Delete the simple gradient texture transfer

The previous approach of transferring simple gradients directly to the texture, and then rendering complex gradients had various inefficiencies: * The texture resource had to be transitioned to and from TRANSFER_DST. * It required a pipeline barrier on the transfer stage. * The gradient texture had to be created with transfer capabilities. * Unnecessary code complexity. It also ran into platform bugs on various Android and Unreal environments. This PR just removes the texture transfer entirely and renders simple gradients at the same time we render complex ones. Rendering the simple ramps requires us to upload more data to the GPU, but the tradeoffs should be worth it. Diffs= 792049458c Delete the simple gradient texture transfer (#8844) Co-authored-by: Chris Dalton <[email protected]>
rive-app · Jan 10, 2025 · c0cd593 · c0cd593
1 parent 9d2535c
commit c0cd593
Show file tree

Hide file tree

Showing 22 changed files with 172 additions and 439 deletions.
diff --git a/.rive_head b/.rive_head
@@ -1 +1 @@
-6f306199cad05604ab7c0293b5268e00ca1abc84
+792049458cf90c0fb3f06a34080db079281bd2e7
diff --git a/renderer/include/rive/renderer/d3d/render_context_d3d_impl.hpp b/renderer/include/rive/renderer/d3d/render_context_d3d_impl.hpp
@@ -131,8 +131,6 @@ class RenderContextD3DImpl : public RenderContextHelperImpl
         gpu::StorageBufferStructure) override;
     std::unique_ptr<BufferRing> makeVertexBufferRing(
         size_t capacityInBytes) override;
-    std::unique_ptr<BufferRing> makeTextureTransferBufferRing(
-        size_t capacityInBytes) override;
 
     void resizeGradientTexture(uint32_t width, uint32_t height) override;
     void resizeTessellationTexture(uint32_t width, uint32_t height) override;

diff --git a/renderer/include/rive/renderer/gl/render_context_gl_impl.hpp b/renderer/include/rive/renderer/gl/render_context_gl_impl.hpp
@@ -207,8 +207,6 @@ class RenderContextGLImpl : public RenderContextHelperImpl
         gpu::StorageBufferStructure) override;
     std::unique_ptr<BufferRing> makeVertexBufferRing(
         size_t capacityInBytes) override;
-    std::unique_ptr<BufferRing> makeTextureTransferBufferRing(
-        size_t capacityInBytes) override;
 
     void resizeGradientTexture(uint32_t width, uint32_t height) override;
     void resizeTessellationTexture(uint32_t width, uint32_t height) override;

diff --git a/renderer/include/rive/renderer/gpu.hpp b/renderer/include/rive/renderer/gpu.hpp
@@ -324,11 +324,10 @@ struct ColorRampLocation
 // and images reference an additional Gradient* and Texture* respectively.
 union SimplePaintValue
 {
-    ColorInt color = 0xff000000; // PaintType::solidColor
-    ColorRampLocation
-        colorRampLocation; // Paintype::linearGradient, Paintype::radialGradient
-    float imageOpacity;    // PaintType::image
-    uint32_t outerClipID;  // Paintype::clipUpdate
+    ColorInt color = 0xff000000;         // PaintType::solidColor
+    ColorRampLocation colorRampLocation; // Paintype::linear/radialGradient
+    float imageOpacity;                  // PaintType::image
+    uint32_t outerClipID;                // Paintype::clipUpdate
 };
 static_assert(sizeof(SimplePaintValue) == 4);
 
@@ -834,10 +833,10 @@ struct TwoTexelRamp
 {
     void set(const ColorInt colors[2])
     {
-        UnpackColorToRGBA8(colors[0], colorData);
-        UnpackColorToRGBA8(colors[1], colorData + 4);
+        color0 = colors[0];
+        color1 = colors[1];
     }
-    uint8_t colorData[8];
+    ColorInt color0, color1;
 };
 static_assert(sizeof(TwoTexelRamp) == 8 * sizeof(uint8_t));
 
