AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

P4 to Git Change 1507569 by gandryey@gera-w8 on 2018/01/24 17:56:10

Browse Source 
SWDEV-142271 - Performance drop is observed in Ocean Surface Simulation of Compubenchcl in 17.50 when compared to 17.Q4.1
	- Rewrite the adaptive mode for waveliimiter. Make sure the performance feedback corresponds to the right wave count. Add the new sampling logic to find the best number, based on average performance.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#295 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuwavelimiter.cpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpuwavelimiter.hpp#10 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.hpp#15 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#71 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#39 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palwavelimiter.cpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palwavelimiter.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.cpp#80 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/command.hpp#88 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#282 edit
This commit is contained in:
foreman
2018-01-24 18:07:37 -05:00
parent 111c3fc45c
commit 1e90e581d6
11 changed files with 136 additions and 257 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/pal/palvirtual.cpp
+5 -2
View File
@@ -204,10 +204,12 @@ void VirtualGPU::Queue::addCmdDoppRef(Pal::IGpuMemory* iMem, bool lastDoppCmd, b
palDoppRefs_.push_back(doppRef);
}
template <bool avoidBarrierSubmit>
uint VirtualGPU::Queue::submit(bool forceFlush) {
cmdCnt_++;
uint id = cmdBufIdCurrent_;
if ((cmdCnt_ > MaxCommands) || forceFlush) {
bool flushCmd = ((cmdCnt_ > MaxCommands) || forceFlush) && !avoidBarrierSubmit;
if (flushCmd) {
if (!flush()) {
return GpuEvent::InvalidID;
}
@@ -1972,7 +1974,8 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
}
dispatchParam.pCpuAqlCode = hsaKernel.cpuAqlCode();
dispatchParam.hsaQueueVa = hsaQueueMem_->vmAddress();
dispatchParam.wavesPerSh = hsaKernel.getWavesPerSH(this);
dispatchParam.wavesPerSh = (enqueueEvent != nullptr) ?
enqueueEvent->profilingInfo().waves_ : 0;
dispatchParam.useAtc = dev().settings().svmFineGrainSystem_ ? true : false;
// Run AQL dispatch in HW
eventBegin(MainEngine);