fix: track registered CSR clients

BartoszDunajski · Compute-Runtime-Automation · commit 43b887234197 · 2023-09-25T12:00:13.000+02:00
In L0 its not possible to track objects relations. For example CmdList may be removed before Event. In such case, Event needs to safely skip unregister call, without accessing CmdList/CmdQueue object. Related-To: NEO-8884 Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com> Source: 53f635e
diff --git a/level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl b/level_zero/core/source/cmdlist/cmdlist_hw_immediate.inl
@@ -339,9 +339,7 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand
         svmAllocMgr->prefetchSVMAllocs(*this->device->getNEODevice(), *csr);
     }
 
-    if (cmdQ->getClientId() == CommandQueue::clientNotRegistered) {
-        cmdQ->setClientId(csr->registerClient());
-    }
+    cmdQ->registerCsrClient();
 
     std::unique_lock<std::mutex> lockForIndirect;
     if (this->hasIndirectAllocationsAllowed()) {
diff --git a/level_zero/core/source/cmdqueue/cmdqueue.cpp b/level_zero/core/source/cmdqueue/cmdqueue.cpp
@@ -253,10 +253,11 @@ CommandQueue *CommandQueue::create(uint32_t productFamily, Device *device, NEO::
 }
 
 void CommandQueueImp::unregisterCsrClient() {
-    if (getClientId() != CommandQueue::clientNotRegistered) {
-        this->csr->unregisterClient();
-        setClientId(CommandQueue::clientNotRegistered);
-    }
+    this->csr->unregisterClient(this);
+}
+
+void CommandQueueImp::registerCsrClient() {
+    this->csr->registerClient(this);
 }
 
 ze_command_queue_mode_t CommandQueueImp::getSynchronousMode() const {
diff --git a/level_zero/core/source/cmdqueue/cmdqueue.h b/level_zero/core/source/cmdqueue/cmdqueue.h
@@ -62,19 +62,15 @@ struct CommandQueue : _ze_command_queue_handle_t {
 
     bool peekIsCopyOnlyCommandQueue() const { return this->isCopyOnlyCommandQueue; }
 
-    uint32_t getClientId() const { return this->clientId; }
-    void setClientId(uint32_t value) { this->clientId = value; }
     virtual void unregisterCsrClient() = 0;
+    virtual void registerCsrClient() = 0;
 
     TaskCountType getTaskCount() const { return taskCount; }
     void setTaskCount(TaskCountType newTaskCount) { taskCount = newTaskCount; }
 
-    static constexpr uint32_t clientNotRegistered = std::numeric_limits<uint32_t>::max();
-
   protected:
     bool frontEndTrackingEnabled() const;
 
-    uint32_t clientId = clientNotRegistered;
     uint32_t partitionCount = 1;
     uint32_t activeSubDevices = 1;
     std::atomic<TaskCountType> taskCount = 0;
diff --git a/level_zero/core/source/cmdqueue/cmdqueue_hw.inl b/level_zero/core/source/cmdqueue/cmdqueue_hw.inl
@@ -73,9 +73,7 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandLists(
         svmAllocMgr->prefetchSVMAllocs(*device->getNEODevice(), *csr);
     }
 
-    if (this->clientId == CommandQueue::clientNotRegistered) {
-        this->clientId = this->csr->registerClient();
-    }
+    registerCsrClient();
 
     auto neoDevice = device->getNEODevice();
     auto ctx = CommandListExecutionContext{phCommandLists,
diff --git a/level_zero/core/source/cmdqueue/cmdqueue_imp.h b/level_zero/core/source/cmdqueue/cmdqueue_imp.h
@@ -90,6 +90,7 @@ struct CommandQueueImp : public CommandQueue {
     void printKernelsPrintfOutput(bool hangDetected);
     void checkAssert();
     void unregisterCsrClient() override;
+    void registerCsrClient() override;
 
