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

add vrcpph, vrcpsh

Browse source
This commit is contained in:
MITSUNARI Shigeo 2021-09-07 09:33:07 +09:00
parent 4f543ca0e4
commit b1ff7891d4
3 changed files with 26 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

4
gen/gen_avx512.cpp
View file

@ -331,6 +331,8 @@ void putX_X_XM_IMM()
{ 0x4D, "vrcp14sd", T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_N8, false }, { 0x4D, "vrcp14sd", T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_N8, false },
{ 0x4D, "vrcp14ss", T_66 | T_0F38 | T_MUST_EVEX | T_EW0 | T_N4, false }, { 0x4D, "vrcp14ss", T_66 | T_0F38 | T_MUST_EVEX | T_EW0 | T_N4, false },
{ 0x4D, "vrcpsh", T_66 | T_MAP6 | T_MUST_EVEX | T_EW0 | T_N2, false },
{ 0x4F, "vrsqrt14sd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N8, false }, { 0x4F, "vrsqrt14sd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N8, false },
{ 0x4F, "vrsqrt14ss", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, false }, { 0x4F, "vrsqrt14ss", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, false },
@ -652,6 +654,8 @@ void putX_XM_IMM()
{ 0x4C, "vrcp14pd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false }, { 0x4C, "vrcp14pd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
{ 0x4C, "vrcp14ps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false }, { 0x4C, "vrcp14ps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },
{ 0x4C, "vrcpph", T_66 | T_MAP6 | T_MUST_EVEX | T_YMM | T_EW0 | T_B16, false },
{ 0x4E, "vrsqrt14pd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false }, { 0x4E, "vrsqrt14pd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_B64, false },
{ 0x4E, "vrsqrt14ps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false }, { 0x4E, "vrsqrt14ps", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_B32, false },

20
test/misc.cpp
View file

@ -967,6 +967,15 @@ CYBOZU_TEST_AUTO(vaddph)
vfmulcph(xmm1, xmm2, ptr [rax+0x40]); vfmulcph(xmm1, xmm2, ptr [rax+0x40]);
vfmulcph(ymm1|k1|T_z, ymm2, ptr_b [rax+0x40]); vfmulcph(ymm1|k1|T_z, ymm2, ptr_b [rax+0x40]);
vfmulcph(zmm1, zmm2, ptr_b [rax+0x40]); vfmulcph(zmm1, zmm2, ptr_b [rax+0x40]);
vrcpph(xmm1, ptr [rax+0x40]);
vrcpph(xmm1, ptr_b [rax+0x40]);
vrcpph(ymm1, ptr [rax+0x40]);
vrcpph(ymm1, ptr_b [rax+0x40]);
vrcpph(zmm1, ptr [rax+0x40]);
vrcpph(zmm1, ptr_b [rax+0x40]);
vrcpsh(xmm1, xmm3, ptr [rax+0x40]);
} }
} c; } c;
const uint8_t tbl[] = { const uint8_t tbl[] = {
@ -1086,6 +1095,17 @@ CYBOZU_TEST_AUTO(vaddph)
0x62, 0xf6, 0x6e, 0x08, 0xd6, 0x48, 0x04, 0x62, 0xf6, 0x6e, 0x08, 0xd6, 0x48, 0x04,
0x62, 0xf6, 0x6e, 0xb9, 0xd6, 0x48, 0x10, 0x62, 0xf6, 0x6e, 0xb9, 0xd6, 0x48, 0x10,
0x62, 0xf6, 0x6e, 0x58, 0xd6, 0x48, 0x10, 0x62, 0xf6, 0x6e, 0x58, 0xd6, 0x48, 0x10,
// vrcpph
0x62, 0xf6, 0x7d, 0x08, 0x4c, 0x48, 0x04,
0x62, 0xf6, 0x7d, 0x18, 0x4c, 0x48, 0x20,
0x62, 0xf6, 0x7d, 0x28, 0x4c, 0x48, 0x02,
0x62, 0xf6, 0x7d, 0x38, 0x4c, 0x48, 0x20,
0x62, 0xf6, 0x7d, 0x48, 0x4c, 0x48, 0x01,
0x62, 0xf6, 0x7d, 0x58, 0x4c, 0x48, 0x20,
// vrcpsh
0x62, 0xf6, 0x65, 0x08, 0x4d, 0x48, 0x20,
}; };
const size_t n = sizeof(tbl) / sizeof(tbl[0]); const size_t n = sizeof(tbl) / sizeof(tbl[0]);
CYBOZU_TEST_EQUAL(c.getSize(), n); CYBOZU_TEST_EQUAL(c.getSize(), n);

2
xbyak/xbyak_mnemonic.h
View file

@ -2183,6 +2183,8 @@ void vrcp28pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_
void vrcp28ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xCA); } void vrcp28ps(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0 | T_B32 | T_SAE_Z, 0xCA); }
void vrcp28sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N8 | T_66 | T_0F38 | T_EW1 | T_SAE_X | T_MUST_EVEX, 0xCB); } void vrcp28sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N8 | T_66 | T_0F38 | T_EW1 | T_SAE_X | T_MUST_EVEX, 0xCB); }
void vrcp28ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N4 | T_66 | T_0F38 | T_EW0 | T_SAE_X | T_MUST_EVEX, 0xCB); } void vrcp28ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N4 | T_66 | T_0F38 | T_EW0 | T_SAE_X | T_MUST_EVEX, 0xCB); }
void vrcpph(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_MAP6 | T_EW0 | T_YMM | T_MUST_EVEX | T_B16, 0x4C); }
void vrcpsh(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N2 | T_66 | T_MAP6 | T_EW0 | T_MUST_EVEX, 0x4D); }
void vreducepd(const Xmm& x, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x56, imm); } void vreducepd(const Xmm& x, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(x, op, T_66 | T_0F3A | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B64, 0x56, imm); }
void vreduceps(const Xmm& x, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x56, imm); } void vreduceps(const Xmm& x, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x56, imm); }
void vreducesd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_N8 | T_66 | T_0F3A | T_EW1 | T_SAE_X | T_MUST_EVEX, 0x57, imm); } void vreducesd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_N8 | T_66 | T_0F3A | T_EW1 | T_SAE_X | T_MUST_EVEX, 0x57, imm); }