performance: align structures for 64-bit platforms

- _tagCLGLBufferInfo 56 -> 48 bytes
- InlineSamplerBaseT 16 -> 12 bytes
- CsrSelectionArgs 88 -> 80 bytes
- EngineInfo 64 -> 48 bytes
- FragmentStorage 48 -> 40 bytes
- SvmFreeUserData 40 -> 32 bytes
- PayloadArgumentBaseT 40 -> 36 bytes
- ImageDescriptor 72 -> 64 bytes
- PerThreadMemoryBufferBaseT 16 -> 12 bytes
- CopyEngineState 24 -> 16 bytes
- BatchBuffer 120 -> 112 bytes
- DebugMetadata 32 -> 24 bytes
- ImmediateFlushData 48 -> 40 bytes

Author: GermanAizek

opencl/extensions/public/cl_gl_private_intel_structures.h

@@ -48,15 +48,15 @@ typedef struct _tagGLCLResourceInfo {
 } GL_CL_RESOURCE_INFO, *PGL_CL_RESOURCE_INFO;
 
 typedef struct _tagCLGLBufferInfo {
-    GLenum bufferName;
-    unsigned int globalShareHandle;
     GMM_RESOURCE_INFO *pGmmResInfo; /// Pointer to GMMResInfo from GL that will be copied in CL (GL)
     GLvoid *pSysMem;
+    GLvoid *pReleaseData;
+    GLenum bufferName;
+    unsigned int globalShareHandle;
     GLint bufferSize;
     GLint bufferOffset;
-    GLboolean oglSynchronized;
     GMM_STATUS status;
-    GLvoid *pReleaseData;
+    GLboolean oglSynchronized;
     GLboolean createOrDestroy;
 } CL_GL_BUFFER_INFO, *PCL_GL_BUFFER_INFO;
 

opencl/source/command_queue/copy_engine_state.h

@@ -13,8 +13,8 @@
 
 namespace NEO {
 struct CopyEngineState {
-    aub_stream::EngineType engineType = aub_stream::EngineType::NUM_ENGINES;
     TaskCountType taskCount = 0;
+    aub_stream::EngineType engineType = aub_stream::EngineType::NUM_ENGINES;
     bool csrClientRegistered = false;
 
     bool isValid() const {

opencl/source/command_queue/csr_selection_args.h

@@ -24,21 +24,21 @@ struct CsrSelectionArgs {
         const size_t *imageOrigin = nullptr;
     };
 
-    cl_command_type cmdType;
-    const size_t *size = nullptr;
     Resource srcResource;
     Resource dstResource;
+    const size_t *size = nullptr;
+    cl_command_type cmdType;
     TransferDirection direction;
 
     CsrSelectionArgs(cl_command_type cmdType, const size_t *size)
-        : cmdType(cmdType),
-          size(size),
+        : size(size),
+          cmdType(cmdType),
           direction(TransferDirection::hostToHost) {}
 
     template <typename ResourceType>
     CsrSelectionArgs(cl_command_type cmdType, ResourceType *src, ResourceType *dst, uint32_t rootDeviceIndex, const size_t *size)
-        : cmdType(cmdType),
-          size(size) {
+        : size(size),
+          cmdType(cmdType) {
         if (src) {
             processResource(*src, rootDeviceIndex, this->srcResource);
         }

opencl/source/command_queue/enqueue_svm.h

@@ -25,20 +25,20 @@ using SvmFreeClbT = void(CL_CALLBACK *)(cl_command_queue queue,
                                         void *userData);
 
 struct SvmFreeUserData {
-    cl_uint numSvmPointers;
     void **svmPointers;
     SvmFreeClbT clb;
     void *userData;
+    cl_uint numSvmPointers;
     bool ownsEventDeletion;
 
