lee/renderdoc
1
0
Fork 0
mirror of https://github.com/baldurk/renderdoc.git synced 2026-05-04 17:10:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

Spirv Debugger run jobs in a hot spin loop in a single JobQueue job

Browse Source 
Each JobQueue Job runs SimulationJobHelper which is a while loop looking for threads to simulate
This commit is contained in:
Jake Turner
2025-08-19 12:29:37 +01:00
parent e9e081a55f
commit b536bbaf95
3 changed files with 71 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
renderdoc/driver/shaders/spirv/spirv_debug.h
+7 -3
View File
@@ -787,9 +787,10 @@ private:
void QueueDeviceThreadStep(uint32_t lane);
void ProcessQueuedDeviceThreadSteps();
void QueueJob(uint32_t lane, rdcarray<ShaderDebugState> *ret);
void StepThread(uint32_t lane, StepThreadMode stepMode, rdcarray<ShaderDebugState> *ret);
void InternalStepThread(uint32_t lane, rdcarray<ShaderDebugState> *ret);
void QueueJob(uint32_t lane);
void StepThread(uint32_t lane, StepThreadMode stepMode);
void InternalStepThread(uint32_t lane);
void SimulationJobHelper();
void ProcessQueuedDebugMessages();
void ProcessQueuedOps();
@@ -804,6 +805,7 @@ private:
rdcarray<bool> queuedGpuSampleGatherOps;
rdcarray<bool> queuedDeviceThreadSteps;
rdcarray<ShaderDebugState> *shaderChangesReturn;
rdcarray<int32_t> queuedJobs;
bool retireIDs = true;
ShaderDebugState activeDebugState;
@@ -812,6 +814,8 @@ private:
rdcarray<ShaderVariable *> pendingGpuSampleGatherOpsResults;
uint64_t deviceThreadID;
int32_t simulationFinished;
bool mtSimulation;
};
// this does a 'safe' value assignment, by doing parallel depth-first iteration of both variables
renderdoc/driver/shaders/spirv/spirv_debug_setup.cpp
+63 -13
View File
@@ -40,6 +40,9 @@ RDOC_DEBUG_CONFIG(
RDOC_CONFIG(bool, Vulkan_Debug_EnableShaderDebugMT, true,
"Use multiple threads to run the shader debugger simulation.");
RDOC_DEBUG_CONFIG(bool, Vulkan_Hack_ShaderDebugUsesJobSystemQueue, false,
"Work in progress run shader debugging simulation in job system queue.");
using namespace rdcshaders;
// this could be cleaner if ShaderVariable wasn't a very public struct, but it's not worth it so
@@ -293,6 +296,8 @@ Debugger::Debugger() : deviceThreadID(Threading::GetCurrentID())
Debugger::~Debugger()
{
AtomicStore(&simulationFinished, 1);
Threading::JobSystem::SyncAllJobs();
SAFE_DELETE(apiWrapper);
}
@@ -1028,6 +1033,7 @@ ShaderDebugTrace *Debugger::BeginDebug(DebugAPIWrapper *api, const ShaderStage s
queuedGpuMathOps.resize(threadsInWorkgroup);
queuedGpuSampleGatherOps.resize(threadsInWorkgroup);
pendingLanes.resize(threadsInWorkgroup);
queuedJobs.resize(threadsInWorkgroup);
for(uint32_t i = 0; i < threadsInWorkgroup; i++)
{
workgroup.push_back(ThreadState(*this, global));
@@ -1035,6 +1041,7 @@ ShaderDebugTrace *Debugger::BeginDebug(DebugAPIWrapper *api, const ShaderStage s
queuedGpuMathOps[i] = false;
queuedGpuSampleGatherOps[i] = false;
pendingLanes[i] = false;
queuedJobs[i] = 0;
}
ThreadState &active = GetActiveLane();
@@ -1783,6 +1790,20 @@ ShaderDebugTrace *Debugger::BeginDebug(DebugAPIWrapper *api, const ShaderStage s
ret->samplers = global.samplers;
ret->inputs = active.inputs;
mtSimulation = Vulkan_Debug_EnableShaderDebugMT();
if(threadsInWorkgroup < 4)
mtSimulation = false;
AtomicStore(&simulationFinished, 0);
if(mtSimulation)
{
if(!Vulkan_Hack_ShaderDebugUsesJobSystemQueue())
{
uint32_t countJobs = RDCMIN(threadsInWorkgroup, Threading::JobSystem::GetCountWorkers() / 2U);
for(uint32_t i = 0; i < countJobs; ++i)
Threading::JobSystem::AddJob([this]() { SimulationJobHelper(); });
}
}
return ret;
}
@@ -2640,7 +2661,11 @@ rdcarray<ShaderDebugState> Debugger::ContinueDebug()
// if we've finished, return an empty set to signify that
if(active.Finished())
{
AtomicStore(&simulationFinished, 1);
Threading::JobSystem::SyncAllJobs();
return ret;
}
bool allStepsCompleted = true;
shaderChangesReturn = &ret;
@@ -2698,7 +2723,7 @@ rdcarray<ShaderDebugState> Debugger::ContinueDebug()
}
RDCASSERTEQUAL(thread.activeMask.size(), activeMask.size());
