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

Correct HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK query

Browse Source 
We should be returning the max workgroup size calculated by the compiler.

Change-Id: If86590efbb9b291f470bdbe87e5df992e661c539


[ROCm/clr commit: b5f9d2f818]
This commit is contained in:
Vlad Sytchenko
2020-05-08 11:08:08 -04:00
committed by Vladislav Sytchenko
parent f3dffba0fe
commit b2afbd99b3
2 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/hipamd/rocclr/hip_module.cpp
+1 -2
View File
@@ -317,8 +317,7 @@ hipError_t hipFuncGetAttribute(int* value, hipFunction_attribute attrib, hipFunc
*value = static_cast<int>(wrkGrpInfo->localMemSize_);
break;
case HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK:
*value = static_cast<int>(wrkGrpInfo->wavefrontPerSIMD_
* wrkGrpInfo->wavefrontSize_);
*value = static_cast<int>(wrkGrpInfo->size_);
break;
case HIP_FUNC_ATTRIBUTE_CONST_SIZE_BYTES:
*value = 0;
projects/clr/hipamd/rocclr/hip_platform.cpp
+1 -1
View File
@@ -313,7 +313,7 @@ bool ihipGetFuncAttributes(const char* func_name, amd::Program* program, hipFunc
func_attr->localSizeBytes = wginfo->privateMemSize_;
func_attr->sharedSizeBytes = wginfo->localMemSize_;
func_attr->maxDynamicSharedSizeBytes = wginfo->availableLDSSize_ - wginfo->localMemSize_;
func_attr->maxThreadsPerBlock = wginfo->wavefrontSize_;
func_attr->maxThreadsPerBlock = wginfo->size_;
func_attr->numRegs = wginfo->usedVGPRs_;
return true;