Calculate Blocklinear Texture Aligned Size Correctly

The size of blocklinear textures did not consider alignment to Block/ROB boundaries before, it is aligned to them now. Incorrect sizes led to textures not being aliased correctly due to different size calculations for GraphicBufferProducer surfaces and Maxwell3D color RTs.

app/src/main/cpp/skyline/gpu/interconnect/graphics_context.h

@@ -337,7 +337,8 @@ namespace skyline::gpu::interconnect {
                 return &*renderTarget.view;
 
             if (renderTarget.guest.mappings.empty()) {
-                auto size{std::max<u64>(renderTarget.guest.layerStride * (renderTarget.guest.layerCount - renderTarget.guest.baseArrayLayer), renderTarget.guest.format->GetSize(renderTarget.guest.dimensions))};
+                size_t layerStride{renderTarget.guest.GetLayerSize()};
+                size_t size{layerStride * (renderTarget.guest.layerCount - renderTarget.guest.baseArrayLayer)};
                 auto mappings{channelCtx.asCtx->gmmu.TranslateRange(renderTarget.iova, size)};
                 renderTarget.guest.mappings.assign(mappings.begin(), mappings.end());
             }

app/src/main/cpp/skyline/gpu/texture/layout.h

@@ -5,25 +5,38 @@
 
 #include "texture.h"
 
-namespace skyline::gpu {
-    /**
-     * @brief Copies the contents of a blocklinear guest texture to a linear output buffer
-     */
-    void CopyBlockLinearToLinear(GuestTexture &guest, u8 *guestInput, u8 *linearOutput) {
+namespace skyline::gpu::texture {
     // Reference on Block-linear tiling: https://gist.github.com/PixelyIon/d9c35050af0ef5690566ca9f0965bc32
     constexpr u8 SectorWidth{16}; // The width of a sector in bytes
     constexpr u8 SectorHeight{2}; // The height of a sector in lines
     constexpr u8 GobWidth{64}; // The width of a GOB in bytes
     constexpr u8 GobHeight{8}; // The height of a GOB in lines
 
+    size_t GetBlockLinearLayerSize(const GuestTexture &guest) {
         u32 blockHeight{guest.tileConfig.blockHeight}; //!< The height of the blocks in GOBs
         u32 robHeight{GobHeight * blockHeight}; //!< The height of a single ROB (Row of Blocks) in lines
         u32 surfaceHeight{guest.dimensions.height / guest.format->blockHeight}; //!< The height of the surface in lines
         u32 surfaceHeightRobs{util::AlignUp(surfaceHeight, robHeight) / robHeight}; //!< The height of the surface in ROBs (Row Of Blocks)
+
         u32 robWidthBytes{util::AlignUp((guest.dimensions.width / guest.format->blockWidth) * guest.format->bpb, GobWidth)}; //!< The width of a ROB in bytes
         u32 robWidthBlocks{robWidthBytes / GobWidth}; //!< The width of a ROB in blocks (and GOBs because block width == 1 on the Tegra X1)
         u32 robBytes{robWidthBytes * robHeight}; //!< The size of a ROB in bytes
-        u32 gobYOffset{robWidthBytes * GobHeight}; //!< The offset of the next Y-axis GOB from the current one in linear space
+
+        return robBytes * surfaceHeightRobs;
+    }
+
+    /**
+     * @brief Copies the contents of a blocklinear guest texture to a linear output buffer
+     */
+    void CopyBlockLinearToLinear(const GuestTexture &guest, u8 *guestInput, u8 *linearOutput) {
+        u32 blockHeight{guest.tileConfig.blockHeight};
+        u32 robHeight{GobHeight * blockHeight};
+        u32 surfaceHeight{guest.dimensions.height / guest.format->blockHeight};
+        u32 surfaceHeightRobs{util::AlignUp(surfaceHeight, robHeight) / robHeight};
+        u32 robWidthBytes{util::AlignUp((guest.dimensions.width / guest.format->blockWidth) * guest.format->bpb, GobWidth)};
+        u32 robWidthBlocks{robWidthBytes / GobWidth};
+        u32 robBytes{robWidthBytes * robHeight};
+        u32 gobYOffset{robWidthBytes * GobHeight};
 
