rename T_SUPPORT_YMM to T_YMM
diff --git a/gen/avx_type.hpp b/gen/avx_type.hpp
index 58190ee..2134a32 100644
--- a/gen/avx_type.hpp
+++ b/gen/avx_type.hpp
@@ -13,7 +13,7 @@
T_W1 = 1 << 12,
T_EW0 = 1 << 13,
T_EW1 = 1 << 14,
- T_SUPPORT_YMM = 1 << 15
+ T_YMM = 1 << 15
};
const int NONE = 256; // same as Xbyak::CodeGenerator::NONE
@@ -69,9 +69,9 @@
if (!str.empty()) str += " | ";
str += "T_EW1";
}
- if (type & T_SUPPORT_YMM) {
+ if (type & T_YMM) {
if (!str.empty()) str += " | ";
- str += "T_SUPPORT_YMM";
+ str += "T_YMM";
}
return str;
}
\ No newline at end of file
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 35a01c5..4330520 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1003,8 +1003,8 @@
};
for (int i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
- printf("void v%spd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x%02X); }\n", p->name, p->code);
- printf("void v%sps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x%02X); }\n", p->name, p->code);
+ printf("void v%spd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x%02X); }\n", p->name, p->code);
+ printf("void v%sps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x%02X); }\n", p->name, p->code);
if (p->only_pd_ps) continue;
printf("void v%ssd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x%02X); }\n", p->name, p->code);
printf("void v%sss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x%02X); }\n", p->name, p->code);
@@ -1019,157 +1019,157 @@
bool hasIMM;
bool enableOmit;
} tbl[] = {
- { 0x0D, "blendpd", T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, true, true },
- { 0x0C, "blendps", T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, true, true },
+ { 0x0D, "blendpd", T_0F3A | T_66 | T_W0 | T_YMM, true, true },
+ { 0x0C, "blendps", T_0F3A | T_66 | T_W0 | T_YMM, true, true },
{ 0x41, "dppd", T_0F3A | T_66 | T_W0, true, true },
- { 0x40, "dpps", T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, true, true },
- { 0x42, "mpsadbw", T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, true, true },
- { 0x0E, "pblendw", T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, true, true },
- { 0x02, "pblendd", T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, true, true },
+ { 0x40, "dpps", T_0F3A | T_66 | T_W0 | T_YMM, true, true },
+ { 0x42, "mpsadbw", T_0F3A | T_66 | T_W0 | T_YMM, true, true },
+ { 0x0E, "pblendw", T_0F3A | T_66 | T_W0 | T_YMM, true, true },
+ { 0x02, "pblendd", T_0F3A | T_66 | T_W0 | T_YMM, true, true },
{ 0x0B, "roundsd", T_0F3A | T_66 | T_W0, true, true },
{ 0x0A, "roundss", T_0F3A | T_66 | T_W0, true, true },
{ 0x44, "pclmulqdq", T_0F3A | T_66 | T_W0, true, true },
- { 0x0C, "permilps", T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, false, false },
- { 0x0D, "permilpd", T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, false, false },
+ { 0x0C, "permilps", T_0F38 | T_66 | T_W0 | T_YMM, false, false },
+ { 0x0D, "permilpd", T_0F38 | T_66 | T_W0 | T_YMM, false, false },
- { 0x47, "psllvd", T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, false, false },
- { 0x47, "psllvq", T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, false, false },
- { 0x46, "psravd", T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, false, false },
- { 0x45, "psrlvd", T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, false, false },
- { 0x45, "psrlvq", T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, false, false },
+ { 0x47, "psllvd", T_0F38 | T_66 | T_W0 | T_YMM, false, false },
+ { 0x47, "psllvq", T_0F38 | T_66 | T_W1 | T_YMM, false, false },
+ { 0x46, "psravd", T_0F38 | T_66 | T_W0 | T_YMM, false, false },
+ { 0x45, "psrlvd", T_0F38 | T_66 | T_W0 | T_YMM, false, false },
+ { 0x45, "psrlvq", T_0F38 | T_66 | T_W1 | T_YMM, false, false },
- { 0xC2, "cmppd", T_0F | T_66 | T_SUPPORT_YMM, true, true },
- { 0xC2, "cmpps", T_0F | T_SUPPORT_YMM, true, true },
+ { 0xC2, "cmppd", T_0F | T_66 | T_YMM, true, true },
+ { 0xC2, "cmpps", T_0F | T_YMM, true, true },
{ 0xC2, "cmpsd", T_0F | T_F2, true, true },
{ 0xC2, "cmpss", T_0F | T_F3, true, true },
{ 0x5A, "cvtsd2ss", T_0F | T_F2, false, true },
{ 0x5A, "cvtss2sd", T_0F | T_F3, false, true },
{ 0x21, "insertps", T_0F3A | T_66 | T_W0, true, true },
- { 0x63, "packsswb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x6B, "packssdw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x67, "packuswb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x2B, "packusdw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x63, "packsswb", T_0F | T_66 | T_YMM, false, true },
+ { 0x6B, "packssdw", T_0F | T_66 | T_YMM, false, true },
+ { 0x67, "packuswb", T_0F | T_66 | T_YMM, false, true },
+ { 0x2B, "packusdw", T_0F38 | T_66 | T_YMM, false, true },
- { 0xFC, "paddb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xFD, "paddw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xFE, "paddd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xD4, "paddq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xFC, "paddb", T_0F | T_66 | T_YMM, false, true },
+ { 0xFD, "paddw", T_0F | T_66 | T_YMM, false, true },
+ { 0xFE, "paddd", T_0F | T_66 | T_YMM, false, true },
+ { 0xD4, "paddq", T_0F | T_66 | T_YMM, false, true },
- { 0xEC, "paddsb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xED, "paddsw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xEC, "paddsb", T_0F | T_66 | T_YMM, false, true },
+ { 0xED, "paddsw", T_0F | T_66 | T_YMM, false, true },
- { 0xDC, "paddusb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xDD, "paddusw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xDC, "paddusb", T_0F | T_66 | T_YMM, false, true },
+ { 0xDD, "paddusw", T_0F | T_66 | T_YMM, false, true },
- { 0x0F, "palignr", T_0F3A | T_66 | T_SUPPORT_YMM, true, true },
+ { 0x0F, "palignr", T_0F3A | T_66 | T_YMM, true, true },
- { 0xDB, "pand", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xDF, "pandn", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xDB, "pand", T_0F | T_66 | T_YMM, false, true },
+ { 0xDF, "pandn", T_0F | T_66 | T_YMM, false, true },
- { 0xE0, "pavgb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xE3, "pavgw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xE0, "pavgb", T_0F | T_66 | T_YMM, false, true },
+ { 0xE3, "pavgw", T_0F | T_66 | T_YMM, false, true },
- { 0x74, "pcmpeqb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x75, "pcmpeqw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x76, "pcmpeqd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x29, "pcmpeqq", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x74, "pcmpeqb", T_0F | T_66 | T_YMM, false, true },
+ { 0x75, "pcmpeqw", T_0F | T_66 | T_YMM, false, true },
+ { 0x76, "pcmpeqd", T_0F | T_66 | T_YMM, false, true },
+ { 0x29, "pcmpeqq", T_0F38 | T_66 | T_YMM, false, true },
- { 0x64, "pcmpgtb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x65, "pcmpgtw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x66, "pcmpgtd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x37, "pcmpgtq", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x64, "pcmpgtb", T_0F | T_66 | T_YMM, false, true },
+ { 0x65, "pcmpgtw", T_0F | T_66 | T_YMM, false, true },
+ { 0x66, "pcmpgtd", T_0F | T_66 | T_YMM, false, true },
+ { 0x37, "pcmpgtq", T_0F38 | T_66 | T_YMM, false, true },
- { 0x01, "phaddw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x02, "phaddd", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x03, "phaddsw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x01, "phaddw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x02, "phaddd", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x03, "phaddsw", T_0F38 | T_66 | T_YMM, false, true },
- { 0x05, "phsubw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x06, "phsubd", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x07, "phsubsw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0xF5, "pmaddwd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x04, "pmaddubsw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x05, "phsubw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x06, "phsubd", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x07, "phsubsw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0xF5, "pmaddwd", T_0F | T_66 | T_YMM, false, true },
+ { 0x04, "pmaddubsw", T_0F38 | T_66 | T_YMM, false, true },
- { 0x3C, "pmaxsb", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0xEE, "pmaxsw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x3D, "pmaxsd", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x3C, "pmaxsb", T_0F38 | T_66 | T_YMM, false, true },
+ { 0xEE, "pmaxsw", T_0F | T_66 | T_YMM, false, true },
+ { 0x3D, "pmaxsd", T_0F38 | T_66 | T_YMM, false, true },
- { 0xDE, "pmaxub", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x3E, "pmaxuw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x3F, "pmaxud", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xDE, "pmaxub", T_0F | T_66 | T_YMM, false, true },
+ { 0x3E, "pmaxuw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x3F, "pmaxud", T_0F38 | T_66 | T_YMM, false, true },
- { 0x38, "pminsb", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0xEA, "pminsw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x39, "pminsd", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x38, "pminsb", T_0F38 | T_66 | T_YMM, false, true },
+ { 0xEA, "pminsw", T_0F | T_66 | T_YMM, false, true },
+ { 0x39, "pminsd", T_0F38 | T_66 | T_YMM, false, true },
- { 0xDA, "pminub", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x3A, "pminuw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x3B, "pminud", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xDA, "pminub", T_0F | T_66 | T_YMM, false, true },
+ { 0x3A, "pminuw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x3B, "pminud", T_0F38 | T_66 | T_YMM, false, true },
- { 0xE4, "pmulhuw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x0B, "pmulhrsw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0xE5, "pmulhw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xD5, "pmullw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x40, "pmulld", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xE4, "pmulhuw", T_0F | T_66 | T_YMM, false, true },
+ { 0x0B, "pmulhrsw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0xE5, "pmulhw", T_0F | T_66 | T_YMM, false, true },
+ { 0xD5, "pmullw", T_0F | T_66 | T_YMM, false, true },
+ { 0x40, "pmulld", T_0F38 | T_66 | T_YMM, false, true },
{ 0xF4, "pmuludq", T_0F | T_66, false, true },
- { 0x28, "pmuldq", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x28, "pmuldq", T_0F38 | T_66 | T_YMM, false, true },
- { 0xEB, "por", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xF6, "psadbw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xEB, "por", T_0F | T_66 | T_YMM, false, true },
