreadNameData().compare_exchange_weak( data->next, data, std::memory_order_release, std::memory_order_relaxed ) ) {}
     }
 #endif
 }
 
-const char* GetThreadName( uint64_t id )
+TRACY_API const char* GetThreadName( uint64_t id )
 {
     static char buf[256];
-#ifdef TRACY_COLLECT_THREAD_NAMES
-    auto ptr = s_threadNameData.load( std::memory_order_relaxed );
+#ifdef TRACY_ENABLE
+    auto ptr = GetThreadNameData().load( std::memory_order_relaxed );
     while( ptr )
     {
         if( ptr->id == id )
@@ -135,26 +184,23 @@ const char* GetThreadName( uint64_t id )
         ptr = ptr->next;
     }
 #else
-#  ifdef _WIN32
-#    if defined NTDDI_WIN10_RS2 && NTDDI_VERSION >= NTDDI_WIN10_RS2
-    auto hnd = OpenThread( THREAD_QUERY_LIMITED_INFORMATION, FALSE, (DWORD)id );
-    if( hnd != 0 )
+#  if defined _WIN32 || defined __CYGWIN__
+    static auto _GetThreadDescription = (t_GetThreadDescription)GetProcAddress( GetModuleHandleA( "kernel32.dll" ), "GetThreadDescription" );
+    if( _GetThreadDescription )
     {
-        PWSTR tmp;
-        GetThreadDescription( hnd, &tmp );
-        auto ret = wcstombs( buf, tmp, 256 );
-        CloseHandle( hnd );
-        if( ret != 0 )
+        auto hnd = OpenThread( THREAD_QUERY_LIMITED_INFORMATION, FALSE, (DWORD)id );
+        if( hnd != 0 )
         {
-            return buf;
+            PWSTR tmp;
+            _GetThreadDescription( hnd, &tmp );
+            auto ret = wcstombs( buf, tmp, 256 );
+            CloseHandle( hnd );
+            if( ret != 0 )
+            {
+                return buf;
+            }
         }
     }
-#    endif
-#  elif defined __GLIBC__ && !defined __ANDROID__ && !defined __EMSCRIPTEN__
-    if( pthread_getname_np( (pthread_t)id, buf, 256 ) == 0 )
-    {
-        return buf;
-    }
 #  elif defined __linux__
     int cs, fd;
     char path[32];
@@ -170,8 +216,14 @@ const char* GetThreadName( uint64_t id )
 #   endif
     if ( ( fd = open( path, O_RDONLY ) ) > 0) {
         int len = read( fd, buf, 255 );
-        if ( len > 0 )
+        if( len > 0 )
+        {
             buf[len] = 0;
+            if( len > 1 && buf[len-1] == '\n' )
+            {
+                buf[len-1] = 0;
+            }
+        }
         close( fd );
     }
 #   ifndef __ANDROID__

common/TracySystem.hpp

@@ -1,44 +1,29 @@
 #ifndef __TRACYSYSTEM_HPP__
 #define __TRACYSYSTEM_HPP__
 
-#ifdef TRACY_ENABLE
-#  if defined __ANDROID__ || defined __CYGWIN__ || defined __APPLE__ || defined _GNU_SOURCE || ( defined _WIN32 && ( !defined NTDDI_WIN10_RS2 || NTDDI_VERSION < NTDDI_WIN10_RS2 ) )
-#    define TRACY_COLLECT_THREAD_NAMES
-#  endif
-#endif
-
-#ifdef _WIN32
-#ifndef _WINDOWS_
-extern "C" __declspec(dllimport) unsigned long __stdcall GetCurrentThreadId(void);
-#endif
-#else
-#  include <pthread.h>
-#endif
-
 #include <stdint.h>
-#include <thread>
+
+#include "TracyApi.h"
 
 namespace tracy
 {
 
+namespace detail
+{
+TRACY_API uint64_t GetThreadHandleImpl();
+}
+
+#ifdef TRACY_ENABLE
+TRACY_API uint64_t GetThreadHandle();
+#else
 static inline uint64_t GetThreadHandle()
 {
-#ifdef _WIN32
-    static_assert( sizeof( decltype( GetCurrentThreadId() ) ) <= sizeof( uint64_t ), "Thread handle too big to fit in protocol" );
-    return uint64_t( GetCurrentThreadId() );
-#elif defined __APPLE__
-    uint64_t id;
-    pthread_threadid_np( pthread_self(), &id );
-    return id;
-#else
-    static_assert( sizeof( decltype( pthread_self() ) ) <= sizeof( uint64_t ), "Thread handle too big to fit in protocol" );
-    return uint64_t( pthread_self() );
-#endif
+    return detail::GetThreadHandleImpl();
 }
+#endif
 
-void SetThreadName( std::thread& thread, const char* name );
-void SetThreadName( std::thread::native_handle_type handle, const char* name );
-const char* GetThreadName( uint64_t id );
+TRACY_API void SetThreadName( const char* name );
+TRACY_API const char* GetThreadName( uint64_t id );
 
 }
 

common/tracy_benaphore.h

@@ -1,68 +0,0 @@
-// Copyright (c) 2015 Jeff Preshing
-//
-// This software is provided 'as-is', without any express or implied
-// warranty. In no event will the authors be held liable for any damages
-// arising from the use of this software.
-//
-// Permission is granted to anyone to use this software for any purpose,
-// including commercial applications, and to alter it and redistribute it
-// freely, subject to the following restrictions:
-//
-// 1. The origin of this software must not be misrepresented; you must not
-// claim that you wrote the original software. If you use this software
-// in a product, an acknowledgement in the product documentation would be
-// appreciated but is not required.
-// 2. Altered source versions must be plainly marked as such, and must not be
-// misrepresented as being the original software.
-// 3. This notice may not be removed or altered from any source distribution.
-
-#ifndef __TRACY_CPP11OM_BENAPHORE_H__
-#define __TRACY_CPP11OM_BENAPHORE_H__
-
-#include <cassert>
-#include <thread>
-#include <atomic>
-#include "tracy_sema.h"
-
-namespace tracy
-{
-
-class NonRecursiveBenaphore
-{
-private:
-    std::atomic<int> m_contentionCount;
-    DefaultSemaphoreType m_sema;
-
-public:
-    NonRecursiveBenaphore() : m_contentionCount(0) {}
-
-    void lock()
-    {
-        if (m_contentionCount.fetch_add(1, std::memory_order_acquire) > 0)
-        {
-            m_sema.wait();
-        }
-    }
-
-    bool try_lock()
-    {
-        if (m_contentionCount.load(std::memory_order_relaxed) != 0)
-            return false;
-        int expected = 0;
-        return m_contentionCount.compare_exchange_strong(expected, 1, std::memory_order_acquire);
-    }
-
-    void unlock()
-    {
-        int oldCount = m_contentionCount.fetch_sub(1, std::memory_order_release);
-        assert(oldCount > 0);
-        if (oldCount > 1)
-        {
-            m_sema.signal();
-        }
-    }
-};
-
-}
-
-#endif // __CPP11OM_BENAPHORE_H__

common/tracy_lz4.hpp

@@ -52,19 +52,23 @@ namespace tracy
   multiple GB/s per core, typically reaching RAM speed limits on multi-core systems.
 
   The LZ4 compression library provides in-memory compression and decompression functions.
+  It gives full buffer control to user.
   Compression can be done in:
     - a single step (described as Simple Functions)
     - a single step, reusing a context (described in Advanced Functions)
     - unbounded multiple steps (described as Streaming compression)
 
-  lz4.h provides block compression functions. It gives full buffer control to user.
-  Decompressing an lz4-compressed block also requires metadata (such as compressed size).
-  Each application is free to encode such metadata in whichever way it wants.
+  lz4.h generates and decodes LZ4-compressed blocks (doc/lz4_Block_format.md).
+  Decompressing a block requires additional metadata, such as its compressed size.
+  Each application is free to encode and pass such metadata in whichever way it wants.
 
-  An additional format, called LZ4 frame specification (doc/lz4_Frame_format.md),
-  take care of encoding standard metadata alongside LZ4-compressed blocks.
-  Frame format is required for interoperability.
-  It is delivered through a companion API, declared in lz4frame.h.
+  lz4.h only handle blocks, it can not generate Frames.
+
+  Blocks are different from Frames (doc/lz4_Frame_format.md).
+  Frames bundle both blocks and metadata in a specified manner.
+  This are required for compressed data to be self-contained and portable.
+  Frame format is delivered through a companion API, declared in lz4frame.h.
+  Note that the `lz4` CLI can only manage frames.
 */
 
 /*^***************************************************************
@@ -93,8 +97,8 @@ namespace tracy
 
 /*------   Version   ------*/
 #define LZ4_VERSION_MAJOR    1    /* for breaking interface changes  */
-#define LZ4_VERSION_MINOR    8    /* for new (non-breaking) interface capabilities */
-#define LZ4_VERSION_RELEASE  3    /* for tweaks, bug-fixes, or development */
+#define LZ4_VERSION_MINOR    9    /* for new (non-breaking) interface capabilities */
+#define LZ4_VERSION_RELEASE  1    /* for tweaks, bug-fixes, or development */
 
 #define LZ4_VERSION_NUMBER (LZ4_VERSION_MAJOR *100*100 + LZ4_VERSION_MINOR *100 + LZ4_VERSION_RELEASE)
 
@@ -104,7 +108,7 @@ namespace tracy
 #define LZ4_VERSION_STRING LZ4_EXPAND_AND_QUOTE(LZ4_LIB_VERSION)
 
 LZ4LIB_API int LZ4_versionNumber (void);  /**< library version number; useful to check dll version */
-LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string; unseful to check dll version */
+LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string; useful to check dll version */
 
 
 /*-************************************
@@ -113,14 +117,15 @@ LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string;
 /*!
  * LZ4_MEMORY_USAGE :
  * Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)
- * Increasing memory usage improves compression ratio
- * Reduced memory usage may improve speed, thanks to cache effect
+ * Increasing memory usage improves compression ratio.
+ * Reduced memory usage may improve speed, thanks to better cache locality.
  * Default value is 14, for 16KB, which nicely fits into Intel x86 L1 cache
  */
 #ifndef LZ4_MEMORY_USAGE
-# define LZ4_MEMORY_USAGE 12
+# define LZ4_MEMORY_USAGE 14
 #endif
 
+
 /*-************************************
 *  Simple Functions
 **************************************/
@@ -131,21 +136,22 @@ LZ4LIB_API const char* LZ4_versionString (void);   /**< library version string;
     It also runs faster, so it's a recommended setting.
     If the function cannot compress 'src' into a more limited 'dst' budget,
     compression stops *immediately*, and the function result is zero.
-    Note : as a consequence, 'dst' content is not valid.
-    Note 2 : This function is protected against buffer overflow scenarios (never writes outside 'dst' buffer, nor read outside 'source' buffer).
+    In which case, 'dst' content is undefined (invalid).
         srcSize : max supported value is LZ4_MAX_INPUT_SIZE.
         dstCapacity : size of buffer 'dst' (which must be already allocated)
-        return  : the number of bytes written into buffer 'dst' (necessarily <= dstCapacity)
-                  or 0 if compression fails */
+       @return  : the number of bytes written into buffer 'dst' (necessarily <= dstCapacity)
+                  or 0 if compression fails
+    Note : This function is protected against buffer overflow scenarios (never writes outside 'dst' buffer, nor read outside 'source' buffer).
+*/
 LZ4LIB_API int LZ4_compress_default(const char* src, char* dst, int srcSize, int dstCapacity);
 
 /*! LZ4_decompress_safe() :
     compressedSize : is the exact complete size of the compressed block.
     dstCapacity : is the size of destination buffer, which must be already allocated.
-    return : the number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
+   @return : the number of bytes decompressed into destination buffer (necessarily <= dstCapacity)
              If destination buffer is not large enough, decoding will stop and output an error code (negative value).
              If the source stream is detected malformed, the function will stop decoding and return a negative result.
-             This function is protected against malicious data packets.
+    Note : This function is protected against malicious data packets (never writes outside 'dst' buffer, nor read outside 'source' buffer).
 */
 LZ4LIB_API int LZ4_decompress_safe (const char* src, char* dst, int compressedSize, int dstCapacity);
 
@@ -156,8 +162,7 @@ LZ4LIB_API int LZ4_decompress_safe (const char* src, char* dst, int compressedSi
 #define LZ4_MAX_INPUT_SIZE        0x7E000000   /* 2 113 929 216 bytes */
 #define LZ4_COMPRESSBOUND(isize)  ((unsigned)(isize) > (unsigned)LZ4_MAX_INPUT_SIZE ? 0 : (isize) + ((isize)/255) + 16)
 
-/*!
-LZ4_compressBound() :
+/*! LZ4_compressBound() :
     Provides the maximum size that LZ4 compression may output in a "worst case" scenario (input data not compressible)
     This function is primarily useful for memory allocation purposes (destination buffer size).
     Macro LZ4_COMPRESSBOUND() is also provided for compilation-time evaluation (stack memory allocation for example).
@@ -168,8 +173,7 @@ LZ4_compressBound() :
 */
 LZ4LIB_API int LZ4_compressBound(int inputSize);
 
-/*!
-LZ4_compress_fast() :
+/*! LZ4_compress_fast() :
     Same as LZ4_compress_default(), but allows selection of "acceleration" factor.
     The larger the acceleration value, the faster the algorithm, but also the lesser the compression.
     It's a trade-off. It can be fine tuned, with each successive value providing roughly +~3% to speed.
@@ -179,13 +183,12 @@ LZ4_compress_fast() :
 LZ4LIB_API int LZ4_compress_fast (const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
 
 
-/*!
-LZ4_compress_fast_extState() :
-    Same compression function, just using an externally allocated memory space to store compression state.
-    Use LZ4_sizeofState() to know how much memory must be allocated,
-    and allocate it on 8-bytes boundaries (using malloc() typically).
-    Then, provide this buffer as 'void* state' to compression function.
-*/
+/*! LZ4_compress_fast_extState() :
+ *  Same as LZ4_compress_fast(), using an externally allocated memory space for its state.
+ *  Use LZ4_sizeofState() to know how much memory must be allocated,
+ *  and allocate it on 8-bytes boundaries (using `malloc()` typically).
+ *  Then, provide this buffer as `void* state` to compression function.
+ */
 LZ4LIB_API int LZ4_sizeofState(void);
 LZ4LIB_API int LZ4_compress_fast_extState (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
 
@@ -205,27 +208,6 @@ LZ4LIB_API int LZ4_compress_fast_extState (void* state, const char* src, char* d
 LZ4LIB_API int LZ4_compress_destSize (const char* src, char* dst, int* srcSizePtr, int targetDstSize);
 
 
-/*! LZ4_decompress_fast() : **unsafe!**
- *  This function used to be a bit faster than LZ4_decompress_safe(),
- *  though situation has changed in recent versions,
- *  and now `LZ4_decompress_safe()` can be as fast and sometimes faster than `LZ4_decompress_fast()`.
- *  Moreover, LZ4_decompress_fast() is not protected vs malformed input, as it doesn't perform full validation of compressed data.
- *  As a consequence, this function is no longer recommended, and may be deprecated in future versions.
- *  It's only remaining specificity is that it can decompress data without knowing its compressed size.
- *
- *  originalSize : is the uncompressed size to regenerate.
- *                 `dst` must be already allocated, its size must be >= 'originalSize' bytes.
- * @return : number of bytes read from source buffer (== compressed size).
- *           If the source stream is detected malformed, the function stops decoding and returns a negative result.
- *  note : This function requires uncompressed originalSize to be known in advance.
- *         The function never writes past the output buffer.
- *         However, since it doesn't know its 'src' size, it may read past the intended input.
- *         Also, because match offsets are not validated during decoding,
- *         reads from 'src' may underflow.
- *         Use this function in trusted environment **only**.
- */
-LZ4LIB_API int LZ4_decompress_fast (const char* src, char* dst, int originalSize);
-
 /*! LZ4_decompress_safe_partial() :
  *  Decompress an LZ4 compressed block, of size 'srcSize' at position 'src',
  *  into destination buffer 'dst' of size 'dstCapacity'.
@@ -258,30 +240,49 @@ LZ4LIB_API int LZ4_decompress_safe_partial (const char* src, char* dst, int srcS
 ***********************************************/
 typedef union LZ4_stream_u LZ4_stream_t;  /* incomplete type (defined later) */
 
-/*! LZ4_createStream() and LZ4_freeStream() :
- *  LZ4_createStream() will allocate and initialize an `LZ4_stream_t` structure.
- *  LZ4_freeStream() releases its memory.
- */
 LZ4LIB_API LZ4_stream_t* LZ4_createStream(void);
 LZ4LIB_API int           LZ4_freeStream (LZ4_stream_t* streamPtr);
 
-/*! LZ4_resetStream() :
- *  An LZ4_stream_t structure can be allocated once and re-used multiple times.
- *  Use this function to start compressing a new stream.
+/*! LZ4_resetStream_fast() : v1.9.0+
+ *  Use this to prepare an LZ4_stream_t for a new chain of dependent blocks
+ *  (e.g., LZ4_compress_fast_continue()).
+ *
+ *  An LZ4_stream_t must be initialized once before usage.
+ *  This is automatically done when created by LZ4_createStream().
+ *  However, should the LZ4_stream_t be simply declared on stack (for example),
+ *  it's necessary to initialize it first, using LZ4_initStream().
+ *
+ *  After init, start any new stream with LZ4_resetStream_fast().
+ *  A same LZ4_stream_t can be re-used multiple times consecutively
+ *  and compress multiple streams,
+ *  provided that it starts each new stream with LZ4_resetStream_fast().
+ *
+ *  LZ4_resetStream_fast() is much faster than LZ4_initStream(),
+ *  but is not compatible with memory regions containing garbage data.
+ *
+ *  Note: it's only useful to call LZ4_resetStream_fast()
+ *        in the context of streaming compression.
+ *        The *extState* functions perform their own resets.
+ *        Invoking LZ4_resetStream_fast() before is redundant, and even counterproductive.
  */
-LZ4LIB_API void LZ4_resetStream (LZ4_stream_t* streamPtr);
+LZ4LIB_API void LZ4_resetStream_fast (LZ4_stream_t* streamPtr);
 
 /*! LZ4_loadDict() :
- *  Use this function to load a static dictionary into LZ4_stream_t.
- *  Any previous data will be forgotten, only 'dictionary' will remain in memory.
+ *  Use this function to reference a static dictionary into LZ4_stream_t.
+ *  The dictionary must remain available during compression.
+ *  LZ4_loadDict() triggers a reset, so any previous data will be forgotten.
+ *  The same dictionary will have to be loaded on decompression side for successful decoding.
+ *  Dictionary are useful for better compression of small data (KB range).
+ *  While LZ4 accept any input as dictionary,
+ *  results are generally better when using Zstandard's Dictionary Builder.
  *  Loading a size of 0 is allowed, and is the same as reset.
- * @return : dictionary size, in bytes (necessarily <= 64 KB)
+ * @return : loaded dictionary size, in bytes (necessarily <= 64 KB)
  */
 LZ4LIB_API int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, int dictSize);
 
 /*! LZ4_compress_fast_continue() :
  *  Compress 'src' content using data from previously compressed blocks, for better compression ratio.
- *  'dst' buffer must be already allocated.
+ * 'dst' buffer must be already allocated.
  *  If dstCapacity >= LZ4_compressBound(srcSize), compression is guaranteed to succeed, and runs faster.
  *
  * @return : size of compressed block
@@ -289,10 +290,10 @@ LZ4LIB_API int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, in
  *
  *  Note 1 : Each invocation to LZ4_compress_fast_continue() generates a new block.
  *           Each block has precise boundaries.
+ *           Each block must be decompressed separately, calling LZ4_decompress_*() with relevant metadata.
  *           It's not possible to append blocks together and expect a single invocation of LZ4_decompress_*() to decompress them together.
- *           Each block must be decompressed separately, calling LZ4_decompress_*() with associated metadata.
  *
- *  Note 2 : The previous 64KB of source data is __assumed__ to remain present, unmodified, at same address in memory!
+ *  Note 2 : The previous 64KB of source data is __assumed__ to remain present, unmodified, at same address in memory !
  *
  *  Note 3 : When input is structured as a double-buffer, each buffer can have any size, including < 64 KB.
  *           Make sure that buffers are separated, by at least one byte.
@@ -300,7 +301,7 @@ LZ4LIB_API int LZ4_loadDict (LZ4_stream_t* streamPtr, const char* dictionary, in
  *
  *  Note 4 : If input buffer is a ring-buffer, it can have any size, including < 64 KB.
  *
- *  Note 5 : After an error, the stream status is invalid, it can only be reset or freed.
+ *  Note 5 : After an error, the stream status is undefined (invalid), it can only be reset or freed.
  */
 LZ4LIB_API int LZ4_compress_fast_continue (LZ4_stream_t* streamPtr, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
 
@@ -336,7 +337,7 @@ LZ4LIB_API int                 LZ4_freeStreamDecode (LZ4_streamDecode_t* LZ4_str
  */
 LZ4LIB_API int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const char* dictionary, int dictSize);
 
-/*! LZ4_decoderRingBufferSize() : v1.8.2
+/*! LZ4_decoderRingBufferSize() : v1.8.2+
  *  Note : in a ring buffer scenario (optional),
  *  blocks are presumed decompressed next to each other
  *  up to the moment there is not enough remaining space for next block (remainingSize < maxBlockSize),
@@ -348,7 +349,7 @@ LZ4LIB_API int LZ4_setStreamDecode (LZ4_streamDecode_t* LZ4_streamDecode, const
  *           or 0 if there is an error (invalid maxBlockSize).
  */
 LZ4LIB_API int LZ4_decoderRingBufferSize(int maxBlockSize);
-#define LZ4_DECODER_RING_BUFFER_SIZE(mbs) (65536 + 14 + (mbs))  /* for static allocation; mbs presumed valid */
+#define LZ4_DECODER_RING_BUFFER_SIZE(maxBlockSize) (65536 + 14 + (maxBlockSize))  /* for static allocation; maxBlockSize presumed valid */
 
 /*! LZ4_decompress_*_continue() :
  *  These decoding functions allow decompression of consecutive blocks in "streaming" mode.
@@ -376,83 +377,67 @@ LZ4LIB_API int LZ4_decoderRingBufferSize(int maxBlockSize);
  *  then indicate where this data is saved using LZ4_setStreamDecode(), before decompressing next block.
 */
 LZ4LIB_API int LZ4_decompress_safe_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int srcSize, int dstCapacity);
-LZ4LIB_API int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int originalSize);
 
 
 /*! LZ4_decompress_*_usingDict() :
  *  These decoding functions work the same as
  *  a combination of LZ4_setStreamDecode() followed by LZ4_decompress_*_continue()
  *  They are stand-alone, and don't need an LZ4_streamDecode_t structure.
- *  Dictionary is presumed stable : it must remain accessible and unmodified during next decompression.
+ *  Dictionary is presumed stable : it must remain accessible and unmodified during decompression.
+ *  Performance tip : Decompression speed can be substantially increased
+ *                    when dst == dictStart + dictSize.
  */
 LZ4LIB_API int LZ4_decompress_safe_usingDict (const char* src, char* dst, int srcSize, int dstCapcity, const char* dictStart, int dictSize);
-LZ4LIB_API int LZ4_decompress_fast_usingDict (const char* src, char* dst, int originalSize, const char* dictStart, int dictSize);
 
 
-/*^**********************************************
+/*^*************************************
  * !!!!!!   STATIC LINKING ONLY   !!!!!!
- ***********************************************/
+ ***************************************/
 
-/*-************************************
- *  Unstable declarations
- **************************************
- * Declarations in this section should be considered unstable.
- * Use at your own peril, etc., etc.
- * They may be removed in the future.
- * Their signatures may change.
- **************************************/
+/*-****************************************************************************
+ * Experimental section
+ *
+ * Symbols declared in this section must be considered unstable. Their
+ * signatures or semantics may change, or they may be removed altogether in the
+ * future. They are therefore only safe to depend on when the caller is
+ * statically linked against the library.
+ *
+ * To protect against unsafe usage, not only are the declarations guarded,
+ * the definitions are hidden by default
+ * when building LZ4 as a shared/dynamic library.
+ *
+ * In order to access these declarations,
+ * define LZ4_STATIC_LINKING_ONLY in your application
+ * before including LZ4's headers.
+ *
+ * In order to make their implementations accessible dynamically, you must
+ * define LZ4_PUBLISH_STATIC_FUNCTIONS when building the LZ4 library.
+ ******************************************************************************/
+
+#ifdef LZ4_PUBLISH_STATIC_FUNCTIONS
+#define LZ4LIB_STATIC_API LZ4LIB_API
+#else
+#define LZ4LIB_STATIC_API
+#endif
 
 #ifdef LZ4_STATIC_LINKING_ONLY
 
-/*! LZ4_resetStream_fast() :
- *  Use this, like LZ4_resetStream(), to prepare a context for a new chain of
- *  calls to a streaming API (e.g., LZ4_compress_fast_continue()).
- *
- *  Note:
- *  Using this in advance of a non- streaming-compression function is redundant,
- *  and potentially bad for performance, since they all perform their own custom
- *  reset internally.
- *
- *  Differences from LZ4_resetStream():
- *  When an LZ4_stream_t is known to be in a internally coherent state,
- *  it can often be prepared for a new compression with almost no work, only
- *  sometimes falling back to the full, expensive reset that is always required
- *  when the stream is in an indeterminate state (i.e., the reset performed by
- *  LZ4_resetStream()).
- *
- *  LZ4_streams are guaranteed to be in a valid state when:
- *  - returned from LZ4_createStream()
- *  - reset by LZ4_resetStream()
- *  - memset(stream, 0, sizeof(LZ4_stream_t)), though this is discouraged
- *  - the stream was in a valid state and was reset by LZ4_resetStream_fast()
- *  - the stream was in a valid state and was then used in any compression call
- *    that returned success
- *  - the stream was in an indeterminate state and was used in a compression
- *    call that fully reset the state (e.g., LZ4_compress_fast_extState()) and
- *    that returned success
- *
- *  When a stream isn't known to be in a valid state, it is not safe to pass to
- *  any fastReset or streaming function. It must first be cleansed by the full
- *  LZ4_resetStream().
- */
-LZ4LIB_API void LZ4_resetStream_fast (LZ4_stream_t* streamPtr);
 
 /*! LZ4_compress_fast_extState_fastReset() :
  *  A variant of LZ4_compress_fast_extState().
  *
- *  Using this variant avoids an expensive initialization step. It is only safe
- *  to call if the state buffer is known to be correctly initialized already
- *  (see above comment on LZ4_resetStream_fast() for a definition of "correctly
- *  initialized"). From a high level, the difference is that this function
- *  initializes the provided state with a call to something like
- *  LZ4_resetStream_fast() while LZ4_compress_fast_extState() starts with a
- *  call to LZ4_resetStream().
+ *  Using this variant avoids an expensive initialization step.
+ *  It is only safe to call if the state buffer is known to be correctly initialized already
+ *  (see above comment on LZ4_resetStream_fast() for a definition of "correctly initialized").
+ *  From a high level, the difference is that
+ *  this function initializes the provided state with a call to something like LZ4_resetStream_fast()
+ *  while LZ4_compress_fast_extState() starts with a call to LZ4_resetStream().
  */
-LZ4LIB_API int LZ4_compress_fast_extState_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
+LZ4LIB_STATIC_API int LZ4_compress_fast_extState_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int acceleration);
 
 /*! LZ4_attach_dictionary() :
- *  This is an experimental API that allows for the efficient use of a
- *  static dictionary many times.
+ *  This is an experimental API that allows
+ *  efficient use of a static dictionary many times.
  *
  *  Rather than re-loading the dictionary buffer into a working context before
  *  each compression, or copying a pre-loaded dictionary's LZ4_stream_t into a
@@ -463,8 +448,8 @@ LZ4LIB_API int LZ4_compress_fast_extState_fastReset (void* state, const char* sr
  *  Currently, only streams which have been prepared by LZ4_loadDict() should
  *  be expected to work.
  *
- *  Alternatively, the provided dictionary stream pointer may be NULL, in which
- *  case any existing dictionary stream is unset.
+ *  Alternatively, the provided dictionaryStream may be NULL,
+ *  in which case any existing dictionary stream is unset.
  *
  *  If a dictionary is provided, it replaces any pre-existing stream history.
  *  The dictionary contents are the only history that can be referenced and
@@ -476,17 +461,18 @@ LZ4LIB_API int LZ4_compress_fast_extState_fastReset (void* state, const char* sr
  *  stream (and source buffer) must remain in-place / accessible / unchanged
  *  through the completion of the first compression call on the stream.
  */
-LZ4LIB_API void LZ4_attach_dictionary(LZ4_stream_t *working_stream, const LZ4_stream_t *dictionary_stream);
+LZ4LIB_STATIC_API void LZ4_attach_dictionary(LZ4_stream_t* workingStream, const LZ4_stream_t* dictionaryStream);
 
 #endif
 
-/*-************************************
- *  Private definitions
- **************************************
- * Do not use these definitions.
- * They are exposed to allow static allocation of `LZ4_stream_t` and `LZ4_streamDecode_t`.
- * Using these definitions will expose code to API and/or ABI break in future versions of the library.
- **************************************/
+
+/*-************************************************************
+ *  PRIVATE DEFINITIONS
+ **************************************************************
+ * Do not use these definitions directly.
+ * They are only exposed to allow static allocation of `LZ4_stream_t` and `LZ4_streamDecode_t`.
+ * Accessing members will expose code to API and/or ABI break in future versions of the library.
+ **************************************************************/
 #define LZ4_HASHLOG   (LZ4_MEMORY_USAGE-2)
 #define LZ4_HASHTABLESIZE (1 << LZ4_MEMORY_USAGE)
 #define LZ4_HASH_SIZE_U32 (1 << LZ4_HASHLOG)       /* required as macro for static allocation */
@@ -498,7 +484,7 @@ typedef struct LZ4_stream_t_internal LZ4_stream_t_internal;
 struct LZ4_stream_t_internal {
     uint32_t hashTable[LZ4_HASH_SIZE_U32];
     uint32_t currentOffset;
-    uint16_t initCheck;
+    uint16_t dirty;
     uint16_t tableType;
     const uint8_t* dictionary;
     const LZ4_stream_t_internal* dictCtx;
@@ -518,7 +504,7 @@ typedef struct LZ4_stream_t_internal LZ4_stream_t_internal;
 struct LZ4_stream_t_internal {
     unsigned int hashTable[LZ4_HASH_SIZE_U32];
     unsigned int currentOffset;
-    unsigned short initCheck;
+    unsigned short dirty;
     unsigned short tableType;
     const unsigned char* dictionary;
     const LZ4_stream_t_internal* dictCtx;
@@ -527,38 +513,54 @@ struct LZ4_stream_t_internal {
 
 typedef struct {
     const unsigned char* externalDict;
-    size_t extDictSize;
     const unsigned char* prefixEnd;
+    size_t extDictSize;
     size_t prefixSize;
 } LZ4_streamDecode_t_internal;
 
 #endif
 
-/*!
- * LZ4_stream_t :
- * information structure to track an LZ4 stream.
- * init this structure before first use.
- * note : only use in association with static linking !
- *        this definition is not API/ABI safe,
- *        it may change in a future version !
+/*! LZ4_stream_t :
+ *  information structure to track an LZ4 stream.
+ *  LZ4_stream_t can also be created using LZ4_createStream(), which is recommended.
+ *  The structure definition can be convenient for static allocation
+ *  (on stack, or as part of larger structure).
+ *  Init this structure with LZ4_initStream() before first use.
+ *  note : only use this definition in association with static linking !
+ *    this definition is not API/ABI safe, and may change in a future version.
  */
-#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE-3)) + 4)
+#define LZ4_STREAMSIZE_U64 ((1 << (LZ4_MEMORY_USAGE-3)) + 4 + ((sizeof(void*)==16) ? 4 : 0) /*AS-400*/ )
 #define LZ4_STREAMSIZE     (LZ4_STREAMSIZE_U64 * sizeof(unsigned long long))
 union LZ4_stream_u {
     unsigned long long table[LZ4_STREAMSIZE_U64];
     LZ4_stream_t_internal internal_donotuse;
 } ;  /* previously typedef'd to LZ4_stream_t */
 
-
-/*!
- * LZ4_streamDecode_t :
- * information structure to track an LZ4 stream during decompression.
- * init this structure  using LZ4_setStreamDecode (or memset()) before first use
- * note : only use in association with static linking !
- *        this definition is not API/ABI safe,
- *        and may change in a future version !
+/*! LZ4_initStream() : v1.9.0+
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is automatically done when invoking LZ4_createStream(),
+ *  but it's not when the structure is simply declared on stack (for example).
+ *
+ *  Use LZ4_initStream() to properly initialize a newly declared LZ4_stream_t.
+ *  It can also initialize any arbitrary buffer of sufficient size,
+ *  and will @return a pointer of proper type upon initialization.
+ *
+ *  Note : initialization fails if size and alignment conditions are not respected.
+ *         In which case, the function will @return NULL.
+ *  Note2: An LZ4_stream_t structure guarantees correct alignment and size.
+ *  Note3: Before v1.9.0, use LZ4_resetStream() instead
  */
-#define LZ4_STREAMDECODESIZE_U64  4
+LZ4LIB_API LZ4_stream_t* LZ4_initStream (void* buffer, size_t size);
+
+
+/*! LZ4_streamDecode_t :
+ *  information structure to track an LZ4 stream during decompression.
+ *  init this structure  using LZ4_setStreamDecode() before first use.
+ *  note : only use in association with static linking !
+ *         this definition is not API/ABI safe,
+ *         and may change in a future version !
+ */
+#define LZ4_STREAMDECODESIZE_U64 (4 + ((sizeof(void*)==16) ? 2 : 0) /*AS-400*/ )
 #define LZ4_STREAMDECODESIZE     (LZ4_STREAMDECODESIZE_U64 * sizeof(unsigned long long))
 union LZ4_streamDecode_u {
     unsigned long long table[LZ4_STREAMDECODESIZE_U64];
@@ -571,11 +573,16 @@ union LZ4_streamDecode_u {
 **************************************/
 
