3
0
Fork 0
forked from suyu/suyu

Merge pull request #1172 from tech4me/impl_iadd3

Shaders: Implemented IADD3
This commit is contained in:
bunnei 2018-08-30 09:52:27 -04:00 committed by GitHub
commit 6e73039eb5
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 84 additions and 1 deletions

View file

@ -213,6 +213,18 @@ enum class XmadMode : u64 {
CBcc = 4, CBcc = 4,
}; };
enum class IAdd3Mode : u64 {
None = 0,
RightShift = 1,
LeftShift = 2,
};
enum class IAdd3Height : u64 {
None = 0,
LowerHalfWord = 1,
UpperHalfWord = 2,
};
enum class FlowCondition : u64 { enum class FlowCondition : u64 {
Always = 0xF, Always = 0xF,
Fcsm_Tr = 0x1C, // TODO(bunnei): What is this used for? Fcsm_Tr = 0x1C, // TODO(bunnei): What is this used for?
@ -338,6 +350,16 @@ union Instruction {
BitField<48, 1, u64> is_signed; BitField<48, 1, u64> is_signed;
} imnmx; } imnmx;
union {
BitField<31, 2, IAdd3Height> height_c;
BitField<33, 2, IAdd3Height> height_b;
BitField<35, 2, IAdd3Height> height_a;
BitField<37, 2, IAdd3Mode> mode;
BitField<49, 1, u64> neg_c;
BitField<50, 1, u64> neg_b;
BitField<51, 1, u64> neg_a;
} iadd3;
union { union {
BitField<54, 1, u64> saturate; BitField<54, 1, u64> saturate;
BitField<56, 1, u64> negate_a; BitField<56, 1, u64> negate_a;
@ -636,7 +658,7 @@ public:
IADD_C, IADD_C,
IADD_R, IADD_R,
IADD_IMM, IADD_IMM,
IADD3_C, IADD3_C, // Add 3 Integers
IADD3_R, IADD3_R,
IADD3_IMM, IADD3_IMM,
IADD32I, IADD32I,

View file

@ -1287,6 +1287,67 @@ private:
instr.alu.saturate_d); instr.alu.saturate_d);
break; break;
} }
case OpCode::Id::IADD3_C:
case OpCode::Id::IADD3_R:
case OpCode::Id::IADD3_IMM: {
std::string op_c = regs.GetRegisterAsInteger(instr.gpr39);
auto apply_height = [](auto height, auto& oprand) {
switch (height) {
case Tegra::Shader::IAdd3Height::None:
break;
case Tegra::Shader::IAdd3Height::LowerHalfWord:
oprand = "((" + oprand + ") & 0xFFFF)";
break;
case Tegra::Shader::IAdd3Height::UpperHalfWord:
oprand = "((" + oprand + ") >> 16)";
break;
default:
LOG_CRITICAL(HW_GPU, "Unhandled IADD3 height: {}",
static_cast<u32>(height.Value()));
UNREACHABLE();
}
};
if (opcode->GetId() == OpCode::Id::IADD3_R) {
apply_height(instr.iadd3.height_a, op_a);
apply_height(instr.iadd3.height_b, op_b);
apply_height(instr.iadd3.height_c, op_c);
}
if (instr.iadd3.neg_a)
op_a = "-(" + op_a + ')';
if (instr.iadd3.neg_b)
op_b = "-(" + op_b + ')';
if (instr.iadd3.neg_c)
op_c = "-(" + op_c + ')';
std::string result;
if (opcode->GetId() == OpCode::Id::IADD3_R) {
switch (instr.iadd3.mode) {
case Tegra::Shader::IAdd3Mode::RightShift:
// TODO(tech4me): According to
// https://envytools.readthedocs.io/en/latest/hw/graph/maxwell/cuda/int.html?highlight=iadd3
// The addition between op_a and op_b should be done in uint33, more
// investigation required
result = "(((" + op_a + " + " + op_b + ") >> 16) + " + op_c + ')';
break;
case Tegra::Shader::IAdd3Mode::LeftShift:
result = "(((" + op_a + " + " + op_b + ") << 16) + " + op_c + ')';
break;
default:
result = '(' + op_a + " + " + op_b + " + " + op_c + ')';
break;
}
} else {
result = '(' + op_a + " + " + op_b + " + " + op_c + ')';
}
regs.SetRegisterToInteger(instr.gpr0, true, 0, result, 1, 1);
break;
}
case OpCode::Id::ISCADD_C: case OpCode::Id::ISCADD_C:
case OpCode::Id::ISCADD_R: case OpCode::Id::ISCADD_R:
case OpCode::Id::ISCADD_IMM: { case OpCode::Id::ISCADD_IMM: {