     SvmFreeUserData(cl_uint numSvmPointers,
                     void **svmPointers, SvmFreeClbT clb,
                     void *userData,
                     bool ownsEventDeletion)
-        : numSvmPointers(numSvmPointers),
-          svmPointers(svmPointers),
+        : svmPointers(svmPointers),
           clb(clb),
           userData(userData),
+          numSvmPointers(numSvmPointers),
           ownsEventDeletion(ownsEventDeletion){};
 };
 

opencl/test/unit_test/command_queue/blit_enqueue_1_tests.cpp

@@ -1251,7 +1251,7 @@ HWTEST_TEMPLATED_F(BlitEnqueueTaskCountTests, whenWaitUntilCompletionCalledThenW
     uint32_t gpgpuTaskCount = 123;
     uint32_t bcsTaskCount = 123;
 
-    CopyEngineState bcsState{bcsCsr->getOsContext().getEngineType(), bcsTaskCount};
+    CopyEngineState bcsState{bcsTaskCount, bcsCsr->getOsContext().getEngineType()};
     commandQueue->waitUntilComplete(gpgpuTaskCount, Range{&bcsState}, 0, false);
 
     EXPECT_EQ(gpgpuTaskCount, static_cast<UltCommandStreamReceiver<FamilyType> *>(gpgpuCsr)->latestWaitForCompletionWithTimeoutTaskCount.load());

opencl/test/unit_test/command_queue/command_queue_hw_1_tests.cpp

@@ -55,7 +55,7 @@ HWTEST_F(CommandQueueHwTest, whenCallingIsCompletedThenTestTaskCountValue) {
     bcsCsr->setupContext(*osContext);
     bcsCsr->initializeTagAllocation();
     EngineControl control(bcsCsr.get(), osContext.get());
-    CopyEngineState state{aub_stream::EngineType::ENGINE_BCS, 1, false};
+    CopyEngineState state{1, aub_stream::EngineType::ENGINE_BCS, false};
 
     MockCommandQueueHw<FamilyType> cmdQ(context, pClDevice, nullptr);
 

opencl/test/unit_test/command_stream/cl_tbx_command_stream_tests.cpp

@@ -47,7 +47,7 @@ HWTEST_F(ClTbxCommandStreamTests, givenTbxCsrWhenDispatchBlitEnqueueThenProcessC
     cmdQ.clearBcsEngines();
     cmdQ.bcsEngines[0] = &engineControl1;
 
-    cmdQ.bcsStates[0] = {aub_stream::ENGINE_BCS, 0, false};
+    cmdQ.bcsStates[0] = {0, aub_stream::ENGINE_BCS, false};
 
     cl_int error = CL_SUCCESS;
     std::unique_ptr<Buffer> buffer(Buffer::create(&context, 0, 1, nullptr, error));

opencl/test/unit_test/mem_obj/buffer_bcs_tests.cpp

@@ -619,7 +619,7 @@ HWTEST_TEMPLATED_F(BcsBufferTests, givenAllBcsEnginesReadyWhenWaitingForEventThe
     ultCsr2.initializeTagAllocation();
     ultCsr2.setupContext(osContext);
 
-    CopyEngineState copyEngineState = {aub_stream::EngineType::ENGINE_BCS2, 2, false};
+    CopyEngineState copyEngineState = {2, aub_stream::EngineType::ENGINE_BCS2, false};
     EngineControl engineControl = {&ultCsr2, &osContext};
     auto bcs2Index = EngineHelpers::getBcsIndex(aub_stream::EngineType::ENGINE_BCS2);
     mockCmdQ->bcsStates[bcs2Index] = copyEngineState;

opencl/test/unit_test/mem_obj/image_tests.cpp

@@ -1559,7 +1559,7 @@ TEST(ImageConvertDescriptorTest, givenClImageDescWhenConvertedThenCorrectImageDe
 }
 
 TEST(ImageConvertDescriptorTest, givenImageDescriptorWhenConvertedThenCorrectClImageDescIsReturned) {
-    ImageDescriptor desc = {ImageType::image2D, 16, 24, 1, 1, 1024, 2048, 1, 3, false};
+    ImageDescriptor desc = {16, 24, 1, 1, 1024, 2048, ImageType::image2D, 1, 3, false};
     auto clDesc = Image::convertDescriptor(desc);
 
