Fix confusion in hipFuncGetAttribute()

Cuda shared == OpenCL local

Cuda local == OpenCL private

Change-Id: I5a204945ecde35919b9e9def20bbb2662fffea2b


[ROCm/hip commit: a373538d72]

projects/hip/rocclr/hip_module.cpp

@@ -314,8 +314,7 @@ hipError_t hipFuncGetAttribute(int* value, hipFunction_attribute attrib, hipFunc
 
   switch(attrib) {
     case HIP_FUNC_ATTRIBUTE_SHARED_SIZE_BYTES:
-      *value = static_cast<int>(wrkGrpInfo->localMemSize_
+      *value = static_cast<int>(wrkGrpInfo->localMemSize_);
-                      - wrkGrpInfo->privateMemSize_);
       break;
     case HIP_FUNC_ATTRIBUTE_MAX_THREADS_PER_BLOCK:
       *value = static_cast<int>(wrkGrpInfo->wavefrontPerSIMD_
@@ -325,7 +324,7 @@ hipError_t hipFuncGetAttribute(int* value, hipFunction_attribute attrib, hipFunc
       *value = 0;
       break;
     case HIP_FUNC_ATTRIBUTE_LOCAL_SIZE_BYTES:
-      *value = static_cast<int>(wrkGrpInfo->localMemSize_);
+      *value = static_cast<int>(wrkGrpInfo->privateMemSize_);
       break;
     case HIP_FUNC_ATTRIBUTE_NUM_REGS:
       *value = static_cast<int>(wrkGrpInfo->availableGPRs_);
@@ -340,7 +339,7 @@ hipError_t hipFuncGetAttribute(int* value, hipFunction_attribute attrib, hipFunc
       *value = 0;
       break;
     case HIP_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES:
-      *value = static_cast<int>(wrkGrpInfo->availableLDSSize_);
+      *value = static_cast<int>(wrkGrpInfo->availableLDSSize_ - wrkGrpInfo->localMemSize_);
       break;
     case HIP_FUNC_ATTRIBUTE_PREFERRED_SHARED_MEMORY_CARVEOUT:
       *value = 0;

projects/hip/rocclr/hip_platform.cpp

@@ -310,8 +310,9 @@ bool ihipGetFuncAttributes(const char* func_name, amd::Program* program, hipFunc
   }
 
   const device::Kernel::WorkGroupInfo* wginfo = it->second->workGroupInfo();
-  func_attr->localSizeBytes = wginfo->localMemSize_;
+  func_attr->localSizeBytes = wginfo->privateMemSize_;
-  func_attr->sharedSizeBytes = wginfo->size_;
+  func_attr->sharedSizeBytes = wginfo->localMemSize_;
+  func_attr->maxDynamicSharedSizeBytes = wginfo->availableLDSSize_ - wginfo->localMemSize_;
   func_attr->maxThreadsPerBlock = wginfo->wavefrontSize_;
   func_attr->numRegs = wginfo->usedVGPRs_;