video_core: Use un-shifted block sizes to avoid integer divisions

Instead of storing all block width, height and depths in their shifted
form:

block_width = 1U << block_shift;

Store them like they are provided by the emulated hardware (their
block_shift form). This way we can avoid doing the costly
Common::AlignUp operation to align texture sizes and drop CPU integer
divisions with bitwise logic (defined in Common::AlignBits).

src/common/alignment.h

@@ -19,6 +19,11 @@ constexpr T AlignDown(T value, std::size_t size) {
     return static_cast<T>(value - value % size);
 }
 
+template <typename T>
+constexpr T AlignBits(T value, T align) {
+    return (value + ((1 << align) - 1)) >> align << align;
+}
+
 template <typename T>
 constexpr bool Is4KBAligned(T value) {
     static_assert(std::is_unsigned_v<T>, "T must be an unsigned value.");

src/video_core/engines/fermi_2d.h

@@ -63,18 +63,15 @@ public:
             }
 
             u32 BlockWidth() const {
-                // The block width is stored in log2 format.
-                return 1 << block_width;
+                return block_width;
             }
 
             u32 BlockHeight() const {
-                // The block height is stored in log2 format.
-                return 1 << block_height;
+                return block_height;
             }
 
             u32 BlockDepth() const {
-                // The block depth is stored in log2 format.
-                return 1 << block_depth;
+                return block_depth;
             }
         };
         static_assert(sizeof(Surface) == 0x28, "Surface has incorrect size");

src/video_core/engines/maxwell_dma.h

@@ -59,11 +59,11 @@ public:
             };
 
             u32 BlockHeight() const {
-                return 1 << block_height;
+                return block_height;
             }
 
             u32 BlockDepth() const {
-                return 1 << block_depth;
+                return block_depth;
             }
         };
 

src/video_core/texture_cache/surface_base.cpp

@@ -22,7 +22,6 @@ SurfaceBaseImpl::SurfaceBaseImpl(GPUVAddr gpu_addr, const SurfaceParams& params)
     : params{params}, mipmap_sizes(params.num_levels),
       mipmap_offsets(params.num_levels), gpu_addr{gpu_addr}, host_memory_size{
                                                                  params.GetHostSizeInBytes()} {
-
     std::size_t offset = 0;
     for (u32 level = 0; level < params.num_levels; ++level) {
         const std::size_t mipmap_size{params.GetGuestMipmapSize(level)};
@@ -75,7 +74,7 @@ void SurfaceBaseImpl::LoadBuffer(Tegra::MemoryManager& memory_manager,
         return;
     }
     if (params.is_tiled) {
-        ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture target {}",
+        ASSERT_MSG(params.block_width == 0, "Block width is defined as {} on texture target {}",
                    params.block_width, static_cast<u32>(params.target));
         for (u32 level = 0; level < params.num_levels; ++level) {
             const std::size_t host_offset{params.GetHostMipmapLevelOffset(level)};

