Merge pull request #2682 from Xewar313/add-append-usm-memory-v2-commandbuffer

Add USMMemcpy to v2 command buffer

Author: pbalcer

source/adapters/level_zero/v2/api.cpp

@@ -239,17 +239,6 @@ ur_result_t urBindlessImagesReleaseExternalSemaphoreExp(
   return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
 }
 
-ur_result_t 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,
-    uint32_t NumEventsInWaitList, const ur_event_handle_t *phEventWaitList,
-    ur_exp_command_buffer_sync_point_t *pSyncPoint, ur_event_handle_t *phEvent,
-    ur_exp_command_buffer_command_handle_t *phCommand) {
-  logger::error("{} function not implemented!", __FUNCTION__);
-  return UR_RESULT_ERROR_UNSUPPORTED_FEATURE;
-}
-
 ur_result_t urCommandBufferAppendUSMFillExp(
     ur_exp_command_buffer_handle_t hCommandBuffer, void *pMemory,
     const void *pPattern, size_t patternSize, size_t size,

source/adapters/level_zero/v2/command_buffer.cpp

@@ -138,6 +138,33 @@ ur_result_t urCommandBufferAppendKernelLaunchExp(
   return exceptionToResult(std::current_exception());
 }
 
+ur_result_t 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,
+    uint32_t numEventsInWaitList, const ur_event_handle_t *phEventWaitList,
+    ur_exp_command_buffer_sync_point_t *pSyncPoint, ur_event_handle_t *phEvent,
+    ur_exp_command_buffer_command_handle_t *phCommand) try {
+
+  // the same issue as in urCommandBufferAppendKernelLaunchExp
+  std::ignore = numEventsInWaitList;
+  std::ignore = phEventWaitList;
+  std::ignore = phEvent;
+  // sync mechanic can be ignored, because all lists are in-order
+  std::ignore = numSyncPointsInWaitList;
+  std::ignore = pSyncPointWaitList;
+  std::ignore = pSyncPoint;
+
+  std::ignore = phCommand;
+  // Responsibility of UMD to offload to copy engine
+  UR_CALL(hCommandBuffer->commandListManager.appendUSMMemcpy(
+      false, pDst, pSrc, size, 0, nullptr, nullptr));
+
+  return UR_RESULT_SUCCESS;
+} catch (...) {
+  return exceptionToResult(std::current_exception());
+}
+
 ur_result_t
 urCommandBufferGetInfoExp(ur_exp_command_buffer_handle_t hCommandBuffer,
                           ur_exp_command_buffer_info_t propName,

source/adapters/level_zero/v2/command_list_manager.cpp

@@ -102,6 +102,30 @@ ur_result_t ur_command_list_manager::appendKernelLaunch(
   return UR_RESULT_SUCCESS;
 }
 
+ur_result_t ur_command_list_manager::appendUSMMemcpy(
+    bool blocking, void *pDst, const void *pSrc, size_t size,
+    uint32_t numEventsInWaitList, const ur_event_handle_t *phEventWaitList,
+    ur_event_handle_t *phEvent) {
+  TRACK_SCOPE_LATENCY("ur_command_list_manager::appendUSMMemcpy");
+
+  std::scoped_lock<ur_shared_mutex> lock(this->Mutex);
+
+  auto zeSignalEvent = getSignalEvent(phEvent, UR_COMMAND_USM_MEMCPY);
+
+  auto [pWaitEvents, numWaitEvents] =
+      getWaitListView(phEventWaitList, numEventsInWaitList);
+
+  ZE2UR_CALL(zeCommandListAppendMemoryCopy,
+             (zeCommandList.get(), pDst, pSrc, size, zeSignalEvent,
+              numWaitEvents, pWaitEvents));
+
+  if (blocking) {
+    ZE2UR_CALL(zeCommandListHostSynchronize, (zeCommandList.get(), UINT64_MAX));
+  }
+
+  return UR_RESULT_SUCCESS;
+}
+
 ze_command_list_handle_t ur_command_list_manager::getZeCommandList() {
   return zeCommandList.get();
 }

source/adapters/level_zero/v2/command_list_manager.hpp

@@ -47,6 +47,11 @@ struct ur_command_list_manager : public _ur_object {
                                  const ur_event_handle_t *phEventWaitList,
                                  ur_event_handle_t *phEvent);
 
+  ur_result_t appendUSMMemcpy(bool blocking, void *pDst, const void *pSrc,
+                              size_t size, uint32_t numEventsInWaitList,
+                              const ur_event_handle_t *phEventWaitList,
+                              ur_event_handle_t *phEvent);
+
   ze_command_list_handle_t getZeCommandList();
 
   wait_list_view getWaitListView(const ur_event_handle_t *phWaitEvents,

source/adapters/level_zero/v2/queue_immediate_in_order.cpp

@@ -698,21 +698,9 @@ ur_result_t ur_queue_immediate_in_order_t::enqueueUSMMemcpy(
   // TODO: parametrize latency tracking with 'blocking'
   TRACK_SCOPE_LATENCY("ur_queue_immediate_in_order_t::enqueueUSMMemcpy");
 
-  std::scoped_lock<ur_shared_mutex> lock(this->Mutex);
-
-  auto zeSignalEvent = getSignalEvent(phEvent, UR_COMMAND_USM_MEMCPY);
-
-  auto [pWaitEvents, numWaitEvents] =
-      getWaitListView(phEventWaitList, numEventsInWaitList);
-
-  ZE2UR_CALL(zeCommandListAppendMemoryCopy,
-             (commandListManager.getZeCommandList(), pDst, pSrc, size,
-              zeSignalEvent, numWaitEvents, pWaitEvents));
-
-  if (blocking) {
-    ZE2UR_CALL(zeCommandListHostSynchronize,
-               (commandListManager.getZeCommandList(), UINT64_MAX));
-  }
+  UR_CALL(commandListManager.appendUSMMemcpy(blocking, pDst, pSrc, size,
+                                             numEventsInWaitList,
+                                             phEventWaitList, phEvent));
 
   return UR_RESULT_SUCCESS;
 }