intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 17:13:29 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix add pipe controll before marker profiling

Browse Source 
Resolves: NEO-6065

Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka
2021-07-19 15:07:12 +00:00
committed by Compute-Runtime-Automation
parent 2d6b8061a8
commit dfe578754b
16 changed files with 156 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/source/command_queue/command_queue_hw.h
View File

@@ -470,7 +470,7 @@ class CommandQueueHw : public CommandQueue {
blockedCommandsData = std::make_unique<KernelOperation>(commandStream, *gpgpuCsr.getInternalAllocationStorage());
} else {
commandStream = &getCommandStream<GfxFamily, commandType>(*this, csrDependencies, profilingRequired, perfCountersRequired,
blitEnqueue, multiDispatchInfo, surfaces, numSurfaces, isMarkerWithProfiling);
blitEnqueue, multiDispatchInfo, surfaces, numSurfaces, isMarkerWithProfiling, eventsRequest.numEventsInWaitList > 0);
}
return commandStream;
}

2
opencl/source/command_queue/command_queue_hw_xehp_plus.inl
View File

@@ -49,7 +49,7 @@ bool CommandQueueHw<Family>::isCacheFlushCommand(uint32_t commandType) const {
}
template <>
LinearStream &getCommandStream<Family, CL_COMMAND_RESOURCE_BARRIER>(CommandQueue &commandQueue, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounterCmdsSpace, bool blitEnqueue, const MultiDispatchInfo &multiDispatchInfo, Surface **surfaces, size_t numSurfaces, bool isMarkerWithProfiling) {
LinearStream &getCommandStream<Family, CL_COMMAND_RESOURCE_BARRIER>(CommandQueue &commandQueue, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounterCmdsSpace, bool blitEnqueue, const MultiDispatchInfo &multiDispatchInfo, Surface **surfaces, size_t numSurfaces, bool isMarkerWithProfiling, bool eventsInWaitList) {
size_t expectedSizeCS = 0;
bool usePostSync = false;
if (commandQueue.getGpgpuCommandStreamReceiver().peekTimestampPacketWriteEnabled()) {

4
opencl/source/command_queue/enqueue_common.h
View File

@@ -280,6 +280,10 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
}
if (isMarkerWithProfiling) {
if (numEventsInWaitList == 0) {
PipeControlArgs args(false);
MemorySynchronizationCommands<GfxFamily>::addPipeControl(commandStream, args);
}
processDispatchForMarkerWithTimestampPacket(*this, &commandStream, eventsRequest, csrDeps);
}
} else if (isMarkerWithProfiling) {

6
opencl/source/command_queue/gpgpu_walker.h
View File

@@ -167,7 +167,7 @@ class GpgpuWalkerHelper {
template <typename GfxFamily>
struct EnqueueOperation {
using PIPE_CONTROL = typename GfxFamily::PIPE_CONTROL;
static size_t getTotalSizeRequiredCS(uint32_t eventType, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounters, bool blitEnqueue, CommandQueue &commandQueue, const MultiDispatchInfo &multiDispatchInfo, bool isMarkerWithProfiling);
static size_t getTotalSizeRequiredCS(uint32_t eventType, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounters, bool blitEnqueue, CommandQueue &commandQueue, const MultiDispatchInfo &multiDispatchInfo, bool isMarkerWithProfiling, bool eventsInWaitList);
static size_t getSizeRequiredCS(uint32_t cmdType, bool reserveProfilingCmdsSpace, bool reservePerfCounters, CommandQueue &commandQueue, const Kernel *pKernel, const DispatchInfo &dispatchInfo);
static size_t getSizeRequiredForTimestampPacketWrite();
static size_t getSizeForCacheFlushAfterWalkerCommands(const Kernel &kernel, const CommandQueue &commandQueue);
@@ -180,8 +180,8 @@ struct EnqueueOperation {
template <typename GfxFamily, uint32_t eventType>
LinearStream &getCommandStream(CommandQueue &commandQueue, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace,
bool reservePerfCounterCmdsSpace, bool blitEnqueue, const MultiDispatchInfo &multiDispatchInfo,
Surface **surfaces, size_t numSurfaces, bool isMarkerWithProfiling) {
size_t expectedSizeCS = EnqueueOperation<GfxFamily>::getTotalSizeRequiredCS(eventType, csrDeps, reserveProfilingCmdsSpace, reservePerfCounterCmdsSpace, blitEnqueue, commandQueue, multiDispatchInfo, isMarkerWithProfiling);
Surface **surfaces, size_t numSurfaces, bool isMarkerWithProfiling, bool eventsInWaitList) {
size_t expectedSizeCS = EnqueueOperation<GfxFamily>::getTotalSizeRequiredCS(eventType, csrDeps, reserveProfilingCmdsSpace, reservePerfCounterCmdsSpace, blitEnqueue, commandQueue, multiDispatchInfo, isMarkerWithProfiling, eventsInWaitList);
return commandQueue.getCS(expectedSizeCS);
}

5
opencl/source/command_queue/gpgpu_walker_base.inl
View File

@@ -178,7 +178,7 @@ size_t GpgpuWalkerHelper<GfxFamily>::getSizeForWaDisableRccRhwoOptimization(cons
}
template <typename GfxFamily>
size_t EnqueueOperation<GfxFamily>::getTotalSizeRequiredCS(uint32_t eventType, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounters, bool blitEnqueue, CommandQueue &commandQueue, const MultiDispatchInfo &multiDispatchInfo, bool isMarkerWithProfiling) {
size_t EnqueueOperation<GfxFamily>::getTotalSizeRequiredCS(uint32_t eventType, const CsrDependencies &csrDeps, bool reserveProfilingCmdsSpace, bool reservePerfCounters, bool blitEnqueue, CommandQueue &commandQueue, const MultiDispatchInfo &multiDispatchInfo, bool isMarkerWithProfiling, bool eventsInWaitlist) {
size_t expectedSizeCS = 0;
auto &hwInfo = commandQueue.getDevice().getHardwareInfo();
auto &commandQueueHw = static_cast<CommandQueueHw<GfxFamily> &>(commandQueue);
@@ -207,6 +207,9 @@ size_t EnqueueOperation<GfxFamily>::getTotalSizeRequiredCS(uint32_t eventType, c
expectedSizeCS += TimestampPacketHelper::getRequiredCmdStreamSize<GfxFamily>(csrDeps);
expectedSizeCS += EnqueueOperation<GfxFamily>::getSizeRequiredForTimestampPacketWrite();
if (isMarkerWithProfiling) {
if (!eventsInWaitlist) {
expectedSizeCS += MemorySynchronizationCommands<GfxFamily>::getSizeForSinglePipeControl();
}
expectedSizeCS += 4 * EncodeStoreMMIO<GfxFamily>::size;
}
} else if (isMarkerWithProfiling) {