Move barrier programming to Encode class

zzdanowicz · Compute-Runtime-Automation · commit 134462919df0 · 2020-07-17T14:28:46.000+02:00
Related-To: NEO-4576

Change-Id: I34b93b3118528b449c4e1b81826f9784633377a9
Signed-off-by: Zbigniew Zdanowicz &lt;zbigniew.zdanowicz@intel.com&gt;
diff --git a/opencl/source/device_queue/device_queue_hw_bdw_plus.inl b/opencl/source/device_queue/device_queue_hw_bdw_plus.inl
@@ -5,6 +5,8 @@
  *
  */
 
+#include "shared/source/command_container/command_encoder.h"
+
 #include "opencl/source/cl_device/cl_device.h"
 #include "opencl/source/device_queue/device_queue_hw_base.inl"
 #include "opencl/source/program/block_kernel_manager.h"
@@ -203,9 +205,9 @@ void DeviceQueueHw<GfxFamily>::setupIndirectState(IndirectHeap &surfaceStateHeap
         pIDDestination[blockIndex + i].setKernelStartPointerHigh(blockKernelStartPointer >> 32);
         pIDDestination[blockIndex + i].setKernelStartPointer(static_cast<uint32_t>(blockKernelStartPointer));
         pIDDestination[blockIndex + i].setDenormMode(INTERFACE_DESCRIPTOR_DATA::DENORM_MODE_SETBYKERNEL);
-        HardwareCommandsHelper<GfxFamily>::programBarrierEnable(&pIDDestination[blockIndex + i],
-                                                                pBlockInfo->patchInfo.executionEnvironment->HasBarriers,
-                                                                parentKernel->getDevice().getHardwareInfo());
+        EncodeDispatchKernel<GfxFamily>::programBarrierEnable(&pIDDestination[blockIndex + i],
+                                                              pBlockInfo->patchInfo.executionEnvironment->HasBarriers,
+                                                              parentKernel->getDevice().getHardwareInfo());
 
         // Set offset to sampler states, block's DHSOffset is added by scheduler
         pIDDestination[blockIndex + i].setSamplerStatePointer(static_cast<uint32_t>(pBlockInfo->getBorderColorStateSize()));
diff --git a/opencl/source/helpers/hardware_commands_helper.h b/opencl/source/helpers/hardware_commands_helper.h
@@ -155,7 +155,6 @@ struct HardwareCommandsHelper : public PerThreadDataHelper {
     static void programMiAtomic(LinearStream &commandStream, uint64_t writeAddress, typename MI_ATOMIC::ATOMIC_OPCODES opcode, typename MI_ATOMIC::DATA_SIZE dataSize);
     static void programMiAtomic(MI_ATOMIC &atomic, uint64_t writeAddress, typename MI_ATOMIC::ATOMIC_OPCODES opcode, typename MI_ATOMIC::DATA_SIZE dataSize);
     static void programCacheFlushAfterWalkerCommand(LinearStream *commandStream, const CommandQueue &commandQueue, const Kernel *kernel, uint64_t postSyncAddress);
-    static void programBarrierEnable(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, uint32_t value, const HardwareInfo &hwInfo);
     static void adjustInterfaceDescriptorData(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const HardwareInfo &hwInfo);
 
     static const size_t alignInterfaceDescriptorData = 64 * sizeof(uint8_t);
diff --git a/opencl/source/helpers/hardware_commands_helper_base.inl b/opencl/source/helpers/hardware_commands_helper_base.inl
@@ -191,8 +191,9 @@ size_t HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(
         static_cast<typename INTERFACE_DESCRIPTOR_DATA::SHARED_LOCAL_MEMORY_SIZE>(HwHelperHw<GfxFamily>::get().computeSlmValues(kernel.slmTotalSize));
 
     interfaceDescriptor.setSharedLocalMemorySize(programmableIDSLMSize);
-    programBarrierEnable(&interfaceDescriptor, kernel.getKernelInfo().patchInfo.executionEnvironment->HasBarriers,
-                         kernel.getDevice().getHardwareInfo());
+    EncodeDispatchKernel<GfxFamily>::programBarrierEnable(&interfaceDescriptor,
+                                                          kernel.getKernelInfo().patchInfo.executionEnvironment->HasBarriers,
+                                                          kernel.getDevice().getHardwareInfo());
 
