diff options
author | bunnei <bunneidev@gmail.com> | 2018-06-12 05:19:37 +0200 |
---|---|---|
committer | GitHub <noreply@github.com> | 2018-06-12 05:19:37 +0200 |
commit | 2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab (patch) | |
tree | b6f3b5e2cdaedf6ef94be6437f67fc07054484cd /src | |
parent | Merge pull request #555 from Subv/gpu_sysregs (diff) | |
parent | gl_shader_decompiler: Implement saturate for float instructions. (diff) | |
download | yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.tar yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.tar.gz yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.tar.bz2 yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.tar.lz yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.tar.xz yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.tar.zst yuzu-2dc8b5c2241f6a18f636bbb7a1ff6403c76e7eab.zip |
Diffstat (limited to 'src')
-rw-r--r-- | src/video_core/engines/shader_bytecode.h | 3 | ||||
-rw-r--r-- | src/video_core/renderer_opengl/gl_shader_decompiler.cpp | 68 |
2 files changed, 32 insertions, 39 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h index ec8dbd370..c158ffed2 100644 --- a/src/video_core/engines/shader_bytecode.h +++ b/src/video_core/engines/shader_bytecode.h @@ -213,7 +213,6 @@ union Instruction { BitField<28, 8, Register> gpr28; BitField<39, 8, Register> gpr39; BitField<48, 16, u64> opcode; - BitField<50, 1, u64> saturate_a; union { BitField<20, 19, u64> imm20_19; @@ -222,7 +221,7 @@ union Instruction { BitField<46, 1, u64> abs_a; BitField<48, 1, u64> negate_a; BitField<49, 1, u64> abs_b; - BitField<50, 1, u64> abs_d; + BitField<50, 1, u64> saturate_d; BitField<56, 1, u64> negate_imm; union { diff --git a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp index 67726e7c6..b94b79384 100644 --- a/src/video_core/renderer_opengl/gl_shader_decompiler.cpp +++ b/src/video_core/renderer_opengl/gl_shader_decompiler.cpp @@ -299,13 +299,15 @@ public: * @param value The code representing the value to assign. * @param dest_num_components Number of components in the destination. * @param value_num_components Number of components in the value. - * @param is_abs Optional, when True, applies absolute value to output. + * @param is_saturated Optional, when True, saturates the provided value. * @param dest_elem Optional, the destination element to use for the operation. */ void SetRegisterToFloat(const Register& reg, u64 elem, const std::string& value, - u64 dest_num_components, u64 value_num_components, bool is_abs = false, - u64 dest_elem = 0) { - SetRegister(reg, elem, value, dest_num_components, value_num_components, is_abs, dest_elem); + u64 dest_num_components, u64 value_num_components, + bool is_saturated = false, u64 dest_elem = 0) { + + SetRegister(reg, elem, is_saturated ? "clamp(" + value + ", 0.0, 1.0)" : value, + dest_num_components, value_num_components, dest_elem); } /** @@ -315,18 +317,21 @@ public: * @param value The code representing the value to assign. * @param dest_num_components Number of components in the destination. * @param value_num_components Number of components in the value. - * @param is_abs Optional, when True, applies absolute value to output. + * @param is_saturated Optional, when True, saturates the provided value. * @param dest_elem Optional, the destination element to use for the operation. */ void SetRegisterToInteger(const Register& reg, bool is_signed, u64 elem, const std::string& value, u64 dest_num_components, - u64 value_num_components, bool is_abs = false, u64 dest_elem = 0) { + u64 value_num_components, bool is_saturated = false, + u64 dest_elem = 0) { + ASSERT_MSG(!is_saturated, "Unimplemented"); + const std::string func = GetGLSLConversionFunc( is_signed ? GLSLRegister::Type::Integer : GLSLRegister::Type::UnsignedInteger, GLSLRegister::Type::Float); SetRegister(reg, elem, func + '(' + value + ')', dest_num_components, value_num_components, - is_abs, dest_elem); + dest_elem); } /** @@ -500,12 +505,10 @@ private: * @param value The code representing the value to assign. * @param dest_num_components Number of components in the destination. * @param value_num_components Number of components in the value. - * @param is_abs Optional, when True, applies absolute value to output. * @param dest_elem Optional, the destination