// Copyright 2020 The Dawn Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

#include <vector>

#include "dawn/common/Math.h"
#include "dawn/tests/DawnTest.h"
#include "dawn/utils/TestUtils.h"
#include "dawn/utils/TextureUtils.h"
#include "dawn/utils/WGPUHelpers.h"

class QueueTests : public DawnTest {};

// Test that GetQueue always returns the same object.
TEST_P(QueueTests, GetQueueSameObject) {
    wgpu::Queue q1 = device.GetQueue();
    wgpu::Queue q2 = device.GetQueue();
    EXPECT_EQ(q1.Get(), q2.Get());
}

DAWN_INSTANTIATE_TEST(QueueTests,
                      D3D12Backend(),
                      MetalBackend(),
                      NullBackend(),
                      OpenGLBackend(),
                      OpenGLESBackend(),
                      VulkanBackend());

class QueueWriteBufferTests : public DawnTest {};

// Test the simplest WriteBuffer setting one u32 at offset 0.
TEST_P(QueueWriteBufferTests, SmallDataAtZero) {
    wgpu::BufferDescriptor descriptor;
    descriptor.size = 4;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    uint32_t value = 0x01020304;
    queue.WriteBuffer(buffer, 0, &value, sizeof(value));

    EXPECT_BUFFER_U32_EQ(value, buffer, 0);
}

// Test an empty WriteBuffer
TEST_P(QueueWriteBufferTests, ZeroSized) {
    wgpu::BufferDescriptor descriptor;
    descriptor.size = 4;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    uint32_t initialValue = 0x42;
    queue.WriteBuffer(buffer, 0, &initialValue, sizeof(initialValue));

    queue.WriteBuffer(buffer, 0, nullptr, 0);

    // The content of the buffer isn't changed
    EXPECT_BUFFER_U32_EQ(initialValue, buffer, 0);
}

// Call WriteBuffer at offset 0 via a u32 twice. Test that data is updated accoordingly.
TEST_P(QueueWriteBufferTests, SetTwice) {
    wgpu::BufferDescriptor descriptor;
    descriptor.size = 4;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    uint32_t value = 0x01020304;
    queue.WriteBuffer(buffer, 0, &value, sizeof(value));

    EXPECT_BUFFER_U32_EQ(value, buffer, 0);

    value = 0x05060708;
    queue.WriteBuffer(buffer, 0, &value, sizeof(value));

    EXPECT_BUFFER_U32_EQ(value, buffer, 0);
}

// Test that WriteBuffer offset works.
TEST_P(QueueWriteBufferTests, SmallDataAtOffset) {
    wgpu::BufferDescriptor descriptor;
    descriptor.size = 4000;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    constexpr uint64_t kOffset = 2000;
    uint32_t value = 0x01020304;
    queue.WriteBuffer(buffer, kOffset, &value, sizeof(value));

    EXPECT_BUFFER_U32_EQ(value, buffer, kOffset);
}

// Stress test for many calls to WriteBuffer
TEST_P(QueueWriteBufferTests, ManyWriteBuffer) {
    // Note: Increasing the size of the buffer will likely cause timeout issues.
    // In D3D12, timeout detection occurs when the GPU scheduler tries but cannot preempt the task
    // executing these commands in-flight. If this takes longer than ~2s, a device reset occurs and
    // fails the test. Since GPUs may or may not complete by then, this test must be disabled OR
    // modified to be well-below the timeout limit.

    // TODO(crbug.com/dawn/228): Re-enable once the issue with Metal on 10.14.6 is fixed.
    DAWN_SUPPRESS_TEST_IF(IsMacOS() && IsIntel() && IsMetal());

    // The Vulkan Validation Layers' memory barrier validation keeps track of every range written
    // to independently which causes validation of each WriteBuffer to take increasing time, and
    // this test to take forever. Skip it when VVLs are enabled.
    DAWN_SUPPRESS_TEST_IF(IsVulkan() && IsBackendValidationEnabled());

    constexpr uint64_t kSize = 4000 * 1000;
    constexpr uint32_t kElements = 250 * 250;
    wgpu::BufferDescriptor descriptor;
    descriptor.size = kSize;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    std::vector<uint32_t> expectedData;
    for (uint32_t i = 0; i < kElements; ++i) {
        queue.WriteBuffer(buffer, i * sizeof(uint32_t), &i, sizeof(i));
        expectedData.push_back(i);
    }

