fix: Copy tiled 1D array per array element with BLT

MaciejPlewka · Compute-Runtime-Automation · commit c5e1fcf313ea · 2025-03-07T17:29:04.000+01:00
Related-To: NEO-14147, HSD-14024424096, HSD-14024424178
Signed-off-by: Maciej Plewka &lt;maciej.plewka@intel.com&gt;
diff --git a/level_zero/core/source/cmdlist/cmdlist_hw.inl b/level_zero/core/source/cmdlist/cmdlist_hw.inl
@@ -1459,6 +1459,8 @@ ze_result_t CommandListCoreFamily<gfxCoreFamily>::appendCopyImageBlit(NEO::Graph
     commandContainer.addToResidencyContainer(clearColorAllocation);
 
     appendEventForProfiling(signalEvent, nullptr, true, false, false, true);
+    blitProperties.transform1DArrayTo2DArrayIfNeeded();
+
     NEO::BlitCommandsHelper<GfxFamily>::dispatchBlitCommandsForImageRegion(blitProperties, *commandContainer.getCommandStream(), *dummyBlitWa.rootDeviceEnvironment);
     dummyBlitWa.isWaRequired = true;
 
diff --git a/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_blit.cpp b/level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_blit.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2020-2024 Intel Corporation
+ * Copyright (C) 2020-2025 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -8,6 +8,9 @@
 #include "shared/source/helpers/register_offsets.h"
 #include "shared/test/common/cmd_parse/gen_cmd_parse.h"
 #include "shared/test/common/mocks/mock_device.h"
+#include "shared/test/common/mocks/mock_gmm.h"
+#include "shared/test/common/mocks/mock_gmm_client_context.h"
+#include "shared/test/common/mocks/mock_gmm_resource_info.h"
 #include "shared/test/common/mocks/mock_graphics_allocation.h"
 #include "shared/test/common/test_macros/hw_test.h"
 
@@ -352,5 +355,79 @@ HWTEST2_F(AppendMemoryCopyFromContext, givenCommandListThenUpOnPerformingAppendM
     EXPECT_EQ(ZE_RESULT_SUCCESS, result);
 }
 