   protected:
     MOCKABLE_VIRTUAL NEO::SubmissionStatus submitBatchBuffer(size_t offset, NEO::ResidencyContainer &residencyContainer, void *endingCmdPtr,
diff --git a/level_zero/core/source/event/event.cpp b/level_zero/core/source/event/event.cpp
@@ -402,9 +402,10 @@ void Event::setLatestUsedCmdQueue(CommandQueue *newCmdQ) {
 }
 
 void Event::unsetCmdQueue() {
-    if (latestUsedCmdQueue) {
-        latestUsedCmdQueue->unregisterCsrClient();
+    for (auto &csr : csrs) {
+        csr->unregisterClient(latestUsedCmdQueue);
     }
+
     latestUsedCmdQueue = nullptr;
 }
 
diff --git a/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_1.cpp b/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_1.cpp
@@ -1200,7 +1200,6 @@ HWTEST2_F(CommandListCreate, givenDirectSubmissionAndImmCmdListWhenDispatchingDi
 
     auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(whiteBoxCmdList->csr);
     ultCsr->recordFlusheBatchBuffer = true;
-    ultCsr->unregisterClient();
 
     EXPECT_FALSE(NEO::RelaxedOrderingHelper::isRelaxedOrderingDispatchAllowed(*ultCsr, 1));
 
@@ -1299,8 +1298,9 @@ HWTEST2_F(CommandListCreate, whenDispatchingThenPassNumCsrClients, IsAtLeastXeHp
     auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(whiteBoxCmdList->csr);
     ultCsr->recordFlusheBatchBuffer = true;
 
-    ultCsr->registerClient();
-    ultCsr->registerClient();
+    int client1, client2;
+    ultCsr->registerClient(&client1);
+    ultCsr->registerClient(&client2);
 
     auto result = commandList->appendLaunchKernel(kernel.toHandle(), &groupCount, nullptr, 0, nullptr, launchParams, false);
 
@@ -1458,8 +1458,9 @@ HWTEST2_F(CommandListCreate, givenDirectSubmissionAndImmCmdListWhenDispatchingTh
 
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(*ultCsr);
     ultCsr->directSubmission.reset(directSubmission);
-    ultCsr->registerClient();
-    ultCsr->registerClient();
+    int client1, client2;
+    ultCsr->registerClient(&client1);
+    ultCsr->registerClient(&client2);
 
     auto verifyFlags = [&ultCsr, useImmediateFlushTask](ze_result_t result, bool dispatchFlag, bool bbFlag) {
         EXPECT_EQ(ZE_RESULT_SUCCESS, result);
@@ -1580,8 +1581,9 @@ HWTEST2_F(CommandListCreate, givenInOrderExecutionWhenDispatchingRelaxedOrdering
 
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(*ultCsr);
     ultCsr->directSubmission.reset(directSubmission);
-    ultCsr->registerClient();
-    ultCsr->registerClient();
+    int client1, client2;
+    ultCsr->registerClient(&client1);
+    ultCsr->registerClient(&client2);
 
     commandList->appendLaunchKernel(kernel.toHandle(), &groupCount, event, 0, nullptr, launchParams, false);
 
@@ -1628,8 +1630,9 @@ HWTEST2_F(CommandListCreate, givenInOrderExecutionWhenDispatchingBarrierThenAllo
 
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(*ultCsr);
     ultCsr->directSubmission.reset(directSubmission);
-    ultCsr->registerClient();
-    ultCsr->registerClient();
+    int client1, client2;
+    ultCsr->registerClient(&client1);
+    ultCsr->registerClient(&client2);
 
     // Initialize NP state
     commandList->appendBarrier(nullptr, 1, &event, false);
@@ -1687,8 +1690,9 @@ HWTEST2_F(CommandListCreate, givenInOrderExecutionWhenDispatchingRelaxedOrdering
 