    EXPECT_BUFFER_U32_RANGE_EQ(expectedData.data(), buffer, 0, kElements);
}

// Test using WriteBuffer for lots of data
TEST_P(QueueWriteBufferTests, LargeWriteBuffer) {
    constexpr uint64_t kSize = 4000 * 1000;
    constexpr uint32_t kElements = 1000 * 1000;
    wgpu::BufferDescriptor descriptor;
    descriptor.size = kSize;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    std::vector<uint32_t> expectedData;
    for (uint32_t i = 0; i < kElements; ++i) {
        expectedData.push_back(i);
    }

    queue.WriteBuffer(buffer, 0, expectedData.data(), kElements * sizeof(uint32_t));

    EXPECT_BUFFER_U32_RANGE_EQ(expectedData.data(), buffer, 0, kElements);
}

// Test using WriteBuffer for super large data block
TEST_P(QueueWriteBufferTests, SuperLargeWriteBuffer) {
    constexpr uint64_t kSize = 12000 * 1000;
    constexpr uint64_t kElements = 3000 * 1000;
    wgpu::BufferDescriptor descriptor;
    descriptor.size = kSize;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    std::vector<uint32_t> expectedData;
    for (uint32_t i = 0; i < kElements; ++i) {
        expectedData.push_back(i);
    }

    queue.WriteBuffer(buffer, 0, expectedData.data(), kElements * sizeof(uint32_t));

    EXPECT_BUFFER_U32_RANGE_EQ(expectedData.data(), buffer, 0, kElements);
}

// Test a special code path: writing when dynamic uploader already contatins some unaligned
// data, it might be necessary to use a ring buffer with properly aligned offset.
TEST_P(QueueWriteBufferTests, UnalignedDynamicUploader) {
    utils::UnalignDynamicUploader(device);

    wgpu::BufferDescriptor descriptor;
    descriptor.size = 4;
    descriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer buffer = device.CreateBuffer(&descriptor);

    uint32_t value = 0x01020304;
    queue.WriteBuffer(buffer, 0, &value, sizeof(value));

    EXPECT_BUFFER_U32_EQ(value, buffer, 0);
}

DAWN_INSTANTIATE_TEST(QueueWriteBufferTests,
                      D3D12Backend(),
                      MetalBackend(),
                      OpenGLBackend(),
                      OpenGLESBackend(),
                      VulkanBackend());

// For MinimumDataSpec bytesPerRow and rowsPerImage, compute a default from the copy extent.
constexpr uint32_t kStrideComputeDefault = 0xFFFF'FFFEul;

class QueueWriteTextureTests : public DawnTest {
  protected:
    static constexpr wgpu::TextureFormat kTextureFormat = wgpu::TextureFormat::RGBA8Unorm;

    struct TextureSpec {
        wgpu::Origin3D copyOrigin;
        wgpu::Extent3D textureSize;
        uint32_t level;
    };

    struct DataSpec {
        uint64_t size;
        uint64_t offset;
        uint32_t bytesPerRow;
        uint32_t rowsPerImage;
    };

    static DataSpec MinimumDataSpec(wgpu::Extent3D writeSize,
                                    uint32_t overrideBytesPerRow = kStrideComputeDefault,
                                    uint32_t overrideRowsPerImage = kStrideComputeDefault) {
        uint32_t bytesPerRow = writeSize.width * utils::GetTexelBlockSizeInBytes(kTextureFormat);
        if (overrideBytesPerRow != kStrideComputeDefault) {
            bytesPerRow = overrideBytesPerRow;
        }
        uint32_t rowsPerImage = writeSize.height;
        if (overrideRowsPerImage != kStrideComputeDefault) {
            rowsPerImage = overrideRowsPerImage;
        }

        uint32_t totalDataSize =
            utils::RequiredBytesInCopy(bytesPerRow, rowsPerImage, writeSize, kTextureFormat);
        return {totalDataSize, 0, bytesPerRow, rowsPerImage};
    }

    static void PackTextureData(const uint8_t* srcData,
                                uint32_t width,
                                uint32_t height,
                                uint32_t srcBytesPerRow,
                                utils::RGBA8* dstData,
                                uint32_t dstTexelPerRow,
                                uint32_t texelBlockSize) {
        for (uint64_t y = 0; y < height; ++y) {
            for (uint64_t x = 0; x < width; ++x) {
                uint64_t src = x * texelBlockSize + y * srcBytesPerRow;
                uint64_t dst = x + y * dstTexelPerRow;

                dstData[dst] = {srcData[src], srcData[src + 1], srcData[src + 2], srcData[src + 3]};
            }
        }
    }

    static void FillData(uint8_t* data, size_t count) {
        for (size_t i = 0; i < count; ++i) {
            data[i] = static_cast<uint8_t>(i % 253);
        }
    }

    void DoTest(const TextureSpec& textureSpec,
                const DataSpec& dataSpec,
                const wgpu::Extent3D& copySize) {
        // Create data of size `size` and populate it
        std::vector<uint8_t> data(dataSpec.size);
        FillData(data.data(), data.size());