     EXPECT_EQ(clDesc.image_type, static_cast<cl_mem_object_type>(CL_MEM_OBJECT_IMAGE2D));

shared/source/command_stream/command_stream_receiver_hw.h

@@ -25,10 +25,10 @@ class CommandStreamReceiverHw : public CommandStreamReceiver {
     using STATE_BASE_ADDRESS = typename GfxFamily::STATE_BASE_ADDRESS;
 
     struct ImmediateFlushData {
-        PipelineSelectArgs pipelineSelectArgs{};
-        size_t estimatedSize = 0;
         void *endPtr = nullptr;
+        size_t estimatedSize = 0;
         size_t csrStartOffset = 0;
+        PipelineSelectArgs pipelineSelectArgs{};
 
         bool pipelineSelectFullConfigurationNeeded = false;
         bool pipelineSelectDirty = false;

shared/source/command_stream/command_stream_receiver_simulated_common_hw.h

@@ -81,12 +81,12 @@ class CommandStreamReceiverSimulatedCommonHw : public CommandStreamReceiverHw<Gf
 
     struct EngineInfo {
         void *pLRCA;
-        uint32_t ggttLRCA;
         void *pGlobalHWStatusPage;
-        uint32_t ggttHWSP;
         void *pRingBuffer;
-        uint32_t ggttRingBuffer;
         size_t sizeRingBuffer;
+        uint32_t ggttLRCA;
+        uint32_t ggttHWSP;
+        uint32_t ggttRingBuffer;
         uint32_t tailRingBuffer;
     } engineInfo = {};
 

shared/source/command_stream/submissions_aggregator.cpp

@@ -99,11 +99,11 @@ NEO::BatchBuffer::BatchBuffer(GraphicsAllocation *commandBufferAllocation, size_
                               size_t usedSize, LinearStream *stream, void *endCmdPtr, uint32_t numCsrClients, bool hasStallingCmds,
                               bool hasRelaxedOrderingDependencies, bool dispatchMonitorFence, bool taskCountUpdateOnly)
     : commandBufferAllocation(commandBufferAllocation), startOffset(startOffset),
-      chainedBatchBufferStartOffset(chainedBatchBufferStartOffset), taskStartAddress(taskStartAddress), chainedBatchBuffer(chainedBatchBuffer),
-      lowPriority(lowPriority),
-      throttle(throttle), sliceCount(sliceCount),
-      usedSize(usedSize), stream(stream), endCmdPtr(endCmdPtr), numCsrClients(numCsrClients), hasStallingCmds(hasStallingCmds),
-      hasRelaxedOrderingDependencies(hasRelaxedOrderingDependencies), dispatchMonitorFence(dispatchMonitorFence), taskCountUpdateOnly(taskCountUpdateOnly) {}
+      chainedBatchBufferStartOffset(chainedBatchBufferStartOffset), taskStartAddress(taskStartAddress), stream(stream), endCmdPtr(endCmdPtr),
+      numCsrClients(numCsrClients), hasStallingCmds(hasStallingCmds), hasRelaxedOrderingDependencies(hasRelaxedOrderingDependencies),
+      dispatchMonitorFence(dispatchMonitorFence), taskCountUpdateOnly(taskCountUpdateOnly), lowPriority(lowPriority), throttle(throttle),
+      chainedBatchBuffer(chainedBatchBuffer), sliceCount(sliceCount),
+      usedSize(usedSize) {}
 
 NEO::CommandBuffer::CommandBuffer(Device &device) : device(device) {
     flushStamp.reset(new FlushStampTracker(false));

shared/source/command_stream/submissions_aggregator.h

@@ -46,30 +46,31 @@ struct BatchBuffer {
                 bool dispatchMonitorFence,
                 bool taskCountUpdateOnly);
     BatchBuffer() {}
+
+    PagingFenceSemaphoreInfo pagingFenceSemInfo{};
+
     GraphicsAllocation *commandBufferAllocation = nullptr;
     ResidencyContainer *allocationsForResidency = nullptr;
     size_t startOffset = 0u;
     size_t chainedBatchBufferStartOffset = 0u;
     uint64_t taskStartAddress = 0; // if task not available, use CSR stream
 