src/video_core/texture_cache/surface_params.cpp

@@ -96,9 +96,9 @@ SurfaceParams SurfaceParams::CreateForDepthBuffer(
     SurfaceParams params;
     params.is_tiled = type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear;
     params.srgb_conversion = false;
-    params.block_width = 1 << std::min(block_width, 5U);
-    params.block_height = 1 << std::min(block_height, 5U);
-    params.block_depth = 1 << std::min(block_depth, 5U);
+    params.block_width = std::min(block_width, 5U);
+    params.block_height = std::min(block_height, 5U);
+    params.block_depth = std::min(block_depth, 5U);
     params.tile_width_spacing = 1;
     params.pixel_format = PixelFormatFromDepthFormat(format);
     params.component_type = ComponentTypeFromDepthFormat(format);
@@ -120,9 +120,9 @@ SurfaceParams SurfaceParams::CreateForFramebuffer(Core::System& system, std::siz
         config.memory_layout.type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear;
     params.srgb_conversion = config.format == Tegra::RenderTargetFormat::BGRA8_SRGB ||
                              config.format == Tegra::RenderTargetFormat::RGBA8_SRGB;
-    params.block_width = 1 << config.memory_layout.block_width;
-    params.block_height = 1 << config.memory_layout.block_height;
-    params.block_depth = 1 << config.memory_layout.block_depth;
+    params.block_width = config.memory_layout.block_width;
+    params.block_height = config.memory_layout.block_height;
+    params.block_depth = config.memory_layout.block_depth;
     params.tile_width_spacing = 1;
     params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
     params.component_type = ComponentTypeFromRenderTarget(config.format);
@@ -149,9 +149,9 @@ SurfaceParams SurfaceParams::CreateForFermiCopySurface(
     params.is_tiled = !config.linear;
     params.srgb_conversion = config.format == Tegra::RenderTargetFormat::BGRA8_SRGB ||
                              config.format == Tegra::RenderTargetFormat::RGBA8_SRGB;
-    params.block_width = params.is_tiled ? std::min(config.BlockWidth(), 32U) : 0,
-    params.block_height = params.is_tiled ? std::min(config.BlockHeight(), 32U) : 0,
-    params.block_depth = params.is_tiled ? std::min(config.BlockDepth(), 32U) : 0,
+    params.block_width = params.is_tiled ? std::min(config.BlockWidth(), 5U) : 0,
+    params.block_height = params.is_tiled ? std::min(config.BlockHeight(), 5U) : 0,
+    params.block_depth = params.is_tiled ? std::min(config.BlockDepth(), 5U) : 0,
     params.tile_width_spacing = 1;
     params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
     params.component_type = ComponentTypeFromRenderTarget(config.format);
@@ -190,9 +190,9 @@ u32 SurfaceParams::GetMipBlockHeight(u32 level) const {
     const u32 height{GetMipHeight(level)};
     const u32 default_block_height{GetDefaultBlockHeight()};
     const u32 blocks_in_y{(height + default_block_height - 1) / default_block_height};
-    u32 block_height = 16;
-    while (block_height > 1 && blocks_in_y <= block_height * 4) {
-        block_height >>= 1;
+    u32 block_height = 4;
+    while (block_height > 0 && blocks_in_y <= (1U << block_height) * 4) {
+        --block_height;
     }
     return block_height;
 }
@@ -202,17 +202,17 @@ u32 SurfaceParams::GetMipBlockDepth(u32 level) const {
         return this->block_depth;
     }
     if (is_layered) {
-        return 1;
+        return 0;
     }
 
     const u32 depth{GetMipDepth(level)};
-    u32 block_depth = 32;
-    while (block_depth > 1 && depth * 2 <= block_depth) {
-        block_depth >>= 1;
+    u32 block_depth = 5;
+    while (block_depth > 0 && depth * 2 <= (1U << block_depth)) {
+        --block_depth;
     }
 
-    if (block_depth == 32 && GetMipBlockHeight(level) >= 4) {
-        return 16;
+    if (block_depth == 5 && GetMipBlockHeight(level) >= 2) {
+        return 4;
     }
 
     return block_depth;
@@ -252,7 +252,8 @@ std::size_t SurfaceParams::GetLayerSize(bool as_host_size, bool uncompressed) co
         size += GetInnerMipmapMemorySize(level, as_host_size, uncompressed);
     }
     if (is_tiled && is_layered) {
-        return Common::AlignUp(size, Tegra::Texture::GetGOBSize() * block_height * block_depth);
+        return Common::AlignBits(size,
+                                 Tegra::Texture::GetGOBSizeShift() + block_height + block_depth);
     }
     return size;
 }

src/video_core/texture_cache/surface_params.h

@@ -54,12 +54,12 @@ public:
             constexpr std::size_t rgb8_bpp = 4ULL;
             // ASTC is uncompressed in software, in emulated as RGBA8
             host_size_in_bytes = 0;
-            for (std::size_t level = 0; level < num_levels; level++) {
+            for (u32 level = 0; level < num_levels; ++level) {
                 const std::size_t width =
                     Common::AlignUp(GetMipWidth(level), GetDefaultBlockWidth());
                 const std::size_t height =
                     Common::AlignUp(GetMipHeight(level), GetDefaultBlockHeight());
-                const std::size_t depth = is_layered ? depth : GetMipDepth(level);
+                const std::size_t depth = is_layered ? this->depth : GetMipDepth(level);
                 host_size_in_bytes += width * height * depth * rgb8_bpp;
             }
         } else {
@@ -96,7 +96,8 @@ public:
     // Helper used for out of class size calculations
     static std::size_t AlignLayered(const std::size_t out_size, const u32 block_height,
                                     const u32 block_depth) {
-        return Common::AlignUp(out_size, Tegra::Texture::GetGOBSize() * block_height * block_depth);
+        return Common::AlignBits(out_size,
+                                 Tegra::Texture::GetGOBSizeShift() + block_height + block_depth);
     }
 