memcpy(thread.activeMask.data(), activeMask.data(), activeMask.size() * sizeof(bool));
QueueJob(lane, &ret);
QueueJob(lane);
}
}
@@ -4579,7 +4604,7 @@ void Debugger::SyncPendingLanes()
pendingLanes[lane] = false;
ThreadState &thread = workgroup[lane];
thread.SetPendingResultReady();
QueueJob(lane, shaderChangesReturn);
QueueJob(lane);
}
}
}
@@ -4675,8 +4700,22 @@ DebugAPIWrapper *Debugger::GetAPIWrapper() const
return apiWrapper;
}
void Debugger::SimulationJobHelper()
{
while(AtomicLoad(&simulationFinished) == 0)
{
for(uint32_t lane = 0; lane < workgroup.size(); ++lane)
{
if(Atomic::CmpExch32(&queuedJobs[lane], 1, 0) == 1)
{
StepThread(lane, StepThreadMode::RUN_MULTIPLE_STEPS);
}
}
};
}
// Called from any thread
void Debugger::StepThread(uint32_t lane, StepThreadMode stepMode, rdcarray<ShaderDebugState> *ret)
void Debugger::StepThread(uint32_t lane, StepThreadMode stepMode)
{
ThreadState &thread = workgroup[lane];
bool isActiveThread = lane == activeLaneIndex;
@@ -4691,7 +4730,7 @@ void Debugger::StepThread(uint32_t lane, StepThreadMode stepMode, rdcarray<Shade
thread.ClearPendingDebugState();
if(isActiveThread)
activeDebugState.stepIndex = curActiveSteps;
InternalStepThread(lane, ret);
InternalStepThread(lane);
thread.ClearPendingDebugState();
}
if(thread.StepNeedsGpuSampleGatherOp())
@@ -4747,7 +4786,7 @@ void Debugger::StepThread(uint32_t lane, StepThreadMode stepMode, rdcarray<Shade
if(simulateStep)
{
RDCASSERTEQUAL(stepMode, StepThreadMode::QUEUE_MULTIPLE_STEPS);
QueueJob(lane, ret);
QueueJob(lane);
return;
}
RDCASSERT(!thread.IsPendingResultPending());
@@ -4755,7 +4794,7 @@ void Debugger::StepThread(uint32_t lane, StepThreadMode stepMode, rdcarray<Shade
}
// Called from any thread
void Debugger::InternalStepThread(uint32_t lane, rdcarray<ShaderDebugState> *ret)
void Debugger::InternalStepThread(uint32_t lane)
{
ThreadState &thread = workgroup[lane];
if(lane == activeLaneIndex)
@@ -4862,7 +4901,7 @@ void Debugger::InternalStepThread(uint32_t lane, rdcarray<ShaderDebugState> *ret
}
}
ret->push_back(activeDebugState);
shaderChangesReturn->push_back(activeDebugState);
{
activeDebugState.callstack.clear();
activeDebugState.changes.clear();
@@ -4886,16 +4925,27 @@ void Debugger::InternalStepThread(uint32_t lane, rdcarray<ShaderDebugState> *ret
}
// Must be called from the replay manager thread (the debugger thread)
void Debugger::QueueJob(uint32_t lane, rdcarray<ShaderDebugState> *ret)
void Debugger::QueueJob(uint32_t lane)
{
CHECK_DEBUGGER_THREAD();
ThreadState &thread = workgroup[lane];
thread.SetStepQueued();
if(Vulkan_Debug_EnableShaderDebugMT())
Threading::JobSystem::AddJob(
[this, lane, ret]() { StepThread(lane, StepThreadMode::RUN_MULTIPLE_STEPS, ret); });
if(mtSimulation)
{
if(Vulkan_Hack_ShaderDebugUsesJobSystemQueue())
{
Threading::JobSystem::AddJob(
[this, lane]() { StepThread(lane, StepThreadMode::RUN_MULTIPLE_STEPS); });
}
else
{
RDCASSERT(Atomic::CmpExch32(&queuedJobs[lane], 0, 1) == 0);
}
}
else
StepThread(lane, StepThreadMode::RUN_SINGLE_STEP, ret);
{
StepThread(lane, StepThreadMode::RUN_SINGLE_STEP);
}
}
// Must be called from the replay manager thread (the debugger thread)
@@ -4939,7 +4989,7 @@ void Debugger::ProcessQueuedDeviceThreadSteps()
ThreadState &thread = workgroup[lane];
thread.SetPendingResultUnknown();
RDCASSERT(thread.IsSimulationStepActive());
StepThread(lane, StepThreadMode::QUEUE_MULTIPLE_STEPS, shaderChangesReturn);
StepThread(lane, StepThreadMode::QUEUE_MULTIPLE_STEPS);
}
}
}
renderdoc/driver/vulkan/vk_shaderdebug.cpp
+1 -1
View File
@@ -6893,7 +6893,6 @@ rdcarray<ShaderDebugState> VulkanReplay::ContinueDebug(ShaderDebugger *debugger)
}
rdcarray<ShaderDebugState> ret = spvDebugger->ContinueDebug();
Threading::JobSystem::SyncAllJobs();
VulkanAPIWrapper *api = (VulkanAPIWrapper *)spvDebugger->GetAPIWrapper();
api->ResetReplay();
@@ -6904,4 +6903,5 @@ rdcarray<ShaderDebugState> VulkanReplay::ContinueDebug(ShaderDebugger *debugger)
void VulkanReplay::FreeDebugger(ShaderDebugger *debugger)
{
delete debugger;
Threading::JobSystem::SyncAllJobs();
}