 /*! Deprecation warnings
-   Should deprecation warnings be a problem,
-   it is generally possible to disable them,
-   typically with -Wno-deprecated-declarations for gcc
-   or _CRT_SECURE_NO_WARNINGS in Visual.
-   Otherwise, it's also possible to define LZ4_DISABLE_DEPRECATE_WARNINGS */
+ *
+ *  Deprecated functions make the compiler generate a warning when invoked.
+ *  This is meant to invite users to update their source code.
+ *  Should deprecation warnings be a problem, it is generally possible to disable them,
+ *  typically with -Wno-deprecated-declarations for gcc
+ *  or _CRT_SECURE_NO_WARNINGS in Visual.
+ *
+ *  Another method is to define LZ4_DISABLE_DEPRECATE_WARNINGS
+ *  before including the header file.
+ */
 #ifdef LZ4_DISABLE_DEPRECATE_WARNINGS
 #  define LZ4_DEPRECATED(message)   /* disable deprecation warnings */
 #else
@@ -595,8 +602,8 @@ union LZ4_streamDecode_u {
 #endif /* LZ4_DISABLE_DEPRECATE_WARNINGS */
 
 /* Obsolete compression functions */
-LZ4_DEPRECATED("use LZ4_compress_default() instead") LZ4LIB_API int LZ4_compress               (const char* source, char* dest, int sourceSize);
-LZ4_DEPRECATED("use LZ4_compress_default() instead") LZ4LIB_API int LZ4_compress_limitedOutput (const char* source, char* dest, int sourceSize, int maxOutputSize);
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress               (const char* source, char* dest, int sourceSize);
+LZ4_DEPRECATED("use LZ4_compress_default() instead")       LZ4LIB_API int LZ4_compress_limitedOutput (const char* source, char* dest, int sourceSize, int maxOutputSize);
 LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_withState               (void* state, const char* source, char* dest, int inputSize);
 LZ4_DEPRECATED("use LZ4_compress_fast_extState() instead") LZ4LIB_API int LZ4_compress_limitedOutput_withState (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
 LZ4_DEPRECATED("use LZ4_compress_fast_continue() instead") LZ4LIB_API int LZ4_compress_continue                (LZ4_stream_t* LZ4_streamPtr, const char* source, char* dest, int inputSize);
@@ -617,13 +624,56 @@ LZ4_DEPRECATED("use LZ4_decompress_safe() instead") LZ4LIB_API int LZ4_uncompres
  */
 LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API void* LZ4_create (char* inputBuffer);
 LZ4_DEPRECATED("Use LZ4_createStream() instead") LZ4LIB_API int   LZ4_sizeofStreamState(void);
-LZ4_DEPRECATED("Use LZ4_resetStream() instead") LZ4LIB_API  int   LZ4_resetStreamState(void* state, char* inputBuffer);
-LZ4_DEPRECATED("Use LZ4_saveDict() instead") LZ4LIB_API     char* LZ4_slideInputBuffer (void* state);
+LZ4_DEPRECATED("Use LZ4_resetStream() instead")  LZ4LIB_API int   LZ4_resetStreamState(void* state, char* inputBuffer);
+LZ4_DEPRECATED("Use LZ4_saveDict() instead")     LZ4LIB_API char* LZ4_slideInputBuffer (void* state);
 
 /* Obsolete streaming decoding functions */
 LZ4_DEPRECATED("use LZ4_decompress_safe_usingDict() instead") LZ4LIB_API int LZ4_decompress_safe_withPrefix64k (const char* src, char* dst, int compressedSize, int maxDstSize);
 LZ4_DEPRECATED("use LZ4_decompress_fast_usingDict() instead") LZ4LIB_API int LZ4_decompress_fast_withPrefix64k (const char* src, char* dst, int originalSize);
 
+/*! LZ4_decompress_fast() : **unsafe!**
+ *  These functions used to be faster than LZ4_decompress_safe(),
+ *  but it has changed, and they are now slower than LZ4_decompress_safe().
+ *  This is because LZ4_decompress_fast() doesn't know the input size,
+ *  and therefore must progress more cautiously in the input buffer to not read beyond the end of block.
+ *  On top of that `LZ4_decompress_fast()` is not protected vs malformed or malicious inputs, making it a security liability.
+ *  As a consequence, LZ4_decompress_fast() is strongly discouraged, and deprecated.
+ *
+ *  The last remaining LZ4_decompress_fast() specificity is that
+ *  it can decompress a block without knowing its compressed size.
+ *  Such functionality could be achieved in a more secure manner,
+ *  by also providing the maximum size of input buffer,
+ *  but it would require new prototypes, and adaptation of the implementation to this new use case.
+ *
+ *  Parameters:
+ *  originalSize : is the uncompressed size to regenerate.
+ *                 `dst` must be already allocated, its size must be >= 'originalSize' bytes.
+ * @return : number of bytes read from source buffer (== compressed size).
+ *           The function expects to finish at block's end exactly.
+ *           If the source stream is detected malformed, the function stops decoding and returns a negative result.
+ *  note : LZ4_decompress_fast*() requires originalSize. Thanks to this information, it never writes past the output buffer.
+ *         However, since it doesn't know its 'src' size, it may read an unknown amount of input, past input buffer bounds.
+ *         Also, since match offsets are not validated, match reads from 'src' may underflow too.
+ *         These issues never happen if input (compressed) data is correct.
+ *         But they may happen if input data is invalid (error or intentional tampering).
+ *         As a consequence, use these functions in trusted environments with trusted data **only**.
+ */
+
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe() instead")
+LZ4LIB_API int LZ4_decompress_fast (const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_continue() instead")
+LZ4LIB_API int LZ4_decompress_fast_continue (LZ4_streamDecode_t* LZ4_streamDecode, const char* src, char* dst, int originalSize);
+LZ4_DEPRECATED("This function is deprecated and unsafe. Consider using LZ4_decompress_safe_usingDict() instead")
+LZ4LIB_API int LZ4_decompress_fast_usingDict (const char* src, char* dst, int originalSize, const char* dictStart, int dictSize);
+
+/*! LZ4_resetStream() :
+ *  An LZ4_stream_t structure must be initialized at least once.
+ *  This is done with LZ4_initStream(), or LZ4_resetStream().
+ *  Consider switching to LZ4_initStream(),
+ *  invoking LZ4_resetStream() will trigger deprecation warnings in the future.
+ */
+LZ4LIB_API void LZ4_resetStream (LZ4_stream_t* streamPtr);
+
 }
 
 #endif /* LZ4_H_2983827168210 */

common/tracy_lz4hc.cpp

@@ -61,8 +61,18 @@
 #  pragma clang diagnostic ignored "-Wunused-function"
 #endif
 
+namespace tracy
+{
+
+/*===   Enums   ===*/
+typedef enum { noDictCtx, usingDictCtxHc } dictCtx_directive;
+
+}
+
 #define LZ4_COMMONDEFS_ONLY
+#ifndef LZ4_SRC_INCLUDED
 #include "tracy_lz4.cpp"   /* LZ4_count, constants, mem */
+#endif
 
 namespace tracy
 {
@@ -78,12 +88,11 @@ namespace tracy
 #define HASH_FUNCTION(i)         (((i) * 2654435761U) >> ((MINMATCH*8)-LZ4HC_HASH_LOG))
 #define DELTANEXTMAXD(p)         chainTable[(p) & LZ4HC_MAXD_MASK]    /* flexible, LZ4HC_MAXD dependent */
 #define DELTANEXTU16(table, pos) table[(U16)(pos)]   /* faster */
+/* Make fields passed to, and updated by LZ4HC_encodeSequence explicit */
+#define UPDATABLE(ip, op, anchor) &ip, &op, &anchor
 
 static U32 LZ4HC_hashPtr(const void* ptr) { return HASH_FUNCTION(LZ4_read32(ptr)); }
 
-/*===   Enums   ===*/
-typedef enum { noDictCtx, usingDictCtx } dictCtx_directive;
-
 
 /**************************************
 *  HC Compression
@@ -94,9 +103,9 @@ static void LZ4HC_clearTables (LZ4HC_CCtx_internal* hc4)
     MEM_INIT(hc4->chainTable, 0xFF, sizeof(hc4->chainTable));
 }
 
-static void LZ4HC_init (LZ4HC_CCtx_internal* hc4, const BYTE* start)
+static void LZ4HC_init_internal (LZ4HC_CCtx_internal* hc4, const BYTE* start)
 {
-    uptrval startingOffset = hc4->end - hc4->base;
+    uptrval startingOffset = (uptrval)(hc4->end - hc4->base);
     if (startingOffset > 1 GB) {
         LZ4HC_clearTables(hc4);
         startingOffset = 0;
@@ -123,7 +132,7 @@ LZ4_FORCE_INLINE void LZ4HC_Insert (LZ4HC_CCtx_internal* hc4, const BYTE* ip)
     while (idx < target) {
         U32 const h = LZ4HC_hashPtr(base+idx);
         size_t delta = idx - hashTable[h];
-        if (delta>MAX_DISTANCE) delta = MAX_DISTANCE;
+        if (delta>LZ4_DISTANCE_MAX) delta = LZ4_DISTANCE_MAX;
         DELTANEXTU16(chainTable, idx) = (U16)delta;
         hashTable[h] = idx;
         idx++;
@@ -226,14 +235,13 @@ LZ4HC_InsertAndGetWiderMatch (
     const U32 dictLimit = hc4->dictLimit;
     const BYTE* const lowPrefixPtr = base + dictLimit;
     const U32 ipIndex = (U32)(ip - base);
-    const U32 lowestMatchIndex = (hc4->lowLimit + 64 KB > ipIndex) ? hc4->lowLimit : ipIndex - MAX_DISTANCE;
+    const U32 lowestMatchIndex = (hc4->lowLimit + 64 KB > ipIndex) ? hc4->lowLimit : ipIndex - LZ4_DISTANCE_MAX;
     const BYTE* const dictBase = hc4->dictBase;
     int const lookBackLength = (int)(ip-iLowLimit);
     int nbAttempts = maxNbAttempts;
-    int matchChainPos = 0;
+    U32 matchChainPos = 0;
     U32 const pattern = LZ4_read32(ip);
     U32 matchIndex;
-    U32 dictMatchIndex;
     repeat_state_e repeat = rep_untested;
     size_t srcPatternLength = 0;
 
@@ -258,7 +266,7 @@ LZ4HC_InsertAndGetWiderMatch (
             if (LZ4_read16(iLowLimit + longest - 1) == LZ4_read16(matchPtr - lookBackLength + longest - 1)) {
                 if (LZ4_read32(matchPtr) == pattern) {
                     int const back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, lowPrefixPtr) : 0;
-                    matchLength = MINMATCH + LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
+                    matchLength = MINMATCH + (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, iHighLimit);
                     matchLength -= back;
                     if (matchLength > longest) {
                         longest = matchLength;
@@ -272,7 +280,7 @@ LZ4HC_InsertAndGetWiderMatch (
                 int back = 0;
                 const BYTE* vLimit = ip + (dictLimit - matchIndex);
                 if (vLimit > iHighLimit) vLimit = iHighLimit;
-                matchLength = LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                matchLength = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
                 if ((ip+matchLength == vLimit) && (vLimit < iHighLimit))
                     matchLength += LZ4_count(ip+matchLength, lowPrefixPtr, iHighLimit);
                 back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictStart) : 0;
@@ -285,14 +293,14 @@ LZ4HC_InsertAndGetWiderMatch (
 
         if (chainSwap && matchLength==longest) {    /* better match => select a better chain */
             assert(lookBackLength==0);   /* search forward only */
-            if (matchIndex + longest <= ipIndex) {
+            if (matchIndex + (U32)longest <= ipIndex) {
                 U32 distanceToNextMatch = 1;
                 int pos;
                 for (pos = 0; pos <= longest - MINMATCH; pos++) {
-                    U32 const candidateDist = DELTANEXTU16(chainTable, matchIndex + pos);
+                    U32 const candidateDist = DELTANEXTU16(chainTable, matchIndex + (U32)pos);
                     if (candidateDist > distanceToNextMatch) {
                         distanceToNextMatch = candidateDist;
-                        matchChainPos = pos;
+                        matchChainPos = (U32)pos;
                 }   }
                 if (distanceToNextMatch > 1) {
                     if (distanceToNextMatch > matchIndex) break;   /* avoid overflow */
@@ -317,7 +325,7 @@ LZ4HC_InsertAndGetWiderMatch (
                     const BYTE* const matchPtr = base + matchCandidateIdx;
                     if (LZ4_read32(matchPtr) == pattern) {  /* good candidate */
                         size_t const forwardPatternLength = LZ4HC_countPattern(matchPtr+sizeof(pattern), iHighLimit, pattern) + sizeof(pattern);
-                        const BYTE* const lowestMatchPtr = (lowPrefixPtr + MAX_DISTANCE >= ip) ? lowPrefixPtr : ip - MAX_DISTANCE;
+                        const BYTE* const lowestMatchPtr = (lowPrefixPtr + LZ4_DISTANCE_MAX >= ip) ? lowPrefixPtr : ip - LZ4_DISTANCE_MAX;
                         size_t const backLength = LZ4HC_reverseCountPattern(matchPtr, lowestMatchPtr, pattern);
                         size_t const currentSegmentLength = backLength + forwardPatternLength;
 
@@ -330,7 +338,7 @@ LZ4HC_InsertAndGetWiderMatch (
                                 size_t const maxML = MIN(currentSegmentLength, srcPatternLength);
                                 if ((size_t)longest < maxML) {
                                     assert(base + matchIndex < ip);
-                                    if (ip - (base+matchIndex) > MAX_DISTANCE) break;
+                                    if (ip - (base+matchIndex) > LZ4_DISTANCE_MAX) break;
                                     assert(maxML < 2 GB);
                                     longest = (int)maxML;
                                     *matchpos = base + matchIndex;   /* virtual pos, relative to ip, to retrieve offset */
@@ -345,16 +353,18 @@ LZ4HC_InsertAndGetWiderMatch (
         }   }   /* PA optimization */
 
         /* follow current chain */
-        matchIndex -= DELTANEXTU16(chainTable, matchIndex+matchChainPos);
+        matchIndex -= DELTANEXTU16(chainTable, matchIndex + matchChainPos);
 
     }  /* while ((matchIndex>=lowestMatchIndex) && (nbAttempts)) */
 
-    if (dict == usingDictCtx && nbAttempts && ipIndex - lowestMatchIndex < MAX_DISTANCE) {
-        size_t const dictEndOffset = dictCtx->end - dictCtx->base;
+    if ( dict == usingDictCtxHc
+      && nbAttempts
+      && ipIndex - lowestMatchIndex < LZ4_DISTANCE_MAX) {
+        size_t const dictEndOffset = (size_t)(dictCtx->end - dictCtx->base);
+        U32 dictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
         assert(dictEndOffset <= 1 GB);
-        dictMatchIndex = dictCtx->hashTable[LZ4HC_hashPtr(ip)];
         matchIndex = dictMatchIndex + lowestMatchIndex - (U32)dictEndOffset;
-        while (ipIndex - matchIndex <= MAX_DISTANCE && nbAttempts--) {
+        while (ipIndex - matchIndex <= LZ4_DISTANCE_MAX && nbAttempts--) {
             const BYTE* const matchPtr = dictCtx->base + dictMatchIndex;
 
             if (LZ4_read32(matchPtr) == pattern) {
@@ -362,22 +372,19 @@ LZ4HC_InsertAndGetWiderMatch (
                 int back = 0;
                 const BYTE* vLimit = ip + (dictEndOffset - dictMatchIndex);
                 if (vLimit > iHighLimit) vLimit = iHighLimit;
-                mlt = LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
+                mlt = (int)LZ4_count(ip+MINMATCH, matchPtr+MINMATCH, vLimit) + MINMATCH;
                 back = lookBackLength ? LZ4HC_countBack(ip, matchPtr, iLowLimit, dictCtx->base + dictCtx->dictLimit) : 0;
                 mlt -= back;
                 if (mlt > longest) {
                     longest = mlt;
                     *matchpos = base + matchIndex + back;
                     *startpos = ip + back;
-                }
-            }
+            }   }
 
             {   U32 const nextOffset = DELTANEXTU16(dictCtx->chainTable, dictMatchIndex);
                 dictMatchIndex -= nextOffset;
                 matchIndex -= nextOffset;
-            }
-        }
-    }
+    }   }   }
 
     return longest;
 }
@@ -397,14 +404,6 @@ int LZ4HC_InsertAndFindBestMatch(LZ4HC_CCtx_internal* const hc4,   /* Index tabl
     return LZ4HC_InsertAndGetWiderMatch(hc4, ip, ip, iLimit, MINMATCH-1, matchpos, &uselessPtr, maxNbAttempts, patternAnalysis, 0 /*chainSwap*/, dict, favorCompressionRatio);
 }
 
-
-
-typedef enum {
-    noLimit = 0,
-    limitedOutput = 1,
-    limitedDestSize = 2,
-} limitedOutput_directive;
-
 /* LZ4HC_encodeSequence() :
  * @return : 0 if ok,
  *           1 if buffer issue detected */
@@ -439,7 +438,7 @@ LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
 
     /* Encode Literal length */
     length = (size_t)(*ip - *anchor);
-    if ((limit) && ((*op + (length >> 8) + length + (2 + 1 + LASTLITERALS)) > oend)) return 1;   /* Check output limit */
+    if ((limit) && ((*op + (length / 255) + length + (2 + 1 + LASTLITERALS)) > oend)) return 1;   /* Check output limit */
     if (length >= RUN_MASK) {
         size_t len = length - RUN_MASK;
         *token = (RUN_MASK << ML_BITS);
@@ -450,17 +449,17 @@ LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
     }
 
     /* Copy Literals */
-    LZ4_wildCopy(*op, *anchor, (*op) + length);
+    LZ4_wildCopy8(*op, *anchor, (*op) + length);
     *op += length;
 
     /* Encode Offset */
-    assert( (*ip - match) <= MAX_DISTANCE );   /* note : consider providing offset as a value, rather than as a pointer difference */
+    assert( (*ip - match) <= LZ4_DISTANCE_MAX );   /* note : consider providing offset as a value, rather than as a pointer difference */
     LZ4_writeLE16(*op, (U16)(*ip-match)); *op += 2;
 
     /* Encode MatchLength */
     assert(matchLength >= MINMATCH);
-    length = (size_t)(matchLength - MINMATCH);
-    if ((limit) && (*op + (length >> 8) + (1 + LASTLITERALS) > oend)) return 1;   /* Check output limit */
+    length = (size_t)matchLength - MINMATCH;
+    if ((limit) && (*op + (length / 255) + (1 + LASTLITERALS) > oend)) return 1;   /* Check output limit */
     if (length >= ML_MASK) {
         *token += ML_MASK;
         length -= ML_MASK;
@@ -513,12 +512,12 @@ LZ4_FORCE_INLINE int LZ4HC_compress_hashChain (
 
     /* init */
     *srcSizePtr = 0;
-    if (limit == limitedDestSize) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
+    if (limit == fillOutput) oend -= LASTLITERALS;                  /* Hack for support LZ4 format restriction */
     if (inputSize < LZ4_minLength) goto _last_literals;                  /* Input too small, no compression (all literals) */
 
     /* Main Loop */
     while (ip <= mflimit) {
-        ml = LZ4HC_InsertAndFindBestMatch (ctx, ip, matchlimit, &ref, maxNbAttempts, patternAnalysis, dict);
+        ml = LZ4HC_InsertAndFindBestMatch(ctx, ip, matchlimit, &ref, maxNbAttempts, patternAnalysis, dict);
         if (ml<MINMATCH) { ip++; continue; }
 
         /* saved, in case we would skip too much */
@@ -535,7 +534,7 @@ _Search2:
 
         if (ml2 == ml) { /* No better match => encode ML1 */
             optr = op;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
             continue;
         }
 
@@ -583,10 +582,10 @@ _Search3:
             if (start2 < ip+ml)  ml = (int)(start2 - ip);
             /* Now, encode 2 sequences */
             optr = op;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
             ip = start2;
             optr = op;
-            if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml2, ref2, limit, oend)) goto _dest_overflow;
+            if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml2, ref2, limit, oend)) goto _dest_overflow;
             continue;
         }
 
@@ -605,7 +604,7 @@ _Search3:
                 }
 
                 optr = op;
-                if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+                if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
                 ip  = start3;
                 ref = ref3;
                 ml  = ml3;
@@ -643,7 +642,7 @@ _Search3:
             }
         }
         optr = op;
-        if (LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ref, limit, oend)) goto _dest_overflow;
+        if (LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ref, limit, oend)) goto _dest_overflow;
 
         /* ML2 becomes ML1 */
         ip = start2; ref = ref2; ml = ml2;
@@ -660,7 +659,7 @@ _last_literals:
     {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
         size_t litLength = (lastRunSize + 255 - RUN_MASK) / 255;
         size_t const totalSize = 1 + litLength + lastRunSize;
-        if (limit == limitedDestSize) oend += LASTLITERALS;  /* restore correct value */
+        if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
         if (limit && (op + totalSize > oend)) {
             if (limit == limitedOutput) return 0;  /* Check output limit */
             /* adapt lastRunSize to fill 'dest' */
@@ -687,7 +686,7 @@ _last_literals:
     return (int) (((char*)op)-dest);
 
 _dest_overflow:
-    if (limit == limitedDestSize) {
+    if (limit == fillOutput) {
         op = optr;  /* restore correct out pointer */
         goto _last_literals;
     }
@@ -737,56 +736,64 @@ LZ4_FORCE_INLINE int LZ4HC_compress_generic_internal (
         { lz4opt,16384,LZ4_OPT_NUM },  /* 12==LZ4HC_CLEVEL_MAX */
     };
 
-    DEBUGLOG(4, "LZ4HC_compress_generic(%p, %p, %d)", ctx, src, *srcSizePtr);
+    DEBUGLOG(4, "LZ4HC_compress_generic(ctx=%p, src=%p, srcSize=%d)", ctx, src, *srcSizePtr);
 
-    if (limit == limitedDestSize && dstCapacity < 1) return 0;         /* Impossible to store anything */
-    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;          /* Unsupported input size (too large or negative) */
+    if (limit == fillOutput && dstCapacity < 1) return 0;   /* Impossible to store anything */
+    if ((U32)*srcSizePtr > (U32)LZ4_MAX_INPUT_SIZE) return 0;    /* Unsupported input size (too large or negative) */
 
     ctx->end += *srcSizePtr;
     if (cLevel < 1) cLevel = LZ4HC_CLEVEL_DEFAULT;   /* note : convention is different from lz4frame, maybe something to review */
     cLevel = MIN(LZ4HC_CLEVEL_MAX, cLevel);
     {   cParams_t const cParam = clTable[cLevel];
         HCfavor_e const favor = ctx->favorDecSpeed ? favorDecompressionSpeed : favorCompressionRatio;
-        if (cParam.strat == lz4hc)
-            return LZ4HC_compress_hashChain(ctx,
+        int result;
+
+        if (cParam.strat == lz4hc) {
+            result = LZ4HC_compress_hashChain(ctx,
                                 src, dst, srcSizePtr, dstCapacity,
                                 cParam.nbSearches, limit, dict);
-        assert(cParam.strat == lz4opt);
-        return LZ4HC_compress_optimal(ctx,
-                            src, dst, srcSizePtr, dstCapacity,
-                            cParam.nbSearches, cParam.targetLength, limit,
-                            cLevel == LZ4HC_CLEVEL_MAX,   /* ultra mode */
-                            dict, favor);
+        } else {
+            assert(cParam.strat == lz4opt);
+            result = LZ4HC_compress_optimal(ctx,
+                                src, dst, srcSizePtr, dstCapacity,
+                                (int)cParam.nbSearches, cParam.targetLength, limit,
+                                cLevel == LZ4HC_CLEVEL_MAX,   /* ultra mode */
+                                dict, favor);
+        }
+        if (result <= 0) ctx->dirty = 1;
+        return result;
     }
 }
 
 static void LZ4HC_setExternalDict(LZ4HC_CCtx_internal* ctxPtr, const BYTE* newBlock);
 
-static int LZ4HC_compress_generic_noDictCtx (
-    LZ4HC_CCtx_internal* const ctx,
-    const char* const src,
-    char* const dst,
-    int* const srcSizePtr,
-    int const dstCapacity,
-    int cLevel,
-    limitedOutput_directive limit
-    )
+static int
+LZ4HC_compress_generic_noDictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
 {
     assert(ctx->dictCtx == NULL);
     return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, noDictCtx);
 }
 
-static int LZ4HC_compress_generic_dictCtx (
-    LZ4HC_CCtx_internal* const ctx,
-    const char* const src,
-    char* const dst,
-    int* const srcSizePtr,
-    int const dstCapacity,
-    int cLevel,
-    limitedOutput_directive limit
-    )
+static int
+LZ4HC_compress_generic_dictCtx (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
 {
-    const size_t position = ctx->end - ctx->base - ctx->lowLimit;
+    const size_t position = (size_t)(ctx->end - ctx->base) - ctx->lowLimit;
     assert(ctx->dictCtx != NULL);
     if (position >= 64 KB) {
         ctx->dictCtx = NULL;
@@ -797,19 +804,20 @@ static int LZ4HC_compress_generic_dictCtx (
         ctx->compressionLevel = (short)cLevel;
         return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
     } else {
-        return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, usingDictCtx);
+        return LZ4HC_compress_generic_internal(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit, usingDictCtxHc);
     }
 }
 
-static int LZ4HC_compress_generic (
-    LZ4HC_CCtx_internal* const ctx,
-    const char* const src,
-    char* const dst,
-    int* const srcSizePtr,
-    int const dstCapacity,
-    int cLevel,
-    limitedOutput_directive limit
-    )
+static int
+LZ4HC_compress_generic (
+        LZ4HC_CCtx_internal* const ctx,
+        const char* const src,
+        char* const dst,
+        int* const srcSizePtr,
+        int const dstCapacity,
+        int cLevel,
+        limitedOutput_directive limit
+        )
 {
     if (ctx->dictCtx == NULL) {
         return LZ4HC_compress_generic_noDictCtx(ctx, src, dst, srcSizePtr, dstCapacity, cLevel, limit);
@@ -819,24 +827,41 @@ static int LZ4HC_compress_generic (
 }
 
 
-int LZ4_sizeofStateHC(void) { return sizeof(LZ4_streamHC_t); }
+int LZ4_sizeofStateHC(void) { return (int)sizeof(LZ4_streamHC_t); }
 
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                   * it reports an aligment of 8-bytes,
+                   * while actually aligning LZ4_streamHC_t on 4 bytes. */
+static size_t LZ4_streamHC_t_alignment(void)
+{
+    struct { char c; LZ4_streamHC_t t; } t_a;
+    return sizeof(t_a) - sizeof(t_a.t);
+}
+#endif
+
+/* state is presumed correctly initialized,
+ * in which case its size and alignment have already been validate */
 int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
 {
     LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                   * it reports an aligment of 8-bytes,
+                   * while actually aligning LZ4_streamHC_t on 4 bytes. */
+    assert(((size_t)state & (LZ4_streamHC_t_alignment() - 1)) == 0);  /* check alignment */
+#endif
     if (((size_t)(state)&(sizeof(void*)-1)) != 0) return 0;   /* Error : state is not aligned for pointers (32 or 64 bits) */
     LZ4_resetStreamHC_fast((LZ4_streamHC_t*)state, compressionLevel);
-    LZ4HC_init (ctx, (const BYTE*)src);
+    LZ4HC_init_internal (ctx, (const BYTE*)src);
     if (dstCapacity < LZ4_compressBound(srcSize))
         return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, limitedOutput);
     else
-        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, noLimit);
+        return LZ4HC_compress_generic (ctx, src, dst, &srcSize, dstCapacity, compressionLevel, notLimited);
 }
 
 int LZ4_compress_HC_extStateHC (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel)
 {
-    if (((size_t)(state)&(sizeof(void*)-1)) != 0) return 0;   /* Error : state is not aligned for pointers (32 or 64 bits) */
-    LZ4_resetStreamHC ((LZ4_streamHC_t*)state, compressionLevel);
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
     return LZ4_compress_HC_extStateHC_fastReset(state, src, dst, srcSize, dstCapacity, compressionLevel);
 }
 
@@ -850,19 +875,19 @@ int LZ4_compress_HC(const char* src, char* dst, int srcSize, int dstCapacity, in
 #endif
     int const cSize = LZ4_compress_HC_extStateHC(statePtr, src, dst, srcSize, dstCapacity, compressionLevel);
 #if defined(LZ4HC_HEAPMODE) && LZ4HC_HEAPMODE==1
-    free(statePtr);
+    FREEMEM(statePtr);
 #endif
     return cSize;
 }
 
-/* LZ4_compress_HC_destSize() :
- * only compatible with regular HC parser */
-int LZ4_compress_HC_destSize(void* LZ4HC_Data, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
+/* state is presumed sized correctly (>= sizeof(LZ4_streamHC_t)) */
+int LZ4_compress_HC_destSize(void* state, const char* source, char* dest, int* sourceSizePtr, int targetDestSize, int cLevel)
 {
-    LZ4HC_CCtx_internal* const ctx = &((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse;
-    LZ4_resetStreamHC((LZ4_streamHC_t*)LZ4HC_Data, cLevel);
-    LZ4HC_init(ctx, (const BYTE*) source);
-    return LZ4HC_compress_generic(ctx, source, dest, sourceSizePtr, targetDestSize, cLevel, limitedDestSize);
+    LZ4_streamHC_t* const ctx = LZ4_initStreamHC(state, sizeof(*ctx));
+    if (ctx==NULL) return 0;   /* init failure */
+    LZ4HC_init_internal(&ctx->internal_donotuse, (const BYTE*) source);
+    LZ4_setCompressionLevel(ctx, cLevel);
+    return LZ4HC_compress_generic(&ctx->internal_donotuse, source, dest, sourceSizePtr, targetDestSize, cLevel, fillOutput);
 }
 
 
@@ -871,44 +896,70 @@ int LZ4_compress_HC_destSize(void* LZ4HC_Data, const char* source, char* dest, i
 *  Streaming Functions
 **************************************/
 /* allocation */
-LZ4_streamHC_t* LZ4_createStreamHC(void) {
+LZ4_streamHC_t* LZ4_createStreamHC(void)
+{
     LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
     if (LZ4_streamHCPtr==NULL) return NULL;
-    LZ4_resetStreamHC(LZ4_streamHCPtr, LZ4HC_CLEVEL_DEFAULT);
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));  /* full initialization, malloc'ed buffer can be full of garbage */
     return LZ4_streamHCPtr;
 }
 
-int LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr) {
+int LZ4_freeStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr)
+{
     DEBUGLOG(4, "LZ4_freeStreamHC(%p)", LZ4_streamHCPtr);
     if (!LZ4_streamHCPtr) return 0;  /* support free on NULL */
-    free(LZ4_streamHCPtr);
+    FREEMEM(LZ4_streamHCPtr);
     return 0;
 }
 
 
-/* initialization */
-void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size)
 {
-    LZ4_STATIC_ASSERT(sizeof(LZ4HC_CCtx_internal) <= sizeof(size_t) * LZ4_STREAMHCSIZE_SIZET);   /* if compilation fails here, LZ4_STREAMHCSIZE must be increased */
-    DEBUGLOG(4, "LZ4_resetStreamHC(%p, %d)", LZ4_streamHCPtr, compressionLevel);
+    LZ4_streamHC_t* const LZ4_streamHCPtr = (LZ4_streamHC_t*)buffer;
+    if (buffer == NULL) return NULL;
+    if (size < sizeof(LZ4_streamHC_t)) return NULL;
+#ifndef _MSC_VER  /* for some reason, Visual fails the aligment test on 32-bit x86 :
+                   * it reports an aligment of 8-bytes,
+                   * while actually aligning LZ4_streamHC_t on 4 bytes. */
+    if (((size_t)buffer) & (LZ4_streamHC_t_alignment() - 1)) return NULL;  /* alignment check */
+#endif
+    /* if compilation fails here, LZ4_STREAMHCSIZE must be increased */
+    LZ4_STATIC_ASSERT(sizeof(LZ4HC_CCtx_internal) <= LZ4_STREAMHCSIZE);
+    DEBUGLOG(4, "LZ4_initStreamHC(%p, %u)", LZ4_streamHCPtr, (unsigned)size);
+    /* end-base will trigger a clearTable on starting compression */
     LZ4_streamHCPtr->internal_donotuse.end = (const BYTE *)(ptrdiff_t)-1;
     LZ4_streamHCPtr->internal_donotuse.base = NULL;
     LZ4_streamHCPtr->internal_donotuse.dictCtx = NULL;
     LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = 0;
+    LZ4_streamHCPtr->internal_donotuse.dirty = 0;
+    LZ4_setCompressionLevel(LZ4_streamHCPtr, LZ4HC_CLEVEL_DEFAULT);
+    return LZ4_streamHCPtr;
+}
+
+/* just a stub */
+void LZ4_resetStreamHC (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
+{
+    LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
     LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
 }
 
 void LZ4_resetStreamHC_fast (LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
 {
     DEBUGLOG(4, "LZ4_resetStreamHC_fast(%p, %d)", LZ4_streamHCPtr, compressionLevel);
-    LZ4_streamHCPtr->internal_donotuse.end -= (uptrval)LZ4_streamHCPtr->internal_donotuse.base;
-    LZ4_streamHCPtr->internal_donotuse.base = NULL;
-    LZ4_streamHCPtr->internal_donotuse.dictCtx = NULL;
+    if (LZ4_streamHCPtr->internal_donotuse.dirty) {
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+    } else {
+        /* preserve end - base : can trigger clearTable's threshold */
+        LZ4_streamHCPtr->internal_donotuse.end -= (uptrval)LZ4_streamHCPtr->internal_donotuse.base;
+        LZ4_streamHCPtr->internal_donotuse.base = NULL;
+        LZ4_streamHCPtr->internal_donotuse.dictCtx = NULL;
+    }
     LZ4_setCompressionLevel(LZ4_streamHCPtr, compressionLevel);
 }
 
 void LZ4_setCompressionLevel(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel)
 {
+    DEBUGLOG(5, "LZ4_setCompressionLevel(%p, %d)", LZ4_streamHCPtr, compressionLevel);
     if (compressionLevel < 1) compressionLevel = LZ4HC_CLEVEL_DEFAULT;
     if (compressionLevel > LZ4HC_CLEVEL_MAX) compressionLevel = LZ4HC_CLEVEL_MAX;
     LZ4_streamHCPtr->internal_donotuse.compressionLevel = (short)compressionLevel;
@@ -919,16 +970,24 @@ void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor)
     LZ4_streamHCPtr->internal_donotuse.favorDecSpeed = (favor!=0);
 }
 
