Do not prefer copy engine for local to local transfers.

Execution Units are faster.
Signed-off-by: Michal Mrozek <[email protected]>

Author: MichalMrozek

opencl/source/helpers/cl_hw_helper_pvc_and_later.inl

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2021 Intel Corporation
+ * Copyright (C) 2021-2022 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -9,7 +9,7 @@
 
 template <>
 inline bool ClHwHelperHw<Family>::preferBlitterForLocalToLocalTransfers() const {
-    return true;
+    return false;
 }
 
 template <>

opencl/test/unit_test/command_queue/command_queue_tests_pvc_and_later.cpp

@@ -285,8 +285,7 @@ HWTEST2_F(BcsCsrSelectionCommandQueueTests, givenBcsSelectedWithForceBcsEngineIn
         dstGraphicsAllocation.memoryPool = MemoryPool::LocalMemory;
         CsrSelectionArgs args{CL_COMMAND_COPY_BUFFER, &srcMemObj, &dstMemObj, 0u, nullptr};
         CommandStreamReceiver &selectedCsr = queue->selectCsrForBuiltinOperation(args);
-        EXPECT_EQ(queue->getBcsCommandStreamReceiver(linkBcsType), &selectedCsr);
-        EXPECT_EQ(linkBcsType, selectedCsr.getOsContext().getEngineType());
+        EXPECT_EQ(&queue->getGpgpuCommandStreamReceiver(), &selectedCsr);
     }
 }
 
@@ -383,12 +382,11 @@ HWTEST2_F(BcsCsrSelectionCommandQueueTests, givenOneBcsEngineInQueueWhenSelectin
         dstGraphicsAllocation.memoryPool = MemoryPool::LocalMemory;
         CsrSelectionArgs args{CL_COMMAND_COPY_BUFFER, &srcMemObj, &dstMemObj, 0u, nullptr};
         CommandStreamReceiver &selectedCsr = queue->selectCsrForBuiltinOperation(args);
-        EXPECT_EQ(queue->getBcsCommandStreamReceiver(linkBcsType), &selectedCsr);
-        EXPECT_EQ(linkBcsType, selectedCsr.getOsContext().getEngineType());
+        EXPECT_EQ(&queue->getGpgpuCommandStreamReceiver(), &selectedCsr);
     }
 }
 
-HWTEST2_F(BcsCsrSelectionCommandQueueTests, givenMultipleEnginesInQueueWhenSelectingCsrForLocalToLocalOperationThenSelectProperBcsCsr, IsAtLeastXeHpcCore) {
+HWTEST2_F(BcsCsrSelectionCommandQueueTests, givenMultipleEnginesInQueueWhenSelectingCsrForLocalToLocalOperationThenSelectProperGpGpuCsr, IsAtLeastXeHpcCore) {
     DebugManagerStateRestore restore{};
     DebugManager.flags.EnableBlitterForEnqueueOperations.set(1);
 
@@ -416,8 +414,7 @@ HWTEST2_F(BcsCsrSelectionCommandQueueTests, givenMultipleEnginesInQueueWhenSelec
             aub_stream::ENGINE_BCS8,
         });
         CommandStreamReceiver &selectedCsr = queue->selectCsrForBuiltinOperation(args);
-        EXPECT_EQ(queue->getBcsCommandStreamReceiver(aub_stream::ENGINE_BCS), &selectedCsr);
-        EXPECT_EQ(aub_stream::ENGINE_BCS, selectedCsr.getOsContext().getEngineType());
+        EXPECT_EQ(&queue->getGpgpuCommandStreamReceiver(), &selectedCsr);
     }
     {
         auto queue = createQueueWithEngines({
@@ -427,8 +424,7 @@ HWTEST2_F(BcsCsrSelectionCommandQueueTests, givenMultipleEnginesInQueueWhenSelec
             aub_stream::ENGINE_BCS8,
         });
         CommandStreamReceiver &selectedCsr = queue->selectCsrForBuiltinOperation(args);
-        EXPECT_EQ(queue->getBcsCommandStreamReceiver(aub_stream::ENGINE_BCS5), &selectedCsr);
-        EXPECT_EQ(aub_stream::ENGINE_BCS5, selectedCsr.getOsContext().getEngineType());
+        EXPECT_EQ(&queue->getGpgpuCommandStreamReceiver(), &selectedCsr);
     }
 }
 
@@ -503,4 +499,4 @@ HWTEST2_F(OoqCommandQueueHwBlitTest, givenBarrierBeforeFirstKernelWhenEnqueueNDR
     const auto memFenceItor = find<MI_MEM_FENCE *>(memFenceStateItor, ccsHwParser.cmdList.end());
     EXPECT_NE(ccsHwParser.cmdList.end(), memFenceItor);
     EXPECT_NE(ccsHwParser.cmdList.end(), memFenceStateItor);
-}
+}

opencl/test/unit_test/xe_hpc_core/hw_helper_tests_xe_hpc_core.cpp

@@ -919,8 +919,8 @@ XE_HPC_CORETEST_F(HwHelperTestsXeHpcCore, WhenCheckingSipWAThenFalseIsReturned)
     EXPECT_FALSE(HwHelper::get(renderCoreFamily).isSipWANeeded(*defaultHwInfo));
 }
 
-XE_HPC_CORETEST_F(HwHelperTestsXeHpcCore, WhenCheckingPreferenceForBlitterForLocalToLocalTransfersThenReturnTrue) {
-    EXPECT_TRUE(ClHwHelper::get(renderCoreFamily).preferBlitterForLocalToLocalTransfers());
+XE_HPC_CORETEST_F(HwHelperTestsXeHpcCore, WhenCheckingPreferenceForBlitterForLocalToLocalTransfersThenReturnFalse) {
+    EXPECT_FALSE(ClHwHelper::get(renderCoreFamily).preferBlitterForLocalToLocalTransfers());
 }
 
 XE_HPC_CORETEST_F(HwHelperTestsXeHpcCore, givenBdA0WhenBcsSubDeviceSupportIsCheckedThenReturnFalse) {