diff options
Diffstat (limited to 'src/video_core/renderer_opengl')
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer.cpp | 116 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_rasterizer.h | 6 | ||||
-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 | 903 | ||||
-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 |
7 files changed, 563 insertions, 519 deletions
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp index 9e93bd609..2b29fc45f 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.cpp +++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp @@ -79,6 +79,26 @@ struct DrawParameters { } }; +struct FramebufferCacheKey { + bool is_single_buffer = false; + bool stencil_enable = false; + + std::array<GLenum, Maxwell::NumRenderTargets> color_attachments{}; + std::array<GLuint, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> colors{}; + u32 colors_count = 0; + + GLuint zeta = 0; + + auto Tie() const { + return std::tie(is_single_buffer, stencil_enable, color_attachments, colors, colors_count, + zeta); + } + + bool operator<(const FramebufferCacheKey& rhs) const { + return Tie() < rhs.Tie(); + } +}; + RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo& info) : res_cache{*this}, shader_cache{*this}, emu_window{window}, screen_info{info}, buffer_cache(*this, STREAM_BUFFER_SIZE) { @@ -90,9 +110,6 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, ScreenInfo OpenGLState::ApplyDefaultState(); - // Create render framebuffer - framebuffer.Create(); - shader_program_manager = std::make_unique<GLShader::ProgramManager>(); state.draw.shader_program = 0; state.Apply(); @@ -361,6 +378,44 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) { SyncClipEnabled(clip_distances); } +void RasterizerOpenGL::SetupCachedFramebuffer(const FramebufferCacheKey& fbkey, + OpenGLState& current_state) { + const auto [entry, is_cache_miss] = framebuffer_cache.try_emplace(fbkey); + auto& framebuffer = entry->second; + + if (is_cache_miss) + framebuffer.Create(); + + current_state.draw.draw_framebuffer = framebuffer.handle; + current_state.ApplyFramebufferState(); + + if (!is_cache_miss) + return; + + if (fbkey.is_single_buffer) { + if (fbkey.color_attachments[0] != GL_NONE) { + glFramebufferTexture(GL_DRAW_FRAMEBUFFER, fbkey.color_attachments[0], fbkey.colors[0], + 0); + } + glDrawBuffer(fbkey.color_attachments[0]); + } else { + for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) { + if (fbkey.colors[index]) { + glFramebufferTexture(GL_DRAW_FRAMEBUFFER, + GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index), + fbkey.colors[index], 0); + } + } + glDrawBuffers(fbkey.colors_count, fbkey.color_attachments.data()); + } + + if (fbkey.zeta) { + GLenum zeta_attachment = + fbkey.stencil_enable ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT; + glFramebufferTexture(GL_DRAW_FRAMEBUFFER, zeta_attachment, fbkey.zeta, 0); + } +} + std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const { const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs; @@ -444,10 +499,10 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us UNIMPLEMENTED_IF(regs.rt_separate_frag_data != 0); // Bind the framebuffer surfaces - current_state.draw.draw_framebuffer = framebuffer.handle; - current_state.ApplyFramebufferState(); current_state.framebuffer_srgb.enabled = regs.framebuffer_srgb != 0; + FramebufferCacheKey fbkey; + if (using_color_fb) { if (single_color_target) { // Used when just a single color attachment is enabled, e.g. for clearing a color buffer @@ -463,14 +518,12 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us state.framebuffer_srgb.enabled |= color_surface->GetSurfaceParams().srgb_conversion; } - glFramebufferTexture2D( - GL_DRAW_FRAMEBUFFER, - GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target), GL_TEXTURE_2D, - color_surface != nullptr ? color_surface->Texture().handle : 0, 0); - glDrawBuffer(GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target)); + fbkey.is_single_buffer = true; + fbkey.color_attachments[0] = + GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(*single_color_target); + fbkey.colors[0] = color_surface != nullptr ? color_surface->Texture().handle : 0; } else { // Multiple color attachments are enabled - std::array<GLenum, Maxwell::NumRenderTargets> buffers; for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) { Surface color_surface = res_cache.GetColorBufferSurface(index, preserve_contents); @@ -485,22 +538,17 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us color_surface->GetSurfaceParams().srgb_conversion; } - buffers[index] = GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index); - glFramebufferTexture2D( - GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index), - GL_TEXTURE_2D, color_surface != nullptr ? color_surface->Texture().handle : 0, - 0); + fbkey.color_attachments[index] = + GL_COLOR_ATTACHMENT0 + regs.rt_control.GetMap(index); + fbkey.colors[index] = + color_surface != nullptr ? color_surface->Texture().handle : 0; } - glDrawBuffers(regs.rt_control.count, buffers.data()); + fbkey.is_single_buffer = false; + fbkey.colors_count = regs.rt_control.count; } } else { - // No color attachments are enabled - zero out all of them - for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) { - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, - GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index), GL_TEXTURE_2D, - 0, 0); - } - glDrawBuffer(GL_NONE); + // No color attachments are enabled - leave them as zero + fbkey.is_single_buffer = true; } if (depth_surface) { @@ -508,22 +556,12 @@ void RasterizerOpenGL::ConfigureFramebuffers(OpenGLState& current_state, bool us // the shader doesn't actually write to it. depth_surface->MarkAsModified(true, res_cache); - if (regs.stencil_enable) { - // Attach both depth and stencil - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, - depth_surface->Texture().handle, 0); - } else { - // Attach depth - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, - depth_surface->Texture().handle, 0); - // Clear stencil attachment - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0); - } - } else { - // Clear both depth and stencil attachment - glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, - 0); + fbkey.zeta = depth_surface->Texture().handle; + fbkey.stencil_enable = regs.stencil_enable; } + + SetupCachedFramebuffer(fbkey, current_state); + SyncViewport(current_state); } diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h index 988fa3e27..8a891ffc7 100644 --- a/src/video_core/renderer_opengl/gl_rasterizer.h +++ b/src/video_core/renderer_opengl/gl_rasterizer.h @@ -40,6 +40,7 @@ namespace OpenGL { struct ScreenInfo; struct DrawParameters; +struct FramebufferCacheKey; class RasterizerOpenGL : public VideoCore::RasterizerInterface { public: @@ -195,11 +196,12 @@ private: OGLVertexArray> vertex_array_cache; + std::map<FramebufferCacheKey, OGLFramebuffer> framebuffer_cache; + std::array<SamplerInfo, Tegra::Engines::Maxwell3D::Regs::NumTextureSamplers> texture_samplers; static constexpr std::size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024; OGLBufferCache buffer_cache; - OGLFramebuffer framebuffer; PrimitiveAssembler primitive_assembler{buffer_cache}; GLint uniform_buffer_alignment; @@ -214,6 +216,8 @@ private: void SetupShaders(GLenum primitive_mode); + void SetupCachedFramebuffer(const FramebufferCacheKey& fbkey, OpenGLState& current_state); + enum class AccelDraw { Disabled, Arrays, Indexed }; AccelDraw accelerate_draw = AccelDraw::Disabled; diff --git a/src/video_core/renderer_opengl/gl_shader_cache.cpp b/src/video_core/renderer_opengl/gl_shader_cache.cpp index 038b25c75..a4265f498 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(u32 offset, u32 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) { + const 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 += 8; + 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 0c1632bd1..e7057a9cb 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -50,6 +50,14 @@ public: using std::runtime_error::runtime_error; }; +/// Generates code to use for a swizzle operation. +static std::string GetSwizzle(u64 elem) { + ASSERT(elem <= 3); + std::string swizzle = "."; + swizzle += "xyzw"[elem]; + return swizzle; +} + /// Translate topology static std::string GetTopologyName(Tegra::Shader::OutputTopology topology) { switch (topology) { @@ -201,14 +209,53 @@ private: } }; +template <typename T> +class ShaderScopedScope { +public: + explicit ShaderScopedScope(T& writer, std::string_view begin_expr, std::string end_expr) + : writer(writer), end_expr(std::move(end_expr)) { + + if (begin_expr.empty()) { + writer.AddLine('{'); + } else { + writer.AddExpression(begin_expr); + writer.AddLine(" {"); + } + ++writer.scope; + } + + ShaderScopedScope(const ShaderScopedScope&) = delete; + + ~ShaderScopedScope() { + --writer.scope; + if (end_expr.empty()) { + writer.AddLine('}'); + } else { + writer.AddExpression("} "); + writer.AddExpression(end_expr); + writer.AddLine(';'); + } + } + + ShaderScopedScope& operator=(const ShaderScopedScope&) = delete; + +private: + T& writer; + std::string end_expr; +}; + class ShaderWriter { public: - void AddLine(std::string_view text) { + void AddExpression(std::string_view text) { DEBUG_ASSERT(scope >= 0); if (!text.empty()) { AppendIndentation(); } shader_source += text; + } + + void AddLine(std::string_view text) { + AddExpression(text); AddNewLine(); } @@ -228,6 +275,11 @@ public: return std::move(shader_source); } + ShaderScopedScope<ShaderWriter> Scope(std::string_view begin_expr = {}, + std::string