Frames whose symbol data is shipped inline with the callstack payload
(sel=1, e.g. Lua-side stack entries) were being passed to
GetCanonicalPointer() in the AddCallstackAllocPayload() query loop,
tripping its sel==0 assertion. They have no native pointer to query
and were already registered in callstackFrameMap earlier in the same
function, so just skip them.
Regression from c704f909, which hoisted the per-call-site dedup into
QueryCallstackFrame(). Three of the four updated call sites were
equivalent before and after, because the old guard and the new one
keyed on the same value. The fourth, this one, was not: the old guard
tested the frame as-is and matched the entry inserted a few lines above,
short-circuiting before GetCanonicalPointer() ran. The new guard keys on
PackPointer(addr), so GetCanonicalPointer() must run first to compute
addr, and the assert fires.
The profiler will typically want to send bursts of queries (e.g. 3 queries
to retrieve source location strings, or multiple queries to get all the call
stack frames, etc.).
Each of these queries will be sent immediately, if available space in the
network buffer permits. Each of these sends is a separate syscall.
Remove this and instead batch all queries with the already existing network
buffer overflow handling functionality.
There are two changes to the protocol:
- `QueueMessageLiteral*` were changed and what used to be addresses are now addresses+metadata
- Other messages now send `QueueMessage*Metadata` with added metadata.
This will later be used to store and transmit message sources, level, etc.
This was causing issues in the Infos -> Trace Statistics window as `GetCallstackFrameCount` uses `m_pendingCallstackFrames`. Just in case, init those all those variables where declared instead of constructor.
This provides some instructions and tips for the manual. Also:
* Made the calibration feature a CMake option
* Cleaned up some minor code issues
* Fixed an issue with the calibration
* Incremented patch number
Callstack frames will now have nullptr as the value in the callstackFrameMap
map, as a way to signal that a query for given key is already pending.
Duplicate queries should no longer happen.
@slomp provided alternative implementation, which produced the following
results:
Queries made: 195,778
Duplicate queries skipped: 9,518,910
Co-authored-by: Marcos Slomp <slomp@adobe.com>
This value is not used for anything, it was just a number displayed in
the UI without much meaning to anyone.
Operations on the queue during early init may not work correctly, stopping
some programs from running past the calibration loop.
Instead of crashing when reaching the maximum number of source locations, display an empty source location ( with "???" everywhere).
Keeping the assert for discoverability of the limit in debug, but ensure profiler won't crash later on (or in release).
This PR does the following things:
- Allow to lock the highlighted thread using left mouse click or lane popup menu
- Add more descriptions for wait reasons in View::DecodeContextSwitchReason
- Stores the Wakeup CPU so that we may display it in the CPU data view
- `QueueThreadWakeup` now contains cpu and adjust info (currently unused). Fits in the padding, no struct size change.
- `QueueContextSwitch` now also contains thread priorities information (windows+linux) and CState (windows), but currently unused by the profiler. Fits in the padding, no struct size change.
- Fixes Linux `EventWakeup` which tries to remove PERF_SAMPLE_CALLCHAIN on the wrong member. Set the whole `perf_event_attr` explicitely instead.
- Fixes Windows ReadyThread event being dropped if triggered before the thread switches out (this is confirmed to be normal behaviour, probably due to thread being switched out but context switch actually happening later)
- Bumps tracy file version and protocol versions
