diff options
| author | MITSUNARI Shigeo <[email protected]> | 2024-10-13 11:16:27 +0900 |
|---|---|---|
| committer | MITSUNARI Shigeo <[email protected]> | 2024-10-13 13:51:06 +0900 |
| commit | dedb7f52af9ee648c41ae27fb48978a5ae68901e (patch) | |
| tree | e1b6709531172ad9538d11cee9b94937f3c15048 | |
| parent | 6e083527507052145d99585eb028342cba81df60 (diff) | |
| download | xbyak-dedb7f52af9ee648c41ae27fb48978a5ae68901e.tar.gz xbyak-dedb7f52af9ee648c41ae27fb48978a5ae68901e.zip | |
add vdpphps
| -rw-r--r-- | gen/gen_avx512.cpp | 4 | ||||
| -rw-r--r-- | test/avx10/misc.txt | 13 | ||||
| -rw-r--r-- | xbyak/xbyak_mnemonic.h | 1 |
3 files changed, 18 insertions, 0 deletions
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp index 44ef945..e7b2336 100644 --- a/gen/gen_avx512.cpp +++ b/gen/gen_avx512.cpp @@ -445,6 +445,9 @@ void putX_X_XM_IMM() { 0x74, "vcvtne2ph2bf8s", T_MUST_EVEX | T_F2 | T_MAP5 | T_EW0 | T_YMM | T_B16 | T_N1, false }, { 0x18, "vcvtne2ph2hf8", T_MUST_EVEX | T_F2 | T_MAP5 | T_EW0 | T_YMM | T_B16 | T_N1, false }, { 0x1B, "vcvtne2ph2hf8s", T_MUST_EVEX | T_F2 | T_MAP5 | T_EW0 | T_YMM | T_B16 | T_N1, false }, + + { 0x52, "vdpphps", T_MUST_EVEX | T_0F38 | T_EW0 | T_YMM | T_B32, false }, +// { 0x42, "vmpsadbw", T_MUST_EVEX | T_F3 | T_0F3A | T_EW0 | T_YMM | T_B32, true }, }; for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) { const Tbl *p = &tbl[i]; @@ -452,6 +455,7 @@ void putX_X_XM_IMM() printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op%s) { opAVX_X_X_XM(x1, x2, op, %s, 0x%02X%s); }\n" , p->name, p->hasIMM ? ", uint8_t imm" : "", s.c_str(), p->code, p->hasIMM ? ", imm" : ""); } +// puts("void vmpsadbw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_0F3A | T_YMM, 0x42, encoding, imm, T_66 | T_W0 | T_YMM, T_F3 | T_EW0 | T_B32); }"); } void putShift() diff --git a/test/avx10/misc.txt b/test/avx10/misc.txt new file mode 100644 index 0000000..f7d1351 --- /dev/null +++ b/test/avx10/misc.txt @@ -0,0 +1,13 @@ +vdpphps(xm1, xm2, xm3); +vdpphps(xm1, xm2, ptr[rax+128]); +vdpphps(xm1, xm2, ptr_b[rax+128]); + +vdpphps(ym1, ym2, ym3); +vdpphps(ym1, ym2, ptr[rax+128]); +vdpphps(ym1, ym2, ptr_b[rax+128]); + +vdpphps(zm1, zm2, zm3); +vdpphps(zm1, zm2, ptr[rax+128]); +vdpphps(zm1, zm2, ptr_b[rax+128]); + +// skip vmpsadbw diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h index 891deea..cc88238 100644 --- a/xbyak/xbyak_mnemonic.h +++ b/xbyak/xbyak_mnemonic.h @@ -2280,6 +2280,7 @@ void vdivnepbf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM void vdivph(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_EW0 | T_YMM | T_MUST_EVEX | T_ER_Z | T_B16, 0x5E); } void vdivsh(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_MAP5 | T_F3 | T_EW0 | T_MUST_EVEX | T_ER_X | T_N2, 0x5E); } void vdpbf16ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3|T_0F38|T_EW0|T_YMM|T_SAE_Z|T_MUST_EVEX|T_B32, 0x52); } +void vdpphps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38|T_EW0|T_YMM|T_MUST_EVEX|T_B32, 0x52); } void vexp2pd(const Zmm& z, const Operand& op) { opAVX_X_XM_IMM(z, op, T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64 | T_SAE_Z, 0xC8); } void vexp2ps(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, 0xC8); } void vexpandpd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_N8|T_66|T_0F38|T_EW1|T_YMM|T_MUST_EVEX, 0x88); } |