@@ -859,7 +858,7 @@ class CommandBufferCompletionFence : public RefCnt<CommandBufferCompletionFence>
 //
 //  1. Render the complex gradients from the gradSpanBuffer to the gradient
 //  texture
-//     (complexGradSpanCount, firstComplexGradSpan, complexGradRowsTop,
+//     (gradSpanCount, firstComplexGradSpan, complexGradRowsTop,
 //     complexGradRowsHeight).
 //
 //  2. Transfer the simple gradient texels from the simpleColorRampsBuffer to
@@ -909,15 +908,11 @@ struct FlushDescriptor
     size_t firstPaintAux = 0;
     uint32_t contourCount = 0;
     size_t firstContour = 0;
-    uint32_t complexGradSpanCount = 0;
-    size_t firstComplexGradSpan = 0;
+    uint32_t gradSpanCount = 0;
+    size_t firstGradSpan = 0;
     uint32_t tessVertexSpanCount = 0;
     size_t firstTessVertexSpan = 0;
-    uint32_t simpleGradTexelsWidth = 0;
-    uint32_t simpleGradTexelsHeight = 0;
-    size_t simpleGradDataOffsetInBytes = 0;
-    uint32_t complexGradRowsTop = 0;
-    uint32_t complexGradRowsHeight = 0;
+    uint32_t gradDataHeight = 0;
     uint32_t tessDataHeight = 0;
     // Override path fill rules with "clockwise".
     bool clockwiseFillOverride = false;

diff --git a/renderer/include/rive/renderer/metal/render_context_metal_impl.h b/renderer/include/rive/renderer/metal/render_context_metal_impl.h
@@ -164,8 +164,6 @@ class RenderContextMetalImpl : public RenderContextHelperImpl
         size_t capacityInBytes, StorageBufferStructure) override;
     std::unique_ptr<BufferRing> makeVertexBufferRing(
         size_t capacityInBytes) override;
-    std::unique_ptr<BufferRing> makeTextureTransferBufferRing(
-        size_t capacityInBytes) override;
 
 private:
     // Renders paths to the main render target.

diff --git a/renderer/include/rive/renderer/render_context.hpp b/renderer/include/rive/renderer/render_context.hpp
@@ -281,7 +281,7 @@ class RenderContext : public RiveRenderFactory
     // LogicalFlush::LayoutCounters.
     struct ResourceAllocationCounts
     {
-        constexpr static int NUM_ELEMENTS = 13;
+        constexpr static int NUM_ELEMENTS = 12;
         using VecType = simd::gvec<size_t, NUM_ELEMENTS>;
 
         RIVE_ALWAYS_INLINE VecType toVec() const
@@ -308,8 +308,7 @@ class RenderContext : public RiveRenderFactory
         size_t paintBufferCount = 0;
         size_t paintAuxBufferCount = 0;
         size_t contourBufferCount = 0;
-        size_t simpleGradientBufferCount = 0;
-        size_t complexGradSpanBufferCount = 0;
+        size_t gradSpanBufferCount = 0;
         size_t tessSpanBufferCount = 0;
         size_t triangleVertexBufferCount = 0;
         size_t gradTextureHeight = 0;
@@ -366,9 +365,6 @@ class RenderContext : public RiveRenderFactory
     WriteOnlyMappedMemory<gpu::PaintData> m_paintData;
     WriteOnlyMappedMemory<gpu::PaintAuxData> m_paintAuxData;
     WriteOnlyMappedMemory<gpu::ContourData> m_contourData;
-    // Simple gradients get written by the CPU.
-    WriteOnlyMappedMemory<gpu::TwoTexelRamp> m_simpleColorRampsData;
-    // Complex gradients get rendered by the GPU.
     WriteOnlyMappedMemory<gpu::GradientSpan> m_gradSpanData;
     WriteOnlyMappedMemory<gpu::TessVertexSpan> m_tessSpanData;
     WriteOnlyMappedMemory<gpu::TriangleVertex> m_triangleVertexData;
@@ -527,8 +523,7 @@ class RenderContext : public RiveRenderFactory
             uint32_t paintPaddingCount = 0;
             uint32_t paintAuxPaddingCount = 0;
             uint32_t contourPaddingCount = 0;
-            uint32_t simpleGradCount = 0;
-            uint32_t complexGradSpanCount = 0;
+            uint32_t gradSpanCount = 0;
             uint32_t gradSpanPaddingCount = 0;
             uint32_t maxGradTextureHeight = 0;
             uint32_t maxTessTextureHeight = 0;
@@ -703,16 +698,19 @@ class RenderContext : public RiveRenderFactory
         // implemented with 2 texels.
         std::unordered_map<uint64_t, uint32_t>
             m_simpleGradients; // [color0, color1] -> texelsIdx.
-        std::vector<gpu::TwoTexelRamp> m_pendingSimpleGradientWrites;
+        std::vector<gpu::TwoTexelRamp> m_pendingSimpleGradDraws;
 
