Support buffer lazy initialization before CopyTextureToBuffer

BUG=dawn:414
TEST=dawn_end2end_tests

Change-Id: I5bdd6333029170d47ea240388e7b7d80750ae5d9
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/25643
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Bryan Bernhart <bryan.bernhart@intel.com>
Reviewed-by: Austin Eng <enga@chromium.org>
Commit-Queue: Jiawei Shao <jiawei.shao@intel.com>

src/dawn_native/Buffer.cpp

@@ -15,6 +15,7 @@
 #include "dawn_native/Buffer.h"
 
 #include "common/Assert.h"
+#include "dawn_native/Commands.h"
 #include "dawn_native/Device.h"
 #include "dawn_native/DynamicUploader.h"
 #include "dawn_native/ErrorData.h"
@@ -661,5 +662,4 @@ namespace dawn_native {
     bool BufferBase::IsFullBufferRange(uint64_t offset, uint64_t size) const {
         return offset == 0 && size == GetSize();
     }
-
 }  // namespace dawn_native

src/dawn_native/Buffer.h

@@ -25,6 +25,8 @@
 
 namespace dawn_native {
 
+    struct CopyTextureToBufferCmd;
+
     enum class MapType : uint32_t;
 
     MaybeError ValidateBufferDescriptor(DeviceBase* device, const BufferDescriptor* descriptor);

src/dawn_native/CommandBuffer.cpp

@@ -15,6 +15,7 @@
 #include "dawn_native/CommandBuffer.h"
 
 #include "common/BitSetIterator.h"
+#include "dawn_native/Buffer.h"
 #include "dawn_native/CommandEncoder.h"
 #include "dawn_native/Commands.h"
 #include "dawn_native/Format.h"
@@ -145,4 +146,33 @@ namespace dawn_native {
         }
     }
 
+    // TODO(jiawei.shao@intel.com): support copying with depth stencil textures
+    bool IsFullBufferOverwrittenInTextureToBufferCopy(const CopyTextureToBufferCmd* copy) {
+        ASSERT(copy != nullptr);
+
+        if (copy->destination.offset > 0) {
+            return false;
+        }
+
+        if (copy->destination.rowsPerImage > copy->copySize.height) {
+            return false;
+        }
+
+        const TextureBase* texture = copy->source.texture.Get();
+        const uint64_t copyTextureDataSizePerRow = copy->copySize.width /
+                                                   texture->GetFormat().blockWidth *
+                                                   texture->GetFormat().blockByteSize;
+        if (copy->destination.bytesPerRow > copyTextureDataSizePerRow) {
+            return false;
+        }
+
+        const uint64_t overwrittenRangeSize =
+            copyTextureDataSizePerRow * (copy->copySize.height / texture->GetFormat().blockHeight) *
+            copy->copySize.depth;
+        if (copy->destination.buffer->GetSize() > overwrittenRangeSize) {
+            return false;
+        }
+
+        return true;
+    }
 }  // namespace dawn_native

src/dawn_native/CommandBuffer.h

@@ -25,6 +25,7 @@
 namespace dawn_native {
 
     struct BeginRenderPassCmd;
+    struct CopyTextureToBufferCmd;
     struct TextureCopy;
 
     class CommandBufferBase : public ObjectBase {
@@ -48,6 +49,8 @@ namespace dawn_native {
 
     void LazyClearRenderPassAttachments(BeginRenderPassCmd* renderPass);
 
+    bool IsFullBufferOverwrittenInTextureToBufferCopy(const CopyTextureToBufferCmd* copy);
+
 }  // namespace dawn_native
 
 #endif  // DAWNNATIVE_COMMANDBUFFER_H_

src/dawn_native/CommandValidation.h

@@ -53,7 +53,7 @@ namespace dawn_native {
                                         const Extent3D& copySize);
 
     MaybeError ValidateBufferCopyView(DeviceBase const* device,
-                                      const BufferCopyView& bufferCopyViewt);
+                                      const BufferCopyView& bufferCopyView);
     MaybeError ValidateTextureCopyView(DeviceBase const* device,
                                        const TextureCopyView& textureCopyView);
 

src/dawn_native/d3d12/BufferD3D12.cpp

@@ -17,6 +17,7 @@
 #include "common/Assert.h"
 #include "common/Constants.h"
 #include "common/Math.h"
+#include "dawn_native/CommandBuffer.h"
 #include "dawn_native/DynamicUploader.h"
 #include "dawn_native/d3d12/CommandRecordingContext.h"
 #include "dawn_native/d3d12/D3D12Error.h"
@@ -371,6 +372,24 @@ namespace dawn_native { namespace d3d12 {
         return {};
     }
 