element to use for the operation. */ void SetRegister(const Register& reg, u64 elem, const std::string& value, - u64 dest_num_components, u64 value_num_components, bool is_abs, - u64 dest_elem) { + u64 dest_num_components, u64 value_num_components, u64 dest_elem) { std::string dest = GetRegister(reg, dest_elem); if (dest_num_components > 1) { dest += GetSwizzle(elem); @@ -516,8 +519,6 @@ private: src += GetSwizzle(elem); } - src = is_abs ? "abs(" + src + ')' : src; - shader.AddLine(dest + " = " + src + ';'); } @@ -808,9 +809,8 @@ private: case OpCode::Id::FMUL_C: case OpCode::Id::FMUL_R: case OpCode::Id::FMUL_IMM: { - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); - - regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " * " + op_b, 1, 1, instr.alu.abs_d); + regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " * " + op_b, 1, 1, + instr.alu.saturate_d); break; } case OpCode::Id::FMUL32_IMM: { @@ -823,41 +823,39 @@ private: case OpCode::Id::FADD_C: case OpCode::Id::FADD_R: case OpCode::Id::FADD_IMM: { - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); - - regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " + " + op_b, 1, 1, instr.alu.abs_d); + regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " + " + op_b, 1, 1, + instr.alu.saturate_d); break; } case OpCode::Id::MUFU: { - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); - switch (instr.sub_op) { case SubOp::Cos: regs.SetRegisterToFloat(instr.gpr0, 0, "cos(" + op_a + ')', 1, 1, - instr.alu.abs_d); + instr.alu.saturate_d); break; case SubOp::Sin: regs.SetRegisterToFloat(instr.gpr0, 0, "sin(" + op_a + ')', 1, 1, - instr.alu.abs_d); + instr.alu.saturate_d); break; case SubOp::Ex2: regs.SetRegisterToFloat(instr.gpr0, 0, "exp2(" + op_a + ')', 1, 1, - instr.alu.abs_d); + instr.alu.saturate_d); break; case SubOp::Lg2: regs.SetRegisterToFloat(instr.gpr0, 0, "log2(" + op_a + ')', 1, 1, - instr.alu.abs_d); + instr.alu.saturate_d); break; case SubOp::Rcp: - regs.SetRegisterToFloat(instr.gpr0, 0, "1.0 / " + op_a, 1, 1, instr.alu.abs_d); + regs.SetRegisterToFloat(instr.gpr0, 0, "1.0 / " + op_a, 1, 1, + instr.alu.saturate_d); break; case SubOp::Rsq: regs.SetRegisterToFloat(instr.gpr0, 0, "inversesqrt(" + op_a + ')', 1, 1, - instr.alu.abs_d); + instr.alu.saturate_d); break; case SubOp::Min: regs.SetRegisterToFloat(instr.gpr0, 0, "min(" + op_a + "," + op_b + ')', 1, 1, - instr.alu.abs_d); + instr.alu.saturate_d); break; default: NGLOG_CRITICAL(HW_GPU, "Unhandled MUFU sub op: {0:x}", @@ -1028,8 +1026,8 @@ private: case OpCode::Id::IADD_C: case OpCode::Id::IADD_R: case OpCode::Id::IADD_IMM: { - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); - regs.SetRegisterToInteger(instr.gpr0, true, 0, op_a + " + " + op_b, 1, 1); + regs.SetRegisterToInteger(instr.gpr0, true, 0, op_a + " + " + op_b, 1, 1, + instr.alu.saturate_d); break; } case OpCode::Id::ISCADD_C: @@ -1051,8 +1049,6 @@ private: break; } case OpCode::Type::Ffma: { - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); - std::string op_a = regs.GetRegisterAsFloat(instr.gpr8); std::string op_b = instr.ffma.negate_b ? "-" : ""; std::string op_c = instr.ffma.negate_c ? "-" : ""; @@ -1086,13 +1082,13 @@ private: } } - regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " * " + op_b + " + " + op_c, 1, 1); + regs.SetRegisterToFloat(instr.gpr0, 0, op_a + " * " + op_b + " + " + op_c, 1, 1, + instr.alu.saturate_d); break; } case OpCode::Type::Conversion: { ASSERT_MSG(instr.conversion.size == Register::Size::Word, "Unimplemented"); ASSERT_MSG(!instr.conversion.negate_a, "Unimplemented"); - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); switch (opcode->GetId()) { case OpCode::Id::I2I_R: { @@ -1106,7 +1102,7 @@ private: } regs.SetRegisterToInteger(instr.gpr0, instr.conversion.is_output_signed, 0, op_a, 1, - 1); + 1, instr.alu.saturate_d); break; } case OpCode::Id::I2F_R: { @@ -1122,8 +1118,6 @@ private: break; } case OpCode::Id::F2F_R: { - ASSERT_MSG(!instr.saturate_a, "Unimplemented"); - std::string op_a = regs.GetRegisterAsFloat(instr.gpr20); switch (instr.conversion.f2f.rounding) { @@ -1149,7 +1143,7 @@ private: op_a = "abs(" + op_a + ')'; } - regs.SetRegisterToFloat(instr.gpr0, 0, op_a, 1, 1); + regs.SetRegisterToFloat(instr.gpr0, 0, op_a, 1, 1, instr.alu.saturate_d); break; } case OpCode::Id::F2I_R: { |