+ { 0xF6, "psadbw", T_0F | T_66 | T_YMM, false, true },
- { 0x00, "pshufb", T_0F38 | T_66 | T_SUPPORT_YMM, false, false },
+ { 0x00, "pshufb", T_0F38 | T_66 | T_YMM, false, false },
- { 0x08, "psignb", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x09, "psignw", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
- { 0x0A, "psignd", T_0F38 | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x08, "psignb", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x09, "psignw", T_0F38 | T_66 | T_YMM, false, true },
+ { 0x0A, "psignd", T_0F38 | T_66 | T_YMM, false, true },
- { 0xF1, "psllw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xF2, "pslld", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xF3, "psllq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xF1, "psllw", T_0F | T_66 | T_YMM, false, true },
+ { 0xF2, "pslld", T_0F | T_66 | T_YMM, false, true },
+ { 0xF3, "psllq", T_0F | T_66 | T_YMM, false, true },
- { 0xE1, "psraw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xE2, "psrad", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xD1, "psrlw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xD2, "psrld", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xD3, "psrlq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xE1, "psraw", T_0F | T_66 | T_YMM, false, true },
+ { 0xE2, "psrad", T_0F | T_66 | T_YMM, false, true },
+ { 0xD1, "psrlw", T_0F | T_66 | T_YMM, false, true },
+ { 0xD2, "psrld", T_0F | T_66 | T_YMM, false, true },
+ { 0xD3, "psrlq", T_0F | T_66 | T_YMM, false, true },
- { 0xF8, "psubb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xF9, "psubw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xFA, "psubd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xFB, "psubq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xF8, "psubb", T_0F | T_66 | T_YMM, false, true },
+ { 0xF9, "psubw", T_0F | T_66 | T_YMM, false, true },
+ { 0xFA, "psubd", T_0F | T_66 | T_YMM, false, true },
+ { 0xFB, "psubq", T_0F | T_66 | T_YMM, false, true },
- { 0xE8, "psubsb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xE9, "psubsw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xE8, "psubsb", T_0F | T_66 | T_YMM, false, true },
+ { 0xE9, "psubsw", T_0F | T_66 | T_YMM, false, true },
- { 0xD8, "psubusb", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0xD9, "psubusw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xD8, "psubusb", T_0F | T_66 | T_YMM, false, true },
+ { 0xD9, "psubusw", T_0F | T_66 | T_YMM, false, true },
- { 0x68, "punpckhbw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x69, "punpckhwd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x6A, "punpckhdq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x6D, "punpckhqdq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x68, "punpckhbw", T_0F | T_66 | T_YMM, false, true },
+ { 0x69, "punpckhwd", T_0F | T_66 | T_YMM, false, true },
+ { 0x6A, "punpckhdq", T_0F | T_66 | T_YMM, false, true },
+ { 0x6D, "punpckhqdq", T_0F | T_66 | T_YMM, false, true },
- { 0x60, "punpcklbw", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x61, "punpcklwd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x62, "punpckldq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x6C, "punpcklqdq", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0x60, "punpcklbw", T_0F | T_66 | T_YMM, false, true },
+ { 0x61, "punpcklwd", T_0F | T_66 | T_YMM, false, true },
+ { 0x62, "punpckldq", T_0F | T_66 | T_YMM, false, true },
+ { 0x6C, "punpcklqdq", T_0F | T_66 | T_YMM, false, true },
- { 0xEF, "pxor", T_0F | T_66 | T_SUPPORT_YMM, false, true },
+ { 0xEF, "pxor", T_0F | T_66 | T_YMM, false, true },
{ 0x53, "rcpss", T_0F | T_F3, false, true },
{ 0x52, "rsqrtss", T_0F | T_F3, false, true },
- { 0xC6, "shufpd", T_0F | T_66 | T_SUPPORT_YMM, true, true },
- { 0xC6, "shufps", T_0F | T_SUPPORT_YMM, true, true },
+ { 0xC6, "shufpd", T_0F | T_66 | T_YMM, true, true },
+ { 0xC6, "shufps", T_0F | T_YMM, true, true },
{ 0x51, "sqrtsd", T_0F | T_F2, false, true },
{ 0x51, "sqrtss", T_0F | T_F3, false, true },
- { 0x15, "unpckhpd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x15, "unpckhps", T_0F | T_SUPPORT_YMM, false, true },
+ { 0x15, "unpckhpd", T_0F | T_66 | T_YMM, false, true },
+ { 0x15, "unpckhps", T_0F | T_YMM, false, true },
- { 0x14, "unpcklpd", T_0F | T_66 | T_SUPPORT_YMM, false, true },
- { 0x14, "unpcklps", T_0F | T_SUPPORT_YMM, false, true },
+ { 0x14, "unpcklpd", T_0F | T_66 | T_YMM, false, true },
+ { 0x14, "unpcklps", T_0F | T_YMM, false, true },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -1189,60 +1189,60 @@
bool hasIMM;
} tbl[] = {
{ 0xDF, "aeskeygenassist", T_0F3A | T_66, true },
- { 0x09, "roundpd", T_0F3A | T_66 | T_SUPPORT_YMM, true },
- { 0x08, "roundps", T_0F3A | T_66 | T_SUPPORT_YMM, true },
- { 0x05, "permilpd", T_0F3A | T_66 | T_SUPPORT_YMM, true },
- { 0x04, "permilps", T_0F3A | T_66 | T_SUPPORT_YMM, true },
+ { 0x09, "roundpd", T_0F3A | T_66 | T_YMM, true },
+ { 0x08, "roundps", T_0F3A | T_66 | T_YMM, true },
+ { 0x05, "permilpd", T_0F3A | T_66 | T_YMM, true },
+ { 0x04, "permilps", T_0F3A | T_66 | T_YMM, true },
{ 0x61, "pcmpestri", T_0F3A | T_66, true },
{ 0x60, "pcmpestrm", T_0F3A | T_66, true },
{ 0x63, "pcmpistri", T_0F3A | T_66, true },
{ 0x62, "pcmpistrm", T_0F3A | T_66, true },
- { 0x0E, "testps", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x0F, "testpd", T_0F38 | T_66 | T_SUPPORT_YMM, false },
+ { 0x0E, "testps", T_0F38 | T_66 | T_YMM, false },
+ { 0x0F, "testpd", T_0F38 | T_66 | T_YMM, false },
{ 0x2F, "comisd", T_0F | T_66, false },
{ 0x2F, "comiss", T_0F, false },
- { 0x5B, "cvtdq2ps", T_0F | T_SUPPORT_YMM, false },
- { 0x5B, "cvtps2dq", T_0F | T_66 | T_SUPPORT_YMM, false },
- { 0x5B, "cvttps2dq", T_0F | T_F3 | T_SUPPORT_YMM, false },
- { 0x28, "movapd", T_0F | T_66 | T_SUPPORT_YMM, false },
- { 0x28, "movaps", T_0F | T_SUPPORT_YMM, false },
- { 0x12, "movddup", T_0F | T_F2 | T_SUPPORT_YMM, false },
- { 0x6F, "movdqa", T_0F | T_66 | T_SUPPORT_YMM, false },
- { 0x6F, "movdqu", T_0F | T_F3 | T_SUPPORT_YMM, false },
- { 0x16, "movshdup", T_0F | T_F3 | T_SUPPORT_YMM, false },
- { 0x12, "movsldup", T_0F | T_F3 | T_SUPPORT_YMM, false },
- { 0x10, "movupd", T_0F | T_66 | T_SUPPORT_YMM, false },
- { 0x10, "movups", T_0F | T_SUPPORT_YMM, false },
+ { 0x5B, "cvtdq2ps", T_0F | T_YMM, false },
+ { 0x5B, "cvtps2dq", T_0F | T_66 | T_YMM, false },
+ { 0x5B, "cvttps2dq", T_0F | T_F3 | T_YMM, false },
+ { 0x28, "movapd", T_0F | T_66 | T_YMM, false },
+ { 0x28, "movaps", T_0F | T_YMM, false },
+ { 0x12, "movddup", T_0F | T_F2 | T_YMM, false },
+ { 0x6F, "movdqa", T_0F | T_66 | T_YMM, false },
+ { 0x6F, "movdqu", T_0F | T_F3 | T_YMM, false },
+ { 0x16, "movshdup", T_0F | T_F3 | T_YMM, false },
+ { 0x12, "movsldup", T_0F | T_F3 | T_YMM, false },
+ { 0x10, "movupd", T_0F | T_66 | T_YMM, false },
+ { 0x10, "movups", T_0F | T_YMM, false },
- { 0x1C, "pabsb", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x1D, "pabsw", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x1E, "pabsd", T_0F38 | T_66 | T_SUPPORT_YMM, false },
+ { 0x1C, "pabsb", T_0F38 | T_66 | T_YMM, false },
+ { 0x1D, "pabsw", T_0F38 | T_66 | T_YMM, false },
+ { 0x1E, "pabsd", T_0F38 | T_66 | T_YMM, false },
{ 0x41, "phminposuw", T_0F38 | T_66, false },
- { 0x20, "pmovsxbw", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x21, "pmovsxbd", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x22, "pmovsxbq", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x23, "pmovsxwd", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x24, "pmovsxwq", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x25, "pmovsxdq", T_0F38 | T_66 | T_SUPPORT_YMM, false },
+ { 0x20, "pmovsxbw", T_0F38 | T_66 | T_YMM, false },
+ { 0x21, "pmovsxbd", T_0F38 | T_66 | T_YMM, false },
+ { 0x22, "pmovsxbq", T_0F38 | T_66 | T_YMM, false },
+ { 0x23, "pmovsxwd", T_0F38 | T_66 | T_YMM, false },
+ { 0x24, "pmovsxwq", T_0F38 | T_66 | T_YMM, false },
+ { 0x25, "pmovsxdq", T_0F38 | T_66 | T_YMM, false },
- { 0x30, "pmovzxbw", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x31, "pmovzxbd", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x32, "pmovzxbq", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x33, "pmovzxwd", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x34, "pmovzxwq", T_0F38 | T_66 | T_SUPPORT_YMM, false },
- { 0x35, "pmovzxdq", T_0F38 | T_66 | T_SUPPORT_YMM, false },
+ { 0x30, "pmovzxbw", T_0F38 | T_66 | T_YMM, false },
+ { 0x31, "pmovzxbd", T_0F38 | T_66 | T_YMM, false },
+ { 0x32, "pmovzxbq", T_0F38 | T_66 | T_YMM, false },
+ { 0x33, "pmovzxwd", T_0F38 | T_66 | T_YMM, false },
+ { 0x34, "pmovzxwq", T_0F38 | T_66 | T_YMM, false },
+ { 0x35, "pmovzxdq", T_0F38 | T_66 | T_YMM, false },
- { 0x70, "pshufd", T_0F | T_66 | T_SUPPORT_YMM, true },
- { 0x70, "pshufhw", T_0F | T_F3 | T_SUPPORT_YMM, true },
- { 0x70, "pshuflw", T_0F | T_F2 | T_SUPPORT_YMM, true },
+ { 0x70, "pshufd", T_0F | T_66 | T_YMM, true },
+ { 0x70, "pshufhw", T_0F | T_F3 | T_YMM, true },
+ { 0x70, "pshuflw", T_0F | T_F2 | T_YMM, true },
{ 0x17, "ptest", T_0F38 | T_66, false },
- { 0x53, "rcpps", T_0F | T_SUPPORT_YMM, false },
- { 0x52, "rsqrtps", T_0F | T_SUPPORT_YMM, false },
+ { 0x53, "rcpps", T_0F | T_YMM, false },
+ { 0x52, "rsqrtps", T_0F | T_YMM, false },
- { 0x51, "sqrtpd", T_0F | T_66 | T_SUPPORT_YMM, false },
- { 0x51, "sqrtps", T_0F | T_SUPPORT_YMM, false },
+ { 0x51, "sqrtpd", T_0F | T_66 | T_YMM, false },
+ { 0x51, "sqrtps", T_0F | T_YMM, false },
{ 0x2E, "ucomisd", T_0F | T_66, false },
{ 0x2E, "ucomiss", T_0F, false },
@@ -1261,12 +1261,12 @@
const char *name;
int type;
} tbl[] = {
- { 0x29, "movapd", T_0F | T_66 | T_SUPPORT_YMM},
- { 0x29, "movaps", T_0F | T_SUPPORT_YMM},
- { 0x7F, "movdqa", T_0F | T_66 | T_SUPPORT_YMM},
- { 0x7F, "movdqu", T_0F | T_F3 | T_SUPPORT_YMM},
- { 0x11, "movupd", T_0F | T_66 | T_SUPPORT_YMM},
- { 0x11, "movups", T_0F | T_SUPPORT_YMM},
+ { 0x29, "movapd", T_0F | T_66 | T_YMM},
+ { 0x29, "movaps", T_0F | T_YMM},
+ { 0x7F, "movdqa", T_0F | T_66 | T_YMM},
+ { 0x7F, "movdqu", T_0F | T_F3 | T_YMM},
+ { 0x11, "movupd", T_0F | T_66 | T_YMM},
+ { 0x11, "movups", T_0F | T_YMM},
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -1282,12 +1282,12 @@
const char *name;
int type;
} tbl[] = {
- { 0xD0, "addsubpd", T_0F | T_66 | T_SUPPORT_YMM},
- { 0xD0, "addsubps", T_0F | T_F2 | T_SUPPORT_YMM},
- { 0x7C, "haddpd", T_0F | T_66 | T_SUPPORT_YMM},
- { 0x7C, "haddps", T_0F | T_F2 | T_SUPPORT_YMM},
- { 0x7D, "hsubpd", T_0F | T_66 | T_SUPPORT_YMM},
- { 0x7D, "hsubps", T_0F | T_F2 | T_SUPPORT_YMM},
+ { 0xD0, "addsubpd", T_0F | T_66 | T_YMM},
+ { 0xD0, "addsubps", T_0F | T_F2 | T_YMM},
+ { 0x7C, "haddpd", T_0F | T_66 | T_YMM},
