OpenCL Queue Families extension 2/n

Disable CL_QUEUE_CAPABILITY_KERNEL_INTEL capability for copy queues

Signed-off-by: Maciej Dziuban <maciej.dziuban@intel.com>
Related-To: NEO-5120

opencl/source/cl_device/cl_device.cpp

@@ -11,6 +11,7 @@
 #include "shared/source/device/device.h"
 #include "shared/source/device/sub_device.h"
 #include "shared/source/execution_environment/root_device_environment.h"
+#include "shared/source/helpers/hw_helper.h"
 #include "shared/source/os_interface/driver_info.h"
 #include "shared/source/os_interface/os_interface.h"
 #include "shared/source/program/sync_buffer_handler.h"
@@ -201,4 +202,33 @@ bool ClDevice::arePipesSupported() const {
     return device.getHardwareInfo().capabilityTable.supportsPipes;
 }
 
+cl_command_queue_capabilities_intel ClDevice::getQueueFamilyCapabilitiesAll() {
+    return CL_QUEUE_CAPABILITY_EVENT_WAIT_LIST_INTEL |
+           CL_QUEUE_CAPABILITY_EVENTS_INTEL |
+           CL_QUEUE_CAPABILITY_TRANSFER_BUFFER_INTEL |
+           CL_QUEUE_CAPABILITY_TRANSFER_BUFFER_RECT_INTEL |
+           CL_QUEUE_CAPABILITY_MAP_BUFFER_INTEL |
+           CL_QUEUE_CAPABILITY_FILL_BUFFER_INTEL |
+           CL_QUEUE_CAPABILITY_TRANSFER_IMAGE_INTEL |
+           CL_QUEUE_CAPABILITY_MAP_IMAGE_INTEL |
+           CL_QUEUE_CAPABILITY_FILL_IMAGE_INTEL |
+           CL_QUEUE_CAPABILITY_MARKER_INTEL |
+           CL_QUEUE_CAPABILITY_BARRIER_INTEL |
+           CL_QUEUE_CAPABILITY_KERNEL_INTEL;
+}
+
+cl_command_queue_capabilities_intel ClDevice::getQueueFamilyCapabilities(EngineGroupType type) {
+    auto &hwHelper = NEO::HwHelper::get(getHardwareInfo().platform.eRenderCoreFamily);
+
+    cl_command_queue_capabilities_intel disabledProperties = 0u;
+    if (hwHelper.isCopyOnlyEngineType(type)) {
+        disabledProperties |= CL_QUEUE_CAPABILITY_KERNEL_INTEL;
+    }
+
+    if (disabledProperties != 0) {
+        return getQueueFamilyCapabilitiesAll() & ~disabledProperties;
+    }
+    return CL_QUEUE_CAPABILITY_ALL_INTEL;
+}
+
 } // namespace NEO

opencl/source/cl_device/cl_device.h

@@ -123,6 +123,9 @@ class ClDevice : public BaseObject<_cl_device_id> {
     bool isDeviceEnqueueSupported() const;
     bool arePipesSupported() const;
 
+    static cl_command_queue_capabilities_intel getQueueFamilyCapabilitiesAll();
+    MOCKABLE_VIRTUAL cl_command_queue_capabilities_intel getQueueFamilyCapabilities(EngineGroupType type);
+
   protected:
     void initializeCaps();
     void initializeExtensionsWithVersion();

opencl/source/cl_device/cl_device_caps.cpp

@@ -365,8 +365,9 @@ void ClDevice::initializeCaps() {
         for (int queueFamilyIndex = 0; queueFamilyIndex < static_cast<int>(EngineGroupType::MaxEngineGroups); queueFamilyIndex++) {
             const std::vector<EngineControl> &enginesInFamily = queueFamilies.at(queueFamilyIndex);
             if (enginesInFamily.size() > 0) {
+                const auto engineGroupType = static_cast<EngineGroupType>(queueFamilyIndex);
                 cl_queue_family_properties_intel properties;
-                properties.capabilities = CL_QUEUE_CAPABILITY_ALL_INTEL;
+                properties.capabilities = getQueueFamilyCapabilities(engineGroupType);
                 properties.count = static_cast<cl_uint>(enginesInFamily.size());
                 properties.properties = deviceInfo.queueOnHostProperties;
                 deviceInfo.queueFamilyProperties.push_back(properties);

