diff options
author | bunnei <bunneidev@gmail.com> | 2018-06-03 18:05:38 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2018-06-03 18:05:38 +0200 |
commit | bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595 (patch) | |
tree | d75cdd5f1f50dc9e32e0ca28e167c1680f8f75d8 | |
parent | Merge pull request #495 from bunnei/improve-rro (diff) | |
parent | gl_shader_decompiler: Implement TEX instruction. (diff) | |
download | yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.tar yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.tar.gz yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.tar.bz2 yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.tar.lz yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.tar.xz yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.tar.zst yuzu-bb9d39b8fe89c7619ffbca2a0a6ecae1eb3f4595.zip |
-rw-r--r-- | src/video_core/engines/shader_bytecode.h | 15 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 45 |
2 files changed, 58 insertions, 2 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index 26c891356..fd4e0746e 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -261,6 +261,19 @@ union Instruction { BitField<50, 1, u64> saturate_a; } conversion; + union { + BitField<31, 4, u64> component_mask; + + bool IsComponentEnabled(size_t component) const { + return ((1 << component) & component_mask) != 0; + } + } tex; + + union { + // TODO(bunnei): This is just a guess, needs to be verified + BitField<52, 1, u64> enable_g_component; + } texs; + BitField<61, 1, u64> is_b_imm; BitField<60, 1, u64> is_b_gpr; BitField<59, 1, u64> is_c_gpr; @@ -281,6 +294,7 @@ public: KIL, LD_A, ST_A, + TEX, TEXQ, // Texture Query TEXS, // Texture Fetch with scalar/non-vec4 source/destinations TLDS, // Texture Load with scalar/non-vec4 source/destinations @@ -444,6 +458,7 @@ private: INST("111000110011----", Id::KIL, Type::Flow, "KIL"), INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"), INST("1110111111110---", Id::ST_A, Type::Memory, "ST_A"), + INST("1100000000111---", Id::TEX, Type::Memory, "TEX"), INST("1101111101001---", Id::TEXQ, Type::Memory, "TEXQ"), INST("1101100---------", Id::TEXS, Type::Memory, "TEXS"), INST("1101101---------", Id::TLDS, Type::Memory, "TLDS"), diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index cb4b68b26..86d880dfc 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -896,6 +896,32 @@ private: instr.gpr0); break; } + case OpCode::Id::TEX: { + ASSERT_MSG(instr.attribute.fmt20.size == 4, "untested"); + const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); + const std::string op_b = regs.GetRegisterAsFloat(instr.gpr8.Value() + 1); + const std::string sampler = GetSampler(instr.sampler); + const std::string coord = "vec2 coords = vec2(" + op_a + ", " + op_b + ");"; + // 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 = "texture(" + sampler + ", coords)"; + + size_t dest_elem{}; + for (size_t elem = 0; elem < instr.attribute.fmt20.size; ++elem) { + if (!instr.tex.IsComponentEnabled(elem)) { + // Skip disabled components + continue; + } + regs.SetRegisterToFloat(instr.gpr0, elem, texture, 1, 4, false, dest_elem); + ++dest_elem; + } + --shader.scope; + shader.AddLine("}"); + break; + } case OpCode::Id::TEXS: { ASSERT_MSG(instr.attribute.fmt20.size == 4, "untested"); const std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); @@ -908,8 +934,23 @@ private: ++shader.scope; shader.AddLine(coord); const std::string texture = "texture(" + sampler + ", coords)"; - for (unsigned elem = 0; elem < instr.attribute.fmt20.size; ++elem) { - regs.SetRegisterToFloat(instr.gpr0, elem, texture, 1, 4, false, elem); + + // TEXS has two destination registers. RG goes into gpr0+0 and gpr0+1, and BA goes + // into gpr28+0 and gpr28+1 + size_t offset{}; + for (const auto& dest : {instr.gpr0.Value(), instr.gpr28.Value()}) { + for (unsigned elem = 0; elem < 2; ++elem) { + if (dest + elem >= Register::ZeroIndex) { + // Skip invalid register values + break; + } + regs.SetRegisterToFloat(dest, elem + offset, texture, 1, 4, false, elem); + if (!instr.texs.enable_g_component) { + // Skip the second component + break; + } + } + offset += 2; } --shader.scope; shader.AddLine("}"); |