livie
/
phoenix-firestorm
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

#3488 Reduce locking 

1. 'sActive' variables are atomic, no locks needed
2. Fix trylocks. There are internal locks inside loadMeshLOD so without
checking locks 3 and 4 viewer would be locked on each loadMeshLOD,
potentially making main thread wait for threads to unlock.
master
Andrey Kleshchev 2025-04-21 15:56:01 +03:00 committed by Andrey Kleshchev
parent 9a333e65c4
commit 44bbdc549a
1 changed files with 8 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
indra/newview/llmeshrepository.cpp
View File

@ -1809,42 +1809,36 @@ bool LLMeshRepoThread::fetchMeshPhysicsShape(const LLUUID& mesh_id)
//static
void LLMeshRepoThread::incActiveLODRequests()
{
LLMutexLock lock(gMeshRepo.mThread->mMutex);
++LLMeshRepoThread::sActiveLODRequests;
}
//static
void LLMeshRepoThread::decActiveLODRequests()
{
LLMutexLock lock(gMeshRepo.mThread->mMutex);
--LLMeshRepoThread::sActiveLODRequests;
}
//static
void LLMeshRepoThread::incActiveHeaderRequests()
{
LLMutexLock lock(gMeshRepo.mThread->mMutex);
++LLMeshRepoThread::sActiveHeaderRequests;
}
//static
void LLMeshRepoThread::decActiveHeaderRequests()
{
LLMutexLock lock(gMeshRepo.mThread->mMutex);
--LLMeshRepoThread::sActiveHeaderRequests;
}
//static
void LLMeshRepoThread::incActiveSkinRequests()
{
LLMutexLock lock(gMeshRepo.mThread->mMutex);
++LLMeshRepoThread::sActiveSkinRequests;
}
//static
void LLMeshRepoThread::decActiveSkinRequests()
{
LLMutexLock lock(gMeshRepo.mThread->mMutex);
--LLMeshRepoThread::sActiveSkinRequests;
}
@ -4495,13 +4489,20 @@ void LLMeshRepository::notifyLoadedMeshes()
{
LLMutexTrylock lock1(mMeshMutex);
LLMutexTrylock lock2(mThread->mMutex);
LLMutexTrylock lock3(mThread->mHeaderMutex);
LLMutexTrylock lock4(mThread->mPendingMutex);
static U32 hold_offs(0);
if (! lock1.isLocked() || ! lock2.isLocked())
if (! lock1.isLocked() || ! lock2.isLocked() || ! lock3.isLocked() || ! lock4.isLocked())
{
// If we can't get the locks, skip and pick this up later.
// Eventually thread queue will be free enough
++hold_offs;
sMaxLockHoldoffs = llmax(sMaxLockHoldoffs, hold_offs);
if (hold_offs > 4)
{
LL_WARNS_ONCE() << "High mesh thread holdoff" << LL_ENDL;
}
return;
}
hold_offs = 0;
@ -4598,8 +4599,6 @@ void LLMeshRepository::notifyLoadedMeshes()
std::partial_sort(mPendingRequests.begin(), mPendingRequests.begin() + push_count,
mPendingRequests.end(), PendingRequestBase::CompareScoreGreater());
}
LLMutexTrylock lock3(mThread->mHeaderMutex);
LLMutexTrylock lock4(mThread->mPendingMutex);
while (!mPendingRequests.empty() && push_count > 0)
{
std::unique_ptr<PendingRequestBase>& req_p = mPendingRequests.front();