+ { 0x7C, "haddps", T_0F | T_F2 | T_YMM},
+ { 0x7D, "hsubpd", T_0F | T_66 | T_YMM},
+ { 0x7D, "hsubps", T_0F | T_F2 | T_YMM},
{ 0xDC, "aesenc", T_0F38 | T_66 | T_W0},
{ 0xDD, "aesenclast", T_0F38 | T_66 | T_W0},
@@ -1305,16 +1305,16 @@
{
const char suf[][8] = { "ps", "pd" };
for (int i = 0; i < 2; i++) {
- printf("void vmaskmov%s(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x%02X); }\n", suf[i], 0x2C + i);
- printf("void vmaskmov%s(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x%02X); }\n", suf[i], 0x2E + i);
+ printf("void vmaskmov%s(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x%02X); }\n", suf[i], 0x2C + i);
+ printf("void vmaskmov%s(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x%02X); }\n", suf[i], 0x2E + i);
}
}
// vpmaskmov
{
const char suf[][8] = { "d", "q" };
for (int i = 0; i < 2; i++) {
- printf("void vpmaskmov%s(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W%d | T_SUPPORT_YMM, 0x%02X); }\n", suf[i], i, 0x8C);
- printf("void vpmaskmov%s(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W%d | T_SUPPORT_YMM, 0x%02X); }\n", suf[i], i, 0x8E);
+ printf("void vpmaskmov%s(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W%d | T_YMM, 0x%02X); }\n", suf[i], i, 0x8C);
+ printf("void vpmaskmov%s(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W%d | T_YMM, 0x%02X); }\n", suf[i], i, 0x8E);
}
}
// vpermd, vpermps
@@ -1327,7 +1327,7 @@
{ "ps", 0x16 },
};
for (int i = 0; i < NUM_OF_ARRAY(tbl); i++) {
- printf("void vperm%s(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x%02X); }\n", tbl[i].suf, tbl[i].code);
+ printf("void vperm%s(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x%02X); }\n", tbl[i].suf, tbl[i].code);
}
}
// vpermq, vpermpd
@@ -1340,7 +1340,7 @@
{ "pd", 0x01 },
};
for (int i = 0; i < NUM_OF_ARRAY(tbl); i++) {
- printf("void vperm%s(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_SUPPORT_YMM, 0x%02X, imm); }\n", tbl[i].suf, tbl[i].code);
+ printf("void vperm%s(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_YMM, 0x%02X, imm); }\n", tbl[i].suf, tbl[i].code);
}
}
// vcmpeqps
@@ -1420,7 +1420,7 @@
{ "231", 0xB0 },
};
printf("void %s%s%s(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W%d%s, 0x%02X); }\n"
- , tbl[i].name, ord[k].str, suf[tbl[i].supportYMM ? 0 : 1][j], j == 0 ? 1 : 0, tbl[i].supportYMM ? " | T_SUPPORT_YMM" : "", tbl[i].code + ord[k].code);
+ , tbl[i].name, ord[k].str, suf[tbl[i].supportYMM ? 0 : 1][j], j == 0 ? 1 : 0, tbl[i].supportYMM ? " | T_YMM" : "", tbl[i].code + ord[k].code);
}
}
}
@@ -1429,9 +1429,9 @@
{
printf("void vaesimc(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0, 0xDB); }\n");
- printf("void vbroadcastf128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x1A); }\n");
- printf("void vbroadcasti128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x5A); }\n");
- printf("void vbroadcastsd(const Ymm& y, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(y, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x19); }\n");
+ printf("void vbroadcastf128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x1A); }\n");
+ printf("void vbroadcasti128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x5A); }\n");
+ printf("void vbroadcastsd(const Ymm& y, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(y, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x19); }\n");
const struct Tbl {
const char *name;
uint8 code;
@@ -1444,18 +1444,18 @@
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];
- printf("void %s(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x%02X); }\n", p.name, p.code);
+ printf("void %s(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x%02X); }\n", p.name, p.code);
}
- printf("void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x19, imm); }\n");
- printf("void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x39, imm); }\n");
+ printf("void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }\n");
+ printf("void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x39, imm); }\n");
printf("void vextractps(const Operand& op, const Xmm& x, uint8 imm) { if (!(op.isREG(32) || op.isMEM()) || x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, op.isREG(), Operand::XMM, T_0F3A | T_66 | T_W0, 0x17, imm); }\n");
- printf("void vinsertf128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x18, imm); }\n");
- printf("void vinserti128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x38, imm); }\n");
- printf("void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x06, imm); }\n");
- printf("void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x46, imm); }\n");
+ printf("void vinsertf128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x18, imm); }\n");
+ printf("void vinserti128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x38, imm); }\n");
+ printf("void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x06, imm); }\n");
+ printf("void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x46, imm); }\n");
- printf("void vlddqu(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_F2 | T_W0 | T_SUPPORT_YMM, 0xF0); }\n");
+ printf("void vlddqu(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_F2 | T_W0 | T_YMM, 0xF0); }\n");
printf("void vldmxcsr(const Address& addr) { opAVX_X_X_XM(xm2, xm0, addr, T_0F, 0xAE); }\n");
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");
@@ -1474,7 +1474,7 @@
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, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0, 0x22, imm); }\n");
printf("void vpinsrd(const Xmm& x, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0, 0x22, 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_SUPPORT_YMM, 0xD7); }\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");
}
// (x, x, imm), (x, imm)
@@ -1497,8 +1497,8 @@
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];
- printf("void v%s(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym%d : xm%d, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x%02X, imm); }\n", p.name, p.idx, p.idx, p.code);
- printf("void v%s(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym%d : xm%d, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x%02X, imm); }\n", p.name, p.idx, p.idx, p.code);
+ printf("void v%s(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym%d : xm%d, x1, x2, T_0F | T_66 | T_YMM, 0x%02X, imm); }\n", p.name, p.idx, p.idx, p.code);
+ printf("void v%s(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym%d : xm%d, x, x, T_0F | T_66 | T_YMM, 0x%02X, imm); }\n", p.name, p.idx, p.idx, p.code);
}
}
// 4-op
@@ -1513,7 +1513,7 @@
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];
- printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x%02X, x4.getIdx() << 4); }\n", p.name, p.code);
+ printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_YMM, 0x%02X, x4.getIdx() << 4); }\n", p.name, p.code);
printf("void %s(const Xmm& x1, const Operand& op, const Xmm& x4) { %s(x1, x1, op, x4); }\n", p.name, p.name);
}
}
@@ -1531,13 +1531,13 @@
printf("void vmovhlps(const Xmm& x1, const Xmm& x2, const Operand& op = Operand()) { if (!op.isNone() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x1, x2, op, T_0F, 0x12); }\n");
printf("void vmovlhps(const Xmm& x1, const Xmm& x2, const Operand& op = Operand()) { if (!op.isNone() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x1, x2, op, T_0F, 0x16); }\n");
- printf("void vmovmskpd(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_66 | T_W0 | T_SUPPORT_YMM, 0x50); }\n");
- printf("void vmovmskps(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_W0 | T_SUPPORT_YMM, 0x50); }\n");
+ printf("void vmovmskpd(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_66 | T_W0 | T_YMM, 0x50); }\n");
+ printf("void vmovmskps(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_W0 | T_YMM, 0x50); }\n");
- printf("void vmovntdq(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_SUPPORT_YMM, 0xE7); }\n");
- printf("void vmovntpd(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_SUPPORT_YMM, 0x2B); }\n");
- printf("void vmovntps(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_SUPPORT_YMM, 0x2B); }\n");
- printf("void vmovntdqa(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ymm0, addr, T_0F38 | T_66 | T_SUPPORT_YMM, 0x2A); }\n");
+ printf("void vmovntdq(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_YMM, 0xE7); }\n");
+ printf("void vmovntpd(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_YMM, 0x2B); }\n");
+ printf("void vmovntps(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_YMM, 0x2B); }\n");
+ printf("void vmovntdqa(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ymm0, addr, T_0F38 | T_66 | T_YMM, 0x2A); }\n");
// vmovsd, vmovss
for (int i = 0; i < 2; i++) {
@@ -1558,12 +1558,12 @@
printf("void vcvtsi2ss(const Xmm& x, const Operand& op1, const Operand& op2 = Operand()) { if (!op2.isNone() && !(op2.isREG(i32e) || op2.isMEM())) throw Error(ERR_BAD_COMBINATION); int type = T_0F | T_F3; if (!op1.isMEM() && !op2.isMEM()) type |= (op1.isREG(32) || op2.isREG(32)) ? T_W0 : T_W1; opAVX_X_X_XMcvt(x, op1, op2, op2.isREG(), Operand::XMM, type, 0x2A); }\n");
printf("void vcvtsi2sd(const Xmm& x, const Operand& op1, const Operand& op2 = Operand()) { if (!op2.isNone() && !(op2.isREG(i32e) || op2.isMEM())) throw Error(ERR_BAD_COMBINATION); int type = T_0F | T_F2; if (!op1.isMEM() && !op2.isMEM()) type |= (op1.isREG(32) || op2.isREG(32)) ? T_W0 : T_W1; opAVX_X_X_XMcvt(x, op1, op2, op2.isREG(), Operand::XMM, type, 0x2A); }\n");
- printf("void vcvtps2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_SUPPORT_YMM, 0x5A); }\n");
- printf("void vcvtdq2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_F3 | T_SUPPORT_YMM, 0xE6); }\n");
+ printf("void vcvtps2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_YMM, 0x5A); }\n");
+ printf("void vcvtdq2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_F3 | T_YMM, 0xE6); }\n");
- printf("void vcvtpd2ps(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_SUPPORT_YMM, 0x5A); }\n");
- printf("void vcvtpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_F2 | T_SUPPORT_YMM, 0xE6); }\n");
- printf("void vcvttpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE6); }\n");
+ printf("void vcvtpd2ps(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_YMM, 0x5A); }\n");
+ printf("void vcvtpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_F2 | T_YMM, 0xE6); }\n");
+ printf("void vcvttpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_YMM, 0xE6); }\n");
printf("void vcvtph2ps(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F38 | T_66 | T_W0, 0x13); }\n");
printf("void vcvtps2ph(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W0, 0x1d, imm); }\n");