-int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, const char* dictionary, int dictSize)
+/* LZ4_loadDictHC() :
+ * LZ4_streamHCPtr is presumed properly initialized */
+int LZ4_loadDictHC (LZ4_streamHC_t* LZ4_streamHCPtr,
+              const char* dictionary, int dictSize)
 {
     LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
     DEBUGLOG(4, "LZ4_loadDictHC(%p, %p, %d)", LZ4_streamHCPtr, dictionary, dictSize);
+    assert(LZ4_streamHCPtr != NULL);
     if (dictSize > 64 KB) {
-        dictionary += dictSize - 64 KB;
+        dictionary += (size_t)dictSize - 64 KB;
         dictSize = 64 KB;
     }
-    LZ4_resetStreamHC(LZ4_streamHCPtr, ctxPtr->compressionLevel);
-    LZ4HC_init (ctxPtr, (const BYTE*)dictionary);
+    /* need a full initialization, there are bad side-effects when using resetFast() */
+    {   int const cLevel = ctxPtr->compressionLevel;
+        LZ4_initStreamHC(LZ4_streamHCPtr, sizeof(*LZ4_streamHCPtr));
+        LZ4_setCompressionLevel(LZ4_streamHCPtr, cLevel);
+    }
+    LZ4HC_init_internal (ctxPtr, (const BYTE*)dictionary);
     ctxPtr->end = (const BYTE*)dictionary + dictSize;
     if (dictSize >= 4) LZ4HC_Insert (ctxPtr, ctxPtr->end-3);
     return dictSize;
@@ -961,9 +1020,11 @@ static int LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
                                             limitedOutput_directive limit)
 {
     LZ4HC_CCtx_internal* const ctxPtr = &LZ4_streamHCPtr->internal_donotuse;
-    DEBUGLOG(4, "LZ4_compressHC_continue_generic(%p, %p, %d)", LZ4_streamHCPtr, src, *srcSizePtr);
+    DEBUGLOG(4, "LZ4_compressHC_continue_generic(ctx=%p, src=%p, srcSize=%d)",
+                LZ4_streamHCPtr, src, *srcSizePtr);
+    assert(ctxPtr != NULL);
     /* auto-init if forgotten */
-    if (ctxPtr->base == NULL) LZ4HC_init (ctxPtr, (const BYTE*) src);
+    if (ctxPtr->base == NULL) LZ4HC_init_internal (ctxPtr, (const BYTE*) src);
 
     /* Check overflow */
     if ((size_t)(ctxPtr->end - ctxPtr->base) > 2 GB) {
@@ -973,7 +1034,8 @@ static int LZ4_compressHC_continue_generic (LZ4_streamHC_t* LZ4_streamHCPtr,
     }
 
     /* Check if blocks follow each other */
-    if ((const BYTE*)src != ctxPtr->end) LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
+    if ((const BYTE*)src != ctxPtr->end)
+        LZ4HC_setExternalDict(ctxPtr, (const BYTE*)src);
 
     /* Check overlapping input/dictionary space */
     {   const BYTE* sourceEnd = (const BYTE*) src + *srcSizePtr;
@@ -994,12 +1056,12 @@ int LZ4_compress_HC_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src,
     if (dstCapacity < LZ4_compressBound(srcSize))
         return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, limitedOutput);
     else
-        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, noLimit);
+        return LZ4_compressHC_continue_generic (LZ4_streamHCPtr, src, dst, &srcSize, dstCapacity, notLimited);
 }
 
 int LZ4_compress_HC_continue_destSize (LZ4_streamHC_t* LZ4_streamHCPtr, const char* src, char* dst, int* srcSizePtr, int targetDestSize)
 {
-    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, limitedDestSize);
+    return LZ4_compressHC_continue_generic(LZ4_streamHCPtr, src, dst, srcSizePtr, targetDestSize, fillOutput);
 }
 
 
@@ -1018,19 +1080,21 @@ int LZ4_saveDictHC (LZ4_streamHC_t* LZ4_streamHCPtr, char* safeBuffer, int dictS
     {   U32 const endIndex = (U32)(streamPtr->end - streamPtr->base);
         streamPtr->end = (const BYTE*)safeBuffer + dictSize;
         streamPtr->base = streamPtr->end - endIndex;
-        streamPtr->dictLimit = endIndex - dictSize;
-        streamPtr->lowLimit = endIndex - dictSize;
+        streamPtr->dictLimit = endIndex - (U32)dictSize;
+        streamPtr->lowLimit = endIndex - (U32)dictSize;
         if (streamPtr->nextToUpdate < streamPtr->dictLimit) streamPtr->nextToUpdate = streamPtr->dictLimit;
     }
     return dictSize;
 }
 
 
-/***********************************
+/***************************************************
 *  Deprecated Functions
-***********************************/
+***************************************************/
+
 /* These functions currently generate deprecation warnings */
-/* Deprecated compression functions */
+
+/* Wrappers for deprecated compression functions */
 int LZ4_compressHC(const char* src, char* dst, int srcSize) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), 0); }
 int LZ4_compressHC_limitedOutput(const char* src, char* dst, int srcSize, int maxDstSize) { return LZ4_compress_HC(src, dst, srcSize, maxDstSize, 0); }
 int LZ4_compressHC2(const char* src, char* dst, int srcSize, int cLevel) { return LZ4_compress_HC (src, dst, srcSize, LZ4_compressBound(srcSize), cLevel); }
@@ -1046,25 +1110,26 @@ int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* ctx, const char* src,
 /* Deprecated streaming functions */
 int LZ4_sizeofStreamStateHC(void) { return LZ4_STREAMHCSIZE; }
 
+/* state is presumed correctly sized, aka >= sizeof(LZ4_streamHC_t)
+ * @return : 0 on success, !=0 if error */
 int LZ4_resetStreamStateHC(void* state, char* inputBuffer)
 {
-    LZ4HC_CCtx_internal *ctx = &((LZ4_streamHC_t*)state)->internal_donotuse;
-    if ((((size_t)state) & (sizeof(void*)-1)) != 0) return 1;   /* Error : pointer is not aligned for pointer (32 or 64 bits) */
-    LZ4_resetStreamHC((LZ4_streamHC_t*)state, ((LZ4_streamHC_t*)state)->internal_donotuse.compressionLevel);
-    LZ4HC_init(ctx, (const BYTE*)inputBuffer);
+    LZ4_streamHC_t* const hc4 = LZ4_initStreamHC(state, sizeof(*hc4));
+    if (hc4 == NULL) return 1;   /* init failed */
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
     return 0;
 }
 
 void* LZ4_createHC (const char* inputBuffer)
 {
-    LZ4_streamHC_t* hc4 = (LZ4_streamHC_t*)ALLOC(sizeof(LZ4_streamHC_t));
+    LZ4_streamHC_t* const hc4 = LZ4_createStreamHC();
     if (hc4 == NULL) return NULL;   /* not enough memory */
-    LZ4_resetStreamHC(hc4, 0 /* compressionLevel */);
-    LZ4HC_init (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
+    LZ4HC_init_internal (&hc4->internal_donotuse, (const BYTE*)inputBuffer);
     return hc4;
 }
 
-int LZ4_freeHC (void* LZ4HC_Data) {
+int LZ4_freeHC (void* LZ4HC_Data)
+{
     if (!LZ4HC_Data) return 0;  /* support free on NULL */
     FREEMEM(LZ4HC_Data);
     return 0;
@@ -1072,7 +1137,7 @@ int LZ4_freeHC (void* LZ4HC_Data) {
 
 int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int cLevel)
 {
-    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, noLimit);
+    return LZ4HC_compress_generic (&((LZ4_streamHC_t*)LZ4HC_Data)->internal_donotuse, src, dst, &srcSize, 0, cLevel, notLimited);
 }
 
 int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* src, char* dst, int srcSize, int dstCapacity, int cLevel)
@@ -1091,7 +1156,7 @@ char* LZ4_slideInputBufferHC(void* LZ4HC_Data)
 
 
 /* ================================================
- * LZ4 Optimal parser (levels 10-12)
+ *  LZ4 Optimal parser (levels [LZ4HC_CLEVEL_OPT_MIN - LZ4HC_CLEVEL_MAX])
  * ===============================================*/
 typedef struct {
     int price;
@@ -1104,8 +1169,9 @@ typedef struct {
 LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen)
 {
     int price = litlen;
+    assert(litlen >= 0);
     if (litlen >= (int)RUN_MASK)
-        price += 1 + (litlen-RUN_MASK)/255;
+        price += 1 + ((litlen-(int)RUN_MASK) / 255);
     return price;
 }
 
@@ -1114,11 +1180,13 @@ LZ4_FORCE_INLINE int LZ4HC_literalsPrice(int const litlen)
 LZ4_FORCE_INLINE int LZ4HC_sequencePrice(int litlen, int mlen)
 {
     int price = 1 + 2 ; /* token + 16-bit offset */
+    assert(litlen >= 0);
+    assert(mlen >= MINMATCH);
 
     price += LZ4HC_literalsPrice(litlen);
 
     if (mlen >= (int)(ML_MASK+MINMATCH))
-        price += 1 + (mlen-(ML_MASK+MINMATCH))/255;
+        price += 1 + ((mlen-(int)(ML_MASK+MINMATCH)) / 255);
 
     return price;
 }
@@ -1177,9 +1245,9 @@ static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
     BYTE* oend = op + dstCapacity;
 
     /* init */
-    DEBUGLOG(5, "LZ4HC_compress_optimal");
+    DEBUGLOG(5, "LZ4HC_compress_optimal(dst=%p, dstCapa=%u)", dst, (unsigned)dstCapacity);
     *srcSizePtr = 0;
-    if (limit == limitedDestSize) oend -= LASTLITERALS;   /* Hack for support LZ4 format restriction */
+    if (limit == fillOutput) oend -= LASTLITERALS;   /* Hack for support LZ4 format restriction */
     if (sufficient_len >= LZ4_OPT_NUM) sufficient_len = LZ4_OPT_NUM-1;
 
     /* Main Loop */
@@ -1197,7 +1265,7 @@ static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
              int const firstML = firstMatch.len;
              const BYTE* const matchPos = ip - firstMatch.off;
              opSaved = op;
-             if ( LZ4HC_encodeSequence(&ip, &op, &anchor, firstML, matchPos, limit, oend) )   /* updates ip, op and anchor */
+             if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), firstML, matchPos, limit, oend) )   /* updates ip, op and anchor */
                  goto _dest_overflow;
              continue;
          }
@@ -1335,6 +1403,7 @@ static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
              }   }
          }  /* for (cur = 1; cur <= last_match_pos; cur++) */
 
+         assert(last_match_pos < LZ4_OPT_NUM + TRAILING_LITERALS);
          best_mlen = opt[last_match_pos].mlen;
          best_off = opt[last_match_pos].off;
          cur = last_match_pos - best_mlen;
@@ -1367,9 +1436,9 @@ static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
                  if (ml == 1) { ip++; rPos++; continue; }  /* literal; note: can end up with several literals, in which case, skip them */
                  rPos += ml;
                  assert(ml >= MINMATCH);
-                 assert((offset >= 1) && (offset <= MAX_DISTANCE));
+                 assert((offset >= 1) && (offset <= LZ4_DISTANCE_MAX));
                  opSaved = op;
-                 if ( LZ4HC_encodeSequence(&ip, &op, &anchor, ml, ip - offset, limit, oend) )   /* updates ip, op and anchor */
+                 if ( LZ4HC_encodeSequence(UPDATABLE(ip, op, anchor), ml, ip - offset, limit, oend) )   /* updates ip, op and anchor */
                      goto _dest_overflow;
          }   }
      }  /* while (ip <= mflimit) */
@@ -1379,7 +1448,7 @@ static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
      {   size_t lastRunSize = (size_t)(iend - anchor);  /* literals */
          size_t litLength = (lastRunSize + 255 - RUN_MASK) / 255;
          size_t const totalSize = 1 + litLength + lastRunSize;
-         if (limit == limitedDestSize) oend += LASTLITERALS;  /* restore correct value */
+         if (limit == fillOutput) oend += LASTLITERALS;  /* restore correct value */
          if (limit && (op + totalSize > oend)) {
              if (limit == limitedOutput) return 0;  /* Check output limit */
              /* adapt lastRunSize to fill 'dst' */
@@ -1406,7 +1475,7 @@ static int LZ4HC_compress_optimal ( LZ4HC_CCtx_internal* ctx,
      return (int) ((char*)op-dst);
 
  _dest_overflow:
-     if (limit == limitedDestSize) {
+     if (limit == fillOutput) {
          op = opSaved;  /* restore correct out pointer */
          goto _last_literals;
      }

common/tracy_lz4hc.hpp

@@ -31,8 +31,8 @@
    - LZ4 source repository : https://github.com/lz4/lz4
    - LZ4 public forum : https://groups.google.com/forum/#!forum/lz4c
 */
-#ifndef LZ4_HC_H_19834876238432
-#define LZ4_HC_H_19834876238432
+#ifndef TRACY_LZ4_HC_H_19834876238432
+#define TRACY_LZ4_HC_H_19834876238432
 
 /* --- Dependency --- */
 /* note : lz4hc requires lz4.h/lz4.c for compilation */
@@ -52,7 +52,7 @@ namespace tracy
  *  Block Compression
  **************************************/
 /*! LZ4_compress_HC() :
- *  Compress data from `src` into `dst`, using the more powerful but slower "HC" algorithm.
+ *  Compress data from `src` into `dst`, using the powerful but slower "HC" algorithm.
  * `dst` must be already allocated.
  *  Compression is guaranteed to succeed if `dstCapacity >= LZ4_compressBound(srcSize)` (see "lz4.h")
  *  Max supported `srcSize` value is LZ4_MAX_INPUT_SIZE (see "lz4.h")
@@ -75,7 +75,21 @@ LZ4LIB_API int LZ4_compress_HC (const char* src, char* dst, int srcSize, int dst
  *  Memory segment must be aligned on 8-bytes boundaries (which a normal malloc() should do properly).
  */
 LZ4LIB_API int LZ4_sizeofStateHC(void);
-LZ4LIB_API int LZ4_compress_HC_extStateHC(void* state, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
+LZ4LIB_API int LZ4_compress_HC_extStateHC(void* stateHC, const char* src, char* dst, int srcSize, int maxDstSize, int compressionLevel);
+
+
+/*! LZ4_compress_HC_destSize() : v1.9.0+
+ *  Will compress as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided in 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr is updated to indicate how much bytes were read from `src`
+ */
+LZ4LIB_API int LZ4_compress_HC_destSize(void* stateHC,
+                                  const char* src, char* dst,
+                                        int* srcSizePtr, int targetDstSize,
+                                        int compressionLevel);
 
 
 /*-************************************
@@ -87,46 +101,92 @@ LZ4LIB_API int LZ4_compress_HC_extStateHC(void* state, const char* src, char* ds
 /*! LZ4_createStreamHC() and LZ4_freeStreamHC() :
  *  These functions create and release memory for LZ4 HC streaming state.
  *  Newly created states are automatically initialized.
- *  Existing states can be re-used several times, using LZ4_resetStreamHC().
- *  These methods are API and ABI stable, they can be used in combination with a DLL.
+ *  A same state can be used multiple times consecutively,
+ *  starting with LZ4_resetStreamHC_fast() to start a new stream of blocks.
  */
 LZ4LIB_API LZ4_streamHC_t* LZ4_createStreamHC(void);
 LZ4LIB_API int             LZ4_freeStreamHC (LZ4_streamHC_t* streamHCPtr);
 
-LZ4LIB_API void LZ4_resetStreamHC (LZ4_streamHC_t* streamHCPtr, int compressionLevel);
+/*
+  These functions compress data in successive blocks of any size,
+  using previous blocks as dictionary, to improve compression ratio.
+  One key assumption is that previous blocks (up to 64 KB) remain read-accessible while compressing next blocks.
+  There is an exception for ring buffers, which can be smaller than 64 KB.
+  Ring-buffer scenario is automatically detected and handled within LZ4_compress_HC_continue().
+
+  Before starting compression, state must be allocated and properly initialized.
+  LZ4_createStreamHC() does both, though compression level is set to LZ4HC_CLEVEL_DEFAULT.
+
+  Selecting the compression level can be done with LZ4_resetStreamHC_fast() (starts a new stream)
+  or LZ4_setCompressionLevel() (anytime, between blocks in the same stream) (experimental).
+  LZ4_resetStreamHC_fast() only works on states which have been properly initialized at least once,
+  which is automatically the case when state is created using LZ4_createStreamHC().
+
+  After reset, a first "fictional block" can be designated as initial dictionary,
+  using LZ4_loadDictHC() (Optional).
+
+  Invoke LZ4_compress_HC_continue() to compress each successive block.
+  The number of blocks is unlimited.
+  Previous input blocks, including initial dictionary when present,
+  must remain accessible and unmodified during compression.
+
+  It's allowed to update compression level anytime between blocks,
+  using LZ4_setCompressionLevel() (experimental).
+
+  'dst' buffer should be sized to handle worst case scenarios
+  (see LZ4_compressBound(), it ensures compression success).
+  In case of failure, the API does not guarantee recovery,
+  so the state _must_ be reset.
+  To ensure compression success
+  whenever `dst` buffer size cannot be made >= LZ4_compressBound(),
+  consider using LZ4_compress_HC_continue_destSize().
+
+  Whenever previous input blocks can't be preserved unmodified in-place during compression of next blocks,
+  it's possible to copy the last blocks into a more stable memory space, using LZ4_saveDictHC().
+  Return value of LZ4_saveDictHC() is the size of dictionary effectively saved into 'safeBuffer' (<= 64 KB)
+
+  After completing a streaming compression,
+  it's possible to start a new stream of blocks, using the same LZ4_streamHC_t state,
+  just by resetting it, using LZ4_resetStreamHC_fast().
+*/
+
+LZ4LIB_API void LZ4_resetStreamHC_fast(LZ4_streamHC_t* streamHCPtr, int compressionLevel);   /* v1.9.0+ */
 LZ4LIB_API int  LZ4_loadDictHC (LZ4_streamHC_t* streamHCPtr, const char* dictionary, int dictSize);
 
-LZ4LIB_API int LZ4_compress_HC_continue (LZ4_streamHC_t* streamHCPtr, const char* src, char* dst, int srcSize, int maxDstSize);
+LZ4LIB_API int LZ4_compress_HC_continue (LZ4_streamHC_t* streamHCPtr,
+                                   const char* src, char* dst,
+                                         int srcSize, int maxDstSize);
+
+/*! LZ4_compress_HC_continue_destSize() : v1.9.0+
+ *  Similar to LZ4_compress_HC_continue(),
+ *  but will read as much data as possible from `src`
+ *  to fit into `targetDstSize` budget.
+ *  Result is provided into 2 parts :
+ * @return : the number of bytes written into 'dst' (necessarily <= targetDstSize)
+ *           or 0 if compression fails.
+ * `srcSizePtr` : on success, *srcSizePtr will be updated to indicate how much bytes were read from `src`.
+ *           Note that this function may not consume the entire input.
+ */
+LZ4LIB_API int LZ4_compress_HC_continue_destSize(LZ4_streamHC_t* LZ4_streamHCPtr,
+                                           const char* src, char* dst,
+                                                 int* srcSizePtr, int targetDstSize);
 
 LZ4LIB_API int LZ4_saveDictHC (LZ4_streamHC_t* streamHCPtr, char* safeBuffer, int maxDictSize);
 
-/*
-  These functions compress data in successive blocks of any size, using previous blocks as dictionary.
-  One key assumption is that previous blocks (up to 64 KB) remain read-accessible while compressing next blocks.
-  There is an exception for ring buffers, which can be smaller than 64 KB.
-  Ring buffers scenario is automatically detected and handled by LZ4_compress_HC_continue().
-
-  Before starting compression, state must be properly initialized, using LZ4_resetStreamHC().
-  A first "fictional block" can then be designated as initial dictionary, using LZ4_loadDictHC() (Optional).
-
-  Then, use LZ4_compress_HC_continue() to compress each successive block.
-  Previous memory blocks (including initial dictionary when present) must remain accessible and unmodified during compression.
-  'dst' buffer should be sized to handle worst case scenarios (see LZ4_compressBound()), to ensure operation success.
-  Because in case of failure, the API does not guarantee context recovery, and context will have to be reset.
-  If `dst` buffer budget cannot be >= LZ4_compressBound(), consider using LZ4_compress_HC_continue_destSize() instead.
-
-  If, for any reason, previous data block can't be preserved unmodified in memory for next compression block,
-  you can save it to a more stable memory space, using LZ4_saveDictHC().
-  Return value of LZ4_saveDictHC() is the size of dictionary effectively saved into 'safeBuffer'.
-*/
 
 
-/*-**************************************************************
+/*^**********************************************
+ * !!!!!!   STATIC LINKING ONLY   !!!!!!
+ ***********************************************/
+
+/*-******************************************************************
  * PRIVATE DEFINITIONS :
- * Do not use these definitions.
- * They are exposed to allow static allocation of `LZ4_streamHC_t`.
- * Using these definitions makes the code vulnerable to potential API break when upgrading LZ4
- ****************************************************************/
+ * Do not use these definitions directly.
+ * They are merely exposed to allow static allocation of `LZ4_streamHC_t`.
+ * Declare an `LZ4_streamHC_t` directly, rather than any type below.
+ * Even then, only do so in the context of static linking, as definitions may change between versions.
+ ********************************************************************/
+
 #define LZ4HC_DICTIONARY_LOGSIZE 16
 #define LZ4HC_MAXD (1<<LZ4HC_DICTIONARY_LOGSIZE)
 #define LZ4HC_MAXD_MASK (LZ4HC_MAXD - 1)
@@ -151,7 +211,9 @@ struct LZ4HC_CCtx_internal
     uint32_t   lowLimit;        /* below that point, no more dict */
     uint32_t   nextToUpdate;    /* index from which to continue dictionary update */
     short      compressionLevel;
-    short      favorDecSpeed;
+    int8_t     favorDecSpeed;   /* favor decompression speed if this flag set,
+                                   otherwise, favor compression ratio */
+    int8_t     dirty;           /* stream has to be fully reset if this flag is set */
     const LZ4HC_CCtx_internal* dictCtx;
 };
 
@@ -169,26 +231,43 @@ struct LZ4HC_CCtx_internal
     unsigned int   lowLimit;         /* below that point, no more dict */
     unsigned int   nextToUpdate;     /* index from which to continue dictionary update */
     short          compressionLevel;
-    short          favorDecSpeed;
+    char           favorDecSpeed;    /* favor decompression speed if this flag set,
+                                        otherwise, favor compression ratio */
+    char           dirty;            /* stream has to be fully reset if this flag is set */
     const LZ4HC_CCtx_internal* dictCtx;
 };
 
 #endif
 
-#define LZ4_STREAMHCSIZE       (4*LZ4HC_HASHTABLESIZE + 2*LZ4HC_MAXD + 56) /* 262200 */
+
+/* Do not use these definitions directly !
+ * Declare or allocate an LZ4_streamHC_t instead.
+ */
+#define LZ4_STREAMHCSIZE       (4*LZ4HC_HASHTABLESIZE + 2*LZ4HC_MAXD + 56 + ((sizeof(void*)==16) ? 56 : 0) /* AS400*/ ) /* 262200 or 262256*/
 #define LZ4_STREAMHCSIZE_SIZET (LZ4_STREAMHCSIZE / sizeof(size_t))
 union LZ4_streamHC_u {
     size_t table[LZ4_STREAMHCSIZE_SIZET];
     LZ4HC_CCtx_internal internal_donotuse;
-};   /* previously typedef'd to LZ4_streamHC_t */
-/*
-  LZ4_streamHC_t :
-  This structure allows static allocation of LZ4 HC streaming state.
-  State must be initialized using LZ4_resetStreamHC() before first use.
+}; /* previously typedef'd to LZ4_streamHC_t */
 
-  Static allocation shall only be used in combination with static linking.
-  When invoking LZ4 from a DLL, use create/free functions instead, which are API and ABI stable.
-*/
+/* LZ4_streamHC_t :
+ * This structure allows static allocation of LZ4 HC streaming state.
+ * This can be used to allocate statically, on state, or as part of a larger structure.
+ *
+ * Such state **must** be initialized using LZ4_initStreamHC() before first use.
+ *
+ * Note that invoking LZ4_initStreamHC() is not required when
+ * the state was created using LZ4_createStreamHC() (which is recommended).
+ * Using the normal builder, a newly created state is automatically initialized.
+ *
+ * Static allocation shall only be used in combination with static linking.
+ */
+
+/* LZ4_initStreamHC() : v1.9.0+
+ * Required before first use of a statically allocated LZ4_streamHC_t.
+ * Before v1.9.0 : use LZ4_resetStreamHC() instead
+ */
+LZ4LIB_API LZ4_streamHC_t* LZ4_initStreamHC (void* buffer, size_t size);
 
 
 /*-************************************
@@ -199,11 +278,11 @@ union LZ4_streamHC_u {
 /* deprecated compression functions */
 LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC               (const char* source, char* dest, int inputSize);
 LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput (const char* source, char* dest, int inputSize, int maxOutputSize);
-LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2 (const char* source, char* dest, int inputSize, int compressionLevel);
-LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput (const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2              (const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput(const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
 LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_withStateHC               (void* state, const char* source, char* dest, int inputSize);
 LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_withStateHC (void* state, const char* source, char* dest, int inputSize, int maxOutputSize);
-LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_withStateHC (void* state, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_withStateHC              (void* state, const char* source, char* dest, int inputSize, int compressionLevel);
 LZ4_DEPRECATED("use LZ4_compress_HC_extStateHC() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_withStateHC(void* state, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
 LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_continue               (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize);
 LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC_limitedOutput_continue (LZ4_streamHC_t* LZ4_streamHCPtr, const char* source, char* dest, int inputSize, int maxOutputSize);
@@ -219,10 +298,22 @@ LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_comp
 LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API void* LZ4_createHC (const char* inputBuffer);
 LZ4_DEPRECATED("use LZ4_saveDictHC() instead") LZ4LIB_API     char* LZ4_slideInputBufferHC (void* LZ4HC_Data);
 LZ4_DEPRECATED("use LZ4_freeStreamHC() instead") LZ4LIB_API   int   LZ4_freeHC (void* LZ4HC_Data);
-LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel);
+LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_continue               (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int compressionLevel);
 LZ4_DEPRECATED("use LZ4_compress_HC_continue() instead") LZ4LIB_API int LZ4_compressHC2_limitedOutput_continue (void* LZ4HC_Data, const char* source, char* dest, int inputSize, int maxOutputSize, int compressionLevel);
 LZ4_DEPRECATED("use LZ4_createStreamHC() instead") LZ4LIB_API int   LZ4_sizeofStreamStateHC(void);
-LZ4_DEPRECATED("use LZ4_resetStreamHC() instead") LZ4LIB_API  int   LZ4_resetStreamStateHC(void* state, char* inputBuffer);
+LZ4_DEPRECATED("use LZ4_initStreamHC() instead") LZ4LIB_API  int   LZ4_resetStreamStateHC(void* state, char* inputBuffer);
+
+
+/* LZ4_resetStreamHC() is now replaced by LZ4_initStreamHC().
+ * The intention is to emphasize the difference with LZ4_resetStreamHC_fast(),
+ * which is now the recommended function to start a new stream of blocks,
+ * but cannot be used to initialize a memory segment containing arbitrary garbage data.
+ *
+ * It is recommended to switch to LZ4_initStreamHC().
+ * LZ4_resetStreamHC() will generate deprecation warnings in a future version.
+ */
+LZ4LIB_API void LZ4_resetStreamHC (LZ4_streamHC_t* streamHCPtr, int compressionLevel);
+
 
 }
 
@@ -244,44 +335,22 @@ LZ4_DEPRECATED("use LZ4_resetStreamHC() instead") LZ4LIB_API  int   LZ4_resetStr
 namespace tracy
 {
 
-/*! LZ4_compress_HC_destSize() : v1.8.0 (experimental)
- *  Will try to compress as much data from `src` as possible
- *  that can fit into `targetDstSize` budget.
- *  Result is provided in 2 parts :
- * @return : the number of bytes written into 'dst'
- *           or 0 if compression fails.
- * `srcSizePtr` : value will be updated to indicate how much bytes were read from `src`
+/*! LZ4_setCompressionLevel() : v1.8.0+ (experimental)
+ *  It's possible to change compression level
+ *  between successive invocations of LZ4_compress_HC_continue*()
+ *  for dynamic adaptation.
  */
-int LZ4_compress_HC_destSize(void* LZ4HC_Data,
-                             const char* src, char* dst,
-                             int* srcSizePtr, int targetDstSize,
-                             int compressionLevel);
+LZ4LIB_STATIC_API void LZ4_setCompressionLevel(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
 
-/*! LZ4_compress_HC_continue_destSize() : v1.8.0 (experimental)
- *  Similar as LZ4_compress_HC_continue(),
- *  but will read a variable nb of bytes from `src`
- *  to fit into `targetDstSize` budget.
- *  Result is provided in 2 parts :
- * @return : the number of bytes written into 'dst'
- *           or 0 if compression fails.
- * `srcSizePtr` : value will be updated to indicate how much bytes were read from `src`.
+/*! LZ4_favorDecompressionSpeed() : v1.8.2+ (experimental)
+ *  Opt. Parser will favor decompression speed over compression ratio.
+ *  Only applicable to levels >= LZ4HC_CLEVEL_OPT_MIN.
  */
-int LZ4_compress_HC_continue_destSize(LZ4_streamHC_t* LZ4_streamHCPtr,
-                            const char* src, char* dst,
-                            int* srcSizePtr, int targetDstSize);
+LZ4LIB_STATIC_API void LZ4_favorDecompressionSpeed(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int favor);
 
-/*! LZ4_setCompressionLevel() : v1.8.0 (experimental)
- *  It's possible to change compression level between 2 invocations of LZ4_compress_HC_continue*()
- */
-void LZ4_setCompressionLevel(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
-
-/*! LZ4_favorDecompressionSpeed() : v1.8.2 (experimental)
- *  Parser will select decisions favoring decompression over compression ratio.
- *  Only work at highest compression settings (level >= LZ4HC_CLEVEL_OPT_MIN)
- */
-void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor);
-
-/*! LZ4_resetStreamHC_fast() :
+/*! LZ4_resetStreamHC_fast() : v1.9.0+
  *  When an LZ4_streamHC_t is known to be in a internally coherent state,
  *  it can often be prepared for a new compression with almost no work, only
  *  sometimes falling back to the full, expensive reset that is always required
@@ -298,8 +367,14 @@ void LZ4_favorDecompressionSpeed(LZ4_streamHC_t* LZ4_streamHCPtr, int favor);
  *  - the stream was in an indeterminate state and was used in a compression
  *    call that fully reset the state (LZ4_compress_HC_extStateHC()) and that
  *    returned success
+ *
+ *  Note:
+ *  A stream that was last used in a compression call that returned an error
+ *  may be passed to this function. However, it will be fully reset, which will
+ *  clear any existing history and settings from the context.
  */
-void LZ4_resetStreamHC_fast(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
+LZ4LIB_STATIC_API void LZ4_resetStreamHC_fast(
+    LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLevel);
 
 /*! LZ4_compress_HC_extStateHC_fastReset() :
  *  A variant of LZ4_compress_HC_extStateHC().
@@ -312,7 +387,11 @@ void LZ4_resetStreamHC_fast(LZ4_streamHC_t* LZ4_streamHCPtr, int compressionLeve
  *  LZ4_resetStreamHC_fast() while LZ4_compress_HC_extStateHC() starts with a
  *  call to LZ4_resetStreamHC().
  */
-int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* dst, int srcSize, int dstCapacity, int compressionLevel);
+LZ4LIB_STATIC_API int LZ4_compress_HC_extStateHC_fastReset (
+    void* state,
+    const char* src, char* dst,
+    int srcSize, int dstCapacity,
+    int compressionLevel);
 
 /*! LZ4_attach_HC_dictionary() :
  *  This is an experimental API that allows for the efficient use of a
@@ -339,7 +418,9 @@ int LZ4_compress_HC_extStateHC_fastReset (void* state, const char* src, char* ds
  *  stream (and source buffer) must remain in-place / accessible / unchanged
  *  through the lifetime of the stream session.
  */
-LZ4LIB_API void LZ4_attach_HC_dictionary(LZ4_streamHC_t *working_stream, const LZ4_streamHC_t *dictionary_stream);
+LZ4LIB_STATIC_API void LZ4_attach_HC_dictionary(
+          LZ4_streamHC_t *working_stream,
+    const LZ4_streamHC_t *dictionary_stream);
 
