eden-emu/xbyak
20
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add vreduceph

Browse source
This commit is contained in:
MITSUNARI Shigeo 2021-09-09 10:58:46 +09:00
parent 37bf3bb49d
commit 043c94ae95
3 changed files with 21 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

1
gen/gen_avx512.cpp
View file

@ -679,6 +679,7 @@ void putX_XM_IMM()
{ 0x56, "vreducepd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_SAE_Z, true }, { 0x56, "vreducepd", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW1 | T_B64 | T_SAE_Z, true },
{ 0x56, "vreduceps", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32 | T_SAE_Z, true }, { 0x56, "vreduceps", T_66 | T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B32 | T_SAE_Z, true },
{ 0x56, "vreduceph", T_0F3A | T_YMM | T_MUST_EVEX | T_EW0 | T_B16 | T_SAE_Z, true },
{ 0x54, "vpopcntb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, false }, { 0x54, "vpopcntb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z, false },
{ 0x54, "vpopcntw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false }, { 0x54, "vpopcntw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z, false },

19
test/misc.cpp
View file

@ -1006,6 +1006,16 @@ CYBOZU_TEST_AUTO(vaddph)
vscalefsh(xmm1, xmm5, ptr [rax+0x40]); vscalefsh(xmm1, xmm5, ptr [rax+0x40]);
vscalefsh(xmm1|k1|T_z|T_rd_sae, xmm5, xmm7); vscalefsh(xmm1|k1|T_z|T_rd_sae, xmm5, xmm7);
vreduceph(xmm1, ptr [rax+0x40], 0x1);
vreduceph(xmm1, ptr_b [rax+0x40], 0x2);
vreduceph(ymm1, ptr [rax+0x40], 0x3);
vreduceph(ymm1, ptr_b [rax+0x40], 0x4);
vreduceph(zmm1, ptr [rax+0x40], 0x5);
vreduceph(zmm1, ptr_b [rax+0x40], 0x6);
vreduceph(zmm1|k1|T_z|T_sae, zmm5, 0x7);
@ -1176,6 +1186,15 @@ CYBOZU_TEST_AUTO(vaddph)
// vscalefsh // vscalefsh
0x62, 0xf6, 0x55, 0x08, 0x2d, 0x48, 0x20, 0x62, 0xf6, 0x55, 0x08, 0x2d, 0x48, 0x20,
0x62, 0xf6, 0x55, 0xb9, 0x2d, 0xcf, 0x62, 0xf6, 0x55, 0xb9, 0x2d, 0xcf,
// vreduceph
0x62, 0xf3, 0x7c, 0x08, 0x56, 0x48, 0x04, 0x01,
0x62, 0xf3, 0x7c, 0x18, 0x56, 0x48, 0x20, 0x02,
0x62, 0xf3, 0x7c, 0x28, 0x56, 0x48, 0x02, 0x03,
0x62, 0xf3, 0x7c, 0x38, 0x56, 0x48, 0x20, 0x04,
0x62, 0xf3, 0x7c, 0x48, 0x56, 0x48, 0x01, 0x05,
0x62, 0xf3, 0x7c, 0x58, 0x56, 0x48, 0x20, 0x06,
0x62, 0xf3, 0x7c, 0x99, 0x56, 0xcd, 0x07,
}; };
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);

1
xbyak/xbyak_mnemonic.h
View file

@ -2186,6 +2186,7 @@ void vrcp28ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1
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 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 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 vreduceph(const Xmm& x, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(x, op, T_0F3A | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B16, 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); }
void vreducess(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_N4 | T_66 | T_0F3A | T_EW0 | T_SAE_X | T_MUST_EVEX, 0x57, imm); } void vreducess(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_N4 | T_66 | T_0F3A | T_EW0 | T_SAE_X | T_MUST_EVEX, 0x57, imm); }