         // Complex gradients have stop(s) between t=0 and t=1. In theory they
         // should be scaled to a ramp where every stop lands exactly on a pixel
         // center, but for now we just always scale them to the entire gradient
         // texture width.
         std::unordered_map<GradientContentKey, uint16_t, DeepHashGradient>
             m_complexGradients; // [colors[0..n], stops[0..n]] -> rowIdx
-        std::vector<const Gradient*> m_pendingComplexColorRampDraws;
-        size_t m_pendingComplexGradSpanCount;
+        std::vector<const Gradient*> m_pendingComplexGradDraws;
+
+        // Simple and complex gradients both get uploaded to the GPU as sets of
+        // "GradientSpan" instances.
+        size_t m_pendingGradSpanCount;
 
         std::vector<ClipInfo> m_clips;
 
@@ -732,10 +730,10 @@ class RenderContext : public RiveRenderFactory
         uint32_t m_outerCubicTessVertexIdx;
         uint32_t m_midpointFanTessVertexIdx;
 
-        gpu::FlushDescriptor m_flushDesc;
-        // Not determined until writeResources().
         gpu::GradTextureLayout m_gradTextureLayout;
 
+        gpu::FlushDescriptor m_flushDesc;
+
         BlockAllocatedLinkedList<DrawBatch> m_drawList;
         gpu::ShaderFeatures m_combinedShaderFeatures;
 