+struct IsAtLeastXeHpCoreAndNotXe2HpgCoreWith2DArrayImageSupport {
+    template <PRODUCT_FAMILY productFamily>
+    static constexpr bool isMatched() {
+        return IsAtLeastGfxCore<IGFX_XE_HP_CORE>::isMatched<productFamily>() && !IsXe2HpgCore::isMatched<productFamily>() && NEO::HwMapper<productFamily>::GfxProduct::supportsSampler;
+    }
+};
+HWTEST2_F(AppendMemoryCopyTests, givenCopyCommandListWhenTiled1DArrayImagePassedToImageCopyBlitThenTransformedTo2DArrayCopy, IsAtLeastXeHpCoreAndNotXe2HpgCoreWith2DArrayImageSupport) {
+    using XY_BLOCK_COPY_BLT = typename FamilyType::XY_BLOCK_COPY_BLT;
+    auto commandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
+    commandList->initialize(device, NEO::EngineGroupType::copy, 0u);
+
+    auto gmmSrc = std::make_unique<MockGmm>(device->getNEODevice()->getGmmHelper());
+    auto resourceInfoSrc = static_cast<MockGmmResourceInfo *>(gmmSrc->gmmResourceInfo.get());
+    resourceInfoSrc->getResourceFlags()->Info.Tile64 = 1;
+    resourceInfoSrc->mockResourceCreateParams.Type = GMM_RESOURCE_TYPE::RESOURCE_1D;
+    resourceInfoSrc->mockResourceCreateParams.ArraySize = 8;
+
+    NEO::MockGraphicsAllocation mockAllocationSrc(0, 1u /*num gmms*/, NEO::AllocationType::internalHostMemory,
+                                                  reinterpret_cast<void *>(0x1234), 0x1000, 0, sizeof(uint32_t),
+                                                  MemoryPool::system4KBPages, MemoryManager::maxOsContextCount);
+    NEO::MockGraphicsAllocation mockAllocationDst(0, 1u /*num gmms*/, NEO::AllocationType::internalHostMemory,
+                                                  reinterpret_cast<void *>(0x1234), 0x1000, 0, sizeof(uint32_t),
+                                                  MemoryPool::system4KBPages, MemoryManager::maxOsContextCount);
+
+    mockAllocationSrc.setGmm(gmmSrc.get(), 0);
+    mockAllocationDst.setGmm(gmmSrc.get(), 0);
+
+    size_t arrayLevels = 8;
+    size_t depth = 1;
+    commandList->appendCopyImageBlit(&mockAllocationDst, &mockAllocationSrc, {0, 0, 0}, {0, 0, 0}, 4, 4, 4, 4, 1, {1, arrayLevels, depth}, {1, arrayLevels, depth}, {1, arrayLevels, depth}, nullptr);
+    GenCmdList cmdList;
+    ASSERT_TRUE(FamilyType::Parse::parseCommandBuffer(
+        cmdList, ptrOffset(commandList->getCmdContainer().getCommandStream()->getCpuBase(), 0), commandList->getCmdContainer().getCommandStream()->getUsed()));
+    auto itor = find<XY_BLOCK_COPY_BLT *>(cmdList.begin(), cmdList.end());
+    EXPECT_NE(cmdList.end(), itor);
+    auto cmd = genCmdCast<XY_BLOCK_COPY_BLT *>(*itor);
+    EXPECT_EQ(cmd->getSourceSurfaceDepth(), arrayLevels);
+    EXPECT_EQ(cmd->getSourceSurfaceHeight(), depth);
+}
+
+HWTEST2_F(AppendMemoryCopyTests, givenCopyCommandListWhenNotTiled1DArrayImagePassedToImageCopyBlitThenNotTransformedTo2DArrayCopy, IsAtLeastXeHpCoreAndNotXe2HpgCoreWith2DArrayImageSupport) {
+    using XY_BLOCK_COPY_BLT = typename FamilyType::XY_BLOCK_COPY_BLT;
+    auto commandList = std::make_unique<WhiteBox<::L0::CommandListCoreFamily<gfxCoreFamily>>>();
+    commandList->initialize(device, NEO::EngineGroupType::copy, 0u);
+
+    auto gmmSrc = std::make_unique<MockGmm>(device->getNEODevice()->getGmmHelper());
+    auto resourceInfoSrc = static_cast<MockGmmResourceInfo *>(gmmSrc->gmmResourceInfo.get());
+    resourceInfoSrc->getResourceFlags()->Info.Tile64 = 0;
+    resourceInfoSrc->mockResourceCreateParams.Type = GMM_RESOURCE_TYPE::RESOURCE_1D;
+    resourceInfoSrc->mockResourceCreateParams.ArraySize = 8;
+
+    NEO::MockGraphicsAllocation mockAllocationSrc(0, 1u /*num gmms*/, NEO::AllocationType::internalHostMemory,
+                                                  reinterpret_cast<void *>(0x1234), 0x1000, 0, sizeof(uint32_t),
+                                                  MemoryPool::system4KBPages, MemoryManager::maxOsContextCount);
+    NEO::MockGraphicsAllocation mockAllocationDst(0, 1u /*num gmms*/, NEO::AllocationType::internalHostMemory,
+                                                  reinterpret_cast<void *>(0x1234), 0x1000, 0, sizeof(uint32_t),
+                                                  MemoryPool::system4KBPages, MemoryManager::maxOsContextCount);
+
+    mockAllocationSrc.setGmm(gmmSrc.get(), 0);
+    mockAllocationDst.setGmm(gmmSrc.get(), 0);
+
+    size_t arrayLevels = 8;
+    size_t depth = 1;
+    commandList->appendCopyImageBlit(&mockAllocationDst, &mockAllocationSrc, {0, 0, 0}, {0, 0, 0}, 1, 1, 1, 1, 1, {1, arrayLevels, depth}, {1, arrayLevels, depth}, {1, arrayLevels, depth}, nullptr);
+    GenCmdList cmdList;
+    ASSERT_TRUE(FamilyType::Parse::parseCommandBuffer(
+        cmdList, ptrOffset(commandList->getCmdContainer().getCommandStream()->getCpuBase(), 0), commandList->getCmdContainer().getCommandStream()->getUsed()));
+    auto itor = find<XY_BLOCK_COPY_BLT *>(cmdList.begin(), cmdList.end());
+    EXPECT_NE(cmdList.end(), itor);
+    auto cmd = genCmdCast<XY_BLOCK_COPY_BLT *>(*itor);
+    EXPECT_EQ(cmd->getSourceSurfaceDepth(), depth);
+    EXPECT_EQ(cmd->getSourceSurfaceHeight(), arrayLevels);
+}
+
 } // namespace ult
 } // namespace L0
diff --git a/opencl/source/helpers/cl_blit_properties.h b/opencl/source/helpers/cl_blit_properties.h
@@ -56,6 +56,7 @@ struct ClBlitProperties {
                 blitProperties.blitDirection = blitDirection;
                 setBlitPropertiesForImage(blitProperties, builtinOpParams);
             }
+            blitProperties.transform1DArrayTo2DArrayIfNeeded();
             return blitProperties;
         }
 