         auto inputSector{guestInput};
         auto outputRob{linearOutput};
@@ -55,53 +68,47 @@ namespace skyline::gpu {
     /**
      * @brief Copies the contents of a blocklinear guest texture to a linear output buffer
      */
-    void CopyLinearToBlockLinear(GuestTexture &guest, u8 *linearInput, u8 *guestOutput) {
-        // Reference on Block-linear tiling: https://gist.github.com/PixelyIon/d9c35050af0ef5690566ca9f0965bc32
-        constexpr u8 SectorWidth{16}; // The width of a sector in bytes
-        constexpr u8 SectorHeight{2}; // The height of a sector in lines
-        constexpr u8 GobWidth{64}; // The width of a GOB in bytes
-        constexpr u8 GobHeight{8}; // The height of a GOB in lines
-
-        u32 blockHeight{guest.tileConfig.blockHeight}; //!< The height of the blocks in GOBs
-        u32 robHeight{GobHeight * blockHeight}; //!< The height of a single ROB (Row of Blocks) in lines
-        u32 surfaceHeight{guest.dimensions.height / guest.format->blockHeight}; //!< The height of the surface in lines
-        u32 surfaceHeightRobs{util::AlignUp(surfaceHeight, robHeight) / robHeight}; //!< The height of the surface in ROBs (Row Of Blocks)
-        u32 robWidthBytes{util::AlignUp((guest.dimensions.width / guest.format->blockWidth) * guest.format->bpb, GobWidth)}; //!< The width of a ROB in bytes
-        u32 robWidthBlocks{robWidthBytes / GobWidth}; //!< The width of a ROB in blocks (and GOBs because block width == 1 on the Tegra X1)
-        u32 robBytes{robWidthBytes * robHeight}; //!< The size of a ROB in bytes
-        u32 gobYOffset{robWidthBytes * GobHeight}; //!< The offset of the next Y-axis GOB from the current one in linear space
+    void CopyLinearToBlockLinear(const GuestTexture &guest, u8 *linearInput, u8 *guestOutput) {
+        u32 blockHeight{guest.tileConfig.blockHeight};
+        u32 robHeight{GobHeight * blockHeight};
+        u32 surfaceHeight{guest.dimensions.height / guest.format->blockHeight};
+        u32 surfaceHeightRobs{util::AlignUp(surfaceHeight, robHeight) / robHeight};
+        u32 robWidthBytes{util::AlignUp((guest.dimensions.width / guest.format->blockWidth) * guest.format->bpb, GobWidth)};
+        u32 robWidthBlocks{robWidthBytes / GobWidth};
+        u32 robBytes{robWidthBytes * robHeight};
+        u32 gobYOffset{robWidthBytes * GobHeight};
 
         auto outputSector{guestOutput};
         auto inputRob{linearInput};
 