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(*ultCsr);
     ultCsr->directSubmission.reset(directSubmission);
-    ultCsr->registerClient();
-    ultCsr->registerClient();
+    int client1, client2;
+    ultCsr->registerClient(&client1);
+    ultCsr->registerClient(&client2);
 
     auto cmdStream = cmdList->getCmdContainer().getCommandStream();
 
diff --git a/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_7.cpp b/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_7.cpp
@@ -2964,7 +2964,8 @@ HWTEST2_F(ImmediateCommandListHostSynchronize, givenCsrClientCountWhenCallingSyn
 
     auto cmdList = createCmdList<gfxCoreFamily>(csr);
 
-    cmdList->cmdQImmediate->setClientId(csr->registerClient());
+    cmdList->cmdQImmediate->registerCsrClient();
+
     auto clientCount = csr->getNumClients();
 
     EXPECT_EQ(cmdList->hostSynchronize(0), ZE_RESULT_SUCCESS);
@@ -2975,7 +2976,8 @@ HWTEST2_F(ImmediateCommandListHostSynchronize, givenCsrClientCountWhenCallingSyn
 
     EXPECT_EQ(clientCount - 1, csr->getNumClients());
 
-    cmdList->cmdQImmediate->setClientId(csr->registerClient());
+    cmdList->cmdQImmediate->registerCsrClient();
+
     clientCount = csr->getNumClients();
 
     csr->callBaseWaitForCompletionWithTimeout = false;
diff --git a/level_zero/core/test/unit_tests/xe_hpc_core/test_cmdqueue_xe_hpc_core.cpp b/level_zero/core/test/unit_tests/xe_hpc_core/test_cmdqueue_xe_hpc_core.cpp
@@ -715,7 +715,8 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
     context->allocHostMem(&hostDesc, size, alignment, &dstPtr);
     auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(whiteBoxCmdList->csr);
     ultCsr->recordFlusheBatchBuffer = true;
-    ultCsr->registerClient();
+    int client;
+    ultCsr->registerClient(&client);
 
     auto result = commandList0->appendMemoryCopy(dstPtr, srcPtr, size, nullptr, 0, nullptr, false, false);
     ASSERT_EQ(ZE_RESULT_SUCCESS, result);
@@ -778,8 +779,9 @@ HWTEST2_F(CommandQueueCommandsXeHpc, givenFlushTaskSubmissionEnabledAndSplitBcsC
     context->allocHostMem(&hostDesc, size, alignment, &dstPtr);
     auto ultCsr = static_cast<NEO::UltCommandStreamReceiver<FamilyType> *>(whiteBoxCmdList->csr);
     ultCsr->recordFlusheBatchBuffer = true;
-    ultCsr->registerClient();
-    ultCsr->registerClient();
+    int client1, client2;
+    ultCsr->registerClient(&client1);
+    ultCsr->registerClient(&client2);
 
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(*ultCsr);
     ultCsr->directSubmission.reset(directSubmission);
diff --git a/opencl/source/command_queue/command_queue.cpp b/opencl/source/command_queue/command_queue.cpp
@@ -1406,7 +1406,7 @@ void CommandQueue::registerGpgpuCsrClient() {
     if (!gpgpuCsrClientRegistered) {
         gpgpuCsrClientRegistered = true;
 
-        getGpgpuCommandStreamReceiver().registerClient();
+        getGpgpuCommandStreamReceiver().registerClient(this);
     }
 }
 
@@ -1417,13 +1417,13 @@ void CommandQueue::registerBcsCsrClient(CommandStreamReceiver &bcsCsr) {
 
     if (!bcsState.csrClientRegistered) {
         bcsState.csrClientRegistered = true;
-        bcsCsr.registerClient();
+        bcsCsr.registerClient(this);
     }
 }
 
 void CommandQueue::unregisterGpgpuCsrClient() {
     if (gpgpuCsrClientRegistered) {
-        gpgpuEngine->commandStreamReceiver->unregisterClient();
+        gpgpuEngine->commandStreamReceiver->unregisterClient(this);
         gpgpuCsrClientRegistered = false;
     }
 }
@@ -1434,7 +1434,7 @@ void CommandQueue::unregisterBcsCsrClient(CommandStreamReceiver &bcsCsr) {
     auto &bcsState = bcsStates[EngineHelpers::getBcsIndex(engineType)];
 