@@ -142,6 +143,7 @@ struct ClBlitProperties {
             setBlitPropertiesForImage(blitProperties, builtinOpParams);
         }
 
+        blitProperties.transform1DArrayTo2DArrayIfNeeded();
         return blitProperties;
     }
 
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
@@ -20,6 +20,9 @@
 #include "shared/test/common/mocks/mock_allocation_properties.h"
 #include "shared/test/common/mocks/mock_direct_submission_hw.h"
 #include "shared/test/common/mocks/mock_gfx_core_helper.h"
+#include "shared/test/common/mocks/mock_gmm.h"
+#include "shared/test/common/mocks/mock_gmm_client_context.h"
+#include "shared/test/common/mocks/mock_gmm_resource_info.h"
 #include "shared/test/common/mocks/mock_internal_allocation_storage.h"
 #include "shared/test/common/mocks/mock_memory_manager.h"
 #include "shared/test/common/mocks/mock_timestamp_container.h"
@@ -2201,3 +2204,57 @@ HWTEST_F(BcsTests, givenHostPtrToImageWhenBlitBufferIsCalledThenBlitCmdIsFound)
     auto cmdIterator = find<typename FamilyType::XY_BLOCK_COPY_BLT *>(hwParser.cmdList.begin(), hwParser.cmdList.end());
     EXPECT_NE(hwParser.cmdList.end(), cmdIterator);
 }
+HWTEST_F(BcsTests, given1DTiledArrayImageWhenConstructPropertiesThenImageTransformedTo2DArray) {
+    if (!pDevice->getHardwareInfo().capabilityTable.supportsImages) {
+        GTEST_SKIP();
+    }
+
+    auto gmmSrc = std::make_unique<MockGmm>(pDevice->getGmmHelper());
+    auto resourceInfoSrc = static_cast<MockGmmResourceInfo *>(gmmSrc->gmmResourceInfo.get());
+    resourceInfoSrc->getResourceFlags()->Info.Tile64 = 1;
+    resourceInfoSrc->mockResourceCreateParams.Type = GMM_RESOURCE_TYPE::RESOURCE_1D;
+    resourceInfoSrc->mockResourceCreateParams.ArraySize = 8;
+
+    std::unique_ptr<Image> image(Image2dHelper<>::create(context.get()));
+    auto oldGmm = std::unique_ptr<Gmm>(image->getGraphicsAllocation(pDevice->getRootDeviceIndex())->getDefaultGmm());
+    image->getGraphicsAllocation(pDevice->getRootDeviceIndex())->setGmm(gmmSrc.release(), 0);
+    BuiltinOpParams builtinOpParams{};
+    builtinOpParams.srcMemObj = image.get();
+    builtinOpParams.dstMemObj = image.get();
+    builtinOpParams.size = {1, 8, 1};
+
+    auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
+    csr.getCmdSizeForComputeMode();
+    auto blitProperties = ClBlitProperties::constructProperties(BlitterConstants::BlitDirection::imageToImage,
+                                                                csr,
+                                                                builtinOpParams);
+    EXPECT_EQ(blitProperties.copySize.z, builtinOpParams.size.y);
+    EXPECT_EQ(blitProperties.copySize.y, builtinOpParams.size.z);
+}
+HWTEST_F(BcsTests, given1DNotTiledArrayImageWhenConstructPropertiesThenImageNotTransformedTo2DArray) {
+    if (!pDevice->getHardwareInfo().capabilityTable.supportsImages) {
+        GTEST_SKIP();
+    }
+
+    auto gmmSrc = std::make_unique<MockGmm>(pDevice->getGmmHelper());
