From 80eacdf89b55528a66b2e94391e640e641e8cb57 Mon Sep 17 00:00:00 2001 From: ReinUsesLisp Date: Sat, 9 Nov 2019 03:26:30 -0300 Subject: texture_cache: Use a table instead of switch for texture formats Use a large flat array to look up texture formats. This allows us to properly implement formats with different component types. It should also be faster. --- .../texture_cache/format_lookup_table.cpp | 214 +++++++++++++++++++++ src/video_core/texture_cache/format_lookup_table.h | 51 +++++ src/video_core/texture_cache/surface_params.cpp | 20 +- src/video_core/texture_cache/surface_params.h | 8 +- src/video_core/texture_cache/texture_cache.h | 7 +- 5 files changed, 288 insertions(+), 12 deletions(-) create mode 100644 src/video_core/texture_cache/format_lookup_table.cpp create mode 100644 src/video_core/texture_cache/format_lookup_table.h (limited to 'src/video_core/texture_cache') diff --git a/src/video_core/texture_cache/format_lookup_table.cpp b/src/video_core/texture_cache/format_lookup_table.cpp new file mode 100644 index 000000000..ee050cc31 --- /dev/null +++ b/src/video_core/texture_cache/format_lookup_table.cpp @@ -0,0 +1,214 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include +#include "common/common_types.h" +#include "common/logging/log.h" +#include "video_core/texture_cache/format_lookup_table.h" + +namespace VideoCommon { + +using Tegra::Texture::ComponentType; +using Tegra::Texture::TextureFormat; +using VideoCore::Surface::PixelFormat; + +namespace { + +static constexpr auto SNORM = ComponentType::SNORM; +static constexpr auto UNORM = ComponentType::UNORM; +static constexpr auto SINT = ComponentType::SINT; +static constexpr auto UINT = ComponentType::UINT; +static constexpr auto SNORM_FORCE_FP16 = ComponentType::SNORM_FORCE_FP16; +static constexpr auto UNORM_FORCE_FP16 = ComponentType::UNORM_FORCE_FP16; +static constexpr auto FLOAT = ComponentType::FLOAT; +static constexpr bool C = false; // Normal color +static constexpr bool S = true; // Srgb + +struct Table { + constexpr Table(TextureFormat texture_format, bool is_srgb, ComponentType red_component, + ComponentType green_component, ComponentType blue_component, + ComponentType alpha_component, PixelFormat pixel_format) + : texture_format{static_cast(texture_format)}, + pixel_format{static_cast(pixel_format)}, red_component{static_cast( + red_component)}, + green_component{static_cast(green_component)}, blue_component{static_cast( + blue_component)}, + alpha_component{static_cast(alpha_component)}, is_srgb{is_srgb ? 1U : 0U} {} + + u32 texture_format : 8; + u32 pixel_format : 8; + u32 red_component : 3; + u32 green_component : 3; + u32 blue_component : 3; + u32 alpha_component : 3; + u32 is_srgb : 1; +}; +constexpr std::array DefinitionTable = {{ + {TextureFormat::A8R8G8B8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ABGR8U}, + {TextureFormat::A8R8G8B8, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::ABGR8S}, + {TextureFormat::A8R8G8B8, C, UINT, UINT, UINT, UINT, PixelFormat::ABGR8UI}, + {TextureFormat::A8R8G8B8, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::RGBA8_SRGB}, + + {TextureFormat::B5G6R5, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::B5G6R5U}, + + {TextureFormat::A2B10G10R10, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::A2B10G10R10U}, + + {TextureFormat::A1B5G5R5, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::A1B5G5R5U}, + + {TextureFormat::A4B4G4R4, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::R4G4B4A4U}, + + {TextureFormat::R8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::R8U}, + {TextureFormat::R8, C, UINT, UINT, UINT, UINT, PixelFormat::R8UI}, + + {TextureFormat::G8R8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::RG8U}, + {TextureFormat::G8R8, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::RG8S}, + + {TextureFormat::R16_G16_B16_A16, