Cool_Tools/yuzu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

gl_shader_decompiler: Implement HFMA2 instructions

Browse Source
This commit is contained in:
ReinUsesLisp 2018-10-15 02:08:31 -03:00
parent d93cdc2750
commit 3d65aa4caf
2 changed files with 85 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
src/video_core/engines/shader_bytecode.h
View File

@ -614,6 +614,29 @@ union Instruction {
}
} half_imm;
union {
union {
BitField<37, 2, HalfPrecision> precision;
BitField<32, 1, u64> saturate;
BitField<30, 1, u64> negate_c;
BitField<35, 2, HalfType> type_c;
} rr;
BitField<57, 2, HalfPrecision> precision;
BitField<52, 1, u64> saturate;
BitField<49, 2, HalfMerge> merge;
BitField<47, 2, HalfType> type_a;
BitField<56, 1, u64> negate_b;
BitField<28, 2, HalfType> type_b;
BitField<51, 1, u64> negate_c;
BitField<53, 2, HalfType> type_reg39;
} hfma2;
union {
BitField<40, 1, u64> invert;
} popc;
@ -1212,6 +1235,10 @@ public:
HMUL2_C,
HMUL2_R,
HMUL2_IMM,
HFMA2_CR,
HFMA2_RC,
HFMA2_RR,
HFMA2_IMM_R,
POPC_C,
POPC_R,
POPC_IMM,
@ -1290,6 +1317,7 @@ public:
Bfe,
Shift,
Ffma,
Hfma2,
Flow,
Synch,
Memory,
@ -1464,6 +1492,10 @@ private:
INST("0111100-1-------", Id::HMUL2_C, Type::ArithmeticHalf, "HMUL2_C"),
INST("0101110100001---", Id::HMUL2_R, Type::ArithmeticHalf, "HMUL2_R"),
INST("0111100-0-------", Id::HMUL2_IMM, Type::ArithmeticHalfImmediate, "HMUL2_IMM"),
INST("01110---1-------", Id::HFMA2_CR, Type::Hfma2, "HFMA2_CR"),
INST("01100---1-------", Id::HFMA2_RC, Type::Hfma2, "HFMA2_RC"),
INST("0101110100000---", Id::HFMA2_RR, Type::Hfma2, "HFMA2_RR"),
INST("01110---0-------", Id::HFMA2_IMM_R, Type::Hfma2, "HFMA2_R_IMM"),
INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),
INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"),

53
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View File

@ -1964,6 +1964,59 @@ private:
instr.alu.saturate_d);
break;
}
case OpCode::Type::Hfma2: {
if (opcode->GetId() == OpCode::Id::HFMA2_RR) {
ASSERT_MSG(instr.hfma2.rr.precision == Tegra::Shader::HalfPrecision::None,
"Unimplemented");
} else {
ASSERT_MSG(instr.hfma2.precision == Tegra::Shader::HalfPrecision::None,
"Unimplemented");
}
const bool saturate = opcode->GetId() == OpCode::Id::HFMA2_RR
? instr.hfma2.rr.saturate != 0
: instr.hfma2.saturate != 0;
const std::string op_a =
GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr8, 0, false), instr.hfma2.type_a);
std::string op_b, op_c;
switch (opcode->GetId()) {
case OpCode::Id::HFMA2_CR:
op_b = GetHalfFloat(regs.GetUniform(instr.cbuf34.index, instr.cbuf34.offset,
GLSLRegister::Type::UnsignedInteger),
instr.hfma2.type_b, false, instr.hfma2.negate_b);
op_c = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr39, 0, false),
instr.hfma2.type_reg39, false, instr.hfma2.negate_c);
break;
case OpCode::Id::HFMA2_RC:
op_b = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr39, 0, false),
instr.hfma2.type_reg39, false, instr.hfma2.negate_b);
op_c = GetHalfFloat(regs.GetUniform(instr.cbuf34.index, instr.cbuf34.offset,
GLSLRegister::Type::UnsignedInteger),
instr.hfma2.type_b, false, instr.hfma2.negate_c);
break;
case OpCode::Id::HFMA2_RR:
op_b = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr20, 0, false),
instr.hfma2.type_b, false, instr.hfma2.negate_b);
op_c = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr39, 0, false),
instr.hfma2.rr.type_c, false, instr.hfma2.rr.negate_c);
break;
case OpCode::Id::HFMA2_IMM_R:
op_b = UnpackHalfImmediate(instr, true);
op_c = GetHalfFloat(regs.GetRegisterAsInteger(instr.gpr39, 0, false),
instr.hfma2.type_reg39, false, instr.hfma2.negate_c);
break;
default:
UNREACHABLE();
op_c = op_b = "vec2(0)";
break;
}
const std::string result = '(' + op_a + " * " + op_b + " + " + op_c + ')';
regs.SetRegisterToHalfFloat(instr.gpr0, 0, result, instr.hfma2.merge, 1, 1, saturate);
break;
}
case OpCode::Type::Conversion: {
switch (opcode->GetId()) {
case OpCode::Id::I2I_R: {