end_expr = {}) { + return ShaderScopedScope(*this, begin_expr, end_expr); + } + int scope = 0; private: @@ -311,7 +363,7 @@ public: // Default - do nothing return value; default: - UNIMPLEMENTED_MSG("Unimplemented conversion size: {}", static_cast<u32>(size)); + UNREACHABLE_MSG("Unimplemented conversion size: {}", static_cast<u32>(size)); } } @@ -816,14 +868,12 @@ private: } if (precise && stage != Maxwell3D::Regs::ShaderStage::Fragment) { - shader.AddLine('{'); - ++shader.scope; + const auto scope = shader.Scope(); + // This avoids optimizations of constant propagation and keeps the code as the original // Sadly using the precise keyword causes "linking" errors on fragment shaders. shader.AddLine("precise float tmp = " + src + ';'); shader.AddLine(dest + " = tmp;"); - --shader.scope; - shader.AddLine('}'); } else { shader.AddLine(dest + " = " + src + ';'); } @@ -962,14 +1012,6 @@ private: } } - /// Generates code to use for a swizzle operation. - static std::string GetSwizzle(u64 elem) { - ASSERT(elem <= 3); - std::string swizzle = "."; - swizzle += "xyzw"[elem]; - return swizzle; - } - ShaderWriter& shader; ShaderWriter& declarations; std::vector<GLSLRegister> regs; @@ -1301,15 +1343,7 @@ private: regs.SetRegisterToInteger(dest, true, 0, result, 1, 1); } - void WriteTexsInstruction(const Instruction& instr, const std::string& coord, - const std::string& texture) { - // 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. - shader.AddLine('{'); - ++shader.scope; - shader.AddLine(coord); - shader.AddLine("vec4 texture_tmp = " + texture + ';'); - + void WriteTexsInstructionFloat(const Instruction& instr, const std::string& texture) { // TEXS has two destination registers and a swizzle. The first two elements in the swizzle // go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1 @@ -1321,19 +1355,49 @@ private: if (written_components < 2) { // Write the first two swizzle components to gpr0 and gpr0+1 - regs.SetRegisterToFloat(instr.gpr0, component, "texture_tmp", 1, 4, false, + regs.SetRegisterToFloat(instr.gpr0, component, texture, 1, 4, false, written_components % 2); } else { ASSERT(instr.texs.HasTwoDestinations()); // Write the rest of the swizzle components to gpr28 and gpr28+1 - regs.SetRegisterToFloat(instr.gpr28, component, "texture_tmp", 1, 4, false, + regs.SetRegisterToFloat(instr.gpr28, component, texture, 1, 4, false, written_components % 2); } ++written_components; } - --shader.scope; - shader.AddLine('}'); + } + + void WriteTexsInstructionHalfFloat(const Instruction& instr, const std::string& texture) { + // TEXS.F16 destionation registers are packed in two registers in pairs (just like any half + // float instruction). + + std::array<std::string, 4> components; + u32 written_components = 0; + + for (u32 component = 0; component < 4; ++component) { + if (!instr.texs.IsComponentEnabled(component)) + continue; + components[written_components++] = texture + GetSwizzle(component); + } + if (written_components == 0) + return; + + const auto BuildComponent = [&](std::string low, std::string high, bool high_enabled) { + return "vec2(" + low + ", " + (high_enabled ? high : "0") + ')'; + }; + + regs.SetRegisterToHalfFloat( + instr.gpr0, 0, BuildComponent(components[0], components[1], written_components > 1), + Tegra::Shader::HalfMerge::H0_H1, 1, 1); + + if (written_components > 2) { + ASSERT(instr.texs.HasTwoDestinations()); + regs.SetRegisterToHalfFloat( + instr.gpr28, 0, + BuildComponent(components[2], components[3], written_components > 3), + Tegra::Shader::HalfMerge::H0_H1, 1, 1); + } } static u32 TextureCoordinates(Tegra::Shader::TextureType texture_type) { @@ -1356,12 +1420,10 @@ private: * top. */ void EmitPushToFlowStack(u32 target) { - shader.AddLine('{'); - ++shader.scope; + const auto scope = shader.Scope(); + shader.AddLine("flow_stack[flow_stack_top] = " + std::to_string(target) + "u;"); shader.AddLine("flow_stack_top++;"); - --shader.scope; - shader.AddLine('}'); } /* @@ -1369,13 +1431,11 @@ private: * popped address and decrementing the stack top. */ void EmitPopFromFlowStack() { - shader.AddLine('{'); - ++shader.scope; + const auto scope = shader.Scope(); + shader.AddLine("flow_stack_top--;"); shader.AddLine("jmp_to = flow_stack[flow_stack_top];"); shader.AddLine("break;"); - --shader.scope; - shader.AddLine('}'); } /// Writes the output values from a fragment shader to the corresponding GLSL output variables. @@ -1487,6 +1547,254 @@ private: } } + std::pair<size_t, std::string> ValidateAndGetCoordinateElement( + const Tegra::Shader::TextureType texture_type, const bool depth_compare, + const bool is_array, const bool lod_bias_enabled, size_t max_coords, size_t max_inputs) { + const size_t coord_count = TextureCoordinates(texture_type); + + size_t total_coord_count = coord_count + (is_array ? 