fix: request for task count should enable monitor fence dispatch

Related-To: NEO-10356

Signed-off-by: Zbigniew Zdanowicz <[email protected]>

Author: zzdanowicz

shared/source/command_stream/command_stream_receiver_hw_base.inl

@@ -2238,6 +2238,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::handleImmediateFlushSendBatc
     size_t chainedBatchBufferStartOffset = startFromCsr ? csrStream.getUsed() : 0;
     uint64_t taskStartAddress = immediateCommandStream.getGpuBase() + immediateCommandStreamStart;
     bool hasStallingCmds = (startFromCsr || dispatchFlags.blockingAppend || dispatchFlags.hasStallingCmds);
+    bool dispatchMonitorFence = dispatchFlags.blockingAppend || dispatchFlags.requireTaskCountUpdate;
 
     constexpr bool immediateLowPriority = false;
     const QueueThrottle immediateThrottle = getThrottleFromPowerSavingUint(this->getUmdPowerHintValue());
@@ -2246,7 +2247,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::handleImmediateFlushSendBatc
     BatchBuffer batchBuffer{streamToSubmit.getGraphicsAllocation(), startOffset, chainedBatchBufferStartOffset, taskStartAddress, chainedBatchBuffer,
                             immediateLowPriority, immediateThrottle, immediateSliceCount,
                             streamToSubmit.getUsed(), &streamToSubmit, flushData.endPtr, this->getNumClients(), hasStallingCmds,
-                            dispatchFlags.hasRelaxedOrderingDependencies, dispatchFlags.blockingAppend, false};
+                            dispatchFlags.hasRelaxedOrderingDependencies, dispatchMonitorFence, false};
     updateStreamTaskCount(streamToSubmit, taskCount + 1);
 
     auto submissionStatus = flushHandler(batchBuffer, this->getResidencyAllocations());

shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp

@@ -6368,6 +6368,8 @@ HWTEST_F(CommandStreamReceiverHwTest, givenEpilogueStreamAvailableWhenFlushImmed
     LinearStream epilogueStream(commandBuffer);
 
     commandStreamReceiver.storeMakeResidentAllocations = true;
+    commandStreamReceiver.recordFlushedBatchBuffer = true;
+
     immediateFlushTaskFlags.requireTaskCountUpdate = true;
     immediateFlushTaskFlags.optionalEpilogueCmdStream = &epilogueStream;
 
@@ -6377,6 +6379,7 @@ HWTEST_F(CommandStreamReceiverHwTest, givenEpilogueStreamAvailableWhenFlushImmed
                                              *pDevice);
 
     EXPECT_TRUE(commandStreamReceiver.isMadeResident(commandBuffer));
+    EXPECT_TRUE(commandStreamReceiver.latestFlushedBatchBuffer.dispatchMonitorFence);
 
     HardwareParse hwParser;