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index b8aac22..9e1c82e 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1360,7 +1360,7 @@
T_W1 = 1 << 12,
T_EW0 = 1 << 13,
T_EW1 = 1 << 14,
- T_SUPPORT_YMM = 1 << 15
+ T_YMM = 1 << 15
};
void vex(const Reg& reg, const Reg& base, const Operand *v, int type, int code, bool x = false)
{
@@ -1760,12 +1760,12 @@
x2 = &x1;
op = &op1;
} else {
- if (!(op1.isXMM() || ((type & T_SUPPORT_YMM) && op1.is(Operand::YMM | Operand::ZMM)))) throw Error(ERR_BAD_COMBINATION);
+ if (!(op1.isXMM() || ((type & T_YMM) && op1.is(Operand::YMM | Operand::ZMM)))) throw Error(ERR_BAD_COMBINATION);
x2 = static_cast<const Xmm*>(&op1);
op = &op2;
}
// (x1, x2, op)
- if (!((x1.isXMM() && x2->isXMM()) || ((type & T_SUPPORT_YMM) && ((x1.isYMM() && x2->isYMM()) || (x1.isZMM() && x2->isZMM()))))) throw Error(ERR_BAD_COMBINATION);
+ if (!((x1.isXMM() && x2->isXMM()) || ((type & T_YMM) && ((x1.isYMM() && x2->isYMM()) || (x1.isZMM() && x2->isZMM()))))) throw Error(ERR_BAD_COMBINATION);
opVex(x1, x2, *op, type, code0, imm8);
}
// if cvt then return pointer to Xmm(idx) (or Ymm(idx)), otherwise return op
@@ -1807,7 +1807,7 @@
if (!isOK) throw Error(ERR_BAD_VSIB_ADDRESSING);
}
addr.permitVsib();
- opAVX_X_X_XM(isAddrYMM ? Ymm(x1.getIdx()) : x1, isAddrYMM ? Ymm(x2.getIdx()) : x2, addr, type | T_SUPPORT_YMM, code);
+ opAVX_X_X_XM(isAddrYMM ? Ymm(x1.getIdx()) : x1, isAddrYMM ? Ymm(x2.getIdx()) : x2, addr, type | T_YMM, code);
}
public:
unsigned int getVersion() const { return VERSION; }
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 40f0b2e..ffc043a 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -704,51 +704,51 @@
void fucom(const Fpu& reg) { opFpu(reg, 0xDD, 0xE0); }
void fucomp(const Fpu& reg) { opFpu(reg, 0xDD, 0xE8); }
void fxch(const Fpu& reg) { opFpu(reg, 0xD9, 0xC8); }
-void vaddpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x58); }
-void vaddps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x58); }
+void vaddpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x58); }
+void vaddps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x58); }
void vaddsd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x58); }
void vaddss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x58); }
-void vsubpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x5C); }
-void vsubps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x5C); }
+void vsubpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x5C); }
+void vsubps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x5C); }
void vsubsd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x5C); }
void vsubss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x5C); }
-void vmulpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x59); }
-void vmulps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x59); }
+void vmulpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x59); }
+void vmulps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x59); }
void vmulsd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x59); }
void vmulss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x59); }
-void vdivpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x5E); }
-void vdivps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x5E); }
+void vdivpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x5E); }
+void vdivps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x5E); }
void vdivsd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x5E); }
void vdivss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x5E); }
-void vmaxpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x5F); }
-void vmaxps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x5F); }
+void vmaxpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x5F); }
+void vmaxps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x5F); }
void vmaxsd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x5F); }
void vmaxss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x5F); }
-void vminpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x5D); }
-void vminps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x5D); }
+void vminpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x5D); }
+void vminps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x5D); }
void vminsd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_EW1, 0x5D); }
void vminss(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F3 | T_EW0, 0x5D); }
-void vandpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x54); }
-void vandps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x54); }
-void vandnpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x55); }
-void vandnps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x55); }
-void vorpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x56); }
-void vorps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x56); }
-void vxorpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_SUPPORT_YMM, 0x57); }
-void vxorps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_SUPPORT_YMM, 0x57); }
-void vblendpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x0D, imm); }
+void vandpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x54); }
+void vandps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x54); }
+void vandnpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x55); }
+void vandnps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x55); }
+void vorpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x56); }
+void vorps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x56); }
+void vxorpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM, 0x57); }
+void vxorps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM, 0x57); }
+void vblendpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x0D, imm); }
void vblendpd(const Xmm& x, const Operand& op, uint8 imm) { vblendpd(x, x, op, imm); }
-void vblendps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x0C, imm); }
+void vblendps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x0C, imm); }
void vblendps(const Xmm& x, const Operand& op, uint8 imm) { vblendps(x, x, op, imm); }
void vdppd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0, 0x41, imm); }
void vdppd(const Xmm& x, const Operand& op, uint8 imm) { vdppd(x, x, op, imm); }
-void vdpps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x40, imm); }
+void vdpps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x40, imm); }
void vdpps(const Xmm& x, const Operand& op, uint8 imm) { vdpps(x, x, op, imm); }
-void vmpsadbw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x42, imm); }
+void vmpsadbw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x42, imm); }
void vmpsadbw(const Xmm& x, const Operand& op, uint8 imm) { vmpsadbw(x, x, op, imm); }
-void vpblendw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x0E, imm); }
+void vpblendw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x0E, imm); }
void vpblendw(const Xmm& x, const Operand& op, uint8 imm) { vpblendw(x, x, op, imm); }
-void vpblendd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x02, imm); }
+void vpblendd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x02, imm); }
void vpblendd(const Xmm& x, const Operand& op, uint8 imm) { vpblendd(x, x, op, imm); }
void vroundsd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0, 0x0B, imm); }
void vroundsd(const Xmm& x, const Operand& op, uint8 imm) { vroundsd(x, x, op, imm); }
@@ -756,16 +756,16 @@
void vroundss(const Xmm& x, const Operand& op, uint8 imm) { vroundss(x, x, op, imm); }
void vpclmulqdq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0, 0x44, imm); }
void vpclmulqdq(const Xmm& x, const Operand& op, uint8 imm) { vpclmulqdq(x, x, op, imm); }
-void vpermilps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x0C); }
-void vpermilpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x0D); }
-void vpsllvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x47); }
-void vpsllvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x47); }
-void vpsravd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x46); }
-void vpsrlvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x45); }
-void vpsrlvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x45); }
-void vcmppd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xC2, imm); }
+void vpermilps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x0C); }
+void vpermilpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x0D); }
+void vpsllvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x47); }
+void vpsllvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W1 | T_YMM, 0x47); }
+void vpsravd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x46); }
+void vpsrlvd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x45); }
+void vpsrlvq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_W1 | T_YMM, 0x45); }
+void vcmppd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xC2, imm); }
void vcmppd(const Xmm& x, const Operand& op, uint8 imm) { vcmppd(x, x, op, imm); }
-void vcmpps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_SUPPORT_YMM, 0xC2, imm); }
+void vcmpps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_YMM, 0xC2, imm); }
void vcmpps(const Xmm& x, const Operand& op, uint8 imm) { vcmpps(x, x, op, imm); }
void vcmpsd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_F2, 0xC2, imm); }
void vcmpsd(const Xmm& x, const Operand& op, uint8 imm) { vcmpsd(x, x, op, imm); }
@@ -777,270 +777,270 @@
void vcvtss2sd(const Xmm& x, const Operand& op) { vcvtss2sd(x, x, op); }
void vinsertps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_W0, 0x21, imm); }
void vinsertps(const Xmm& x, const Operand& op, uint8 imm) { vinsertps(x, x, op, imm); }
-void vpacksswb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x63); }
+void vpacksswb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x63); }
void vpacksswb(const Xmm& x, const Operand& op) { vpacksswb(x, x, op); }
-void vpackssdw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x6B); }
+void vpackssdw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x6B); }
void vpackssdw(const Xmm& x, const Operand& op) { vpackssdw(x, x, op); }
-void vpackuswb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x67); }
+void vpackuswb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x67); }
void vpackuswb(const Xmm& x, const Operand& op) { vpackuswb(x, x, op); }
-void vpackusdw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x2B); }
+void vpackusdw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x2B); }
void vpackusdw(const Xmm& x, const Operand& op) { vpackusdw(x, x, op); }
-void vpaddb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xFC); }
+void vpaddb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xFC); }