        // Create a texture that is `width` x `height` with (`level` + 1) mip levels.
        wgpu::TextureDescriptor descriptor = {};
        descriptor.dimension = wgpu::TextureDimension::e2D;
        descriptor.size = textureSpec.textureSize;
        descriptor.format = kTextureFormat;
        descriptor.mipLevelCount = textureSpec.level + 1;
        descriptor.usage = wgpu::TextureUsage::CopyDst | wgpu::TextureUsage::CopySrc;
        wgpu::Texture texture = device.CreateTexture(&descriptor);

        wgpu::TextureDataLayout textureDataLayout = utils::CreateTextureDataLayout(
            dataSpec.offset, dataSpec.bytesPerRow, dataSpec.rowsPerImage);

        wgpu::ImageCopyTexture imageCopyTexture =
            utils::CreateImageCopyTexture(texture, textureSpec.level, textureSpec.copyOrigin);

        queue.WriteTexture(&imageCopyTexture, data.data(), dataSpec.size, &textureDataLayout,
                           &copySize);

        const uint32_t bytesPerTexel = utils::GetTexelBlockSizeInBytes(kTextureFormat);
        wgpu::Extent3D mipSize = {textureSpec.textureSize.width >> textureSpec.level,
                                  textureSpec.textureSize.height >> textureSpec.level,
                                  textureSpec.textureSize.depthOrArrayLayers};
        uint32_t bytesPerRow = dataSpec.bytesPerRow;
        if (bytesPerRow == wgpu::kCopyStrideUndefined) {
            bytesPerRow = mipSize.width * bytesPerTexel;
        }
        uint32_t alignedBytesPerRow = Align(bytesPerRow, bytesPerTexel);
        uint32_t appliedRowsPerImage =
            dataSpec.rowsPerImage > 0 ? dataSpec.rowsPerImage : mipSize.height;
        uint32_t bytesPerImage = bytesPerRow * appliedRowsPerImage;

        const uint32_t maxArrayLayer = textureSpec.copyOrigin.z + copySize.depthOrArrayLayers;

        uint64_t dataOffset = dataSpec.offset;
        const uint32_t texelCountLastLayer =
            (alignedBytesPerRow / bytesPerTexel) * (mipSize.height - 1) + mipSize.width;
        for (uint32_t slice = textureSpec.copyOrigin.z; slice < maxArrayLayer; ++slice) {
            // Pack the data in the specified copy region to have the same
            // format as the expected texture data.
            std::vector<utils::RGBA8> expected(texelCountLastLayer);
            PackTextureData(data.data() + dataOffset, copySize.width, copySize.height,
                            dataSpec.bytesPerRow, expected.data(), copySize.width, bytesPerTexel);

            EXPECT_TEXTURE_EQ(expected.data(), texture,
                              {textureSpec.copyOrigin.x, textureSpec.copyOrigin.y, slice},
                              {copySize.width, copySize.height}, textureSpec.level)
                << "Write to texture failed copying " << dataSpec.size << "-byte data with offset "
                << dataSpec.offset << " and bytes per row " << dataSpec.bytesPerRow << " to [("
                << textureSpec.copyOrigin.x << ", " << textureSpec.copyOrigin.y << "), ("
                << textureSpec.copyOrigin.x + copySize.width << ", "
                << textureSpec.copyOrigin.y + copySize.height << ")) region of "
                << textureSpec.textureSize.width << " x " << textureSpec.textureSize.height
                << " texture at mip level " << textureSpec.level << " layer " << slice << std::endl;

            dataOffset += bytesPerImage;
        }
    }

    void DoSimpleWriteTextureTest(uint32_t width, uint32_t height) {
        constexpr wgpu::TextureFormat kFormat = wgpu::TextureFormat::RGBA8Unorm;
        constexpr uint32_t kPixelSize = 4;

        std::vector<uint32_t> data(width * height);
        for (size_t i = 0; i < data.size(); i++) {
            data[i] = 0xFFFFFFFF;
        }

        wgpu::TextureDescriptor descriptor = {};
        descriptor.size = {width, height, 1};
        descriptor.format = kFormat;
        descriptor.usage = wgpu::TextureUsage::CopyDst | wgpu::TextureUsage::CopySrc;
        wgpu::Texture texture = device.CreateTexture(&descriptor);

        wgpu::ImageCopyTexture imageCopyTexture =
            utils::CreateImageCopyTexture(texture, 0, {0, 0, 0});
        wgpu::TextureDataLayout textureDataLayout =
            utils::CreateTextureDataLayout(0, width * kPixelSize);
        wgpu::Extent3D copyExtent = {width, height, 1};
        device.GetQueue().WriteTexture(&imageCopyTexture, data.data(), width * height * kPixelSize,
                                       &textureDataLayout, &copyExtent);