-    GraphicsAllocation *chainedBatchBuffer = nullptr;
-    bool lowPriority = false;
-    QueueThrottle throttle = QueueThrottle::MEDIUM;
-    uint64_t sliceCount = QueueSliceCount::defaultSliceCount;
-    size_t usedSize = 0u;
-
     // only used in drm csr in gem close worker active mode
     LinearStream *stream = nullptr;
     void *endCmdPtr = nullptr;
     uint32_t numCsrClients = 0;
 
-    PagingFenceSemaphoreInfo pagingFenceSemInfo{};
-
     bool hasStallingCmds = false;
     bool hasRelaxedOrderingDependencies = false;
     bool disableFlatRingBuffer = false;
     bool dispatchMonitorFence = false;
     bool taskCountUpdateOnly = false;
+
+    bool lowPriority = false;
+    QueueThrottle throttle = QueueThrottle::MEDIUM;
+    GraphicsAllocation *chainedBatchBuffer = nullptr;
+    uint64_t sliceCount = QueueSliceCount::defaultSliceCount;
+    size_t usedSize = 0u;
 };
 
 struct CommandBuffer : public IDNode<CommandBuffer> {

shared/source/device_binary_format/zebin/zeinfo.h

@@ -634,12 +634,12 @@ inline constexpr BtiValueT btiValue = -1;
 } // namespace Defaults
 
 struct PayloadArgumentBaseT {
-    ArgTypeT argType = argTypeUnknown;
     OffsetT offset = Defaults::offset;
     SourceOffseT sourceOffset = Defaults::sourceOffset;
     SizeT size = 0;
     ArgIndexT argIndex = Defaults::argIndex;
     BtiValueT btiValue = Defaults::btiValue;
+    ArgTypeT argType = argTypeUnknown;
     AddrmodeT addrmode = memoryAddressingModeUnknown;
     AddrspaceT addrspace = addressSpaceUnknown;
     AccessTypeT accessType = accessTypeUnknown;
@@ -692,9 +692,9 @@ inline constexpr Slot slot = 0U;
 } // namespace Defaults
 
 struct PerThreadMemoryBufferBaseT {
+    SizeT size = 0U;
     AllocationType allocationType = AllocationTypeUnknown;
     MemoryUsageT memoryUsage = MemoryUsageUnknown;
-    SizeT size = 0U;
     IsSimtThreadT isSimtThread = Defaults::isSimtThread;
     Slot slot = Defaults::slot;
 };
@@ -732,8 +732,8 @@ inline constexpr NormalizedT normalized = false;
 
 struct InlineSamplerBaseT {
     SamplerIndexT samplerIndex = Defaults::samplerIndex;
-    AddrModeT addrMode = Defaults::addrMode;
     FilterModeT filterMode = Defaults::filterMode;
+    AddrModeT addrMode = Defaults::addrMode;
     NormalizedT normalized = Defaults::normalized;
 };
 } // namespace InlineSamplers

shared/source/helpers/surface_format_info.h

@@ -218,13 +218,13 @@ enum class ImageType {
 };
 
 struct ImageDescriptor {
-    ImageType imageType;
     size_t imageWidth;
     size_t imageHeight;
     size_t imageDepth;
     size_t imageArraySize;
     size_t imageRowPitch;
     size_t imageSlicePitch;
+    ImageType imageType;
     uint32_t numMipLevels;
     uint32_t numSamples;
     bool fromParent;

shared/source/memory_manager/host_ptr_defines.h

@@ -51,9 +51,9 @@ struct AllocationRequirements {
 struct FragmentStorage {
     const void *fragmentCpuPointer = nullptr;
     size_t fragmentSize = 0;
-    int refCount = 0;
     OsHandle *osInternalStorage = nullptr;
     ResidencyData *residency = nullptr;
+    int refCount = 0;
     bool driverAllocation = false;
 };
 

shared/source/os_interface/linux/xe/ioctl_helper_xe.h

@@ -190,9 +190,9 @@ class IoctlHelperXe : public IoctlHelper {
 
     std::unique_ptr<XeDrm::drm_xe_engine_class_instance> defaultEngine;
     struct DebugMetadata {
-        DrmResourceClass type;
         uint64_t offset;
         uint64_t size;
+        DrmResourceClass type;
         bool isCookie;
     };