fma is ok
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 05fb20c..6170483 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1435,8 +1435,10 @@
t |= T_B64;
} else if (suf == "ps") {
t |= T_B32;
- } else {
- t |= T_ER_X;
+ } else if (suf == "sd") {
+ t |= T_ER_X | T_N8;
+ } else { // ss
+ t |= T_ER_X | T_N4;
}
std::string type = type2String(t);
printf("void %s%s%s(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, %s, 0x%02X); }\n"
@@ -1483,17 +1485,17 @@
printf("void vstmxcsr(const Address& addr) { opAVX_X_X_XM(xm3, xm0, addr, T_0F, 0xAE); }\n");
printf("void vmaskmovdqu(const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x1, xm0, x2, T_0F | T_66, 0xF7); }\n");
- printf("void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(8|16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX, 0x14, imm); }\n");
- printf("void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX, 0x15, imm); } }\n");
- printf("void vpextrd(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0, 0x16, imm); }\n");
+ printf("void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(8|16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX | T_N4, 0x14, imm); }\n");
+ printf("void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX | T_N4, 0x15, imm); } }\n");
+ printf("void vpextrd(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x16, imm); }\n");
- printf("void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX, 0x20, imm); }\n");
+ printf("void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX | T_N4, 0x20, imm); }\n");
printf("void vpinsrb(const Xmm& x, const Operand& op, uint8 imm) { vpinsrb(x, x, op, imm); }\n");
- printf("void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F | T_66 | T_EVEX, 0xC4, imm); }\n");
+ printf("void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F | T_66 | T_EVEX | T_N4, 0xC4, imm); }\n");
printf("void vpinsrw(const Xmm& x, const Operand& op, uint8 imm) { vpinsrw(x, x, op, imm); }\n");
- printf("void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0, 0x22, imm); }\n");
+ printf("void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x22, imm); }\n");
printf("void vpinsrd(const Xmm& x, const Operand& op, uint8 imm) { vpinsrd(x, x, op, imm); }\n");
printf("void vpmovmskb(const Reg32e& r, const Xmm& x) { bool isYMM= x.isYMM(); opAVX_X_X_XM(isYMM ? Ymm(r.getIdx()) : Xmm(r.getIdx()), isYMM ? ym0 : xm0, x, T_0F | T_66 | T_YMM, 0xD7); }\n");
@@ -1604,9 +1606,9 @@
printf("#ifdef XBYAK64\n");
printf("void vmovq(const Xmm& x, const Reg64& r) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x6E); }\n");
printf("void vmovq(const Reg64& r, const Xmm& x) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x7E); }\n");
- printf("void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1, 0x16, imm); }\n");
+ printf("void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }\n");
- printf("void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1, 0x22, imm); }\n");
+ printf("void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }\n");
printf("void vpinsrq(const Xmm& x, const Operand& op, uint8 imm) { vpinsrq(x, x, op, imm); }\n");
printf("void vcvtss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1, 0x2D); }\n");
diff --git a/test/make_512.cpp b/test/make_512.cpp
index 4f46fb0..80ed9a2 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -1211,14 +1211,14 @@
put(p.name, p.x1, p.x2, p.xm, IMM8);
}
#ifdef XBYAK64
- put("vpextrb", _REG64, _XMM3, IMM8);
- put("vpextrw", _REG64|MEM, _XMM3, IMM8);
- put("vpextrd", _REG32, _XMM3, IMM8);
- put("vpextrq", _REG64, _XMM3, IMM8);
- put("vpinsrb", _XMM3, _XMM3, _REG32, IMM8);
- put("vpinsrw", _XMM3, _XMM3, _REG32, IMM8);
- put("vpinsrd", _XMM3, _XMM3, _REG32, IMM8);
- put("vpinsrq", _XMM3, _XMM3, _REG64, IMM8);
+ put("vpextrb", _REG64 | _MEM, _XMM3, IMM8);
+ put("vpextrw", _REG64 | _MEM, _XMM3, IMM8);
+ put("vpextrd", _REG32 | _MEM, _XMM3, IMM8);
+ put("vpextrq", _REG64 | _MEM, _XMM3, IMM8);
+ put("vpinsrb", _XMM3, _XMM3, _REG32 | _MEM, IMM8);
+ put("vpinsrw", _XMM3, _XMM3, _REG32 | _MEM, IMM8);
+ put("vpinsrd", _XMM3, _XMM3, _REG32 | _MEM, IMM8);
+ put("vpinsrq", _XMM3, _XMM3, _REG64 | _MEM, IMM8);
#endif
}
void put512_FMA()
@@ -1422,7 +1422,7 @@
#ifdef MIN_TEST
putMin();
#else
-#if 1
+#if 0
putOpmask();
separateFunc();
putCombi();
@@ -1441,12 +1441,12 @@
separateFunc();
put512_X3();
separateFunc();
-#endif
-#if 0
put512_X3_I();
separateFunc();
+#endif
put512_FMA();
separateFunc();
+#if 0
put512_Y_XM();
separateFunc();
put512_AVX1();
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index da61db4..0c520c1 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1345,12 +1345,12 @@
void vfmadd132ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x98); }
void vfmadd213ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xA8); }
void vfmadd231ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xB8); }
-void vfmadd132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0x99); }
-void vfmadd213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xA9); }
-void vfmadd231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xB9); }
-void vfmadd132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0x99); }
-void vfmadd213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xA9); }
-void vfmadd231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xB9); }
+void vfmadd132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0x99); }
+void vfmadd213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xA9); }
+void vfmadd231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xB9); }
+void vfmadd132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0x99); }
+void vfmadd213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xA9); }
+void vfmadd231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xB9); }
void vfmaddsub132pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0x96); }
void vfmaddsub213pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0xA6); }
void vfmaddsub231pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0xB6); }
@@ -1369,36 +1369,36 @@
void vfmsub132ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x9A); }
void vfmsub213ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xAA); }
void vfmsub231ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xBA); }
-void vfmsub132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0x9B); }
-void vfmsub213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xAB); }
-void vfmsub231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xBB); }
-void vfmsub132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0x9B); }
-void vfmsub213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xAB); }