+    auto resourceInfoSrc = static_cast<MockGmmResourceInfo *>(gmmSrc->gmmResourceInfo.get());
+    resourceInfoSrc->getResourceFlags()->Info.Tile64 = 0;
+    resourceInfoSrc->mockResourceCreateParams.Type = GMM_RESOURCE_TYPE::RESOURCE_1D;
+    resourceInfoSrc->mockResourceCreateParams.ArraySize = 8;
+
+    std::unique_ptr<Image> image(Image2dHelper<>::create(context.get()));
+    auto oldGmm = std::unique_ptr<Gmm>(image->getGraphicsAllocation(pDevice->getRootDeviceIndex())->getDefaultGmm());
+    image->getGraphicsAllocation(pDevice->getRootDeviceIndex())->setGmm(gmmSrc.release(), 0);
+    BuiltinOpParams builtinOpParams{};
+    builtinOpParams.srcMemObj = image.get();
+    builtinOpParams.dstMemObj = image.get();
+    builtinOpParams.size = {1, 8, 1};
+
+    auto &csr = pDevice->getUltCommandStreamReceiver<FamilyType>();
+    csr.getCmdSizeForComputeMode();
+    auto blitProperties = ClBlitProperties::constructProperties(BlitterConstants::BlitDirection::imageToImage,
+                                                                csr,
+                                                                builtinOpParams);
+    EXPECT_EQ(blitProperties.copySize.y, builtinOpParams.size.y);
+    EXPECT_EQ(blitProperties.copySize.z, builtinOpParams.size.z);
+}
diff --git a/shared/source/helpers/blit_properties.cpp b/shared/source/helpers/blit_properties.cpp
@@ -8,6 +8,8 @@
 #include "shared/source/helpers/blit_properties.h"
 
 #include "shared/source/command_stream/command_stream_receiver.h"
+#include "shared/source/gmm_helper/gmm.h"
+#include "shared/source/gmm_helper/resource_info.h"
 #include "shared/source/helpers/timestamp_packet.h"
 #include "shared/source/memory_manager/surface.h"
 
@@ -179,5 +181,43 @@ bool BlitProperties::isImageOperation() const {
            blitDirection == BlitterConstants::BlitDirection::imageToHostPtr ||
            blitDirection == BlitterConstants::BlitDirection::imageToImage;
 }
+bool BlitProperties::isSrc1DTiledArray() const {
+    if (srcAllocation->getDefaultGmm()) {
+        return is1DTiledArray(srcAllocation->getDefaultGmm()->gmmResourceInfo.get());
+    }
+    return false;
+}
+bool BlitProperties::isDst1DTiledArray() const {
+    if (dstAllocation->getDefaultGmm()) {
+        return is1DTiledArray(dstAllocation->getDefaultGmm()->gmmResourceInfo.get());
+    }
+    return false;
+}
+bool BlitProperties::is1DTiledArray(GmmResourceInfo *resInfo) const {
+    auto resourceType = resInfo->getResourceType();
+    auto isArray = resInfo->getArraySize() > 1;
+    auto isTiled = resInfo->getResourceFlags()->Info.Tile4 || resInfo->getResourceFlags()->Info.Tile64;
+    if (resourceType == GMM_RESOURCE_TYPE::RESOURCE_1D && isTiled && isArray) {
+        return true;
+    }
+    return false;
+}
+void BlitProperties::transform1DArrayTo2DArrayIfNeeded() {
+    if (this->isSrc1DTiledArray() || this->isDst1DTiledArray()) {
+        this->srcSize.z = this->srcSize.y;
+        this->srcSize.y = 1;
 
