Add bcs split control mask

Introduce debug variable to control which engines
the tranfser will be split into

Related-To: NEO-7173

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

Author: LukaszJobczyk

opencl/source/command_queue/command_queue.cpp

@@ -305,21 +305,28 @@ void CommandQueue::initializeBcsEngine(bool internalUsage) {
 }
 
 void CommandQueue::constructBcsEnginesForSplit() {
-    if (!this->bcsSplitInitialized) {
-        for (auto i = static_cast<uint32_t>(aub_stream::EngineType::ENGINE_BCS2); i <= static_cast<uint32_t>(aub_stream::EngineType::ENGINE_BCS8); i += 2) {
-            auto index = EngineHelpers::getBcsIndex(static_cast<aub_stream::EngineType>(i));
-            if (!bcsEngines[index]) {
-                auto &neoDevice = device->getNearestGenericSubDevice(0)->getDevice();
-                bcsEngines[index] = neoDevice.tryGetEngine(static_cast<aub_stream::EngineType>(i), EngineUsage::Regular);
-                bcsEngineTypes.push_back(static_cast<aub_stream::EngineType>(i));
-                if (bcsEngines[index]) {
-                    bcsEngines[index]->osContext->ensureContextInitialized();
-                    bcsEngines[index]->commandStreamReceiver->initDirectSubmission();
-                }
+    if (this->bcsSplitInitialized) {
+        return;
+    }
+
+    if (DebugManager.flags.SplitBcsMask.get() > 0) {
+        this->splitEngines = DebugManager.flags.SplitBcsMask.get();
+    }
+
+    for (uint32_t i = 0; i < bcsInfoMaskSize; i++) {
+        if (this->splitEngines.test(i) && !bcsEngines[i]) {
+            auto &neoDevice = device->getNearestGenericSubDevice(0)->getDevice();
+            auto engineType = EngineHelpers::mapBcsIndexToEngineType(i, true);
+            bcsEngines[i] = neoDevice.tryGetEngine(engineType, EngineUsage::Regular);
+            bcsEngineTypes.push_back(engineType);
+            if (bcsEngines[i]) {
+                bcsEngines[i]->osContext->ensureContextInitialized();
+                bcsEngines[i]->commandStreamReceiver->initDirectSubmission();
             }
         }
-        this->bcsSplitInitialized = true;
     }
+
+    this->bcsSplitInitialized = true;
 }
 
 void CommandQueue::prepareHostPtrSurfaceForSplit(bool split, GraphicsAllocation &allocation) {

opencl/source/command_queue/command_queue.h

@@ -420,7 +420,9 @@ class CommandQueue : public BaseObject<_cl_command_queue> {
     bool isCopyOnly = false;
     bool bcsAllowed = false;
     bool bcsInitialized = false;
+
     bool bcsSplitInitialized = false;
+    BcsInfoMask splitEngines = EngineHelpers::evenLinkedCopyEnginesMask;
 
     LinearStream *commandStream = nullptr;
 

opencl/source/command_queue/enqueue_common.h

@@ -1136,13 +1136,18 @@ cl_int CommandQueueHw<GfxFamily>::enqueueBlitSplit(MultiDispatchInfo &dispatchIn
 
     StackVec<std::unique_lock<CommandStreamReceiver::MutexType>, 3u> locks;
     StackVec<CommandStreamReceiver *, 3u> copyEngines;
-    for (auto i = static_cast<uint32_t>(aub_stream::EngineType::ENGINE_BCS2); i <= static_cast<uint32_t>(aub_stream::EngineType::ENGINE_BCS8); i += 2) {
-        auto bcs = getBcsCommandStreamReceiver(static_cast<aub_stream::EngineType>(i));
-        if (bcs) {
-            locks.push_back(std::move(bcs->obtainUniqueOwnership()));
-            copyEngines.push_back(bcs);
+
+    for (uint32_t i = 0; i < bcsInfoMaskSize; i++) {
+        if (this->splitEngines.test(i)) {
+            auto engineType = EngineHelpers::mapBcsIndexToEngineType(i, true);
+            auto bcs = getBcsCommandStreamReceiver(engineType);
+            if (bcs) {
+                locks.push_back(std::move(bcs->obtainUniqueOwnership()));
+                copyEngines.push_back(bcs);
+            }
         }
     }
+
     DEBUG_BREAK_IF(copyEngines.size() == 0);
     TakeOwnershipWrapper<CommandQueueHw<GfxFamily>> queueOwnership(*this);
 

opencl/test/unit_test/command_queue/command_queue_tests_pvc_and_later.cpp

@@ -129,6 +129,37 @@ HWTEST2_F(CommandQueuePvcAndLaterTests, whenConstructBcsEnginesForSplitThenConta
     EXPECT_EQ(4u, queue->countBcsEngines());
 }
 