        EXPECT_TEXTURE_EQ(data.data(), texture, {0, 0}, {width, height});
    }
};

// Test writing the whole texture for varying texture sizes.
TEST_P(QueueWriteTextureTests, VaryingTextureSize) {
    for (unsigned int w : {127, 128}) {
        for (unsigned int h : {63, 64}) {
            for (unsigned int d : {1, 3, 4}) {
                TextureSpec textureSpec;
                textureSpec.textureSize = {w, h, d};
                textureSpec.copyOrigin = {0, 0, 0};
                textureSpec.level = 0;

                DoTest(textureSpec, MinimumDataSpec({w, h, d}), {w, h, d});
            }
        }
    }
}

// Test uploading a large amount of data with writeTexture.
TEST_P(QueueWriteTextureTests, LargeWriteTexture) {
    TextureSpec textureSpec;
    textureSpec.textureSize = {2048, 2048, 2};
    textureSpec.copyOrigin = {0, 0, 0};
    textureSpec.level = 0;

    DoTest(textureSpec, MinimumDataSpec(textureSpec.textureSize), textureSpec.textureSize);
}

// Test writing a pixel with an offset.
TEST_P(QueueWriteTextureTests, VaryingTextureOffset) {
    constexpr uint32_t kWidth = 259;
    constexpr uint32_t kHeight = 127;
    DataSpec pixelData = MinimumDataSpec({1, 1, 1});

    constexpr wgpu::Extent3D kCopySize = {1, 1, 1};
    constexpr wgpu::Extent3D kTextureSize = {kWidth, kHeight, 1};
    TextureSpec defaultTextureSpec;
    defaultTextureSpec.textureSize = kTextureSize;
    defaultTextureSpec.level = 0;

    for (unsigned int w : {0u, kWidth / 7, kWidth / 3, kWidth - 1}) {
        for (unsigned int h : {0u, kHeight / 7, kHeight / 3, kHeight - 1}) {
            TextureSpec textureSpec = defaultTextureSpec;
            textureSpec.copyOrigin = {w, h, 0};
            DoTest(textureSpec, pixelData, kCopySize);
        }
    }
}

// Test writing a pixel with an offset to a texture array
TEST_P(QueueWriteTextureTests, VaryingTextureArrayOffset) {
    constexpr uint32_t kWidth = 259;
    constexpr uint32_t kHeight = 127;
    constexpr uint32_t kDepth = 62;
    DataSpec pixelData = MinimumDataSpec({1, 1, 1});

    constexpr wgpu::Extent3D kCopySize = {1, 1, 1};
    constexpr wgpu::Extent3D kTextureSize = {kWidth, kHeight, kDepth};
    TextureSpec defaultTextureSpec;
    defaultTextureSpec.textureSize = kTextureSize;
    defaultTextureSpec.level = 0;

    for (unsigned int w : {0u, kWidth / 7, kWidth / 3, kWidth - 1}) {
        for (unsigned int h : {0u, kHeight / 7, kHeight / 3, kHeight - 1}) {
            for (unsigned int d : {0u, kDepth / 7, kDepth / 3, kDepth - 1}) {
                TextureSpec textureSpec = defaultTextureSpec;
                textureSpec.copyOrigin = {w, h, d};
                DoTest(textureSpec, pixelData, kCopySize);
            }
        }
    }
}

// Test writing with varying write sizes.
TEST_P(QueueWriteTextureTests, VaryingWriteSize) {
    constexpr uint32_t kWidth = 257;
    constexpr uint32_t kHeight = 127;
    for (unsigned int w : {13, 63, 128, 256}) {
        for (unsigned int h : {16, 19, 32, 63}) {
            TextureSpec textureSpec;
            textureSpec.copyOrigin = {0, 0, 0};
            textureSpec.level = 0;
            textureSpec.textureSize = {kWidth, kHeight, 1};
            DoTest(textureSpec, MinimumDataSpec({w, h, 1}), {w, h, 1});
        }
    }
}

