diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_cache.cpp | 45 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_cache.h | 2 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 242 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_gen.cpp | 7 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_gen.h | 3 |
5 files changed, 192 insertions, 107 deletions
diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 038b25c75..aea6bf1af 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.cpp +++ b/src/video_core/renderer_opengl/gl_shader_cache.cpp @@ -2,7 +2,9 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <boost/functional/hash.hpp> #include "common/assert.h" +#include "common/hash.h" #include "core/core.h" #include "core/memory.h" #include "video_core/engines/maxwell_3d.h" @@ -66,14 +68,17 @@ CachedShader::CachedShader(VAddr addr, Maxwell::ShaderProgram program_type) // stage here. setup.SetProgramB(GetShaderCode(GetShaderAddress(Maxwell::ShaderProgram::VertexB))); case Maxwell::ShaderProgram::VertexB: + CalculateProperties(); program_result = GLShader::GenerateVertexShader(setup); gl_type = GL_VERTEX_SHADER; break; case Maxwell::ShaderProgram::Geometry: + CalculateProperties(); program_result = GLShader::GenerateGeometryShader(setup); gl_type = GL_GEOMETRY_SHADER; break; case Maxwell::ShaderProgram::Fragment: + CalculateProperties(); program_result = GLShader::GenerateFragmentShader(setup); gl_type = GL_FRAGMENT_SHADER; break; @@ -140,6 +145,46 @@ GLuint CachedShader::LazyGeometryProgram(OGLProgram& target_program, return target_program.handle; }; +static bool IsSchedInstruction(std::size_t offset, std::size_t main_offset) { + // sched instructions appear once every 4 instructions. + static constexpr std::size_t SchedPeriod = 4; + const std::size_t absolute_offset = offset - main_offset; + return (absolute_offset % SchedPeriod) == 0; +} + +static std::size_t CalculateProgramSize(const GLShader::ProgramCode& program) { + constexpr std::size_t start_offset = 10; + std::size_t offset = start_offset; + std::size_t size = start_offset * sizeof(u64); + while (offset < program.size()) { + const u64 inst = program[offset]; + if (!IsSchedInstruction(offset, start_offset)) { + if (inst == 0 || (inst >> 52) == 0x50b) { + break; + } + } + size += sizeof(inst); + offset++; + } + return size; +} + +void CachedShader::CalculateProperties() { + setup.program.real_size = CalculateProgramSize(setup.program.code); + setup.program.real_size_b = 0; + setup.program.unique_identifier = Common::CityHash64( + reinterpret_cast<const char*>(setup.program.code.data()), setup.program.real_size); + if (program_type == Maxwell::ShaderProgram::VertexA) { + std::size_t seed = 0; + boost::hash_combine(seed, setup.program.unique_identifier); + setup.program.real_size_b = CalculateProgramSize(setup.program.code_b); + const u64 identifier_b = Common::CityHash64( + reinterpret_cast<const char*>(setup.program.code_b.data()), setup.program.real_size_b); + boost::hash_combine(seed, identifier_b); + setup.program.unique_identifier = static_cast<u64>(seed); + } +} + ShaderCacheOpenGL::ShaderCacheOpenGL(RasterizerOpenGL& rasterizer) : RasterizerCache{rasterizer} {} Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) { diff --git a/src/video_core/renderer_opengl/gl_shader_cache.h b/src/video_core/renderer_opengl/gl_shader_cache.h index 08f470de3..b4ef6030d 100644 --- a/src/video_core/renderer_opengl/gl_shader_cache.h +++ b/src/video_core/renderer_opengl/gl_shader_cache.h @@ -81,6 +81,8 @@ private: GLuint LazyGeometryProgram(OGLProgram& target_program, const std::string& glsl_topology, u32 max_vertices, const std::string& debug_name); + void CalculateProperties(); + VAddr addr; std::size_t shader_length; Maxwell::ShaderProgram program_type; diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 4fc09cac6..a5cfa0070 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -928,7 +928,7 @@ private: case Attribute::Index::FrontFacing: // TODO(Subv): Find out what the values are for the other elements. ASSERT(stage == Maxwell3D::Regs::ShaderStage::Fragment); - return "vec4(0, 0, 0, uintBitsToFloat(gl_FrontFacing ? 