diff --git a/renderer/include/rive/renderer/render_context_helper_impl.hpp b/renderer/include/rive/renderer/render_context_helper_impl.hpp
@@ -26,7 +26,6 @@ class RenderContextHelperImpl : public RenderContextImpl
                               gpu::StorageBufferStructure) override;
     void resizeContourBuffer(size_t sizeInBytes,
                              gpu::StorageBufferStructure) override;
-    void resizeSimpleColorRampsBuffer(size_t sizeInBytes) override;
     void resizeGradSpanBuffer(size_t sizeInBytes) override;
     void resizeTessVertexSpanBuffer(size_t sizeInBytes) override;
     void resizeTriangleVertexBuffer(size_t sizeInBytes) override;
@@ -37,7 +36,6 @@ class RenderContextHelperImpl : public RenderContextImpl
     void* mapPaintBuffer(size_t mapSizeInBytes) override;
     void* mapPaintAuxBuffer(size_t mapSizeInBytes) override;
     void* mapContourBuffer(size_t mapSizeInBytes) override;
-    void* mapSimpleColorRampsBuffer(size_t mapSizeInBytes) override;
     void* mapGradSpanBuffer(size_t mapSizeInBytes) override;
     void* mapTessVertexSpanBuffer(size_t mapSizeInBytes) override;
     void* mapTriangleVertexBuffer(size_t mapSizeInBytes) override;
@@ -48,7 +46,6 @@ class RenderContextHelperImpl : public RenderContextImpl
     void unmapPaintBuffer() override;
     void unmapPaintAuxBuffer() override;
     void unmapContourBuffer() override;
-    void unmapSimpleColorRampsBuffer() override;
     void unmapGradSpanBuffer() override;
     void unmapTessVertexSpanBuffer() override;
     void unmapTriangleVertexBuffer() override;
@@ -69,10 +66,6 @@ class RenderContextHelperImpl : public RenderContextImpl
     BufferRing* paintBufferRing() { return m_paintBuffer.get(); }
     BufferRing* paintAuxBufferRing() { return m_paintAuxBuffer.get(); }
     BufferRing* contourBufferRing() { return m_contourBuffer.get(); }
-    BufferRing* simpleColorRampsBufferRing()
-    {
-        return m_simpleColorRampsBuffer.get();
-    }
     BufferRing* gradSpanBufferRing() { return m_gradSpanBuffer.get(); }
     BufferRing* tessSpanBufferRing() { return m_tessSpanBuffer.get(); }
     BufferRing* triangleBufferRing() { return m_triangleBuffer.get(); }
@@ -89,8 +82,6 @@ class RenderContextHelperImpl : public RenderContextImpl
         gpu::StorageBufferStructure) = 0;
     virtual std::unique_ptr<BufferRing> makeVertexBufferRing(
         size_t capacityInBytes) = 0;
-    virtual std::unique_ptr<BufferRing> makeTextureTransferBufferRing(
-        size_t capacityInBytes) = 0;
 
 private:
     std::unique_ptr<BufferRing> m_flushUniformBuffer;
@@ -99,7 +90,6 @@ class RenderContextHelperImpl : public RenderContextImpl
     std::unique_ptr<BufferRing> m_paintBuffer;
     std::unique_ptr<BufferRing> m_paintAuxBuffer;
     std::unique_ptr<BufferRing> m_contourBuffer;
-    std::unique_ptr<BufferRing> m_simpleColorRampsBuffer;
     std::unique_ptr<BufferRing> m_gradSpanBuffer;
     std::unique_ptr<BufferRing> m_tessSpanBuffer;
     std::unique_ptr<BufferRing> m_triangleBuffer;

diff --git a/renderer/include/rive/renderer/render_context_impl.hpp b/renderer/include/rive/renderer/render_context_impl.hpp
@@ -50,7 +50,6 @@ class RenderContextImpl
                                       gpu::StorageBufferStructure) = 0;
     virtual void resizeContourBuffer(size_t sizeInBytes,
                                      gpu::StorageBufferStructure) = 0;
-    virtual void resizeSimpleColorRampsBuffer(size_t sizeInBytes) = 0;
     virtual void resizeGradSpanBuffer(size_t sizeInBytes) = 0;
     virtual void resizeTessVertexSpanBuffer(size_t sizeInBytes) = 0;
     virtual void resizeTriangleVertexBuffer(size_t sizeInBytes) = 0;
@@ -68,7 +67,6 @@ class RenderContextImpl
     virtual void* mapPaintBuffer(size_t mapSizeInBytes) = 0;
     virtual void* mapPaintAuxBuffer(size_t mapSizeInBytes) = 0;
     virtual void* mapContourBuffer(size_t mapSizeInBytes) = 0;
-    virtual void* mapSimpleColorRampsBuffer(size_t mapSizeInBytes) = 0;
     virtual void* mapGradSpanBuffer(size_t mapSizeInBytes) = 0;
     virtual void* mapTessVertexSpanBuffer(size_t mapSizeInBytes) = 0;
     virtual void* mapTriangleVertexBuffer(size_t mapSizeInBytes) = 0;
@@ -80,7 +78,6 @@ class RenderContextImpl
     virtual void unmapPaintBuffer() = 0;
     virtual void unmapPaintAuxBuffer() = 0;
     virtual void unmapContourBuffer() = 0;
-    virtual void unmapSimpleColorRampsBuffer() = 0;
     virtual void unmapGradSpanBuffer() = 0;
     virtual void unmapTessVertexSpanBuffer() = 0;
     virtual void unmapTriangleVertexBuffer() = 0;