void vpaddb(const Xmm& x, const Operand& op) { vpaddb(x, x, op); }
-void vpaddw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xFD); }
+void vpaddw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xFD); }
void vpaddw(const Xmm& x, const Operand& op) { vpaddw(x, x, op); }
-void vpaddd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xFE); }
+void vpaddd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xFE); }
void vpaddd(const Xmm& x, const Operand& op) { vpaddd(x, x, op); }
-void vpaddq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD4); }
+void vpaddq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD4); }
void vpaddq(const Xmm& x, const Operand& op) { vpaddq(x, x, op); }
-void vpaddsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xEC); }
+void vpaddsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xEC); }
void vpaddsb(const Xmm& x, const Operand& op) { vpaddsb(x, x, op); }
-void vpaddsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xED); }
+void vpaddsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xED); }
void vpaddsw(const Xmm& x, const Operand& op) { vpaddsw(x, x, op); }
-void vpaddusb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xDC); }
+void vpaddusb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xDC); }
void vpaddusb(const Xmm& x, const Operand& op) { vpaddusb(x, x, op); }
-void vpaddusw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xDD); }
+void vpaddusw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xDD); }
void vpaddusw(const Xmm& x, const Operand& op) { vpaddusw(x, x, op); }
-void vpalignr(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x0F, imm); }
+void vpalignr(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_YMM, 0x0F, imm); }
void vpalignr(const Xmm& x, const Operand& op, uint8 imm) { vpalignr(x, x, op, imm); }
-void vpand(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xDB); }
+void vpand(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xDB); }
void vpand(const Xmm& x, const Operand& op) { vpand(x, x, op); }
-void vpandn(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xDF); }
+void vpandn(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xDF); }
void vpandn(const Xmm& x, const Operand& op) { vpandn(x, x, op); }
-void vpavgb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE0); }
+void vpavgb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE0); }
void vpavgb(const Xmm& x, const Operand& op) { vpavgb(x, x, op); }
-void vpavgw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE3); }
+void vpavgw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE3); }
void vpavgw(const Xmm& x, const Operand& op) { vpavgw(x, x, op); }
-void vpcmpeqb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x74); }
+void vpcmpeqb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x74); }
void vpcmpeqb(const Xmm& x, const Operand& op) { vpcmpeqb(x, x, op); }
-void vpcmpeqw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x75); }
+void vpcmpeqw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x75); }
void vpcmpeqw(const Xmm& x, const Operand& op) { vpcmpeqw(x, x, op); }
-void vpcmpeqd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x76); }
+void vpcmpeqd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x76); }
void vpcmpeqd(const Xmm& x, const Operand& op) { vpcmpeqd(x, x, op); }
-void vpcmpeqq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x29); }
+void vpcmpeqq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x29); }
void vpcmpeqq(const Xmm& x, const Operand& op) { vpcmpeqq(x, x, op); }
-void vpcmpgtb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x64); }
+void vpcmpgtb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x64); }
void vpcmpgtb(const Xmm& x, const Operand& op) { vpcmpgtb(x, x, op); }
-void vpcmpgtw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x65); }
+void vpcmpgtw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x65); }
void vpcmpgtw(const Xmm& x, const Operand& op) { vpcmpgtw(x, x, op); }
-void vpcmpgtd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x66); }
+void vpcmpgtd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x66); }
void vpcmpgtd(const Xmm& x, const Operand& op) { vpcmpgtd(x, x, op); }
-void vpcmpgtq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x37); }
+void vpcmpgtq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x37); }
void vpcmpgtq(const Xmm& x, const Operand& op) { vpcmpgtq(x, x, op); }
-void vphaddw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x01); }
+void vphaddw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x01); }
void vphaddw(const Xmm& x, const Operand& op) { vphaddw(x, x, op); }
-void vphaddd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x02); }
+void vphaddd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x02); }
void vphaddd(const Xmm& x, const Operand& op) { vphaddd(x, x, op); }
-void vphaddsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x03); }
+void vphaddsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x03); }
void vphaddsw(const Xmm& x, const Operand& op) { vphaddsw(x, x, op); }
-void vphsubw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x05); }
+void vphsubw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x05); }
void vphsubw(const Xmm& x, const Operand& op) { vphsubw(x, x, op); }
-void vphsubd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x06); }
+void vphsubd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x06); }
void vphsubd(const Xmm& x, const Operand& op) { vphsubd(x, x, op); }
-void vphsubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x07); }
+void vphsubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x07); }
void vphsubsw(const Xmm& x, const Operand& op) { vphsubsw(x, x, op); }
-void vpmaddwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF5); }
+void vpmaddwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF5); }
void vpmaddwd(const Xmm& x, const Operand& op) { vpmaddwd(x, x, op); }
-void vpmaddubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x04); }
+void vpmaddubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x04); }
void vpmaddubsw(const Xmm& x, const Operand& op) { vpmaddubsw(x, x, op); }
-void vpmaxsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x3C); }
+void vpmaxsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x3C); }
void vpmaxsb(const Xmm& x, const Operand& op) { vpmaxsb(x, x, op); }
-void vpmaxsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xEE); }
+void vpmaxsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xEE); }
void vpmaxsw(const Xmm& x, const Operand& op) { vpmaxsw(x, x, op); }
-void vpmaxsd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x3D); }
+void vpmaxsd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x3D); }
void vpmaxsd(const Xmm& x, const Operand& op) { vpmaxsd(x, x, op); }
-void vpmaxub(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xDE); }
+void vpmaxub(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xDE); }
void vpmaxub(const Xmm& x, const Operand& op) { vpmaxub(x, x, op); }
-void vpmaxuw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x3E); }
+void vpmaxuw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x3E); }
void vpmaxuw(const Xmm& x, const Operand& op) { vpmaxuw(x, x, op); }
-void vpmaxud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x3F); }
+void vpmaxud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x3F); }
void vpmaxud(const Xmm& x, const Operand& op) { vpmaxud(x, x, op); }
-void vpminsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x38); }
+void vpminsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x38); }
void vpminsb(const Xmm& x, const Operand& op) { vpminsb(x, x, op); }
-void vpminsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xEA); }
+void vpminsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xEA); }
void vpminsw(const Xmm& x, const Operand& op) { vpminsw(x, x, op); }
-void vpminsd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x39); }
+void vpminsd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x39); }
void vpminsd(const Xmm& x, const Operand& op) { vpminsd(x, x, op); }
-void vpminub(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xDA); }
+void vpminub(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xDA); }
void vpminub(const Xmm& x, const Operand& op) { vpminub(x, x, op); }
-void vpminuw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x3A); }
+void vpminuw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x3A); }
void vpminuw(const Xmm& x, const Operand& op) { vpminuw(x, x, op); }
-void vpminud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x3B); }
+void vpminud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x3B); }
void vpminud(const Xmm& x, const Operand& op) { vpminud(x, x, op); }
-void vpmulhuw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE4); }
+void vpmulhuw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE4); }
void vpmulhuw(const Xmm& x, const Operand& op) { vpmulhuw(x, x, op); }
-void vpmulhrsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x0B); }
+void vpmulhrsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x0B); }
void vpmulhrsw(const Xmm& x, const Operand& op) { vpmulhrsw(x, x, op); }
-void vpmulhw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE5); }
+void vpmulhw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE5); }
void vpmulhw(const Xmm& x, const Operand& op) { vpmulhw(x, x, op); }
-void vpmullw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD5); }
+void vpmullw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD5); }
void vpmullw(const Xmm& x, const Operand& op) { vpmullw(x, x, op); }
-void vpmulld(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x40); }
+void vpmulld(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x40); }
void vpmulld(const Xmm& x, const Operand& op) { vpmulld(x, x, op); }
void vpmuludq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66, 0xF4); }
void vpmuludq(const Xmm& x, const Operand& op) { vpmuludq(x, x, op); }
-void vpmuldq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x28); }
+void vpmuldq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x28); }
void vpmuldq(const Xmm& x, const Operand& op) { vpmuldq(x, x, op); }
-void vpor(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xEB); }
+void vpor(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xEB); }
void vpor(const Xmm& x, const Operand& op) { vpor(x, x, op); }
-void vpsadbw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF6); }
+void vpsadbw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF6); }
void vpsadbw(const Xmm& x, const Operand& op) { vpsadbw(x, x, op); }
-void vpshufb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x00); }