1 : 0))"; + return "vec4(0, 0, 0, intBitsToFloat(gl_FrontFacing ? -1 : 0))"; default: const u32 index{static_cast<u32>(attribute) - static_cast<u32>(Attribute::Index::Attribute_0)}; @@ -1681,7 +1681,7 @@ private: for (size_t i = 0; i < coord_count; ++i) { const bool last = (i == (coord_count - 1)) && (coord_count > 1); coord += regs.GetRegisterAsFloat(last ? last_coord_register : coord_register + i); - if (!last) { + if (i < coord_count - 1) { coord += ','; } } @@ -1702,6 +1702,99 @@ private: is_array, (coord_count > 2 ? 1 : 0))); } + std::pair<std::string, std::string> GetTLD4Code(const Instruction& instr, + const Tegra::Shader::TextureType texture_type, + const bool depth_compare, const bool is_array) { + + const size_t coord_count = TextureCoordinates(texture_type); + const size_t total_coord_count = coord_count + (is_array ? 1 : 0); + const size_t total_reg_count = total_coord_count + (depth_compare ? 1 : 0); + + constexpr std::array<const char*, 5> coord_container{ + {"", "", "vec2 coord = vec2(", "vec3 coord = vec3(", "vec4 coord = vec4("}}; + + // If enabled arrays index is always stored in the gpr8 field + const u64 array_register = instr.gpr8.Value(); + // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used + const u64 coord_register = array_register + (is_array ? 1 : 0); + + std::string coord = coord_container[total_coord_count]; + for (size_t i = 0; i < coord_count;) { + coord += regs.GetRegisterAsFloat(coord_register + i); + ++i; + if (i != coord_count) { + coord += ','; + } + } + + if (is_array) { + coord += ',' + regs.GetRegisterAsInteger(array_register); + } + coord += ");"; + + const std::string sampler = + GetSampler(instr.sampler, texture_type, is_array, depth_compare); + + std::string texture = "textureGather(" + sampler + ", coord, "; + if (depth_compare) { + // Depth is always stored in the register signaled by gpr20 + texture += regs.GetRegisterAsFloat(instr.gpr20.Value()) + ')'; + } else { + texture += std::to_string(instr.tld4.component) + ')'; + } + return std::make_pair(coord, texture); + } + + std::pair<std::string, std::string> GetTLDSCode(const Instruction& instr, + const Tegra::Shader::TextureType texture_type, + const bool is_array) { + + const size_t coord_count = TextureCoordinates(texture_type); + const size_t total_coord_count = coord_count + (is_array ? 1 : 0); + const bool lod_enabled = + instr.tlds.GetTextureProcessMode() == Tegra::Shader::TextureProcessMode::LL; + + constexpr std::array<const char*, 4> coord_container{ + {"", "int coord = (", "ivec2 coord = ivec2(", "ivec3 coord = ivec3("}}; + + std::string coord = coord_container[total_coord_count]; + + // If enabled arrays index is always stored in the gpr8 field + const u64 array_register = instr.gpr8.Value(); + + // if is array gpr20 is used + const u64 coord_register = is_array ? instr.gpr20.Value() : instr.gpr8.Value(); + + const u64 last_coord_register = + ((coord_count > 2) || (coord_count == 2 && !lod_enabled)) && !is_array + ? static_cast<u64>(instr.gpr20.Value()) + : coord_register + 1; + + for (size_t i = 0; i < coord_count; ++i) { + const bool last = (i == (coord_count - 1)) && (coord_count > 1); + coord += regs.GetRegisterAsInteger(last ? last_coord_register : coord_register + i); + if (i < coord_count - 1) { + coord += ','; + } + } + if (is_array) { + coord += ',' + regs.GetRegisterAsInteger(array_register); + } + coord += ");"; + + const std::string sampler = GetSampler(instr.sampler, texture_type, is_array, false); + + std::string texture = "texelFetch(" + sampler + ", coords"; + + if (lod_enabled) { + // When lod is used always is in grp20 + texture += ", " + regs.GetRegisterAsInteger(instr.gpr20) + ')'; + } else { + texture += ", 0)"; + } + return std::make_pair(coord, texture); + } + /** * Compiles a single instruction from Tegra to GLSL. * @param offset the offset of the Tegra shader instruction. @@ -2825,9 +2918,6 @@ private: const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()}; const bool is_array{instr.tlds.IsArrayTexture()}; - ASSERT(texture_type == Tegra::Shader::TextureType::Texture2D); - ASSERT(is_array == false); - UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), @@ -2835,54 +2925,16 @@ private: UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::MZ), "MZ is not implemented"); - u32 extra_op_offset = 0; - - ShaderScopedScope scope = shader.Scope(); - - switch (texture_type) { - case Tegra::Shader::TextureType::Texture1D: { - const std::string x = regs.GetRegisterAsInteger(instr.gpr8); - shader.AddLine("float coords = " + x + ';'); - break; - } - case Tegra::Shader::TextureType::Texture2D: { - UNIMPLEMENTED_IF_MSG(is_array, "Unhandled 2d array texture"); - - const std::string x = regs.GetRegisterAsInteger(instr.gpr8); - const std::string y = regs.GetRegisterAsInteger(instr.gpr20); - // shader.AddLine("ivec2 coords = ivec2(" + x + ", " + y + ");"); - shader.AddLine("ivec2 coords = ivec2(" + x + ", " + y + ");"); - extra_op_offset = 1; - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<u32>(texture_type)); - } - const std::string sampler = - GetSampler(instr.sampler, texture_type, is_array, false); + const auto [coord, texture] = GetTLDSCode(instr, texture_type, is_array); - const std::string texture = [&]() { - switch (instr.tlds.GetTextureProcessMode()) { - case Tegra::Shader::TextureProcessMode::LZ: - return "texelFetch(" + sampler + ", coords, 0)"; - case Tegra::Shader::TextureProcessMode::LL: - shader.AddLine( - "float lod = " + - regs.GetRegisterAsInteger(instr.gpr20.Value() + extra_op_offset) + ';'); - return "texelFetch(" + sampler + ", coords, lod)"; - default: - UNIMPLEMENTED_MSG("Unhandled texture process mode {}", - static_cast<u32>(instr.tlds.GetTextureProcessMode())); - return "texelFetch(" + sampler + ", coords, 0)"; - } - }(); + const auto scope = shader.Scope(); - WriteTexsInstructionFloat(instr, texture); + shader.AddLine(coord); + shader.AddLine("vec4 texture_tmp = " + texture + ';'); + WriteTexsInstructionFloat(instr, "texture_tmp"); break; } case OpCode::Id::TLD4: { - ASSERT(instr.tld4.texture_type == Tegra::Shader::TextureType::Texture2D); - ASSERT(instr.tld4.array == 0); UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); @@ -2892,56 +2944,29 @@ private: "NDV is not implemented"); UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::PTP), "PTP is not implemented"); + + auto texture_type = instr.tld4.texture_type.Value(); const bool depth_compare = instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); - auto texture_type = instr.tld4.texture_type.Value(); - u32 num_coordinates = TextureCoordinates(texture_type); - if (depth_compare) - num_coordinates += 1; - - const auto scope = shader.Scope(); + const bool is_array = instr.tld4.array != 0; - switch (num_coordinates) { - case 2: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - shader.AddLine("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.gpr8.Value() + 2); - shader.AddLine("vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"); - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled coordinates number {}", - static_cast<u32>(num_coordinates)); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); - texture_type = Tegra::Shader::TextureType::Texture2D; - } + const auto [coord, texture] = + GetTLD4Code(instr, texture_type, depth_compare, is_array); - const std::string sampler = - GetSampler(instr.sampler, texture_type, false, depth_compare); + const auto scope = shader.Scope(); - const std::string texture = "textureGather(" + sampler + ", coords, " + - std::to_string(instr.tld4.component) + ')'; + shader.AddLine(coord); + std::size_t dest_elem{}; - if (depth_compare) { - regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); - } else { - std::size_t dest_elem{}; - for (std::size_t elem = 0; elem < 4; ++elem) { - if (!instr.tex.IsComponentEnabled(elem)) { - // Skip disabled components - continue; - } - regs.SetRegisterToFloat(instr.gpr0, elem, texture, 1, 4, false, dest_elem); - ++dest_elem; + shader.AddLine("vec4 texture_tmp = " + texture + ';'); + for (std::size_t elem = 0; elem < 4; ++elem) { + if (!instr.tex.IsComponentEnabled(elem)) { + // Skip disabled components + continue; } + regs.SetRegisterToFloat(instr.gpr0, elem, "texture_tmp", 