diff options
Diffstat (limited to 'src/video_core')
-rw-r--r-- | src/video_core/engines/fermi_2d.cpp | 14 | ||||
-rw-r--r-- | src/video_core/engines/maxwell_dma.cpp | 10 | ||||
-rw-r--r-- | src/video_core/engines/maxwell_dma.h | 4 | ||||
-rw-r--r-- | src/video_core/engines/shader_bytecode.h | 2 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer.cpp | 3 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer_cache.cpp | 59 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer_cache.h | 20 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 77 | ||||
-rw-r--r-- | src/video_core/textures/decoders.cpp | 193 | ||||
-rw-r--r-- | src/video_core/textures/decoders.h | 15 | ||||
-rw-r--r-- | src/video_core/textures/texture.h | 1 |
11 files changed, 266 insertions, 132 deletions
diff --git a/src/video_core/engines/fermi_2d.cpp b/src/video_core/engines/fermi_2d.cpp index 912e785b9..597b279b9 100644 --- a/src/video_core/engines/fermi_2d.cpp +++ b/src/video_core/engines/fermi_2d.cpp @@ -62,14 +62,16 @@ void Fermi2D::HandleSurfaceCopy() { u8* dst_buffer = Memory::GetPointer(dest_cpu); if (!regs.src.linear && regs.dst.linear) { // If the input is tiled and the output is linear, deswizzle the input and copy it over. - Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel, - dst_bytes_per_pixel, src_buffer, dst_buffer, true, - regs.src.BlockHeight()); + Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth, + src_bytes_per_pixel, dst_bytes_per_pixel, src_buffer, + dst_buffer, true, regs.src.BlockHeight(), + regs.src.BlockDepth()); } else { // If the input is linear and the output is tiled, swizzle the input and copy it over. - Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel, - dst_bytes_per_pixel, dst_buffer, src_buffer, false, - regs.dst.BlockHeight()); + Texture::CopySwizzledData(regs.src.width, regs.src.height, regs.src.depth, + src_bytes_per_pixel, dst_bytes_per_pixel, dst_buffer, + src_buffer, false, regs.dst.BlockHeight(), + regs.dst.BlockDepth()); } } } diff --git a/src/video_core/engines/maxwell_dma.cpp b/src/video_core/engines/maxwell_dma.cpp index aa7481b8c..bf2a21bb6 100644 --- a/src/video_core/engines/maxwell_dma.cpp +++ b/src/video_core/engines/maxwell_dma.cpp @@ -68,12 +68,14 @@ void MaxwellDMA::HandleCopy() { if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) { // If the input is tiled and the output is linear, deswizzle the input and copy it over. - Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y, 1, 1, src_buffer, - dst_buffer, true, regs.src_params.BlockHeight()); + Texture::CopySwizzledData(regs.src_params.size_x, regs.src_params.size_y, + regs.src_params.size_z, 1, 1, src_buffer, dst_buffer, true, + regs.src_params.BlockHeight(), regs.src_params.BlockDepth()); } else { // If the input is linear and the output is tiled, swizzle the input and copy it over. - Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y, 1, 1, dst_buffer, - src_buffer, false, regs.dst_params.BlockHeight()); + Texture::CopySwizzledData(regs.dst_params.size_x, regs.dst_params.size_y, + regs.dst_params.size_z, 1, 1, dst_buffer, src_buffer, false, + regs.dst_params.BlockHeight(), regs.dst_params.BlockDepth()); } } diff --git a/src/video_core/engines/maxwell_dma.h b/src/video_core/engines/maxwell_dma.h index 311ccb616..df19e02e2 100644 --- a/src/video_core/engines/maxwell_dma.h +++ b/src/video_core/engines/maxwell_dma.h @@ -43,6 +43,10 @@ public: u32 BlockHeight() const { return 1 << block_height; } + + u32 BlockDepth() const { + return 1 << block_depth; + } }; static_assert(sizeof(Parameters) == 24, "Parameters has wrong size"); diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 9a59b65b3..f356f9a03 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -267,7 +267,7 @@ enum class ControlCode : u64 { GTU = 12, NEU = 13, GEU = 14, - // + T = 15, OFF = 16, LO = 17, SFF = 18, diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 84582c777..8d5f277e2 