// Test writing with varying write sizes to texture arrays.
TEST_P(QueueWriteTextureTests, VaryingArrayWriteSize) {
    constexpr uint32_t kWidth = 257;
    constexpr uint32_t kHeight = 127;
    constexpr uint32_t kDepth = 65;
    for (unsigned int w : {13, 63, 128, 256}) {
        for (unsigned int h : {16, 19, 32, 63}) {
            for (unsigned int d : {3, 6}) {
                TextureSpec textureSpec;
                textureSpec.copyOrigin = {0, 0, 0};
                textureSpec.level = 0;
                textureSpec.textureSize = {kWidth, kHeight, kDepth};
                DoTest(textureSpec, MinimumDataSpec({w, h, d}), {w, h, d});
            }
        }
    }
}

// Test writing to varying mips
TEST_P(QueueWriteTextureTests, TextureWriteToMip) {
    constexpr uint32_t kWidth = 259;
    constexpr uint32_t kHeight = 127;

    TextureSpec defaultTextureSpec;
    defaultTextureSpec.copyOrigin = {0, 0, 0};
    defaultTextureSpec.textureSize = {kWidth, kHeight, 1};

    for (unsigned int i = 1; i < 4; ++i) {
        TextureSpec textureSpec = defaultTextureSpec;
        textureSpec.level = i;
        DoTest(textureSpec, MinimumDataSpec({kWidth >> i, kHeight >> i, 1}),
               {kWidth >> i, kHeight >> i, 1});
    }
}

// Test writing with different multiples of texel block size as data offset
TEST_P(QueueWriteTextureTests, VaryingDataOffset) {
    constexpr uint32_t kWidth = 259;
    constexpr uint32_t kHeight = 127;

    TextureSpec textureSpec;
    textureSpec.copyOrigin = {0, 0, 0};
    textureSpec.textureSize = {kWidth, kHeight, 1};
    textureSpec.level = 0;

    for (uint64_t offset : {1, 2, 4, 17, 64, 128, 300}) {
        DataSpec dataSpec = MinimumDataSpec({kWidth, kHeight, 1});
        dataSpec.size += offset;
        dataSpec.offset += offset;
        DoTest(textureSpec, dataSpec, {kWidth, kHeight, 1});
    }
}

// Test writing with rowsPerImage greater than needed.
TEST_P(QueueWriteTextureTests, VaryingRowsPerImage) {
    constexpr uint32_t kWidth = 65;
    constexpr uint32_t kHeight = 31;
    constexpr uint32_t kDepth = 17;

    constexpr wgpu::Extent3D copySize = {kWidth - 1, kHeight - 1, kDepth - 1};

    for (unsigned int r : {1, 2, 3, 64, 200}) {
        TextureSpec textureSpec;
        textureSpec.copyOrigin = {1, 1, 1};
        textureSpec.textureSize = {kWidth, kHeight, kDepth};
        textureSpec.level = 0;

        DataSpec dataSpec = MinimumDataSpec(copySize, kStrideComputeDefault, copySize.height + r);
        DoTest(textureSpec, dataSpec, copySize);
    }
}

// Test with bytesPerRow greater than needed
TEST_P(QueueWriteTextureTests, VaryingBytesPerRow) {
    constexpr uint32_t kWidth = 257;
    constexpr uint32_t kHeight = 129;

    TextureSpec textureSpec;
    textureSpec.textureSize = {kWidth, kHeight, 1};
    textureSpec.copyOrigin = {1, 2, 0};
    textureSpec.level = 0;

    constexpr wgpu::Extent3D copyExtent = {17, 19, 1};

    for (unsigned int b : {1, 2, 3, 4}) {
        uint32_t bytesPerRow =
            copyExtent.width * utils::GetTexelBlockSizeInBytes(kTextureFormat) + b;
        DoTest(textureSpec, MinimumDataSpec(copyExtent, bytesPerRow), copyExtent);
    }
}

// Test that writing with bytesPerRow = 0 and bytesPerRow < bytesInACompleteRow works
// when we're copying one row only
TEST_P(QueueWriteTextureTests, BytesPerRowWithOneRowCopy) {
    constexpr uint32_t kWidth = 259;
    constexpr uint32_t kHeight = 127;

    TextureSpec textureSpec;
    textureSpec.copyOrigin = {0, 0, 0};
    textureSpec.textureSize = {kWidth, kHeight, 1};
    textureSpec.level = 0;

    {
        constexpr wgpu::Extent3D copyExtent = {5, 1, 1};
        DataSpec dataSpec = MinimumDataSpec(copyExtent);

        // bytesPerRow undefined
        dataSpec.bytesPerRow = wgpu::kCopyStrideUndefined;
        DoTest(textureSpec, dataSpec, copyExtent);
    }
}

