|
|
|
@ -17,33 +17,60 @@ u32 ShaderIR::DecodeBfe(NodeBlock& bb, u32 pc) { |
|
|
|
const Instruction instr = {program_code[pc]}; |
|
|
|
const auto opcode = OpCode::Decode(instr); |
|
|
|
|
|
|
|
UNIMPLEMENTED_IF(instr.bfe.negate_b); |
|
|
|
|
|
|
|
Node op_a = GetRegister(instr.gpr8); |
|
|
|
op_a = GetOperandAbsNegInteger(op_a, false, instr.bfe.negate_a, false); |
|
|
|
|
|
|
|
switch (opcode->get().GetId()) { |
|
|
|
case OpCode::Id::BFE_IMM: { |
|
|
|
UNIMPLEMENTED_IF_MSG(instr.generates_cc, |
|
|
|
"Condition codes generation in BFE is not implemented"); |
|
|
|
Node op_b = [&] { |
|
|
|
switch (opcode->get().GetId()) { |
|
|
|
case OpCode::Id::BFE_R: |
|
|
|
return GetRegister(instr.gpr20); |
|
|
|
case OpCode::Id::BFE_C: |
|
|
|
return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset()); |
|
|
|
case OpCode::Id::BFE_IMM: |
|
|
|
return Immediate(instr.alu.GetSignedImm20_20()); |
|
|
|
default: |
|
|
|
UNREACHABLE(); |
|
|
|
return Immediate(0); |
|
|
|
} |
|
|
|
}(); |
|
|
|
|
|
|
|
const Node inner_shift_imm = Immediate(static_cast<u32>(instr.bfe.GetLeftShiftValue())); |
|
|
|
const Node outer_shift_imm = |
|
|
|
Immediate(static_cast<u32>(instr.bfe.GetLeftShiftValue() + instr.bfe.shift_position)); |
|
|
|
UNIMPLEMENTED_IF_MSG(instr.bfe.rd_cc, "Condition codes in BFE is not implemented"); |
|
|
|
|
|
|
|
const Node inner_shift = |
|
|
|
Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, op_a, inner_shift_imm); |
|
|
|
const Node outer_shift = |
|
|
|
Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, inner_shift, outer_shift_imm); |
|
|
|
const bool is_signed = instr.bfe.is_signed; |
|
|
|
|
|
|
|
SetInternalFlagsFromInteger(bb, outer_shift, instr.generates_cc); |
|
|
|
SetRegister(bb, instr.gpr0, outer_shift); |
|
|
|
break; |
|
|
|
} |
|
|
|
default: |
|
|
|
UNIMPLEMENTED_MSG("Unhandled BFE instruction: {}", opcode->get().GetName()); |
|
|
|
// using reverse parallel method in
|
|
|
|
// https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel
|
|
|
|
// note for later if possible to implement faster method.
|
|
|
|
if (instr.bfe.brev) { |
|
|
|
const auto swap = [&](u32 s, u32 mask) { |
|
|
|
Node v1 = |
|
|
|
SignedOperation(OperationCode::ILogicalShiftRight, is_signed, op_a, Immediate(s)); |
|
|
|
if (mask != 0) { |
|
|
|
v1 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v1), |
|
|
|
Immediate(mask)); |
|
|
|
} |
|
|
|
Node v2 = op_a; |
|
|
|
if (mask != 0) { |
|
|
|
v2 = SignedOperation(OperationCode::IBitwiseAnd, is_signed, std::move(v2), |
|
|
|
Immediate(mask)); |
|
|
|
} |
|
|
|
v2 = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed, std::move(v2), |
|
|
|
Immediate(s)); |
|
|
|
return SignedOperation(OperationCode::IBitwiseOr, is_signed, std::move(v1), |
|
|
|
std::move(v2)); |
|
|
|
}; |
|
|
|
op_a = swap(1, 0x55555555U); |
|
|
|
op_a = swap(2, 0x33333333U); |
|
|
|
op_a = swap(4, 0x0F0F0F0FU); |
|
|
|
op_a = swap(8, 0x00FF00FFU); |
|
|
|
op_a = swap(16, 0); |
|
|
|
} |
|
|
|
|
|
|
|
const auto offset = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b, |
|
|
|
Immediate(0), Immediate(8)); |
|
|
|
const auto bits = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_b, |
|
|
|
Immediate(8), Immediate(8)); |
|
|
|
auto result = SignedOperation(OperationCode::IBitfieldExtract, is_signed, op_a, offset, bits); |
|
|
|
SetRegister(bb, instr.gpr0, std::move(result)); |
|
|
|
|
|
|
|
return pc; |
|
|
|
} |
|
|
|
|
|
|
|
|