diff options
| author | MITSUNARI Shigeo <[email protected]> | 2024-10-13 11:30:39 +0900 |
|---|---|---|
| committer | MITSUNARI Shigeo <[email protected]> | 2024-10-13 13:51:06 +0900 |
| commit | 749aa31dd4a31acf5095cc7db8dfd2d24f0b8787 (patch) | |
| tree | ac6afe00babdf67db73eb342c0d74b5f9a04bac3 | |
| parent | dedb7f52af9ee648c41ae27fb48978a5ae68901e (diff) | |
| download | xbyak-749aa31dd4a31acf5095cc7db8dfd2d24f0b8787.tar.gz xbyak-749aa31dd4a31acf5095cc7db8dfd2d24f0b8787.zip | |
mov vcvtneps2bf16 to avx512
| -rw-r--r-- | gen/gen_avx512.cpp | 2 | ||||
| -rw-r--r-- | gen/gen_code.cpp | 1 | ||||
| -rw-r--r-- | xbyak/xbyak_mnemonic.h | 2 |
3 files changed, 3 insertions, 2 deletions
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp index e7b2336..109afc6 100644 --- a/gen/gen_avx512.cpp +++ b/gen/gen_avx512.cpp @@ -619,6 +619,8 @@ void putCvt() { 0x2A, "vcvtsi2sh", T_F3 | T_MAP5 | T_MUST_EVEX | T_ER_R | T_M_K, 6 }, { 0x7B, "vcvtusi2sh", T_F3 | T_MAP5 | T_MUST_EVEX | T_ER_R | T_M_K, 6 }, + + { 0x72, "vcvtneps2bf16", T_MUST_EVEX | T_F3 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 2 }, }; for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) { const Tbl& p = tbl[i]; diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp index c7ee311..764d118 100644 --- a/gen/gen_code.cpp +++ b/gen/gen_code.cpp @@ -1802,7 +1802,6 @@ void put() const Tbl& p = tbl[i]; printf("void %s(const Xmm& x, const Address& addr) { opVex(x, 0, addr, %s, 0x%02X); }\n", p.name, type2String(p.type).c_str(), p.code); } - printf("void vcvtneps2bf16(const Xmm& x, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opCvt2(x, op, %s|orEvexIf(encoding), 0x72); }\n", type2String(T_F3 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32).c_str()); } // haswell gpr(reg, reg, r/m) { diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h index cc88238..daafcd1 100644 --- a/xbyak/xbyak_mnemonic.h +++ b/xbyak/xbyak_mnemonic.h @@ -1213,7 +1213,6 @@ void vcvtneebf162ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_F3| void vcvtneeph2ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_66|T_0F38|T_W0|T_YMM, 0xB0); } void vcvtneobf162ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_F2|T_0F38|T_W0|T_YMM, 0xB0); } void vcvtneoph2ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_0F38|T_W0|T_YMM, 0xB0); } -void vcvtneps2bf16(const Xmm& x, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opCvt2(x, op, T_F3|T_0F38|T_EW0|T_YMM|T_SAE_Z|T_B32|orEvexIf(encoding), 0x72); } void vcvtpd2dq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_F2 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0xE6); } void vcvtpd2ps(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0x5A); } void vcvtph2ps(const Xmm& x, const Operand& op) { checkCvt1(x, op); opVex(x, 0, op, T_0F38 | T_66 | T_W0 | T_EVEX | T_EW0 | T_N8 | T_N_VL | T_SAE_Y, 0x13); } @@ -2220,6 +2219,7 @@ void vcvtneph2bf8(const Xmm& x, const Operand& op) { opCvt2(x, op, T_F3|T_0F38|T void vcvtneph2bf8s(const Xmm& x, const Operand& op) { opCvt2(x, op, T_F3|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x74); } void vcvtneph2hf8(const Xmm& x, const Operand& op) { opCvt2(x, op, T_F3|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x18); } void vcvtneph2hf8s(const Xmm& x, const Operand& op) { opCvt2(x, op, T_F3|T_MAP5|T_EW0|T_YMM|T_MUST_EVEX|T_B16, 0x1B); } +void vcvtneps2bf16(const Xmm& x, const Operand& op) { opCvt2(x, op, T_F3|T_0F38|T_EW0|T_YMM|T_SAE_Z|T_MUST_EVEX|T_B32, 0x72); } void vcvtpd2ph(const Xmm& x, const Operand& op) { opCvt5(x, op, T_N16|T_N_VL|T_66|T_MAP5|T_EW1|T_ER_Z|T_MUST_EVEX|T_B64, 0x5A); } void vcvtpd2qq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66|T_0F|T_EW1|T_YMM|T_ER_Z|T_MUST_EVEX|T_B64, 0x7B); } void vcvtpd2udq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F|T_EW1|T_YMM|T_ER_Z|T_MUST_EVEX|T_B64, 0x79); } |