+    MaybeError Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext,
+                                                          const CopyTextureToBufferCmd* copy) {
+        // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse
+        // instead when buffer lazy initialization is completely supported.
+        if (IsDataInitialized() ||
+            !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) {
+            return {};
+        }
+
+        if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) {
+            SetIsDataInitialized();
+        } else {
+            DAWN_TRY(InitializeToZero(commandContext));
+        }
+
+        return {};
+    }
+
     MaybeError Buffer::InitializeToZero(CommandRecordingContext* commandContext) {
         ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse));
         ASSERT(!IsDataInitialized());

src/dawn_native/d3d12/BufferD3D12.h

@@ -48,6 +48,8 @@ namespace dawn_native { namespace d3d12 {
         MaybeError EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext,
                                                       uint64_t offset,
                                                       uint64_t size);
+        MaybeError EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext,
+                                                      const CopyTextureToBufferCmd* copy);
 
       private:
         ~Buffer() override;

src/dawn_native/d3d12/CommandBufferD3D12.cpp

@@ -724,6 +724,8 @@ namespace dawn_native { namespace d3d12 {
                     Texture* texture = ToBackend(copy->source.texture.Get());
                     Buffer* buffer = ToBackend(copy->destination.buffer.Get());
 
+                    DAWN_TRY(buffer->EnsureDataInitializedAsDestination(commandContext, copy));
+
                     ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D);
                     SubresourceRange subresources = {copy->source.mipLevel, 1,
                                                      copy->source.origin.z, copy->copySize.depth};

src/dawn_native/metal/BufferMTL.h

@@ -35,6 +35,8 @@ namespace dawn_native { namespace metal {
         void EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext,
                                                 uint64_t offset,
                                                 uint64_t size);
+        void EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext,
+                                                const CopyTextureToBufferCmd* copy);
 
       private:
         using BufferBase::BufferBase;

src/dawn_native/metal/BufferMTL.mm

@@ -15,6 +15,7 @@
 #include "dawn_native/metal/BufferMTL.h"
 
 #include "common/Math.h"
+#include "dawn_native/CommandBuffer.h"
 #include "dawn_native/metal/CommandRecordingContext.h"
 #include "dawn_native/metal/DeviceMTL.h"
 
@@ -176,6 +177,22 @@ namespace dawn_native { namespace metal {
         }
     }
 
+    void Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* commandContext,
+                                                    const CopyTextureToBufferCmd* copy) {
+        // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse
+        // instead when buffer lazy initialization is completely supported.
+        if (IsDataInitialized() ||
+            !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) {
+            return;
+        }
+
+        if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) {
+            SetIsDataInitialized();
+        } else {
+            InitializeToZero(commandContext);
+        }
+    }
+
     void Buffer::InitializeToZero(CommandRecordingContext* commandContext) {
         ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse));
         ASSERT(!IsDataInitialized());

src/dawn_native/metal/CommandBufferMTL.mm

@@ -645,6 +645,8 @@ namespace dawn_native { namespace metal {
                     Texture* texture = ToBackend(src.texture.Get());
                     Buffer* buffer = ToBackend(dst.buffer.Get());
 
+                    buffer->EnsureDataInitializedAsDestination(commandContext, copy);
+
                     texture->EnsureSubresourceContentInitialized(
                         GetSubresourcesAffectedByCopy(src, copySize));
 

src/dawn_native/opengl/BufferGL.cpp

@@ -14,6 +14,7 @@
 
 #include "dawn_native/opengl/BufferGL.h"
 
+#include "dawn_native/CommandBuffer.h"
 #include "dawn_native/opengl/DeviceGL.h"
 
 namespace dawn_native { namespace opengl {
@@ -77,6 +78,21 @@ namespace dawn_native { namespace opengl {
         }
     }
 