// Test with bytesPerRow greater than needed in a write to a texture array.
TEST_P(QueueWriteTextureTests, VaryingArrayBytesPerRow) {
    constexpr uint32_t kWidth = 257;
    constexpr uint32_t kHeight = 129;
    constexpr uint32_t kLayers = 65;

    TextureSpec textureSpec;
    textureSpec.textureSize = {kWidth, kHeight, kLayers};
    textureSpec.copyOrigin = {1, 2, 3};
    textureSpec.level = 0;

    constexpr wgpu::Extent3D copyExtent = {17, 19, 21};

    // Test with bytesPerRow divisible by blockWidth
    for (unsigned int b : {1, 2, 3, 65, 300}) {
        uint32_t bytesPerRow =
            (copyExtent.width + b) * utils::GetTexelBlockSizeInBytes(kTextureFormat);
        uint32_t rowsPerImage = 23;
        DoTest(textureSpec, MinimumDataSpec(copyExtent, bytesPerRow, rowsPerImage), copyExtent);
    }

    // Test with bytesPerRow not divisible by blockWidth
    for (unsigned int b : {1, 2, 3, 19, 301}) {
        uint32_t bytesPerRow =
            copyExtent.width * utils::GetTexelBlockSizeInBytes(kTextureFormat) + b;
        uint32_t rowsPerImage = 23;
        DoTest(textureSpec, MinimumDataSpec(copyExtent, bytesPerRow, rowsPerImage), copyExtent);
    }
}

// Test valid special cases of bytesPerRow and rowsPerImage (0 or undefined).
TEST_P(QueueWriteTextureTests, StrideSpecialCases) {
    TextureSpec textureSpec;
    textureSpec.copyOrigin = {0, 0, 0};
    textureSpec.textureSize = {4, 4, 4};
    textureSpec.level = 0;

    // bytesPerRow 0
    for (const wgpu::Extent3D copyExtent :
         {wgpu::Extent3D{0, 2, 2}, {0, 0, 2}, {0, 2, 0}, {0, 0, 0}}) {
        DoTest(textureSpec, MinimumDataSpec(copyExtent, 0, 2), copyExtent);
    }

    // bytesPerRow undefined
    for (const wgpu::Extent3D copyExtent :
         {wgpu::Extent3D{2, 1, 1}, {2, 0, 1}, {2, 1, 0}, {2, 0, 0}}) {
        DoTest(textureSpec, MinimumDataSpec(copyExtent, wgpu::kCopyStrideUndefined, 2), copyExtent);
    }

    // rowsPerImage 0
    for (const wgpu::Extent3D copyExtent :
         {wgpu::Extent3D{2, 0, 2}, {2, 0, 0}, {0, 0, 2}, {0, 0, 0}}) {
        DoTest(textureSpec, MinimumDataSpec(copyExtent, 256, 0), copyExtent);
    }

    // rowsPerImage undefined
    for (const wgpu::Extent3D copyExtent : {wgpu::Extent3D{2, 2, 1}, {2, 2, 0}}) {
        DoTest(textureSpec, MinimumDataSpec(copyExtent, 256, wgpu::kCopyStrideUndefined),
               copyExtent);
    }
}

// Testing a special code path: writing when dynamic uploader already contatins some unaligned
// data, it might be necessary to use a ring buffer with properly aligned offset.
TEST_P(QueueWriteTextureTests, UnalignedDynamicUploader) {
    utils::UnalignDynamicUploader(device);

    constexpr wgpu::Extent3D size = {10, 10, 1};

    TextureSpec textureSpec;
    textureSpec.textureSize = size;
    textureSpec.copyOrigin = {0, 0, 0};
    textureSpec.level = 0;

    DoTest(textureSpec, MinimumDataSpec(size), size);
}

// This tests for a bug that occurred within the D3D12 CopyTextureSplitter, which incorrectly copied
// data when the internal offset was larger than 256, but less than 512 and the copy size was 64
// width or less with a height of 1.
TEST_P(QueueWriteTextureTests, WriteTo64x1TextureFromUnalignedDynamicUploader) {
    // First, WriteTexture with 96 pixels, or 384 bytes to create an offset in the dynamic uploader.
    DoSimpleWriteTextureTest(96, 1);

    // Now test writing to a 64x1 texture. Because a 64x1 texture's row pitch is equal to its slice
    // pitch, the texture copy offset could be calculated incorrectly inside the internal D3D12
    // TextureCopySplitter.
    DoSimpleWriteTextureTest(64, 1);
}