     if (bcsState.isValid() && bcsState.csrClientRegistered) {
-        bcsCsr.unregisterClient();
+        bcsCsr.unregisterClient(this);
         bcsState.csrClientRegistered = false;
     }
 }
diff --git a/opencl/test/unit_test/command_queue/command_queue_hw_1_tests.cpp b/opencl/test/unit_test/command_queue/command_queue_hw_1_tests.cpp
@@ -1254,11 +1254,12 @@ HWTEST_F(CommandQueueHwTest, givenRelaxedOrderingEnabledWhenCheckingIfAllowedByC
     EXPECT_FALSE(mockCmdQueueHw.relaxedOrderingForGpgpuAllowed(0));
     EXPECT_FALSE(mockCmdQueueHw.relaxedOrderingForGpgpuAllowed(1));
 
-    ultCsr.registerClient();
+    int client1, client2;
+    ultCsr.registerClient(&client1);
     EXPECT_FALSE(mockCmdQueueHw.relaxedOrderingForGpgpuAllowed(0));
     EXPECT_FALSE(mockCmdQueueHw.relaxedOrderingForGpgpuAllowed(1));
 
-    ultCsr.registerClient();
+    ultCsr.registerClient(&client2);
 
     EXPECT_FALSE(mockCmdQueueHw.relaxedOrderingForGpgpuAllowed(0));
     EXPECT_TRUE(mockCmdQueueHw.relaxedOrderingForGpgpuAllowed(1));
diff --git a/opencl/test/unit_test/command_queue/enqueue_kernel_2_tests.cpp b/opencl/test/unit_test/command_queue/enqueue_kernel_2_tests.cpp
@@ -1165,8 +1165,9 @@ HWTEST2_F(RelaxedOrderingEnqueueKernelTests, givenEnqueueKernelWhenProgrammingDe
     auto &ultCsr = pDevice->getUltCommandStreamReceiver<FamilyType>();
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(ultCsr);
     ultCsr.directSubmission.reset(directSubmission);
-    ultCsr.registerClient();
-    ultCsr.registerClient();
+    int client1, client2;
+    ultCsr.registerClient(&client1);
+    ultCsr.registerClient(&client2);
 
     MockCommandQueueHw<FamilyType> mockCmdQueueHw{context, pClDevice, nullptr};
 
@@ -1264,8 +1265,9 @@ HWTEST2_F(RelaxedOrderingEnqueueKernelTests, givenBarrierWithDependenciesWhenFlu
     auto &ultCsr = pDevice->getUltCommandStreamReceiver<FamilyType>();
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(ultCsr);
     ultCsr.directSubmission.reset(directSubmission);
-    ultCsr.registerClient();
-    ultCsr.registerClient();
+    int client1, client2;
+    ultCsr.registerClient(&client1);
+    ultCsr.registerClient(&client2);
 
     MockCommandQueueHw<FamilyType> mockCmdQueueHw{context, pClDevice, nullptr};
 
@@ -1311,8 +1313,9 @@ HWTEST2_F(RelaxedOrderingEnqueueKernelTests, givenPipeControlForIoqDependencyRes
     auto &ultCsr = pDevice->getUltCommandStreamReceiver<FamilyType>();
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(ultCsr);
     ultCsr.directSubmission.reset(directSubmission);
-    ultCsr.registerClient();
-    ultCsr.registerClient();
+    int client1, client2;
+    ultCsr.registerClient(&client1);
+    ultCsr.registerClient(&client2);
 
     MockKernelWithInternals mockKernel(*pClDevice);
 