 }
 

common/tracy_sema.h

@@ -1,255 +0,0 @@
-// Copyright (c) 2015 Jeff Preshing
-//
-// This software is provided 'as-is', without any express or implied
-// warranty. In no event will the authors be held liable for any damages
-// arising from the use of this software.
-//
-// Permission is granted to anyone to use this software for any purpose,
-// including commercial applications, and to alter it and redistribute it
-// freely, subject to the following restrictions:
-//
-// 1. The origin of this software must not be misrepresented; you must not
-// claim that you wrote the original software. If you use this software
-// in a product, an acknowledgement in the product documentation would be
-// appreciated but is not required.
-// 2. Altered source versions must be plainly marked as such, and must not be
-// misrepresented as being the original software.
-// 3. This notice may not be removed or altered from any source distribution.
-
-#ifndef __TRACY_CPP11OM_SEMAPHORE_H__
-#define __TRACY_CPP11OM_SEMAPHORE_H__
-
-#include <atomic>
-#include <cassert>
-
-#if defined(__MACH__)
-    #include <mach/mach.h>
-#elif defined(__unix__)
-    #include <semaphore.h>
-#endif
-
-namespace tracy
-{
-
-#if defined(_WIN32)
-//---------------------------------------------------------
-// Semaphore (Windows)
-//---------------------------------------------------------
-#ifndef MAXLONG
-enum { MAXLONG = 0x7fffffff };
-#endif
-
-#ifndef INFINITE
-enum { INFINITE = 0xFFFFFFFF };
-#endif
-
-#ifndef _WINDOWS_
-typedef void* HANDLE;
-
-extern "C" __declspec(dllimport) HANDLE __stdcall CreateSemaphoreA( void*, long, long, const char* );
-extern "C" __declspec(dllimport) int __stdcall CloseHandle( HANDLE );
-extern "C" __declspec(dllimport) unsigned long __stdcall WaitForSingleObject( HANDLE, unsigned long );
-extern "C" __declspec(dllimport) int __stdcall ReleaseSemaphore( HANDLE, long, long* );
-#endif
-
-class Semaphore
-{
-private:
-    HANDLE m_hSema;
-
-    Semaphore(const Semaphore& other) = delete;
-    Semaphore& operator=(const Semaphore& other) = delete;
-
-public:
-    Semaphore(int initialCount = 0)
-    {
-        assert(initialCount >= 0);
-        m_hSema = CreateSemaphoreA(NULL, initialCount, MAXLONG, NULL);
-    }
-
-    ~Semaphore()
-    {
-        CloseHandle(m_hSema);
-    }
-
-    void wait()
-    {
-        WaitForSingleObject(m_hSema, INFINITE);
-    }
-
-    void signal(int count = 1)
-    {
-        ReleaseSemaphore(m_hSema, count, NULL);
-    }
-};
-
-
-#elif defined(__MACH__)
-//---------------------------------------------------------
-// Semaphore (Apple iOS and OSX)
-// Can't use POSIX semaphores due to http://lists.apple.com/archives/darwin-kernel/2009/Apr/msg00010.html
-//---------------------------------------------------------
-
-class Semaphore
-{
-private:
-    semaphore_t m_sema;
-
-    Semaphore(const Semaphore& other) = delete;
-    Semaphore& operator=(const Semaphore& other) = delete;
-
-public:
-    Semaphore(int initialCount = 0)
-    {
-        assert(initialCount >= 0);
-        semaphore_create(mach_task_self(), &m_sema, SYNC_POLICY_FIFO, initialCount);
-    }
-
-    ~Semaphore()
-    {
-        semaphore_destroy(mach_task_self(), m_sema);
-    }
-
-    void wait()
-    {
-        semaphore_wait(m_sema);
-    }
-
-    void signal()
-    {
-        semaphore_signal(m_sema);
-    }
-
-    void signal(int count)
-    {
-        while (count-- > 0)
-        {
-            semaphore_signal(m_sema);
-        }
-    }
-};
-
-
-#elif defined(__unix__)
-//---------------------------------------------------------
-// Semaphore (POSIX, Linux)
-//---------------------------------------------------------
-
-class Semaphore
-{
-private:
-    sem_t m_sema;
-
-    Semaphore(const Semaphore& other) = delete;
-    Semaphore& operator=(const Semaphore& other) = delete;
-
-public:
-    Semaphore(int initialCount = 0)
-    {
-        assert(initialCount >= 0);
-        sem_init(&m_sema, 0, initialCount);
-    }
-
-    ~Semaphore()
-    {
-        sem_destroy(&m_sema);
-    }
-
-    void wait()
-    {
-        // http://stackoverflow.com/questions/2013181/gdb-causes-sem-wait-to-fail-with-eintr-error
-        int rc;
-        do
-        {
-            rc = sem_wait(&m_sema);
-        }
-        while (rc == -1 && errno == EINTR);
-    }
-
-    void signal()
-    {
-        sem_post(&m_sema);
-    }
-
-    void signal(int count)
-    {
-        while (count-- > 0)
-        {
-            sem_post(&m_sema);
-        }
-    }
-};
-
-
-#else
-
-#error Unsupported platform!
-
-#endif
-
-
-//---------------------------------------------------------
-// LightweightSemaphore
-//---------------------------------------------------------
-class LightweightSemaphore
-{
-private:
-    std::atomic<int> m_count;
-    Semaphore m_sema;
-
-    void waitWithPartialSpinning()
-    {
-        int oldCount;
-        // Is there a better way to set the initial spin count?
-        // If we lower it to 1000, testBenaphore becomes 15x slower on my Core i7-5930K Windows PC,
-        // as threads start hitting the kernel semaphore.
-        int spin = 10000;
-        while (spin--)
-        {
-            oldCount = m_count.load(std::memory_order_relaxed);
-            if ((oldCount > 0) && m_count.compare_exchange_strong(oldCount, oldCount - 1, std::memory_order_acquire))
-                return;
-            std::atomic_signal_fence(std::memory_order_acquire);     // Prevent the compiler from collapsing the loop.
-        }
-        oldCount = m_count.fetch_sub(1, std::memory_order_acquire);
-        if (oldCount <= 0)
-        {
-            m_sema.wait();
-        }
-    }
-
-public:
-    LightweightSemaphore(int initialCount = 0) : m_count(initialCount)
-    {
-        assert(initialCount >= 0);
-    }
-
-    bool tryWait()
-    {
-        int oldCount = m_count.load(std::memory_order_relaxed);
-        return (oldCount > 0 && m_count.compare_exchange_strong(oldCount, oldCount - 1, std::memory_order_acquire));
-    }
-
-    void wait()
-    {
-        if (!tryWait())
-            waitWithPartialSpinning();
-    }
-
-    void signal(int count = 1)
-    {
-        int oldCount = m_count.fetch_add(count, std::memory_order_release);
-        int toRelease = -oldCount < count ? -oldCount : count;
-        if (toRelease > 0)
-        {
-            m_sema.signal(toRelease);
-        }
-    }
-};
-
-
-typedef LightweightSemaphore DefaultSemaphoreType;
-
-}
-
-#endif // __CPP11OM_SEMAPHORE_H__

examples/OpenCLVectorAdd/CMakeLists.txt

@@ -0,0 +1,14 @@
+cmake_minimum_required(VERSION 3.0)
+
+project(OpenCLVectorAdd)
+
+find_package(OpenCL REQUIRED)
+
+add_executable(OpenCLVectorAdd OpenCLVectorAdd.cpp)
+
+add_library(TracyClient STATIC ../../TracyClient.cpp
+                               ../../TracyOpenCL.hpp)
+target_include_directories(TracyClient PUBLIC ../../)
+target_compile_definitions(TracyClient PUBLIC TRACY_ENABLE=1)
+
+target_link_libraries(OpenCLVectorAdd PUBLIC OpenCL::OpenCL TracyClient)

examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp

@@ -0,0 +1,190 @@
+#include <iostream>
+#include <cassert>
+#include <string>
+#include <vector>
+#include <numeric>
+
+#include <CL/cl.h>
+
+#include <Tracy.hpp>
+#include <TracyOpenCL.hpp>
+
+#define CL_ASSERT(err)                                              \
+    if((err) != CL_SUCCESS)                                         \
+    {                                                               \
+        std::cerr << "OpenCL Call Returned " << err << std::endl;   \
+        assert(false);                                              \
+    }
+
+const char kernelSource[] =
+"   void __kernel vectorAdd(global float* C, global float* A, global float* B, int N)  "
+"   {                                                                                  "
+"       int i = get_global_id(0);                                                      "
+"       if (i < N) {                                                                   "
+"           C[i] = A[i] + B[i];                                                        "
+"       }                                                                              "
+"   }                                                                                  ";
+
+int main()
+{
+    cl_platform_id platform;
+    cl_device_id device;
+    cl_context context;
+    cl_command_queue commandQueue;
+    cl_kernel vectorAddKernel;
+    cl_program program;
+    cl_int err;
+    cl_mem bufferA, bufferB, bufferC;
+
+    TracyCLCtx tracyCLCtx;
+
+    {
+        ZoneScopedN("OpenCL Init");
+
+        cl_uint numPlatforms = 0;
+        CL_ASSERT(clGetPlatformIDs(0, nullptr, &numPlatforms));
+
+        if (numPlatforms == 0)
+        {
+            std::cerr << "Cannot find OpenCL platform to run this application" << std::endl;
+            return 1;
+        }
+
+        CL_ASSERT(clGetPlatformIDs(1, &platform, nullptr));
+
+        size_t platformNameBufferSize = 0;
+        CL_ASSERT(clGetPlatformInfo(platform, CL_PLATFORM_NAME, 0, nullptr, &platformNameBufferSize));
+        std::string platformName(platformNameBufferSize, '\0');
+        CL_ASSERT(clGetPlatformInfo(platform, CL_PLATFORM_NAME, platformNameBufferSize, &platformName[0], nullptr));
+
+        std::cout << "OpenCL Platform: " << platformName << std::endl;
+
+        CL_ASSERT(clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 1, &device, nullptr));
+        size_t deviceNameBufferSize = 0;
+        CL_ASSERT(clGetDeviceInfo(device, CL_DEVICE_NAME, 0, nullptr, &deviceNameBufferSize));
+        std::string deviceName(deviceNameBufferSize, '\0');
+        CL_ASSERT(clGetDeviceInfo(device, CL_DEVICE_NAME, deviceNameBufferSize, &deviceName[0], nullptr));
+
+        std::cout << "OpenCL Device: " << deviceName << std::endl;
+
+        err = CL_SUCCESS;
+        context = clCreateContext(nullptr, 1, &device, nullptr, nullptr, &err);
+        CL_ASSERT(err);
+
+        size_t kernelSourceLength = sizeof(kernelSource);
+        const char* kernelSourceArray = { kernelSource };
+        program = clCreateProgramWithSource(context, 1, &kernelSourceArray, &kernelSourceLength, &err);
+        CL_ASSERT(err);
+
+        if (clBuildProgram(program, 1, &device, nullptr, nullptr, nullptr) != CL_SUCCESS)
+        {
+            size_t programBuildLogBufferSize = 0;
+            CL_ASSERT(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, nullptr, &programBuildLogBufferSize));
+            std::string programBuildLog(programBuildLogBufferSize, '\0');
+            CL_ASSERT(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, programBuildLogBufferSize, &programBuildLog[0], nullptr));
+            std::clog << programBuildLog << std::endl;
+            return 1;
+        }
+
+        vectorAddKernel = clCreateKernel(program, "vectorAdd", &err);
+        CL_ASSERT(err);
+
+        commandQueue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
+        CL_ASSERT(err);
+    }
+
+    tracyCLCtx = TracyCLContext(context, device);
+
+    size_t N = 10 * 1024 * 1024 / sizeof(float); // 10MB of floats
+    std::vector<float> hostA, hostB, hostC;
+
+    {
+        ZoneScopedN("Host Data Init");
+        hostA.resize(N);
+        hostB.resize(N);
+        hostC.resize(N);
+
+        std::iota(std::begin(hostA), std::end(hostA), 0);
+        std::iota(std::begin(hostB), std::end(hostB), 0);
+    }
+
+    {
+        ZoneScopedN("Host to Device Memory Copy");
+
+        bufferA = clCreateBuffer(context, CL_MEM_READ_WRITE, N * sizeof(float), nullptr, &err);
+        CL_ASSERT(err);
+        bufferB = clCreateBuffer(context, CL_MEM_READ_WRITE, N * sizeof(float), nullptr, &err);
+        CL_ASSERT(err);
+        bufferC = clCreateBuffer(context, CL_MEM_READ_WRITE, N * sizeof(float), nullptr, &err);
+        CL_ASSERT(err);
+
+        cl_event writeBufferAEvent, writeBufferBEvent;
+        {
+            ZoneScopedN("Write Buffer A");
+            TracyCLZoneS(tracyCLCtx, "Write BufferA", 5);
+
+            CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferA, CL_TRUE, 0, N * sizeof(float), hostA.data(), 0, nullptr, &writeBufferAEvent));
+
+            TracyCLZoneSetEvent(writeBufferAEvent);
+        }
+        {
+            ZoneScopedN("Write Buffer B");
+            TracyCLZone(tracyCLCtx, "Write BufferB");
+
+            CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferB, CL_TRUE, 0, N * sizeof(float), hostB.data(), 0, nullptr, &writeBufferBEvent));
+
+            TracyCLZoneSetEvent(writeBufferBEvent);
+        }
+    }
+
+    for (int i = 0; i < 10; ++i)
+    {
+        ZoneScopedN("VectorAdd Kernel Launch");
+        TracyCLZoneC(tracyCLCtx, "VectorAdd Kernel", tracy::Color::Blue4);
+
+        CL_ASSERT(clSetKernelArg(vectorAddKernel, 0, sizeof(cl_mem), &bufferC));
+        CL_ASSERT(clSetKernelArg(vectorAddKernel, 1, sizeof(cl_mem), &bufferA));
+        CL_ASSERT(clSetKernelArg(vectorAddKernel, 2, sizeof(cl_mem), &bufferB));
+        CL_ASSERT(clSetKernelArg(vectorAddKernel, 3, sizeof(int), &static_cast<int>(N)));
+
+        cl_event vectorAddKernelEvent;
+        CL_ASSERT(clEnqueueNDRangeKernel(commandQueue, vectorAddKernel, 1, nullptr, &N, nullptr, 0, nullptr, &vectorAddKernelEvent));
+
+        CL_ASSERT(clWaitForEvents(1, &vectorAddKernelEvent));
+
+        TracyCLZoneSetEvent(vectorAddKernelEvent);
+
+        cl_ulong kernelStartTime, kernelEndTime;
+        CL_ASSERT(clGetEventProfilingInfo(vectorAddKernelEvent, CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &kernelStartTime, nullptr));
+        CL_ASSERT(clGetEventProfilingInfo(vectorAddKernelEvent, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &kernelEndTime, nullptr));
+        std::cout << "VectorAdd Kernel Elapsed: " << ((kernelEndTime - kernelStartTime) / 1000) << " us" << std::endl;
+    }
+
+    {
+        ZoneScopedN("Device to Host Memory Copy");
+        TracyCLZone(tracyCLCtx, "Read Buffer C");
+
+        cl_event readbufferCEvent;
+        CL_ASSERT(clEnqueueReadBuffer(commandQueue, bufferC, CL_TRUE, 0, N * sizeof(float), hostC.data(), 0, nullptr, &readbufferCEvent));
+        TracyCLZoneSetEvent(readbufferCEvent);
+    }
+
+    CL_ASSERT(clFinish(commandQueue));
+
+    TracyCLCollect(tracyCLCtx);
+
+    {
+        ZoneScopedN("Checking results");
+
+        for (int i = 0; i < N; ++i)
+        {
+            assert(hostC[i] == hostA[i] + hostB[i]);
+        }
+    }
+
+    std::cout << "Results are correct!" << std::endl;
+
+    TracyCLDestroy(tracyCLCtx);
+
+    return 0;
+}

examples/ToyPathTracer/.gitignore

@@ -0,0 +1 @@
+Windows/Compiled*Shader.h

examples/ToyPathTracer/README

@@ -0,0 +1,4 @@
+https://github.com/aras-p/ToyPathTracer
+
+Modified to render only 10 frames. Client part requires 12 GB, server part
+requires 6.4 GB.

examples/ToyPathTracer/Source/Config.h

@@ -0,0 +1,33 @@
+
+#if defined(__APPLE__) && !defined(__METAL_VERSION__)
+#include <TargetConditionals.h>
+#endif
+
+#define kBackbufferWidth 1280
+#define kBackbufferHeight 720
+
+#if defined(__EMSCRIPTEN__)
+#define CPU_CAN_DO_SIMD 0
+#define CPU_CAN_DO_THREADS 0
+#else
+#define CPU_CAN_DO_SIMD 1
+#define CPU_CAN_DO_THREADS 1
+#endif
+
+
+#define DO_SAMPLES_PER_PIXEL 4
+#define DO_ANIMATE_SMOOTHING 0.9f
+#define DO_LIGHT_SAMPLING 1
+#define DO_MITSUBA_COMPARE 0
+
+// Should path tracing be done on the GPU with a compute shader?
+#define DO_COMPUTE_GPU 0
+#define kCSGroupSizeX 8
+#define kCSGroupSizeY 8
+#define kCSMaxObjects 64
+
+// Should float3 struct use SSE/NEON?
+#define DO_FLOAT3_WITH_SIMD (!(DO_COMPUTE_GPU) && CPU_CAN_DO_SIMD && 1)
+
+// Should HitSpheres function use SSE/NEON?
+#define DO_HIT_SPHERES_SIMD (CPU_CAN_DO_SIMD && 1)

examples/ToyPathTracer/Source/MathSimd.h

@@ -0,0 +1,192 @@
+#pragma once
+
+#if defined(_MSC_VER)
+#define VM_INLINE __forceinline
+#else
+#define VM_INLINE __attribute__((unused, always_inline, nodebug)) inline
+#endif
+
+#define kSimdWidth 4
+
+#if !defined(__arm__) && !defined(__arm64__) && !defined(__EMSCRIPTEN__)
+
+// ---- SSE implementation
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+#include <smmintrin.h>
+
+#define SHUFFLE4(V, X,Y,Z,W) float4(_mm_shuffle_ps((V).m, (V).m, _MM_SHUFFLE(W,Z,Y,X)))
+
+struct float4
+{
+    VM_INLINE float4() {}
+    VM_INLINE explicit float4(const float *p) { m = _mm_loadu_ps(p); }
+    VM_INLINE explicit float4(float x, float y, float z, float w) { m = _mm_set_ps(w, z, y, x); }
+    VM_INLINE explicit float4(float v) { m = _mm_set_ps1(v); }
+    VM_INLINE explicit float4(__m128 v) { m = v; }
+    
+    VM_INLINE float getX() const { return _mm_cvtss_f32(m); }
+    VM_INLINE float getY() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(1, 1, 1, 1))); }
+    VM_INLINE float getZ() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(2, 2, 2, 2))); }
+    VM_INLINE float getW() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(3, 3, 3, 3))); }
+    
+    __m128 m;
+};
+
+typedef float4 bool4;
+
+VM_INLINE float4 operator+ (float4 a, float4 b) { a.m = _mm_add_ps(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a, float4 b) { a.m = _mm_sub_ps(a.m, b.m); return a; }
+VM_INLINE float4 operator* (float4 a, float4 b) { a.m = _mm_mul_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator==(float4 a, float4 b) { a.m = _mm_cmpeq_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator!=(float4 a, float4 b) { a.m = _mm_cmpneq_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator< (float4 a, float4 b) { a.m = _mm_cmplt_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator> (float4 a, float4 b) { a.m = _mm_cmpgt_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator<=(float4 a, float4 b) { a.m = _mm_cmple_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator>=(float4 a, float4 b) { a.m = _mm_cmpge_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator&(bool4 a, bool4 b) { a.m = _mm_and_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator|(bool4 a, bool4 b) { a.m = _mm_or_ps(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a) { a.m = _mm_xor_ps(a.m, _mm_set1_ps(-0.0f)); return a; }
+VM_INLINE float4 min(float4 a, float4 b) { a.m = _mm_min_ps(a.m, b.m); return a; }
+VM_INLINE float4 max(float4 a, float4 b) { a.m = _mm_max_ps(a.m, b.m); return a; }
+
+VM_INLINE float hmin(float4 v)
+{
+    v = min(v, SHUFFLE4(v, 2, 3, 0, 0));
+    v = min(v, SHUFFLE4(v, 1, 0, 0, 0));
+    return v.getX();
+}
+
+// Returns a 4-bit code where bit0..bit3 is X..W
+VM_INLINE unsigned mask(float4 v) { return _mm_movemask_ps(v.m); }
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool4 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool4 v) { return mask(v) == 15; }
+
+// "select", i.e. hibit(cond) ? b : a
+// on SSE4.1 and up this can be done easily via "blend" instruction;
+// on older SSEs has to do a bunch of hoops, see
+// https://fgiesen.wordpress.com/2016/04/03/sse-mind-the-gap/
+
+VM_INLINE float4 select(float4 a, float4 b, bool4 cond)
+{
+#if defined(__SSE4_1__) || defined(_MSC_VER) // on windows assume we always have SSE4.1
+    a.m = _mm_blendv_ps(a.m, b.m, cond.m);
+#else
+    __m128 d = _mm_castsi128_ps(_mm_srai_epi32(_mm_castps_si128(cond.m), 31));
+    a.m = _mm_or_ps(_mm_and_ps(d, b.m), _mm_andnot_ps(d, a.m));
+#endif
+    return a;
+}
+VM_INLINE __m128i select(__m128i a, __m128i b, bool4 cond)
+{
+#if defined(__SSE4_1__) || defined(_MSC_VER) // on windows assume we always have SSE4.1
+    return _mm_blendv_epi8(a, b, _mm_castps_si128(cond.m));
+#else
+    __m128i d = _mm_srai_epi32(_mm_castps_si128(cond.m), 31);
+    return _mm_or_si128(_mm_and_si128(d, b), _mm_andnot_si128(d, a));
+#endif
+}
+
+VM_INLINE float4 sqrtf(float4 v) { return float4(_mm_sqrt_ps(v.m)); }
+
+#elif !defined(__EMSCRIPTEN__)
+
+// ---- NEON implementation
+
+#define USE_NEON 1
+#include <arm_neon.h>
+
+struct float4
+{
+    VM_INLINE float4() {}
+    VM_INLINE explicit float4(const float *p) { m = vld1q_f32(p); }
+    VM_INLINE explicit float4(float x, float y, float z, float w) { float v[4] = {x, y, z, w}; m = vld1q_f32(v); }
+    VM_INLINE explicit float4(float v) { m = vdupq_n_f32(v); }
+    VM_INLINE explicit float4(float32x4_t v) { m = v; }
+    
+    VM_INLINE float getX() const { return vgetq_lane_f32(m, 0); }
+    VM_INLINE float getY() const { return vgetq_lane_f32(m, 1); }
+    VM_INLINE float getZ() const { return vgetq_lane_f32(m, 2); }
+    VM_INLINE float getW() const { return vgetq_lane_f32(m, 3); }
+    
+    float32x4_t m;
+};
+
+typedef float4 bool4;
+
+VM_INLINE float4 operator+ (float4 a, float4 b) { a.m = vaddq_f32(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a, float4 b) { a.m = vsubq_f32(a.m, b.m); return a; }
+VM_INLINE float4 operator* (float4 a, float4 b) { a.m = vmulq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator==(float4 a, float4 b) { a.m = vceqq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator!=(float4 a, float4 b) { a.m = a.m = vmvnq_u32(vceqq_f32(a.m, b.m)); return a; }
+VM_INLINE bool4 operator< (float4 a, float4 b) { a.m = vcltq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator> (float4 a, float4 b) { a.m = vcgtq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator<=(float4 a, float4 b) { a.m = vcleq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator>=(float4 a, float4 b) { a.m = vcgeq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator&(bool4 a, bool4 b) { a.m = vandq_u32(a.m, b.m); return a; }
+VM_INLINE bool4 operator|(bool4 a, bool4 b) { a.m = vorrq_u32(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a) { a.m = vnegq_f32(a.m); return a; }
+VM_INLINE float4 min(float4 a, float4 b) { a.m = vminq_f32(a.m, b.m); return a; }
+VM_INLINE float4 max(float4 a, float4 b) { a.m = vmaxq_f32(a.m, b.m); return a; }
+
+VM_INLINE float hmin(float4 v)
+{
+    float32x2_t minOfHalfs = vpmin_f32(vget_low_f32(v.m), vget_high_f32(v.m));
+    float32x2_t minOfMinOfHalfs = vpmin_f32(minOfHalfs, minOfHalfs);
+    return vget_lane_f32(minOfMinOfHalfs, 0);
+}
+
+// Returns a 4-bit code where bit0..bit3 is X..W
+VM_INLINE unsigned mask(float4 v)
+{
+    static const uint32x4_t movemask = { 1, 2, 4, 8 };
+    static const uint32x4_t highbit = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+    uint32x4_t t0 = vreinterpretq_u32_f32(v.m);
+    uint32x4_t t1 = vtstq_u32(t0, highbit);
+    uint32x4_t t2 = vandq_u32(t1, movemask);
+    uint32x2_t t3 = vorr_u32(vget_low_u32(t2), vget_high_u32(t2));
+    return vget_lane_u32(t3, 0) | vget_lane_u32(t3, 1);
+}
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool4 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool4 v) { return mask(v) == 15; }
+
+// "select", i.e. hibit(cond) ? b : a
+// on SSE4.1 and up this can be done easily via "blend" instruction;
+// on older SSEs has to do a bunch of hoops, see
+// https://fgiesen.wordpress.com/2016/04/03/sse-mind-the-gap/
+
+VM_INLINE float4 select(float4 a, float4 b, bool4 cond)
+{
+    a.m = vbslq_f32(cond.m, b.m, a.m);
+    return a;
+}
+VM_INLINE int32x4_t select(int32x4_t a, int32x4_t b, bool4 cond)
+{
+    return vbslq_f32(cond.m, b, a);
+}
+
+VM_INLINE float4 sqrtf(float4 v)
+{
+    float32x4_t V = v.m;
+    float32x4_t S0 = vrsqrteq_f32(V);
+    float32x4_t P0 = vmulq_f32( V, S0 );
+    float32x4_t R0 = vrsqrtsq_f32( P0, S0 );
+    float32x4_t S1 = vmulq_f32( S0, R0 );
+    float32x4_t P1 = vmulq_f32( V, S1 );
+    float32x4_t R1 = vrsqrtsq_f32( P1, S1 );
+    float32x4_t S2 = vmulq_f32( S1, R1 );
+    float32x4_t P2 = vmulq_f32( V, S2 );
+    float32x4_t R2 = vrsqrtsq_f32( P2, S2 );
+    float32x4_t S3 = vmulq_f32( S2, R2 );
+    return float4(vmulq_f32(V, S3));
+}
+
+VM_INLINE float4 splatX(float32x4_t v) { return float4(vdupq_lane_f32(vget_low_f32(v), 0)); }
+VM_INLINE float4 splatY(float32x4_t v) { return float4(vdupq_lane_f32(vget_low_f32(v), 1)); }
+VM_INLINE float4 splatZ(float32x4_t v) { return float4(vdupq_lane_f32(vget_high_f32(v), 0)); }
+VM_INLINE float4 splatW(float32x4_t v) { return float4(vdupq_lane_f32(vget_high_f32(v), 1)); }
+
+#endif

examples/ToyPathTracer/Source/Maths.cpp

@@ -0,0 +1,203 @@
+#include "Maths.h"
+#include <stdlib.h>
+#include <stdint.h>
+
+static uint32_t XorShift32(uint32_t& state)
+{
+    uint32_t x = state;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 15;
+    state = x;
+    return x;
+}
+
+float RandomFloat01(uint32_t& state)
+{
+    return (XorShift32(state) & 0xFFFFFF) / 16777216.0f;
+}
+
+float3 RandomInUnitDisk(uint32_t& state)
+{
+    float3 p;
+    do
+    {
+        p = 2.0 * float3(RandomFloat01(state),RandomFloat01(state),0) - float3(1,1,0);
+    } while (dot(p,p) >= 1.0);
+    return p;
+}
+
+float3 RandomInUnitSphere(uint32_t& state)
+{
+    float3 p;
+    do {
+        p = 2.0*float3(RandomFloat01(state),RandomFloat01(state),RandomFloat01(state)) - float3(1,1,1);
+    } while (sqLength(p) >= 1.0);
+    return p;
+}
+
+float3 RandomUnitVector(uint32_t& state)
+{
+    float z = RandomFloat01(state) * 2.0f - 1.0f;
+    float a = RandomFloat01(state) * 2.0f * kPI;
+    float r = sqrtf(1.0f - z * z);
+    float x = r * cosf(a);
+    float y = r * sinf(a);
+    return float3(x, y, z);
+}
+
+
+int HitSpheres(const Ray& r, const SpheresSoA& spheres, float tMin, float tMax, Hit& outHit)
+{
+#if DO_HIT_SPHERES_SIMD
+    float4 hitT = float4(tMax);
+#if USE_NEON
+    int32x4_t id = vdupq_n_s32(-1);
+#else
+    __m128i id = _mm_set1_epi32(-1);
+#endif
+
+#if DO_FLOAT3_WITH_SIMD && !USE_NEON
+    float4 rOrigX = SHUFFLE4(r.orig, 0, 0, 0, 0);
+    float4 rOrigY = SHUFFLE4(r.orig, 1, 1, 1, 1);
+    float4 rOrigZ = SHUFFLE4(r.orig, 2, 2, 2, 2);
+    float4 rDirX = SHUFFLE4(r.dir, 0, 0, 0, 0);
+    float4 rDirY = SHUFFLE4(r.dir, 1, 1, 1, 1);
+    float4 rDirZ = SHUFFLE4(r.dir, 2, 2, 2, 2);
+#elif DO_FLOAT3_WITH_SIMD
+    float4 rOrigX = splatX(r.orig.m);
+    float4 rOrigY = splatY(r.orig.m);
+    float4 rOrigZ = splatZ(r.orig.m);
+    float4 rDirX = splatX(r.dir.m);
+    float4 rDirY = splatY(r.dir.m);
+    float4 rDirZ = splatZ(r.dir.m);
+#else
+    float4 rOrigX = float4(r.orig.x);
+    float4 rOrigY = float4(r.orig.y);
+    float4 rOrigZ = float4(r.orig.z);
+    float4 rDirX = float4(r.dir.x);
+    float4 rDirY = float4(r.dir.y);
+    float4 rDirZ = float4(r.dir.z);
+#endif
+    float4 tMin4 = float4(tMin);
+#if USE_NEON
+    int32x4_t curId = vcombine_u32(vcreate_u32(0ULL | (1ULL<<32)), vcreate_u32(2ULL | (3ULL<<32)));
+#else
+    __m128i curId = _mm_set_epi32(3, 2, 1, 0);
+#endif
+    // process 4 spheres at once
+    for (int i = 0; i < spheres.simdCount; i += kSimdWidth)
+    {
+        // load data for 4 spheres
+        float4 sCenterX = float4(spheres.centerX + i);
+        float4 sCenterY = float4(spheres.centerY + i);
+        float4 sCenterZ = float4(spheres.centerZ + i);
+        float4 sSqRadius = float4(spheres.sqRadius + i);
+        // note: we flip this vector and calculate -b (nb) since that happens to be slightly preferable computationally
+        float4 coX = sCenterX - rOrigX;
+        float4 coY = sCenterY - rOrigY;
+        float4 coZ = sCenterZ - rOrigZ;
+        float4 nb = coX * rDirX + coY * rDirY + coZ * rDirZ;
+        float4 c = coX * coX + coY * coY + coZ * coZ - sSqRadius;
+        float4 discr = nb * nb - c;
+        bool4 discrPos = discr > float4(0.0f);
+        // if ray hits any of the 4 spheres
+        if (any(discrPos))
+        {
+            float4 discrSq = sqrtf(discr);
+
+            // ray could hit spheres at t0 & t1
+            float4 t0 = nb - discrSq;
+            float4 t1 = nb + discrSq;
+
+            float4 t = select(t1, t0, t0 > tMin4); // if t0 is above min, take it (since it's the earlier hit); else try t1.
+            bool4 msk = discrPos & (t > tMin4) & (t < hitT);
+            // if hit, take it
+            id = select(id, curId, msk);
+            hitT = select(hitT, t, msk);
+        }
+#if USE_NEON
+        curId = vaddq_s32(curId, vdupq_n_s32(kSimdWidth));
+#else
+        curId = _mm_add_epi32(curId, _mm_set1_epi32(kSimdWidth));
+#endif
+    }
+    // now we have up to 4 hits, find and return closest one
+    float minT = hmin(hitT);
+    if (minT < tMax) // any actual hits?
+    {
+        int minMask = mask(hitT == float4(minT));
+        if (minMask != 0)
+        {
+            int id_scalar[4];
+            float hitT_scalar[4];
+#if USE_NEON
+            vst1q_s32(id_scalar, id);
+            vst1q_f32(hitT_scalar, hitT.m);
+#else
+            _mm_storeu_si128((__m128i *)id_scalar, id);
+            _mm_storeu_ps(hitT_scalar, hitT.m);
+#endif
+
+            // In general, you would do this with a bit scan (first set/trailing zero count).
+            // But who cares, it's only 16 options.
+            static const int laneId[16] =
+            {
+                0, 0, 1, 0, // 00xx
+                2, 0, 1, 0, // 01xx
+                3, 0, 1, 0, // 10xx
+                2, 0, 1, 0, // 11xx
+            };
+
+            int lane = laneId[minMask];
+            int hitId = id_scalar[lane];
+            float finalHitT = hitT_scalar[lane];
+
+            outHit.pos = r.pointAt(finalHitT);
+            outHit.normal = (outHit.pos - float3(spheres.centerX[hitId], spheres.centerY[hitId], spheres.centerZ[hitId])) * spheres.invRadius[hitId];
+            outHit.t = finalHitT;
+            return hitId;
+        }
+    }
+
+    return -1;
+
+#else // #if DO_HIT_SPHERES_SIMD
+
+    float hitT = tMax;
+    int id = -1;
+    for (int i = 0; i < spheres.count; ++i)
+    {
+        float coX = spheres.centerX[i] - r.orig.getX();
+        float coY = spheres.centerY[i] - r.orig.getY();
+        float coZ = spheres.centerZ[i] - r.orig.getZ();
+        float nb = coX * r.dir.getX() + coY * r.dir.getY() + coZ * r.dir.getZ();
+        float c = coX * coX + coY * coY + coZ * coZ - spheres.sqRadius[i];
+        float discr = nb * nb - c;
+        if (discr > 0)
+        {
+            float discrSq = sqrtf(discr);
+
+            // Try earlier t
+            float t = nb - discrSq;
+            if (t <= tMin) // before min, try later t!
+                t = nb + discrSq;
+
+            if (t > tMin && t < hitT)
+            {
+                id = i;
+                hitT = t;
+            }
+        }
+    }
+    if (id != -1)
+    {
+        outHit.pos = r.pointAt(hitT);
+        outHit.normal = (outHit.pos - float3(spheres.centerX[id], spheres.centerY[id], spheres.centerZ[id])) * spheres.invRadius[id];
+        outHit.t = hitT;
+        return id;
+    }
+    else
+        return -1;
+#endif // #else of #if DO_HIT_SPHERES_SIMD
+}