     /// Returns the offset in bytes in guest memory of a given mipmap level.

src/video_core/texture_cache/texture_cache.h

@@ -81,6 +81,9 @@ public:
         if (!gpu_addr) {
             return {};
         }
+        if (gpu_addr == 0x1b7ec0000) {
+            // __debugbreak();
+        }
         const auto params{SurfaceParams::CreateForTexture(system, config, entry)};
         return GetSurface(gpu_addr, params, true).second;
     }

src/video_core/textures/decoders.cpp

@@ -36,10 +36,16 @@ struct alignas(64) SwizzleTable {
     std::array<std::array<u16, M>, N> values{};
 };
 
-constexpr u32 gob_size_x = 64;
-constexpr u32 gob_size_y = 8;
-constexpr u32 gob_size_z = 1;
-constexpr u32 gob_size = gob_size_x * gob_size_y * gob_size_z;
+constexpr u32 gob_size_x_shift = 6;
+constexpr u32 gob_size_y_shift = 3;
+constexpr u32 gob_size_z_shift = 0;
+constexpr u32 gob_size_shift = gob_size_x_shift + gob_size_y_shift + gob_size_z_shift;
+
+constexpr u32 gob_size_x = 1U << gob_size_x_shift;
+constexpr u32 gob_size_y = 1U << gob_size_y_shift;
+constexpr u32 gob_size_z = 1U << gob_size_z_shift;
+constexpr u32 gob_size = 1U << gob_size_shift;
+
 constexpr u32 fast_swizzle_align = 16;
 
 constexpr auto legacy_swizzle_table = SwizzleTable<gob_size_y, gob_size_x, gob_size_z>();
@@ -171,14 +177,16 @@ void SwizzledData(u8* const swizzled_data, u8* const unswizzled_data, const bool
 void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel,
                       u32 out_bytes_per_pixel, u8* const swizzled_data, u8* const unswizzled_data,
                       bool unswizzle, u32 block_height, u32 block_depth, u32 width_spacing) {
+    const u32 block_height_size{1U << block_height};
+    const u32 block_depth_size{1U << block_depth};
     if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % fast_swizzle_align == 0) {
         SwizzledData<true>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
-                           bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth,
-                           width_spacing);
+                           bytes_per_pixel, out_bytes_per_pixel, block_height_size,
+                           block_depth_size, width_spacing);
     } else {
         SwizzledData<false>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
-                            bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth,
-                            width_spacing);
+                            bytes_per_pixel, out_bytes_per_pixel, block_height_size,
+                            block_depth_size, width_spacing);
     }
 }
 