+HWTEST2_F(CommandQueuePvcAndLaterTests, givenSplitBcsMaskWhenConstructBcsEnginesForSplitThenContainsGivenBcsEngines, IsAtLeastXeHpcCore) {
+    DebugManagerStateRestore restorer;
+    std::bitset<bcsInfoMaskSize> bcsMask = 0b100110101;
+    DebugManager.flags.DeferCmdQBcsInitialization.set(1u);
+    DebugManager.flags.SplitBcsMask.set(static_cast<int>(bcsMask.to_ulong()));
+    HardwareInfo hwInfo = *defaultHwInfo;
+    hwInfo.featureTable.ftrBcsInfo = maxNBitValue(9);
+    hwInfo.capabilityTable.blitterOperationsSupported = true;
+    MockDevice *device = MockDevice::createWithNewExecutionEnvironment<MockDevice>(&hwInfo, 0);
+    MockClDevice clDevice{device};
+    cl_device_id clDeviceId = static_cast<cl_device_id>(&clDevice);
+    ClDeviceVector clDevices{&clDeviceId, 1u};
+    cl_int retVal{};
+    auto context = std::unique_ptr<Context>{Context::create<Context>(nullptr, clDevices, nullptr, nullptr, retVal)};
+    EXPECT_EQ(CL_SUCCESS, retVal);
+    auto queue = std::make_unique<MockCommandQueue>(*context);
+    EXPECT_EQ(0u, queue->countBcsEngines());
+
+    queue->constructBcsEnginesForSplit();
+
+    EXPECT_EQ(5u, queue->countBcsEngines());
+
+    for (uint32_t i = 0; i < bcsInfoMaskSize; i++) {
+        if (bcsMask.test(i)) {
+            EXPECT_NE(queue->bcsEngines[i], nullptr);
+        } else {
+            EXPECT_EQ(queue->bcsEngines[i], nullptr);
+        }
+    }
+}
+
 HWTEST2_F(CommandQueuePvcAndLaterTests, whenSelectCsrForHostPtrAllocationThenReturnProperEngine, IsAtLeastXeHpcCore) {
     DebugManagerStateRestore restorer;
     DebugManager.flags.DeferCmdQBcsInitialization.set(1u);

shared/source/debug_settings/debug_variables_base.inl

@@ -288,6 +288,7 @@ DECLARE_DEBUG_VARIABLE(int32_t, DeferCmdQGpgpuInitialization, -1, "-1: default,
 DECLARE_DEBUG_VARIABLE(int32_t, DeferCmdQBcsInitialization, -1, "-1: default, 0:disabled, 1: enabled.")
 DECLARE_DEBUG_VARIABLE(int32_t, PreferInternalBcsEngine, -1, "-1: default, 0:disabled, 1: enabled. When enabled use internal BCS engine for internal transfers, when disabled use regular engine")
 DECLARE_DEBUG_VARIABLE(int32_t, SplitBcsCopy, -1, "-1: default, 0:disabled, 1: enabled. When enqueues copy to main copy engine then split between even linked copy engines")
+DECLARE_DEBUG_VARIABLE(int32_t, SplitBcsMask, 0, "0: default, >0: bitmask: indicates bcs engines for split")
 DECLARE_DEBUG_VARIABLE(int32_t, ReuseKernelBinaries, -1, "-1: default, 0:disabled, 1: enabled. If enabled, driver reuses kernel binaries.")
 
 /*DIRECT SUBMISSION FLAGS*/

shared/source/helpers/engine_node_helper.h

@@ -65,6 +65,7 @@ constexpr bool isLinkBcs(aub_stream::EngineType engineType) {
 }
 
 constexpr uint32_t numLinkedCopyEngines = 8u;
+constexpr size_t evenLinkedCopyEnginesMask = 0b101010100;
 
 bool linkCopyEnginesSupported(const HardwareInfo &hwInfo, const DeviceBitfield &deviceBitfield);
 

shared/test/common/test_files/igdrcl.config

@@ -394,6 +394,7 @@ AssignBCSAtEnqueue = -1
 DeferCmdQGpgpuInitialization = -1
 DeferCmdQBcsInitialization = -1
 SplitBcsCopy = -1
+SplitBcsMask = 0
 PreferInternalBcsEngine = -1
 ReuseKernelBinaries = -1
 EnableChipsetUniqueUUID = -1