examples/ToyPathTracer/Source/Maths.h

@@ -0,0 +1,436 @@
+#pragma once
+
+#include <math.h>
+#include <assert.h>
+#include <stdint.h>
+#include "Config.h"
+#include "MathSimd.h"
+
+#define kPI 3.1415926f
+
+// SSE/SIMD vector largely based on http://www.codersnotes.com/notes/maths-lib-2016/
+#if DO_FLOAT3_WITH_SIMD
+
+
+#if !defined(__arm__) && !defined(__arm64__)
+
+// ---- SSE implementation
+
+// SHUFFLE3(v, 0,1,2) leaves the vector unchanged (v.xyz).
+// SHUFFLE3(v, 0,0,0) splats the X (v.xxx).
+#define SHUFFLE3(V, X,Y,Z) float3(_mm_shuffle_ps((V).m, (V).m, _MM_SHUFFLE(Z,Z,Y,X)))
+
+struct float3
+{
+    VM_INLINE float3() {}
+    VM_INLINE explicit float3(const float *p) { m = _mm_set_ps(p[2], p[2], p[1], p[0]); }
+    VM_INLINE explicit float3(float x, float y, float z) { m = _mm_set_ps(z, z, y, x); }
+    VM_INLINE explicit float3(float v) { m = _mm_set1_ps(v); }
+    VM_INLINE explicit float3(__m128 v) { m = v; }
+
+    VM_INLINE float getX() const { return _mm_cvtss_f32(m); }
+    VM_INLINE float getY() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(1, 1, 1, 1))); }
+    VM_INLINE float getZ() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(2, 2, 2, 2))); }
+
+    VM_INLINE float3 yzx() const { return SHUFFLE3(*this, 1, 2, 0); }
+    VM_INLINE float3 zxy() const { return SHUFFLE3(*this, 2, 0, 1); }
+
+    VM_INLINE void store(float *p) const { p[0] = getX(); p[1] = getY(); p[2] = getZ(); }
+
+    void setX(float x)
+    {
+        m = _mm_move_ss(m, _mm_set_ss(x));
+    }
+    void setY(float y)
+    {
+        __m128 t = _mm_move_ss(m, _mm_set_ss(y));
+        t = _mm_shuffle_ps(t, t, _MM_SHUFFLE(3, 2, 0, 0));
+        m = _mm_move_ss(t, m);
+    }
+    void setZ(float z)
+    {
+        __m128 t = _mm_move_ss(m, _mm_set_ss(z));
+        t = _mm_shuffle_ps(t, t, _MM_SHUFFLE(3, 0, 1, 0));
+        m = _mm_move_ss(t, m);
+    }
+
+    __m128 m;
+};
+
+typedef float3 bool3;
+
+VM_INLINE float3 operator+ (float3 a, float3 b) { a.m = _mm_add_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator- (float3 a, float3 b) { a.m = _mm_sub_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator* (float3 a, float3 b) { a.m = _mm_mul_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator/ (float3 a, float3 b) { a.m = _mm_div_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator* (float3 a, float b) { a.m = _mm_mul_ps(a.m, _mm_set1_ps(b)); return a; }
+VM_INLINE float3 operator/ (float3 a, float b) { a.m = _mm_div_ps(a.m, _mm_set1_ps(b)); return a; }
+VM_INLINE float3 operator* (float a, float3 b) { b.m = _mm_mul_ps(_mm_set1_ps(a), b.m); return b; }
+VM_INLINE float3 operator/ (float a, float3 b) { b.m = _mm_div_ps(_mm_set1_ps(a), b.m); return b; }
+VM_INLINE float3& operator+= (float3 &a, float3 b) { a = a + b; return a; }
+VM_INLINE float3& operator-= (float3 &a, float3 b) { a = a - b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float3 b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float3 b) { a = a / b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float b) { a = a / b; return a; }
+VM_INLINE bool3 operator==(float3 a, float3 b) { a.m = _mm_cmpeq_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator!=(float3 a, float3 b) { a.m = _mm_cmpneq_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator< (float3 a, float3 b) { a.m = _mm_cmplt_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator> (float3 a, float3 b) { a.m = _mm_cmpgt_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator<=(float3 a, float3 b) { a.m = _mm_cmple_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator>=(float3 a, float3 b) { a.m = _mm_cmpge_ps(a.m, b.m); return a; }
+VM_INLINE float3 min(float3 a, float3 b) { a.m = _mm_min_ps(a.m, b.m); return a; }
+VM_INLINE float3 max(float3 a, float3 b) { a.m = _mm_max_ps(a.m, b.m); return a; }
+
+VM_INLINE float3 operator- (float3 a) { return float3(_mm_setzero_ps()) - a; }
+
+VM_INLINE float hmin(float3 v)
+{
+    v = min(v, SHUFFLE3(v, 1, 0, 2));
+    return min(v, SHUFFLE3(v, 2, 0, 1)).getX();
+}
+VM_INLINE float hmax(float3 v)
+{
+    v = max(v, SHUFFLE3(v, 1, 0, 2));
+    return max(v, SHUFFLE3(v, 2, 0, 1)).getX();
+}
+
+VM_INLINE float3 cross(float3 a, float3 b)
+{
+    // x  <-  a.y*b.z - a.z*b.y
+    // y  <-  a.z*b.x - a.x*b.z
+    // z  <-  a.x*b.y - a.y*b.x
+    // We can save a shuffle by grouping it in this wacky order:
+    return (a.zxy()*b - a*b.zxy()).zxy();
+}
+
+// Returns a 3-bit code where bit0..bit2 is X..Z
+VM_INLINE unsigned mask(float3 v) { return _mm_movemask_ps(v.m) & 7; }
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool3 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool3 v) { return mask(v) == 7; }
+
+VM_INLINE float3 clamp(float3 t, float3 a, float3 b) { return min(max(t, a), b); }
+VM_INLINE float sum(float3 v) { return v.getX() + v.getY() + v.getZ(); }
+VM_INLINE float dot(float3 a, float3 b) { return sum(a*b); }
+
+#else // #if !defined(__arm__) && !defined(__arm64__)
+
+// ---- NEON implementation
+
+#include <arm_neon.h>
+
+struct float3
+{
+    VM_INLINE float3() {}
+    VM_INLINE explicit float3(const float *p) { float v[4] = {p[0], p[1], p[2], 0}; m = vld1q_f32(v); }
+    VM_INLINE explicit float3(float x, float y, float z) { float v[4] = {x, y, z, 0}; m = vld1q_f32(v); }
+    VM_INLINE explicit float3(float v) { m = vdupq_n_f32(v); }
+    VM_INLINE explicit float3(float32x4_t v) { m = v; }
+    
+    VM_INLINE float getX() const { return vgetq_lane_f32(m, 0); }
+    VM_INLINE float getY() const { return vgetq_lane_f32(m, 1); }
+    VM_INLINE float getZ() const { return vgetq_lane_f32(m, 2); }
+    
+    VM_INLINE float3 yzx() const
+    {
+        float32x2_t low = vget_low_f32(m);
+        float32x4_t yzx = vcombine_f32(vext_f32(low, vget_high_f32(m), 1), low);
+        return float3(yzx);
+    }
+    VM_INLINE float3 zxy() const
+    {
+        float32x4_t p = m;
+        p = vuzpq_f32(vreinterpretq_f32_s32(vextq_s32(vreinterpretq_s32_f32(p), vreinterpretq_s32_f32(p), 1)), p).val[1];
+        return float3(p);
+    }
+    
+    VM_INLINE void store(float *p) const { p[0] = getX(); p[1] = getY(); p[2] = getZ(); }
+    
+    void setX(float x)
+    {
+        m = vsetq_lane_f32(x, m, 0);
+    }
+    void setY(float y)
+    {
+        m = vsetq_lane_f32(y, m, 1);
+    }
+    void setZ(float z)
+    {
+        m = vsetq_lane_f32(z, m, 2);
+    }
+    
+    float32x4_t m;
+};
+
+typedef float3 bool3;
+
+VM_INLINE float32x4_t rcp_2(float32x4_t v)
+{
+    float32x4_t e = vrecpeq_f32(v);
+    e = vmulq_f32(vrecpsq_f32(e, v), e);
+    e = vmulq_f32(vrecpsq_f32(e, v), e);
+    return e;
+}
+
+VM_INLINE float3 operator+ (float3 a, float3 b) { a.m = vaddq_f32(a.m, b.m); return a; }
+VM_INLINE float3 operator- (float3 a, float3 b) { a.m = vsubq_f32(a.m, b.m); return a; }
+VM_INLINE float3 operator* (float3 a, float3 b) { a.m = vmulq_f32(a.m, b.m); return a; }
+VM_INLINE float3 operator/ (float3 a, float3 b) { float32x4_t recip = rcp_2(b.m); a.m = vmulq_f32(a.m, recip); return a; }
+VM_INLINE float3 operator* (float3 a, float b) { a.m = vmulq_f32(a.m, vdupq_n_f32(b)); return a; }
+VM_INLINE float3 operator/ (float3 a, float b) { float32x4_t recip = rcp_2(vdupq_n_f32(b)); a.m = vmulq_f32(a.m, recip); return a; }
+VM_INLINE float3 operator* (float a, float3 b) { b.m = vmulq_f32(vdupq_n_f32(a), b.m); return b; }
+VM_INLINE float3 operator/ (float a, float3 b) { float32x4_t recip = rcp_2(b.m); b.m = vmulq_f32(vdupq_n_f32(a), recip); return b; }
+VM_INLINE float3& operator+= (float3 &a, float3 b) { a = a + b; return a; }
+VM_INLINE float3& operator-= (float3 &a, float3 b) { a = a - b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float3 b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float3 b) { a = a / b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float b) { a = a / b; return a; }
+VM_INLINE bool3 operator==(float3 a, float3 b) { a.m = vceqq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator!=(float3 a, float3 b) { a.m = vmvnq_u32(vceqq_f32(a.m, b.m)); return a; }
+VM_INLINE bool3 operator< (float3 a, float3 b) { a.m = vcltq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator> (float3 a, float3 b) { a.m = vcgtq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator<=(float3 a, float3 b) { a.m = vcleq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator>=(float3 a, float3 b) { a.m = vcgeq_f32(a.m, b.m); return a; }
+VM_INLINE float3 min(float3 a, float3 b) { a.m = vminq_f32(a.m, b.m); return a; }
+VM_INLINE float3 max(float3 a, float3 b) { a.m = vmaxq_f32(a.m, b.m); return a; }
+
+VM_INLINE float3 operator- (float3 a) { a.m = vnegq_f32(a.m); return a; }
+
+VM_INLINE float hmin(float3 v)
+{
+    float32x2_t minOfHalfs = vpmin_f32(vget_low_f32(v.m), vget_high_f32(v.m));
+    float32x2_t minOfMinOfHalfs = vpmin_f32(minOfHalfs, minOfHalfs);
+    return vget_lane_f32(minOfMinOfHalfs, 0);
+}
+VM_INLINE float hmax(float3 v)
+{
+    float32x2_t maxOfHalfs = vpmax_f32(vget_low_f32(v.m), vget_high_f32(v.m));
+    float32x2_t maxOfMaxOfHalfs = vpmax_f32(maxOfHalfs, maxOfHalfs);
+    return vget_lane_f32(maxOfMaxOfHalfs, 0);
+}
+
+VM_INLINE float3 cross(float3 a, float3 b)
+{
+    // x  <-  a.y*b.z - a.z*b.y
+    // y  <-  a.z*b.x - a.x*b.z
+    // z  <-  a.x*b.y - a.y*b.x
+    // We can save a shuffle by grouping it in this wacky order:
+    return (a.zxy()*b - a*b.zxy()).zxy();
+}
+
+// Returns a 3-bit code where bit0..bit2 is X..Z
+VM_INLINE unsigned mask(float3 v)
+{
+    static const uint32x4_t movemask = { 1, 2, 4, 8 };
+    static const uint32x4_t highbit = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+    uint32x4_t t0 = vreinterpretq_u32_f32(v.m);
+    uint32x4_t t1 = vtstq_u32(t0, highbit);
+    uint32x4_t t2 = vandq_u32(t1, movemask);
+    uint32x2_t t3 = vorr_u32(vget_low_u32(t2), vget_high_u32(t2));
+    return vget_lane_u32(t3, 0) | vget_lane_u32(t3, 1);
+}
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool3 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool3 v) { return mask(v) == 7; }
+
+VM_INLINE float3 clamp(float3 t, float3 a, float3 b) { return min(max(t, a), b); }
+VM_INLINE float sum(float3 v) { return v.getX() + v.getY() + v.getZ(); }
+VM_INLINE float dot(float3 a, float3 b) { return sum(a*b); }
+
+
+#endif // #else of #if !defined(__arm__) && !defined(__arm64__)
+
+#else // #if DO_FLOAT3_WITH_SIMD
+
+// ---- Simple scalar C implementation
+
+
+struct float3
+{
+    float3() : x(0), y(0), z(0) {}
+    float3(float x_, float y_, float z_) : x(x_), y(y_), z(z_) {}
+
+    float3 operator-() const { return float3(-x, -y, -z); }
+    float3& operator+=(const float3& o) { x+=o.x; y+=o.y; z+=o.z; return *this; }
+    float3& operator-=(const float3& o) { x-=o.x; y-=o.y; z-=o.z; return *this; }
+    float3& operator*=(const float3& o) { x*=o.x; y*=o.y; z*=o.z; return *this; }
+    float3& operator*=(float o) { x*=o; y*=o; z*=o; return *this; }
+
+    VM_INLINE float getX() const { return x; }
+    VM_INLINE float getY() const { return y; }
+    VM_INLINE float getZ() const { return z; }
+    VM_INLINE void setX(float x_) { x = x_; }
+    VM_INLINE void setY(float y_) { y = y_; }
+    VM_INLINE void setZ(float z_) { z = z_; }
+    VM_INLINE void store(float *p) const { p[0] = getX(); p[1] = getY(); p[2] = getZ(); }
+
+    float x, y, z;
+};
+
+VM_INLINE float3 operator+(const float3& a, const float3& b) { return float3(a.x+b.x,a.y+b.y,a.z+b.z); }
+VM_INLINE float3 operator-(const float3& a, const float3& b) { return float3(a.x-b.x,a.y-b.y,a.z-b.z); }
+VM_INLINE float3 operator*(const float3& a, const float3& b) { return float3(a.x*b.x,a.y*b.y,a.z*b.z); }
+VM_INLINE float3 operator*(const float3& a, float b) { return float3(a.x*b,a.y*b,a.z*b); }
+VM_INLINE float3 operator*(float a, const float3& b) { return float3(a*b.x,a*b.y,a*b.z); }
+VM_INLINE float dot(const float3& a, const float3& b) { return a.x*b.x+a.y*b.y+a.z*b.z; }
+VM_INLINE float3 cross(const float3& a, const float3& b)
+{
+    return float3(
+                  a.y*b.z - a.z*b.y,
+                  -(a.x*b.z - a.z*b.x),
+                  a.x*b.y - a.y*b.x
+                  );
+}
+#endif // #else of #if DO_FLOAT3_WITH_SIMD
+
+VM_INLINE float length(float3 v) { return sqrtf(dot(v, v)); }
+VM_INLINE float sqLength(float3 v) { return dot(v, v); }
+VM_INLINE float3 normalize(float3 v) { return v * (1.0f / length(v)); }
+VM_INLINE float3 lerp(float3 a, float3 b, float t) { return a + (b-a)*t; }
+
+
+inline void AssertUnit(float3 v)
+{
+    assert(fabsf(sqLength(v) - 1.0f) < 0.01f);
+}
+
+inline float3 reflect(float3 v, float3 n)
+{
+    return v - 2*dot(v,n)*n;
+}
+
+inline bool refract(float3 v, float3 n, float nint, float3& outRefracted)
+{
+    AssertUnit(v);
+    float dt = dot(v, n);
+    float discr = 1.0f - nint*nint*(1-dt*dt);
+    if (discr > 0)
+    {
+        outRefracted = nint * (v - n*dt) - n*sqrtf(discr);
+        return true;
+    }
+    return false;
+}
+inline float schlick(float cosine, float ri)
+{
+    float r0 = (1-ri) / (1+ri);
+    r0 = r0*r0;
+    return r0 + (1-r0)*powf(1-cosine, 5);
+}
+
+struct Ray
+{
+    Ray() {}
+    Ray(float3 orig_, float3 dir_) : orig(orig_), dir(dir_) { AssertUnit(dir); }
+
+    float3 pointAt(float t) const { return orig + dir * t; }
+
+    float3 orig;
+    float3 dir;
+};
+
+
+struct Hit
+{
+    float3 pos;
+    float3 normal;
+    float t;
+};
+
+
+struct Sphere
+{
+    Sphere() : radius(1.0f), invRadius(0.0f) {}
+    Sphere(float3 center_, float radius_) : center(center_), radius(radius_), invRadius(0.0f) {}
+
+    void UpdateDerivedData() { invRadius = 1.0f/radius; }
+
+    float3 center;
+    float radius;
+    float invRadius;
+};
+
+
+// data for all spheres in a "structure of arrays" layout
+struct SpheresSoA
+{
+    SpheresSoA(int c)
+    {
+        count = c;
+        // we'll be processing spheres in kSimdWidth chunks, so make sure to allocate
+        // enough space
+        simdCount = (c + (kSimdWidth - 1)) / kSimdWidth * kSimdWidth;
+        centerX = new float[simdCount];
+        centerY = new float[simdCount];
+        centerZ = new float[simdCount];
+        sqRadius = new float[simdCount];
+        invRadius = new float[simdCount];
+        // set all data to "impossible sphere" state
+        for (int i = count; i < simdCount; ++i)
+        {
+            centerX[i] = centerY[i] = centerZ[i] = 10000.0f;
+            sqRadius[i] = 0.0f;
+            invRadius[i] = 0.0f;
+        }
+    }
+    ~SpheresSoA()
+    {
+        delete[] centerX;
+        delete[] centerY;
+        delete[] centerZ;
+        delete[] sqRadius;
+        delete[] invRadius;
+    }
+    float* centerX;
+    float* centerY;
+    float* centerZ;
+    float* sqRadius;
+    float* invRadius;
+    int simdCount;
+    int count;
+};
+
+
+int HitSpheres(const Ray& r, const SpheresSoA& spheres, float tMin, float tMax, Hit& outHit);
+
+float RandomFloat01(uint32_t& state);
+float3 RandomInUnitDisk(uint32_t& state);
+float3 RandomInUnitSphere(uint32_t& state);
+float3 RandomUnitVector(uint32_t& state);
+
+struct Camera
+{
+    Camera() {}
+    // vfov is top to bottom in degrees
+    Camera(const float3& lookFrom, const float3& lookAt, const float3& vup, float vfov, float aspect, float aperture, float focusDist)
+    {
+        lensRadius = aperture / 2;
+        float theta = vfov*kPI/180;
+        float halfHeight = tanf(theta/2);
+        float halfWidth = aspect * halfHeight;
+        origin = lookFrom;
+        w = normalize(lookFrom - lookAt);
+        u = normalize(cross(vup, w));
+        v = cross(w, u);
+        lowerLeftCorner = origin - halfWidth*focusDist*u - halfHeight*focusDist*v - focusDist*w;
+        horizontal = 2*halfWidth*focusDist*u;
+        vertical = 2*halfHeight*focusDist*v;
+    }
+
+    Ray GetRay(float s, float t, uint32_t& state) const
+    {
+        float3 rd = lensRadius * RandomInUnitDisk(state);
+        float3 offset = u * rd.getX() + v * rd.getY();
+        return Ray(origin + offset, normalize(lowerLeftCorner + s*horizontal + t*vertical - origin - offset));
+    }
+
+    float3 origin;
+    float3 lowerLeftCorner;
+    float3 horizontal;
+    float3 vertical;
+    float3 u, v, w;
+    float lensRadius;
+};
+

examples/ToyPathTracer/Source/Test.cpp

@@ -0,0 +1,392 @@
+#include "Config.h"
+#include "Test.h"
+#include "Maths.h"
+#include <algorithm>
+#if CPU_CAN_DO_THREADS
+#include "enkiTS/TaskScheduler_c.h"
+#include <thread>
+#endif
+#include <atomic>
+
+#include "../../../Tracy.hpp"
+
+// 46 spheres (2 emissive) when enabled; 9 spheres (1 emissive) when disabled
+#define DO_BIG_SCENE 1
+
+static Sphere s_Spheres[] =
+{
+    {float3(0,-100.5,-1), 100},
+    {float3(2,0,-1), 0.5f},
+    {float3(0,0,-1), 0.5f},
+    {float3(-2,0,-1), 0.5f},
+    {float3(2,0,1), 0.5f},
+    {float3(0,0,1), 0.5f},
+    {float3(-2,0,1), 0.5f},
+    {float3(0.5f,1,0.5f), 0.5f},
+    {float3(-1.5f,1.5f,0.f), 0.3f},
+#if DO_BIG_SCENE
+    {float3(4,0,-3), 0.5f}, {float3(3,0,-3), 0.5f}, {float3(2,0,-3), 0.5f}, {float3(1,0,-3), 0.5f}, {float3(0,0,-3), 0.5f}, {float3(-1,0,-3), 0.5f}, {float3(-2,0,-3), 0.5f}, {float3(-3,0,-3), 0.5f}, {float3(-4,0,-3), 0.5f},
+    {float3(4,0,-4), 0.5f}, {float3(3,0,-4), 0.5f}, {float3(2,0,-4), 0.5f}, {float3(1,0,-4), 0.5f}, {float3(0,0,-4), 0.5f}, {float3(-1,0,-4), 0.5f}, {float3(-2,0,-4), 0.5f}, {float3(-3,0,-4), 0.5f}, {float3(-4,0,-4), 0.5f},
+    {float3(4,0,-5), 0.5f}, {float3(3,0,-5), 0.5f}, {float3(2,0,-5), 0.5f}, {float3(1,0,-5), 0.5f}, {float3(0,0,-5), 0.5f}, {float3(-1,0,-5), 0.5f}, {float3(-2,0,-5), 0.5f}, {float3(-3,0,-5), 0.5f}, {float3(-4,0,-5), 0.5f},
+    {float3(4,0,-6), 0.5f}, {float3(3,0,-6), 0.5f}, {float3(2,0,-6), 0.5f}, {float3(1,0,-6), 0.5f}, {float3(0,0,-6), 0.5f}, {float3(-1,0,-6), 0.5f}, {float3(-2,0,-6), 0.5f}, {float3(-3,0,-6), 0.5f}, {float3(-4,0,-6), 0.5f},
+    {float3(1.5f,1.5f,-2), 0.3f},
+#endif // #if DO_BIG_SCENE
+};
+const int kSphereCount = sizeof(s_Spheres) / sizeof(s_Spheres[0]);
+
+static SpheresSoA s_SpheresSoA(kSphereCount);
+
+struct Material
+{
+    enum Type { Lambert, Metal, Dielectric };
+    Type type;
+    float3 albedo;
+    float3 emissive;
+    float roughness;
+    float ri;
+};
+
+static Material s_SphereMats[kSphereCount] =
+{
+    { Material::Lambert, float3(0.8f, 0.8f, 0.8f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.8f, 0.4f, 0.4f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0, 0, },
+    { Material::Metal, float3(0.4f, 0.4f, 0.8f), float3(0,0,0), 0, 0 },
+    { Material::Metal, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0, 0 },
+    { Material::Metal, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0.2f, 0 },
+    { Material::Metal, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0.6f, 0 },
+    { Material::Dielectric, float3(0.4f, 0.4f, 0.4f), float3(0,0,0), 0, 1.5f },
+    { Material::Lambert, float3(0.8f, 0.6f, 0.2f), float3(30,25,15), 0, 0 },
+#if DO_BIG_SCENE
+    { Material::Lambert, float3(0.1f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.2f, 0.2f, 0.2f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.3f, 0.3f, 0.3f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.4f, 0.4f, 0.4f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.5f, 0.5f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.6f, 0.6f, 0.6f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.7f, 0.7f, 0.7f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.8f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.9f, 0.9f, 0.9f), float3(0,0,0), 0, 0, },
+    { Material::Metal, float3(0.1f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.2f, 0.2f, 0.2f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.3f, 0.3f, 0.3f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.4f, 0.4f, 0.4f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.5f, 0.5f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.6f, 0.6f, 0.6f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.7f, 0.7f, 0.7f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.8f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.9f, 0.9f, 0.9f), float3(0,0,0), 0, 0, },
+    { Material::Metal, float3(0.8f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.8f, 0.5f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.8f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.4f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.8f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.1f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.5f, 0.1f, 0.8f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.8f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.8f, 0.5f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.8f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.4f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.8f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.1f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.5f, 0.1f, 0.8f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.1f, 0.2f, 0.5f), float3(3,10,20), 0, 0 },
+#endif
+};
+
+static int s_EmissiveSpheres[kSphereCount];
+static int s_EmissiveSphereCount;
+
+static Camera s_Cam;
+
+const float kMinT = 0.001f;
+const float kMaxT = 1.0e7f;
+const int kMaxDepth = 10;
+
+
+bool HitWorld(const Ray& r, float tMin, float tMax, Hit& outHit, int& outID)
+{
+    outID = HitSpheres(r, s_SpheresSoA, tMin, tMax, outHit);
+    return outID != -1;
+}
+
+
+static bool Scatter(const Material& mat, const Ray& r_in, const Hit& rec, float3& attenuation, Ray& scattered, float3& outLightE, int& inoutRayCount, uint32_t& state)
+{
+    ZoneScoped;
+    outLightE = float3(0,0,0);
+    if (mat.type == Material::Lambert)
+    {
+        // random point on unit sphere that is tangent to the hit point
+        float3 target = rec.pos + rec.normal + RandomUnitVector(state);
+        scattered = Ray(rec.pos, normalize(target - rec.pos));
+        attenuation = mat.albedo;
+
+        // sample lights
+#if DO_LIGHT_SAMPLING
+        for (int j = 0; j < s_EmissiveSphereCount; ++j)
+        {
+            int i = s_EmissiveSpheres[j];
+            const Material& smat = s_SphereMats[i];
+            if (&mat == &smat)
+                continue; // skip self
+            const Sphere& s = s_Spheres[i];
+
+            // create a random direction towards sphere
+            // coord system for sampling: sw, su, sv
+            float3 sw = normalize(s.center - rec.pos);
+            float3 su = normalize(cross(fabs(sw.getX())>0.01f ? float3(0,1,0):float3(1,0,0), sw));
+            float3 sv = cross(sw, su);
+            // sample sphere by solid angle
+            float cosAMax = sqrtf(1.0f - s.radius*s.radius / sqLength(rec.pos-s.center));
+            float eps1 = RandomFloat01(state), eps2 = RandomFloat01(state);
+            float cosA = 1.0f - eps1 + eps1 * cosAMax;
+            float sinA = sqrtf(1.0f - cosA*cosA);
+            float phi = 2 * kPI * eps2;
+            float3 l = su * (cosf(phi) * sinA) + sv * (sinf(phi) * sinA) + sw * cosA;
+            //l = normalize(l); // NOTE(fg): This is already normalized, by construction.
+
+            // shoot shadow ray
+            Hit lightHit;
+            int hitID;
+            ++inoutRayCount;
+            if (HitWorld(Ray(rec.pos, l), kMinT, kMaxT, lightHit, hitID) && hitID == i)
+            {
+                float omega = 2 * kPI * (1-cosAMax);
+
+                float3 rdir = r_in.dir;
+                AssertUnit(rdir);
+                float3 nl = dot(rec.normal, rdir) < 0 ? rec.normal : -rec.normal;
+                outLightE += (mat.albedo * smat.emissive) * (std::max(0.0f, dot(l, nl)) * omega / kPI);
+            }
+        }
+#endif
+        return true;
+    }
+    else if (mat.type == Material::Metal)
+    {
+        AssertUnit(r_in.dir); AssertUnit(rec.normal);
+        float3 refl = reflect(r_in.dir, rec.normal);
+        // reflected ray, and random inside of sphere based on roughness
+        float roughness = mat.roughness;
+#if DO_MITSUBA_COMPARE
+        roughness = 0; // until we get better BRDF for metals
+#endif
+        scattered = Ray(rec.pos, normalize(refl + roughness*RandomInUnitSphere(state)));
+        attenuation = mat.albedo;
+        return dot(scattered.dir, rec.normal) > 0;
+    }
+    else if (mat.type == Material::Dielectric)
+    {
+        AssertUnit(r_in.dir); AssertUnit(rec.normal);
+        float3 outwardN;
+        float3 rdir = r_in.dir;
+        float3 refl = reflect(rdir, rec.normal);
+        float nint;
+        attenuation = float3(1,1,1);
+        float3 refr;
+        float reflProb;
+        float cosine;
+        if (dot(rdir, rec.normal) > 0)
+        {
+            outwardN = -rec.normal;
+            nint = mat.ri;
+            cosine = mat.ri * dot(rdir, rec.normal);
+        }
+        else
+        {
+            outwardN = rec.normal;
+            nint = 1.0f / mat.ri;
+            cosine = -dot(rdir, rec.normal);
+        }
+        if (refract(rdir, outwardN, nint, refr))
+        {
+            reflProb = schlick(cosine, mat.ri);
+        }
+        else
+        {
+            reflProb = 1;
+        }
+        if (RandomFloat01(state) < reflProb)
+            scattered = Ray(rec.pos, normalize(refl));
+        else
+            scattered = Ray(rec.pos, normalize(refr));
+    }
+    else
+    {
+        attenuation = float3(1,0,1);
+        return false;
+    }
+    return true;
+}
+
+static float3 Trace(const Ray& r, int depth, int& inoutRayCount, uint32_t& state, bool doMaterialE = true)
+{
+    ZoneScoped;
+    Hit rec;
+    int id = 0;
+    ++inoutRayCount;
+    if (HitWorld(r, kMinT, kMaxT, rec, id))
+    {
+        Ray scattered;
+        float3 attenuation;
+        float3 lightE;
+        const Material& mat = s_SphereMats[id];
+        float3 matE = mat.emissive;
+        if (depth < kMaxDepth && Scatter(mat, r, rec, attenuation, scattered, lightE, inoutRayCount, state))
+        {
+#if DO_LIGHT_SAMPLING
+            if (!doMaterialE) matE = float3(0,0,0); // don't add material emission if told so
+            // dor Lambert materials, we just did explicit light (emissive) sampling and already
+            // for their contribution, so if next ray bounce hits the light again, don't add
+            // emission
+            doMaterialE = (mat.type != Material::Lambert);
+#endif
+            return matE + lightE + attenuation * Trace(scattered, depth+1, inoutRayCount, state, doMaterialE);
+        }
+        else
+        {
+            return matE;
+        }
+    }
+    else
+    {
+        // sky
+#if DO_MITSUBA_COMPARE
+        return float3(0.15f,0.21f,0.3f); // easier compare with Mitsuba's constant environment light
+#else
+        float3 unitDir = r.dir;
+        float t = 0.5f*(unitDir.getY() + 1.0f);
+        return ((1.0f-t)*float3(1.0f, 1.0f, 1.0f) + t*float3(0.5f, 0.7f, 1.0f)) * 0.3f;
+#endif
+    }
+}
+
+#if CPU_CAN_DO_THREADS
+static enkiTaskScheduler* g_TS;
+#endif
+
+void InitializeTest()
+{
+    ZoneScoped;
+    #if CPU_CAN_DO_THREADS
+    g_TS = enkiNewTaskScheduler();
+    enkiInitTaskSchedulerNumThreads(g_TS, std::max<int>( 2, std::thread::hardware_concurrency() - 2));
+    #endif
+}
+
+void ShutdownTest()
+{
+    ZoneScoped;
+    #if CPU_CAN_DO_THREADS
+    enkiDeleteTaskScheduler(g_TS);
+    #endif
+}
+
+struct JobData
+{
+    float time;
+    int frameCount;
+    int screenWidth, screenHeight;
+    float* backbuffer;
+    Camera* cam;
+    std::atomic<int> rayCount;
+    unsigned testFlags;
+};
+
+static void TraceRowJob(uint32_t start, uint32_t end, uint32_t threadnum, void* data_)
+{
+    ZoneScoped;
+    JobData& data = *(JobData*)data_;
+    float* backbuffer = data.backbuffer + start * data.screenWidth * 4;
+    float invWidth = 1.0f / data.screenWidth;
+    float invHeight = 1.0f / data.screenHeight;
+    float lerpFac = float(data.frameCount) / float(data.frameCount+1);
+    if (data.testFlags & kFlagAnimate)
+        lerpFac *= DO_ANIMATE_SMOOTHING;
+    if (!(data.testFlags & kFlagProgressive))
+        lerpFac = 0;
+    int rayCount = 0;
+    for (uint32_t y = start; y < end; ++y)
+    {
+        uint32_t state = (y * 9781 + data.frameCount * 6271) | 1;
+        for (int x = 0; x < data.screenWidth; ++x)
+        {
+            float3 col(0, 0, 0);
+            for (int s = 0; s < DO_SAMPLES_PER_PIXEL; s++)
+            {
+                float u = float(x + RandomFloat01(state)) * invWidth;
+                float v = float(y + RandomFloat01(state)) * invHeight;
+                Ray r = data.cam->GetRay(u, v, state);
+                col += Trace(r, 0, rayCount, state);
+            }
+            col *= 1.0f / float(DO_SAMPLES_PER_PIXEL);
+
+            float3 prev(backbuffer[0], backbuffer[1], backbuffer[2]);
+            col = prev * lerpFac + col * (1-lerpFac);
+            col.store(backbuffer);
+            backbuffer += 4;
+        }
+    }
+    data.rayCount += rayCount;
+}
+
+void UpdateTest(float time, int frameCount, int screenWidth, int screenHeight, unsigned testFlags)
+{
+    ZoneScoped;
+    if (testFlags & kFlagAnimate)
+    {
+        s_Spheres[1].center.setY(cosf(time) + 1.0f);
+        s_Spheres[8].center.setZ(sinf(time)*0.3f);
+    }
+    float3 lookfrom(0, 2, 3);
+    float3 lookat(0, 0, 0);
+    float distToFocus = 3;
+#if DO_MITSUBA_COMPARE
+    float aperture = 0.0f;
+#else
+    float aperture = 0.1f;
+#endif
+#if DO_BIG_SCENE
+    aperture *= 0.2f;
+#endif
+
+    s_EmissiveSphereCount = 0;
+    for (int i = 0; i < kSphereCount; ++i)
+    {
+        Sphere& s = s_Spheres[i];
+        s.UpdateDerivedData();
+        s_SpheresSoA.centerX[i] = s.center.getX();
+        s_SpheresSoA.centerY[i] = s.center.getY();
+        s_SpheresSoA.centerZ[i] = s.center.getZ();
+        s_SpheresSoA.sqRadius[i] = s.radius * s.radius;
+        s_SpheresSoA.invRadius[i] = s.invRadius;
+
+        // Remember IDs of emissive spheres (light sources)
+        const Material& smat = s_SphereMats[i];
+        if (smat.emissive.getX() > 0 || smat.emissive.getY() > 0 || smat.emissive.getZ() > 0)
+        {
+            s_EmissiveSpheres[s_EmissiveSphereCount] = i;
+            s_EmissiveSphereCount++;
+        }
+    }
+
+    s_Cam = Camera(lookfrom, lookat, float3(0, 1, 0), 60, float(screenWidth) / float(screenHeight), aperture, distToFocus);
+}
+
+void DrawTest(float time, int frameCount, int screenWidth, int screenHeight, float* backbuffer, int& outRayCount, unsigned testFlags)
+{
+    ZoneScoped;
+    JobData args;
+    args.time = time;
+    args.frameCount = frameCount;
+    args.screenWidth = screenWidth;
+    args.screenHeight = screenHeight;
+    args.backbuffer = backbuffer;
+    args.cam = &s_Cam;
+    args.testFlags = testFlags;
+    args.rayCount = 0;
+
+    #if CPU_CAN_DO_THREADS
+    enkiTaskSet* task = enkiCreateTaskSet(g_TS, TraceRowJob);
+    bool threaded = true;
+    enkiAddTaskSetToPipeMinRange(g_TS, task, &args, screenHeight, threaded ? 4 : screenHeight);
+    enkiWaitForTaskSet(g_TS, task);
+    enkiDeleteTaskSet(task);
+    #else
+    TraceRowJob(0, screenHeight, 0, &args);
+    #endif
+
+    outRayCount = args.rayCount;
+}
+
+void GetObjectCount(int& outCount, int& outObjectSize, int& outMaterialSize, int& outCamSize)
+{
+    ZoneScoped;
+    outCount = kSphereCount;
+    outObjectSize = sizeof(Sphere);
+    outMaterialSize = sizeof(Material);
+    outCamSize = sizeof(Camera);
+}
+
+void GetSceneDesc(void* outObjects, void* outMaterials, void* outCam, void* outEmissives, int* outEmissiveCount)
+{
+    ZoneScoped;
+    memcpy(outObjects, s_Spheres, kSphereCount * sizeof(s_Spheres[0]));
+    memcpy(outMaterials, s_SphereMats, kSphereCount * sizeof(s_SphereMats[0]));
+    memcpy(outCam, &s_Cam, sizeof(s_Cam));
+    memcpy(outEmissives, s_EmissiveSpheres, s_EmissiveSphereCount * sizeof(s_EmissiveSpheres[0]));
+    *outEmissiveCount = s_EmissiveSphereCount;
+}

