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

add tests of avx-vnni-int{8,16}

Browse source
This commit is contained in:
MITSUNARI Shigeo 2023-08-07 10:59:12 +09:00
parent 4a6132d660
commit dd66cfb764
3 changed files with 78 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

8
gen/gen_code.cpp
View file

@ -1817,6 +1817,7 @@ void put()
} }
} }
// avx-vnni-int8 // avx-vnni-int8
// avx-vnni-int16
{ {
const struct Tbl { const struct Tbl {
uint8_t code; uint8_t code;
@ -1829,6 +1830,13 @@ void put()
{ 0x51, "vpdpbsuds", T_F3 | T_0F38 | T_W0 | T_YMM }, { 0x51, "vpdpbsuds", T_F3 | T_0F38 | T_W0 | T_YMM },
{ 0x50, "vpdpbuud", T_0F38 | T_W0 | T_YMM }, { 0x50, "vpdpbuud", T_0F38 | T_W0 | T_YMM },
{ 0x51, "vpdpbuuds", T_0F38 | T_W0 | T_YMM }, { 0x51, "vpdpbuuds", T_0F38 | T_W0 | T_YMM },
{ 0xD2, "vpdpwsud", T_F3 | T_0F38 | T_W0 | T_YMM },
{ 0xD3, "vpdpwsuds", T_F3 | T_0F38 | T_W0 | T_YMM },
{ 0xD2, "vpdpwusd", T_66 | T_0F38 | T_W0 | T_YMM },
{ 0xD3, "vpdpwusds", T_66 | T_0F38 | T_W0 | T_YMM },
{ 0xD2, "vpdpwuud", T_0F38 | T_W0 | T_YMM },
{ 0xD3, "vpdpwuuds", T_0F38 | T_W0 | T_YMM },
}; };
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) { for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i]; const Tbl *p = &tbl[i];

64
test/misc.cpp
View file

@ -2217,4 +2217,68 @@ CYBOZU_TEST_AUTO(crypto)
CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n); CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
} }
CYBOZU_TEST_AUTO(avx_vnni_int)
{
struct Code : Xbyak::CodeGenerator {
Code()
{
vpdpbssd(xmm1, xmm2, xmm3);
vpdpbssd(ymm1, ymm2, ptr [rax]);
vpdpbssds(xmm1, xmm2, xmm3);
vpdpbssds(ymm1, ymm2, ptr [rax]);
vpdpbsud(xmm1, xmm2, xmm3);
vpdpbsud(ymm1, ymm2, ptr [rax]);
vpdpbsuds(xmm1, xmm2, xmm3);
vpdpbsuds(ymm1, ymm2, ptr [rax]);
vpdpbuud(xmm1, xmm2, xmm3);
vpdpbuud(ymm1, ymm2, ptr [rax]);
vpdpbuuds(xmm1, xmm2, xmm3);
vpdpbuuds(ymm1, ymm2, ptr [rax]);
vpdpwsud(xmm1, xmm2, xmm3);
vpdpwsud(ymm1, ymm2, ptr [rax]);
vpdpwsuds(xmm1, xmm2, xmm3);
vpdpwsuds(ymm1, ymm2, ptr [rax]);
vpdpwusd(xmm1, xmm2, xmm3);
vpdpwusd(ymm1, ymm2, ptr [rax]);
vpdpwusds(xmm1, xmm2, xmm3);
vpdpwusds(ymm1, ymm2, ptr [rax]);
vpdpwuud(xmm1, xmm2, xmm3);
vpdpwuud(ymm1, ymm2, ptr [rax]);
vpdpwuuds(xmm1, xmm2, xmm3);
vpdpwuuds(ymm1, ymm2, ptr [rax]);
}
} c;
const uint8_t tbl[] = {
0xc4, 0xe2, 0x6b, 0x50, 0xcb,
0xc4, 0xe2, 0x6f, 0x50, 0x08,
0xc4, 0xe2, 0x6b, 0x51, 0xcb,
0xc4, 0xe2, 0x6f, 0x51, 0x08,
0xc4, 0xe2, 0x6a, 0x50, 0xcb,
0xc4, 0xe2, 0x6e, 0x50, 0x08,
0xc4, 0xe2, 0x6a, 0x51, 0xcb,
0xc4, 0xe2, 0x6e, 0x51, 0x08,
0xc4, 0xe2, 0x68, 0x50, 0xcb,
0xc4, 0xe2, 0x6c, 0x50, 0x08,
0xc4, 0xe2, 0x68, 0x51, 0xcb,
0xc4, 0xe2, 0x6c, 0x51, 0x08,
0xc4, 0xe2, 0x6a, 0xd2, 0xcb,
0xc4, 0xe2, 0x6e, 0xd2, 0x08,
0xc4, 0xe2, 0x6a, 0xd3, 0xcb,
0xc4, 0xe2, 0x6e, 0xd3, 0x08,
0xc4, 0xe2, 0x69, 0xd2, 0xcb,
0xc4, 0xe2, 0x6d, 0xd2, 0x08,
0xc4, 0xe2, 0x69, 0xd3, 0xcb,
0xc4, 0xe2, 0x6d, 0xd3, 0x08,
0xc4, 0xe2, 0x68, 0xd2, 0xcb,
0xc4, 0xe2, 0x6c, 0xd2, 0x08,
0xc4, 0xe2, 0x68, 0xd3, 0xcb,
0xc4, 0xe2, 0x6c, 0xd3, 0x08,
};
const size_t n = sizeof(tbl) / sizeof(tbl[0]);
CYBOZU_TEST_EQUAL(c.getSize(), n);
CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
}
#endif #endif

6
xbyak/xbyak_mnemonic.h
View file

@ -1219,6 +1219,12 @@ void vpdpbuud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1
void vpdpbuuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0x51); } void vpdpbuuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0x51); }
void vpdpwssd(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x52, encoding); } void vpdpwssd(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x52, encoding); }
void vpdpwssds(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x53, encoding); } void vpdpwssds(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x53, encoding); }
void vpdpwsud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_YMM, 0xD2); }
void vpdpwsuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_YMM, 0xD3); }
void vpdpwusd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_YMM, 0xD2); }
void vpdpwusds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_YMM, 0xD3); }
void vpdpwuud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0xD2); }
void vpdpwuuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0xD3); }
void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { if (!(y1.isYMM() && y2.isYMM() && op.isYMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(y1, &y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x06, imm); } void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { if (!(y1.isYMM() && y2.isYMM() && op.isYMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(y1, &y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x06, imm); }
void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { if (!(y1.isYMM() && y2.isYMM() && op.isYMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(y1, &y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x46, imm); } void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { if (!(y1.isYMM() && y2.isYMM() && op.isYMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(y1, &y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x46, imm); }
void vpermd(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x36); } void vpermd(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x36); }