Scarfski/filament
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

improve JobSystem::parallel_for + minor optimizations

Browse Source 
- parallel_for doesn't use recursion anymore to create the "leaf"
jobs, this is now done linearly on N thread (one thread per CPU).
This uses less stack space, and reduces miss-predicted branches.

- remove almost all SYSTRACE calls because they have a huge impact
on things like parallel_for() and are misleading. They can be
enabled again by setting HEAVY_SYSTRACE to true.
This commit is contained in:
Pixelflinger
2019-07-03 15:28:52 -07:00
committed by Mathias Agopian
parent 2df639133b
commit 8170ca7cd1
3 changed files with 66 additions and 56 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
libs/utils/src/JobSystem.cpp
View File

@@ -21,7 +21,6 @@
// when SYSTRACE_TAG_JOBSYSTEM is used, enables even heavier systraces
#define HEAVY_SYSTRACE 0
#include <utils/JobSystem.h>
#include <cmath>
@@ -55,6 +54,16 @@
# define gettid() syscall(SYS_gettid)
#endif
#if HEAVY_SYSTRACE
# define HEAVY_SYSTRACE_CALL() SYSTRACE_CALL()
# define HEAVY_SYSTRACE_NAME(name) SYSTRACE_NAME(name)
# define HEAVY_SYSTRACE_VALUE32(name, v) SYSTRACE_VALUE32(name, v)
#else
# define HEAVY_SYSTRACE_CALL()
# define HEAVY_SYSTRACE_NAME(name)
# define HEAVY_SYSTRACE_VALUE32(name, v)
#endif
namespace utils {
UTILS_DEFINE_TLS(JobSystem::ThreadState *) JobSystem::sThreadState(nullptr);
@@ -202,7 +211,6 @@ inline bool JobSystem::hasJobCompleted(JobSystem::Job const* job) noexcept {
return job->runningJobCount.load(std::memory_order_relaxed) <= 0;
}
template <typename Mutex>
void JobSystem::wait(std::unique_lock<Mutex>& lock) noexcept {
++mWaiterCount;
mWaiterCondition.wait(lock);
@@ -210,9 +218,10 @@ void JobSystem::wait(std::unique_lock<Mutex>& lock) noexcept {
}
void JobSystem::wake() noexcept {
mWaiterLock.lock();
Mutex& lock = mWaiterLock;
lock.lock();
const uint32_t waiterCount = mWaiterCount;
mWaiterLock.unlock();
lock.unlock();
mWaiterCondition.notify_n(waiterCount);
}
@@ -245,7 +254,7 @@ inline JobSystem::ThreadState* JobSystem::getStateToStealFrom(JobSystem::ThreadS
}
JobSystem::Job* JobSystem::steal(JobSystem::ThreadState& state) noexcept {
SYSTRACE_CALL();
HEAVY_SYSTRACE_CALL();
Job* job = nullptr;
do {
ThreadState* const stateToStealFrom = getStateToStealFrom(state);
@@ -257,7 +266,7 @@ JobSystem::Job* JobSystem::steal(JobSystem::ThreadState& state) noexcept {
}
bool JobSystem::execute(JobSystem::ThreadState& state) noexcept {
SYSTRACE_CALL();
HEAVY_SYSTRACE_CALL();
Job* job = pop(state.workQueue);
if (UTILS_UNLIKELY(job == nullptr)) {
@@ -269,10 +278,10 @@ bool JobSystem::execute(JobSystem::ThreadState& state) noexcept {
UTILS_UNUSED_IN_RELEASE
uint32_t activeJobs = mActiveJobs.fetch_sub(1, std::memory_order_relaxed);
assert(activeJobs); // whoops, we were already at 0
SYSTRACE_VALUE32("JobSystem::activeJobs", activeJobs - 1);
HEAVY_SYSTRACE_VALUE32("JobSystem::activeJobs", activeJobs - 1);
if (UTILS_LIKELY(job->function)) {
SYSTRACE_NAME("job->function");
HEAVY_SYSTRACE_NAME("job->function");
job->function(job->storage, *this, job);
}
finish(job);
@@ -305,7 +314,7 @@ void JobSystem::loop(ThreadState* state) noexcept {
UTILS_NOINLINE
void JobSystem::finish(Job* job) noexcept {
SYSTRACE_CALL();
HEAVY_SYSTRACE_CALL();
bool notify = false;
@@ -378,11 +387,7 @@ void JobSystem::release(JobSystem::Job*& job) noexcept {
}
void JobSystem::run(JobSystem::Job*& job, uint32_t flags) noexcept {
#if HEAVY_SYSTRACE
SYSTRACE_CALL();
#else
SYSTRACE_CONTEXT();
#endif
HEAVY_SYSTRACE_CALL();
ThreadState& state(getState());
@@ -393,7 +398,7 @@ void JobSystem::run(JobSystem::Job*& job, uint32_t flags) noexcept {
put(state.workQueue, job);
SYSTRACE_VALUE32("JobSystem::activeJobs", activeJobs + 1);
HEAVY_SYSTRACE_VALUE32("JobSystem::activeJobs", activeJobs + 1);
// wake-up a thread if needed...
if (!(flags & DONT_SIGNAL)) {