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::RGBA16U}, + {TextureFormat::R16_G16_B16_A16, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::RGBA16F}, + {TextureFormat::R16_G16_B16_A16, C, UINT, UINT, UINT, UINT, PixelFormat::RGBA16UI}, + + {TextureFormat::R16_G16, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::RG16F}, + {TextureFormat::R16_G16, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::RG16}, + {TextureFormat::R16_G16, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::RG16S}, + {TextureFormat::R16_G16, C, UINT, UINT, UINT, UINT, PixelFormat::RG16UI}, + {TextureFormat::R16_G16, C, SINT, SINT, SINT, SINT, PixelFormat::RG16I}, + + {TextureFormat::R16, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::R16F}, + {TextureFormat::R16, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::R16U}, + {TextureFormat::R16, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::R16S}, + {TextureFormat::R16, C, UINT, UINT, UINT, UINT, PixelFormat::R16UI}, + {TextureFormat::R16, C, SINT, SINT, SINT, SINT, PixelFormat::R16I}, + + {TextureFormat::BF10GF11RF11, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::R11FG11FB10F}, + + {TextureFormat::R32_G32_B32_A32, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::RGBA32F}, + {TextureFormat::R32_G32_B32_A32, C, UINT, UINT, UINT, UINT, PixelFormat::RGBA32UI}, + + {TextureFormat::R32_G32_B32, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::RGB32F}, + + {TextureFormat::R32_G32, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::RG32F}, + {TextureFormat::R32_G32, C, UINT, UINT, UINT, UINT, PixelFormat::RG32UI}, + + {TextureFormat::R32, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::R32F}, + {TextureFormat::R32, C, UINT, UINT, UINT, UINT, PixelFormat::R32UI}, + + {TextureFormat::E5B9G9R9_SHAREDEXP, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::E5B9G9R9F}, + + {TextureFormat::ZF32, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::Z32F}, + {TextureFormat::Z16, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::Z16}, + {TextureFormat::S8Z24, C, UINT, UNORM, UNORM, UNORM, PixelFormat::S8Z24}, + {TextureFormat::ZF32_X24S8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::Z32FS8}, + + {TextureFormat::DXT1, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT1}, + {TextureFormat::DXT1, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT1_SRGB}, + + {TextureFormat::DXT23, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT23}, + {TextureFormat::DXT23, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT23_SRGB}, + + {TextureFormat::DXT45, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT45}, + {TextureFormat::DXT45, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXT45_SRGB}, + + // TODO: Use a different pixel format for SNORM + {TextureFormat::DXN1, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXN1}, + {TextureFormat::DXN1, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::DXN1}, + + {TextureFormat::DXN2, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::DXN2UNORM}, + {TextureFormat::DXN2, C, SNORM, SNORM, SNORM, SNORM, PixelFormat::DXN2SNORM}, + + {TextureFormat::BC7U, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::BC7U}, + {TextureFormat::BC7U, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::BC7U_SRGB}, + + {TextureFormat::BC6H_SF16, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::BC6H_SF16}, + {TextureFormat::BC6H_UF16, C, FLOAT, FLOAT, FLOAT, FLOAT, PixelFormat::BC6H_UF16}, + + {TextureFormat::ASTC_2D_4X4, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_4X4}, + {TextureFormat::ASTC_2D_4X4, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_4X4_SRGB}, + + {TextureFormat::ASTC_2D_5X4, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_5X4}, + {TextureFormat::ASTC_2D_5X4, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_5X4_SRGB}, + + {TextureFormat::ASTC_2D_5X5, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_5X5}, + {TextureFormat::ASTC_2D_5X5, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_5X5_SRGB}, + + {TextureFormat::ASTC_2D_8X8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_8X8}, + {TextureFormat::ASTC_2D_8X8, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_8X8_SRGB}, + + {TextureFormat::ASTC_2D_8X5, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_8X5}, + {TextureFormat::ASTC_2D_8X5, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_8X5_SRGB}, + + {TextureFormat::ASTC_2D_10X8, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_10X8}, + {TextureFormat::ASTC_2D_10X8, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_10X8_SRGB}, + + {TextureFormat::ASTC_2D_6X6, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_6X6}, + {TextureFormat::ASTC_2D_6X6, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_6X6_SRGB}, + + {TextureFormat::ASTC_2D_10X10, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_10X10}, + {TextureFormat::ASTC_2D_10X10, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_10X10_SRGB}, + + {TextureFormat::ASTC_2D_12X12, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_12X12}, + {TextureFormat::ASTC_2D_12X12, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_12X12_SRGB}, + + {TextureFormat::ASTC_2D_8X6, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_8X6}, + {TextureFormat::ASTC_2D_8X6, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_8X6_SRGB}, + + {TextureFormat::ASTC_2D_6X5, C, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_6X5}, + {TextureFormat::ASTC_2D_6X5, S, UNORM, UNORM, UNORM, UNORM, PixelFormat::ASTC_2D_6X5_SRGB}, +}}; + +} // Anonymous namespace + +FormatLookupTable::FormatLookupTable() { + table.fill(static_cast(PixelFormat::Invalid)); + + for (const auto entry : DefinitionTable) { + table[CalculateIndex(static_cast(entry.texture_format), entry.is_srgb != 0, + static_cast(entry.red_component), + static_cast(entry.green_component), + static_cast(entry.blue_component), + static_cast(entry.alpha_component))] = + static_cast(entry.pixel_format); + } +} + +PixelFormat FormatLookupTable::GetPixelFormat(TextureFormat format, bool is_srgb, + ComponentType red_component, + ComponentType green_component, + ComponentType blue_component, + ComponentType alpha_component) const noexcept { + const auto pixel_format = static_cast(table[CalculateIndex( + format, is_srgb, red_component, green_component, blue_component, alpha_component)]); + // [[likely]] + if (pixel_format != PixelFormat::Invalid) { + return pixel_format; + } + UNIMPLEMENTED_MSG("texture format={} srgb={} components={{{} {} {} {}}}", + static_cast(format), is_srgb, static_cast(red_component), + static_cast(green_component), static_cast(blue_component), + static_cast(alpha_component)); + return PixelFormat::ABGR8U; +} + +void FormatLookupTable::Set(TextureFormat format, bool is_srgb, ComponentType red_component, + ComponentType green_component, ComponentType blue_component, + ComponentType alpha_component, PixelFormat pixel_format) {} + +std::size_t FormatLookupTable::CalculateIndex(TextureFormat format, bool is_srgb, + ComponentType red_component, + ComponentType green_component, + ComponentType blue_component, + ComponentType alpha_component) noexcept { + const auto format_index = static_cast(format); + const auto red_index = static_cast(red_component); + const auto green_index = static_cast(red_component); + const auto blue_index = static_cast(red_component); + const auto alpha_index = static_cast(red_component); + const std::size_t srgb_index = is_srgb ? 