P4 to Git Change 1077336 by emankov@em-hsa-amd on 2014/09/16 11:13:20

ECR #333753 - Compiler Lib/RT: Fix RT_DEVICE_ENQUEUE & RT_KERNEL_INDEX querying

	tests: pre check-in, compiler ocl conformance tests

	Reviewer: German Andryeyev, Vinay Madhusudan

Affected files ...

... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/v0_8/if_acl.cpp#48 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpukernel.cpp#263 edit


[ROCm/clr commit: 09fb6c961b]

projects/clr/rocclr/compiler/lib/backends/common/v0_8/if_acl.cpp

@@ -2457,10 +2457,10 @@ if_aclQueryInfo(aclCompiler *cl,
            }
     case RT_DEVICE_ENQUEUE: {
            if (size != NULL && ptr == NULL) {
-             (*size) = sizeof(uint32_t);
-           } else if (ptr != NULL && (*size) >= (sizeof(uint32_t))) {
+             (*size) = sizeof(aclMetadata().enqueue_kernel);
+           } else if (ptr != NULL && (*size) >= (sizeof(aclMetadata().enqueue_kernel))) {
                const aclMetadata *md = reinterpret_cast<const aclMetadata*>(roSec);
-              (*reinterpret_cast<uint32_t*>(ptr)) = md->enqueue_kernel;
+               memcpy(ptr, &md->enqueue_kernel, sizeof(md->enqueue_kernel));
            } else {
              success = false;
            }
@@ -2469,10 +2469,10 @@ if_aclQueryInfo(aclCompiler *cl,
     // Temporary approach till the "ldk" instruction is supported.
     case RT_KERNEL_INDEX: {
            if (size != NULL && ptr == NULL) {
-             (*size) = sizeof(uint32_t);
-           } else if (ptr != NULL && (*size) >= (sizeof(uint32_t))) {
+             (*size) = sizeof(aclMetadata().kernel_index);
+           } else if (ptr != NULL && (*size) >= (sizeof(aclMetadata().kernel_index))) {
                const aclMetadata *md = reinterpret_cast<const aclMetadata*>(roSec);
-              (*reinterpret_cast<uint32_t*>(ptr)) = md->kernel_index;
+               memcpy(ptr, &md->kernel_index, sizeof(md->kernel_index));
            } else {
              success = false;
            }

projects/clr/rocclr/runtime/device/gpu/gpukernel.cpp

@@ -3612,20 +3612,21 @@ HSAILKernel::init()
         delete [] aclPrintfList;
     }
 
-    size_t sizeOfDevice;
     bool hasKernelEnqueue = false;
+    size_t sizeOfDeviceEnqueue = sizeof(hasKernelEnqueue);
     error = aclQueryInfo(dev().hsaCompiler(), prog().binaryElf(),
         RT_DEVICE_ENQUEUE, openClKernelName.c_str(),
-        &hasKernelEnqueue, &sizeOfDevice);
+        &hasKernelEnqueue, &sizeOfDeviceEnqueue);
     if (error != ACL_SUCCESS) {
         return false;
     }
     flags_.dynamicParallelism_ = hasKernelEnqueue;
 
     int index = -1;
+    size_t sizeOfIndex = sizeof(index);
     error = aclQueryInfo(dev().hsaCompiler(), prog().binaryElf(),
         RT_KERNEL_INDEX, openClKernelName.c_str(),
-        &index, &sizeOfDevice);
+        &index, &sizeOfIndex);
     if (error != ACL_SUCCESS) {
         return false;
     }