lux/yuzu
1
0
Fork 0
mirror of https://git.h3cjp.net/H3cJP/yuzu.git synced 2024-12-15 15:25:31 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #1040 from bunnei/xmad

Browse source 
gl_shader_decompiler: Implement XMAD instruction.
This commit is contained in:
bunnei 2018-08-12 21:56:40 -04:00 committed by GitHub
parent a970709d5d 534abf9d97
commit 8da753ab81
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 120 additions and 4 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

29
src/video_core/engines/shader_bytecode.h
View file

@ -200,6 +200,14 @@ enum class IMinMaxExchange : u64 {
XHi = 3,
};
enum class XmadMode : u64 {
None = 0,
CLo = 1,
CHi = 2,
CSfu = 3,
CBcc = 4,
};
enum class FlowCondition : u64 {
Always = 0xF,
Fcsm_Tr = 0x1C, // TODO(bunnei): What is this used for?
@ -456,6 +464,18 @@ union Instruction {
}
} bra;
union {
BitField<20, 16, u64> imm20_16;
BitField<36, 1, u64> product_shift_left;
BitField<37, 1, u64> merge_37;
BitField<48, 1, u64> sign_a;
BitField<49, 1, u64> sign_b;
BitField<50, 3, XmadMode> mode;
BitField<52, 1, u64> high_b;
BitField<53, 1, u64> high_a;
BitField<56, 1, u64> merge_56;
} xmad;
union {
BitField<20, 14, u64> offset;
BitField<34, 5, u64> index;
@ -593,6 +613,7 @@ public:
IntegerSetPredicate,
PredicateSetPredicate,
Conversion,
Xmad,
Unknown,
};
@ -782,10 +803,10 @@ private:
INST("010010110101----", Id::ISET_C, Type::IntegerSet, "ISET_C"),
INST("0011011-0101----", Id::ISET_IMM, Type::IntegerSet, "ISET_IMM"),
INST("0101000010010---", Id::PSETP, Type::PredicateSetPredicate, "PSETP"),
INST("0011011-00------", Id::XMAD_IMM, Type::Arithmetic, "XMAD_IMM"),
INST("0100111---------", Id::XMAD_CR, Type::Arithmetic, "XMAD_CR"),
INST("010100010-------", Id::XMAD_RC, Type::Arithmetic, "XMAD_RC"),
INST("0101101100------", Id::XMAD_RR, Type::Arithmetic, "XMAD_RR"),
INST("0011011-00------", Id::XMAD_IMM, Type::Xmad, "XMAD_IMM"),
INST("0100111---------", Id::XMAD_CR, Type::Xmad, "XMAD_CR"),
INST("010100010-------", Id::XMAD_RC, Type::Xmad, "XMAD_RC"),
INST("0101101100------", Id::XMAD_RR, Type::Xmad, "XMAD_RR"),
};
#undef INST
std::stable_sort(table.begin(), table.end(), [](const auto& a, const auto& b) {

95
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View file

@ -376,6 +376,8 @@ public:
return value;
} else if (type == GLSLRegister::Type::Integer) {
return "floatBitsToInt(" + value + ')';
} else if (type == GLSLRegister::Type::UnsignedInteger) {
return "floatBitsToUint(" + value + ')';
} else {
UNREACHABLE();
}
@ -1630,6 +1632,99 @@ private:
}
break;
}
case OpCode::Type::Xmad: {
ASSERT_MSG(!instr.xmad.sign_a, "Unimplemented");
ASSERT_MSG(!instr.xmad.sign_b, "Unimplemented");
std::string op_a{regs.GetRegisterAsInteger(instr.gpr8, 0, instr.xmad.sign_a)};
std::string op_b;
std::string op_c;
// TODO(bunnei): Needs to be fixed once op_a or op_b is signed
ASSERT_MSG(instr.xmad.sign_a == instr.xmad.sign_b, "Unimplemented");
const bool is_signed{instr.xmad.sign_a == 1};
bool is_merge{};
switch (opcode->GetId()) {
case OpCode::Id::XMAD_CR: {
is_merge = instr.xmad.merge_56;
op_b += regs.GetUniform(instr.cbuf34.index, instr.cbuf34.offset,
instr.xmad.sign_b ? GLSLRegister::Type::Integer
: GLSLRegister::Type::UnsignedInteger);
op_c += regs.GetRegisterAsInteger(instr.gpr39, 0, is_signed);
break;
}
case OpCode::Id::XMAD_RR: {
is_merge = instr.xmad.merge_37;
op_b += regs.GetRegisterAsInteger(instr.gpr20, 0, instr.xmad.sign_b);
op_c += regs.GetRegisterAsInteger(instr.gpr39, 0, is_signed);
break;
}
case OpCode::Id::XMAD_RC: {
op_b += regs.GetRegisterAsInteger(instr.gpr39, 0, instr.xmad.sign_b);
op_c += regs.GetUniform(instr.cbuf34.index, instr.cbuf34.offset,
is_signed ? GLSLRegister::Type::Integer
: GLSLRegister::Type::UnsignedInteger);
break;
}
case OpCode::Id::XMAD_IMM: {
is_merge = instr.xmad.merge_37;
op_b += std::to_string(instr.xmad.imm20_16);
op_c += regs.GetRegisterAsInteger(instr.gpr39, 0, is_signed);
break;
}
default: {
LOG_CRITICAL(HW_GPU, "Unhandled XMAD instruction: {}", opcode->GetName());
UNREACHABLE();
}
}
// TODO(bunnei): Ensure this is right with signed operands
if (instr.xmad.high_a) {
op_a = "((" + op_a + ") >> 16)";
} else {
op_a = "((" + op_a + ") & 0xFFFF)";
}
std::string src2 = '(' + op_b + ')'; // Preserve original source 2
if (instr.xmad.high_b) {
op_b = '(' + src2 + " >> 16)";
} else {
op_b = '(' + src2 + " & 0xFFFF)";
}
std::string product = '(' + op_a + " * " + op_b + ')';
if (instr.xmad.product_shift_left) {
product = '(' + product + " << 16)";
}
switch (instr.xmad.mode) {
case Tegra::Shader::XmadMode::None:
break;
case Tegra::Shader::XmadMode::CLo:
op_c = "((" + op_c + ") & 0xFFFF)";
break;
case Tegra::Shader::XmadMode::CHi:
op_c = "((" + op_c + ") >> 16)";
break;
case Tegra::Shader::XmadMode::CBcc:
op_c = "((" + op_c + ") + (" + src2 + "<< 16))";
break;
default: {
LOG_CRITICAL(HW_GPU, "Unhandled XMAD mode: {}",
static_cast<u32>(instr.xmad.mode.Value()));
UNREACHABLE();
}
}
std::string sum{'(' + product + " + " + op_c + ')'};
if (is_merge) {
sum = "((" + sum + " & 0xFFFF) | (" + src2 + "<< 16))";
}
regs.SetRegisterToInteger(instr.gpr0, is_signed, 0, sum, 1, 1);
break;
}
default: {
switch (opcode->GetId()) {
case OpCode::Id::EXIT: {