1 : 0; + + return format_index * PerFormat + + srgb_index * PerComponent * PerComponent * PerComponent * PerComponent + + alpha_index * PerComponent * PerComponent * PerComponent + + blue_index * PerComponent * PerComponent + green_index * PerComponent + red_index; +} + +} // namespace VideoCommon diff --git a/src/video_core/texture_cache/format_lookup_table.h b/src/video_core/texture_cache/format_lookup_table.h new file mode 100644 index 000000000..8da7481cd --- /dev/null +++ b/src/video_core/texture_cache/format_lookup_table.h @@ -0,0 +1,51 @@ +// Copyright 2019 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include +#include +#include "video_core/surface.h" +#include "video_core/textures/texture.h" + +namespace VideoCommon { + +class FormatLookupTable { +public: + explicit FormatLookupTable(); + + VideoCore::Surface::PixelFormat GetPixelFormat( + Tegra::Texture::TextureFormat format, bool is_srgb, + Tegra::Texture::ComponentType red_component, Tegra::Texture::ComponentType green_component, + Tegra::Texture::ComponentType blue_component, + Tegra::Texture::ComponentType alpha_component) const noexcept; + +private: + static_assert(VideoCore::Surface::MaxPixelFormat <= std::numeric_limits::max()); + + static constexpr std::size_t NumTextureFormats = 128; + + static constexpr std::size_t PerComponent = 8; + static constexpr std::size_t PerComponents2 = PerComponent * PerComponent; + static constexpr std::size_t PerComponents3 = PerComponents2 * PerComponent; + static constexpr std::size_t PerComponents4 = PerComponents3 * PerComponent; + static constexpr std::size_t PerFormat = PerComponents4 * 2; + + static std::size_t CalculateIndex(Tegra::Texture::TextureFormat format, bool is_srgb, + Tegra::Texture::ComponentType red_component, + Tegra::Texture::ComponentType green_component, + Tegra::Texture::ComponentType blue_component, + Tegra::Texture::ComponentType alpha_component) noexcept; + + void Set(Tegra::Texture::TextureFormat format, bool is_srgb, + Tegra::Texture::ComponentType red_component, + Tegra::Texture::ComponentType green_component, + Tegra::Texture::ComponentType blue_component, + Tegra::Texture::ComponentType alpha_component, + VideoCore::Surface::PixelFormat pixel_format); + + std::array table; +}; + +} // namespace VideoCommon diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp index 04b5e8ba3..858e17e08 100644 --- a/src/video_core/texture_cache/surface_params.cpp +++ b/src/video_core/texture_cache/surface_params.cpp @@ -2,13 +2,16 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. -#include +#include +#include +#include #include "common/alignment.h" #include "common/bit_util.h" #include "core/core.h" #include "video_core/engines/shader_bytecode.h" #include "video_core/surface.h" +#include "video_core/texture_cache/format_lookup_table.h" #include "video_core/texture_cache/surface_params.h" namespace VideoCommon { @@ -16,7 +19,6 @@ namespace VideoCommon { using VideoCore::Surface::PixelFormat; using VideoCore::Surface::PixelFormatFromDepthFormat; using VideoCore::Surface::PixelFormatFromRenderTargetFormat; -using VideoCore::Surface::PixelFormatFromTextureFormat; using VideoCore::Surface::SurfaceTarget; using VideoCore::Surface::SurfaceTargetFromTextureType; using VideoCore::Surface::SurfaceType; @@ -66,7 +68,8 @@ constexpr u32 GetMipmapSize(bool uncompressed, u32 mip_size, u32 tile) { } // Anonymous namespace -SurfaceParams SurfaceParams::CreateForTexture(const Tegra::Texture::TICEntry& tic, +SurfaceParams SurfaceParams::CreateForTexture(const FormatLookupTable& lookup_table, + const Tegra::Texture::TICEntry& tic, const VideoCommon::Shader::Sampler& entry) { SurfaceParams params; params.is_tiled = tic.IsTiled(); @@ -75,8 +78,8 @@ SurfaceParams SurfaceParams::CreateForTexture(const Tegra::Texture::TICEntry& ti params.block_height = params.is_tiled ? tic.BlockHeight() : 0, params.block_depth = params.is_tiled ? tic.BlockDepth() : 0, params.tile_width_spacing = params.is_tiled ? (1 << tic.tile_width_spacing.Value()) : 1; - params.pixel_format = - PixelFormatFromTextureFormat(tic.format, tic.r_type.Value(), params.srgb_conversion); + params.pixel_format = lookup_table.GetPixelFormat( + tic.format, params.srgb_conversion, tic.r_type, tic.g_type, tic.b_type, tic.a_type); params.type = GetFormatType(params.pixel_format); if (entry.IsShadow() && params.type == SurfaceType::ColorTexture) { switch (params.pixel_format) { @@ -124,7 +127,8 @@ SurfaceParams SurfaceParams::CreateForTexture(const Tegra::Texture::TICEntry& ti return params; } -SurfaceParams SurfaceParams::CreateForImage(const Tegra::Texture::TICEntry& tic, +SurfaceParams SurfaceParams::CreateForImage(const FormatLookupTable& lookup_table, + const Tegra::Texture::TICEntry& tic, const VideoCommon::Shader::Image& entry) { SurfaceParams params; params.is_tiled = tic.IsTiled(); @@ -133,8 +137,8 @@ SurfaceParams SurfaceParams::CreateForImage(const Tegra::Texture::TICEntry& tic, params.block_height = params.is_tiled ? tic.BlockHeight() : 0, params.block_depth = params.is_tiled ? tic.BlockDepth() : 0, params.tile_width_spacing = params.is_tiled ? (1 << tic.tile_width_spacing.Value()) : 1; - params.pixel_format = - PixelFormatFromTextureFormat(tic.format, tic.r_type.Value(), params.srgb_conversion); + params.pixel_format = lookup_table.GetPixelFormat( + tic.format, params.srgb_conversion, tic.r_type, tic.g_type, tic.b_type, tic.a_type); params.type = GetFormatType(params.pixel_format); params.type = GetFormatType(params.pixel_format); params.target = ImageTypeToSurfaceTarget(entry.GetType()); diff --git a/src/video_core/texture_cache/surface_params.h b/src/video_core/texture_cache/surface_params.h index a84058869..709aa0dc2 100644 --- a/src/video_core/texture_cache/surface_params.h +++ b/src/video_core/texture_cache/surface_params.h @@ -16,16 +16,20 @@ namespace VideoCommon { +class FormatLookupTable; + using VideoCore::Surface::SurfaceCompression; class SurfaceParams { public: /// Creates SurfaceCachedParams from a texture configuration. - static SurfaceParams CreateForTexture(const Tegra::Texture::TICEntry& tic, + static SurfaceParams CreateForTexture(const FormatLookupTable& lookup_table, + const Tegra::Texture::TICEntry& tic, const VideoCommon::Shader::Sampler& entry); /// Creates SurfaceCachedParams from an image configuration. - static SurfaceParams CreateForImage(const Tegra::Texture::TICEntry& tic, + static SurfaceParams CreateForImage(const FormatLookupTable& lookup_table, + const Tegra::Texture::TICEntry& tic, const VideoCommon::Shader::Image& entry); /// Creates SurfaceCachedParams for a depth buffer configuration. diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h index 8074cc945..41309ebea 100644 --- a/src/video_core/texture_cache/texture_cache.h +++ b/src/video_core/texture_cache/texture_cache.h @@ -29,6 +29,7 @@ #include "video_core/rasterizer_interface.h" #include "video_core/surface.h" #include "video_core/texture_cache/copy_params.h" +#include "video_core/texture_cache/format_lookup_table.h" #include "video_core/texture_cache/surface_base.h" #include "video_core/texture_cache/surface_params.h" #include "video_core/texture_cache/surface_view.h" @@ -96,7 +97,7 @@ public: if (!gpu_addr) { return {}; } - const auto params{SurfaceParams::CreateForTexture(tic, entry)}; + const auto params{SurfaceParams::CreateForTexture(format_lookup_table, tic, entry)}; const auto [surface, view] = GetSurface(gpu_addr, params, true, false); if (guard_samplers) { sampled_textures.push_back(surface); @@ -111,7 +112,7 @@ public: if (!gpu_addr) { return {}; } - const auto params{SurfaceParams::CreateForImage(tic, entry)}; + const auto params{SurfaceParams::CreateForImage(format_lookup_table, tic, entry)}; const auto [surface, view] = GetSurface(gpu_addr, params, true, false); if (guard_samplers) { sampled_textures.push_back(surface); @@ -953,6 +954,8 @@ private: VideoCore::RasterizerInterface& rasterizer; + FormatLookupTable format_lookup_table; + u64 ticks{}; // Guards the cache for protection conflicts. -- cgit v1.2.3