diff --git a/renderer/include/rive/renderer/vulkan/render_context_vulkan_impl.hpp b/renderer/include/rive/renderer/vulkan/render_context_vulkan_impl.hpp
@@ -173,7 +173,6 @@ class RenderContextVulkanImpl : public RenderContextImpl
     IMPLEMENT_PLS_STRUCTURED_BUFFER(PaintBuffer, m_paintBufferRing)
     IMPLEMENT_PLS_STRUCTURED_BUFFER(PaintAuxBuffer, m_paintAuxBufferRing)
     IMPLEMENT_PLS_STRUCTURED_BUFFER(ContourBuffer, m_contourBufferRing)
-    IMPLEMENT_PLS_BUFFER(SimpleColorRampsBuffer, m_simpleColorRampsBufferRing)
     IMPLEMENT_PLS_BUFFER(GradSpanBuffer, m_gradSpanBufferRing)
     IMPLEMENT_PLS_BUFFER(TessVertexSpanBuffer, m_tessSpanBufferRing)
     IMPLEMENT_PLS_BUFFER(TriangleVertexBuffer, m_triangleBufferRing)
@@ -230,7 +229,6 @@ class RenderContextVulkanImpl : public RenderContextImpl
     vkutil::BufferRing m_paintBufferRing;
     vkutil::BufferRing m_paintAuxBufferRing;
     vkutil::BufferRing m_contourBufferRing;
-    vkutil::BufferRing m_simpleColorRampsBufferRing;
     vkutil::BufferRing m_gradSpanBufferRing;
     vkutil::BufferRing m_tessSpanBufferRing;
     vkutil::BufferRing m_triangleBufferRing;

diff --git a/renderer/include/rive/renderer/webgpu/render_context_webgpu_impl.hpp b/renderer/include/rive/renderer/webgpu/render_context_webgpu_impl.hpp
@@ -148,8 +148,6 @@ class RenderContextWebGPUImpl : public RenderContextHelperImpl
         gpu::StorageBufferStructure) override;
     std::unique_ptr<BufferRing> makeVertexBufferRing(
         size_t capacityInBytes) override;
-    std::unique_ptr<BufferRing> makeTextureTransferBufferRing(
-        size_t capacityInBytes) override;
 
     void resizeGradientTexture(uint32_t width, uint32_t height) override;
     void resizeTessellationTexture(uint32_t width, uint32_t height) override;

diff --git a/renderer/premake5_pls_renderer.lua b/renderer/premake5_pls_renderer.lua
@@ -126,29 +126,9 @@ function execute_and_check(cmd)
     end
 end
 
--- Wipe out the shader directory if this make command differs from the one that built it.
-local existing_makecommand = nil
-local makecommand_filename = pls_shaders_absolute_dir .. '/./.makecommand'
-local makecommand_file = io.open(makecommand_filename)
-if makecommand_file then
-    existing_makecommand = makecommand_file:read('*all')
-    -- Trim whitespace
-    existing_makecommand = string.gsub(existing_makecommand, '^%s*(.-)%s*$', '%1')
-    makecommand_file:close()
-end
-if not existing_makecommand or existing_makecommand ~= makecommand then
-    if existing_makecommand then
-        print('"make" command for PLS shaders differs from before: cleaning...')
-    end
-    execute_and_check('rm -fr ' .. pls_shaders_absolute_dir)
-end
-
 -- Build shaders.
 execute_and_check(makecommand)
 