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -286,7 +286,8 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { &ubo, sizeof(ubo), static_cast<std::size_t>(uniform_buffer_alignment)); // Bind the buffer - glBindBufferRange(GL_UNIFORM_BUFFER, stage, buffer_cache.GetHandle(), offset, sizeof(ubo)); + glBindBufferRange(GL_UNIFORM_BUFFER, static_cast<GLuint>(stage), buffer_cache.GetHandle(), + offset, static_cast<GLsizeiptr>(sizeof(ubo))); Shader shader{shader_cache.GetStageProgram(program)}; diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp index 65a220c41..801d45144 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp @@ -231,6 +231,8 @@ static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_form {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 + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_8X5 + {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE, ComponentType::UNorm, false}, // ASTC_2D_5X4 // Depth formats {GL_DEPTH_COMPONENT32F, GL_DEPTH_COMPONENT, GL_FLOAT, ComponentType::Float, false}, // Z32F @@ -277,7 +279,9 @@ static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType static bool IsPixelFormatASTC(PixelFormat format) { switch (format) { case PixelFormat::ASTC_2D_4X4: + case PixelFormat::ASTC_2D_5X4: case PixelFormat::ASTC_2D_8X8: + case PixelFormat::ASTC_2D_8X5: return true; default: return false; @@ -288,8 +292,12 @@ static std::pair<u32, u32> GetASTCBlockSize(PixelFormat format) { switch (format) { case PixelFormat::ASTC_2D_4X4: return {4, 4}; + case PixelFormat::ASTC_2D_5X4: + return {5, 4}; case PixelFormat::ASTC_2D_8X8: return {8, 8}; + case PixelFormat::ASTC_2D_8X5: + return {8, 5}; default: LOG_CRITICAL(HW_GPU, "Unhandled format: {}", static_cast<u32>(format)); UNREACHABLE(); @@ -323,8 +331,8 @@ static bool IsFormatBCn(PixelFormat format) { } template <bool morton_to_gl, PixelFormat format> -void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::size_t gl_buffer_size, - VAddr addr) { +void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth, u8* gl_buffer, + std::size_t gl_buffer_size, VAddr addr) { constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / CHAR_BIT; constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format); @@ -333,7 +341,7 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::si // pixel values. const u32 tile_size{IsFormatBCn(format) ? 4U : 1U}; const std::vector<u8> data = Tegra::Texture::UnswizzleTexture( - addr, tile_size, bytes_per_pixel, stride, height, block_height); + addr, tile_size, bytes_per_pixel, stride, height, depth, block_height, block_depth); const std::size_t size_to_copy{std::min(gl_buffer_size, data.size())}; memcpy(gl_buffer, data.data(), size_to_copy); } else { @@ -345,7 +353,7 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, std::si } } -static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), +static constexpr std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr), SurfaceParams::MaxPixelFormat> morton_to_gl_fns = { // clang-format off @@ -395,6 +403,8 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), MortonCopy<true, PixelFormat::RG32UI>, MortonCopy<true, PixelFormat::R32UI>, MortonCopy<true, PixelFormat::ASTC_2D_8X8>, + MortonCopy<true, PixelFormat::ASTC_2D_8X5>, + MortonCopy<true, PixelFormat::ASTC_2D_5X4>, MortonCopy<true, PixelFormat::Z32F>, MortonCopy<true, PixelFormat::Z16>, MortonCopy<true, PixelFormat::Z24S8>, @@ -403,7 +413,7 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), // clang-format on }; -static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), +static constexpr std::array<void (*)(u32, u32, u32, u32, u32, u8*, std::size_t, VAddr), SurfaceParams::MaxPixelFormat> gl_to_morton_fns = { // clang-format off @@ -455,6 +465,8 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, std::size_t, VAddr), MortonCopy<false, PixelFormat::RG32UI>, MortonCopy<false, PixelFormat::R32UI>, nullptr, + nullptr, + nullptr, MortonCopy<false, PixelFormat::Z32F>, MortonCopy<false, PixelFormat::Z16>, MortonCopy<false, PixelFormat::Z24S8>, @@ -790,7 +802,9 @@ static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelForma u32 width, u32 height) { switch (pixel_format) { case PixelFormat::ASTC_2D_4X4: - case PixelFormat::ASTC_2D_8X8: { + case PixelFormat::ASTC_2D_8X8: + case PixelFormat::ASTC_2D_8X5: + case PixelFormat::ASTC_2D_5X4: { // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC. u32 block_width{}; u32 block_height{}; @@ -827,36 +841,23 @@ void CachedSurface::LoadGLBuffer() { if (params.is_tiled) { gl_buffer.resize(total_size); + u32 depth = params.depth; + u32 block_depth = params.block_depth; ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}", params.block_width, static_cast<u32>(params.target)); - ASSERT_MSG(params.block_depth == 1, "Block depth is defined as {} on texture type {}", - params.block_depth, static_cast<u32>(params.target)); - // TODO(bunnei): This only unswizzles and copies a 2D texture - we do not yet know how to do - // this for 3D textures, etc. - switch (params.target) { - case SurfaceParams::SurfaceTarget::Texture2D: - // Pass impl. to the fallback code below - break; - case SurfaceParams::SurfaceTarget::Texture2DArray: - case SurfaceParams::SurfaceTarget::TextureCubemap: - for (std::size_t index = 0; index < params.depth; ++index) { - const std::size_t offset{index * copy_size}; - morton_to_gl_fns[static_cast<std::size_t>(params.pixel_format)]( - params.width, params.block_height, params.height, gl_buffer.data() + offset, - copy_size, params.addr + offset); - } - break; - default: - LOG_CRITICAL(HW_GPU, "Unimplemented tiled load for target={}", - static_cast<u32>(params.target)); - UNREACHABLE(); + if (params.target == SurfaceParams::SurfaceTarget::Texture2D) { + // TODO(Blinkhawk): Eliminate this condition once all texture types are implemented. + depth = 1U; + block_depth = 1U; } + const std::size_t size = copy_size * depth; + morton_to_gl_fns[static_cast<std::size_t>(params.pixel_format)]( - params.width, params.block_height, params.height, gl_buffer.data(), copy_size, - params.addr); + params.width, params.block_height, params.height, block_depth, depth, gl_buffer.data(), + size, params.addr); } else { const u8* const texture_src_data_end{texture_src_data + total_size}; gl_buffer.assign(texture_src_data, texture_src_data_end); diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h index 66d98ad4e..0b8ae3eb4 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h +++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h @@ -74,19 +74,21 @@ struct SurfaceParams { RG32UI = 43, R32UI = 44, ASTC_2D_8X8 = 45, + ASTC_2D_8X5 = 46, + ASTC_2D_5X4 = 47, MaxColorFormat, // Depth formats - Z32F = 46, - Z16 = 47, + Z32F = 48, + Z16 = 49, MaxDepthFormat, // DepthStencil formats - Z24S8 = 48, - S8Z24 = 49, - Z32FS8 = 50, + Z24S8 = 50, + S8Z24 = 51, + Z32FS8 = 52, MaxDepthStencilFormat, @@ -220,6 +222,8 @@ struct SurfaceParams { 1, // RG32UI 1, // R32UI 4, // ASTC_2D_8X8 + 4, // ASTC_2D_8X5 + 4, // ASTC_2D_5X4 1, // Z32F 1, // Z16 1, // Z24S8 @@ -282,6 +286,8 @@ struct SurfaceParams { 64, // RG32UI 32, // R32UI 16, // ASTC_2D_8X8 + 32, // ASTC_2D_8X5 + 32, // ASTC_2D_5X4 32, // Z32F 16, // Z16 32, // Z24S8 @@ -553,8 +559,12 @@ struct SurfaceParams { return PixelFormat::BC6H_SF16; case Tegra::Texture::TextureFormat::ASTC_2D_4X4: return PixelFormat::ASTC_2D_4X4; + case Tegra::Texture::TextureFormat::ASTC_2D_5X4: + return PixelFormat::ASTC_2D_5X4; case Tegra::Texture::TextureFormat::ASTC_2D_8X8: return PixelFormat::ASTC_2D_8X8; + case Tegra::Texture::TextureFormat::ASTC_2D_8X5: + return PixelFormat::ASTC_2D_8X5; case Tegra::Texture::TextureFormat::R16_G16: switch (component_type) { case Tegra::Texture::ComponentType::FLOAT: diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 8dfb49507..28dba0084 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -1436,7 +1436,6 @@ private: break; } - case OpCode::Type::Shift: { std::string op_a = regs.GetRegisterAsInteger(instr.gpr8, 0, true); std::string op_b; @@ -1478,7 +1477,6 @@ private: } break; } - case OpCode::Type::ArithmeticIntegerImmediate: { std::string op_a = regs.GetRegisterAsInteger(instr.gpr8); std::string op_b = std::to_string(instr.alu.imm20_32.Value()); @@ -2038,9 +2036,9 @@ private: break; } case OpCode::Id::TEX: { - ASSERT_MSG(instr.tex.array == 0, "TEX arrays unimplemented"); Tegra::Shader::TextureType texture_type{instr.tex.texture_type}; std::string coord; + const bool is_array = instr.tex.array != 0; ASSERT_MSG(!instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); @@ -2055,21 +2053,59 @@ private: switch (num_coordinates) { case 1: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - coord = "float coords = " + x + ';'; + if (is_array) { + const std::string index = regs.GetRegisterAsInteger(instr.gpr8); + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + coord = "vec2 coords = vec2(" + x + ", " + index + ");"; + } else { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + coord = "float coords = " + x + ';'; + } break; } case 2: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + if (is_array) { + const std::string index = regs.GetRegisterAsInteger(instr.gpr8); + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + index + ");"; + } else { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + } break; } case 3: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string z = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; + if (depth_compare) { + if (is_array) { + const std::string index = regs.GetRegisterAsInteger(instr.gpr8); + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string y = regs.GetRegisterAsFloat(instr.gpr20); + const std::string z = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index + + ");"; + } else { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string z = regs.GetRegisterAsFloat(instr.gpr20); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; + } + } else { + if (is_array) { + const std::string index = regs.GetRegisterAsInteger(instr.gpr8); + const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); + const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 3); + coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + index + + ");"; + } else { + const std::string x = regs.GetRegisterAsFloat(instr.gpr8); + const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string z = regs.GetRegisterAsFloat(instr.gpr8.Value() + 2); + coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; + } + } break; } default: @@ -2088,7 +2124,7 @@ private: std::string op_c; const std::string sampler = - GetSampler(instr.sampler, texture_type, false, depth_compare); + GetSampler(instr.sampler, texture_type, is_array, depth_compare); // Add an extra scope and declare the texture coords inside to prevent // overwriting them in case they are used as outputs of the texs instruction. @@ -2108,10 +2144,13 @@ private: } case Tegra::Shader::TextureProcessMode::LB: case Tegra::Shader::TextureProcessMode::LBA: { - if (num_coordinates <= 2) { - op_c = regs.GetRegisterAsFloat(instr.gpr20); + if (depth_compare) { + if (is_array) + op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 2); + else + op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); } else { - op_c = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); + op_c = regs.GetRegisterAsFloat(instr.gpr20); } // TODO: Figure if A suffix changes the equation at all. texture = "texture(" + sampler + ", coords, " + op_c + ')'; @@ -2626,14 +2665,14 @@ private: const std::string pred = GetPredicateCondition(instr.csetp.pred39, instr.csetp.neg_pred39 != 0); const std::string combiner = GetPredicateCombiner(instr.csetp.op); - const std::string controlCode = regs.GetControlCode(instr.csetp.cc); + const std::string control_code = regs.GetControlCode(instr.csetp.cc); if (instr.csetp.pred3 != static_cast<u64>(Pred::UnusedIndex)) { SetPredicate(instr.csetp.pred3, - '(' + controlCode + ") " + combiner + " (" + pred + ')'); + '(' + control_code + ") " + combiner + " (" + pred + ')'); } if (instr.csetp.pred0 != static_cast<u64>(Pred::UnusedIndex)) { SetPredicate(instr.csetp.pred0, - "!