opencl/test/unit_test/device/device_caps_tests.cpp

@@ -1458,3 +1458,25 @@ TEST_F(DeviceCapsWithModifiedHwInfoTest, givenPlatformWithSourceLevelDebuggerNot
     EXPECT_EQ(nullptr, device->getDebugger());
     EXPECT_FALSE(caps.debuggerActive);
 }
+
+TEST_F(DeviceGetCapsTest, givenClDeviceWhenInitializingCapsThenUseGetQueueFamilyCapabilitiesMethod) {
+    struct ClDeviceWithCustomQueueCaps : MockClDevice {
+        using MockClDevice::MockClDevice;
+        cl_command_queue_capabilities_intel queueCaps{};
+        cl_command_queue_capabilities_intel getQueueFamilyCapabilities(EngineGroupType type) override {
+            return queueCaps;
+        }
+    };
+
+    auto device = std::make_unique<ClDeviceWithCustomQueueCaps>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(defaultHwInfo.get()));
+
+    device->deviceInfo = {};
+    device->queueCaps = CL_QUEUE_CAPABILITY_FILL_IMAGE_INTEL;
+    device->initializeCaps();
+    EXPECT_EQ(device->queueCaps, device->getDeviceInfo().queueFamilyProperties[0].capabilities);
+
+    device->deviceInfo = {};
+    device->queueCaps = CL_QUEUE_CAPABILITY_MAP_BUFFER_INTEL | CL_QUEUE_CAPABILITY_TRANSFER_BUFFER_INTEL;
+    device->initializeCaps();
+    EXPECT_EQ(device->queueCaps, device->getDeviceInfo().queueFamilyProperties[0].capabilities);
+}

opencl/test/unit_test/device/device_tests.cpp

@@ -466,3 +466,30 @@ TEST(DeviceGenEngineTest, givenEmptyGroupsWhenGettingNonEmptyGroupsThenReturnCor
     EXPECT_EQ(&engineGroups[3], device->getNonEmptyEngineGroup(1));
     EXPECT_EQ(nullptr, device->getNonEmptyEngineGroup(2));
 }
+
+TEST(DeviceGenEngineTest, whenGettingQueueFamilyCapabilitiesAllThenReturnCorrectValue) {
+    const cl_command_queue_capabilities_intel expectedProperties = CL_QUEUE_CAPABILITY_EVENT_WAIT_LIST_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_EVENTS_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_TRANSFER_BUFFER_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_TRANSFER_BUFFER_RECT_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_MAP_BUFFER_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_FILL_BUFFER_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_TRANSFER_IMAGE_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_MAP_IMAGE_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_FILL_IMAGE_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_MARKER_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_BARRIER_INTEL |
+                                                                   CL_QUEUE_CAPABILITY_KERNEL_INTEL;
+    EXPECT_EQ(expectedProperties, MockClDevice::getQueueFamilyCapabilitiesAll());
+}
+
+TEST(DeviceGenEngineTest, givenCopyQueueWhenGettingQueueFamilyCapabilitiesThenDoNotReturnKernel) {
+    const auto propertiesAll = MockClDevice::getQueueFamilyCapabilitiesAll();
+    const auto propertiesWithoutKernel = setBits(propertiesAll, false, CL_QUEUE_CAPABILITY_KERNEL_INTEL);
+
+    auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(defaultHwInfo.get()));
+
+    EXPECT_EQ(CL_QUEUE_CAPABILITY_ALL_INTEL, device->getQueueFamilyCapabilities(NEO::EngineGroupType::Compute));
+    EXPECT_EQ(CL_QUEUE_CAPABILITY_ALL_INTEL, device->getQueueFamilyCapabilities(NEO::EngineGroupType::RenderCompute));
+    EXPECT_EQ(propertiesWithoutKernel, device->getQueueFamilyCapabilities(NEO::EngineGroupType::Copy));
+}

opencl/test/unit_test/device/get_device_info_tests.cpp

@@ -735,7 +735,7 @@ HWTEST_F(GetDeviceInfoQueueFamilyTest, givenSingleDeviceWhenInitializingCapsThen
     EXPECT_EQ(3u, families[0].count);
     EXPECT_EQ(clDevice.getDeviceInfo().queueOnHostProperties, families[0].properties);
 
-    EXPECT_EQ(CL_QUEUE_CAPABILITY_ALL_INTEL, families[1].capabilities);
+    EXPECT_EQ(clDevice.getQueueFamilyCapabilities(EngineGroupType::Copy), families[1].capabilities);
     EXPECT_EQ(1u, families[1].count);
     EXPECT_EQ(clDevice.getDeviceInfo().queueOnHostProperties, families[1].properties);
 }
@@ -760,7 +760,7 @@ HWTEST_F(GetDeviceInfoQueueFamilyTest, givenSubDeviceWhenInitializingCapsThenRet
     EXPECT_EQ(3u, families[0].count);
     EXPECT_EQ(clDevice.getDeviceInfo().queueOnHostProperties, families[0].properties);
 
-    EXPECT_EQ(CL_QUEUE_CAPABILITY_ALL_INTEL, families[1].capabilities);
+    EXPECT_EQ(clDevice.getQueueFamilyCapabilities(EngineGroupType::Copy), families[1].capabilities);
     EXPECT_EQ(1u, families[1].count);
     EXPECT_EQ(clDevice.getDeviceInfo().queueOnHostProperties, families[1].properties);
 }

opencl/test/unit_test/mocks/mock_cl_device.h

@@ -34,6 +34,8 @@ class MockClDevice : public ClDevice {
     using ClDevice::driverInfo;
     using ClDevice::enabledClVersion;
     using ClDevice::getClDeviceName;
+    using ClDevice::getQueueFamilyCapabilities;
+    using ClDevice::getQueueFamilyCapabilitiesAll;
     using ClDevice::initializeCaps;
     using ClDevice::name;
     using ClDevice::ocl21FeaturesEnabled;