     PreemptionHelper::programInterfaceDescriptorDataPreemption<GfxFamily>(&interfaceDescriptor, preemptionMode);
     HardwareCommandsHelper<GfxFamily>::adjustInterfaceDescriptorData(&interfaceDescriptor, kernel.getDevice().getHardwareInfo());
diff --git a/opencl/source/helpers/hardware_commands_helper_bdw_plus.inl b/opencl/source/helpers/hardware_commands_helper_bdw_plus.inl
@@ -164,11 +164,6 @@ void HardwareCommandsHelper<GfxFamily>::programCacheFlushAfterWalkerCommand(Line
     *pipeControl = cmd;
 }
 
-template <typename GfxFamily>
-void HardwareCommandsHelper<GfxFamily>::programBarrierEnable(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, uint32_t value, const HardwareInfo &hwInfo) {
-    pInterfaceDescriptor->setBarrierEnable(value);
-}
-
 template <typename GfxFamily>
 void HardwareCommandsHelper<GfxFamily>::adjustInterfaceDescriptorData(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, const HardwareInfo &hwInfo) {}
 } // namespace NEO
diff --git a/opencl/test/unit_test/gen_common/CMakeLists.txt b/opencl/test/unit_test/gen_common/CMakeLists.txt
@@ -9,7 +9,6 @@ set(IGDRCL_SRCS_tests_gen_common
   ${CMAKE_CURRENT_SOURCE_DIR}/exclude_tests/exclude_test_declare.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/exclude_tests/exclude_test_exclude.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/gen_commands_common_validation.h
-  ${CMAKE_CURRENT_SOURCE_DIR}/hw_cmds_tests.cpp
   ${CMAKE_CURRENT_SOURCE_DIR}/matchers.h
   ${NEO_SHARED_TEST_DIRECTORY}/unit_test/test_macros/header${BRANCH_DIR_SUFFIX}/test.h
 )
diff --git a/opencl/test/unit_test/gen_common/hw_cmds_tests.cpp b/opencl/test/unit_test/gen_common/hw_cmds_tests.cpp
diff --git a/shared/source/command_container/command_encoder.h b/shared/source/command_container/command_encoder.h
@@ -57,6 +57,8 @@ struct EncodeDispatchKernel {
                                  bool inlineDataProgrammingRequired,
                                  bool isIndirect,
                                  uint32_t requiredWorkGroupOrder);
+
+    static void programBarrierEnable(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor, uint32_t value, const HardwareInfo &hwInfo);
 };
 
 template <typename GfxFamily>
diff --git a/shared/source/command_container/command_encoder_base.inl b/shared/source/command_container/command_encoder_base.inl
@@ -66,7 +66,9 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container,
     auto numThreadsPerThreadGroup = dispatchInterface->getNumThreadsPerThreadGroup();
     idd.setNumberOfThreadsInGpgpuThreadGroup(numThreadsPerThreadGroup);
 
-    idd.setBarrierEnable(kernelDescriptor.kernelAttributes.flags.usesBarriers);
+    EncodeDispatchKernel<Family>::programBarrierEnable(&idd,
+                                                       kernelDescriptor.kernelAttributes.hasBarriers,
+                                                       container.getDevice()->getHardwareInfo());
     auto slmSize = static_cast<typename INTERFACE_DESCRIPTOR_DATA::SHARED_LOCAL_MEMORY_SIZE>(
         HwHelperHw<Family>::get().computeSlmValues(dispatchInterface->getSlmTotalSize()));
     idd.setSharedLocalMemorySize(
@@ -339,6 +341,13 @@ void EncodeDispatchKernel<GfxFamily>::encodeThreadData(WALKER_TYPE &walkerCmd,
     walkerCmd.setBottomExecutionMask(maxDword);
 }
 
+template <typename GfxFamily>
+void EncodeDispatchKernel<GfxFamily>::programBarrierEnable(INTERFACE_DESCRIPTOR_DATA *pInterfaceDescriptor,
+                                                           uint32_t value,
+                                                           const HardwareInfo &hwInfo) {
+    pInterfaceDescriptor->setBarrierEnable(value);
+}
+
 template <typename GfxFamily>
 void EncodeMiFlushDW<GfxFamily>::appendMiFlushDw(MI_FLUSH_DW *miFlushDwCmd) {}
 
