eden-emu/xbyak
19
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add vperm{b,w,d}

Browse source
This commit is contained in:
MITSUNARI Shigeo 2016-07-14 12:02:41 +09:00
parent 4bc46763ab
commit b7d044b7bc
5 changed files with 22 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

3
gen/gen_avx512.cpp
View file

@ -198,6 +198,9 @@ void putX_X_XM_IMM()
{ 0xEF, "vpxorq", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
{ 0x40, "vpmullq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
{ 0x8D, "vpermb", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0, false },
{ 0x8D, "vpermw", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, false },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];

13
gen/gen_code.cpp
View file

@ -1319,15 +1319,18 @@ void put()
}
// vpermd, vpermps
{
const struct {
const char *suf;
const struct Tbl {
uint8 code;
const char *name;
int type;
} tbl[] = {
{ "d", 0x36 },
{ "ps", 0x16 },
{ 0x36, "vpermd", T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_EW0 | T_B32 },
{ 0x16, "vpermps", T_0F38 | T_66 | T_W0 | T_YMM },
};
for (int i = 0; i < NUM_OF_ARRAY(tbl); i++) {
printf("void vperm%s(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x%02X); }\n", tbl[i].suf, tbl[i].code);
const Tbl& p = tbl[i];
std::string type = type2String(p.type);
printf("void %s(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, %s, 0x%02X); }\n", p.name, type.c_str(), p.code);
}
}
// vpermq, vpermpd

8
test/make_nm.cpp
View file

@ -2983,6 +2983,14 @@ public:
{ "vextractps", REG32 | _MEM, _XMM3, IMM8 },
{ "vpermb", XMM_KZ, _XMM, _XMM },
{ "vpermb", ZMM_KZ, _ZMM, _ZMM | _MEM },
{ "vpermw", XMM_KZ, _XMM, _XMM },
{ "vpermw", ZMM_KZ, _ZMM, _ZMM | _MEM },
{ "vpermd", YMM_KZ, _YMM, _YMM | M_1to8 },
{ "vpermd", ZMM_KZ, _ZMM, _ZMM | M_1to16 },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];

2
xbyak/xbyak_avx512.h
View file

@ -109,6 +109,8 @@ void vporq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x
void vpxord(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0xEF); }
void vpxorq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0xEF); }
void vpmullq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x40); }
void vpermb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW0 | T_YMM | T_MUST_EVEX, 0x8D); }
void vpermw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX, 0x8D); }
void vpsraq(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_X_XM(x.copyAndSetIdx(4), x, op, T_0F | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x72, imm); }
void vextractf32x4(const Operand& op, const Ymm& r, uint8 imm) { opAVX_X_X_XMcvt(r, true, cvtIdx0(r), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_EW0 | T_YMM | T_MUST_EVEX, 0x19, imm); }
void vextractf64x2(const Operand& op, const Ymm& r, uint8 imm) { opAVX_X_X_XMcvt(r, true, cvtIdx0(r), op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_EW1 | T_YMM | T_MUST_EVEX, 0x19, imm); }

2
xbyak/xbyak_mnemonic.h
View file

@ -1037,7 +1037,7 @@ void vpmaskmovd(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_X
void vpmaskmovd(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x8E); }
void vpmaskmovq(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W1 | T_YMM, 0x8C); }
void vpmaskmovq(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W1 | T_YMM, 0x8E); }
void vpermd(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x36); }
void vpermd(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x36); }
void vpermps(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x16); }
void vpermq(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_YMM, 0x00, imm); }
void vpermpd(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_YMM, 0x01, imm); }