@@ -1331,8 +1334,9 @@ HWTEST2_F(RelaxedOrderingEnqueueKernelTests, givenEnqueueWithPipeControlWhenSend
     auto &ultCsr = pDevice->getUltCommandStreamReceiver<FamilyType>();
     auto directSubmission = new MockDirectSubmissionHw<FamilyType, RenderDispatcher<FamilyType>>(ultCsr);
     ultCsr.directSubmission.reset(directSubmission);
-    ultCsr.registerClient();
-    ultCsr.registerClient();
+    int client1, client2;
+    ultCsr.registerClient(&client1);
+    ultCsr.registerClient(&client2);
 
     ultCsr.recordFlusheBatchBuffer = true;
 
diff --git a/opencl/test/unit_test/command_stream/command_stream_receiver_hw_1_tests.cpp b/opencl/test/unit_test/command_stream/command_stream_receiver_hw_1_tests.cpp
@@ -1091,8 +1091,9 @@ HWTEST2_F(RelaxedOrderingBcsTests, givenDependenciesWhenFlushingThenProgramCorre
     ASSERT_EQ(CL_SUCCESS, retVal);
 
     auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
-    csr.registerClient();
-    csr.registerClient();
+    int client1, client2;
+    csr.registerClient(&client1);
+    csr.registerClient(&client2);
     csr.recordFlusheBatchBuffer = true;
 
     csr.blitterDirectSubmission = std::make_unique<MockDirectSubmissionHw<FamilyType, BlitterDispatcher<FamilyType>>>(csr);
@@ -1133,8 +1134,9 @@ HWTEST2_F(RelaxedOrderingBcsTests, givenDependenciesWhenFlushingThenProgramProgr
     ASSERT_EQ(CL_SUCCESS, retVal);
 
     auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
-    csr.registerClient();
-    csr.registerClient();
+    int client1, client2;
+    csr.registerClient(&client1);
+    csr.registerClient(&client1);
     csr.recordFlusheBatchBuffer = true;
 
     csr.blitterDirectSubmission = std::make_unique<MockDirectSubmissionHw<FamilyType, BlitterDispatcher<FamilyType>>>(csr);
@@ -1158,12 +1160,12 @@ HWTEST2_F(RelaxedOrderingBcsTests, givenDependenciesWhenFlushingThenProgramProgr
     EXPECT_FALSE(csr.bcsRelaxedOrderingAllowed(blitPropertiesContainer, false));
 
     blitPropertiesContainer.push_back(blitProperties);
-    csr.unregisterClient();
-    csr.unregisterClient();
+    csr.unregisterClient(&client1);
+    csr.unregisterClient(&client2);
     EXPECT_FALSE(csr.bcsRelaxedOrderingAllowed(blitPropertiesContainer, false));
 
-    csr.registerClient();
-    csr.registerClient();
+    csr.registerClient(&client1);
+    csr.registerClient(&client2);
     csr.blitterDirectSubmission.reset();
     EXPECT_FALSE(csr.bcsRelaxedOrderingAllowed(blitPropertiesContainer, false));
 }
@@ -1174,8 +1176,9 @@ HWTEST2_F(RelaxedOrderingBcsTests, givenTagUpdateWhenFlushingThenDisableRelaxedO
     ASSERT_EQ(CL_SUCCESS, retVal);
 
     auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
-    csr.registerClient();
-    csr.registerClient();
+    int client1, client2;
+    csr.registerClient(&client1);
+    csr.registerClient(&client2);
     csr.recordFlusheBatchBuffer = true;
 
     csr.blitterDirectSubmission = std::make_unique<MockDirectSubmissionHw<FamilyType, BlitterDispatcher<FamilyType>>>(csr);
diff --git a/opencl/test/unit_test/command_stream/command_stream_receiver_hw_2_tests.cpp b/opencl/test/unit_test/command_stream/command_stream_receiver_hw_2_tests.cpp
@@ -709,8 +709,11 @@ HWTEST_F(BcsTests, givenTaskStreamWhenFlushingThenPassNumClients) {
                                                                           graphicsAllocation->getGpuAddress(), 0,
                                                                           0, 0, {1, 1, 1}, 0, 0, 0, 0);
 
