From f8e994354f62e823d2a480eb00e0b3e34eff1a65 Mon Sep 17 00:00:00 2001 From: FernandoS27 Date: Thu, 13 Sep 2018 22:58:40 -0400 Subject: Optimized Texture Swizzling --- src/video_core/textures/decoders.cpp | 51 ++++++++++++++++++++++++++++++++++-- 1 file changed, 49 insertions(+), 2 deletions(-) (limited to 'src/video_core/textures') diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 272294c62..55926cca4 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -46,6 +46,48 @@ void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_ } } +template +struct alignas(64) SwizzleTable { + constexpr SwizzleTable() { + for (u32 y = 0; y < N; ++y) { + for (u32 x = 0; x < M; ++x) { + const u32 x2 = x * 16; + values[y][x] = static_cast(((x2 % 64) / 32) * 256 + ((y % 8) / 2) * 64 + + ((x2 % 32) / 16) * 32 + (y % 2) * 16); + } + } + } + const std::array& operator[](std::size_t index) const { + return values[index]; + } + std::array, N> values{}; +}; + +constexpr auto swizzle_table = SwizzleTable<8, 4>(); + +void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u8* swizzled_data, + u8* unswizzled_data, bool unswizzle, u32 block_height) { + std::array data_ptrs; + const std::size_t stride{width * bytes_per_pixel}; + const std::size_t image_width_in_gobs{(stride + 63) / 64}; + const std::size_t copy_size{16}; + for (std::size_t y = 0; y < height; ++y) { + const std::size_t initial_gob = + (y / (8 * block_height)) * 512 * block_height * image_width_in_gobs + + (y % (8 * block_height) / 8) * 512; + const std::size_t pixel_base{y * width * bytes_per_pixel}; + const auto& table = swizzle_table[y % 8]; + for (std::size_t xb = 0; xb < stride; xb += copy_size) { + const std::size_t gob_address{initial_gob + (xb / 64) * 512 * block_height}; + const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]}; + const std::size_t pixel_index{xb + pixel_base}; + data_ptrs[unswizzle] = swizzled_data + swizzle_offset; + data_ptrs[!unswizzle] = unswizzled_data + pixel_index; + std::memcpy(data_ptrs[0], data_ptrs[1], copy_size); + } + } +} + u32 BytesPerPixel(TextureFormat format) { switch (format) { case TextureFormat::DXT1: @@ -91,8 +133,13 @@ u32 BytesPerPixel(TextureFormat format) { std::vector UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, u32 height, u32 block_height) { std::vector unswizzled_data(width * height * bytes_per_pixel); - CopySwizzledData(width / tile_size, height / tile_size, bytes_per_pixel, bytes_per_pixel, - Memory::GetPointer(address), unswizzled_data.data(), true, block_height); + if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) { + FastSwizzleData(width / tile_size, height / tile_size, bytes_per_pixel, + Memory::GetPointer(address), unswizzled_data.data(), true, block_height); + } else { + CopySwizzledData(width / tile_size, height / tile_size, bytes_per_pixel, bytes_per_pixel, + Memory::GetPointer(address), unswizzled_data.data(), true, block_height); + } return unswizzled_data; } -- cgit v1.2.3 From 2845348608ccccd0ce36a351cc4a35e753d84837 Mon Sep 17 00:00:00 2001 From: raven02 Date: Sun, 16 Sep 2018 22:47:02 +0800 Subject: Implement ASTC_2D_8X8 (Bayonetta 2) --- src/video_core/renderer_opengl/gl_rasterizer_cache.cpp | 9 ++++++++- src/video_core/renderer_opengl/gl_rasterizer_cache.h | 15 ++++++++++----- src/video_core/textures/decoders.cpp | 2 ++ 3 files changed, 20 insertions(+), 6 deletions(-) (limited to 'src/video_core/textures') diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 32001e44b..63bbcb666 