-void vfmsub231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xBB); }
+void vfmsub132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0x9B); }
+void vfmsub213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xAB); }
+void vfmsub231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xBB); }
+void vfmsub132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0x9B); }
+void vfmsub213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xAB); }
+void vfmsub231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xBB); }
void vfnmadd132pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0x9C); }
void vfnmadd213pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0xAC); }
void vfnmadd231pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0xBC); }
void vfnmadd132ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x9C); }
void vfnmadd213ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xAC); }
void vfnmadd231ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xBC); }
-void vfnmadd132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0x9D); }
-void vfnmadd213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xAD); }
-void vfnmadd231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xBD); }
-void vfnmadd132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0x9D); }
-void vfnmadd213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xAD); }
-void vfnmadd231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xBD); }
+void vfnmadd132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0x9D); }
+void vfnmadd213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xAD); }
+void vfnmadd231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xBD); }
+void vfnmadd132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0x9D); }
+void vfnmadd213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xAD); }
+void vfnmadd231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xBD); }
void vfnmsub132pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0x9E); }
void vfnmsub213pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0xAE); }
void vfnmsub231pd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0xBE); }
void vfnmsub132ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x9E); }
void vfnmsub213ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xAE); }
void vfnmsub231ps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0xBE); }
-void vfnmsub132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0x9F); }
-void vfnmsub213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xAF); }
-void vfnmsub231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X, 0xBF); }
-void vfnmsub132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0x9F); }
-void vfnmsub213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xAF); }
-void vfnmsub231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X, 0xBF); }
+void vfnmsub132sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0x9F); }
+void vfnmsub213sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xAF); }
+void vfnmsub231sd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W1 | T_EW1 | T_EVEX | T_ER_X | T_N8, 0xBF); }
+void vfnmsub132ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0x9F); }
+void vfnmsub213ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xAF); }
+void vfnmsub231ss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX | T_ER_X | T_N4, 0xBF); }
void vaesimc(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0, 0xDB); }
void vbroadcastf128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x1A); }
void vbroadcasti128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x5A); }
@@ -1419,14 +1419,14 @@
void vldmxcsr(const Address& addr) { opAVX_X_X_XM(xm2, xm0, addr, T_0F, 0xAE); }
void vstmxcsr(const Address& addr) { opAVX_X_X_XM(xm3, xm0, addr, T_0F, 0xAE); }
void vmaskmovdqu(const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x1, xm0, x2, T_0F | T_66, 0xF7); }
-void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(8|16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX, 0x14, imm); }
-void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX, 0x15, imm); } }
-void vpextrd(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0, 0x16, imm); }
-void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX, 0x20, imm); }
+void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(8|16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX | T_N4, 0x14, imm); }
+void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(16|i32e) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX | T_N4, 0x15, imm); } }
+void vpextrd(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x16, imm); }
+void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_EVEX | T_N4, 0x20, imm); }
void vpinsrb(const Xmm& x, const Operand& op, uint8 imm) { vpinsrb(x, x, op, imm); }
-void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F | T_66 | T_EVEX, 0xC4, imm); }
+void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F | T_66 | T_EVEX | T_N4, 0xC4, imm); }
void vpinsrw(const Xmm& x, const Operand& op, uint8 imm) { vpinsrw(x, x, op, imm); }
-void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0, 0x22, imm); }
+void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x22, imm); }
void vpinsrd(const Xmm& x, const Operand& op, uint8 imm) { vpinsrd(x, x, op, imm); }
void vpmovmskb(const Reg32e& r, const Xmm& x) { bool isYMM= x.isYMM(); opAVX_X_X_XM(isYMM ? Ymm(r.getIdx()) : Xmm(r.getIdx()), isYMM ? ym0 : xm0, x, T_0F | T_66 | T_YMM, 0xD7); }
void vpslldq(const Xmm& x, const Operand& op, uint8 imm) { opAVX_X_X_XM(x.copyAndSetIdx(7), x, op, T_66 | T_0F | T_YMM | T_EVEX, 0x73, imm); }
@@ -1490,8 +1490,8 @@
#ifdef XBYAK64
void vmovq(const Xmm& x, const Reg64& r) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x6E); }
void vmovq(const Reg64& r, const Xmm& x) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x7E); }
-void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1, 0x16, imm); }
-void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1, 0x22, imm); }
+void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, false, xm0, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }
+void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!op.isREG(64) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x1, false, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }
void vpinsrq(const Xmm& x, const Operand& op, uint8 imm) { vpinsrq(x, x, op, imm); }
void vcvtss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1, 0x2D); }
void vcvttss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1, 0x2C); }
