Intel USM Properly annotate indirect kernel access

Have to annotate kernels with all types CL_KERNEL_EXEC_INFO_INDIRECT_HOST_ACCESS_INTEL CL_KERNEL_EXEC_INFO_INDIRECT_DEVICE_ACCESS_INTEL CL_KERNEL_EXEC_INFO_INDIRECT_SHARED_ACCESS_INTEL since a kernel might have allocations of varied types
CHIP-SPV · Aug 16, 2024 · 7664b80 · 7664b80
1 parent fc6d412
commit 7664b80
Show file tree

Hide file tree

Showing 2 changed files with 19 additions and 14 deletions.
diff --git a/src/backend/OpenCL/CHIPBackendOpenCL.cc b/src/backend/OpenCL/CHIPBackendOpenCL.cc
@@ -281,6 +281,19 @@ annotateIndirectPointers(const CHIPContextOpenCL &Ctx,
                                    AnnotationList.size() * sizeof(void *),
                                    AnnotationList.data());
     CHIPERR_CHECK_LOG_AND_THROW_TABLE(clSetKernelExecInfo);
+
+    if (Ctx.getAllocStrategy() == AllocationStrategy::IntelUSM) {
+      cl_bool param = CL_TRUE;
+      clStatus = clSetKernelExecInfo(KernelAPIHandle, CL_KERNEL_EXEC_INFO_INDIRECT_HOST_ACCESS_INTEL,
+                                     sizeof(cl_bool), &param);
+      CHIPERR_CHECK_LOG_AND_THROW_TABLE(clSetKernelExecInfo);
+      clStatus = clSetKernelExecInfo(KernelAPIHandle, CL_KERNEL_EXEC_INFO_INDIRECT_DEVICE_ACCESS_INTEL,
+                                     sizeof(cl_bool), &param);
+      CHIPERR_CHECK_LOG_AND_THROW_TABLE(clSetKernelExecInfo);
+      clStatus = clSetKernelExecInfo(KernelAPIHandle, CL_KERNEL_EXEC_INFO_INDIRECT_SHARED_ACCESS_INTEL,
+                                     sizeof(cl_bool), &param);
+      CHIPERR_CHECK_LOG_AND_THROW_TABLE(clSetKernelExecInfo); 
+    }
   }
 
   return AllocKeepAlives;
@@ -1259,13 +1272,8 @@ CHIPQueueOpenCL::launchImpl(chipstar::ExecItem *ExecItem) {
   LOCK(Backend->DubiousLockOpenCL);
 #endif
 
-  std::unique_ptr<std::vector<std::shared_ptr<void>>> AllocationsToKeepAlive;
-  // Disable annotation for CPU device. Runtime errror.
-  // https://community.intel.com/t5/OpenCL-for-CPU/OpenCL-Using-USM-on-Intel-CPU-Runtime-along-with/m-p/1621136#M7350
-  if (ChipEnvVars.getDevice().getType() != DeviceType::CPU) {
-    AllocationsToKeepAlive = annotateIndirectPointers(
-        *OclContext, Kernel->getModule()->getInfo(), KernelHandle);
-  }
+  auto AllocationsToKeepAlive = annotateIndirectPointers(
+      *OclContext, Kernel->getModule()->getInfo(), KernelHandle);
 
   auto [EventsToWait, EventLocks] = getSyncQueuesLastEvents(LaunchEvent, false);
   std::vector<cl_event> SyncQueuesEventHandles = getOpenCLHandles(EventsToWait);
@@ -1891,7 +1899,8 @@ void CHIPBackendOpenCL::initializeImpl() {
   std::vector<cl::Device> SupportedDevices;
   std::vector<cl::Device> Dev;
   clStatus = SelectedPlatform.getDevices(SelectedDevType, &Dev);
-  CHIPERR_CHECK_LOG_AND_THROW_TABLE(clGetPlatformIDs);
+  CHIPERR_CHECK_LOG_AND_THROW_TABLE(
+      clGetDeviceIDs); // C equivalent of getDevices
 
   for (auto D : Dev) {
 

diff --git a/src/backend/OpenCL/clHipErrorConversion.hh b/src/backend/OpenCL/clHipErrorConversion.hh
@@ -94,7 +94,8 @@ const std::unordered_map<void *, cl_hip_error_map_t> CL_HIP_ERROR_MAPS = {
       {CL_OUT_OF_RESOURCES, hipErrorOutOfMemory},
       {CL_MEM_OBJECT_ALLOCATION_FAILURE, hipErrorOutOfMemory},
       {CL_INVALID_EVENT_WAIT_LIST, hipErrorInvalidResourceHandle},
-      {CL_OUT_OF_HOST_MEMORY, hipErrorOutOfMemory}}},
+      {CL_OUT_OF_HOST_MEMORY, hipErrorOutOfMemory},
+      {CL_INVALID_OPERATION, hipErrorInvalidValue}}},
 
     {(void *)&clEnqueueReadBuffer,
      {{CL_SUCCESS, hipSuccess},
@@ -447,11 +448,6 @@ const std::unordered_map<void *, cl_hip_error_map_t> CL_HIP_ERROR_MAPS = {
       {CL_INVALID_EVENT_WAIT_LIST, hipErrorInvalidResourceHandle},
       {CL_OUT_OF_HOST_MEMORY, hipErrorOutOfMemory}}},
 
-    {(void *)&clGetPlatformIDs,
-     {{CL_SUCCESS, hipSuccess},
-      {CL_INVALID_VALUE, hipErrorInvalidValue},
-      {CL_OUT_OF_HOST_MEMORY, hipErrorOutOfMemory}}},
-
     {(void *)&clCreateProgramWithIL,
      {{CL_SUCCESS, hipSuccess},
       {CL_INVALID_CONTEXT, hipErrorInvalidResourceHandle},