(" + controlCode + ") " + combiner + " (" + pred + ')'); + "!(" + control_code + ") " + combiner + " (" + pred + ')'); } break; } diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp index 0d2456b56..18ab723f7 100644 --- a/src/video_core/textures/decoders.cpp +++ b/src/video_core/textures/decoders.cpp @@ -40,72 +40,146 @@ struct alignas(64) SwizzleTable { constexpr auto legacy_swizzle_table = SwizzleTable<8, 64, 1>(); constexpr auto fast_swizzle_table = SwizzleTable<8, 4, 16>(); -static void LegacySwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, - u32 block_height) { +/** + * This function manages ALL the GOBs(Group of Bytes) Inside a single block. + * Instead of going gob by gob, we map the coordinates inside a block and manage from + * those. Block_Width is assumed to be 1. + */ +void PreciseProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, + const u32 y_end, const u32 z_end, const u32 tile_offset, + const u32 xy_block_size, const u32 layer_z, const u32 stride_x, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array<u8*, 2> data_ptrs; - const std::size_t stride = width * bytes_per_pixel; - const std::size_t gobs_in_x = 64; - const std::size_t gobs_in_y = 8; - const std::size_t gobs_size = gobs_in_x * gobs_in_y; - const std::size_t image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x}; - for (std::size_t y = 0; y < height; ++y) { - const std::size_t gob_y_address = - (y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs + - (y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size; - const auto& table = legacy_swizzle_table[y % gobs_in_y]; - for (std::size_t x = 0; x < width; ++x) { - const std::size_t gob_address = - gob_y_address + (x * bytes_per_pixel / gobs_in_x) * gobs_size * block_height; - const std::size_t x2 = x * bytes_per_pixel; - const std::size_t swizzle_offset = gob_address + table[x2 % gobs_in_x]; - const std::size_t pixel_index = (x + y * width) * out_bytes_per_pixel; - - data_ptrs[unswizzle] = swizzled_data + swizzle_offset; - data_ptrs[!unswizzle] = unswizzled_data + pixel_index; - - std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); + u32 z_address = tile_offset; + const u32 gob_size_x = 64; + const u32 gob_size_y = 8; + const u32 gob_size_z = 1; + const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; + for (u32 z = z_start; z < z_end; z++) { + u32 y_address = z_address; + u32 pixel_base = layer_z * z + y_start * stride_x; + for (u32 y = y_start; y < y_end; y++) { + const auto& table = legacy_swizzle_table[y % gob_size_y]; + for (u32 x = x_start; x < x_end; x++) { + const u32 swizzle_offset{y_address + table[x * bytes_per_pixel % gob_size_x]}; + const u32 pixel_index{x * out_bytes_per_pixel + pixel_base}; + data_ptrs[unswizzle] = swizzled_data + swizzle_offset; + data_ptrs[!unswizzle] = unswizzled_data + pixel_index; + std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel); + } + pixel_base += stride_x; + if ((y + 1) % gob_size_y == 0) + y_address += gob_size; } + z_address += xy_block_size; } } -static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, - u32 block_height) { +/** + * This function manages ALL the GOBs(Group of Bytes) Inside a single block. + * Instead of going gob by gob, we map the coordinates inside a block and manage from + * those. Block_Width is assumed to be 1. + */ +void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, + const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end, + const u32 y_end, const u32 z_end, const u32 tile_offset, + const u32 xy_block_size, const u32 layer_z, const u32 stride_x, + const u32 bytes_per_pixel, const u32 