1, 4, false, + dest_elem); + ++dest_elem; } break; } @@ -2955,28 +2980,31 @@ private: const auto scope = shader.Scope(); + std::string coords; + const bool depth_compare = instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); - const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); - const std::string op_b = regs.GetRegisterAsFloat(instr.gpr20); - // TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction. + const std::string sampler = GetSampler( instr.sampler, Tegra::Shader::TextureType::Texture2D, false, depth_compare); - if (depth_compare) { - // Note: TLD4S coordinate encoding works just like TEXS's - const std::string op_y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - shader.AddLine("vec3 coords = vec3(" + op_a + ", " + op_y + ", " + op_b + ");"); - } else { - shader.AddLine("vec2 coords = vec2(" + op_a + ", " + op_b + ");"); - } - std::string texture = "textureGather(" + sampler + ", coords, " + - std::to_string(instr.tld4s.component) + ')'; - if (depth_compare) { - texture = "vec4(" + texture + ')'; - } + const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); + coords = "vec2 coords = vec2(" + op_a + ", "; + std::string texture = "textureGather(" + sampler + ", coords, "; - WriteTexsInstructionFloat(instr, texture); + if (!depth_compare) { + const std::string op_b = regs.GetRegisterAsFloat(instr.gpr20); + coords += op_b + ");"; + texture += std::to_string(instr.tld4s.component) + ')'; + } else { + const std::string op_b = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string op_c = regs.GetRegisterAsFloat(instr.gpr20); + coords += op_b + ");"; + texture += op_c + ')'; + } + shader.AddLine(coords); + shader.AddLine("vec4 texture_tmp = " + texture + ';'); + WriteTexsInstructionFloat(instr, "texture_tmp"); break; } case OpCode::Id::TXQ: { diff --git a/src/video_core/renderer_opengl/gl_shader_gen.cpp b/src/video_core/renderer_opengl/gl_shader_gen.cpp index 23ed91e27..5d0819dc5 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.cpp +++ b/src/video_core/renderer_opengl/gl_shader_gen.cpp @@ -2,6 +2,7 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <fmt/format.h> #include "common/assert.h" #include "video_core/engines/maxwell_3d.h" #include "video_core/renderer_opengl/gl_shader_decompiler.h" @@ -16,6 +17,8 @@ static constexpr u32 PROGRAM_OFFSET{10}; ProgramResult GenerateVertexShader(const ShaderSetup& setup) { std::string out = "#version 430 core\n"; out += "#extension GL_ARB_separate_shader_objects : enable\n\n"; + const std::string id = fmt::format("{:016x}", setup.program.unique_identifier); + out += "// Shader Unique Id: VS" + id + "\n\n"; out += Decompiler::GetCommonDeclarations(); out += R"( @@ -84,6 +87,8 @@ void main() { ProgramResult GenerateGeometryShader(const ShaderSetup& setup) { // Version is intentionally skipped in shader generation, it's added by the lazy compilation. std::string out = "#extension GL_ARB_separate_shader_objects : enable\n\n"; + const std::string id = fmt::format("{:016x}", setup.program.unique_identifier); + out += "// Shader Unique Id: GS" + id + "\n\n"; out += Decompiler::GetCommonDeclarations(); out += "bool exec_geometry();\n"; @@ -117,6 +122,8 @@ void main() { ProgramResult GenerateFragmentShader(const ShaderSetup& setup) { std::string out = "#version 430 core\n"; out += "#extension GL_ARB_separate_shader_objects : enable\n\n"; + const std::string id = fmt::format("{:016x}", setup.program.unique_identifier); + out += "// Shader Unique Id: FS" + id + "\n\n"; out += Decompiler::GetCommonDeclarations(); out += "bool exec_fragment();\n"; diff --git a/src/video_core/renderer_opengl/gl_shader_gen.h b/src/video_core/renderer_opengl/gl_shader_gen.h index 4fa6d7612..fcc20d3b4 100644 --- a/src/video_core/renderer_opengl/gl_shader_gen.h +++ b/src/video_core/renderer_opengl/gl_shader_gen.h @@ -177,6 +177,9 @@ struct ShaderSetup { struct { ProgramCode code; ProgramCode code_b; // Used for dual vertex shaders + u64 unique_identifier; + std::size_t real_size; + std::size_t real_size_b; } program; /// Used in scenarios where we have a dual vertex shaders |