-    csr.registerClient();
-    csr.registerClient();
+    int client1, client2;
+    csr.registerClient(&client1);
+    csr.registerClient(&client2);
+
+    EXPECT_EQ(2u, csr.getNumClients());
 
     flushBcsTask(&csr, blitProperties, true, *pDevice);
 
diff --git a/shared/source/command_stream/command_stream_receiver.cpp b/shared/source/command_stream/command_stream_receiver.cpp
@@ -95,6 +95,8 @@ CommandStreamReceiver::CommandStreamReceiver(ExecutionEnvironment &executionEnvi
     productHelper.fillStateBaseAddressPropertiesSupportStructure(sbaSupportFlags);
     this->doubleSbaWa = productHelper.isAdditionalStateBaseAddressWARequired(hwInfo);
     this->l1CachePolicyData.init(productHelper);
+
+    registeredClients.reserve(16);
 }
 
 CommandStreamReceiver::~CommandStreamReceiver() {
@@ -1080,5 +1082,24 @@ bool CommandStreamReceiver::isRayTracingStateProgramingNeeded(Device &device) co
     return device.getRTMemoryBackedBuffer() && getBtdCommandDirty();
 }
 
+void CommandStreamReceiver::registerClient(void *client) {
+    std::unique_lock<MutexType> lock(registeredClientsMutex);
+    auto element = std::find(registeredClients.begin(), registeredClients.end(), client);
+    if (element == registeredClients.end()) {
+        registeredClients.push_back(client);
+        numClients++;
+    }
+}
+
+void CommandStreamReceiver::unregisterClient(void *client) {
+    std::unique_lock<MutexType> lock(registeredClientsMutex);
+
+    auto element = std::find(registeredClients.begin(), registeredClients.end(), client);
+    if (element != registeredClients.end()) {
+        registeredClients.erase(element);
+        numClients--;
+    }
+}
+
 std::function<void()> CommandStreamReceiver::debugConfirmationFunction = []() { std::cin.get(); };
 } // namespace NEO
diff --git a/shared/source/command_stream/command_stream_receiver.h b/shared/source/command_stream/command_stream_receiver.h
diff --git a/shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp b/shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp
diff --git a/shared/test/unit_test/direct_submission/direct_submission_tests_2.cpp b/shared/test/unit_test/direct_submission/direct_submission_tests_2.cpp