1 : 0) + (depth_compare ? 1 : 0); + const size_t total_reg_count = total_coord_count + (lod_bias_enabled ? 1 : 0); + if (total_coord_count > max_coords || total_reg_count > max_inputs) { + UNIMPLEMENTED_MSG("Unsupported Texture operation"); + total_coord_count = std::min(total_coord_count, max_coords); + } + // 1D.DC opengl is using a vec3 but 2nd component is ignored later. + total_coord_count += + (depth_compare && !is_array && texture_type == Tegra::Shader::TextureType::Texture1D) + ? 1 + : 0; + + constexpr std::array<const char*, 5> coord_container{ + {"", "float coord = (", "vec2 coord = vec2(", "vec3 coord = vec3(", + "vec4 coord = vec4("}}; + + return std::pair<size_t, std::string>(coord_count, coord_container[total_coord_count]); + } + + std::string GetTextureCode(const Tegra::Shader::Instruction& instr, + const Tegra::Shader::TextureType texture_type, + const Tegra::Shader::TextureProcessMode process_mode, + const bool depth_compare, const bool is_array, + const size_t bias_offset) { + + if ((texture_type == Tegra::Shader::TextureType::Texture3D && + (is_array || depth_compare)) || + (texture_type == Tegra::Shader::TextureType::TextureCube && is_array && + depth_compare)) { + UNIMPLEMENTED_MSG("This method is not supported."); + } + + const std::string sampler = + GetSampler(instr.sampler, texture_type, is_array, depth_compare); + + const bool lod_needed = process_mode == Tegra::Shader::TextureProcessMode::LZ || + process_mode == Tegra::Shader::TextureProcessMode::LL || + process_mode == Tegra::Shader::TextureProcessMode::LLA; + + const bool gl_lod_supported = !( + (texture_type == Tegra::Shader::TextureType::Texture2D && is_array && depth_compare) || + (texture_type == Tegra::Shader::TextureType::TextureCube && !is_array && + depth_compare)); + + const std::string read_method = lod_needed && gl_lod_supported ? "textureLod(" : "texture("; + std::string texture = read_method + sampler + ", coord"; + + if (process_mode != Tegra::Shader::TextureProcessMode::None) { + if (process_mode == Tegra::Shader::TextureProcessMode::LZ) { + if (gl_lod_supported) { + texture += ", 0"; + } else { + // Lod 0 is emulated by a big negative bias + // in scenarios that are not supported by glsl + texture += ", -1000"; + } + } else { + // If present, lod or bias are always stored in the register indexed by the + // gpr20 + // field with an offset depending on the usage of the other registers + texture += ',' + regs.GetRegisterAsFloat(instr.gpr20.Value() + bias_offset); + } + } + texture += ")"; + return texture; + } + + std::pair<std::string, std::string> GetTEXCode( + const Instruction& instr, const Tegra::Shader::TextureType texture_type, + const Tegra::Shader::TextureProcessMode process_mode, const bool depth_compare, + const bool is_array) { + const bool lod_bias_enabled = (process_mode != Tegra::Shader::TextureProcessMode::None && + process_mode != Tegra::Shader::TextureProcessMode::LZ); + + const auto [coord_count, coord_dcl] = ValidateAndGetCoordinateElement( + texture_type, depth_compare, is_array, lod_bias_enabled, 4, 5); + // 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_dcl; + for (size_t i = 0; i < coord_count;) { + coord += regs.GetRegisterAsFloat(coord_register + i); + ++i; + if (i != coord_count) { + coord += ','; + } + } + // 1D.DC in opengl the 2nd component is ignored. + if (depth_compare && !is_array && texture_type == Tegra::Shader::TextureType::Texture1D) { + coord += ",0.0"; + } + if (depth_compare) { + // Depth is always stored in the register signaled by gpr20 + // or in the next register if lod or bias are used + const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0); + coord += ',' + regs.GetRegisterAsFloat(depth_register); + } + if (is_array) { + coord += ',' + regs.GetRegisterAsInteger(array_register); + } + coord += ");"; + return std::make_pair( + coord, GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, 0)); + } + + std::pair<std::string, std::string> GetTEXSCode( + const Instruction& instr, const Tegra::Shader::TextureType texture_type, + const Tegra::Shader::TextureProcessMode process_mode, const bool depth_compare, + const bool is_array) { + const bool lod_bias_enabled = (process_mode != Tegra::Shader::TextureProcessMode::None && + process_mode != Tegra::Shader::TextureProcessMode::LZ); + + const auto [coord_count, coord_dcl] = ValidateAndGetCoordinateElement( + texture_type, depth_compare, is_array, lod_bias_enabled, 4, 4); + // If enabled arrays index is always stored in the gpr8 field + const u64 array_register = instr.gpr8.Value(); + // First coordinate index is stored in gpr8 field or (gpr8 + 1) when arrays are used + const u64 coord_register = array_register + (is_array ? 1 : 0); + const u64 last_coord_register = + (is_array || !