examples/ToyPathTracer/Source/Test.h

@@ -0,0 +1,17 @@
+#pragma once
+#include <stdint.h>
+
+enum TestFlags
+{
+    kFlagAnimate = (1 << 0),
+    kFlagProgressive = (1 << 1),
+};
+
+void InitializeTest();
+void ShutdownTest();
+
+void UpdateTest(float time, int frameCount, int screenWidth, int screenHeight, unsigned testFlags);
+void DrawTest(float time, int frameCount, int screenWidth, int screenHeight, float* backbuffer, int& outRayCount, unsigned testFlags);
+
+void GetObjectCount(int& outCount, int& outObjectSize, int& outMaterialSize, int& outCamSize);
+void GetSceneDesc(void* outObjects, void* outMaterials, void* outCam, void* outEmissives, int* outEmissiveCount);

examples/ToyPathTracer/Source/enkiTS/Atomics.h

@@ -0,0 +1,79 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+
+#ifdef _WIN32
+    #define WIN32_LEAN_AND_MEAN
+    #include <Windows.h>
+	#undef GetObject
+    #include <intrin.h>
+
+    extern "C" void _ReadWriteBarrier();
+    #pragma intrinsic(_ReadWriteBarrier)
+    #pragma intrinsic(_InterlockedCompareExchange)
+    #pragma intrinsic(_InterlockedExchangeAdd)
+
+    // Memory Barriers to prevent CPU and Compiler re-ordering
+    #define BASE_MEMORYBARRIER_ACQUIRE() _ReadWriteBarrier()
+    #define BASE_MEMORYBARRIER_RELEASE() _ReadWriteBarrier()
+    #define BASE_ALIGN(x) __declspec( align( x ) ) 
+
+#else
+    #define BASE_MEMORYBARRIER_ACQUIRE() __asm__ __volatile__("": : :"memory")  
+    #define BASE_MEMORYBARRIER_RELEASE() __asm__ __volatile__("": : :"memory")  
+	#define BASE_ALIGN(x)  __attribute__ ((aligned( x )))
+#endif
+
+namespace enki
+{
+    // Atomically performs: if( *pDest == compareWith ) { *pDest = swapTo; }
+    // returns old *pDest (so if successfull, returns compareWith)
+    inline uint32_t AtomicCompareAndSwap( volatile uint32_t* pDest, uint32_t swapTo, uint32_t compareWith )
+    {
+       #ifdef _WIN32
+			// assumes two's complement - unsigned / signed conversion leads to same bit pattern
+            return _InterlockedCompareExchange( (volatile long*)pDest,swapTo, compareWith );
+        #else
+            return __sync_val_compare_and_swap( pDest, compareWith, swapTo );
+        #endif      
+    }
+
+    inline uint64_t AtomicCompareAndSwap( volatile uint64_t* pDest, uint64_t swapTo, uint64_t compareWith )
+    {
+       #ifdef _WIN32
+			// assumes two's complement - unsigned / signed conversion leads to same bit pattern
+            return _InterlockedCompareExchange64( (__int64 volatile*)pDest, swapTo, compareWith );
+        #else
+            return __sync_val_compare_and_swap( pDest, compareWith, swapTo );
+        #endif      
+    }	
+
+    // Atomically performs: tmp = *pDest; *pDest += value; return tmp;
+    inline int32_t AtomicAdd( volatile int32_t* pDest, int32_t value )
+    {
+       #ifdef _WIN32
+            return _InterlockedExchangeAdd( (long*)pDest, value );
+        #else
+            return __sync_fetch_and_add( pDest, value );
+        #endif      
+    }
+
+}

examples/ToyPathTracer/Source/enkiTS/LockLessMultiReadPipe.h

@@ -0,0 +1,240 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+#include <assert.h>
+
+#include "Atomics.h"
+#include <string.h>
+
+
+namespace enki
+{
+    // LockLessMultiReadPipe - Single writer, multiple reader thread safe pipe using (semi) lockless programming
+    // Readers can only read from the back of the pipe
+    // The single writer can write to the front of the pipe, and read from both ends (a writer can be a reader)
+    // for many of the principles used here, see http://msdn.microsoft.com/en-us/library/windows/desktop/ee418650(v=vs.85).aspx
+    // Note: using log2 sizes so we do not need to clamp (multi-operation)
+    // T is the contained type
+    // Note this is not true lockless as the use of flags as a form of lock state.
+    template<uint8_t cSizeLog2, typename T> class LockLessMultiReadPipe
+    {
+    public:
+        LockLessMultiReadPipe();
+        ~LockLessMultiReadPipe() {}
+
+        // ReaderTryReadBack returns false if we were unable to read
+        // This is thread safe for both multiple readers and the writer
+        bool ReaderTryReadBack(   T* pOut );
+
+        // WriterTryReadFront returns false if we were unable to read
+        // This is thread safe for the single writer, but should not be called by readers
+        bool WriterTryReadFront(  T* pOut );
+
+        // WriterTryWriteFront returns false if we were unable to write
+        // This is thread safe for the single writer, but should not be called by readers
+        bool WriterTryWriteFront( const T& in );
+
+        // IsPipeEmpty() is a utility function, not intended for general use
+        // Should only be used very prudently.
+        bool IsPipeEmpty() const
+        {
+            return 0 == m_WriteIndex - m_ReadCount;
+        }
+
+        void Clear()
+        {
+            m_WriteIndex = 0;
+            m_ReadIndex = 0;
+            m_ReadCount = 0;
+            memset( (void*)m_Flags, 0, sizeof( m_Flags ) );
+        }
+
+    private:
+        const static uint32_t           ms_cSize        = ( 1 << cSizeLog2 );
+        const static uint32_t           ms_cIndexMask   = ms_cSize - 1;
+        const static uint32_t           FLAG_INVALID    = 0xFFFFFFFF; // 32bit for CAS
+        const static uint32_t           FLAG_CAN_WRITE  = 0x00000000; // 32bit for CAS
+        const static uint32_t           FLAG_CAN_READ   = 0x11111111; // 32bit for CAS
+
+        T                               m_Buffer[ ms_cSize ];
+
+        // read and write indexes allow fast access to the pipe, but actual access
+        // controlled by the access flags. 
+        volatile uint32_t BASE_ALIGN(4) m_WriteIndex;
+        volatile uint32_t BASE_ALIGN(4) m_ReadCount;
+        volatile uint32_t               m_Flags[  ms_cSize ];
+        volatile uint32_t BASE_ALIGN(4) m_ReadIndex;
+    };
+
+    template<uint8_t cSizeLog2, typename T> inline
+        LockLessMultiReadPipe<cSizeLog2,T>::LockLessMultiReadPipe()
+        : m_WriteIndex(0)
+        , m_ReadIndex(0)
+        , m_ReadCount(0)
+    {
+        assert( cSizeLog2 < 32 );
+        memset( (void*)m_Flags, 0, sizeof( m_Flags ) );
+    }
+
+    template<uint8_t cSizeLog2, typename T> inline
+        bool LockLessMultiReadPipe<cSizeLog2,T>::ReaderTryReadBack(   T* pOut )
+    {
+
+        uint32_t actualReadIndex;
+
+        uint32_t readCount  = m_ReadCount;
+
+        // We get hold of read index for consistency,
+        // and do first pass starting at read count
+        uint32_t readIndexToUse  = readCount;
+
+
+        while(true)
+        {
+
+            uint32_t writeIndex = m_WriteIndex;
+            // power of two sizes ensures we can use a simple calc without modulus
+            uint32_t numInPipe = writeIndex - readCount;
+            if( 0 == numInPipe )
+            {
+                return false;
+            }
+            if( readIndexToUse >= writeIndex )
+            {
+                // move back to start
+                readIndexToUse = m_ReadIndex;
+            }
+
+
+            // power of two sizes ensures we can perform AND for a modulus
+            actualReadIndex    = readIndexToUse & ms_cIndexMask;
+
+            // Multiple potential readers mean we should check if the data is valid,
+            // using an atomic compare exchange
+            uint32_t previous = AtomicCompareAndSwap( &m_Flags[  actualReadIndex ], FLAG_INVALID, FLAG_CAN_READ );
+            if( FLAG_CAN_READ == previous )
+            {
+                break;
+            }
+            ++readIndexToUse;
+
+            //update known readcount
+            readCount  = m_ReadCount;
+        }
+
+        // we update the read index using an atomic add, as we've only read one piece of data.
+        // this ensure consistency of the read index, and the above loop ensures readers
+        // only read from unread data
+        AtomicAdd(  (volatile int32_t*)&m_ReadCount, 1 );
+
+        BASE_MEMORYBARRIER_ACQUIRE();
+        // now read data, ensuring we do so after above reads & CAS
+        *pOut = m_Buffer[ actualReadIndex ];
+
+        m_Flags[  actualReadIndex ] = FLAG_CAN_WRITE;
+
+        return true;
+    }
+
+    template<uint8_t cSizeLog2, typename T> inline
+        bool LockLessMultiReadPipe<cSizeLog2,T>::WriterTryReadFront(  T* pOut )
+    {
+        uint32_t writeIndex = m_WriteIndex;
+        uint32_t frontReadIndex  = writeIndex;
+
+        // Multiple potential readers mean we should check if the data is valid,
+        // using an atomic compare exchange - which acts as a form of lock (so not quite lockless really).
+        uint32_t previous = FLAG_INVALID;
+        uint32_t actualReadIndex = 0;
+        while( true )
+        {
+            // power of two sizes ensures we can use a simple calc without modulus
+            uint32_t readCount = m_ReadCount;
+            uint32_t numInPipe = writeIndex - readCount;
+            if( 0 == numInPipe || 0 == frontReadIndex )
+            {
+                // frontReadIndex can get to 0 here if that item was just being read by another thread.
+                m_ReadIndex = readCount;
+                return false;
+            }
+            --frontReadIndex;
+            actualReadIndex = frontReadIndex & ms_cIndexMask;
+            previous = AtomicCompareAndSwap( &m_Flags[  actualReadIndex ], FLAG_INVALID, FLAG_CAN_READ );
+            if( FLAG_CAN_READ == previous )
+            {
+                break;
+            }
+            else if( m_ReadIndex >= frontReadIndex  )
+            {
+                return false;
+            }
+        }
+
+        // now read data, ensuring we do so after above reads & CAS
+        *pOut = m_Buffer[ actualReadIndex ];
+
+        m_Flags[  actualReadIndex ] = FLAG_CAN_WRITE;
+
+        BASE_MEMORYBARRIER_RELEASE();
+
+        // 32-bit aligned stores are atomic, and writer owns the write index
+        // we only move one back as this is as many as we have read, not where we have read from.
+        --m_WriteIndex;
+        return true;
+    }
+
+
+    template<uint8_t cSizeLog2, typename T> inline
+        bool LockLessMultiReadPipe<cSizeLog2,T>::WriterTryWriteFront( const T& in )
+    {
+        // The writer 'owns' the write index, and readers can only reduce
+        // the amount of data in the pipe.
+        // We get hold of both values for consistency and to reduce false sharing
+        // impacting more than one access
+        uint32_t writeIndex = m_WriteIndex;
+
+
+        // power of two sizes ensures we can perform AND for a modulus
+        uint32_t actualWriteIndex    = writeIndex & ms_cIndexMask;
+
+        // a reader may still be reading this item, as there are multiple readers
+        if( m_Flags[ actualWriteIndex ] != FLAG_CAN_WRITE ) 
+        {
+            return false; // still being read, so have caught up with tail. 
+        }
+
+
+        // as we are the only writer we can update the data without atomics
+        //  whilst the write index has not been updated
+        m_Buffer[ actualWriteIndex ] = in;
+        m_Flags[  actualWriteIndex ] = FLAG_CAN_READ;
+
+        // We need to ensure the above writes occur prior to updating the write index,
+        // otherwise another thread might read before it's finished
+        BASE_MEMORYBARRIER_RELEASE();
+
+        // 32-bit aligned stores are atomic, and the writer controls the write index
+        ++writeIndex;
+        m_WriteIndex = writeIndex;
+        return true;
+    }
+
+}

examples/ToyPathTracer/Source/enkiTS/TaskScheduler.cpp

@@ -0,0 +1,437 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#include <assert.h>
+
+#include "TaskScheduler.h"
+#include "LockLessMultiReadPipe.h"
+
+
+
+using namespace enki;
+
+
+static const uint32_t PIPESIZE_LOG2              = 8;
+static const uint32_t SPIN_COUNT                 = 100;
+static const uint32_t SPIN_BACKOFF_MULTIPLIER    = 10;
+static const uint32_t MAX_NUM_INITIAL_PARTITIONS = 8;
+
+// each software thread gets it's own copy of gtl_threadNum, so this is safe to use as a static variable
+static THREAD_LOCAL uint32_t                             gtl_threadNum       = 0;
+
+namespace enki 
+{
+	struct SubTaskSet
+	{
+		ITaskSet*           pTask;
+		TaskSetPartition    partition;
+	};
+
+	// we derive class TaskPipe rather than typedef to get forward declaration working easily
+	class TaskPipe : public LockLessMultiReadPipe<PIPESIZE_LOG2,enki::SubTaskSet> {};
+
+	struct ThreadArgs
+	{
+		uint32_t		threadNum;
+		TaskScheduler*  pTaskScheduler;
+	};
+}
+
+namespace
+{
+	SubTaskSet       SplitTask( SubTaskSet& subTask_, uint32_t rangeToSplit_ )
+	{
+		SubTaskSet splitTask = subTask_;
+		uint32_t rangeLeft = subTask_.partition.end - subTask_.partition.start;
+
+        if( rangeToSplit_ > rangeLeft )
+        {
+            rangeToSplit_ = rangeLeft;
+        }
+        splitTask.partition.end = subTask_.partition.start + rangeToSplit_;
+		subTask_.partition.start = splitTask.partition.end;
+		return splitTask;
+	}
+
+	#if defined _WIN32
+		#if defined _M_IX86  || defined _M_X64
+			#pragma intrinsic(_mm_pause)
+			inline void Pause() { _mm_pause(); }
+		#endif
+	#elif defined __i386__ || defined __x86_64__
+		inline void Pause() { __asm__ __volatile__("pause;"); }
+	#else
+		inline void Pause() { ;} // may have NOP or yield equiv
+	#endif
+}
+
+
+static void SafeCallback(ProfilerCallbackFunc func_, uint32_t threadnum_)
+{
+	if( func_ )
+	{
+		func_(threadnum_);
+	}
+}
+
+ProfilerCallbacks* TaskScheduler::GetProfilerCallbacks()
+{
+	return &m_ProfilerCallbacks;
+}
+
+THREADFUNC_DECL TaskScheduler::TaskingThreadFunction( void* pArgs )
+{
+	ThreadArgs args					= *(ThreadArgs*)pArgs;
+	uint32_t threadNum				= args.threadNum;
+	TaskScheduler*  pTS				= args.pTaskScheduler;
+    gtl_threadNum      = threadNum;
+
+	SafeCallback( pTS->m_ProfilerCallbacks.threadStart, threadNum );
+    
+    uint32_t spinCount = 0;
+	uint32_t hintPipeToCheck_io = threadNum + 1;	// does not need to be clamped.
+    while( pTS->m_bRunning )
+    {
+        if(!pTS->TryRunTask( threadNum, hintPipeToCheck_io ) )
+        {
+            // no tasks, will spin then wait
+            ++spinCount;
+            if( spinCount > SPIN_COUNT )
+            {
+				pTS->WaitForTasks( threadNum );
+				spinCount = 0;
+            }
+			else
+			{
+				uint32_t spinBackoffCount = spinCount * SPIN_BACKOFF_MULTIPLIER;
+				while( spinBackoffCount )
+				{
+					Pause();
+					--spinBackoffCount;
+				}
+			}
+        }
+        else
+        {
+            spinCount = 0;
+        }
+    }
+
+    AtomicAdd( &pTS->m_NumThreadsRunning, -1 );
+	SafeCallback( pTS->m_ProfilerCallbacks.threadStop, threadNum );
+
+    return 0;
+}
+
+
+void TaskScheduler::StartThreads()
+{
+    if( m_bHaveThreads )
+    {
+        return;
+    }
+    m_bRunning = true;
+
+    SemaphoreCreate( m_NewTaskSemaphore );
+
+    // we create one less thread than m_NumThreads as the main thread counts as one
+    m_pThreadNumStore = new ThreadArgs[m_NumThreads];
+    m_pThreadIDs      = new threadid_t[m_NumThreads];
+	m_pThreadNumStore[0].threadNum      = 0;
+	m_pThreadNumStore[0].pTaskScheduler = this;
+	m_pThreadIDs[0] = 0;
+    m_NumThreadsWaiting = 0;
+    m_NumThreadsRunning = 1;// acount for main thread
+    for( uint32_t thread = 1; thread < m_NumThreads; ++thread )
+    {
+		m_pThreadNumStore[thread].threadNum      = thread;
+		m_pThreadNumStore[thread].pTaskScheduler = this;
+        ThreadCreate( &m_pThreadIDs[thread], TaskingThreadFunction, &m_pThreadNumStore[thread] );
+        ++m_NumThreadsRunning;
+    }
+
+    // ensure we have sufficient tasks to equally fill either all threads including main
+    // or just the threads we've launched, this is outside the firstinit as we want to be able
+    // to runtime change it
+	if( 1 == m_NumThreads )
+	{
+		m_NumPartitions = 1;
+		m_NumInitialPartitions = 1;
+	}
+	else
+	{
+		m_NumPartitions = m_NumThreads * (m_NumThreads - 1);
+		m_NumInitialPartitions = m_NumThreads - 1;
+		if( m_NumInitialPartitions > MAX_NUM_INITIAL_PARTITIONS )
+		{
+			m_NumInitialPartitions = MAX_NUM_INITIAL_PARTITIONS;
+		}
+	}
+
+    m_bHaveThreads = true;
+}
+
+void TaskScheduler::StopThreads( bool bWait_ )
+{
+    if( m_bHaveThreads )
+    {
+        // wait for them threads quit before deleting data
+        m_bRunning = false;
+        while( bWait_ && m_NumThreadsRunning > 1 )
+        {
+            // keep firing event to ensure all threads pick up state of m_bRunning
+            SemaphoreSignal( m_NewTaskSemaphore, m_NumThreadsRunning );
+        }
+
+        for( uint32_t thread = 1; thread < m_NumThreads; ++thread )
+        {
+            ThreadTerminate( m_pThreadIDs[thread] );
+        }
+
+		m_NumThreads = 0;
+        delete[] m_pThreadNumStore;
+        delete[] m_pThreadIDs;
+        m_pThreadNumStore = 0;
+        m_pThreadIDs = 0;
+        SemaphoreClose( m_NewTaskSemaphore );
+
+        m_bHaveThreads = false;
+		m_NumThreadsWaiting = 0;
+		m_NumThreadsRunning = 0;
+    }
+}
+
+bool TaskScheduler::TryRunTask( uint32_t threadNum, uint32_t& hintPipeToCheck_io_ )
+{
+    // check for tasks
+    SubTaskSet subTask;
+    bool bHaveTask = m_pPipesPerThread[ threadNum ].WriterTryReadFront( &subTask );
+
+	uint32_t threadToCheck = hintPipeToCheck_io_;
+	uint32_t checkCount = 0;
+    while( !bHaveTask && checkCount < m_NumThreads )
+    {
+		threadToCheck = ( hintPipeToCheck_io_ + checkCount ) % m_NumThreads;
+		if( threadToCheck != threadNum )
+		{
+			bHaveTask = m_pPipesPerThread[ threadToCheck ].ReaderTryReadBack( &subTask );
+		}
+		++checkCount;
+    }
+        
+    if( bHaveTask )
+    {
+		// update hint, will preserve value unless actually got task from another thread.
+		hintPipeToCheck_io_ = threadToCheck;
+
+		uint32_t partitionSize = subTask.partition.end - subTask.partition.start;
+		if( subTask.pTask->m_RangeToRun < partitionSize )
+		{
+			SubTaskSet taskToRun = SplitTask( subTask, subTask.pTask->m_RangeToRun );
+			SplitAndAddTask( gtl_threadNum, subTask, subTask.pTask->m_RangeToRun, 0 );
+			taskToRun.pTask->ExecuteRange( taskToRun.partition, threadNum );
+			AtomicAdd( &taskToRun.pTask->m_RunningCount, -1 );
+		}
+		else
+		{
+
+			// the task has already been divided up by AddTaskSetToPipe, so just run it
+			subTask.pTask->ExecuteRange( subTask.partition, threadNum );
+			AtomicAdd( &subTask.pTask->m_RunningCount, -1 );
+		}
+    }
+
+    return bHaveTask;
+
+}
+
+void TaskScheduler::WaitForTasks( uint32_t threadNum )
+{
+	// We incrememt the number of threads waiting here in order
+	// to ensure that the check for tasks occurs after the increment
+	// to prevent a task being added after a check, then the thread waiting.
+	// This will occasionally result in threads being mistakenly awoken,
+	// but they will then go back to sleep.
+	AtomicAdd( &m_NumThreadsWaiting, 1 );
+
+    bool bHaveTasks = false;
+    for( uint32_t thread = 0; thread < m_NumThreads; ++thread )
+    {
+        if( !m_pPipesPerThread[ thread ].IsPipeEmpty() )
+        {
+            bHaveTasks = true;
+            break;
+        }
+    }
+    if( !bHaveTasks )
+    {
+        SafeCallback( m_ProfilerCallbacks.waitStart, threadNum );
+        SemaphoreWait( m_NewTaskSemaphore );
+        SafeCallback( m_ProfilerCallbacks.waitStop, threadNum );
+    }
+
+    int32_t prev = AtomicAdd( &m_NumThreadsWaiting, -1 );
+    assert( prev != 0 );
+}
+
+void TaskScheduler::WakeThreads()
+{
+	SemaphoreSignal( m_NewTaskSemaphore, m_NumThreadsWaiting );
+}
+
+void TaskScheduler::SplitAndAddTask( uint32_t threadNum_, SubTaskSet subTask_,
+	uint32_t rangeToSplit_, int32_t runningCountOffset_ )
+{
+    int32_t numAdded = 0;
+    while( subTask_.partition.start != subTask_.partition.end )
+    {
+        SubTaskSet taskToAdd = SplitTask( subTask_, rangeToSplit_ );
+
+        // add the partition to the pipe
+        ++numAdded;
+        if( !m_pPipesPerThread[ gtl_threadNum ].WriterTryWriteFront( taskToAdd ) )
+        {
+			if( numAdded > 1 )
+			{
+				WakeThreads();
+			}
+			// alter range to run the appropriate fraction
+			if( taskToAdd.pTask->m_RangeToRun < rangeToSplit_ )
+			{
+				taskToAdd.partition.end = taskToAdd.partition.start + taskToAdd.pTask->m_RangeToRun;
+				subTask_.partition.start = taskToAdd.partition.end;
+			}
+            taskToAdd.pTask->ExecuteRange( taskToAdd.partition, threadNum_ );
+            --numAdded;
+        }
+    }
+
+    // increment running count by number added
+    AtomicAdd( &subTask_.pTask->m_RunningCount, numAdded + runningCountOffset_ );
+
+	WakeThreads();
+}
+
+void    TaskScheduler::AddTaskSetToPipe( ITaskSet* pTaskSet )
+{
+	// set running count to -1 to guarantee it won't be found complete until all subtasks added
+    pTaskSet->m_RunningCount = -1;
+
+    // divide task up and add to pipe
+    pTaskSet->m_RangeToRun = pTaskSet->m_SetSize / m_NumPartitions;
+    if( pTaskSet->m_RangeToRun < pTaskSet->m_MinRange ) { pTaskSet->m_RangeToRun = pTaskSet->m_MinRange; }
+
+	uint32_t rangeToSplit = pTaskSet->m_SetSize / m_NumInitialPartitions;
+	if( rangeToSplit < pTaskSet->m_MinRange ) { rangeToSplit = pTaskSet->m_MinRange; }
+
+    SubTaskSet subTask;
+    subTask.pTask = pTaskSet;
+    subTask.partition.start = 0;
+    subTask.partition.end = pTaskSet->m_SetSize;
+	SplitAndAddTask( gtl_threadNum, subTask, rangeToSplit, 1 );
+}
+
+void    TaskScheduler::WaitforTaskSet( const ITaskSet* pTaskSet )
+{
+	uint32_t hintPipeToCheck_io = gtl_threadNum + 1;	// does not need to be clamped.
+	if( pTaskSet )
+	{
+		while( pTaskSet->m_RunningCount )
+		{
+			TryRunTask( gtl_threadNum, hintPipeToCheck_io );
+			// should add a spin then wait for task completion event.
+		}
+	}
+	else
+	{
+			TryRunTask( gtl_threadNum, hintPipeToCheck_io );
+	}
+}
+
+void    TaskScheduler::WaitforAll()
+{
+    bool bHaveTasks = true;
+ 	uint32_t hintPipeToCheck_io = gtl_threadNum  + 1;	// does not need to be clamped.
+	int32_t threadsRunning = m_NumThreadsRunning - 1;
+    while( bHaveTasks || m_NumThreadsWaiting < threadsRunning )
+    {
+        TryRunTask( gtl_threadNum, hintPipeToCheck_io );
+        bHaveTasks = false;
+        for( uint32_t thread = 0; thread < m_NumThreads; ++thread )
+        {
+            if( !m_pPipesPerThread[ thread ].IsPipeEmpty() )
+            {
+                bHaveTasks = true;
+                break;
+            }
+        }
+     }
+}
+
+void    TaskScheduler::WaitforAllAndShutdown()
+{
+    WaitforAll();
+    StopThreads(true);
+	delete[] m_pPipesPerThread;
+    m_pPipesPerThread = 0;
+}
+
+uint32_t        TaskScheduler::GetNumTaskThreads() const
+{
+    return m_NumThreads;
+}
+
+TaskScheduler::TaskScheduler()
+		: m_pPipesPerThread(NULL)
+		, m_NumThreads(0)
+		, m_pThreadNumStore(NULL)
+		, m_pThreadIDs(NULL)
+		, m_bRunning(false)
+		, m_NumThreadsRunning(0)
+		, m_NumThreadsWaiting(0)
+		, m_NumPartitions(0)
+		, m_bHaveThreads(false)
+{
+	memset(&m_ProfilerCallbacks, 0, sizeof(m_ProfilerCallbacks));
+}
+
+TaskScheduler::~TaskScheduler()
+{
+    StopThreads( true ); // Stops threads, waiting for them.
+
+    delete[] m_pPipesPerThread;
+    m_pPipesPerThread = 0;
+}
+
+void    TaskScheduler::Initialize( uint32_t numThreads_ )
+{
+	assert( numThreads_ );
+    StopThreads( true ); // Stops threads, waiting for them.
+    delete[] m_pPipesPerThread;
+
+	m_NumThreads = numThreads_;
+
+    m_pPipesPerThread = new TaskPipe[ m_NumThreads ];
+
+    StartThreads();
+}
+
+void   TaskScheduler::Initialize()
+{
+	Initialize( GetNumHardwareThreads() );
+}

examples/ToyPathTracer/Source/enkiTS/TaskScheduler.h

@@ -0,0 +1,177 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+#include "Threads.h"
+
+namespace enki
+{
+
+	struct TaskSetPartition
+	{
+		uint32_t start;
+		uint32_t end;
+	};
+
+	class  TaskScheduler;
+	class  TaskPipe;
+	struct ThreadArgs;
+	struct SubTaskSet;
+
+	// Subclass ITaskSet to create tasks.
+	// TaskSets can be re-used, but check
+	class ITaskSet
+	{
+	public:
+        ITaskSet()
+            : m_SetSize(1)
+			, m_MinRange(1)
+            , m_RunningCount(0)
+			, m_RangeToRun(1)
+        {}
+
+        ITaskSet( uint32_t setSize_ )
+            : m_SetSize( setSize_ )
+			, m_MinRange(1)
+            , m_RunningCount(0)
+			, m_RangeToRun(1)
+        {}
+
+		ITaskSet( uint32_t setSize_, uint32_t minRange_ )
+            : m_SetSize( setSize_ )
+			, m_MinRange( minRange_ )
+            , m_RunningCount(0)
+			, m_RangeToRun(minRange_)
+        {}
+
+		// Execute range should be overloaded to process tasks. It will be called with a
+		// range_ where range.start >= 0; range.start < range.end; and range.end < m_SetSize;
+		// The range values should be mapped so that linearly processing them in order is cache friendly
+		// i.e. neighbouring values should be close together.
+		// threadnum should not be used for changing processing of data, it's intended purpose
+		// is to allow per-thread data buckets for output.
+		virtual void            ExecuteRange( TaskSetPartition range, uint32_t threadnum  ) = 0;
+
+		// Size of set - usually the number of data items to be processed, see ExecuteRange. Defaults to 1
+		uint32_t                m_SetSize;
+
+		// Minimum size of of TaskSetPartition range when splitting a task set into partitions.
+		// This should be set to a value which results in computation effort of at least 10k
+		// clock cycles to minimize tast scheduler overhead.
+		// NOTE: The last partition will be smaller than m_MinRange if m_SetSize is not a multiple
+		// of m_MinRange.
+		// Also known as grain size in literature.
+		uint32_t                m_MinRange;
+
+		bool                    GetIsComplete()
+		{
+			return 0 == m_RunningCount;
+		}
+	private:
+		friend class           TaskScheduler;
+		volatile int32_t        m_RunningCount;
+		uint32_t                m_RangeToRun;
+	};
+
+	// TaskScheduler implements several callbacks intended for profilers
+	typedef void (*ProfilerCallbackFunc)( uint32_t threadnum_ );
+	struct ProfilerCallbacks
+	{
+		ProfilerCallbackFunc threadStart;
+		ProfilerCallbackFunc threadStop;
+		ProfilerCallbackFunc waitStart;
+		ProfilerCallbackFunc waitStop;
+	};
+
+	class TaskScheduler
+	{
+	public:
+		TaskScheduler();
+		~TaskScheduler();
+
+		// Call either Initialize() or Initialize( numThreads_ ) before adding tasks.
+
+		// Initialize() will create GetNumHardwareThreads()-1 threads, which is
+		// sufficient to fill the system when including the main thread.
+		// Initialize can be called multiple times - it will wait for completion
+		// before re-initializing.
+		void			Initialize();
+
+		// Initialize( numThreads_ ) - numThreads_ (must be > 0)
+		// will create numThreads_-1 threads, as thread 0 is
+		// the thread on which the initialize was called.
+		void			Initialize( uint32_t numThreads_ );
+
+
+		// Adds the TaskSet to pipe and returns if the pipe is not full.
+		// If the pipe is full, pTaskSet is run.
+		// should only be called from main thread, or within a task
+		void            AddTaskSetToPipe( ITaskSet* pTaskSet );
+
+		// Runs the TaskSets in pipe until true == pTaskSet->GetIsComplete();
+		// should only be called from thread which created the taskscheduler , or within a task
+		// if called with 0 it will try to run tasks, and return if none available.
+		void            WaitforTaskSet( const ITaskSet* pTaskSet );
+
+		// Waits for all task sets to complete - not guaranteed to work unless we know we
+		// are in a situation where tasks aren't being continuosly added.
+		void            WaitforAll();
+
+		// Waits for all task sets to complete and shutdown threads - not guaranteed to work unless we know we
+		// are in a situation where tasks aren't being continuosly added.
+		void            WaitforAllAndShutdown();
+
+		// Returns the number of threads created for running tasks + 1
+		// to account for the main thread.
+		uint32_t        GetNumTaskThreads() const;
+
+		// Returns the ProfilerCallbacks structure so that it can be modified to
+		// set the callbacks.
+		ProfilerCallbacks* GetProfilerCallbacks();
+
+	private:
+		static THREADFUNC_DECL  TaskingThreadFunction( void* pArgs );
+        void             WaitForTasks( uint32_t threadNum );
+		bool             TryRunTask( uint32_t threadNum, uint32_t& hintPipeToCheck_io_ );
+		void             StartThreads();
+		void             StopThreads( bool bWait_ );
+		void             SplitAndAddTask( uint32_t threadNum_, SubTaskSet subTask_,
+										  uint32_t rangeToSplit_, int32_t runningCountOffset_ );
+		void             WakeThreads();
+
+		TaskPipe*                                                m_pPipesPerThread;
+
+		uint32_t                                                 m_NumThreads;
+		ThreadArgs*                                              m_pThreadNumStore;
+		threadid_t*                                              m_pThreadIDs;
+		volatile bool                                            m_bRunning;
+		volatile int32_t                                         m_NumThreadsRunning;
+		volatile int32_t                                         m_NumThreadsWaiting;
+		uint32_t                                                 m_NumPartitions;
+		uint32_t                                                 m_NumInitialPartitions;
+		semaphoreid_t                                            m_NewTaskSemaphore;
+		bool                                                     m_bHaveThreads;
+		ProfilerCallbacks										 m_ProfilerCallbacks;
+
+		TaskScheduler( const TaskScheduler& nocopy );
+		TaskScheduler& operator=( const TaskScheduler& nocopy );
+	};
+
+}