Original file line number	Diff line number	Diff line change
`@@ -339,9 +339,7 @@ inline ze_result_t CommandListCoreFamilyImmediate<gfxCoreFamily>::executeCommand`
`339`	`339`	`svmAllocMgr->prefetchSVMAllocs(this->device->getNEODevice(), csr);`
`340`	`340`	`}`
`341`	`341`
`342`		`- if (cmdQ->getClientId() == CommandQueue::clientNotRegistered) {`
`343`		`- cmdQ->setClientId(csr->registerClient());`
`344`		`- }`
	`342`	`+ cmdQ->registerCsrClient();`
`345`	`343`
`346`	`344`	`std::unique_lock<std::mutex> lockForIndirect;`
`347`	`345`	`if (this->hasIndirectAllocationsAllowed()) {`
Original file line number	Diff line number	Diff line change
`@@ -253,10 +253,11 @@ CommandQueue CommandQueue::create(uint32_t productFamily, Device device, NEO::`
`253`	`253`	`}`
`254`	`254`
`255`	`255`	`void CommandQueueImp::unregisterCsrClient() {`
`256`		`- if (getClientId() != CommandQueue::clientNotRegistered) {`
`257`		`- this->csr->unregisterClient();`
`258`		`- setClientId(CommandQueue::clientNotRegistered);`
`259`		`- }`
	`256`	`+ this->csr->unregisterClient(this);`
	`257`	`+}`
	`258`	`+`
	`259`	`+void CommandQueueImp::registerCsrClient() {`
	`260`	`+ this->csr->registerClient(this);`
`260`	`261`	`}`
`261`	`262`
`262`	`263`	`ze_command_queue_mode_t CommandQueueImp::getSynchronousMode() const {`
Original file line number	Diff line number	Diff line change
`@@ -73,9 +73,7 @@ ze_result_t CommandQueueHw<gfxCoreFamily>::executeCommandLists(`
`73`	`73`	`svmAllocMgr->prefetchSVMAllocs(device->getNEODevice(), csr);`
`74`	`74`	`}`
`75`	`75`
`76`		`- if (this->clientId == CommandQueue::clientNotRegistered) {`
`77`		`- this->clientId = this->csr->registerClient();`
`78`		`- }`
	`76`	`+ registerCsrClient();`
`79`	`77`
`80`	`78`	`auto neoDevice = device->getNEODevice();`
`81`	`79`	`auto ctx = CommandListExecutionContext{phCommandLists,`
Original file line number	Diff line number	Diff line change
`@@ -402,9 +402,10 @@ void Event::setLatestUsedCmdQueue(CommandQueue *newCmdQ) {`
`402`	`402`	`}`
`403`	`403`
`404`	`404`	`void Event::unsetCmdQueue() {`
`405`		`- if (latestUsedCmdQueue) {`
`406`		`- latestUsedCmdQueue->unregisterCsrClient();`
	`405`	`+ for (auto &csr : csrs) {`
	`406`	`+ csr->unregisterClient(latestUsedCmdQueue);`
`407`	`407`	`}`
	`408`	`+`
`408`	`409`	`latestUsedCmdQueue = nullptr;`
`409`	`410`	`}`
`410`	`411`
Original file line number	Diff line number	Diff line change
`@@ -1406,7 +1406,7 @@ void CommandQueue::registerGpgpuCsrClient() {`
`1406`	`1406`	`if (!gpgpuCsrClientRegistered) {`
`1407`	`1407`	`gpgpuCsrClientRegistered = true;`
`1408`	`1408`
`1409`		`- getGpgpuCommandStreamReceiver().registerClient();`
	`1409`	`+ getGpgpuCommandStreamReceiver().registerClient(this);`
`1410`	`1410`	`}`
`1411`	`1411`	`}`
`1412`	`1412`
`@@ -1417,13 +1417,13 @@ void CommandQueue::registerBcsCsrClient(CommandStreamReceiver &bcsCsr) {`
`1417`	`1417`
`1418`	`1418`	`if (!bcsState.csrClientRegistered) {`
`1419`	`1419`	`bcsState.csrClientRegistered = true;`
`1420`		`- bcsCsr.registerClient();`
	`1420`	`+ bcsCsr.registerClient(this);`
`1421`	`1421`	`}`
`1422`	`1422`	`}`
`1423`	`1423`
`1424`	`1424`	`void CommandQueue::unregisterGpgpuCsrClient() {`
`1425`	`1425`	`if (gpgpuCsrClientRegistered) {`
`1426`		`- gpgpuEngine->commandStreamReceiver->unregisterClient();`
	`1426`	`+ gpgpuEngine->commandStreamReceiver->unregisterClient(this);`
`1427`	`1427`	`gpgpuCsrClientRegistered = false;`
`1428`	`1428`	`}`
`1429`	`1429`	`}`
`@@ -1434,7 +1434,7 @@ void CommandQueue::unregisterBcsCsrClient(CommandStreamReceiver &bcsCsr) {`
`1434`	`1434`	`auto &bcsState = bcsStates[EngineHelpers::getBcsIndex(engineType)];`
`1435`	`1435`
`1436`	`1436`	`if (bcsState.isValid() && bcsState.csrClientRegistered) {`
`1437`		`- bcsCsr.unregisterClient();`
	`1437`	`+ bcsCsr.unregisterClient(this);`
`1438`	`1438`	`bcsState.csrClientRegistered = false;`
`1439`	`1439`	`}`
`1440`	`1440`	`}`