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

emit_x64_vector_saturation: Reduce esize noise in EmitVectorSignedSaturated

Browse source
This commit is contained in:
Merry 2021-05-28 15:03:45 +01:00
parent de31caca49
commit a76e8c8827
1 changed files with 26 additions and 46 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

72
src/dynarmic/backend/x64/emit_x64_vector_saturation.cpp
View file

@ -10,6 +10,24 @@
#include "dynarmic/ir/microinstruction.h"
#include "dynarmic/ir/opcodes.h"
#define FCODE(NAME) \
[&code](auto... args) { \
if constexpr (esize == 32) { \
code.NAME##s(args...); \
} else { \
code.NAME##d(args...); \
} \
}
#define ICODE(NAME) \
[&code](auto... args) { \
if constexpr (esize == 32) { \
code.NAME##d(args...); \
} else { \
code.NAME##q(args...); \
} \
}
namespace Dynarmic::Backend::X64 {
using namespace Xbyak::util;
@ -65,43 +83,23 @@ void EmitVectorSignedSaturated(BlockOfCode& code, EmitContext& ctx, IR::Inst* in
code.movaps(tmp, result);
if (code.HasHostFeature(HostFeature::AVX512_Ortho | HostFeature::AVX512DQ)) {
// Do a regular unsigned operation
if constexpr (op == Op::Add) {
if constexpr (esize == 32) {
code.vpaddd(result, result, arg);
} else {
code.vpaddq(result, result, arg);
}
} else {
if constexpr (esize == 32) {
code.vpsubd(result, result, arg);
} else {
code.vpsubq(result, result, arg);
}
}
// Determine if an overflow/underflow happened
if constexpr (op == Op::Add) {
ICODE(vpadd)(result, result, arg);
code.vpternlogd(tmp, result, arg, 0b00100100);
} else {
ICODE(vpsub)(result, result, arg);
code.vpternlogd(tmp, result, arg, 0b00011000);
}
// Masked write
if constexpr (esize == 32) {
code.vpmovd2m(k1, tmp);
code.vpsrad(result | k1, result, 31);
code.vpxord(result | k1, result, code.MConst(xword_b, msb_mask, msb_mask));
} else {
code.vpmovq2m(k1, tmp);
code.vpsraq(result | k1, result, 63);
code.vpxorq(result | k1, result, code.MConst(xword_b, msb_mask, msb_mask));
}
ICODE(vpsra)(result | k1, result, u8(esize - 1));
ICODE(vpxor)(result | k1, result, code.MConst(xword_b, msb_mask, msb_mask));
// Set ZF if an overflow happened
code.ktestb(k1, k1);
// Write Q if overflow/underflow occured
code.setnz(overflow);
code.or_(code.byte[code.r15 + code.GetJitStateInfo().offsetof_fpsr_qc], overflow);
@ -109,22 +107,12 @@ void EmitVectorSignedSaturated(BlockOfCode& code, EmitContext& ctx, IR::Inst* in
return;
}
// TODO AVX2 implementation
code.movaps(xmm0, result);
if constexpr (op == Op::Add) {
if constexpr (esize == 32) {
code.paddd(result, arg);
} else {
code.paddq(result, arg);
}
ICODE(padd)(result, arg);
} else {
if constexpr (esize == 32) {
code.psubd(result, arg);
} else {
code.psubq(result, arg);
}
ICODE(psub)(result, arg);
}
code.pxor(tmp, result);
@ -145,22 +133,14 @@ void EmitVectorSignedSaturated(BlockOfCode& code, EmitContext& ctx, IR::Inst* in
if (code.HasHostFeature(HostFeature::SSE41)) {
code.ptest(xmm0, code.MConst(xword, msb_mask, msb_mask));
} else {
if constexpr (esize == 32) {
code.movmskps(overflow.cvt32(), xmm0);
} else {
code.movmskpd(overflow.cvt32(), xmm0);
}
FCODE(movmskp)(overflow.cvt32(), xmm0);
code.test(overflow.cvt32(), overflow.cvt32());
}
code.setnz(overflow);
code.or_(code.byte[code.r15 + code.GetJitStateInfo().offsetof_fpsr_qc], overflow);
if (code.HasHostFeature(HostFeature::SSE41)) {
if constexpr (esize == 32) {
code.blendvps(result, tmp);
} else {
code.blendvpd(result, tmp);
}
FCODE(blendvp)(result, tmp);
ctx.reg_alloc.DefineValue(inst, result);
} else {