-void vpsignb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x08); }
+void vpshufb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x00); }
+void vpsignb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x08); }
void vpsignb(const Xmm& x, const Operand& op) { vpsignb(x, x, op); }
-void vpsignw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x09); }
+void vpsignw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x09); }
void vpsignw(const Xmm& x, const Operand& op) { vpsignw(x, x, op); }
-void vpsignd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x0A); }
+void vpsignd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_YMM, 0x0A); }
void vpsignd(const Xmm& x, const Operand& op) { vpsignd(x, x, op); }
-void vpsllw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF1); }
+void vpsllw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF1); }
void vpsllw(const Xmm& x, const Operand& op) { vpsllw(x, x, op); }
-void vpslld(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF2); }
+void vpslld(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF2); }
void vpslld(const Xmm& x, const Operand& op) { vpslld(x, x, op); }
-void vpsllq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF3); }
+void vpsllq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF3); }
void vpsllq(const Xmm& x, const Operand& op) { vpsllq(x, x, op); }
-void vpsraw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE1); }
+void vpsraw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE1); }
void vpsraw(const Xmm& x, const Operand& op) { vpsraw(x, x, op); }
-void vpsrad(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE2); }
+void vpsrad(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE2); }
void vpsrad(const Xmm& x, const Operand& op) { vpsrad(x, x, op); }
-void vpsrlw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD1); }
+void vpsrlw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD1); }
void vpsrlw(const Xmm& x, const Operand& op) { vpsrlw(x, x, op); }
-void vpsrld(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD2); }
+void vpsrld(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD2); }
void vpsrld(const Xmm& x, const Operand& op) { vpsrld(x, x, op); }
-void vpsrlq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD3); }
+void vpsrlq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD3); }
void vpsrlq(const Xmm& x, const Operand& op) { vpsrlq(x, x, op); }
-void vpsubb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF8); }
+void vpsubb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF8); }
void vpsubb(const Xmm& x, const Operand& op) { vpsubb(x, x, op); }
-void vpsubw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xF9); }
+void vpsubw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xF9); }
void vpsubw(const Xmm& x, const Operand& op) { vpsubw(x, x, op); }
-void vpsubd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xFA); }
+void vpsubd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xFA); }
void vpsubd(const Xmm& x, const Operand& op) { vpsubd(x, x, op); }
-void vpsubq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xFB); }
+void vpsubq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xFB); }
void vpsubq(const Xmm& x, const Operand& op) { vpsubq(x, x, op); }
-void vpsubsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE8); }
+void vpsubsb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE8); }
void vpsubsb(const Xmm& x, const Operand& op) { vpsubsb(x, x, op); }
-void vpsubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE9); }
+void vpsubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xE9); }
void vpsubsw(const Xmm& x, const Operand& op) { vpsubsw(x, x, op); }
-void vpsubusb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD8); }
+void vpsubusb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD8); }
void vpsubusb(const Xmm& x, const Operand& op) { vpsubusb(x, x, op); }
-void vpsubusw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xD9); }
+void vpsubusw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xD9); }
void vpsubusw(const Xmm& x, const Operand& op) { vpsubusw(x, x, op); }
-void vpunpckhbw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x68); }
+void vpunpckhbw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x68); }
void vpunpckhbw(const Xmm& x, const Operand& op) { vpunpckhbw(x, x, op); }
-void vpunpckhwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x69); }
+void vpunpckhwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x69); }
void vpunpckhwd(const Xmm& x, const Operand& op) { vpunpckhwd(x, x, op); }
-void vpunpckhdq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x6A); }
+void vpunpckhdq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x6A); }
void vpunpckhdq(const Xmm& x, const Operand& op) { vpunpckhdq(x, x, op); }
-void vpunpckhqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x6D); }
+void vpunpckhqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x6D); }
void vpunpckhqdq(const Xmm& x, const Operand& op) { vpunpckhqdq(x, x, op); }
-void vpunpcklbw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x60); }
+void vpunpcklbw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x60); }
void vpunpcklbw(const Xmm& x, const Operand& op) { vpunpcklbw(x, x, op); }
-void vpunpcklwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x61); }
+void vpunpcklwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x61); }
void vpunpcklwd(const Xmm& x, const Operand& op) { vpunpcklwd(x, x, op); }
-void vpunpckldq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x62); }
+void vpunpckldq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x62); }
void vpunpckldq(const Xmm& x, const Operand& op) { vpunpckldq(x, x, op); }
-void vpunpcklqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x6C); }
+void vpunpcklqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x6C); }
void vpunpcklqdq(const Xmm& x, const Operand& op) { vpunpcklqdq(x, x, op); }
-void vpxor(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xEF); }
+void vpxor(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xEF); }
void vpxor(const Xmm& x, const Operand& op) { vpxor(x, x, op); }
void vrcpss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_F3, 0x53); }
void vrcpss(const Xmm& x, const Operand& op) { vrcpss(x, x, op); }
void vrsqrtss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_F3, 0x52); }
void vrsqrtss(const Xmm& x, const Operand& op) { vrsqrtss(x, x, op); }
-void vshufpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0xC6, imm); }
+void vshufpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0xC6, imm); }
void vshufpd(const Xmm& x, const Operand& op, uint8 imm) { vshufpd(x, x, op, imm); }
-void vshufps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_SUPPORT_YMM, 0xC6, imm); }
+void vshufps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_YMM, 0xC6, imm); }
void vshufps(const Xmm& x, const Operand& op, uint8 imm) { vshufps(x, x, op, imm); }
void vsqrtsd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_F2, 0x51); }
void vsqrtsd(const Xmm& x, const Operand& op) { vsqrtsd(x, x, op); }
void vsqrtss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_F3, 0x51); }
void vsqrtss(const Xmm& x, const Operand& op) { vsqrtss(x, x, op); }
-void vunpckhpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x15); }
+void vunpckhpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x15); }
void vunpckhpd(const Xmm& x, const Operand& op) { vunpckhpd(x, x, op); }
-void vunpckhps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_SUPPORT_YMM, 0x15); }
+void vunpckhps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_YMM, 0x15); }
void vunpckhps(const Xmm& x, const Operand& op) { vunpckhps(x, x, op); }
-void vunpcklpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_SUPPORT_YMM, 0x14); }
+void vunpcklpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_YMM, 0x14); }
void vunpcklpd(const Xmm& x, const Operand& op) { vunpcklpd(x, x, op); }
-void vunpcklps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_SUPPORT_YMM, 0x14); }
+void vunpcklps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_YMM, 0x14); }
void vunpcklps(const Xmm& x, const Operand& op) { vunpcklps(x, x, op); }
void vaeskeygenassist(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66, 0xDF, imm); }
-void vroundpd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x09, imm); }
-void vroundps(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x08, imm); }
-void vpermilpd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x05, imm); }
-void vpermilps(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x04, imm); }
+void vroundpd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_YMM, 0x09, imm); }
+void vroundps(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_YMM, 0x08, imm); }
+void vpermilpd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_YMM, 0x05, imm); }
+void vpermilps(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66 | T_YMM, 0x04, imm); }
void vpcmpestri(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66, 0x61, imm); }
void vpcmpestrm(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66, 0x60, imm); }
void vpcmpistri(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66, 0x63, imm); }
void vpcmpistrm(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F3A | T_66, 0x62, imm); }
-void vtestps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x0E); }
-void vtestpd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x0F); }
+void vtestps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x0E); }
+void vtestpd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x0F); }
void vcomisd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66, 0x2F); }
void vcomiss(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F, 0x2F); }
-void vcvtdq2ps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_SUPPORT_YMM, 0x5B); }
-void vcvtps2dq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_SUPPORT_YMM, 0x5B); }
-void vcvttps2dq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_SUPPORT_YMM, 0x5B); }
-void vmovapd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_SUPPORT_YMM, 0x28); }
-void vmovaps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_SUPPORT_YMM, 0x28); }
-void vmovddup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F2 | T_SUPPORT_YMM, 0x12); }
-void vmovdqa(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_SUPPORT_YMM, 0x6F); }
-void vmovdqu(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_SUPPORT_YMM, 0x6F); }
-void vmovshdup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_SUPPORT_YMM, 0x16); }
-void vmovsldup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_SUPPORT_YMM, 0x12); }