+        this->dstSize.z = this->dstSize.y;
+        this->dstSize.y = 1;
+
+        this->srcOffset.z = this->srcOffset.y;
+        this->srcOffset.y = 0;
+
+        this->dstOffset.z = this->dstOffset.y;
+        this->dstOffset.y = 0;
+
+        this->copySize.z = this->copySize.y;
+        this->copySize.y = 1;
+    }
+}
 } // namespace NEO
diff --git a/shared/source/helpers/blit_properties.h b/shared/source/helpers/blit_properties.h
@@ -21,6 +21,7 @@ class TagNodeBase;
 class TimestampPacketContainer;
 class GraphicsAllocation;
 class CommandStreamReceiver;
+class GmmResourceInfo;
 
 enum class BlitSyncMode {
     none = 0,
@@ -66,6 +67,12 @@ struct BlitProperties {
                                                    TimestampPacketContainer &kernelTimestamps, const CsrDependencies &depsFromEvents,
                                                    CommandStreamReceiver &gpguCsr, CommandStreamReceiver &bcsCsr);
 
+    bool isImageOperation() const;
+    bool isSrc1DTiledArray() const;
+    bool isDst1DTiledArray() const;
+    bool is1DTiledArray(GmmResourceInfo *resInfo) const;
+    void transform1DArrayTo2DArrayIfNeeded();
+
     BlitSyncProperties blitSyncProperties = {};
     CsrDependencies csrDependencies;
     TagNodeBase *multiRootDeviceEventSync = nullptr;
@@ -95,8 +102,6 @@ struct BlitProperties {
     GMM_YUV_PLANE_ENUM dstPlane = GMM_YUV_PLANE_ENUM::GMM_NO_PLANE;
     GMM_YUV_PLANE_ENUM srcPlane = GMM_YUV_PLANE_ENUM::GMM_NO_PLANE;
     bool isSystemMemoryPoolUsed = false;
-
-    bool isImageOperation() const;
 };
 
 } // namespace NEO
diff --git a/shared/test/unit_test/blit_properties/CMakeLists.txt b/shared/test/unit_test/blit_properties/CMakeLists.txt
@@ -0,0 +1,12 @@
+#
+# Copyright (C) 2020-2025 Intel Corporation
+#
+# SPDX-License-Identifier: MIT
+#
+
+target_sources(neo_shared_tests PRIVATE
+               ${CMAKE_CURRENT_SOURCE_DIR}/CMakeLists.txt
+               ${CMAKE_CURRENT_SOURCE_DIR}/blitter_properties_tests.cpp
+)
+
+add_subdirectories()
diff --git a/shared/test/unit_test/blit_properties/blitter_properties_tests.cpp b/shared/test/unit_test/blit_properties/blitter_properties_tests.cpp

Original file line number	Diff line number	Diff line change
`@@ -56,6 +56,7 @@ struct ClBlitProperties {`
`56`	`56`	`blitProperties.blitDirection = blitDirection;`
`57`	`57`	`setBlitPropertiesForImage(blitProperties, builtinOpParams);`
`58`	`58`	`}`
	`59`	`+ blitProperties.transform1DArrayTo2DArrayIfNeeded();`
`59`	`60`	`return blitProperties;`
`60`	`61`	`}`
`61`	`62`
`@@ -142,6 +143,7 @@ struct ClBlitProperties {`
`142`	`143`	`setBlitPropertiesForImage(blitProperties, builtinOpParams);`
`143`	`144`	`}`
`144`	`145`
	`146`	`+ blitProperties.transform1DArrayTo2DArrayIfNeeded();`
`145`	`147`	`return blitProperties;`
`146`	`148`	`}`
`147`	`149`