Enable update task count from wait on DG2

Related-To: NEO-6948

Signed-off-by: Dominik Dabek <[email protected]>

Author: ddabek-i

opencl/source/command_queue/command_queue.cpp

@@ -793,7 +793,7 @@ uint32_t CommandQueue::peekBcsTaskCount(aub_stream::EngineType bcsEngineType) co
 }
 
 bool CommandQueue::isTextureCacheFlushNeeded(uint32_t commandType) const {
-    return commandType == CL_COMMAND_COPY_IMAGE && getGpgpuCommandStreamReceiver().isDirectSubmissionEnabled();
+    return (commandType == CL_COMMAND_COPY_IMAGE || commandType == CL_COMMAND_WRITE_IMAGE) && getGpgpuCommandStreamReceiver().isDirectSubmissionEnabled();
 }
 
 IndirectHeap &CommandQueue::getIndirectHeap(IndirectHeap::Type heapType, size_t minRequiredSize) {

opencl/test/unit_test/command_queue/command_queue_tests.cpp

@@ -423,7 +423,7 @@ HWTEST_F(CommandQueueCommandStreamTest, WhenCheckIsTextureCacheFlushNeededThenRe
     EXPECT_FALSE(cmdQ.isTextureCacheFlushNeeded(CL_COMMAND_COPY_BUFFER_RECT));
 
     for (auto i = CL_COMMAND_NDRANGE_KERNEL; i < CL_COMMAND_RELEASE_GL_OBJECTS; i++) {
-        if (i == CL_COMMAND_COPY_IMAGE) {
+        if (i == CL_COMMAND_COPY_IMAGE || i == CL_COMMAND_WRITE_IMAGE) {
             commandStreamReceiver.directSubmissionAvailable = true;
             EXPECT_TRUE(cmdQ.isTextureCacheFlushNeeded(i));
             commandStreamReceiver.directSubmissionAvailable = false;

opencl/test/unit_test/helpers/hw_helper_tests_dg2_and_later.cpp

@@ -154,6 +154,12 @@ HWTEST2_F(HwHelperTestsDg2AndLater, givenXeHPGAndLaterPlatformWhenCheckingIfUnTy
     EXPECT_TRUE(hwHelper.unTypedDataPortCacheFlushRequired());
 }
 
+HWTEST2_F(HwHelperTestsDg2AndLater, givenHwHelperWhenCheckIsUpdateTaskCountFromWaitSupportedThenReturnsTrue, IsAtLeastXeHpgCore) {
+    auto &hwHelper = HwHelper::get(hardwareInfo.platform.eRenderCoreFamily);
+
+    EXPECT_TRUE(hwHelper.isUpdateTaskCountFromWaitSupported());
+}
+
 using HwInfoConfigTestDg2AndLater = ::testing::Test;
 
 HWTEST2_F(HwInfoConfigTestDg2AndLater, givenDg2AndLaterPlatformWhenAskedIfHeapInLocalMemThenTrueIsReturned, IsAtLeastXeHpgCore) {

shared/source/helpers/hw_helper_dg2_and_later.inl

@@ -52,6 +52,11 @@ void MemorySynchronizationCommands<GfxFamily>::setPipeControlWAFlags(PIPE_CONTRO
     pipeControl.setUnTypedDataPortCacheFlush(true);
 }
 
+template <>
+bool HwHelperHw<Family>::isUpdateTaskCountFromWaitSupported() const {
+    return true;
+}
+
 template <>
 bool HwHelperHw<Family>::unTypedDataPortCacheFlushRequired() const {
     return true;

shared/source/helpers/hw_helper_pvc_and_later.inl

@@ -37,11 +37,6 @@ bool HwHelperHw<Family>::isCooperativeDispatchSupported(const EngineGroupType en
     return true;
 }
 
-template <>
-bool HwHelperHw<Family>::isUpdateTaskCountFromWaitSupported() const {
-    return true;
-}
-
 template <>
 uint32_t HwHelperHw<Family>::adjustMaxWorkGroupCount(uint32_t maxWorkGroupCount, const EngineGroupType engineGroupType,
                                                      const HardwareInfo &hwInfo, bool isEngineInstanced) const {