fix: emit uc tag update into epilogue stream when present

shared/source/command_stream/command_stream_receiver_hw_base.inl

@@ -466,8 +466,13 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTaskHeapful(
     if (detectInitProgrammingFlagsRequired(dispatchFlags)) {
         initProgrammingFlags();
     }
+
+    // a regular command list dispatched as the primary batch buffer has no space left in
+    // the task stream, so commands following the task go into the epilogue stream
+    LinearStream &epilogueCommandStream = dispatchFlags.optionalEpilogueCmdStream != nullptr ? *dispatchFlags.optionalEpilogueCmdStream : commandStreamTask;
+
     if (this->ucResourceRequiresTagUpdate) {
-        this->emitTagUpdateWithoutDCFlush(commandStreamTask);
+        this->emitTagUpdateWithoutDCFlush(epilogueCommandStream);
     }
     const auto &hwInfo = peekHwInfo();
 
@@ -478,7 +483,6 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushTaskHeapful(
                                      dispatchFlags.textureCacheFlush ||
                                      this->heapStorageRequiresRecyclingTag;
     if (barrierWithPostSyncNeeded) {
-        LinearStream &epilogueCommandStream = dispatchFlags.optionalEpilogueCmdStream != nullptr ? *dispatchFlags.optionalEpilogueCmdStream : commandStreamTask;
         processBarrierWithPostSync(epilogueCommandStream, dispatchFlags, levelClosed, currentPipeControlForNooping,
                                    epiloguePipeControlLocation, hasStallingCmdsOnTaskStream, args);
     }

shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp

@@ -6473,6 +6473,51 @@ HWTEST_F(CommandStreamReceiverHwTest, givenEpilogueStreamAvailableWhenFlushTaskC
     commandStreamReceiver.getMemoryManager()->freeGraphicsMemoryImpl(commandBuffer);
 }
 
+HWTEST_F(CommandStreamReceiverHwTest, givenUcResourceRequiresTagUpdateAndFullTaskStreamWithEpilogueStreamWhenFlushTaskCalledThenTagUpdateIsDispatchedIntoEpilogueStream) {
+    using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;
+
+    auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
+    if (commandStreamReceiver.heaplessModeEnabled) {
+        GTEST_SKIP();
+    }
+
+    GraphicsAllocation *commandBuffer = commandStreamReceiver.getMemoryManager()->allocateGraphicsMemoryWithProperties(MockAllocationProperties{commandStreamReceiver.getRootDeviceIndex(), MemoryConstants::pageSize});
+    ASSERT_NE(nullptr, commandBuffer);
+    LinearStream epilogueStream(commandBuffer);
+
+    flushTaskFlags.optionalEpilogueCmdStream = &epilogueStream;
+    commandStreamReceiver.ucResourceRequiresTagUpdate = true;
+
+    // a regular command list dispatched as the primary batch buffer fills the task stream
+    commandStream.getSpace(commandStream.getAvailableSpace());
+
+    EXPECT_NO_THROW(commandStreamReceiver.flushTask(commandStream,
+                                                    0,
+                                                    &dsh,
+                                                    &ioh,
+                                                    nullptr,
+                                                    taskLevel,
+                                                    flushTaskFlags,
+                                                    *pDevice));
+    EXPECT_EQ(commandStream.getMaxAvailableSpace(), commandStream.getUsed());
+    EXPECT_FALSE(commandStreamReceiver.ucResourceRequiresTagUpdate);
+
+    HardwareParse hwParser;
+    hwParser.parseCommands<FamilyType>(epilogueStream, 0);
+    auto ucTagAddress = commandStreamReceiver.getUcTagGPUAddress();
+    bool ucTagUpdateFound = false;
+    for (auto &cmd : findAll<PIPE_CONTROL *>(hwParser.cmdList.begin(), hwParser.cmdList.end())) {
+        auto pipeControl = genCmdCast<PIPE_CONTROL *>(*cmd);
+        uint64_t address = pipeControl->getAddressHigh();
+        address <<= 32;
+        address |= pipeControl->getAddress();
+        ucTagUpdateFound |= (address == ucTagAddress);
+    }
+    EXPECT_TRUE(ucTagUpdateFound);
+
+    commandStreamReceiver.getMemoryManager()->freeGraphicsMemoryImpl(commandBuffer);
+}
+
 HWTEST2_F(CommandStreamReceiverHwTest, givenSpecialPipelineSelectModeChangedWhenGetCmdSizeForPielineSelectIsCalledThenCorrectSizeIsReturned, IsAtMostXeCore) {
     using PIPELINE_SELECT = typename FamilyType::PIPELINE_SELECT;
     using PIPE_CONTROL = typename FamilyType::PIPE_CONTROL;