(lod_bias_enabled || depth_compare) || (coord_count > 2)) + ? static_cast<u64>(instr.gpr20.Value()) + : coord_register + 1; + + std::string coord = coord_dcl; + 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 (i < coord_count - 1) { + coord += ','; + } + } + + if (depth_compare) { + // Depth is always stored in the register signaled by gpr20 + // or in the next register if lod or bias are used + const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0); + coord += ',' + regs.GetRegisterAsFloat(depth_register); + } + if (is_array) { + coord += ',' + regs.GetRegisterAsInteger(array_register); + } + coord += ");"; + + return std::make_pair(coord, + GetTextureCode(instr, texture_type, process_mode, depth_compare, + 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. @@ -2287,8 +2595,7 @@ private: UNIMPLEMENTED_IF(instr.conversion.selector); UNIMPLEMENTED_IF_MSG(instr.generates_cc, "Condition codes generation in I2F is not implemented"); - - std::string op_a{}; + std::string op_a; if (instr.is_b_gpr) { op_a = @@ -2444,10 +2751,7 @@ private: case OpCode::Id::LD_C: { UNIMPLEMENTED_IF(instr.ld_c.unknown != 0); - // Add an extra scope and declare the index register inside to prevent - // overwriting it in case it is used as an output of the LD instruction. - shader.AddLine("{"); - ++shader.scope; + const auto scope = shader.Scope(); shader.AddLine("uint index = (" + regs.GetRegisterAsInteger(instr.gpr8, 0, false) + " / 4) & (MAX_CONSTBUFFER_ELEMENTS - 1);"); @@ -2473,19 +2777,13 @@ private: UNIMPLEMENTED_MSG("Unhandled type: {}", static_cast<unsigned>(instr.ld_c.type.Value())); } - - --shader.scope; - shader.AddLine("}"); break; } case OpCode::Id::LD_L: { UNIMPLEMENTED_IF_MSG(instr.ld_l.unknown == 1, "LD_L Unhandled mode: {}", static_cast<unsigned>(instr.ld_l.unknown.Value())); - // Add an extra scope and declare the index register inside to prevent - // overwriting it in case it is used as an output of the LD instruction. - shader.AddLine('{'); - ++shader.scope; + const auto scope = shader.Scope(); std::string op = '(' + regs.GetRegisterAsInteger(instr.gpr8, 0, false) + " + " + std::to_string(instr.smem_imm.Value()) + ')'; @@ -2502,9 +2800,6 @@ private: UNIMPLEMENTED_MSG("LD_L Unhandled type: {}", static_cast<unsigned>(instr.ldst_sl.type.Value())); } - - --shader.scope; - shader.AddLine('}'); break; } case OpCode::Id::ST_A: { @@ -2539,10 +2834,7 @@ private: UNIMPLEMENTED_IF_MSG(instr.st_l.unknown == 0, "ST_L Unhandled mode: {}", static_cast<unsigned>(instr.st_l.unknown.Value())); - // Add an extra scope and declare the index register inside to prevent - // overwriting it in case it is used as an output of the LD instruction. - shader.AddLine('{'); - ++shader.scope; + const auto scope = shader.Scope(); std::string op = '(' + regs.GetRegisterAsInteger(instr.gpr8, 0, false) + " + " + std::to_string(instr.smem_imm.Value()) + ')'; @@ -2557,179 +2849,28 @@ private: UNIMPLEMENTED_MSG("ST_L Unhandled type: {}", static_cast<unsigned>(instr.ldst_sl.type.Value())); } - - --shader.scope; - shader.AddLine('}'); break; } case OpCode::Id::TEX: { Tegra::Shader::TextureType texture_type{instr.tex.texture_type}; - std::string coord; const bool is_array = instr.tex.array != 0; - + const bool depth_compare = + instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); + const auto process_mode = instr.tex.GetTextureProcessMode(); UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), "AOFFI is not implemented"); - const bool depth_compare = - instr.tex.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); - u32 num_coordinates = TextureCoordinates(texture_type); - u32 start_index = 0; - std::string array_elem; - if (is_array) { - array_elem = regs.GetRegisterAsInteger(instr.gpr8); - start_index = 1; - } - const auto process_mode = instr.tex.GetTextureProcessMode(); - u32 start_index_b = 0; - std::string lod_value; - if (process_mode != Tegra::Shader::TextureProcessMode::LZ && - process_mode != Tegra::Shader::TextureProcessMode::None) { - start_index_b = 1; - lod_value = regs.GetRegisterAsFloat(instr.gpr20); - } - - std::string depth_value; - if (depth_compare) { - depth_value = regs.GetRegisterAsFloat(instr.gpr20.Value() + start_index_b); - } - - bool depth_compare_extra = false; + const auto [coord, texture] = + GetTEXCode(instr, texture_type, process_mode, depth_compare, is_array); - switch (num_coordinates) { - case 1: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index); - if (is_array) { - if (depth_compare) { - coord = "vec3 coords = vec3(" + x + ", " + depth_value + ", " + - array_elem + ");"; - } else { - coord = "vec2 coords = vec2(" + x + ", " + array_elem + ");"; - } - } else { - if (depth_compare) { - coord = "vec2 coords = vec2(" + x + ", " + depth_value + ");"; - } else { - coord = "float coords = " + x + ';'; - } - } - break; - } - case 