// This tests for a bug in the allocation of internal staging buffer, which incorrectly copied depth
// stencil data to the internal offset that is not a multiple of 4.
TEST_P(QueueWriteTextureTests, WriteStencilAspectWithSourceOffsetUnalignedTo4) {
    // Copies to a single aspect are unsupported on OpenGL.
    DAWN_SUPPRESS_TEST_IF(IsOpenGL() || IsOpenGLES());

    wgpu::TextureDescriptor textureDescriptor;
    textureDescriptor.format = wgpu::TextureFormat::Depth24PlusStencil8;
    textureDescriptor.usage = wgpu::TextureUsage::CopySrc | wgpu::TextureUsage::CopyDst;
    textureDescriptor.size = {1, 1, 1};
    wgpu::Texture dstTexture1 = device.CreateTexture(&textureDescriptor);
    wgpu::Texture dstTexture2 = device.CreateTexture(&textureDescriptor);

    wgpu::BufferDescriptor bufferDescriptor;
    bufferDescriptor.size = 8u;
    bufferDescriptor.usage = wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst;
    wgpu::Buffer outputBuffer = device.CreateBuffer(&bufferDescriptor);
    wgpu::CommandEncoder encoder = device.CreateCommandEncoder();

    constexpr wgpu::Extent3D kWriteSize = {1, 1, 1};
    constexpr uint8_t kData[] = {1, 2};
    constexpr uint32_t kBytesPerRowForWriteTexture = 1u;

    std::vector<uint8_t> expectedData(8, 0);

    // In the first call of queue.writeTexture(), Dawn will allocate a new staging buffer in its
    // internal ring buffer and write the user data into it at the offset 0.
    {
        constexpr uint32_t kDataOffset1 = 0u;
        wgpu::TextureDataLayout textureDataLayout =
            utils::CreateTextureDataLayout(kDataOffset1, kBytesPerRowForWriteTexture);
        wgpu::ImageCopyTexture imageCopyTexture = utils::CreateImageCopyTexture(
            dstTexture1, 0, {0, 0, 0}, wgpu::TextureAspect::StencilOnly);
        queue.WriteTexture(&imageCopyTexture, kData, sizeof(kData), &textureDataLayout,
                           &kWriteSize);

        constexpr uint32_t kOutputBufferOffset1 = 0u;
        wgpu::ImageCopyBuffer imageCopyBuffer = utils::CreateImageCopyBuffer(
            outputBuffer, kOutputBufferOffset1, kTextureBytesPerRowAlignment);
        encoder.CopyTextureToBuffer(&imageCopyTexture, &imageCopyBuffer, &kWriteSize);

        expectedData[kOutputBufferOffset1] = kData[kDataOffset1];
    }

    // In the second call of queue.writeTexture(), Dawn will still use the same staging buffer
    // allocated in the first call, whose first 2 bytes have been used in the first call of
    // queue.writeTexture(). Dawn should write the user data at the offset 4 bytes since the
    // destination texture aspect is stencil.
    {
        constexpr uint32_t kDataOffset2 = 1u;
        wgpu::TextureDataLayout textureDataLayout =
            utils::CreateTextureDataLayout(kDataOffset2, kBytesPerRowForWriteTexture);
        wgpu::ImageCopyTexture imageCopyTexture = utils::CreateImageCopyTexture(
            dstTexture2, 0, {0, 0, 0}, wgpu::TextureAspect::StencilOnly);
        queue.WriteTexture(&imageCopyTexture, kData, sizeof(kData), &textureDataLayout,
                           &kWriteSize);

        constexpr uint32_t kOutputBufferOffset2 = 4u;
        wgpu::ImageCopyBuffer imageCopyBuffer = utils::CreateImageCopyBuffer(
            outputBuffer, kOutputBufferOffset2, kTextureBytesPerRowAlignment);
        encoder.CopyTextureToBuffer(&imageCopyTexture, &imageCopyBuffer, &kWriteSize);

        expectedData[kOutputBufferOffset2] = kData[kDataOffset2];
    }

    wgpu::CommandBuffer commandBuffer = encoder.Finish();
    queue.Submit(1, &commandBuffer);

