[PI][UR][HIP][CUDA][L0][NCPU] Pull in latest UR fixes. (#11155)

Fetch the latest revision of unified runtime:

Notable changes
* Some command buffer entry-points have been renamed, also 2 additions
of membufferFill and USMFill
* UrInit/urTearDown have been removed - replaced with loader only
versions.

Author: veselypeta

adapter.cpp

@@ -20,15 +20,6 @@ struct ur_adapter_handle_t_ {
 
 ur_adapter_handle_t_ adapter{};
 
-UR_APIEXPORT ur_result_t UR_APICALL urInit(ur_device_init_flags_t,
-                                           ur_loader_config_handle_t) {
-  return UR_RESULT_SUCCESS;
-}
-
-UR_APIEXPORT ur_result_t UR_APICALL urTearDown(void *) {
-  return UR_RESULT_SUCCESS;
-}
-
 UR_APIEXPORT ur_result_t UR_APICALL
 urAdapterGet(uint32_t NumEntries, ur_adapter_handle_t *phAdapters,
              uint32_t *pNumAdapters) {

command_buffer.cpp

@@ -73,7 +73,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendKernelLaunchExp(
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
 
-UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMemcpyUSMExp(
+UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendUSMMemcpyExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, void *pDst, const void *pSrc,
     size_t size, uint32_t numSyncPointsInWaitList,
     const ur_exp_command_buffer_sync_point_t *pSyncPointWaitList,
@@ -91,7 +91,16 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMemcpyUSMExp(
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
 
-UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMembufferCopyExp(
+UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendUSMFillExp(
+    ur_exp_command_buffer_handle_t, void *, const void *, size_t, size_t,
+    uint32_t, const ur_exp_command_buffer_sync_point_t *,
+    ur_exp_command_buffer_sync_point_t *) {
+  detail::ur::die("Experimental Command-buffer feature is not "
+                  "implemented for CUDA adapter.");
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}
+
+UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMemBufferCopyExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, ur_mem_handle_t hSrcMem,
     ur_mem_handle_t hDstMem, size_t srcOffset, size_t dstOffset, size_t size,
     uint32_t numSyncPointsInWaitList,
@@ -112,7 +121,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMembufferCopyExp(
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
 
-UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMembufferCopyRectExp(
+UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMemBufferCopyRectExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, ur_mem_handle_t hSrcMem,
     ur_mem_handle_t hDstMem, ur_rect_offset_t srcOrigin,
     ur_rect_offset_t dstOrigin, ur_rect_region_t region, size_t srcRowPitch,
@@ -140,7 +149,7 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMembufferCopyRectExp(
 }
 
 UR_APIEXPORT
-ur_result_t UR_APICALL urCommandBufferAppendMembufferWriteExp(
+ur_result_t UR_APICALL urCommandBufferAppendMemBufferWriteExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, ur_mem_handle_t hBuffer,
     size_t offset, size_t size, const void *pSrc,
     uint32_t numSyncPointsInWaitList,
@@ -161,7 +170,7 @@ ur_result_t UR_APICALL urCommandBufferAppendMembufferWriteExp(
 }
 
 UR_APIEXPORT
-ur_result_t UR_APICALL urCommandBufferAppendMembufferReadExp(
+ur_result_t UR_APICALL urCommandBufferAppendMemBufferReadExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, ur_mem_handle_t hBuffer,
     size_t offset, size_t size, void *pDst, uint32_t numSyncPointsInWaitList,
     const ur_exp_command_buffer_sync_point_t *pSyncPointWaitList,
@@ -181,7 +190,7 @@ ur_result_t UR_APICALL urCommandBufferAppendMembufferReadExp(
 }
 
 UR_APIEXPORT
-ur_result_t UR_APICALL urCommandBufferAppendMembufferWriteRectExp(
+ur_result_t UR_APICALL urCommandBufferAppendMemBufferWriteRectExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, ur_mem_handle_t hBuffer,
     ur_rect_offset_t bufferOffset, ur_rect_offset_t hostOffset,
     ur_rect_region_t region, size_t bufferRowPitch, size_t bufferSlicePitch,
@@ -209,7 +218,7 @@ ur_result_t UR_APICALL urCommandBufferAppendMembufferWriteRectExp(
 }
 
