suyu/dynarmic
26
1
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity

frontend/ir_emitter: Add half-precision variant of FPVectorRoundInt

Browse source
This commit is contained in:
Lioncash 2019-04-13 17:49:04 -04:00 committed by MerryMage
parent ad0c698f89
commit 5b4673da4b
3 changed files with 34 additions and 22 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
src/backend/x64/emit_x64_vector_floating_point.cpp
View file

@ -1160,28 +1160,30 @@ void EmitFPVectorRoundInt(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
const auto rounding = static_cast<FP::RoundingMode>(inst->GetArg(1).GetU8()); const auto rounding = static_cast<FP::RoundingMode>(inst->GetArg(1).GetU8());
const bool exact = inst->GetArg(2).GetU1(); const bool exact = inst->GetArg(2).GetU1();
if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41) && rounding != FP::RoundingMode::ToNearest_TieAwayFromZero && !exact) { if constexpr (fsize != 16) {
const u8 round_imm = [&]() -> u8 { if (code.DoesCpuSupport(Xbyak::util::Cpu::tSSE41) && rounding != FP::RoundingMode::ToNearest_TieAwayFromZero && !exact) {
switch (rounding) { const u8 round_imm = [&]() -> u8 {
case FP::RoundingMode::ToNearest_TieEven: switch (rounding) {
return 0b00; case FP::RoundingMode::ToNearest_TieEven:
case FP::RoundingMode::TowardsPlusInfinity: return 0b00;
return 0b10; case FP::RoundingMode::TowardsPlusInfinity:
case FP::RoundingMode::TowardsMinusInfinity: return 0b10;
return 0b01; case FP::RoundingMode::TowardsMinusInfinity:
case FP::RoundingMode::TowardsZero: return 0b01;
return 0b11; case FP::RoundingMode::TowardsZero:
default: return 0b11;
UNREACHABLE(); default:
} UNREACHABLE();
return 0; }
}(); return 0;
}();
EmitTwoOpVectorOperation<fsize, DefaultIndexer>(code, ctx, inst, [&](const Xbyak::Xmm& result, const Xbyak::Xmm& xmm_a){ EmitTwoOpVectorOperation<fsize, DefaultIndexer>(code, ctx, inst, [&](const Xbyak::Xmm& result, const Xbyak::Xmm& xmm_a){
FCODE(roundp)(result, xmm_a, round_imm); FCODE(roundp)(result, xmm_a, round_imm);
}); });
return; return;
}
} }
using rounding_list = mp::list< using rounding_list = mp::list<
@ -1218,6 +1220,10 @@ void EmitFPVectorRoundInt(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
EmitTwoOpFallback(code, ctx, inst, lut.at(std::make_tuple(rounding, exact))); EmitTwoOpFallback(code, ctx, inst, lut.at(std::make_tuple(rounding, exact)));
} }
void EmitX64::EmitFPVectorRoundInt16(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorRoundInt<16>(code, ctx, inst);
}
void EmitX64::EmitFPVectorRoundInt32(EmitContext& ctx, IR::Inst* inst) { void EmitX64::EmitFPVectorRoundInt32(EmitContext& ctx, IR::Inst* inst) {
EmitFPVectorRoundInt<32>(code, ctx, inst); EmitFPVectorRoundInt<32>(code, ctx, inst);
} }

9
src/frontend/ir/ir_emitter.cpp
View file

@ -2278,11 +2278,16 @@ U128 IREmitter::FPVectorRecipStepFused(size_t esize, const U128& a, const U128&
} }
U128 IREmitter::FPVectorRoundInt(size_t esize, const U128& operand, FP::RoundingMode rounding, bool exact) { U128 IREmitter::FPVectorRoundInt(size_t esize, const U128& operand, FP::RoundingMode rounding, bool exact) {
const IR::U8 rounding_imm = Imm8(static_cast<u8>(rounding));
const IR::U1 exact_imm = Imm1(exact);
switch (esize) { switch (esize) {
case 16:
return Inst<U128>(Opcode::FPVectorRoundInt16, operand, rounding_imm, exact_imm);
case 32: case 32:
return Inst<U128>(Opcode::FPVectorRoundInt32, operand, Imm8(static_cast<u8>(rounding)), Imm1(exact)); return Inst<U128>(Opcode::FPVectorRoundInt32, operand, rounding_imm, exact_imm);
case 64: case 64:
return Inst<U128>(Opcode::FPVectorRoundInt64, operand, Imm8(static_cast<u8>(rounding)), Imm1(exact)); return Inst<U128>(Opcode::FPVectorRoundInt64, operand, rounding_imm, exact_imm);
} }
UNREACHABLE(); UNREACHABLE();
return {}; return {};

1
src/frontend/ir/opcodes.inc
View file

@ -574,6 +574,7 @@ OPCODE(FPVectorRecipEstimate32, U128, U128
OPCODE(FPVectorRecipEstimate64, U128, U128 ) OPCODE(FPVectorRecipEstimate64, U128, U128 )
OPCODE(FPVectorRecipStepFused32, U128, U128, U128 ) OPCODE(FPVectorRecipStepFused32, U128, U128, U128 )
OPCODE(FPVectorRecipStepFused64, U128, U128, U128 ) OPCODE(FPVectorRecipStepFused64, U128, U128, U128 )
OPCODE(FPVectorRoundInt16, U128, U128, U8, U1 )
OPCODE(FPVectorRoundInt32, U128, U128, U8, U1 ) OPCODE(FPVectorRoundInt32, U128, U128, U8, U1 )
OPCODE(FPVectorRoundInt64, U128, U128, U8, U1 ) OPCODE(FPVectorRoundInt64, U128, U128, U8, U1 )
OPCODE(FPVectorRSqrtEstimate32, U128, U128 ) OPCODE(FPVectorRSqrtEstimate32, U128, U128 )