diff --git a/shared/source/kernel/kernel_descriptor.h b/shared/source/kernel/kernel_descriptor.h
@@ -46,6 +46,7 @@ struct KernelDescriptor final {
         uint32_t perThreadScratchSize[2] = {0U, 0U};
         uint32_t perThreadPrivateMemorySize = 0U;
         uint32_t perThreadSystemThreadSurfaceSize = 0U;
+        uint32_t hasBarriers = 0u;
         uint16_t requiredWorkgroupSize[3] = {0U, 0U, 0U};
         uint16_t crossThreadDataSize = 0U;
         uint16_t perThreadDataSize = 0U;
diff --git a/shared/source/kernel/kernel_descriptor_from_patchtokens.cpp b/shared/source/kernel/kernel_descriptor_from_patchtokens.cpp
@@ -49,6 +49,7 @@ void populateKernelDescriptor(KernelDescriptor &dst, const SPatchExecutionEnviro
     dst.kernelAttributes.simdSize = execEnv.LargestCompiledSIMDSize;
     dst.kernelAttributes.flags.usesDeviceSideEnqueue = (0 != execEnv.HasDeviceEnqueue);
     dst.kernelAttributes.flags.usesBarriers = (0 != execEnv.HasBarriers);
+    dst.kernelAttributes.hasBarriers = execEnv.HasBarriers;
     dst.kernelAttributes.flags.requiresDisabledMidThreadPreemption = (0 != execEnv.DisableMidThreadPreemption);
     dst.kernelMetadata.compiledSubGroupsNumber = execEnv.CompiledSubGroupsNumber;
     dst.kernelAttributes.flags.usesFencesForReadWriteImages = (0 != execEnv.UsesFencesForReadWriteImages);
diff --git a/shared/test/unit_test/encoders/test_encode_dispatch_kernel.cpp b/shared/test/unit_test/encoders/test_encode_dispatch_kernel.cpp
@@ -14,10 +14,13 @@
 #include "shared/test/unit_test/device_binary_format/patchtokens_tests.h"
 #include "shared/test/unit_test/fixtures/command_container_fixture.h"
 #include "shared/test/unit_test/helpers/debug_manager_state_restore.h"
+#include "shared/test/unit_test/mocks/mock_device.h"
 #include "shared/test/unit_test/mocks/mock_dispatch_kernel_encoder_interface.h"
 
 #include "opencl/source/helpers/hardware_commands_helper.h"
 #include "opencl/test/unit_test/gen_common/matchers.h"
+#include "test.h"
+
 using namespace NEO;
 
 using CommandEncodeStatesTest = Test<CommandEncodeStatesFixture>;
@@ -707,3 +710,23 @@ HWTEST_F(WalkerThreadTest, givenDebugFlagEnabledWhenKernelDescriptorInlineDataDi
 
     EXPECT_FALSE(EncodeDispatchKernel<FamilyType>::inlineDataProgrammingRequired(kernelDesc));
 }
+
+using namespace NEO;
+
+using InterfaceDescriptorDataTests = ::testing::Test;
+
+HWCMDTEST_F(IGFX_GEN8_CORE, InterfaceDescriptorDataTests, givenVariousValuesWhenCallingSetBarrierEnableThenCorrectValueIsSet) {
+    using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
+    INTERFACE_DESCRIPTOR_DATA idd = FamilyType::cmdInitInterfaceDescriptorData;
+    MockDevice device;
+    auto hwInfo = device.getHardwareInfo();
+
+    EncodeDispatchKernel<FamilyType>::programBarrierEnable(&idd, 0, hwInfo);
+    EXPECT_FALSE(idd.getBarrierEnable());
+
+    EncodeDispatchKernel<FamilyType>::programBarrierEnable(&idd, 1, hwInfo);
+    EXPECT_TRUE(idd.getBarrierEnable());
+
+    EncodeDispatchKernel<FamilyType>::programBarrierEnable(&idd, 2, hwInfo);
+    EXPECT_TRUE(idd.getBarrierEnable());
+}

Original file line number	Diff line number	Diff line change
`@@ -9,7 +9,6 @@ set(IGDRCL_SRCS_tests_gen_common`
`9`	`9`	`${CMAKE_CURRENT_SOURCE_DIR}/exclude_tests/exclude_test_declare.cpp`
`10`	`10`	`${CMAKE_CURRENT_SOURCE_DIR}/exclude_tests/exclude_test_exclude.cpp`
`11`	`11`	`${CMAKE_CURRENT_SOURCE_DIR}/gen_commands_common_validation.h`
`12`		`- ${CMAKE_CURRENT_SOURCE_DIR}/hw_cmds_tests.cpp`
`13`	`12`	`${CMAKE_CURRENT_SOURCE_DIR}/matchers.h`
`14`	`13`	`${NEO_SHARED_TEST_DIRECTORY}/unit_test/test_macros/header${BRANCH_DIR_SUFFIX}/test.h`
`15`	`14`	`)`