+    void Buffer::EnsureDataInitializedAsDestination(const CopyTextureToBufferCmd* copy) {
+        // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse
+        // instead when buffer lazy initialization is completely supported.
+        if (IsDataInitialized() ||
+            !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) {
+            return;
+        }
+
+        if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) {
+            SetIsDataInitialized();
+        } else {
+            InitializeToZero();
+        }
+    }
+
     void Buffer::InitializeToZero() {
         ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse));
         ASSERT(!IsDataInitialized());

src/dawn_native/opengl/BufferGL.h

@@ -31,6 +31,7 @@ namespace dawn_native { namespace opengl {
 
         void EnsureDataInitialized();
         void EnsureDataInitializedAsDestination(uint64_t offset, uint64_t size);
+        void EnsureDataInitializedAsDestination(const CopyTextureToBufferCmd* copy);
 
       private:
         ~Buffer() override;

src/dawn_native/opengl/CommandBufferGL.cpp

@@ -611,6 +611,8 @@ namespace dawn_native { namespace opengl {
                         UNREACHABLE();
                     }
 
+                    buffer->EnsureDataInitializedAsDestination(copy);
+
                     ASSERT(texture->GetDimension() == wgpu::TextureDimension::e2D);
                     SubresourceRange subresources = {src.mipLevel, 1, src.origin.z,
                                                      copy->copySize.depth};

src/dawn_native/vulkan/BufferVk.cpp

@@ -14,6 +14,7 @@
 
 #include "dawn_native/vulkan/BufferVk.h"
 
+#include "dawn_native/CommandBuffer.h"
 #include "dawn_native/vulkan/DeviceVk.h"
 #include "dawn_native/vulkan/FencedDeleter.h"
 #include "dawn_native/vulkan/ResourceHeapVk.h"
@@ -329,6 +330,22 @@ namespace dawn_native { namespace vulkan {
         }
     }
 
+    void Buffer::EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext,
+                                                    const CopyTextureToBufferCmd* copy) {
+        // TODO(jiawei.shao@intel.com): check Toggle::LazyClearResourceOnFirstUse
+        // instead when buffer lazy initialization is completely supported.
+        if (IsDataInitialized() ||
+            !GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse)) {
+            return;
+        }
+
+        if (IsFullBufferOverwrittenInTextureToBufferCopy(copy)) {
+            SetIsDataInitialized();
+        } else {
+            InitializeToZero(recordingContext);
+        }
+    }
+
     void Buffer::InitializeToZero(CommandRecordingContext* recordingContext) {
         ASSERT(GetDevice()->IsToggleEnabled(Toggle::LazyClearBufferOnFirstUse));
         ASSERT(!IsDataInitialized());

src/dawn_native/vulkan/BufferVk.h

@@ -47,6 +47,8 @@ namespace dawn_native { namespace vulkan {
         void EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext,
                                                 uint64_t offset,
                                                 uint64_t size);
+        void EnsureDataInitializedAsDestination(CommandRecordingContext* recordingContext,
+                                                const CopyTextureToBufferCmd* copy);
 
       private:
         ~Buffer() override;

src/dawn_native/vulkan/CommandBufferVk.cpp

@@ -484,6 +484,9 @@ namespace dawn_native { namespace vulkan {
                     auto& src = copy->source;
                     auto& dst = copy->destination;
 
+                    ToBackend(dst.buffer)
+                        ->EnsureDataInitializedAsDestination(recordingContext, copy);
+
                     VkBufferImageCopy region =
                         ComputeBufferImageCopyRegion(dst, src, copy->copySize);
                     VkImageSubresourceLayers subresource = region.imageSubresource;

src/tests/end2end/BufferZeroInitTests.cpp

@@ -28,6 +28,19 @@
         }                                                                                     \
     } while (0)
 
+namespace {
+
+    struct BufferZeroInitInCopyT2BSpec {
+        wgpu::Extent3D textureSize;
+        uint64_t bufferOffset;
+        uint64_t extraBytes;
+        uint32_t bytesPerRow;
+        uint32_t rowsPerImage;
+        uint32_t lazyClearCount;
+    };
+
+}  // anonymous namespace
+
 class BufferZeroInitTest : public DawnTest {
   public:
     wgpu::Buffer CreateBuffer(uint64_t size,
@@ -60,8 +73,9 @@ class BufferZeroInitTest : public DawnTest {
         }
     }
 
