[SYCL][CUDA] Multiple fixes for the CUDA adapter (#11121)

- Return correct error code in urContextCreateWithNativeHandle
- Add UR_DEVICE_INFO_VIRTUAL_MEMORY_SUPPORT to urDeviceGetInfo
- Move asserts from urEnqueueMemBufferFill to UR validation layers
(commit 2c533e6 on UR repository)
- Make behaviour of urEventSetCallback consistent with other unsupported
entrypoints.
- urProgramGetInfo and urQueueGetInfo now return
UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION for unsupported cases.
- Add entrypoints for urSamplerGetNativeHandle and
urSamplerCreateWithNativeHandle
- Fix UR_USM_ALLOC_INFO_BASE_PTR and UR_USM_ALLOC_INFO_SIZE
implementation.

Author: fabiomestre

context.cpp

@@ -150,7 +150,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urContextCreateWithNativeHandle(
   std::ignore = pProperties;
   std::ignore = phContext;
 
-  return UR_RESULT_ERROR_INVALID_OPERATION;
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
 
 UR_APIEXPORT ur_result_t UR_APICALL urContextSetExtendedDeleter(

device.cpp

@@ -1026,6 +1026,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urDeviceGetInfo(ur_device_handle_t hDevice,
   case UR_DEVICE_INFO_GPU_SUBSLICES_PER_SLICE:
   case UR_DEVICE_INFO_GPU_EU_COUNT_PER_SUBSLICE:
   case UR_DEVICE_INFO_GPU_HW_THREADS_PER_EU:
+  case UR_DEVICE_INFO_VIRTUAL_MEMORY_SUPPORT:
     return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
 
   default:

enqueue.cpp

@@ -731,19 +731,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urEnqueueMemBufferFill(
   UR_ASSERT(size + offset <= hBuffer->Mem.BufferMem.getSize(),
             UR_RESULT_ERROR_INVALID_SIZE);
 
-  auto ArgsAreMultiplesOfPatternSize =
-      (offset % patternSize == 0) || (size % patternSize == 0);
-
-  auto PatternIsValid = (pPattern != nullptr);
-
-  auto PatternSizeIsValid =
-      ((patternSize & (patternSize - 1)) == 0) && // is power of two
-      (patternSize > 0) && (patternSize <= 128);  // falls within valid range
-
-  UR_ASSERT(ArgsAreMultiplesOfPatternSize && PatternIsValid &&
-                PatternSizeIsValid,
-            UR_RESULT_ERROR_INVALID_SIZE);
-
   std::unique_ptr<ur_event_handle_t_> RetImplEvent{nullptr};
 
   try {

event.cpp

@@ -219,7 +219,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urEventSetCallback(ur_event_handle_t,
                                                        ur_execution_info_t,
                                                        ur_event_callback_t,
                                                        void *) {
-  detail::ur::die("Event Callback not implemented in CUDA adapter");
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
 

program.cpp

@@ -315,7 +315,11 @@ urProgramGetInfo(ur_program_handle_t hProgram, ur_program_info_t propName,
   case UR_PROGRAM_INFO_BINARIES:
     return ReturnValue(&hProgram->Binary, 1);
   case UR_PROGRAM_INFO_KERNEL_NAMES:
-    return getKernelNames(hProgram);
+    /* TODO: Add implementation for getKernelNames */
+    UR_ASSERT(getKernelNames(hProgram), UR_RESULT_ERROR_UNSUPPORTED_FEATURE);
+    return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
+  case UR_PROGRAM_INFO_NUM_KERNELS:
+    return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
   default:
     break;
   }

queue.cpp

@@ -311,9 +311,10 @@ UR_APIEXPORT ur_result_t UR_APICALL urQueueGetInfo(ur_queue_handle_t hQueue,
       return UR_RESULT_ERROR_OUT_OF_RESOURCES;
     }
   }
-  default:
+  case UR_QUEUE_INFO_DEVICE_DEFAULT:
+  case UR_QUEUE_INFO_SIZE:
     return UR_RESULT_ERROR_UNSUPPORTED_ENUMERATION;
+  default:
+    return UR_RESULT_ERROR_INVALID_ENUMERATION;
   }
-
-  return UR_RESULT_ERROR_INVALID_ENUMERATION;
 }