--- Save the make command for incremental shader builds.
-execute_and_check('echo \'' .. makecommand .. '\' > ' .. makecommand_filename)
-
 newoption({
     trigger = 'nop-obj-c',
     description = 'include Metal classes, but as no-ops (for compilers that don\'t support Obj-C)',

diff --git a/renderer/src/d3d/render_context_d3d_impl.cpp b/renderer/src/d3d/render_context_d3d_impl.cpp
@@ -809,14 +809,6 @@ std::unique_ptr<BufferRing> RenderContextD3DImpl::makeVertexBufferRing(
                : nullptr;
 }
 
-std::unique_ptr<BufferRing> RenderContextD3DImpl::makeTextureTransferBufferRing(
-    size_t capacityInBytes)
-{
-    // It appears impossible to update a D3D texture from a GPU buffer; store
-    // this data on the heap and upload it to the texture at flush time.
-    return std::make_unique<HeapBufferRing>(capacityInBytes);
-}
-
 RenderTargetD3D::RenderTargetD3D(RenderContextD3DImpl* renderContextImpl,
                                  uint32_t width,
                                  uint32_t height) :
@@ -1420,7 +1412,7 @@ void RenderContextD3DImpl::flush(const FlushDescriptor& desc)
     }
 
     // Render the complex color ramps to the gradient texture.
-    if (desc.complexGradSpanCount > 0)
+    if (desc.gradSpanCount > 0)
     {
         ID3D11Buffer* gradSpanBuffer =
             flush_buffer(m_gpuContext.Get(), gradSpanBufferRing());
@@ -1437,13 +1429,12 @@ void RenderContextD3DImpl::flush(const FlushDescriptor& desc)
 
         m_gpuContext->VSSetShader(m_colorRampVertexShader.Get(), NULL, 0);
 
-        D3D11_VIEWPORT viewport = {
-            0,
-            static_cast<float>(desc.complexGradRowsTop),
-            static_cast<float>(kGradTextureWidth),
-            static_cast<float>(desc.complexGradRowsHeight),
-            0,
-            1};
+        D3D11_VIEWPORT viewport = {0,
+                                   0,
+                                   static_cast<float>(kGradTextureWidth),
+                                   static_cast<float>(desc.gradDataHeight),
+                                   0,
+                                   1};
         m_gpuContext->RSSetViewports(1, &viewport);
 
         // Unbind the gradient texture before rendering it.
@@ -1460,31 +1451,9 @@ void RenderContextD3DImpl::flush(const FlushDescriptor& desc)
 
         m_gpuContext->DrawInstanced(
             4,
-            desc.complexGradSpanCount,
-            0,
-            math::lossless_numeric_cast<UINT>(desc.firstComplexGradSpan));
-    }
-
-    // Copy the simple color ramps to the gradient texture.
-    if (desc.simpleGradTexelsHeight > 0)
-    {
-        assert(desc.simpleGradTexelsHeight * desc.simpleGradTexelsWidth * 4 <=
-               simpleColorRampsBufferRing()->capacityInBytes());
-        D3D11_BOX box;
-        box.left = 0;
-        box.right = desc.simpleGradTexelsWidth;
-        box.top = 0;
-        box.bottom = desc.simpleGradTexelsHeight;
-        box.front = 0;
-        box.back = 1;
-        m_gpuContext->UpdateSubresource(
-            m_gradTexture.Get(),
+            desc.gradSpanCount,
             0,
-            &box,
-            heap_buffer_contents(simpleColorRampsBufferRing()) +
-                desc.simpleGradDataOffsetInBytes,
-            kGradTextureWidth * 4,
-            0);
+            math::lossless_numeric_cast<UINT>(desc.firstGradSpan));
     }
 
     // Tessellate all curves into vertices in the tessellation texture.
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		6f306199cad05604ab7c0293b5268e00ca1abc84
		792049458cf90c0fb3f06a34080db079281bd2e7