-    wgpu::Texture CreateAndInitialize2DTexture(const wgpu::Extent3D& size,
-                                               wgpu::TextureFormat format) {
+    wgpu::Texture CreateAndInitializeTexture(const wgpu::Extent3D& size,
+                                             wgpu::TextureFormat format,
+                                             wgpu::Color color = {0.f, 0.f, 0.f, 0.f}) {
         wgpu::TextureDescriptor descriptor;
         descriptor.size = size;
         descriptor.format = format;
@@ -70,14 +84,73 @@ class BufferZeroInitTest : public DawnTest {
         wgpu::Texture texture = device.CreateTexture(&descriptor);
 
         wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
-        utils::ComboRenderPassDescriptor renderPassDescriptor({texture.CreateView()});
-        wgpu::RenderPassEncoder renderPass = encoder.BeginRenderPass(&renderPassDescriptor);
-        renderPass.EndPass();
+
+        for (uint32_t arrayLayer = 0; arrayLayer < size.depth; ++arrayLayer) {
+            wgpu::TextureViewDescriptor viewDescriptor;
+            viewDescriptor.format = format;
+            viewDescriptor.dimension = wgpu::TextureViewDimension::e2D;
+            viewDescriptor.baseArrayLayer = arrayLayer;
+            viewDescriptor.arrayLayerCount = 1u;
+
+            utils::ComboRenderPassDescriptor renderPassDescriptor(
+                {texture.CreateView(&viewDescriptor)});
+            renderPassDescriptor.cColorAttachments[0].clearColor = color;
+            wgpu::RenderPassEncoder renderPass = encoder.BeginRenderPass(&renderPassDescriptor);
+            renderPass.EndPass();
+        }
+
         wgpu::CommandBuffer commandBuffer = encoder.Finish();
         queue.Submit(1, &commandBuffer);
 
         return texture;
     }
+
+    void TestBufferZeroInitInCopyTextureToBuffer(const BufferZeroInitInCopyT2BSpec& spec) {
+        constexpr wgpu::TextureFormat kTextureFormat = wgpu::TextureFormat::R32Float;
+        ASSERT(utils::GetTexelBlockSizeInBytes(kTextureFormat) * spec.textureSize.width %
+                   kTextureBytesPerRowAlignment ==
+               0);
+
+        constexpr wgpu::Color kClearColor = {0.5f, 0.5f, 0.5f, 0.5f};
+        wgpu::Texture texture =
+            CreateAndInitializeTexture(spec.textureSize, kTextureFormat, kClearColor);
+
+        const wgpu::TextureCopyView textureCopyView =
+            utils::CreateTextureCopyView(texture, 0, {0, 0, 0});
+
+        const uint64_t bufferSize = spec.bufferOffset + spec.extraBytes +
+                                    utils::RequiredBytesInCopy(spec.bytesPerRow, spec.rowsPerImage,
+                                                               spec.textureSize, kTextureFormat);
+        wgpu::BufferDescriptor bufferDescriptor;
+        bufferDescriptor.size = bufferSize;
+        bufferDescriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
+        wgpu::Buffer buffer = device.CreateBuffer(&bufferDescriptor);
+        const wgpu::BufferCopyView bufferCopyView = utils::CreateBufferCopyView(
+            buffer, spec.bufferOffset, spec.bytesPerRow, spec.rowsPerImage);
+
+        wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
+        encoder.CopyTextureToBuffer(&textureCopyView, &bufferCopyView, &spec.textureSize);
+        wgpu::CommandBuffer commandBuffer = encoder.Finish();
+        EXPECT_LAZY_CLEAR(spec.lazyClearCount, queue.Submit(1, &commandBuffer));
+
+        const uint64_t expectedValueCount = bufferSize / sizeof(float);
+        std::vector<float> expectedValues(expectedValueCount, 0.f);
+
+        for (uint32_t slice = 0; slice < spec.textureSize.depth; ++slice) {
+            const uint64_t baseOffsetBytesPerSlice =
+                spec.bufferOffset + spec.bytesPerRow * spec.rowsPerImage * slice;
+            for (uint32_t y = 0; y < spec.textureSize.height; ++y) {
+                const uint64_t baseOffsetBytesPerRow =
+                    baseOffsetBytesPerSlice + spec.bytesPerRow * y;
+                const uint64_t baseOffsetFloatCountPerRow = baseOffsetBytesPerRow / sizeof(float);
+                for (uint32_t x = 0; x < spec.textureSize.width; ++x) {
+                    expectedValues[baseOffsetFloatCountPerRow + x] = 0.5f;
+                }
+            }
+        }
+
+        EXPECT_BUFFER_FLOAT_RANGE_EQ(expectedValues.data(), buffer, 0, expectedValues.size());
+    }
 };
 