out_bytes_per_pixel) { std::array<u8*, 2> data_ptrs; - const std::size_t stride{width * bytes_per_pixel}; - const std::size_t gobs_in_x = 64; - const std::size_t gobs_in_y = 8; - const std::size_t gobs_size = gobs_in_x * gobs_in_y; - const std::size_t image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x}; - const std::size_t copy_size{16}; - for (std::size_t y = 0; y < height; ++y) { - const std::size_t initial_gob = - (y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs + - (y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size; - const std::size_t pixel_base{y * width * out_bytes_per_pixel}; - const auto& table = fast_swizzle_table[y % gobs_in_y]; - for (std::size_t xb = 0; xb < stride; xb += copy_size) { - const std::size_t gob_address{initial_gob + - (xb / gobs_in_x) * gobs_size * block_height}; - const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]}; - const std::size_t out_x = xb * out_bytes_per_pixel / bytes_per_pixel; - const std::size_t pixel_index{out_x + 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 z_address = tile_offset; + const u32 x_startb = x_start * bytes_per_pixel; + const u32 x_endb = x_end * bytes_per_pixel; + const u32 copy_size = 16; + const u32 gob_size_x = 64; + const u32 gob_size_y = 8; + const u32 gob_size_z = 1; + const u32 gob_size = gob_size_x * gob_size_y * gob_size_z; + for (u32 z = z_start; z < z_end; z++) { + u32 y_address = z_address; + u32 pixel_base = layer_z * z + y_start * stride_x; + for (u32 y = y_start; y < y_end; y++) { + const auto& table = fast_swizzle_table[y % gob_size_y]; + for (u32 xb = x_startb; xb < x_endb; xb += copy_size) { + const u32 swizzle_offset{y_address + table[(xb / copy_size) % 4]}; + const u32 out_x = xb * out_bytes_per_pixel / bytes_per_pixel; + const u32 pixel_index{out_x + 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); + } + pixel_base += stride_x; + if ((y + 1) % gob_size_y == 0) + y_address += gob_size; + } + z_address += xy_block_size; + } +} + +/** + * This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue. + * The body of this function takes care of splitting the swizzled texture into blocks, + * and managing the extents of it. Once all the parameters of a single block are obtained, + * the function calls 'ProcessBlock' to process that particular Block. + * + * Documentation for the memory layout and decoding can be found at: + * https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces + */ +template <bool fast> +void SwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, const u32 width, + const u32 height, const u32 depth, const u32 bytes_per_pixel, + const u32 out_bytes_per_pixel, const u32 block_height, const u32 block_depth) { + auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); }; + const u32 stride_x = width * out_bytes_per_pixel; + const u32 layer_z = height * stride_x; + const u32 gob_x_bytes = 64; + const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel; + const u32 gob_elements_y = 8; + const u32 gob_elements_z = 1; + const u32 block_x_elements = gob_elements_x; + const u32 block_y_elements = gob_elements_y * block_height; + const u32 block_z_elements = gob_elements_z * block_depth; + const u32 blocks_on_x = div_ceil(width, block_x_elements); + const u32 blocks_on_y = div_ceil(height, block_y_elements); + const u32 blocks_on_z = div_ceil(depth, block_z_elements); + const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z; + const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z; + const u32 xy_block_size = gob_size * block_height; + const u32 block_size = xy_block_size * block_depth; + u32 tile_offset = 0; + for (u32 zb = 0; zb < blocks_on_z; zb++) { + const u32 z_start = zb * block_z_elements; + const u32 z_end = std::min(depth, z_start + block_z_elements); + for (u32 yb = 0; yb < blocks_on_y; yb++) { + const u32 y_start = yb * block_y_elements; + const u32 y_end = std::min(height, y_start + block_y_elements); + for (u32 xb = 0; xb < blocks_on_x; xb++) { + const u32 x_start = xb * block_x_elements; + const u32 x_end = std::min(width, x_start + block_x_elements); + if (fast) { + FastProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, + layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); + } else { + PreciseProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start, + z_start, x_end, y_end, z_end, tile_offset, xy_block_size, + layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel); + } + tile_offset += block_size; + } } } } -void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) { +void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, + u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, + bool unswizzle, u32 block_height, u32 block_depth) { if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) { - FastSwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data, - unswizzled_data, unswizzle, block_height); + SwizzledData<true>(swizzled_data, unswizzled_data, unswizzle, width, height, depth, + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); } else { - LegacySwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data, - unswizzled_data, unswizzle, block_height); + SwizzledData<false>(swizzled_data, unswizzled_data, unswizzle, width, height, depth, + bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth); } } @@ -126,7 +200,9 @@ u32 BytesPerPixel(TextureFormat format) { case TextureFormat::R32_G32_B32: return 12; case TextureFormat::ASTC_2D_4X4: + case TextureFormat::ASTC_2D_5X4: case TextureFormat::ASTC_2D_8X8: + case TextureFormat::ASTC_2D_8X5: case TextureFormat::A8R8G8B8: case TextureFormat::A2B10G10R10: case TextureFormat::BF10GF11RF11: @@ -153,10 +229,11 @@ u32 BytesPerPixel(TextureFormat format) { } std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, - u32 height, u32 block_height) { - std::vector<u8> 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); + u32 height, u32 depth, u32 block_height, u32 block_depth) { + std::vector<u8> unswizzled_data(width * height * depth * bytes_per_pixel); + CopySwizzledData(width / tile_size, height / tile_size, depth, bytes_per_pixel, bytes_per_pixel, + Memory::GetPointer(address), unswizzled_data.data(), true, block_height, + block_depth); return unswizzled_data; } diff --git a/src/video_core/textures/decoders.h b/src/video_core/textures/decoders.h index 234d250af..aaf316947 100644 --- a/src/video_core/textures/decoders.h +++ b/src/video_core/textures/decoders.h @@ -14,17 +14,14 @@ namespace Tegra::Texture { * Unswizzles a swizzled texture without changing its format. */ std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, - u32 height, u32 block_height = TICEntry::DefaultBlockHeight); - -/** - * Unswizzles a swizzled depth texture without changing its format. - */ -std::vector<u8> UnswizzleDepthTexture(VAddr address, DepthFormat format, u32 width, u32 height, - u32 block_height = TICEntry::DefaultBlockHeight); + u32 height, u32 depth, + u32 block_height = TICEntry::DefaultBlockHeight, + u32 block_depth = TICEntry::DefaultBlockHeight); /// Copies texture data from a buffer and performs swizzling/unswizzling as necessary. -void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, - u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height); +void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel, + u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, + bool unswizzle, u32 block_height, u32 block_depth); /** * Decodes an unswizzled texture into a A8R8G8B8 texture. diff --git a/src/video_core/textures/texture.h b/src/video_core/textures/texture.h index 58d17abcb..5947bd2b9 100644 --- a/src/video_core/textures/texture.h +++ b/src/video_core/textures/texture.h @@ -141,6 +141,7 @@ static_assert(sizeof(TextureHandle) == 4, "TextureHandle has wrong size"); struct TICEntry { static constexpr u32 DefaultBlockHeight = 16; + static constexpr u32 DefaultBlockDepth = 1; union { u32 raw; |