-void vmovupd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_SUPPORT_YMM, 0x10); }
-void vmovups(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_SUPPORT_YMM, 0x10); }
-void vpabsb(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x1C); }
-void vpabsw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x1D); }
-void vpabsd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x1E); }
+void vcvtdq2ps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x5B); }
+void vcvtps2dq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_YMM, 0x5B); }
+void vcvttps2dq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_YMM, 0x5B); }
+void vmovapd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_YMM, 0x28); }
+void vmovaps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x28); }
+void vmovddup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F2 | T_YMM, 0x12); }
+void vmovdqa(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_YMM, 0x6F); }
+void vmovdqu(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_YMM, 0x6F); }
+void vmovshdup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_YMM, 0x16); }
+void vmovsldup(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_YMM, 0x12); }
+void vmovupd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_YMM, 0x10); }
+void vmovups(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x10); }
+void vpabsb(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x1C); }
+void vpabsw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x1D); }
+void vpabsd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x1E); }
void vphminposuw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66, 0x41); }
-void vpmovsxbw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x20); }
-void vpmovsxbd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x21); }
-void vpmovsxbq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x22); }
-void vpmovsxwd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x23); }
-void vpmovsxwq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x24); }
-void vpmovsxdq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x25); }
-void vpmovzxbw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x30); }
-void vpmovzxbd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x31); }
-void vpmovzxbq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x32); }
-void vpmovzxwd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x33); }
-void vpmovzxwq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x34); }
-void vpmovzxdq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_SUPPORT_YMM, 0x35); }
-void vpshufd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_SUPPORT_YMM, 0x70, imm); }
-void vpshufhw(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_SUPPORT_YMM, 0x70, imm); }
-void vpshuflw(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F | T_F2 | T_SUPPORT_YMM, 0x70, imm); }
+void vpmovsxbw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x20); }
+void vpmovsxbd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x21); }
+void vpmovsxbq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x22); }
+void vpmovsxwd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x23); }
+void vpmovsxwq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x24); }
+void vpmovsxdq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x25); }
+void vpmovzxbw(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x30); }
+void vpmovzxbd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x31); }
+void vpmovzxbq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x32); }
+void vpmovzxwd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x33); }
+void vpmovzxwq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x34); }
+void vpmovzxdq(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66 | T_YMM, 0x35); }
+void vpshufd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_YMM, 0x70, imm); }
+void vpshufhw(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F | T_F3 | T_YMM, 0x70, imm); }
+void vpshuflw(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_0F | T_F2 | T_YMM, 0x70, imm); }
void vptest(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F38 | T_66, 0x17); }
-void vrcpps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_SUPPORT_YMM, 0x53); }
-void vrsqrtps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_SUPPORT_YMM, 0x52); }
-void vsqrtpd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_SUPPORT_YMM, 0x51); }
-void vsqrtps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_SUPPORT_YMM, 0x51); }
+void vrcpps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x53); }
+void vrsqrtps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x52); }
+void vsqrtpd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66 | T_YMM, 0x51); }
+void vsqrtps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x51); }
void vucomisd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_66, 0x2E); }
void vucomiss(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F, 0x2E); }
-void vmovapd(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_66 | T_SUPPORT_YMM, 0x29); }
-void vmovaps(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_SUPPORT_YMM, 0x29); }
-void vmovdqa(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_66 | T_SUPPORT_YMM, 0x7F); }
-void vmovdqu(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_F3 | T_SUPPORT_YMM, 0x7F); }
-void vmovupd(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_66 | T_SUPPORT_YMM, 0x11); }
-void vmovups(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_SUPPORT_YMM, 0x11); }
-void vaddsubpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_SUPPORT_YMM, 0xD0); }
-void vaddsubps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_SUPPORT_YMM, 0xD0); }
-void vhaddpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_SUPPORT_YMM, 0x7C); }
-void vhaddps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_SUPPORT_YMM, 0x7C); }
-void vhsubpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_SUPPORT_YMM, 0x7D); }
-void vhsubps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_SUPPORT_YMM, 0x7D); }
+void vmovapd(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_66 | T_YMM, 0x29); }
+void vmovaps(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_YMM, 0x29); }
+void vmovdqa(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_66 | T_YMM, 0x7F); }
+void vmovdqu(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_F3 | T_YMM, 0x7F); }
+void vmovupd(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_66 | T_YMM, 0x11); }
+void vmovups(const Address& addr, const Xmm& xmm) { opAVX_X_XM_IMM(xmm, addr, T_0F | T_YMM, 0x11); }
+void vaddsubpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_YMM, 0xD0); }
+void vaddsubps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_YMM, 0xD0); }
+void vhaddpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_YMM, 0x7C); }
+void vhaddps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_YMM, 0x7C); }
+void vhsubpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_YMM, 0x7D); }
+void vhsubps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_F2 | T_YMM, 0x7D); }
void vaesenc(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xDC); }
void vaesenclast(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xDD); }
void vaesdec(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xDE); }
void vaesdeclast(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xDF); }
-void vmaskmovps(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x2C); }
-void vmaskmovps(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x2E); }
-void vmaskmovpd(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x2D); }
-void vmaskmovpd(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x2F); }
-void vpmaskmovd(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x8C); }
-void vpmaskmovd(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x8E); }
-void vpmaskmovq(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x8C); }
-void vpmaskmovq(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x8E); }
-void vpermd(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x36); }
-void vpermps(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x16); }
-void vpermq(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_SUPPORT_YMM, 0x00, imm); }
-void vpermpd(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_SUPPORT_YMM, 0x01, imm); }
+void vmaskmovps(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x2C); }
+void vmaskmovps(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x2E); }
+void vmaskmovpd(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x2D); }
+void vmaskmovpd(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x2F); }
+void vpmaskmovd(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x8C); }
+void vpmaskmovd(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W0 | T_YMM, 0x8E); }
+void vpmaskmovq(const Xmm& x1, const Xmm& x2, const Address& addr) { opAVX_X_X_XM(x1, x2, addr, T_0F38 | T_66 | T_W1 | T_YMM, 0x8C); }
+void vpmaskmovq(const Address& addr, const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x2, x1, addr, T_0F38 | T_66 | T_W1 | T_YMM, 0x8E); }
+void vpermd(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x36); }
+void vpermps(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x16); }
+void vpermq(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_YMM, 0x00, imm); }
+void vpermpd(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_0F3A | T_66 | T_W1 | T_YMM, 0x01, imm); }
void cmpeqpd(const Xmm& x, const Operand& op) { cmppd(x, op, 0); }
void vcmpeqpd(const Xmm& x1, const Xmm& x2, const Operand& op) { vcmppd(x1, x2, op, 0); }
void vcmpeqpd(const Xmm& x, const Operand& op) { vcmpeqpd(x, x, op); }
@@ -1337,60 +1337,60 @@
void vmovlpd(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, xm0, addr, T_0F | T_66, 0x13); }
void vmovlps(const Xmm& x, const Operand& op1, const Operand& op2 = Operand()) { if (!op2.isNone() && !op2.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x, op1, op2, T_0F, 0x12); }
void vmovlps(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, xm0, addr, T_0F, 0x13); }
-void vfmadd132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x98); }
-void vfmadd213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xA8); }
-void vfmadd231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xB8); }
-void vfmadd132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x98); }
-void vfmadd213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xA8); }
-void vfmadd231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xB8); }
+void vfmadd132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0x98); }
+void vfmadd213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xA8); }