-        for (u32 rob{}, y{}, paddingY{}; rob < surfaceHeightRobs; rob++) { // Every Surface contains `surfaceHeightRobs` ROBs
-            auto outputBlock{inputRob}; // We iterate through a block independently of the ROB
-            for (u32 block{}; block < robWidthBlocks; block++) { // Every ROB contains `surfaceWidthBlocks` Blocks
-                auto inputGob{outputBlock}; // We iterate through a GOB independently of the block
-                for (u32 gobY{}; gobY < blockHeight; gobY++) { // Every Block contains `blockHeight` Y-axis GOBs
-                    for (u32 index{}; index < SectorWidth * SectorHeight; index++) { // Every Y-axis GOB contains `sectorWidth * sectorHeight` sectors
-                        u32 xT{((index << 3) & 0b10000) | ((index << 1) & 0b100000)}; // Morton-Swizzle on the X-axis
-                        u32 yT{((index >> 1) & 0b110) | (index & 0b1)}; // Morton-Swizzle on the Y-axis
+        for (u32 rob{}, y{}, paddingY{}; rob < surfaceHeightRobs; rob++) {
+            auto outputBlock{inputRob};
+            for (u32 block{}; block < robWidthBlocks; block++) {
+                auto inputGob{outputBlock};
+                for (u32 gobY{}; gobY < blockHeight; gobY++) {
+                    for (u32 index{}; index < SectorWidth * SectorHeight; index++) {
+                        u32 xT{((index << 3) & 0b10000) | ((index << 1) & 0b100000)};
+                        u32 yT{((index >> 1) & 0b110) | (index & 0b1)};
                         std::memcpy(outputSector, inputGob + (yT * robWidthBytes) + xT, SectorWidth);
-                        outputSector += SectorWidth; // `sectorWidth` bytes are of sequential image data
+                        outputSector += SectorWidth;
                     }
-                    inputGob += gobYOffset; // Increment the output GOB to the next Y-axis GOB
+                    inputGob += gobYOffset;
                 }
-                outputSector += paddingY; // Increment the input sector to the next sector
-                outputBlock += GobWidth; // Increment the output block to the next block (As Block Width = 1 GOB Width)
+                outputSector += paddingY;
+                outputBlock += GobWidth;
             }
-            inputRob += robBytes; // Increment the output block to the next ROB
+            inputRob += robBytes;
 
-            y += robHeight; // Increment the Y position to the next ROB
-            blockHeight = static_cast<u8>(std::min(static_cast<u32>(blockHeight), (surfaceHeight - y) / GobHeight)); // Calculate the amount of Y GOBs which aren't padding
-            paddingY = (guest.tileConfig.blockHeight - blockHeight) * (SectorWidth * SectorWidth * SectorHeight); // Calculate the amount of padding between contiguous sectors
+            y += robHeight;
+            blockHeight = static_cast<u8>(std::min(static_cast<u32>(blockHeight), (surfaceHeight - y) / GobHeight));
+            paddingY = (guest.tileConfig.blockHeight - blockHeight) * (SectorWidth * SectorWidth * SectorHeight);
         }
     }
 
     /**
      * @brief Copies the contents of a pitch-linear guest texture to a linear output buffer
      */
-    void CopyPitchLinearToLinear(GuestTexture &guest, u8 *guestInput, u8 *linearOutput) {
+    void CopyPitchLinearToLinear(const GuestTexture &guest, u8 *guestInput, u8 *linearOutput) {
         auto sizeLine{guest.format->GetSize(guest.dimensions.width, 1)}; //!< The size of a single line of pixel data
         auto sizeStride{guest.tileConfig.pitch}; //!< The size of a single stride of pixel data
 
@@ -118,7 +125,7 @@ namespace skyline::gpu {
     /**
      * @brief Copies the contents of a linear buffer to a pitch-linear guest texture
      */
-    void CopyLinearToPitchLinear(GuestTexture &guest, u8 *linearInput, u8 *guestOutput) {
+    void CopyLinearToPitchLinear(const GuestTexture &guest, u8 *linearInput, u8 *guestOutput) {
         auto sizeLine{guest.format->GetSize(guest.dimensions.width, 1)}; //!< The size of a single line of pixel data
         auto sizeStride{guest.tileConfig.pitch}; //!< The size of a single stride of pixel data
 

app/src/main/cpp/skyline/gpu/texture/texture.cpp

@@ -5,9 +5,22 @@
 #include <common/trace.h>
 #include <kernel/types/KProcess.h>
 #include "texture.h"
-#include "copy.h"
+#include "layout.h"
 