 // Test that calling writeBuffer to overwrite the entire buffer doesn't need to lazily initialize
@@ -405,7 +478,7 @@ TEST_P(BufferZeroInitTest, CopyBufferToTexture) {
 
     constexpr wgpu::TextureFormat kTextureFormat = wgpu::TextureFormat::R32Uint;
 
-    wgpu::Texture texture = CreateAndInitialize2DTexture(kTextureSize, kTextureFormat);
+    wgpu::Texture texture = CreateAndInitializeTexture(kTextureSize, kTextureFormat);
     const wgpu::TextureCopyView textureCopyView =
         utils::CreateTextureCopyView(texture, 0, {0, 0, 0});
 
@@ -461,6 +534,70 @@ TEST_P(BufferZeroInitTest, CopyBufferToTexture) {
     }
 }
 
+// Test that the code path of CopyTextureToBuffer clears the destination buffer correctly when it is
+// the first use of the buffer and the texture is a 2D non-array texture.
+TEST_P(BufferZeroInitTest, Copy2DTextureToBuffer) {
+    constexpr wgpu::Extent3D kTextureSize = {64u, 8u, 1u};
+
+    // bytesPerRow == texelBlockSizeInBytes * copySize.width && bytesPerRow * copySize.height ==
+    // buffer.size
+    {
+        TestBufferZeroInitInCopyTextureToBuffer(
+            {kTextureSize, 0u, 0u, kTextureBytesPerRowAlignment, kTextureSize.height, 0u});
+    }
+
+    // bytesPerRow > texelBlockSizeInBytes * copySize.width
+    {
+        constexpr uint64_t kBytesPerRow = kTextureBytesPerRowAlignment * 2;
+        TestBufferZeroInitInCopyTextureToBuffer(
+            {kTextureSize, 0u, 0u, kBytesPerRow, kTextureSize.height, 1u});
+    }
+
+    // bufferOffset > 0
+    {
+        constexpr uint64_t kBufferOffset = 16u;
+        TestBufferZeroInitInCopyTextureToBuffer({kTextureSize, kBufferOffset, 0u,
+                                                 kTextureBytesPerRowAlignment, kTextureSize.height,
+                                                 1u});
+    }
+
+    // bytesPerRow * copySize.height < buffer.size
+    {
+        constexpr uint64_t kExtraBufferSize = 16u;
+        TestBufferZeroInitInCopyTextureToBuffer({kTextureSize, 0u, kExtraBufferSize,
+                                                 kTextureBytesPerRowAlignment, kTextureSize.height,
+                                                 1u});
+    }
+}
+
+// Test that the code path of CopyTextureToBuffer clears the destination buffer correctly when it is
+// the first use of the buffer and the texture is a 2D array texture.
+TEST_P(BufferZeroInitTest, Copy2DArrayTextureToBuffer) {
+    constexpr wgpu::Extent3D kTextureSize = {64u, 4u, 3u};
+
+    // bytesPerRow == texelBlockSizeInBytes * copySize.width && rowsPerImage == copySize.height &&
+    // bytesPerRow * (rowsPerImage * (copySize.depth - 1) + copySize.height) == buffer.size
+    {
+        TestBufferZeroInitInCopyTextureToBuffer(
+            {kTextureSize, 0u, 0u, kTextureBytesPerRowAlignment, kTextureSize.height, 0u});
+    }
+
+    // rowsPerImage > copySize.height
+    {
+        constexpr uint64_t kRowsPerImage = kTextureSize.height + 1u;
+        TestBufferZeroInitInCopyTextureToBuffer(
+            {kTextureSize, 0u, 0u, kTextureBytesPerRowAlignment, kRowsPerImage, 1u});
+    }
+
+    // bytesPerRow * rowsPerImage * copySize.depth < buffer.size
+    {
+        constexpr uint64_t kExtraBufferSize = 16u;
+        TestBufferZeroInitInCopyTextureToBuffer({kTextureSize, 0u, kExtraBufferSize,
+                                                 kTextureBytesPerRowAlignment, kTextureSize.height,
+                                                 1u});
+    }
+}
+
 DAWN_INSTANTIATE_TEST(BufferZeroInitTest,
                       D3D12Backend({"nonzero_clear_resources_on_creation_for_testing",
                                     "lazy_clear_buffer_on_first_use"}),