suyu/dynarmic
28
1
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity

emit_x64_vector_floating_point: Implement GetNonSignMaskVector

Browse source
This commit is contained in:
Merry 2023-08-28 12:57:20 +01:00
parent 92a47c8db2
commit ceea80dd59
1 changed files with 16 additions and 11 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

27
src/dynarmic/backend/x64/emit_x64_vector_floating_point.cpp
View file

@ -145,18 +145,24 @@ void HandleNaNs(BlockOfCode& code, EmitContext& ctx, bool fpcr_controlled, std::
template<size_t fsize> template<size_t fsize>
Xbyak::Address GetVectorOf(BlockOfCode& code, u64 value) { Xbyak::Address GetVectorOf(BlockOfCode& code, u64 value) {
if constexpr (fsize == 32) { if constexpr (fsize == 16) {
return code.MConst(xword, (value << 48) | (value << 32) | (value << 16) | value, (value << 48) | (value << 32) | (value << 16) | value);
} else if constexpr (fsize == 32) {
return code.MConst(xword, (value << 32) | value, (value << 32) | value); return code.MConst(xword, (value << 32) | value, (value << 32) | value);
} else { } else {
static_assert(fsize == 64);
return code.MConst(xword, value, value); return code.MConst(xword, value, value);
} }
} }
template<size_t fsize, u64 value> template<size_t fsize, u64 value>
Xbyak::Address GetVectorOf(BlockOfCode& code) { Xbyak::Address GetVectorOf(BlockOfCode& code) {
if constexpr (fsize == 32) { if constexpr (fsize == 16) {
return code.MConst(xword, (value << 48) | (value << 32) | (value << 16) | value, (value << 48) | (value << 32) | (value << 16) | value);
} else if constexpr (fsize == 32) {
return code.MConst(xword, (value << 32) | value, (value << 32) | value); return code.MConst(xword, (value << 32) | value, (value << 32) | value);
} else { } else {
static_assert(fsize == 64);
return code.MConst(xword, value, value); return code.MConst(xword, value, value);
} }
} }
@ -173,6 +179,13 @@ Xbyak::Address GetNegativeZeroVector(BlockOfCode& code) {
return GetVectorOf<fsize, FP::FPInfo<FPT>::Zero(true)>(code); return GetVectorOf<fsize, FP::FPInfo<FPT>::Zero(true)>(code);
} }
template<size_t fsize>
Xbyak::Address GetNonSignMaskVector(BlockOfCode& code) {
using FPT = mcl::unsigned_integer_of_size<fsize>;
constexpr FPT non_sign_mask = FP::FPInfo<FPT>::exponent_mask | FP::FPInfo<FPT>::mantissa_mask;
return GetVectorOf<fsize, non_sign_mask>(code);
}
template<size_t fsize> template<size_t fsize>
Xbyak::Address GetSmallestNormalVector(BlockOfCode& code) { Xbyak::Address GetSmallestNormalVector(BlockOfCode& code) {
using FPT = mcl::unsigned_integer_of_size<fsize>; using FPT = mcl::unsigned_integer_of_size<fsize>;
@ -586,17 +599,9 @@ void EmitFourOpFallback(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, Lam
template<size_t fsize> template<size_t fsize>
void FPVectorAbs(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) { void FPVectorAbs(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst) {
using FPT = mcl::unsigned_integer_of_size<fsize>;
constexpr FPT non_sign_mask = FP::FPInfo<FPT>::sign_mask - FPT(1u);
constexpr u64 non_sign_mask64 = mcl::bit::replicate_element<fsize, u64>(non_sign_mask);
auto args = ctx.reg_alloc.GetArgumentInfo(inst); auto args = ctx.reg_alloc.GetArgumentInfo(inst);
const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]); const Xbyak::Xmm a = ctx.reg_alloc.UseScratchXmm(args[0]);
const Xbyak::Address mask = code.MConst(xword, non_sign_mask64, non_sign_mask64); code.andps(a, GetNonSignMaskVector<fsize>(code));
code.andps(a, mask);
ctx.reg_alloc.DefineValue(inst, a); ctx.reg_alloc.DefineValue(inst, a);
} }