mirror of
https://git.h3cjp.net/H3cJP/citra.git
synced 2024-11-24 02:42:49 +00:00
shader/shift: Implement SHR wrapped and clamped variants
Nvidia defaults to wrapped shifts, but this is undefined behaviour on OpenGL's spec. Explicitly mask/clamp according to what the guest shader requires.
This commit is contained in:
parent
922c7f4e51
commit
77ef4fa907
|
@ -674,6 +674,10 @@ union Instruction {
|
|||
BitField<48, 1, u64> is_signed;
|
||||
} shift;
|
||||
|
||||
union {
|
||||
BitField<39, 1, u64> wrap;
|
||||
} shr;
|
||||
|
||||
union {
|
||||
BitField<39, 5, u64> shift_amount;
|
||||
BitField<48, 1, u64> negate_b;
|
||||
|
|
|
@ -17,8 +17,8 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
|
|||
const Instruction instr = {program_code[pc]};
|
||||
const auto opcode = OpCode::Decode(instr);
|
||||
|
||||
const Node op_a = GetRegister(instr.gpr8);
|
||||
const Node op_b = [&]() {
|
||||
Node op_a = GetRegister(instr.gpr8);
|
||||
Node op_b = [&]() {
|
||||
if (instr.is_b_imm) {
|
||||
return Immediate(instr.alu.GetSignedImm20_20());
|
||||
} else if (instr.is_b_gpr) {
|
||||
|
@ -32,16 +32,23 @@ u32 ShaderIR::DecodeShift(NodeBlock& bb, u32 pc) {
|
|||
case OpCode::Id::SHR_C:
|
||||
case OpCode::Id::SHR_R:
|
||||
case OpCode::Id::SHR_IMM: {
|
||||
const Node value = SignedOperation(OperationCode::IArithmeticShiftRight,
|
||||
instr.shift.is_signed, PRECISE, op_a, op_b);
|
||||
if (instr.shr.wrap) {
|
||||
op_b = Operation(OperationCode::UBitwiseAnd, std::move(op_b), Immediate(0x1f));
|
||||
} else {
|
||||
op_b = Operation(OperationCode::IMax, std::move(op_b), Immediate(0));
|
||||
op_b = Operation(OperationCode::IMin, std::move(op_b), Immediate(31));
|
||||
}
|
||||
|
||||
Node value = SignedOperation(OperationCode::IArithmeticShiftRight, instr.shift.is_signed,
|
||||
std::move(op_a), std::move(op_b));
|
||||
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
|
||||
SetRegister(bb, instr.gpr0, value);
|
||||
SetRegister(bb, instr.gpr0, std::move(value));
|
||||
break;
|
||||
}
|
||||
case OpCode::Id::SHL_C:
|
||||
case OpCode::Id::SHL_R:
|
||||
case OpCode::Id::SHL_IMM: {
|
||||
const Node value = Operation(OperationCode::ILogicalShiftLeft, PRECISE, op_a, op_b);
|
||||
const Node value = Operation(OperationCode::ILogicalShiftLeft, op_a, op_b);
|
||||
SetInternalFlagsFromInteger(bb, value, instr.generates_cc);
|
||||
SetRegister(bb, instr.gpr0, value);
|
||||
break;
|
||||
|
|
Loading…
Reference in a new issue