examples/ToyPathTracer/Source/enkiTS/TaskScheduler_c.cpp

@@ -0,0 +1,122 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#include "TaskScheduler_c.h"
+#include "TaskScheduler.h"
+
+#include <assert.h>
+
+using namespace enki;
+
+struct enkiTaskScheduler : TaskScheduler
+{
+};
+
+struct enkiTaskSet : ITaskSet
+{
+	enkiTaskSet( enkiTaskExecuteRange taskFun_ ) : taskFun(taskFun_), pArgs(NULL) {}
+
+	virtual void ExecuteRange( TaskSetPartition range, uint32_t threadnum  )
+	{
+		taskFun( range.start, range.end, threadnum, pArgs );
+	}
+
+	enkiTaskExecuteRange taskFun;
+	void* pArgs;
+};
+
+enkiTaskScheduler*	enkiNewTaskScheduler()
+{
+	enkiTaskScheduler* pETS = new enkiTaskScheduler();
+    return pETS;
+}
+
+void	            enkiInitTaskScheduler(  enkiTaskScheduler* pETS_ )
+{
+    pETS_->Initialize();
+}
+
+void	            enkiInitTaskSchedulerNumThreads(  enkiTaskScheduler* pETS_, uint32_t numThreads_ )
+{
+    pETS_->Initialize( numThreads_ );
+}
+
+void				enkiDeleteTaskScheduler( enkiTaskScheduler* pETS_ )
+{
+	delete pETS_;
+}
+
+enkiTaskSet*		enkiCreateTaskSet( enkiTaskScheduler* pETS_, enkiTaskExecuteRange taskFunc_  )
+{
+	return new enkiTaskSet( taskFunc_ );
+}
+
+void                enkiDeleteTaskSet( enkiTaskSet* pTaskSet_ )
+{
+	delete pTaskSet_;
+}
+
+void				enkiAddTaskSetToPipe( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_, void* pArgs_, uint32_t setSize_ )
+{
+	assert( pTaskSet_ );
+	assert( pTaskSet_->taskFun );
+
+	pTaskSet_->m_SetSize = setSize_;
+	pTaskSet_->pArgs = pArgs_;
+	pETS_->AddTaskSetToPipe( pTaskSet_ );
+}
+
+void enkiAddTaskSetToPipeMinRange(enkiTaskScheduler * pETS_, enkiTaskSet * pTaskSet_, void * pArgs_, uint32_t setSize_, uint32_t minRange_)
+{
+	assert( pTaskSet_ );
+	assert( pTaskSet_->taskFun );
+
+	pTaskSet_->m_SetSize = setSize_;
+	pTaskSet_->m_MinRange = minRange_;
+	pTaskSet_->pArgs = pArgs_;
+	pETS_->AddTaskSetToPipe( pTaskSet_ );
+}
+
+int				enkiIsTaskSetComplete( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ )
+{
+	assert( pTaskSet_ );
+	return ( pTaskSet_->GetIsComplete() ) ? 1 : 0;
+}
+
+void				enkiWaitForTaskSet( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ )
+{
+	pETS_->WaitforTaskSet( pTaskSet_ );
+}
+
+void				enkiWaitForAll( enkiTaskScheduler* pETS_ )
+{
+	pETS_->WaitforAll();
+}
+
+
+uint32_t			enkiGetNumTaskThreads( enkiTaskScheduler* pETS_ )
+{
+	return pETS_->GetNumTaskThreads();
+}
+
+enkiProfilerCallbacks*	enkiGetProfilerCallbacks( enkiTaskScheduler* pETS_ )
+{
+    assert( sizeof(enkiProfilerCallbacks) == sizeof(enki::ProfilerCallbacks) );
+    return (enkiProfilerCallbacks*)pETS_->GetProfilerCallbacks();
+}
+

examples/ToyPathTracer/Source/enkiTS/TaskScheduler_c.h

@@ -0,0 +1,104 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+
+typedef struct enkiTaskScheduler enkiTaskScheduler;
+typedef struct enkiTaskSet		 enkiTaskSet;
+
+typedef void (* enkiTaskExecuteRange)( uint32_t start_, uint32_t end, uint32_t threadnum_, void* pArgs_ );
+
+
+// Create a new task scheduler
+enkiTaskScheduler*	enkiNewTaskScheduler();
+
+// Initialize task scheduler - will create GetNumHardwareThreads()-1 threads, which is
+// sufficient to fill the system when including the main thread.
+// Initialize can be called multiple times - it will wait for completion
+// before re-initializing.
+void	            enkiInitTaskScheduler(  enkiTaskScheduler* pETS_ );
+
+// Initialize a task scheduler with numThreads_ (must be > 0)
+// will create numThreads_-1 threads, as thread 0 is
+// the thread on which the initialize was called.
+void	            enkiInitTaskSchedulerNumThreads(  enkiTaskScheduler* pETS_, uint32_t numThreads_ );
+
+
+// Delete a task scheduler
+void				enkiDeleteTaskScheduler( enkiTaskScheduler* pETS_ );
+
+// Create a task set.
+enkiTaskSet*		enkiCreateTaskSet( enkiTaskScheduler* pETS_, enkiTaskExecuteRange taskFunc_  );
+
+// Delete a task set.
+void                enkiDeleteTaskSet( enkiTaskSet* pTaskSet_ );
+
+// Schedule the task
+void				enkiAddTaskSetToPipe( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_,
+										   void* pArgs_, uint32_t setSize_ );
+
+// Schedule the task with a minimum range.
+// This should be set to a value which results in computation effort of at least 10k
+// clock cycles to minimize tast scheduler overhead.
+// NOTE: The last partition will be smaller than m_MinRange if m_SetSize is not a multiple
+// of m_MinRange.
+// Also known as grain size in literature.
+void				enkiAddTaskSetToPipeMinRange( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_,
+												  void* pArgs_, uint32_t setSize_, uint32_t minRange_ );
+
+
+// Check if TaskSet is complete. Doesn't wait. Returns 1 if complete, 0 if not.
+int					enkiIsTaskSetComplete( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ );
+
+
+// Wait for a given task.
+// should only be called from thread which created the taskscheduler , or within a task
+// if called with 0 it will try to run tasks, and return if none available.
+void				enkiWaitForTaskSet( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ );
+
+
+// Waits for all task sets to complete - not guaranteed to work unless we know we
+// are in a situation where tasks aren't being continuosly added.
+void				enkiWaitForAll( enkiTaskScheduler* pETS_ );
+
+
+// get number of threads
+uint32_t			enkiGetNumTaskThreads( enkiTaskScheduler* pETS_ );
+
+// TaskScheduler implements several callbacks intended for profilers
+typedef void (*enkiProfilerCallbackFunc)( uint32_t threadnum_ );
+struct enkiProfilerCallbacks
+{
+    enkiProfilerCallbackFunc threadStart;
+    enkiProfilerCallbackFunc threadStop;
+    enkiProfilerCallbackFunc waitStart;
+    enkiProfilerCallbackFunc waitStop;
+};
+
+// Get the callback structure so it can be set 
+struct enkiProfilerCallbacks*	enkiGetProfilerCallbacks( enkiTaskScheduler* pETS_ );
+
+#ifdef __cplusplus
+}
+#endif

examples/ToyPathTracer/Source/enkiTS/Threads.h

@@ -0,0 +1,210 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+#include <assert.h>
+
+#ifdef _WIN32
+
+	#include "Atomics.h"
+
+	#define WIN32_LEAN_AND_MEAN
+	#include <Windows.h>
+	
+	#define THREADFUNC_DECL DWORD WINAPI
+	#define THREAD_LOCAL __declspec( thread )
+
+namespace enki
+{
+    typedef HANDLE threadid_t;
+
+    // declare the thread start function as:
+    // THREADFUNC_DECL MyThreadStart( void* pArg );
+    inline bool ThreadCreate( threadid_t* returnid, DWORD ( WINAPI *StartFunc) (void* ), void* pArg )
+    {
+        // posix equiv pthread_create
+        DWORD threadid;
+        *returnid = CreateThread( 0, 0, StartFunc, pArg, 0, &threadid );
+        return  *returnid != NULL;
+    }
+
+    inline bool ThreadTerminate( threadid_t threadid )
+    {
+        // posix equiv pthread_cancel
+        return CloseHandle( threadid ) == 0;
+    }
+
+    inline uint32_t GetNumHardwareThreads()
+    {
+        SYSTEM_INFO sysInfo;
+        GetSystemInfo(&sysInfo);
+        return sysInfo.dwNumberOfProcessors;
+    }
+}
+
+#else // posix
+
+	#include <pthread.h>
+	#include <unistd.h>
+	#define THREADFUNC_DECL void*
+	#define THREAD_LOCAL __thread
+
+namespace enki
+{
+    typedef pthread_t threadid_t;  
+        
+    // declare the thread start function as:
+    // THREADFUNC_DECL MyThreadStart( void* pArg );
+    inline bool ThreadCreate( threadid_t* returnid, void* ( *StartFunc) (void* ), void* pArg )
+    {
+        // posix equiv pthread_create
+        int32_t retval = pthread_create( returnid, NULL, StartFunc, pArg );
+
+        return  retval == 0;
+    }
+    
+    inline bool ThreadTerminate( threadid_t threadid )
+    {
+        // posix equiv pthread_cancel
+        return pthread_cancel( threadid ) == 0;
+    }
+    
+    inline uint32_t GetNumHardwareThreads()
+    {
+        return (uint32_t)sysconf( _SC_NPROCESSORS_ONLN );
+    }
+}
+
+#endif // posix
+
+
+// Semaphore implementation
+#ifdef _WIN32
+
+namespace enki
+{
+    struct semaphoreid_t
+    {
+        HANDLE      sem;
+    };
+	
+	inline void SemaphoreCreate( semaphoreid_t& semaphoreid )
+    {
+        semaphoreid.sem = CreateSemaphore(NULL, 0, MAXLONG, NULL );
+    }
+
+    inline void SemaphoreClose( semaphoreid_t& semaphoreid )
+    {
+        CloseHandle( semaphoreid.sem );
+    }
+
+    inline void SemaphoreWait( semaphoreid_t& semaphoreid  )
+    {
+        DWORD retval = WaitForSingleObject( semaphoreid.sem, INFINITE );
+
+        assert( retval != WAIT_FAILED );
+    }
+
+    inline void SemaphoreSignal( semaphoreid_t& semaphoreid, int32_t countWaiting )
+    {
+		if( countWaiting )
+		{
+			ReleaseSemaphore( semaphoreid.sem, countWaiting, NULL );
+		}
+    }
+}
+#elif defined(__MACH__)
+
+// OS X does not have POSIX semaphores
+// see https://developer.apple.com/library/content/documentation/Darwin/Conceptual/KernelProgramming/synchronization/synchronization.html
+#include <mach/mach.h>
+
+namespace enki
+{
+    
+    struct semaphoreid_t
+    {
+        semaphore_t   sem;
+    };
+	
+	inline void SemaphoreCreate( semaphoreid_t& semaphoreid )
+    {
+		semaphore_create( mach_task_self(), &semaphoreid.sem, SYNC_POLICY_FIFO, 0 );
+    }
+    
+    inline void SemaphoreClose( semaphoreid_t& semaphoreid )
+    {
+        semaphore_destroy( mach_task_self(), semaphoreid.sem );
+    }
+    
+    inline void SemaphoreWait( semaphoreid_t& semaphoreid  )
+    {
+        semaphore_wait( semaphoreid.sem );
+    }
+    
+    inline void SemaphoreSignal( semaphoreid_t& semaphoreid, int32_t countWaiting )
+    {
+        while( countWaiting-- > 0 )
+		{
+			semaphore_signal( semaphoreid.sem );
+		}
+    }
+}
+
+#else // POSIX
+
+#include <semaphore.h>
+
+namespace enki
+{
+    
+    struct semaphoreid_t
+    {
+        sem_t   sem;
+    };
+	
+	inline void SemaphoreCreate( semaphoreid_t& semaphoreid )
+    {
+		int err = sem_init( &semaphoreid.sem, 0, 0 );
+		assert( err == 0 );
+    }
+    
+    inline void SemaphoreClose( semaphoreid_t& semaphoreid )
+    {
+        sem_destroy( &semaphoreid.sem );
+    }
+    
+    inline void SemaphoreWait( semaphoreid_t& semaphoreid  )
+    {
+        int err = sem_wait( &semaphoreid.sem );
+		assert( err == 0 );
+    }
+    
+    inline void SemaphoreSignal( semaphoreid_t& semaphoreid, int32_t countWaiting )
+    {
+        while( countWaiting-- > 0 )
+		{
+			sem_post( &semaphoreid.sem );
+		}
+    }
+}
+#endif
+
+

examples/ToyPathTracer/Windows/ComputeShader.hlsl

@@ -0,0 +1,395 @@
+#include "../Source/Config.h"
+
+inline uint RNG(inout uint state)
+{
+    uint x = state;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 15;
+    state = x;
+    return x;
+}
+
+float RandomFloat01(inout uint state)
+{
+    return (RNG(state) & 0xFFFFFF) / 16777216.0f;
+}
+
+float3 RandomInUnitDisk(inout uint state)
+{
+    float a = RandomFloat01(state) * 2.0f * 3.1415926f;
+    float2 xy = float2(cos(a), sin(a));
+    xy *= sqrt(RandomFloat01(state));
+    return float3(xy, 0);
+}
+float3 RandomInUnitSphere(inout uint state)
+{
+    float z = RandomFloat01(state) * 2.0f - 1.0f;
+    float t = RandomFloat01(state) * 2.0f * 3.1415926f;
+    float r = sqrt(max(0.0, 1.0f - z * z));
+    float x = r * cos(t);
+    float y = r * sin(t);
+    float3 res = float3(x, y, z);
+    res *= pow(RandomFloat01(state), 1.0 / 3.0);
+    return res;
+}
+float3 RandomUnitVector(inout uint state)
+{
+    float z = RandomFloat01(state) * 2.0f - 1.0f;
+    float a = RandomFloat01(state) * 2.0f * 3.1415926f;
+    float r = sqrt(1.0f - z * z);
+    float x = r * cos(a);
+    float y = r * sin(a);
+    return float3(x, y, z);
+}
+
+
+
+struct Ray
+{
+    float3 orig;
+    float3 dir;
+};
+Ray MakeRay(float3 orig_, float3 dir_) { Ray r; r.orig = orig_; r.dir = dir_; return r; }
+float3 RayPointAt(Ray r, float t) { return r.orig + r.dir * t; }
+
+
+inline bool refract(float3 v, float3 n, float nint, out float3 outRefracted)
+{
+    float dt = dot(v, n);
+    float discr = 1.0f - nint * nint*(1 - dt * dt);
+    if (discr > 0)
+    {
+        outRefracted = nint * (v - n * dt) - n * sqrt(discr);
+        return true;
+    }
+    return false;
+}
+inline float schlick(float cosine, float ri)
+{
+    float r0 = (1 - ri) / (1 + ri);
+    r0 = r0 * r0;
+    // note: saturate to guard against possible tiny negative numbers
+    return r0 + (1 - r0)*pow(saturate(1 - cosine), 5);
+}
+
+struct Hit
+{
+    float3 pos;
+    float3 normal;
+    float t;
+};
+
+struct Sphere
+{
+    float3 center;
+    float radius;
+    float invRadius;
+};
+
+#define MatLambert 0
+#define MatMetal 1
+#define MatDielectric 2
+
+struct Material
+{
+    int type;
+    float3 albedo;
+    float3 emissive;
+    float roughness;
+    float ri;
+};
+
+groupshared Sphere s_GroupSpheres[kCSMaxObjects];
+groupshared Material s_GroupMaterials[kCSMaxObjects];
+groupshared int s_GroupEmissives[kCSMaxObjects];
+
+
+struct Camera
+{
+    float3 origin;
+    float3 lowerLeftCorner;
+    float3 horizontal;
+    float3 vertical;
+    float3 u, v, w;
+    float lensRadius;
+};
+
+Ray CameraGetRay(Camera cam, float s, float t, inout uint state)
+{
+    float3 rd = cam.lensRadius * RandomInUnitDisk(state);
+    float3 offset = cam.u * rd.x + cam.v * rd.y;
+    return MakeRay(cam.origin + offset, normalize(cam.lowerLeftCorner + s * cam.horizontal + t * cam.vertical - cam.origin - offset));
+}
+
+
+int HitSpheres(Ray r, int sphereCount, float tMin, float tMax, inout Hit outHit)
+{
+    float hitT = tMax;
+    int id = -1;
+    for (int i = 0; i < sphereCount; ++i)
+    {
+        Sphere s = s_GroupSpheres[i];
+        float3 co = s.center - r.orig;
+        float nb = dot(co, r.dir);
+        float c = dot(co, co) - s.radius*s.radius;
+        float discr = nb * nb - c;
+        if (discr > 0)
+        {
+            float discrSq = sqrt(discr);
+
+            // Try earlier t
+            float t = nb - discrSq;
+            if (t <= tMin) // before min, try later t!
+                t = nb + discrSq;
+
+            if (t > tMin && t < hitT)
+            {
+                id = i;
+                hitT = t;
+            }
+        }
+    }
+
+    if (id != -1)
+    {
+        outHit.pos = RayPointAt(r, hitT);
+        outHit.normal = (outHit.pos - s_GroupSpheres[id].center) * s_GroupSpheres[id].invRadius;
+        outHit.t = hitT;
+    }
+    return id;
+}
+
+struct Params
+{
+    Camera cam;
+    int sphereCount;
+    int screenWidth;
+    int screenHeight;
+    int frames;
+    float invWidth;
+    float invHeight;
+    float lerpFac;
+    int emissiveCount;
+};
+
+
+#define kMinT 0.001f
+#define kMaxT 1.0e7f
+#define kMaxDepth 10
+
+
+static int HitWorld(int sphereCount, Ray r, float tMin, float tMax, inout Hit outHit)
+{
+    return HitSpheres(r, sphereCount, tMin, tMax, outHit);
+}
+
+
+static bool Scatter(int sphereCount, int emissiveCount, int matID, Ray r_in, Hit rec, out float3 attenuation, out Ray scattered, out float3 outLightE, inout int inoutRayCount, inout uint state)
+{
+    outLightE = float3(0, 0, 0);
+    Material mat = s_GroupMaterials[matID];
+    if (mat.type == MatLambert)
+    {
+        // random point on unit sphere that is tangent to the hit point
+        float3 target = rec.pos + rec.normal + RandomUnitVector(state);
+        scattered = MakeRay(rec.pos, normalize(target - rec.pos));
+        attenuation = mat.albedo;
+
+        // sample lights
+#if DO_LIGHT_SAMPLING
+        for (int j = 0; j < emissiveCount; ++j)
+        {
+            int i = s_GroupEmissives[j];
+            if (matID == i)
+                continue; // skip self
+            Material smat = s_GroupMaterials[i];
+            Sphere s = s_GroupSpheres[i];
+
+            // create a random direction towards sphere
+            // coord system for sampling: sw, su, sv
+            float3 sw = normalize(s.center - rec.pos);
+            float3 su = normalize(cross(abs(sw.x)>0.01f ? float3(0, 1, 0) : float3(1, 0, 0), sw));
+            float3 sv = cross(sw, su);
+            // sample sphere by solid angle
+            float cosAMax = sqrt(1.0f - s.radius*s.radius / dot(rec.pos - s.center, rec.pos - s.center));
+            float eps1 = RandomFloat01(state), eps2 = RandomFloat01(state);
+            float cosA = 1.0f - eps1 + eps1 * cosAMax;
+            float sinA = sqrt(1.0f - cosA * cosA);
+            float phi = 2 * 3.1415926 * eps2;
+            float3 l = su * cos(phi) * sinA + sv * sin(phi) * sinA + sw * cosA;
+
+            // shoot shadow ray
+            Hit lightHit;
+            ++inoutRayCount;
+            int hitID = HitWorld(sphereCount, MakeRay(rec.pos, l), kMinT, kMaxT, lightHit);
+            if (hitID == i)
+            {
+                float omega = 2 * 3.1415926 * (1 - cosAMax);
+
+                float3 rdir = r_in.dir;
+                float3 nl = dot(rec.normal, rdir) < 0 ? rec.normal : -rec.normal;
+                outLightE += (mat.albedo * smat.emissive) * (max(0.0f, dot(l, nl)) * omega / 3.1415926);
+            }
+        }
+#endif
+        return true;
+    }
+    else if (mat.type == MatMetal)
+    {
+        float3 refl = reflect(r_in.dir, rec.normal);
+        // reflected ray, and random inside of sphere based on roughness
+        float roughness = mat.roughness;
+#if DO_MITSUBA_COMPARE
+        roughness = 0; // until we get better BRDF for metals
+#endif
+        scattered = MakeRay(rec.pos, normalize(refl + roughness*RandomInUnitSphere(state)));
+        attenuation = mat.albedo;
+        return dot(scattered.dir, rec.normal) > 0;
+    }
+    else if (mat.type == MatDielectric)
+    {
+        float3 outwardN;
+        float3 rdir = r_in.dir;
+        float3 refl = reflect(rdir, rec.normal);
+        float nint;
+        attenuation = float3(1, 1, 1);
+        float3 refr;
+        float reflProb;
+        float cosine;
+        if (dot(rdir, rec.normal) > 0)
+        {
+            outwardN = -rec.normal;
+            nint = mat.ri;
+            cosine = mat.ri * dot(rdir, rec.normal);
+        }
+        else
+        {
+            outwardN = rec.normal;
+            nint = 1.0f / mat.ri;
+            cosine = -dot(rdir, rec.normal);
+        }
+        if (refract(rdir, outwardN, nint, refr))
+        {
+            reflProb = schlick(cosine, mat.ri);
+        }
+        else
+        {
+            reflProb = 1;
+        }
+        if (RandomFloat01(state) < reflProb)
+            scattered = MakeRay(rec.pos, normalize(refl));
+        else
+            scattered = MakeRay(rec.pos, normalize(refr));
+    }
+    else
+    {
+        attenuation = float3(1, 0, 1);
+        scattered = MakeRay(float3(0,0,0), float3(0, 0, 1));
+        return false;
+    }
+    return true;
+}
+
+static float3 Trace(int sphereCount, int emissiveCount, Ray r, inout int inoutRayCount, inout uint state)
+{
+    float3 col = 0;
+    float3 curAtten = 1;
+    bool doMaterialE = true;
+    // GPUs don't support recursion, so do tracing iterations in a loop up to max depth
+    for (int depth = 0; depth < kMaxDepth; ++depth)
+    {
+        Hit rec;
+        ++inoutRayCount;
+        int id = HitWorld(sphereCount, r, kMinT, kMaxT, rec);
+        if (id >= 0)
+        {
+            Ray scattered;
+            float3 attenuation;
+            float3 lightE;
+            Material mat = s_GroupMaterials[id];
+            float3 matE = mat.emissive;
+            if (Scatter(sphereCount, emissiveCount, id, r, rec, attenuation, scattered, lightE, inoutRayCount, state))
+            {
+#if DO_LIGHT_SAMPLING
+                if (!doMaterialE) matE = 0;
+                doMaterialE = (mat.type != MatLambert);
+#endif
+                col += curAtten * (matE + lightE);
+                curAtten *= attenuation;
+                r = scattered;
+            }
+            else
+            {
+                col += curAtten * matE;
+                break;
+            }
+        }
+        else
+        {
+            // sky
+#if DO_MITSUBA_COMPARE
+            col += curAtten * float3(0.15f, 0.21f, 0.3f); // easier compare with Mitsuba's constant environment light
+#else
+            float3 unitDir = r.dir;
+            float t = 0.5f*(unitDir.y + 1.0f);
+            float3 skyCol = ((1.0f - t)*float3(1.0f, 1.0f, 1.0f) + t * float3(0.5f, 0.7f, 1.0f)) * 0.3f;
+            col += curAtten * skyCol;
+#endif
+            break;
+        }
+    }
+    return col;
+}
+
+Texture2D srcImage : register(t0);
+RWTexture2D<float4> dstImage : register(u0);
+StructuredBuffer<Sphere> g_Spheres : register(t1);
+StructuredBuffer<Material> g_Materials : register(t2);
+StructuredBuffer<Params> g_Params : register(t3);
+StructuredBuffer<int> g_Emissives : register(t4);
+RWByteAddressBuffer g_OutRayCount : register(u1);
+
+[numthreads(kCSGroupSizeX, kCSGroupSizeY, 1)]
+void main(uint3 gid : SV_DispatchThreadID, uint3 tid : SV_GroupThreadID)
+{
+    // First, move scene data (spheres, materials, emissive indices) into group shared
+    // memory. Do this in parallel; each thread in group copies its own chunk of data.
+    uint threadID = tid.y * kCSGroupSizeX + tid.x;
+    uint groupSize = kCSGroupSizeX * kCSGroupSizeY;
+    uint objCount = g_Params[0].sphereCount;
+    uint myObjCount = (objCount + groupSize - 1) / groupSize;
+    uint myObjStart = threadID * myObjCount;
+    for (uint io = myObjStart; io < myObjStart + myObjCount; ++io)
+    {
+        if (io < objCount)
+        {
+            s_GroupSpheres[io] = g_Spheres[io];
+            s_GroupMaterials[io] = g_Materials[io];
+        }
+        if (io < g_Params[0].emissiveCount)
+        {
+            s_GroupEmissives[io] = g_Emissives[io];
+        }
+    }
+    GroupMemoryBarrierWithGroupSync();
+
+    int rayCount = 0;
+    float3 col = 0;
+    Params params = g_Params[0];
+    uint rngState = (gid.x * 1973 + gid.y * 9277 + params.frames * 26699) | 1;
+    for (int s = 0; s < DO_SAMPLES_PER_PIXEL; s++)
+    {
+        float u = float(gid.x + RandomFloat01(rngState)) * params.invWidth;
+        float v = float(gid.y + RandomFloat01(rngState)) * params.invHeight;
+        Ray r = CameraGetRay(params.cam, u, v, rngState);
+        col += Trace(params.sphereCount, params.emissiveCount, r, rayCount, rngState);
+    }
+    col *= 1.0f / float(DO_SAMPLES_PER_PIXEL);
+
+    float3 prev = srcImage.Load(int3(gid.xy,0)).rgb;
+    col = lerp(col, prev, params.lerpFac);
+    dstImage[gid.xy] = float4(col, 1);
+
+    g_OutRayCount.InterlockedAdd(0, rayCount);
+}

examples/ToyPathTracer/Windows/PixelShader.hlsl

@@ -0,0 +1,15 @@
+float3 LinearToSRGB(float3 rgb)
+{
+    rgb = max(rgb, float3(0, 0, 0));
+    return max(1.055 * pow(rgb, 0.416666667) - 0.055, 0.0);
+}
+
+Texture2D tex : register(t0);
+SamplerState smp : register(s0);
+
+float4 main(float2 uv : TEXCOORD0) : SV_Target
+{
+    float3 col = tex.Sample(smp, uv).rgb;
+    col = LinearToSRGB(col);
+    return float4(col, 1.0f);
+}

examples/ToyPathTracer/Windows/TestCpu.sln

@@ -0,0 +1,31 @@
+
+Microsoft Visual Studio Solution File, Format Version 12.00
+# Visual Studio 15
+VisualStudioVersion = 15.0.27130.2036
+MinimumVisualStudioVersion = 10.0.40219.1
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "TestCpu", "TestCpu.vcxproj", "{4F84B756-87F5-4B92-827B-DA087DAE1900}"
+EndProject
+Global
+	GlobalSection(SolutionConfigurationPlatforms) = preSolution
+		Debug|x64 = Debug|x64
+		Debug|x86 = Debug|x86
+		Release|x64 = Release|x64
+		Release|x86 = Release|x86
+	EndGlobalSection
+	GlobalSection(ProjectConfigurationPlatforms) = postSolution
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x64.ActiveCfg = Debug|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x64.Build.0 = Debug|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x86.ActiveCfg = Debug|Win32
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x86.Build.0 = Debug|Win32
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x64.ActiveCfg = Release|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x64.Build.0 = Release|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x86.ActiveCfg = Release|Win32
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x86.Build.0 = Release|Win32
+	EndGlobalSection
+	GlobalSection(SolutionProperties) = preSolution
+		HideSolutionNode = FALSE
+	EndGlobalSection
+	GlobalSection(ExtensibilityGlobals) = postSolution
+		SolutionGuid = {067FB780-37B8-465E-AD7E-E7B238B9C04F}
+	EndGlobalSection
+EndGlobal

examples/ToyPathTracer/Windows/TestCpu.vcxproj

@@ -0,0 +1,243 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="15.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup Label="ProjectConfigurations">
+    <ProjectConfiguration Include="Debug|Win32">
+      <Configuration>Debug</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|Win32">
+      <Configuration>Release</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Debug|x64">
+      <Configuration>Debug</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|x64">
+      <Configuration>Release</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+  </ItemGroup>
+  <PropertyGroup Label="Globals">
+    <VCProjectVersion>15.0</VCProjectVersion>
+    <ProjectGuid>{4F84B756-87F5-4B92-827B-DA087DAE1900}</ProjectGuid>
+    <Keyword>Win32Proj</Keyword>
+    <RootNamespace>TestCpu</RootNamespace>
+    <WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v142</PlatformToolset>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v142</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v142</PlatformToolset>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v142</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+  <ImportGroup Label="ExtensionSettings">
+  </ImportGroup>
+  <ImportGroup Label="Shared">
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <PropertyGroup Label="UserMacros" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <LinkIncremental>false</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <LinkIncremental>false</LinkIncremental>
+  </PropertyGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <PreprocessorDefinitions>WIN32;_DEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <PreprocessorDefinitions>TRACY_ENABLE;_DEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <PreprocessorDefinitions>WIN32;NDEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <ExceptionHandling>false</ExceptionHandling>
+      <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+      <BufferSecurityCheck>false</BufferSecurityCheck>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <PreprocessorDefinitions>TRACY_ENABLE;NDEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <ExceptionHandling>false</ExceptionHandling>
+      <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+      <BufferSecurityCheck>false</BufferSecurityCheck>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemGroup>
+    <ClCompile Include="..\..\..\TracyClient.cpp" />
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler.cpp" />
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler_c.cpp" />
+    <ClCompile Include="..\Source\Maths.cpp" />
+    <ClCompile Include="..\Source\Test.cpp" />
+    <ClCompile Include="TestWin.cpp" />
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="..\Source\Config.h" />
+    <ClInclude Include="..\Source\enkiTS\Atomics.h" />
+    <ClInclude Include="..\Source\enkiTS\LockLessMultiReadPipe.h" />
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler.h" />
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler_c.h" />
+    <ClInclude Include="..\Source\enkiTS\Threads.h" />
+    <ClInclude Include="..\Source\Maths.h" />
+    <ClInclude Include="..\Source\MathSimd.h" />
+    <ClInclude Include="..\Source\Test.h" />
+    <ClInclude Include="..\Source\stb_image.h" />
+  </ItemGroup>
+  <ItemGroup>
+    <None Include="..\.editorconfig" />
+  </ItemGroup>
+  <ItemGroup>
+    <FxCompile Include="ComputeShader.hlsl">
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|x64'">5.0</ShaderModel>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">5.0</ShaderModel>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">5.0</ShaderModel>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">5.0</ShaderModel>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CompiledComputeShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CompiledComputeShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">CompiledComputeShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">CompiledComputeShader.h</HeaderFileOutput>
+    </FxCompile>
+    <FxCompile Include="PixelShader.hlsl">
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Pixel</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Pixel</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Pixel</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Pixel</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|x64'">5.0</ShaderModel>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CompiledPixelShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CompiledPixelShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">CompiledPixelShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">CompiledPixelShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">g_PSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">g_PSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">g_PSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">g_PSBytecode</VariableName>
+    </FxCompile>
+    <FxCompile Include="VertexShader.hlsl">
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Vertex</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Vertex</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Vertex</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Vertex</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|x64'">5.0</ShaderModel>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CompiledVertexShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CompiledVertexShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">CompiledVertexShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">CompiledVertexShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">g_VSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">g_VSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">g_VSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">g_VSBytecode</VariableName>
+    </FxCompile>
+  </ItemGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+  <ImportGroup Label="ExtensionTargets">
+  </ImportGroup>
+</Project>

