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

P4 to Git Change 1533139 by todli@todli-win-opencl-kv1 on 2018/03/27 15:01:15

Browse Source 
SWDEV-147368 - Choose proper engine for markerWrite to SDI ExternalPhysical Memory according to last copy engine

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#80 edit
This commit is contained in:
foreman
2018-03-27 15:17:14 -04:00
parent 987c092500
commit 0f6d9fea87
1 changed files with 22 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/pal/palvirtual.cpp
+22 -11
View File
@@ -2542,29 +2542,40 @@ void VirtualGPU::submitSignal(amd::SignalCommand& vcmd) {
pal::Memory* pGpuMemory = dev().getGpuMemory(&vcmd.memory()); pal::Memory* pGpuMemory = dev().getGpuMemory(&vcmd.memory());
GpuEvent gpuEvent; GpuEvent gpuEvent;
eventBegin(MainEngine);
uint32_t value = vcmd.markerValue(); uint32_t value = vcmd.markerValue();
addVmMemory(pGpuMemory);
if (vcmd.type() == CL_COMMAND_WAIT_SIGNAL_AMD) { if (vcmd.type() == CL_COMMAND_WAIT_SIGNAL_AMD) {
eventBegin(MainEngine);
addVmMemory(pGpuMemory);
iCmd()->CmdWaitBusAddressableMemoryMarker(*(pGpuMemory->iMem()), value, 0xFFFFFFFF, iCmd()->CmdWaitBusAddressableMemoryMarker(*(pGpuMemory->iMem()), value, 0xFFFFFFFF,
Pal::CompareFunc::GreaterEqual); Pal::CompareFunc::GreaterEqual);
eventEnd(MainEngine, gpuEvent);
} else if (vcmd.type() == CL_COMMAND_WRITE_SIGNAL_AMD) { } else if (vcmd.type() == CL_COMMAND_WRITE_SIGNAL_AMD) {
EngineType activeEngineID = engineID_;
engineID_ = static_cast<EngineType>(pGpuMemory->getGpuEvent(*this)->engineId_);
// Make sure GPU finished operation and data reached memory before the marker write // Make sure GPU finished operation and data reached memory before the marker write
static constexpr bool FlushL2 = true; static constexpr bool FlushL2 = true;
addBarrier(FlushL2); addBarrier(FlushL2);
// \todo: Implement the right changes in PAL // Workarounds: We had systems where an extra delay was necessary.
// Workarounds: for CP overfetch issues and the lack of SDMA sync
{ {
// Flush CB associated with the DGMA buffer // Flush CB associated with the DGMA buffer
isDone(pGpuMemory->getGpuEvent(*this)); isDone(pGpuMemory->getGpuEvent(*this));
// Make sure SDMA is done on the DGMA buffer
pGpuMemory->wait(*this, true);
} }
iCmd()->CmdUpdateBusAddressableMemoryMarker(*(pGpuMemory->iMem()), value);
eventBegin(engineID_);
queues_[engineID_]->addCmdMemRef(pGpuMemory->memRef());
queues_[engineID_]->iCmd()->
CmdUpdateBusAddressableMemoryMarker(*(pGpuMemory->iMem()), value);
eventEnd(engineID_, gpuEvent);
// Restore the original engine
engineID_ = activeEngineID;
} }
eventEnd(MainEngine, gpuEvent);
// Update the global GPU event // Update the global GPU event
setGpuEvent(gpuEvent); setGpuEvent(gpuEvent);