 UR_APIEXPORT
-ur_result_t UR_APICALL urCommandBufferAppendMembufferReadRectExp(
+ur_result_t UR_APICALL urCommandBufferAppendMemBufferReadRectExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, ur_mem_handle_t hBuffer,
     ur_rect_offset_t bufferOffset, ur_rect_offset_t hostOffset,
     ur_rect_region_t region, size_t bufferRowPitch, size_t bufferSlicePitch,
@@ -251,3 +260,12 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferEnqueueExp(
                   "implemented for CUDA adapter.");
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
+
+UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferAppendMemBufferFillExp(
+    ur_exp_command_buffer_handle_t, ur_mem_handle_t, const void *, size_t,
+    size_t, size_t, uint32_t, const ur_exp_command_buffer_sync_point_t *,
+    ur_exp_command_buffer_sync_point_t *) {
+  detail::ur::die("Experimental Command-buffer feature is not "
+                  "implemented for CUDA adapter.");
+  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
+}

ur_interface_loader.cpp

@@ -200,8 +200,6 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetGlobalProcAddrTable(
   if (UR_RESULT_SUCCESS != result) {
     return result;
   }
-  pDdiTable->pfnInit = urInit;
-  pDdiTable->pfnTearDown = urTearDown;
   pDdiTable->pfnAdapterGet = urAdapterGet;
   pDdiTable->pfnAdapterRelease = urAdapterRelease;
   pDdiTable->pfnAdapterRetain = urAdapterRetain;
@@ -278,17 +276,19 @@ UR_DLLEXPORT ur_result_t UR_APICALL urGetCommandBufferExpProcAddrTable(
   pDdiTable->pfnReleaseExp = urCommandBufferReleaseExp;
   pDdiTable->pfnFinalizeExp = urCommandBufferFinalizeExp;
   pDdiTable->pfnAppendKernelLaunchExp = urCommandBufferAppendKernelLaunchExp;
-  pDdiTable->pfnAppendMemcpyUSMExp = urCommandBufferAppendMemcpyUSMExp;
-  pDdiTable->pfnAppendMembufferCopyExp = urCommandBufferAppendMembufferCopyExp;
-  pDdiTable->pfnAppendMembufferCopyRectExp =
-      urCommandBufferAppendMembufferCopyRectExp;
-  pDdiTable->pfnAppendMembufferReadExp = urCommandBufferAppendMembufferReadExp;
-  pDdiTable->pfnAppendMembufferReadRectExp =
-      urCommandBufferAppendMembufferReadRectExp;
-  pDdiTable->pfnAppendMembufferWriteExp =
-      urCommandBufferAppendMembufferWriteExp;
-  pDdiTable->pfnAppendMembufferWriteRectExp =
-      urCommandBufferAppendMembufferWriteRectExp;
+  pDdiTable->pfnAppendUSMMemcpyExp = urCommandBufferAppendUSMMemcpyExp;
+  pDdiTable->pfnAppendUSMFillExp = urCommandBufferAppendUSMFillExp;
+  pDdiTable->pfnAppendMemBufferCopyExp = urCommandBufferAppendMemBufferCopyExp;
+  pDdiTable->pfnAppendMemBufferCopyRectExp =
+      urCommandBufferAppendMemBufferCopyRectExp;
+  pDdiTable->pfnAppendMemBufferReadExp = urCommandBufferAppendMemBufferReadExp;
+  pDdiTable->pfnAppendMemBufferReadRectExp =
+      urCommandBufferAppendMemBufferReadRectExp;
+  pDdiTable->pfnAppendMemBufferWriteExp =
+      urCommandBufferAppendMemBufferWriteExp;
+  pDdiTable->pfnAppendMemBufferWriteRectExp =
+      urCommandBufferAppendMemBufferWriteRectExp;
+  pDdiTable->pfnAppendMemBufferFillExp = urCommandBufferAppendMemBufferFillExp;
   pDdiTable->pfnEnqueueExp = urCommandBufferEnqueueExp;
 
   return retVal;