2: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index); - const std::string y = - regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index + 1); - if (is_array) { - if (depth_compare) { - coord = "vec4 coords = vec4(" + x + ", " + y + ", " + depth_value + - ", " + array_elem + ");"; - } else { - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + array_elem + ");"; - } - } else { - if (depth_compare) { - coord = - "vec3 coords = vec3(" + x + ", " + y + ", " + depth_value + ");"; - } else { - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; - } - } - break; - } - case 3: { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index); - const std::string y = - regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index + 1); - const std::string z = - regs.GetRegisterAsFloat(instr.gpr8.Value() + start_index + 2); - if (is_array) { - depth_compare_extra = depth_compare; - coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + - array_elem + ");"; - } else { - if (depth_compare) { - coord = "vec4 coords = vec4(" + x + ", " + y + ", " + z + ", " + - depth_value + ");"; - } else { - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; - } - } - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled coordinates number {}", - static_cast<u32>(num_coordinates)); - - // Fallback to interpreting as a 2D texture for now - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; - texture_type = Tegra::Shader::TextureType::Texture2D; - } - - const std::string sampler = - 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. - - shader.AddLine('{'); - ++shader.scope; + const auto scope = shader.Scope(); shader.AddLine(coord); - std::string texture; - switch (instr.tex.GetTextureProcessMode()) { - case Tegra::Shader::TextureProcessMode::None: { - if (!depth_compare_extra) { - texture = "texture(" + sampler + ", coords)"; - } else { - texture = "texture(" + sampler + ", coords, " + depth_value + ')'; - } - break; - } - case Tegra::Shader::TextureProcessMode::LZ: { - if (!depth_compare_extra) { - texture = "textureLod(" + sampler + ", coords, 0.0)"; - } else { - texture = "texture(" + sampler + ", coords, " + depth_value + ')'; - } - break; - } - case Tegra::Shader::TextureProcessMode::LB: - case Tegra::Shader::TextureProcessMode::LBA: { - // TODO: Figure if A suffix changes the equation at all. - if (!depth_compare_extra) { - texture = "texture(" + sampler + ", coords, " + lod_value + ')'; - } else { - texture = "texture(" + sampler + ", coords, " + depth_value + ')'; - LOG_WARNING(HW_GPU, - "OpenGL Limitation: can't set bias value along depth compare"); - } - break; - } - case Tegra::Shader::TextureProcessMode::LL: - case Tegra::Shader::TextureProcessMode::LLA: { - // TODO: Figure if A suffix changes the equation at all. - if (!depth_compare_extra) { - texture = "textureLod(" + sampler + ", coords, " + lod_value + ')'; - } else { - texture = "texture(" + sampler + ", coords, " + depth_value + ')'; - LOG_WARNING(HW_GPU, - "OpenGL Limitation: can't set lod value along depth compare"); - } - break; - } - default: { - if (!depth_compare_extra) { - texture = "texture(" + sampler + ", coords)"; - } else { - texture = "texture(" + sampler + ", coords, " + depth_value + ')'; - } - UNIMPLEMENTED_MSG("Unhandled texture process mode {}", - static_cast<u32>(instr.tex.GetTextureProcessMode())); - } - } - if (!depth_compare) { + if (depth_compare) { + regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); + } else { shader.AddLine("vec4 texture_tmp = " + texture + ';'); std::size_t dest_elem{}; for (std::size_t elem = 0; elem < 4; ++elem) { @@ -2741,147 +2882,42 @@ private: dest_elem); ++dest_elem; } - } else { - regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); } - --shader.scope; - shader.AddLine('}'); break; } case OpCode::Id::TEXS: { Tegra::Shader::TextureType texture_type{instr.texs.GetTextureType()}; - bool is_array{instr.texs.IsArrayTexture()}; + const bool is_array{instr.texs.IsArrayTexture()}; + const bool depth_compare = + instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); + const auto process_mode = instr.texs.GetTextureProcessMode(); UNIMPLEMENTED_IF_MSG(instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); - const bool depth_compare = - instr.texs.UsesMiscMode(Tegra::Shader::TextureMiscMode::DC); - u32 num_coordinates = TextureCoordinates(texture_type); - const auto process_mode = instr.texs.GetTextureProcessMode(); - std::string lod_value; - std::string coord; - u32 lod_offset = 0; - if (process_mode == Tegra::Shader::TextureProcessMode::LL) { - if (num_coordinates > 2) { - lod_value = regs.GetRegisterAsFloat(instr.gpr20.Value() + 1); - lod_offset = 2; - } else { - lod_value = regs.GetRegisterAsFloat(instr.gpr20); - lod_offset = 1; - } - } + const auto scope = shader.Scope(); - switch (num_coordinates) { - case 1: { - coord = "float coords = " + regs.GetRegisterAsFloat(instr.gpr8) + ';'; - break; - } - case 2: { - if (is_array) { - if (depth_compare) { - 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 index = regs.GetRegisterAsInteger(instr.gpr8); - const std::string x = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + index + ");"; - } - } else { - if (lod_offset != 0) { - if (depth_compare) { - 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.Value() + lod_offset); - coord = "vec3 coords = vec3(" + x + ", " + y + ", " + z + ");"; - } 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 + ");"; - } - } else { - if (depth_compare) { - 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 { - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - 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 + ");"; - break; - } - default: - UNIMPLEMENTED_MSG("Unhandled coordinates number {}", - static_cast<u32>(num_coordinates)); + auto [coord, texture] = + GetTEXSCode(instr, texture_type, process_mode, depth_compare, is_array); - // Fallback to interpreting as a 2D texture for now - const std::string x = regs.GetRegisterAsFloat(instr.gpr8); - const std::string y = regs.GetRegisterAsFloat(instr.gpr20); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; - texture_type = Tegra::Shader::TextureType::Texture2D; - is_array = false; - } - const std::string sampler = - GetSampler(instr.sampler, texture_type, is_array, depth_compare); - std::string texture; - switch (process_mode) { - case Tegra::Shader::TextureProcessMode::None: { - texture = "texture(" + sampler + ", coords)"; - break; - } - case Tegra::Shader::TextureProcessMode::LZ: { - if (depth_compare && is_array) { - texture = "texture(" + sampler + ", coords)"; - } else { - texture = "textureLod(" + sampler + ", coords, 0.0)"; - } - break; - } - case Tegra::Shader::TextureProcessMode::LL: { - texture = "textureLod(" + sampler + ", coords, " + lod_value + ')'; - break; - } - default: { - texture = "texture(" + sampler + ", coords)"; - UNIMPLEMENTED_MSG("Unhandled texture process mode {}", - static_cast<u32>(instr.texs.GetTextureProcessMode())); - } + shader.AddLine(coord); + + if (depth_compare) { + texture = "vec4(" + texture + ')'; } - if (!depth_compare) { - WriteTexsInstruction(instr, coord, texture); + shader.AddLine("vec4 texture_tmp = " + texture + ';'); + + if (instr.texs.fp32_flag) { + WriteTexsInstructionFloat(instr, "texture_tmp"); } else { - WriteTexsInstruction(instr, coord, "vec4(" + texture + ')'); + WriteTexsInstructionHalfFloat(instr, "texture_tmp"); } - break; } case OpCode::Id::TLDS: { 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), @@ -2889,63 +2925,16 @@ private: UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(Tegra::Shader::TextureMiscMode::MZ), "MZ is not implemented"); - u32 extra_op_offset = 0; + const auto [coord, texture] = GetTLDSCode(instr, texture_type, is_array); - // Scope to avoid variable name overlaps. - shader.AddLine('{'); - ++shader.scope; - std::string coords; + const auto scope = shader.Scope(); - switch (texture_type) { - case Tegra::Shader::TextureType::Texture1D: { - const std::string x = regs.GetRegisterAsInteger(instr.gpr8); - coords = "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 + ");"); - coords = "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); - std::string texture = "texelFetch(" + sampler + ", coords, 0)"; - switch (instr.tlds.GetTextureProcessMode()) { - case Tegra::Shader::TextureProcessMode::LZ: { - texture = "texelFetch(" + sampler + ", coords, 0)"; - break; - } - case Tegra::Shader::TextureProcessMode::LL: { - shader.AddLine( - "float lod = " + - regs.GetRegisterAsInteger(instr.gpr20.Value() + extra_op_offset) + ';'); - texture = "texelFetch(" + sampler + ", coords, lod)"; - break; - } - default: { - texture = "texelFetch(" + sampler + ", coords, 0)"; - UNIMPLEMENTED_MSG("Unhandled texture process mode {}", - static_cast<u32>(instr.tlds.GetTextureProcessMode())); - } - } - WriteTexsInstruction(instr, coords, texture); - - --shader.scope; - shader.AddLine('}'); + 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); - std::string coord; UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); @@ -2955,64 +2944,30 @@ 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 bool is_array = instr.tld4.array != 0; - // 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. - shader.AddLine('{'); - ++shader.scope; + const auto [coord, texture] = + GetTLD4Code(instr, texture_type, depth_compare, is_array); - 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 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) { - shader.AddLine("vec4 texture_tmp = " + texture + ';'); - 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_tmp", 