    EXPECT_BUFFER_U8_RANGE_EQ(expectedData.data(), outputBuffer, 0, 8);
}

// Tests calling queue.writeTexture() to a depth texture after calling queue.writeTexture() on
// another texture always works. On some D3D12 backends the buffer offset of buffer-to-texture
// copies must be a multiple of 512 when the destination texture is a depth stencil texture.
TEST_P(QueueWriteTextureTests, WriteDepthAspectAfterOtherQueueWriteTextureCalls) {
    // Copies to a single aspect are unsupported on OpenGL.
    DAWN_SUPPRESS_TEST_IF(IsOpenGL() || IsOpenGLES());

    wgpu::TextureDescriptor textureDescriptor;
    textureDescriptor.format = wgpu::TextureFormat::Depth16Unorm;
    textureDescriptor.usage = wgpu::TextureUsage::CopySrc | wgpu::TextureUsage::CopyDst;
    textureDescriptor.size = {1, 1, 1};
    wgpu::Texture depthTexture1 = device.CreateTexture(&textureDescriptor);
    wgpu::Texture depthTexture2 = device.CreateTexture(&textureDescriptor);

    constexpr uint16_t kExpectedData1 = (204 << 8) | 205;
    wgpu::ImageCopyTexture imageCopyTexture1 = utils::CreateImageCopyTexture(depthTexture1);
    wgpu::TextureDataLayout textureDataLayout =
        utils::CreateTextureDataLayout(0, sizeof(kExpectedData1));
    queue.WriteTexture(&imageCopyTexture1, &kExpectedData1, sizeof(kExpectedData1),
                       &textureDataLayout, &textureDescriptor.size);

    constexpr uint16_t kExpectedData2 = (206 << 8) | 207;
    wgpu::ImageCopyTexture imageCopyTexture2 = utils::CreateImageCopyTexture(depthTexture2);
    queue.WriteTexture(&imageCopyTexture2, &kExpectedData2, sizeof(kExpectedData2),
                       &textureDataLayout, &textureDescriptor.size);

    EXPECT_TEXTURE_EQ(&kExpectedData1, depthTexture1, {0, 0}, {1, 1}, 0,
                      wgpu::TextureAspect::DepthOnly);
    EXPECT_TEXTURE_EQ(&kExpectedData2, depthTexture2, {0, 0}, {1, 1}, 0,
                      wgpu::TextureAspect::DepthOnly);
}

// Tests calling queue.writeTexture() to the stencil aspect after calling queue.writeTexture() on
// another texture always works. On some D3D12 backends the buffer offset of buffer-to-texture
// copies must be a multiple of 512 when the destination texture is a depth stencil texture.
TEST_P(QueueWriteTextureTests, WriteStencilAspectAfterOtherQueueWriteTextureCalls) {
    // Copies to a single aspect are unsupported on OpenGL.
    DAWN_SUPPRESS_TEST_IF(IsOpenGL() || IsOpenGLES());

    wgpu::TextureDescriptor textureDescriptor;
    textureDescriptor.format = wgpu::TextureFormat::Depth24PlusStencil8;
    textureDescriptor.usage = wgpu::TextureUsage::CopySrc | wgpu::TextureUsage::CopyDst;
    textureDescriptor.size = {1, 1, 1};
    wgpu::Texture depthStencilTexture1 = device.CreateTexture(&textureDescriptor);
    wgpu::Texture depthStencilTexture2 = device.CreateTexture(&textureDescriptor);

    constexpr uint8_t kExpectedData1 = 204u;
    wgpu::ImageCopyTexture imageCopyTexture1 = utils::CreateImageCopyTexture(
        depthStencilTexture1, 0, {0, 0, 0}, wgpu::TextureAspect::StencilOnly);
    wgpu::TextureDataLayout textureDataLayout =
        utils::CreateTextureDataLayout(0, sizeof(kExpectedData1));
    queue.WriteTexture(&imageCopyTexture1, &kExpectedData1, sizeof(kExpectedData1),
                       &textureDataLayout, &textureDescriptor.size);

    constexpr uint8_t kExpectedData2 = 205;
    wgpu::ImageCopyTexture imageCopyTexture2 = utils::CreateImageCopyTexture(
        depthStencilTexture2, 0, {0, 0, 0}, wgpu::TextureAspect::StencilOnly);
    queue.WriteTexture(&imageCopyTexture2, &kExpectedData2, sizeof(kExpectedData2),
                       &textureDataLayout, &textureDescriptor.size);

    EXPECT_TEXTURE_EQ(&kExpectedData1, depthStencilTexture1, {0, 0}, {1, 1}, 0,
                      wgpu::TextureAspect::StencilOnly);
    EXPECT_TEXTURE_EQ(&kExpectedData2, depthStencilTexture2, {0, 0}, {1, 1}, 0,
                      wgpu::TextureAspect::StencilOnly);
}

DAWN_INSTANTIATE_TEST(QueueWriteTextureTests,
                      D3D12Backend(),
                      D3D12Backend({"d3d12_use_temp_buffer_in_depth_stencil_texture_and_buffer_"
                                    "copy_with_non_zero_buffer_offset"}),
                      MetalBackend(),
                      OpenGLBackend(),
                      OpenGLESBackend(),
                      VulkanBackend());