From 8e94d568a8dbbd218e2e62f89d6cdeb8e4f2ee76 Mon Sep 17 00:00:00 2001
From: Michal Mrozek <michal.mrozek@intel.com>
Date: Wed, 2 Mar 2022 12:43:02 +0000
Subject: [PATCH] Do not set dcFlush in Immediate dispatch mode.

Some devices do not need dcFlush.
Setting it prevents further optimization of pipe controls which
are not needed.

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
---
 ...and_stream_receiver_flush_task_2_tests.cpp | 44 +++++++++++++++++++
 .../command_stream_receiver_hw_base.inl       |  9 ++--
 2 files changed, 50 insertions(+), 3 deletions(-)

diff --git a/opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_2_tests.cpp b/opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_2_tests.cpp
index e1e8ed4ca8..d5bdea18ea 100644
--- a/opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_2_tests.cpp
+++ b/opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_2_tests.cpp
@@ -80,6 +80,50 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, GivenBlockedKernelNotRequiringDCFl
     buffer->release();
 }
 
+HWTEST_F(CommandStreamReceiverFlushTaskTests, GivenEnableUpdateTaskFromWaitWhenNonBlockingCallIsMadeThenNoPipeControlInsertedOnDevicesWithoutDCFlushRequirements) {
+    DebugManagerStateRestore restorer;
+    DebugManager.flags.UpdateTaskCountFromWait.set(3u);
+    typedef typename FamilyType::PIPE_CONTROL PIPE_CONTROL;
+    MockContext ctx(pClDevice);
+    auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
+    commandStreamReceiver.timestampPacketWriteEnabled = false;
+    CommandQueueHw<FamilyType> commandQueue(&ctx, pClDevice, 0, false);
+    size_t tempBuffer[] = {0, 1, 2};
+    size_t dstBuffer[] = {0, 1, 2};
+    cl_int retVal = 0;
+
+    auto buffer = Buffer::create(&ctx, CL_MEM_USE_HOST_PTR, sizeof(tempBuffer), tempBuffer, retVal);
+
+    commandQueue.enqueueWriteBuffer(buffer, CL_FALSE, 0, sizeof(tempBuffer), dstBuffer, nullptr, 0u, nullptr, 0);
+
+    auto &commandStreamTask = *commandStreamReceiver.lastFlushedCommandStream;
+
+    cmdList.clear();
+    // Parse command list
+    parseCommands<FamilyType>(commandStreamTask, 0);
+
+    auto pipeControlExpected = MemorySynchronizationCommands<FamilyType>::getDcFlushEnable(true, *defaultHwInfo);
+
+    auto itorPC = find<PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
+
+    if (pipeControlExpected) {
+        EXPECT_NE(cmdList.end(), itorPC);
+        if (UnitTestHelper<FamilyType>::isPipeControlWArequired(pDevice->getHardwareInfo())) {
+            itorPC++;
+            itorPC = find<PIPE_CONTROL *>(itorPC, cmdList.end());
+            EXPECT_NE(cmdList.end(), itorPC);
+        }
+
+        // Verify that the dcFlushEnabled bit is set in PC
+        auto pCmdWA = reinterpret_cast<PIPE_CONTROL *>(*itorPC);
+        EXPECT_EQ(MemorySynchronizationCommands<FamilyType>::getDcFlushEnable(true, *defaultHwInfo), pCmdWA->getDcFlushEnable());
+    } else {
+        EXPECT_EQ(cmdList.end(), itorPC);
+    }
+
+    buffer->release();
+}
+
 HWTEST_F(CommandStreamReceiverFlushTaskTests, GivenTaskCsPassedAsCommandStreamParamWhenFlushingTaskThenCompletionStampIsCorrect) {
     CommandQueueHw<FamilyType> commandQueue(nullptr, pClDevice, 0, false);
     auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
diff --git a/shared/source/command_stream/command_stream_receiver_hw_base.inl b/shared/source/command_stream/command_stream_receiver_hw_base.inl
index ef2282622e..302af82324 100644
--- a/shared/source/command_stream/command_stream_receiver_hw_base.inl
+++ b/shared/source/command_stream/command_stream_receiver_hw_base.inl
@@ -187,6 +187,8 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
     }
 
     const auto &hwInfo = peekHwInfo();
+    auto &hwHelper = HwHelper::get(hwInfo.platform.eRenderCoreFamily);
+
     bool updateTag = false;
     if (dispatchFlags.blocking || dispatchFlags.dcFlush || dispatchFlags.guardCommandBufferWithPipeControl) {
         if (this->dispatchMode == DispatchMode::ImmediateDispatch) {
@@ -195,7 +197,10 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
             levelClosed = true;
             //if we guard with ppc, flush dc as well to speed up completion latency
             if (dispatchFlags.guardCommandBufferWithPipeControl) {
-                dispatchFlags.dcFlush = true;
+                const auto &hwInfoConfig = *NEO::HwInfoConfig::get(hwInfo.platform.eProductFamily);
+                if (hwInfoConfig.isDcFlushAllowed()) {
+                    dispatchFlags.dcFlush = true;
+                }
             }
         }
 
@@ -255,8 +260,6 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
                                                                                                             dispatchFlags.pipelineSelectArgs.specialPipelineSelectMode,
                                                                                                             hwInfo);
 
-    auto &hwHelper = HwHelper::get(peekHwInfo().platform.eRenderCoreFamily);
-
     if (dispatchFlags.threadArbitrationPolicy == ThreadArbitrationPolicy::NotPresent) {
         if (this->streamProperties.stateComputeMode.threadArbitrationPolicy.value != -1) {
             // Reuse previous programming