mirrors/dynarmic
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

emit_x64_{vector_}floating_point: Add AVX512 implementation for DenormalsAreZero

Browse source 
Both single and double precision floating point numbers as well as the
packed and unpacked version of this instruction will be able to use the
same memory constant. This takes advantage of the fact that `VFIXUPIMM*`
doesn't just copy from the source, but it will convert to `0.0` if it
turns out that it is a denormal and the `MXCSR.DAZ` flag is set.

```
tsrc[31:0]←((src1[30:23] = 0) AND (MXCSR.DAZ =1)) ? 0.0 : src1[31:0]
...
CASE(token_response[3:0]) {
    ...
    0001: dest[31:0]←tsrc[31:0]; ; pass through src1 normal input value, denormal as zero
    ...
```
This commit is contained in:
Wunkolo 2022-06-18 00:02:59 -07:00 committed by merry
parent 3ed2aebb20
commit 6367a26e62
2 changed files with 43 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

22
src/dynarmic/backend/x64/emit_x64_floating_point.cpp
View file

@ -16,6 +16,7 @@
#include <mcl/mp/typelist/lower_to_tuple.hpp>
#include <mcl/stdint.hpp>
#include <mcl/type_traits/integer_of_size.hpp>
#include <xbyak/xbyak.h>
#include "dynarmic/backend/x64/abi.h"
#include "dynarmic/backend/x64/block_of_code.h"
@ -79,6 +80,27 @@ constexpr u64 f64_max_s64_lim = 0x43e0000000000000u; // 2^63 as a double (actua
template<size_t fsize>
void DenormalsAreZero(BlockOfCode& code, EmitContext& ctx, std::initializer_list<Xbyak::Xmm> to_daz) {
if (ctx.FPCR().FZ()) {
if (code.HasHostFeature(HostFeature::AVX512_OrthoFloat)) {
constexpr u32 denormal_to_zero = FixupLUT(
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src);
constexpr u64 denormal_to_zero64 = mcl::bit::replicate_element<fsize, u64>(denormal_to_zero);
const Xbyak::Xmm tmp = xmm16;
FCODE(vmovap)(tmp, code.MConst(xword, u64(denormal_to_zero64), u64(denormal_to_zero64)));
for (const Xbyak::Xmm& xmm : to_daz) {
FCODE(vfixupimms)(xmm, xmm, tmp, u8(0));
}
return;
}
for (const Xbyak::Xmm& xmm : to_daz) {
code.movaps(xmm0, code.MConst(xword, fsize == 32 ? f32_non_sign_mask : f64_non_sign_mask));
code.andps(xmm0, xmm);

21
src/dynarmic/backend/x64/emit_x64_vector_floating_point.cpp
View file

@ -18,6 +18,7 @@
#include <mcl/mp/typelist/lower_to_tuple.hpp>
#include <mcl/type_traits/function_info.hpp>
#include <mcl/type_traits/integer_of_size.hpp>
#include <xbyak/xbyak.h>
#include "dynarmic/backend/x64/abi.h"
#include "dynarmic/backend/x64/block_of_code.h"
@ -223,6 +224,26 @@ void ZeroIfNaN(BlockOfCode& code, Xbyak::Xmm result) {
template<size_t fsize>
void DenormalsAreZero(BlockOfCode& code, FP::FPCR fpcr, std::initializer_list<Xbyak::Xmm> to_daz, Xbyak::Xmm tmp) {
if (fpcr.FZ()) {
if (code.HasHostFeature(HostFeature::AVX512_OrthoFloat)) {
constexpr u32 denormal_to_zero = FixupLUT(
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src,
FpFixup::Norm_Src);
constexpr u64 denormal_to_zero64 = mcl::bit::replicate_element<fsize, u64>(denormal_to_zero);
FCODE(vmovap)(tmp, code.MConst(xword, u64(denormal_to_zero64), u64(denormal_to_zero64)));
for (const Xbyak::Xmm& xmm : to_daz) {
FCODE(vfixupimmp)(xmm, xmm, tmp, u8(0));
}
return;
}
if (fpcr.RMode() != FP::RoundingMode::TowardsMinusInfinity) {
code.movaps(tmp, GetNegativeZeroVector<fsize>(code));
} else {