Flush tag update if any heap was allocated

Signed-off-by: Lukasz Jobczyk <[email protected]>

Author: LukaszJobczyk

opencl/test/unit_test/command_stream/command_stream_receiver_flush_task_3_tests.cpp

@@ -578,6 +578,34 @@ HWTEST_F(CommandStreamReceiverFlushTaskTests, givenUpdateTaskCountFromWaitWhenFl
     EXPECT_EQ(0u, mockCsr->peekLatestFlushedTaskCount());
 }
 
+HWTEST_F(CommandStreamReceiverFlushTaskTests, givenUpdateTaskCountFromWaitEnabledAndHeapAllocationDirtyWhenFlushTaskThenTaskCountIsFlushed) {
+    DebugManagerStateRestore restorer;
+    DebugManager.flags.UpdateTaskCountFromWait.set(3);
+
+    auto mockCsr = new MockCsrHw2<FamilyType>(*pDevice->executionEnvironment, pDevice->getRootDeviceIndex(), pDevice->getDeviceBitfield());
+    pDevice->resetCommandStreamReceiver(mockCsr);
+    mockCsr->useNewResourceImplicitFlush = false;
+    mockCsr->useGpuIdleImplicitFlush = false;
+    mockCsr->heapStorageReqiuresRecyclingTag = true;
+
+    DispatchFlags dispatchFlags = DispatchFlagsHelper::createDefaultDispatchFlags();
+    dispatchFlags.preemptionMode = PreemptionHelper::getDefaultPreemptionMode(pDevice->getHardwareInfo());
+    dispatchFlags.guardCommandBufferWithPipeControl = false;
+
+    mockCsr->flushTask(commandStream,
+                       0,
+                       &dsh,
+                       &ioh,
+                       &ssh,
+                       taskLevel,
+                       dispatchFlags,
+                       *pDevice);
+
+    parseCommands<FamilyType>(commandStream);
+    auto itorPipeControl = find<typename FamilyType::PIPE_CONTROL *>(cmdList.begin(), cmdList.end());
+    EXPECT_NE(itorPipeControl, cmdList.end());
+}
+
 HWTEST_F(CommandStreamReceiverFlushTaskTests, givenCsrInDefaultModeWhenFlushTaskIsCalledThenFlushedTaskCountIsModifed) {
     CommandQueueHw<FamilyType> commandQueue(nullptr, pClDevice, 0, false);
     auto &commandStream = commandQueue.getCS(4096u);

shared/source/command_stream/command_stream_receiver.cpp

@@ -540,6 +540,7 @@ IndirectHeap &CommandStreamReceiver::getIndirectHeap(IndirectHeap::Type heapType
     if (heap && heap->getAvailableSpace() < minRequiredSize && heapMemory) {
         internalAllocationStorage->storeAllocation(std::unique_ptr<GraphicsAllocation>(heapMemory), REUSABLE_ALLOCATION);
         heapMemory = nullptr;
+        this->heapStorageReqiuresRecyclingTag = true;
     }
 
     if (!heapMemory) {

shared/source/command_stream/command_stream_receiver.h

@@ -481,6 +481,7 @@ class CommandStreamReceiver {
     bool GSBAFor32BitProgrammed = false;
     bool GSBAStateDirty = true;
     bool bindingTableBaseAddressRequired = false;
+    bool heapStorageReqiuresRecyclingTag = false;
     bool mediaVfeStateDirty = true;
     bool lastVmeSubslicesConfig = false;
     bool stallingCommandsOnNextFlushRequired = false;

shared/source/command_stream/command_stream_receiver_hw_base.inl

@@ -206,7 +206,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
     const auto &hwInfo = peekHwInfo();
     auto &hwHelper = HwHelper::get(hwInfo.platform.eRenderCoreFamily);
 
-    if (dispatchFlags.blocking || dispatchFlags.dcFlush || dispatchFlags.guardCommandBufferWithPipeControl) {
+    if (dispatchFlags.blocking || dispatchFlags.dcFlush || dispatchFlags.guardCommandBufferWithPipeControl || this->heapStorageReqiuresRecyclingTag) {
         if (this->dispatchMode == DispatchMode::ImmediateDispatch) {
             // for ImmediateDispatch we will send this right away, therefore this pipe control will close the level
             // for BatchedSubmissions it will be nooped and only last ppc in batch will be emitted.
@@ -220,6 +220,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTask(
             }
         }
 
+        this->heapStorageReqiuresRecyclingTag = false;
         epiloguePipeControlLocation = ptrOffset(commandStreamTask.getCpuBase(), commandStreamTask.getUsed());
 
         if ((dispatchFlags.outOfOrderExecutionAllowed || timestampPacketWriteEnabled) &&

shared/test/common/mocks/mock_command_stream_receiver.h

@@ -239,6 +239,7 @@ class MockCsrHw2 : public CommandStreamReceiverHw<GfxFamily> {
     using CommandStreamReceiver::dispatchMode;
     using CommandStreamReceiver::feSupportFlags;
     using CommandStreamReceiver::globalFenceAllocation;
+    using CommandStreamReceiver::heapStorageReqiuresRecyclingTag;
     using CommandStreamReceiver::isPreambleSent;
     using CommandStreamReceiver::latestFlushedTaskCount;
     using CommandStreamReceiver::mediaVfeStateDirty;