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -167,6 +167,7 @@ static constexpr std::array tex_form {GL_RG8, GL_RG, GL_BYTE, ComponentType::SNorm, false}, // RG8S {GL_RG32UI, GL_RG_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false}, // RG32UI {GL_R32UI, GL_RED_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false}, // R32UI + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X8 // Depth formats {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F @@ -213,6 +214,7 @@ static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType static bool IsPixelFormatASTC(PixelFormat format) { switch (format) { case PixelFormat::ASTC_2D_4X4: + case PixelFormat::ASTC_2D_8X8: return true; default: return false; @@ -223,6 +225,8 @@ static std::pair GetASTCBlockSize(PixelFormat format) { switch (format) { case PixelFormat::ASTC_2D_4X4: return {4, 4}; + case PixelFormat::ASTC_2D_8X8: + return {8, 8}; default: LOG_CRITICAL(HW_GPU, "Unhandled format: {}", static_cast(format)); UNREACHABLE(); @@ -327,6 +331,7 @@ static constexpr std::array, MortonCopy, MortonCopy, + MortonCopy, MortonCopy, MortonCopy, MortonCopy, @@ -386,6 +391,7 @@ static constexpr std::array, MortonCopy, MortonCopy, + nullptr, MortonCopy, MortonCopy, MortonCopy, @@ -544,7 +550,8 @@ static void ConvertG8R8ToR8G8(std::vector& data, u32 width, u32 height) { static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector& data, PixelFormat pixel_format, u32 width, u32 height) { switch (pixel_format) { - case PixelFormat::ASTC_2D_4X4: { + case PixelFormat::ASTC_2D_4X4: + case PixelFormat::ASTC_2D_8X8: { // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC. u32 block_width{}; u32 block_height{}; diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index 57ea8593b..0525e9a94 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -70,19 +70,20 @@ struct SurfaceParams { RG8S = 42, RG32UI = 43, R32UI = 44, + ASTC_2D_8X8 = 45, MaxColorFormat, // Depth formats - Z32F = 45, - Z16 = 46, + Z32F = 46, + Z16 = 47, MaxDepthFormat, // DepthStencil formats - Z24S8 = 47, - S8Z24 = 48, - Z32FS8 = 49, + Z24S8 = 48, + S8Z24 = 49, + Z32FS8 = 50, MaxDepthStencilFormat, @@ -192,6 +193,7 @@ struct SurfaceParams { 1, // RG8S 1, // RG32UI 1, // R32UI + 4, // ASTC_2D_8X8 1, // Z32F 1, // Z16 1, // Z24S8 @@ -253,6 +255,7 @@ struct SurfaceParams { 16, // RG8S 64, // RG32UI 32, // R32UI + 16, // ASTC_2D_8X8 32, // Z32F 16, // Z16 32, // Z24S8 @@ -522,6 +525,8 @@ struct SurfaceParams { return PixelFormat::BC6H_SF16; case Tegra::Texture::TextureFormat::ASTC_2D_4X4: return PixelFormat::ASTC_2D_4X4; + case Tegra::Texture::TextureFormat::ASTC_2D_8X8: + return PixelFormat::ASTC_2D_8X8; case Tegra::Texture::TextureFormat::R16_G16: switch (component_type) { case Tegra::Texture::ComponentType::FLOAT: diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 272294c62..58b65de1a 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -63,6 +63,7 @@ u32 BytesPerPixel(TextureFormat format) { case TextureFormat::R32_G32_B32: return 12; case TextureFormat::ASTC_2D_4X4: + case TextureFormat::ASTC_2D_8X8: case TextureFormat::A8R8G8B8: case TextureFormat::A2B10G10R10: case TextureFormat::BF10GF11RF11: @@ -111,6 +112,7 @@ std::vector DecodeTexture(const std::vector& texture_data, TextureFormat case TextureFormat::BC6H_UF16: case TextureFormat::BC6H_SF16: case TextureFormat::ASTC_2D_4X4: + case TextureFormat::ASTC_2D_8X8: case TextureFormat::A8R8G8B8: case TextureFormat::A2B10G10R10: case TextureFormat::A1B5G5R5: -- cgit v1.2.3