 namespace skyline::gpu {
+    u32 GuestTexture::GetLayerSize() {
+        switch (tileConfig.mode) {
+            case texture::TileMode::Linear:
+                return layerStride = static_cast<u32>(format->GetSize(dimensions));
+
+            case texture::TileMode::Pitch:
+                return layerStride = dimensions.height * tileConfig.pitch;
+
+            case texture::TileMode::Block:
+                return layerStride = static_cast<u32>(texture::GetBlockLinearLayerSize(*this));
+        }
+    }
+
     TextureView::TextureView(std::shared_ptr<Texture> texture, vk::ImageViewType type, vk::ImageSubresourceRange range, texture::Format format, vk::ComponentMapping mapping) : texture(std::move(texture)), type(type), format(format), mapping(mapping), range(range) {}
 
     vk::ImageView TextureView::GetView() {
@@ -103,9 +116,9 @@ namespace skyline::gpu {
         }()};
 
         if (guest->tileConfig.mode == texture::TileMode::Block)
-            CopyBlockLinearToLinear(*guest, pointer, bufferData);
+            texture::CopyBlockLinearToLinear(*guest, pointer, bufferData);
         else if (guest->tileConfig.mode == texture::TileMode::Pitch)
-            CopyPitchLinearToLinear(*guest, pointer, bufferData);
+            texture::CopyPitchLinearToLinear(*guest, pointer, bufferData);
         else if (guest->tileConfig.mode == texture::TileMode::Linear)
             std::memcpy(bufferData, pointer, size);
 
@@ -208,9 +221,9 @@ namespace skyline::gpu {
         auto guestOutput{guest->mappings[0].data()};
 
         if (guest->tileConfig.mode == texture::TileMode::Block)
-            CopyLinearToBlockLinear(*guest, hostBuffer, guestOutput);
+            texture::CopyLinearToBlockLinear(*guest, hostBuffer, guestOutput);
         else if (guest->tileConfig.mode == texture::TileMode::Pitch)
-            CopyLinearToPitchLinear(*guest, hostBuffer, guestOutput);
+            texture::CopyLinearToPitchLinear(*guest, hostBuffer, guestOutput);
         else if (guest->tileConfig.mode == texture::TileMode::Linear)
             std::memcpy(hostBuffer, guestOutput, format->GetSize(dimensions));
     }

app/src/main/cpp/skyline/gpu/texture/texture.h

@@ -261,7 +261,7 @@ namespace skyline::gpu {
         texture::TextureType type{};
         u16 baseArrayLayer{};
         u16 layerCount{};
-        u32 layerStride{}; //!< An optional hint regarding the size of a single layer, it will be set to 0 when not available
+        u32 layerStride{}; //!< An optional hint regarding the size of a single layer, it will be set to 0 when not available, GetLayerSize() should be used to retrieve this value
 
         GuestTexture() {}
 
@@ -284,6 +284,12 @@ namespace skyline::gpu {
               baseArrayLayer(baseArrayLayer),
               layerCount(layerCount),
               layerStride(layerStride) {}
+
+        /**
+         * @note Requires `dimensions`, `format` and `tileConfig` to be filled in
+         * @return The size of a single layer with alignment in bytes
+         */
+        u32 GetLayerSize();
     };
 
     class TextureManager;

app/src/main/cpp/skyline/services/hosbinder/GraphicBufferProducer.cpp

@@ -343,7 +343,9 @@ namespace skyline::service::hosbinder {
                 throw exception("Legacy 16Bx16 tiled surfaces are not supported");
             }
 
-            gpu::GuestTexture guestTexture(span<u8>(nvMapHandleObj->GetPointer() + surface.offset, surface.size), gpu::texture::Dimensions(surface.width, surface.height), format, tileConfig, gpu::texture::TextureType::e2D);
+            gpu::texture::Dimensions dimensions(surface.width, surface.height);
+            gpu::GuestTexture guestTexture(span<u8>{}, dimensions, format, tileConfig, gpu::texture::TextureType::e2D);
+            guestTexture.mappings[0] = span<u8>(nvMapHandleObj->GetPointer() + surface.offset, guestTexture.GetLayerSize());
             buffer.texture = state.gpu->texture.FindOrCreate(guestTexture)->texture;
         }