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; } - } else { - regs.SetRegisterToFloat(instr.gpr0, 0, texture, 1, 1, false); + regs.SetRegisterToFloat(instr.gpr0, elem, "texture_tmp", 1, 4, false, + dest_elem); + ++dest_elem; } - --shader.scope; - shader.AddLine('}'); break; } case OpCode::Id::TLD4S: { @@ -3023,45 +2978,42 @@ private: instr.tld4s.UsesMiscMode(Tegra::Shader::TextureMiscMode::AOFFI), "AOFFI is not implemented"); - // Scope to avoid variable name overlaps. - shader.AddLine('{'); - ++shader.scope; + 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) { - coords = "vec2 coords = vec2(" + op_a + ", " + op_b + ");"; - } else { - // Note: TLD4S coordinate encoding works just like TEXS's - const std::string op_y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coords = "vec3 coords = vec3(" + op_a + ", " + op_y + ", " + op_b + ");"; - } - const std::string texture = "textureGather(" + sampler + ", coords, " + - std::to_string(instr.tld4s.component) + ')'; + + const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); + coords = "vec2 coords = vec2(" + op_a + ", "; + std::string texture = "textureGather(" + sampler + ", coords, "; if (!depth_compare) { - WriteTexsInstruction(instr, coords, texture); + const std::string op_b = regs.GetRegisterAsFloat(instr.gpr20); + coords += op_b + ");"; + texture += std::to_string(instr.tld4s.component) + ')'; } else { - WriteTexsInstruction(instr, coords, "vec4(" + texture + ')'); + 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.scope; - shader.AddLine('}'); + shader.AddLine(coords); + shader.AddLine("vec4 texture_tmp = " + texture + ';'); + WriteTexsInstructionFloat(instr, "texture_tmp"); break; } case OpCode::Id::TXQ: { UNIMPLEMENTED_IF_MSG(instr.txq.UsesMiscMode(Tegra::Shader::TextureMiscMode::NODEP), "NODEP is not implemented"); - ++shader.scope; - shader.AddLine('{'); - // TODO: the new commits on the texture refactor, change the way samplers work. + const auto scope = shader.Scope(); + + // TODO: The new commits on the texture refactor, change the way samplers work. // Sadly, not all texture instructions specify the type of texture their sampler // uses. This must be fixed at a later instance. const std::string sampler = @@ -3072,7 +3024,8 @@ private: regs.GetRegisterAsInteger(instr.gpr8) + ')'; const std::string mip_level = "textureQueryLevels(" + sampler + ')'; shader.AddLine("ivec2 sizes = " + texture + ';'); - regs.SetRegisterToInteger(instr.gpr0, true, 0, "sizes.x", 1, 1); + + regs.SetRegisterToInteger(instr.gpr0.Value() + 0, true, 0, "sizes.x", 1, 1); regs.SetRegisterToInteger(instr.gpr0.Value() + 1, true, 0, "sizes.y", 1, 1); regs.SetRegisterToInteger(instr.gpr0.Value() + 2, true, 0, "0", 1, 1); regs.SetRegisterToInteger(instr.gpr0.Value() + 3, true, 0, mip_level, 1, 1); @@ -3083,8 +3036,6 @@ private: static_cast<u32>(instr.txq.query_type.Value())); } } - --shader.scope; - shader.AddLine('}'); break; } case OpCode::Id::TMML: { @@ -3099,17 +3050,18 @@ private: const std::string sampler = GetSampler(instr.sampler, texture_type, is_array, false); - // TODO: add coordinates for different samplers once other texture types are + const auto scope = shader.Scope(); + + // TODO: Add coordinates for different samplers once other texture types are // implemented. - std::string coord; switch (texture_type) { case Tegra::Shader::TextureType::Texture1D: { - coord = "float coords = " + x + ';'; + shader.AddLine("float coords = " + x + ';'); break; } case Tegra::Shader::TextureType::Texture2D: { const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); break; } default: @@ -3117,22 +3069,15 @@ private: // Fallback to interpreting as a 2D texture for now const std::string y = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); - coord = "vec2 coords = vec2(" + x + ", " + y + ");"; + shader.AddLine("vec2 coords = vec2(" + x + ", " + y + ");"); texture_type = Tegra::Shader::TextureType::Texture2D; } - // 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. - shader.AddLine('{'); - ++shader.scope; - shader.AddLine(coord); + const std::string texture = "textureQueryLod(" + sampler + ", coords)"; - const std::string tmp = "vec2 tmp = " + texture + "*vec2(256.0, 256.0);"; - shader.AddLine(tmp); + shader.AddLine("vec2 tmp = " + texture + " * vec2(256.0, 256.0);"); regs.SetRegisterToInteger(instr.gpr0, true, 0, "int(tmp.y)", 1, 1); regs.SetRegisterToInteger(instr.gpr0.Value() + 1, false, 0, "uint(tmp.x)", 1, 1); - --shader.scope; - shader.AddLine('}'); break; } default: { @@ -3971,4 +3916,4 @@ std::optional<ProgramResult> DecompileProgram(const ProgramCode& program_code, u return {}; } -} // namespace OpenGL::GLShader::Decompiler +} // namespace OpenGL::GLShader::Decompiler
\ No newline at end of file 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 |