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

P4 to Git Change 1537141 by gandryey@gera-w8 on 2018/04/05 11:56:20

Browse Source 
SWDEV-149190 - [CQE OCL][DTB-BLOCKER][WIN][LNX][QR]observed soft hang while running oclruntime sub-test of ocltst due to faulty CL#1529531
	- Pal::PerfExperiment can't works with extra offsets for allocations. Thus disable suballocation for the thread trace.
	- Minor clean-ups: make sure the profiling end is called for threadtrace commands; call buf release on mem copy in the program loader; add comments to the scratch buffer alignment.

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#80 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palprogram.cpp#60 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palthreadtrace.cpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#85 edit
This commit is contained in:
foreman
2018-04-05 12:17:12 -04:00
parent 01e8d43d00
commit 1d972ed59a
4 changed files with 12 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rocclr/runtime/device/pal/palthreadtrace.cpp
+5 -2
View File
@@ -58,8 +58,11 @@ bool PalThreadTraceReference::finalize() {
// Acquire GPU memory for the query from the pool and bind it.
Pal::GpuMemoryRequirements gpuMemReqs = {};
iPerf()->GetGpuMemoryRequirements(&gpuMemReqs);
memory_ = new Memory(gpu().dev(), amd::alignUp(gpuMemReqs.size, gpuMemReqs.alignment));
// Note: Pal:::PerfExperiments can't apply the runtime offsets for suballocaiotns,
// thus disable suballocation.
Pal::gpusize bufSize = std::max(gpuMemReqs.size,
gpu_.dev().settings().subAllocationMaxSize_ + gpuMemReqs.alignment);
memory_ = new Memory(gpu().dev(), amd::alignUp(bufSize, gpuMemReqs.alignment));
if (nullptr == memory_) {
return false;