sampler.cpp

@@ -93,3 +93,14 @@ urSamplerRelease(ur_sampler_handle_t hSampler) {
 
   return UR_RESULT_SUCCESS;
 }
+
+UR_APIEXPORT ur_result_t UR_APICALL
+urSamplerGetNativeHandle(ur_sampler_handle_t, ur_native_handle_t *) {
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urSamplerCreateWithNativeHandle(
+    ur_native_handle_t, ur_context_handle_t,
+    const ur_sampler_native_properties_t *, ur_sampler_handle_t *) {
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}

ur_interface_loader.cpp

@@ -40,7 +40,7 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetPlatformProcAddrTable(
   pDdiTable->pfnGet = urPlatformGet;
   pDdiTable->pfnGetApiVersion = urPlatformGetApiVersion;
   pDdiTable->pfnGetInfo = urPlatformGetInfo;
-  pDdiTable->pfnGetNativeHandle = nullptr;
+  pDdiTable->pfnGetNativeHandle = urPlatformGetNativeHandle;
   pDdiTable->pfnGetBackendOption = urPlatformGetBackendOption;
   return UR_RESULT_SUCCESS;
 }
@@ -132,9 +132,9 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetSamplerProcAddrTable(
     return result;
   }
   pDdiTable->pfnCreate = urSamplerCreate;
-  pDdiTable->pfnCreateWithNativeHandle = nullptr;
+  pDdiTable->pfnCreateWithNativeHandle = urSamplerCreateWithNativeHandle;
   pDdiTable->pfnGetInfo = urSamplerGetInfo;
-  pDdiTable->pfnGetNativeHandle = nullptr;
+  pDdiTable->pfnGetNativeHandle = urSamplerGetNativeHandle;
   pDdiTable->pfnRelease = urSamplerRelease;
   pDdiTable->pfnRetain = urSamplerRetain;
   return UR_RESULT_SUCCESS;

usm.cpp

@@ -232,21 +232,21 @@ urUSMGetMemAllocInfo(ur_context_handle_t hContext, const void *pMem,
 #endif
     }
     case UR_USM_ALLOC_INFO_BASE_PTR: {
-#if __CUDA_API_VERSION >= 10020
+#if CUDA_VERSION >= 10020
       // CU_POINTER_ATTRIBUTE_RANGE_START_ADDR was introduced in CUDA 10.2
-      unsigned int Value;
-      result = UR_CHECK_ERROR(cuPointerGetAttribute(
-          &Value, CU_POINTER_ATTRIBUTE_RANGE_START_ADDR, (CUdeviceptr)pMem));
-      return ReturnValue(Value);
+      void *Base;
+      Result = UR_CHECK_ERROR(cuPointerGetAttribute(
+          &Base, CU_POINTER_ATTRIBUTE_RANGE_START_ADDR, (CUdeviceptr)pMem));
+      return ReturnValue(Base);
 #else
       return UR_RESULT_ERROR_INVALID_VALUE;
 #endif
     }
     case UR_USM_ALLOC_INFO_SIZE: {
-#if __CUDA_API_VERSION >= 10020
+#if CUDA_VERSION >= 10020
       // CU_POINTER_ATTRIBUTE_RANGE_SIZE was introduced in CUDA 10.2
-      unsigned int Value;
-      result = UR_CHECK_ERROR(cuPointerGetAttribute(
+      size_t Value;
+      Result = UR_CHECK_ERROR(cuPointerGetAttribute(
           &Value, CU_POINTER_ATTRIBUTE_RANGE_SIZE, (CUdeviceptr)pMem));
       return ReturnValue(Value);
 #else