@@ -249,16 +257,18 @@ std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y,
 
 void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
                     u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height) {
+    const u32 block_height_size{1U << block_height};
     const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) /
                                   gob_size_x};
     for (u32 line = 0; line < subrect_height; ++line) {
         const u32 gob_address_y =
-            (line / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs +
-            ((line % (gob_size_y * block_height)) / gob_size_y) * gob_size;
+            (line / (gob_size_y * block_height_size)) * gob_size * block_height_size *
+                image_width_in_gobs +
+            ((line % (gob_size_y * block_height_size)) / gob_size_y) * gob_size;
         const auto& table = legacy_swizzle_table[line % gob_size_y];
         for (u32 x = 0; x < subrect_width; ++x) {
             const u32 gob_address =
-                gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height;
+                gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height_size;
             const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % gob_size_x];
             u8* source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel;
             u8* dest_addr = swizzled_data + swizzled_offset;
@@ -271,14 +281,17 @@ void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32
 void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width,
                       u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height,
                       u32 offset_x, u32 offset_y) {
+    const u32 block_height_size{1U << block_height};
     for (u32 line = 0; line < subrect_height; ++line) {
         const u32 y2 = line + offset_y;
-        const u32 gob_address_y = (y2 / (gob_size_y * block_height)) * gob_size * block_height +
-                                  ((y2 % (gob_size_y * block_height)) / gob_size_y) * gob_size;
+        const u32 gob_address_y =
+            (y2 / (gob_size_y * block_height_size)) * gob_size * block_height_size +
+            ((y2 % (gob_size_y * block_height_size)) / gob_size_y) * gob_size;
         const auto& table = legacy_swizzle_table[y2 % gob_size_y];
         for (u32 x = 0; x < subrect_width; ++x) {
             const u32 x2 = (x + offset_x) * bytes_per_pixel;
-            const u32 gob_address = gob_address_y + (x2 / gob_size_x) * gob_size * block_height;
+            const u32 gob_address =
+                gob_address_y + (x2 / gob_size_x) * gob_size * block_height_size;
             const u32 swizzled_offset = gob_address + table[x2 % gob_size_x];
             u8* dest_line = unswizzled_data + line * dest_pitch + x * bytes_per_pixel;
             u8* source_addr = swizzled_data + swizzled_offset;
@@ -291,16 +304,18 @@ void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32
 void SwizzleKepler(const u32 width, const u32 height, const u32 dst_x, const u32 dst_y,
                    const u32 block_height, const std::size_t copy_size, const u8* source_data,
                    u8* swizzle_data) {
+    const u32 block_height_size{1U << block_height};
     const u32 image_width_in_gobs{(width + gob_size_x - 1) / gob_size_x};
     std::size_t count = 0;
     for (std::size_t y = dst_y; y < height && count < copy_size; ++y) {
         const std::size_t gob_address_y =
-            (y / (gob_size_y * block_height)) * gob_size * block_height * image_width_in_gobs +
-            ((y % (gob_size_y * block_height)) / gob_size_y) * gob_size;
+            (y / (gob_size_y * block_height_size)) * gob_size * block_height_size *
+                image_width_in_gobs +
+            ((y % (gob_size_y * block_height_size)) / gob_size_y) * gob_size;
         const auto& table = legacy_swizzle_table[y % gob_size_y];
         for (std::size_t x = dst_x; x < width && count < copy_size; ++x) {
             const std::size_t gob_address =
-                gob_address_y + (x / gob_size_x) * gob_size * block_height;
+                gob_address_y + (x / gob_size_x) * gob_size * block_height_size;
             const std::size_t swizzled_offset = gob_address + table[x % gob_size_x];
             const u8* source_line = source_data + count;
             u8* dest_addr = swizzle_data + swizzled_offset;
@@ -356,9 +371,9 @@ std::vector<u8> DecodeTexture(const std::vector<u8>& texture_data, TextureFormat
 std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
                           u32 block_height, u32 block_depth) {
     if (tiled) {
-        const u32 aligned_width = Common::AlignUp(width * bytes_per_pixel, gob_size_x);
-        const u32 aligned_height = Common::AlignUp(height, gob_size_y * block_height);
-        const u32 aligned_depth = Common::AlignUp(depth, gob_size_z * block_depth);
+        const u32 aligned_width = Common::AlignBits(width * bytes_per_pixel, gob_size_x_shift);
+        const u32 aligned_height = Common::AlignBits(height, gob_size_y_shift + block_height);
+        const u32 aligned_depth = Common::AlignBits(depth, gob_size_z_shift + block_depth);
         return aligned_width * aligned_height * aligned_depth;
     } else {
         return width * height * depth * bytes_per_pixel;

src/video_core/textures/decoders.h

@@ -12,8 +12,8 @@ namespace Tegra::Texture {
 
 // GOBSize constant. Calculated by 64 bytes in x multiplied by 8 y coords, represents
 // an small rect of (64/bytes_per_pixel)X8.
-inline std::size_t GetGOBSize() {
-    return 512;
+inline std::size_t GetGOBSizeShift() {
+    return 9;
 }
 
 /// Unswizzles a swizzled texture without changing its format.

src/video_core/textures/texture.h

@@ -219,20 +219,17 @@ struct TICEntry {
 
     u32 BlockWidth() const {
         ASSERT(IsTiled());
-        // The block height is stored in log2 format.
-        return 1 << block_width;
+        return block_width;
     }
 
     u32 BlockHeight() const {
         ASSERT(IsTiled());
-        // The block height is stored in log2 format.
-        return 1 << block_height;
+        return block_height;
     }
 
     u32 BlockDepth() const {
         ASSERT(IsTiled());
-        // The block height is stored in log2 format.
-        return 1 << block_depth;
+        return block_depth;
     }
 
     bool IsTiled() const {