examples/ToyPathTracer/Windows/TestCpu.vcxproj.filters

@@ -0,0 +1,67 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup>
+    <ClCompile Include="TestWin.cpp" />
+    <ClCompile Include="..\Source\Test.cpp">
+      <Filter>Source</Filter>
+    </ClCompile>
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler.cpp">
+      <Filter>Source\enkiTS</Filter>
+    </ClCompile>
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler_c.cpp">
+      <Filter>Source\enkiTS</Filter>
+    </ClCompile>
+    <ClCompile Include="..\Source\Maths.cpp">
+      <Filter>Source</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\TracyClient.cpp" />
+  </ItemGroup>
+  <ItemGroup>
+    <Filter Include="Source">
+      <UniqueIdentifier>{5f19f217-c1c7-4eeb-be61-8b986fee9375}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source\enkiTS">
+      <UniqueIdentifier>{38c448a8-1dcc-4116-9410-a9f8d068caff}</UniqueIdentifier>
+    </Filter>
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="..\Source\Test.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\stb_image.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\Atomics.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\LockLessMultiReadPipe.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler_c.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\Threads.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\Maths.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\Config.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\MathSimd.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+  </ItemGroup>
+  <ItemGroup>
+    <None Include="..\.editorconfig" />
+  </ItemGroup>
+  <ItemGroup>
+    <FxCompile Include="VertexShader.hlsl" />
+    <FxCompile Include="PixelShader.hlsl" />
+    <FxCompile Include="ComputeShader.hlsl" />
+  </ItemGroup>
+</Project>

examples/ToyPathTracer/Windows/TestWin.cpp

@@ -0,0 +1,557 @@
+#include <stdint.h>
+#define WIN32_LEAN_AND_MEAN
+#define NOMINMAX
+#include <windows.h>
+#include <d3d11_1.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+#include <algorithm>
+
+#include "../Source/Config.h"
+#include "../Source/Maths.h"
+#include "../Source/Test.h"
+#include "CompiledVertexShader.h"
+#include "CompiledPixelShader.h"
+
+#include "../../../Tracy.hpp"
+
+static HINSTANCE g_HInstance;
+static HWND g_Wnd;
+
+ATOM                MyRegisterClass(HINSTANCE hInstance);
+BOOL                InitInstance(HINSTANCE, int);
+LRESULT CALLBACK    WndProc(HWND, UINT, WPARAM, LPARAM);
+INT_PTR CALLBACK    About(HWND, UINT, WPARAM, LPARAM);
+
+static HRESULT InitD3DDevice();
+static void ShutdownD3DDevice();
+static void RenderFrame();
+
+static float* g_Backbuffer;
+
+static D3D_FEATURE_LEVEL g_D3D11FeatureLevel = D3D_FEATURE_LEVEL_11_0;
+static ID3D11Device* g_D3D11Device = nullptr;
+static ID3D11DeviceContext* g_D3D11Ctx = nullptr;
+static IDXGISwapChain* g_D3D11SwapChain = nullptr;
+static ID3D11RenderTargetView* g_D3D11RenderTarget = nullptr;
+static ID3D11VertexShader* g_VertexShader;
+static ID3D11PixelShader* g_PixelShader;
+static ID3D11Texture2D *g_BackbufferTexture, *g_BackbufferTexture2;
+static ID3D11ShaderResourceView *g_BackbufferSRV, *g_BackbufferSRV2;
+static ID3D11UnorderedAccessView *g_BackbufferUAV, *g_BackbufferUAV2;
+static ID3D11SamplerState* g_SamplerLinear;
+static ID3D11RasterizerState* g_RasterState;
+static int g_BackbufferIndex;
+
+
+#if DO_COMPUTE_GPU
+#include "CompiledComputeShader.h"
+struct ComputeParams
+{
+    Camera cam;
+    int sphereCount;
+    int screenWidth;
+    int screenHeight;
+    int frames;
+    float invWidth;
+    float invHeight;
+    float lerpFac;
+    int emissiveCount;
+};
+static ID3D11ComputeShader* g_ComputeShader;
+static ID3D11Buffer* g_DataSpheres;     static ID3D11ShaderResourceView* g_SRVSpheres;
+static ID3D11Buffer* g_DataMaterials;   static ID3D11ShaderResourceView* g_SRVMaterials;
+static ID3D11Buffer* g_DataParams;      static ID3D11ShaderResourceView* g_SRVParams;
+static ID3D11Buffer* g_DataEmissives;   static ID3D11ShaderResourceView* g_SRVEmissives;
+static ID3D11Buffer* g_DataCounter;     static ID3D11UnorderedAccessView* g_UAVCounter;
+static int g_SphereCount, g_ObjSize, g_MatSize;
+static ID3D11Query *g_QueryBegin, *g_QueryEnd, *g_QueryDisjoint;
+#endif // #if DO_COMPUTE_GPU
+
+int APIENTRY wWinMain(_In_ HINSTANCE hInstance, _In_opt_ HINSTANCE, _In_ LPWSTR, _In_ int nCmdShow)
+{
+    g_Backbuffer = new float[kBackbufferWidth * kBackbufferHeight * 4];
+    memset(g_Backbuffer, 0, kBackbufferWidth * kBackbufferHeight * 4 * sizeof(g_Backbuffer[0]));
+
+    InitializeTest();
+
+    MyRegisterClass(hInstance);
+    if (!InitInstance (hInstance, nCmdShow))
+    {
+        return FALSE;
+    }
+
+    if (FAILED(InitD3DDevice()))
+    {
+        ShutdownD3DDevice();
+        return 0;
+    }
+
+    g_D3D11Device->CreateVertexShader(g_VSBytecode, ARRAYSIZE(g_VSBytecode), NULL, &g_VertexShader);
+    g_D3D11Device->CreatePixelShader(g_PSBytecode, ARRAYSIZE(g_PSBytecode), NULL, &g_PixelShader);
+#if DO_COMPUTE_GPU
+    g_D3D11Device->CreateComputeShader(g_CSBytecode, ARRAYSIZE(g_CSBytecode), NULL, &g_ComputeShader);
+#endif
+
+    D3D11_TEXTURE2D_DESC texDesc = {};
+    texDesc.Width = kBackbufferWidth;
+    texDesc.Height = kBackbufferHeight;
+    texDesc.MipLevels = 1;
+    texDesc.ArraySize = 1;
+    texDesc.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
+    texDesc.SampleDesc.Count = 1;
+    texDesc.SampleDesc.Quality = 0;
+#if DO_COMPUTE_GPU
+    texDesc.Usage = D3D11_USAGE_DEFAULT;
+    texDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_UNORDERED_ACCESS;
+    texDesc.CPUAccessFlags = 0;
+#else
+    texDesc.Usage = D3D11_USAGE_DYNAMIC;
+    texDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
+    texDesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
+#endif
+    texDesc.MiscFlags = 0;
+    g_D3D11Device->CreateTexture2D(&texDesc, NULL, &g_BackbufferTexture);
+    g_D3D11Device->CreateTexture2D(&texDesc, NULL, &g_BackbufferTexture2);
+
+    D3D11_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
+    srvDesc.Format = texDesc.Format;
+    srvDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D;
+    srvDesc.Texture2D.MipLevels = 1;
+    srvDesc.Texture2D.MostDetailedMip = 0;
+    g_D3D11Device->CreateShaderResourceView(g_BackbufferTexture, &srvDesc, &g_BackbufferSRV);
+    g_D3D11Device->CreateShaderResourceView(g_BackbufferTexture2, &srvDesc, &g_BackbufferSRV2);
+
+    D3D11_SAMPLER_DESC smpDesc = {};
+    smpDesc.Filter = D3D11_FILTER_MIN_MAG_LINEAR_MIP_POINT;
+    smpDesc.AddressU = smpDesc.AddressV = smpDesc.AddressW = D3D11_TEXTURE_ADDRESS_CLAMP;
+    g_D3D11Device->CreateSamplerState(&smpDesc, &g_SamplerLinear);
+
+    D3D11_RASTERIZER_DESC rasterDesc = {};
+    rasterDesc.FillMode = D3D11_FILL_SOLID;
+    rasterDesc.CullMode = D3D11_CULL_NONE;
+    g_D3D11Device->CreateRasterizerState(&rasterDesc, &g_RasterState);
+
+#if DO_COMPUTE_GPU
+    D3D11_UNORDERED_ACCESS_VIEW_DESC uavDesc = {};
+
+    int camSize;
+    GetObjectCount(g_SphereCount, g_ObjSize, g_MatSize, camSize);
+    assert(g_ObjSize == 20);
+    assert(g_MatSize == 36);
+    assert(camSize == 88);
+    D3D11_BUFFER_DESC bdesc = {};
+    bdesc.ByteWidth = g_SphereCount * g_ObjSize;
+    bdesc.Usage = D3D11_USAGE_DEFAULT;
+    bdesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
+    bdesc.CPUAccessFlags = 0;
+    bdesc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_STRUCTURED;
+    bdesc.StructureByteStride = g_ObjSize;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataSpheres);
+    srvDesc.Format = DXGI_FORMAT_UNKNOWN;
+    srvDesc.ViewDimension = D3D11_SRV_DIMENSION_BUFFER;
+    srvDesc.Buffer.FirstElement = 0;
+    srvDesc.Buffer.NumElements = g_SphereCount;
+    g_D3D11Device->CreateShaderResourceView(g_DataSpheres, &srvDesc, &g_SRVSpheres);
+
+    bdesc.ByteWidth = g_SphereCount * g_MatSize;
+    bdesc.StructureByteStride = g_MatSize;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataMaterials);
+    srvDesc.Buffer.NumElements = g_SphereCount;
+    g_D3D11Device->CreateShaderResourceView(g_DataMaterials, &srvDesc, &g_SRVMaterials);
+
+    bdesc.ByteWidth = sizeof(ComputeParams);
+    bdesc.StructureByteStride = sizeof(ComputeParams);
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataParams);
+    srvDesc.Buffer.NumElements = 1;
+    g_D3D11Device->CreateShaderResourceView(g_DataParams, &srvDesc, &g_SRVParams);
+
+    bdesc.ByteWidth = g_SphereCount * 4;
+    bdesc.StructureByteStride = 4;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataEmissives);
+    srvDesc.Buffer.NumElements = g_SphereCount;
+    g_D3D11Device->CreateShaderResourceView(g_DataEmissives, &srvDesc, &g_SRVEmissives);
+
+    bdesc.ByteWidth = 4;
+    bdesc.BindFlags |= D3D11_BIND_UNORDERED_ACCESS;
+    bdesc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_ALLOW_RAW_VIEWS;
+    bdesc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataCounter);
+    uavDesc.Format = DXGI_FORMAT_R32_TYPELESS;
+    uavDesc.ViewDimension = D3D11_UAV_DIMENSION_BUFFER;
+    uavDesc.Buffer.FirstElement = 0;
+    uavDesc.Buffer.NumElements = 1;
+    uavDesc.Buffer.Flags = D3D11_BUFFER_UAV_FLAG_RAW;
+    g_D3D11Device->CreateUnorderedAccessView(g_DataCounter, &uavDesc, &g_UAVCounter);
+
+    uavDesc.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
+    uavDesc.ViewDimension = D3D11_UAV_DIMENSION_TEXTURE2D;
+    uavDesc.Texture2D.MipSlice = 0;
+    g_D3D11Device->CreateUnorderedAccessView(g_BackbufferTexture, &uavDesc, &g_BackbufferUAV);
+    g_D3D11Device->CreateUnorderedAccessView(g_BackbufferTexture2, &uavDesc, &g_BackbufferUAV2);
+
+    D3D11_QUERY_DESC qDesc = {};
+    qDesc.Query = D3D11_QUERY_TIMESTAMP;
+    g_D3D11Device->CreateQuery(&qDesc, &g_QueryBegin);
+    g_D3D11Device->CreateQuery(&qDesc, &g_QueryEnd);
+    qDesc.Query = D3D11_QUERY_TIMESTAMP_DISJOINT;
+    g_D3D11Device->CreateQuery(&qDesc, &g_QueryDisjoint);
+#endif // #if DO_COMPUTE_GPU
+
+
+    static int framesLeft = 10;
+
+    // Main message loop
+    MSG msg = { 0 };
+    while (msg.message != WM_QUIT)
+    {
+        if (PeekMessage(&msg, NULL, 0U, 0U, PM_REMOVE))
+        {
+            TranslateMessage(&msg);
+            DispatchMessage(&msg);
+        }
+        else
+        {
+            RenderFrame();
+            if( --framesLeft == 0 ) break;
+        }
+    }
+
+    ShutdownTest();
+    ShutdownD3DDevice();
+
+    return (int) msg.wParam;
+}
+
+
+ATOM MyRegisterClass(HINSTANCE hInstance)
+{
+    ZoneScoped;
+
+    WNDCLASSEXW wcex;
+    memset(&wcex, 0, sizeof(wcex));
+    wcex.cbSize = sizeof(WNDCLASSEX);
+    wcex.style          = CS_HREDRAW | CS_VREDRAW;
+    wcex.lpfnWndProc    = WndProc;
+    wcex.cbClsExtra     = 0;
+    wcex.cbWndExtra     = 0;
+    wcex.hInstance      = hInstance;
+    wcex.hCursor        = LoadCursor(nullptr, IDC_ARROW);
+    wcex.hbrBackground  = (HBRUSH)(COLOR_WINDOW+1);
+    wcex.lpszClassName  = L"TestClass";
+    return RegisterClassExW(&wcex);
+}
+
+BOOL InitInstance(HINSTANCE hInstance, int nCmdShow)
+{
+    ZoneScoped;
+
+    g_HInstance = hInstance;
+    RECT rc = { 0, 0, kBackbufferWidth, kBackbufferHeight };
+    DWORD style = WS_OVERLAPPED | WS_CAPTION | WS_SYSMENU | WS_MINIMIZEBOX;
+    AdjustWindowRect(&rc, style, FALSE);
+    HWND hWnd = CreateWindowW(L"TestClass", L"Test", style, CW_USEDEFAULT, CW_USEDEFAULT, rc.right-rc.left, rc.bottom-rc.top, nullptr, nullptr, hInstance, nullptr);
+    if (!hWnd)
+        return FALSE;
+    g_Wnd = hWnd;
+    ShowWindow(hWnd, nCmdShow);
+    UpdateWindow(hWnd);
+    return TRUE;
+}
+
+static uint64_t s_Time;
+static int s_Count;
+static char s_Buffer[200];
+static unsigned s_Flags = kFlagProgressive;
+static int s_FrameCount = 0;
+
+
+static void RenderFrame()
+{
+    ZoneScoped;
+
+    LARGE_INTEGER time1;
+
+#if DO_COMPUTE_GPU
+    QueryPerformanceCounter(&time1);
+    float t = float(clock()) / CLOCKS_PER_SEC;
+    UpdateTest(t, s_FrameCount, kBackbufferWidth, kBackbufferHeight, s_Flags);
+
+    g_BackbufferIndex = 1 - g_BackbufferIndex;
+    void* dataSpheres = alloca(g_SphereCount * g_ObjSize);
+    void* dataMaterials = alloca(g_SphereCount * g_MatSize);
+    void* dataEmissives = alloca(g_SphereCount * 4);
+    ComputeParams dataParams;
+    GetSceneDesc(dataSpheres, dataMaterials, &dataParams.cam, dataEmissives, &dataParams.emissiveCount);
+
+    dataParams.sphereCount = g_SphereCount;
+    dataParams.screenWidth = kBackbufferWidth;
+    dataParams.screenHeight = kBackbufferHeight;
+    dataParams.frames = s_FrameCount;
+    dataParams.invWidth = 1.0f / kBackbufferWidth;
+    dataParams.invHeight = 1.0f / kBackbufferHeight;
+    float lerpFac = float(s_FrameCount) / float(s_FrameCount + 1);
+    if (s_Flags & kFlagAnimate)
+        lerpFac *= DO_ANIMATE_SMOOTHING;
+    if (!(s_Flags & kFlagProgressive))
+        lerpFac = 0;
+    dataParams.lerpFac = lerpFac;
+
+    g_D3D11Ctx->UpdateSubresource(g_DataSpheres, 0, NULL, dataSpheres, 0, 0);
+    g_D3D11Ctx->UpdateSubresource(g_DataMaterials, 0, NULL, dataMaterials, 0, 0);
+    g_D3D11Ctx->UpdateSubresource(g_DataParams, 0, NULL, &dataParams, 0, 0);
+    g_D3D11Ctx->UpdateSubresource(g_DataEmissives, 0, NULL, dataEmissives, 0, 0);
+
+    ID3D11ShaderResourceView* srvs[] = {
+        g_BackbufferIndex == 0 ? g_BackbufferSRV2 : g_BackbufferSRV,
+        g_SRVSpheres,
+        g_SRVMaterials,
+        g_SRVParams,
+        g_SRVEmissives
+    };
+    g_D3D11Ctx->CSSetShaderResources(0, ARRAYSIZE(srvs), srvs);
+    ID3D11UnorderedAccessView* uavs[] = {
+        g_BackbufferIndex == 0 ? g_BackbufferUAV : g_BackbufferUAV2,
+        g_UAVCounter
+    };
+    g_D3D11Ctx->CSSetUnorderedAccessViews(0, ARRAYSIZE(uavs), uavs, NULL);
+    g_D3D11Ctx->CSSetShader(g_ComputeShader, NULL, 0);
+    g_D3D11Ctx->Begin(g_QueryDisjoint);
+    g_D3D11Ctx->End(g_QueryBegin);
+    g_D3D11Ctx->Dispatch(kBackbufferWidth/kCSGroupSizeX, kBackbufferHeight/kCSGroupSizeY, 1);
+    g_D3D11Ctx->End(g_QueryEnd);
+    uavs[0] = NULL;
+    g_D3D11Ctx->CSSetUnorderedAccessViews(0, ARRAYSIZE(uavs), uavs, NULL);
+    ++s_FrameCount;
+
+#else
+    QueryPerformanceCounter(&time1);
+    float t = float(clock()) / CLOCKS_PER_SEC;
+    static size_t s_RayCounter = 0;
+    int rayCount;
+    UpdateTest(t, s_FrameCount, kBackbufferWidth, kBackbufferHeight, s_Flags);
+    DrawTest(t, s_FrameCount, kBackbufferWidth, kBackbufferHeight, g_Backbuffer, rayCount, s_Flags);
+    s_FrameCount++;
+    s_RayCounter += rayCount;
+    LARGE_INTEGER time2;
+    QueryPerformanceCounter(&time2);
+    uint64_t dt = time2.QuadPart - time1.QuadPart;
+    ++s_Count;
+    s_Time += dt;
+    if (s_Count > 10)
+    {
+        LARGE_INTEGER frequency;
+        QueryPerformanceFrequency(&frequency);
+
+        double s = double(s_Time) / double(frequency.QuadPart) / s_Count;
+        sprintf_s(s_Buffer, sizeof(s_Buffer), "%.2fms (%.1f FPS) %.1fMrays/s %.2fMrays/frame frames %i\n", s * 1000.0f, 1.f / s, s_RayCounter / s_Count / s * 1.0e-6f, s_RayCounter / s_Count * 1.0e-6f, s_FrameCount);
+        SetWindowTextA(g_Wnd, s_Buffer);
+        OutputDebugStringA(s_Buffer);
+        s_Count = 0;
+        s_Time = 0;
+        s_RayCounter = 0;
+    }
+
+    D3D11_MAPPED_SUBRESOURCE mapped;
+    g_D3D11Ctx->Map(g_BackbufferTexture, 0, D3D11_MAP_WRITE_DISCARD, 0, &mapped);
+    const uint8_t* src = (const uint8_t*)g_Backbuffer;
+    uint8_t* dst = (uint8_t*)mapped.pData;
+    for (int y = 0; y < kBackbufferHeight; ++y)
+    {
+        memcpy(dst, src, kBackbufferWidth * 16);
+        src += kBackbufferWidth * 16;
+        dst += mapped.RowPitch;
+    }
+    g_D3D11Ctx->Unmap(g_BackbufferTexture, 0);
+#endif
+
+    g_D3D11Ctx->VSSetShader(g_VertexShader, NULL, 0);
+    g_D3D11Ctx->PSSetShader(g_PixelShader, NULL, 0);
+    g_D3D11Ctx->PSSetShaderResources(0, 1, g_BackbufferIndex == 0 ? &g_BackbufferSRV : &g_BackbufferSRV2);
+    g_D3D11Ctx->PSSetSamplers(0, 1, &g_SamplerLinear);
+    g_D3D11Ctx->IASetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
+    g_D3D11Ctx->RSSetState(g_RasterState);
+    g_D3D11Ctx->Draw(3, 0);
+    g_D3D11SwapChain->Present(0, 0);
+
+    FrameMark;
+
+#if DO_COMPUTE_GPU
+    g_D3D11Ctx->End(g_QueryDisjoint);
+
+    // get GPU times
+    while (g_D3D11Ctx->GetData(g_QueryDisjoint, NULL, 0, 0) == S_FALSE) { Sleep(0); }
+    D3D10_QUERY_DATA_TIMESTAMP_DISJOINT tsDisjoint;
+    g_D3D11Ctx->GetData(g_QueryDisjoint, &tsDisjoint, sizeof(tsDisjoint), 0);
+    if (!tsDisjoint.Disjoint)
+    {
+        UINT64 tsBegin, tsEnd;
+        // Note: on some GPUs/drivers, even when the disjoint query above already said "yeah I have data",
+        // might still not return "I have data" for timestamp queries before it.
+        while (g_D3D11Ctx->GetData(g_QueryBegin, &tsBegin, sizeof(tsBegin), 0) == S_FALSE) { Sleep(0); }
+        while (g_D3D11Ctx->GetData(g_QueryEnd, &tsEnd, sizeof(tsEnd), 0) == S_FALSE) { Sleep(0); }
+
+        float s = float(tsEnd - tsBegin) / float(tsDisjoint.Frequency);
+
+        static uint64_t s_RayCounter;
+        D3D11_MAPPED_SUBRESOURCE mapped;
+        g_D3D11Ctx->Map(g_DataCounter, 0, D3D11_MAP_READ, 0, &mapped);
+        s_RayCounter += *(const int*)mapped.pData;
+        g_D3D11Ctx->Unmap(g_DataCounter, 0);
+        int zeroCount = 0;
+        g_D3D11Ctx->UpdateSubresource(g_DataCounter, 0, NULL, &zeroCount, 0, 0);
+
+        static float s_Time;
+        ++s_Count;
+        s_Time += s;
+        if (s_Count > 150)
+        {
+            s = s_Time / s_Count;
+            sprintf_s(s_Buffer, sizeof(s_Buffer), "%.2fms (%.1f FPS) %.1fMrays/s %.2fMrays/frame frames %i\n", s * 1000.0f, 1.f / s, s_RayCounter / s_Count / s * 1.0e-6f, s_RayCounter / s_Count * 1.0e-6f, s_FrameCount);
+            SetWindowTextA(g_Wnd, s_Buffer);
+            s_Count = 0;
+            s_Time = 0;
+            s_RayCounter = 0;
+        }
+
+    }
+#endif // #if DO_COMPUTE_GPU
+}
+
+
+LRESULT CALLBACK WndProc(HWND hWnd, UINT message, WPARAM wParam, LPARAM lParam)
+{
+    switch (message)
+    {
+    case WM_PAINT:
+        {
+            PAINTSTRUCT ps;
+            HDC hdc = BeginPaint(hWnd, &ps);
+            EndPaint(hWnd, &ps);
+        }
+        break;
+    case WM_DESTROY:
+        PostQuitMessage(0);
+        break;
+    case WM_CHAR:
+        if (wParam == 'a')
+            s_Flags = s_Flags ^ kFlagAnimate;
+        if (wParam == 'p')
+        {
+            s_Flags = s_Flags ^ kFlagProgressive;
+            s_FrameCount = 0;
+        }
+        break;
+    default:
+        return DefWindowProc(hWnd, message, wParam, lParam);
+    }
+    return 0;
+}
+
+
+static HRESULT InitD3DDevice()
+{
+    ZoneScoped;
+
+    HRESULT hr = S_OK;
+
+    RECT rc;
+    GetClientRect(g_Wnd, &rc);
+    UINT width = rc.right - rc.left;
+    UINT height = rc.bottom - rc.top;
+
+    UINT createDeviceFlags = 0;
+#ifdef _DEBUG
+    createDeviceFlags |= D3D11_CREATE_DEVICE_DEBUG;
+#endif
+
+    D3D_FEATURE_LEVEL featureLevels[] =
+    {
+        D3D_FEATURE_LEVEL_11_0,
+    };
+    UINT numFeatureLevels = ARRAYSIZE(featureLevels);
+    hr = D3D11CreateDevice(nullptr, D3D_DRIVER_TYPE_HARDWARE, nullptr, createDeviceFlags, featureLevels, numFeatureLevels, D3D11_SDK_VERSION, &g_D3D11Device, &g_D3D11FeatureLevel, &g_D3D11Ctx);
+    if (FAILED(hr))
+        return hr;
+
+    // Get DXGI factory
+    IDXGIFactory1* dxgiFactory = nullptr;
+    {
+        IDXGIDevice* dxgiDevice = nullptr;
+        hr = g_D3D11Device->QueryInterface(__uuidof(IDXGIDevice), reinterpret_cast<void**>(&dxgiDevice));
+        if (SUCCEEDED(hr))
+        {
+            IDXGIAdapter* adapter = nullptr;
+            hr = dxgiDevice->GetAdapter(&adapter);
+            if (SUCCEEDED(hr))
+            {
+                hr = adapter->GetParent(__uuidof(IDXGIFactory1), reinterpret_cast<void**>(&dxgiFactory));
+                adapter->Release();
+            }
+            dxgiDevice->Release();
+        }
+    }
+    if (FAILED(hr))
+        return hr;
+
+    // Create swap chain
+    DXGI_SWAP_CHAIN_DESC sd;
+    ZeroMemory(&sd, sizeof(sd));
+    sd.BufferCount = 1;
+    sd.BufferDesc.Width = width;
+    sd.BufferDesc.Height = height;
+    sd.BufferDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+    sd.BufferDesc.RefreshRate.Numerator = 60;
+    sd.BufferDesc.RefreshRate.Denominator = 1;
+    sd.BufferUsage = DXGI_USAGE_RENDER_TARGET_OUTPUT;
+    sd.OutputWindow = g_Wnd;
+    sd.SampleDesc.Count = 1;
+    sd.SampleDesc.Quality = 0;
+    sd.Windowed = TRUE;
+    hr = dxgiFactory->CreateSwapChain(g_D3D11Device, &sd, &g_D3D11SwapChain);
+
+    // Prevent Alt-Enter
+    dxgiFactory->MakeWindowAssociation(g_Wnd, DXGI_MWA_NO_ALT_ENTER);
+    dxgiFactory->Release();
+
+    if (FAILED(hr))
+        return hr;
+
+    // RTV
+    ID3D11Texture2D* pBackBuffer = nullptr;
+    hr = g_D3D11SwapChain->GetBuffer(0, __uuidof(ID3D11Texture2D), reinterpret_cast<void**>(&pBackBuffer));
+    if (FAILED(hr))
+        return hr;
+    hr = g_D3D11Device->CreateRenderTargetView(pBackBuffer, nullptr, &g_D3D11RenderTarget);
+    pBackBuffer->Release();
+    if (FAILED(hr))
+        return hr;
+
+    g_D3D11Ctx->OMSetRenderTargets(1, &g_D3D11RenderTarget, nullptr);
+
+    // Viewport
+    D3D11_VIEWPORT vp;
+    vp.Width = (float)width;
+    vp.Height = (float)height;
+    vp.MinDepth = 0.0f;
+    vp.MaxDepth = 1.0f;
+    vp.TopLeftX = 0;
+    vp.TopLeftY = 0;
+    g_D3D11Ctx->RSSetViewports(1, &vp);
+
+    return S_OK;
+}
+
+static void ShutdownD3DDevice()
+{
+    ZoneScoped;
+
+    if (g_D3D11Ctx) g_D3D11Ctx->ClearState();
+
+    if (g_D3D11RenderTarget) g_D3D11RenderTarget->Release();
+    if (g_D3D11SwapChain) g_D3D11SwapChain->Release();
+    if (g_D3D11Ctx) g_D3D11Ctx->Release();
+    if (g_D3D11Device) g_D3D11Device->Release();
+}

examples/ToyPathTracer/Windows/VertexShader.hlsl

@@ -0,0 +1,13 @@
+struct vs2ps
+{
+    float2 uv : TEXCOORD0;
+    float4 pos : SV_Position;
+};
+
+vs2ps main(uint vid : SV_VertexID)
+{
+    vs2ps o;
+    o.uv = float2((vid << 1) & 2, vid & 2);
+    o.pos = float4(o.uv * float2(2, 2) + float2(-1, -1), 0, 1);
+    return o;
+}

examples/ToyPathTracer/license.md

@@ -0,0 +1,24 @@
+This is free and unencumbered software released into the public domain.
+
+Anyone is free to copy, modify, publish, use, compile, sell, or
+distribute this software, either in source code form or as a compiled
+binary, for any purpose, commercial or non-commercial, and by any
+means.
+
+In jurisdictions that recognize copyright laws, the author or authors
+of this software dedicate any and all copyright interest in the
+software to the public domain. We make this dedication for the benefit
+of the public at large and to the detriment of our heirs and
+successors. We intend this dedication to be an overt act of
+relinquishment in perpetuity of all present and future rights to this
+software under copyright law.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
+
+For more information, please refer to <http://unlicense.org>

extra/dxt1divtable.c

@@ -0,0 +1,22 @@
+#include <stdint.h>
+#include <stdio.h>
+
+int main()
+{
+    for( int i=0; i<255*3+1; i++ )
+    {
+        // replace 4 with 2 for ARM NEON table
+        uint32_t range = ( 4 << 16 ) / ( 1+i );
+        if( range > 0xFFFF ) range = 0xFFFF;
+        if( i % 16 == 15 )
+        {
+            printf( "0x%04x,\n", range );
+        }
+        else
+        {
+            printf( "0x%04x, ", range );
+        }
+    }
+    printf( "\n" );
+    return 0;
+}

extra/dxt1table.c

@@ -0,0 +1,36 @@
+#include <stdint.h>
+#include <stdio.h>
+
+static const uint8_t IndexTable[4] = { 1, 3, 2, 0 };
+
+int convert( int v )
+{
+    int v0 = v & 0x3;
+    int v1 = ( v >> 2 ) & 0x3;
+    int v2 = ( v >> 4 ) & 0x3;
+    int v3 = ( v >> 6 );
+
+    int t0 = IndexTable[v0];
+    int t1 = IndexTable[v1];
+    int t2 = IndexTable[v2];
+    int t3 = IndexTable[v3];
+
+    return t0 | ( t1 << 2 ) | ( t2 << 4 ) | ( t3 << 6 );
+}
+
+int main()
+{
+    for( int i=0; i<256; i++ )
+    {
+        if( i % 16 == 15 )
+        {
+            printf( "%i,\n", convert( i ) );
+        }
+        else
+        {
+            printf( "%i,\t", convert( i ) );
+        }
+    }
+    printf( "\n" );
+    return 0;
+}

extra/rdotbl.c

@@ -0,0 +1,24 @@
+#include <stdio.h>
+
+int main()
+{
+    //int a = 16, b = 44, s = 4;
+    //int av = 12, bv = 6, cv = 3;
+
+    //int a = 32, b = 48, s = 16;
+    //int av = 12, bv = 6, cv = 3;
+
+    int a = 48, b = 64, s = 16;
+    int av = 48, bv = 32, cv = 24;
+
+    printf( "int TrTbl[] = { " );
+    int first = 1;
+    for( int i=0; i<256; i+=s )
+    {
+        if( first ) first = 0; else printf( ", " );
+        if( i < a ) printf( "%i", av );
+        else if( i < b ) printf( "%i", bv );
+        else printf( "%i", cv );
+    }
+    printf( " };\n" );
+}

extra/systrace/build.sh

@@ -0,0 +1,4 @@
+#!/bin/sh
+clang tracy_systrace.c -s -Os -ffunction-sections -fdata-sections -Wl,--gc-sections -fno-stack-protector -Wl,-z,norelro -Wl,--build-id=none -nostdlib -ldl -o tracy_systrace
+strip --strip-all -R .note.gnu.gold-version -R .comment -R .note -R .note.gnu.build-id -R .note.ABI-tag -R .eh_frame -R .eh_frame_hdr -R .gnu.hash -R .gnu.version -R .got tracy_systrace
+sstrip -z tracy_systrace

extra/systrace/tracy_systrace.c

@@ -0,0 +1,53 @@
+#include <fcntl.h>
+#include <poll.h>
+#include <stdio.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <stdlib.h>
+#include <time.h>
+#include <unistd.h>
+#include <dlfcn.h>
+
+enum { BufSize = 64*1024 };
+
+typedef int (*open_t)( const char*, int, ... );
+typedef void (*exit_t)( int );
+typedef int (*poll_t)( struct pollfd*, nfds_t, int timeout );
+typedef int (*nanosleep_t)( const struct timespec*, struct timespec* );
+typedef ssize_t (*read_t)( int, void*, size_t );
+typedef ssize_t (*write_t)( int, const void*, size_t );
+
+void _start()
+{
+    void* libc = dlopen( "libc.so", RTLD_LAZY );
+
+    open_t sym_open = dlsym( libc, "open" );
+    exit_t sym_exit = dlsym( libc, "exit" );
+    poll_t sym_poll = dlsym( libc, "poll" );
+    nanosleep_t sym_nanosleep = dlsym( libc, "nanosleep" );
+    read_t sym_read = dlsym( libc, "read" );
+    write_t sym_write = dlsym( libc, "write" );
+
+    char buf[BufSize];
+
+    int kernelFd = sym_open( "/sys/kernel/debug/tracing/trace_pipe", O_RDONLY );
+    if( kernelFd < 0 ) sym_exit( 0 );
+
+    struct pollfd pfd;
+    pfd.fd = kernelFd;
+    pfd.events = POLLIN | POLLERR;
+
+    struct timespec sleepTime;
+    sleepTime.tv_sec = 0;
+    sleepTime.tv_nsec = 1000 * 1000 * 10;
+
+    for(;;)
+    {
+        while( sym_poll( &pfd, 1, 0 ) <= 0 ) sym_nanosleep( &sleepTime, NULL );
+        const int rd = sym_read( kernelFd, buf, BufSize );
+        if( rd <= 0 ) break;
+        sym_write( STDOUT_FILENO, buf, rd );
+    }
+
+    sym_exit( 0 );
+}