+void vfmadd231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xB8); }
+void vfmadd132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0x98); }
+void vfmadd213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xA8); }
+void vfmadd231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xB8); }
void vfmadd132sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0x99); }
void vfmadd213sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xA9); }
void vfmadd231sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xB9); }
void vfmadd132ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0x99); }
void vfmadd213ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xA9); }
void vfmadd231ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xB9); }
-void vfmaddsub132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x96); }
-void vfmaddsub213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xA6); }
-void vfmaddsub231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xB6); }
-void vfmaddsub132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x96); }
-void vfmaddsub213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xA6); }
-void vfmaddsub231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xB6); }
-void vfmsubadd132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x97); }
-void vfmsubadd213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xA7); }
-void vfmsubadd231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xB7); }
-void vfmsubadd132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x97); }
-void vfmsubadd213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xA7); }
-void vfmsubadd231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xB7); }
-void vfmsub132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x9A); }
-void vfmsub213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xAA); }
-void vfmsub231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xBA); }
-void vfmsub132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x9A); }
-void vfmsub213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xAA); }
-void vfmsub231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xBA); }
+void vfmaddsub132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0x96); }
+void vfmaddsub213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xA6); }
+void vfmaddsub231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xB6); }
+void vfmaddsub132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0x96); }
+void vfmaddsub213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xA6); }
+void vfmaddsub231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xB6); }
+void vfmsubadd132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0x97); }
+void vfmsubadd213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xA7); }
+void vfmsubadd231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xB7); }
+void vfmsubadd132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0x97); }
+void vfmsubadd213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xA7); }
+void vfmsubadd231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xB7); }
+void vfmsub132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0x9A); }
+void vfmsub213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xAA); }
+void vfmsub231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xBA); }
+void vfmsub132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0x9A); }
+void vfmsub213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xAA); }
+void vfmsub231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xBA); }
void vfmsub132sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0x9B); }
void vfmsub213sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xAB); }
void vfmsub231sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xBB); }
void vfmsub132ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0x9B); }
void vfmsub213ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xAB); }
void vfmsub231ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xBB); }
-void vfnmadd132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x9C); }
-void vfnmadd213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xAC); }
-void vfnmadd231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xBC); }
-void vfnmadd132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x9C); }
-void vfnmadd213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xAC); }
-void vfnmadd231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xBC); }
+void vfnmadd132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0x9C); }
+void vfnmadd213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xAC); }
+void vfnmadd231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xBC); }
+void vfnmadd132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0x9C); }
+void vfnmadd213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xAC); }
+void vfnmadd231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xBC); }
void vfnmadd132sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0x9D); }
void vfnmadd213sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xAD); }
void vfnmadd231sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xBD); }
void vfnmadd132ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0x9D); }
void vfnmadd213ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xAD); }
void vfnmadd231ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xBD); }
-void vfnmsub132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0x9E); }
-void vfnmsub213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xAE); }
-void vfnmsub231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_SUPPORT_YMM, 0xBE); }
-void vfnmsub132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x9E); }
-void vfnmsub213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xAE); }
-void vfnmsub231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0xBE); }
+void vfnmsub132pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0x9E); }
+void vfnmsub213pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xAE); }
+void vfnmsub231pd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1 | T_YMM, 0xBE); }
+void vfnmsub132ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0x9E); }
+void vfnmsub213ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xAE); }
+void vfnmsub231ps(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0 | T_YMM, 0xBE); }
void vfnmsub132sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0x9F); }
void vfnmsub213sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xAF); }
void vfnmsub231sd(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W1, 0xBF); }
@@ -1398,22 +1398,22 @@
void vfnmsub213ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 0xAF); }
void vfnmsub231ss(const Xmm& xmm, const Xmm& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F38 | T_66 | T_W0, 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_SUPPORT_YMM, 0x1A); }
-void vbroadcasti128(const Ymm& y, const Address& addr) { opAVX_X_XM_IMM(y, addr, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x5A); }
-void vbroadcastsd(const Ymm& y, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(y, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x19); }
-void vbroadcastss(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x18); }
-void vpbroadcastb(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x78); }
-void vpbroadcastw(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x79); }
-void vpbroadcastd(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x58); }
-void vpbroadcastq(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_SUPPORT_YMM, 0x59); }
-void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x19, imm); }
-void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x39, imm); }
+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); }
+void vbroadcastsd(const Ymm& y, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(y, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x19); }
+void vbroadcastss(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x18); }
+void vpbroadcastb(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x78); }
+void vpbroadcastw(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x79); }
+void vpbroadcastd(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x58); }
+void vpbroadcastq(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) throw Error(ERR_BAD_COMBINATION); opAVX_X_XM_IMM(x, op, T_0F38 | T_66 | T_W0 | T_YMM, 0x59); }
+void vextractf128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x19, imm); }
+void vextracti128(const Operand& op, const Ymm& y, uint8 imm) { opAVX_X_X_XMcvt(y, y.isXMM() ? xm0 : ym0, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x39, imm); }
void vextractps(const Operand& op, const Xmm& x, uint8 imm) { if (!(op.isREG(32) || op.isMEM()) || x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, op.isREG(), Operand::XMM, T_0F3A | T_66 | T_W0, 0x17, imm); }
-void vinsertf128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x18, imm); }
-void vinserti128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x38, imm); }
-void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x06, imm); }
-void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_SUPPORT_YMM, 0x46, imm); }
-void vlddqu(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_F2 | T_W0 | T_SUPPORT_YMM, 0xF0); }
+void vinsertf128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x18, imm); }
+void vinserti128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XMcvt(y1, y2, op, op.isXMM(), Operand::YMM, T_0F3A | T_66 | T_W0 | T_YMM, 0x38, imm); }
+void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x06, imm); }
+void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8 imm) { opAVX_X_X_XM(y1, y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x46, imm); }
+void vlddqu(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_F2 | T_W0 | T_YMM, 0xF0); }
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); }
@@ -1427,32 +1427,32 @@
void vpinsrw(const Xmm& x, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x, op, !op.isMEM(), Operand::XMM, T_0F | T_66, 0xC4, 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, x2, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0, 0x22, imm); }
void vpinsrd(const Xmm& x, const Operand& op, uint8 imm) { if (!op.isREG(32) && !op.isMEM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x, op, !op.isMEM(), Operand::XMM, T_0F3A | T_66 | T_W0, 0x22, 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_SUPPORT_YMM, 0xD7); }
-void vpslldq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym7 : xm7, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpslldq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym7 : xm7, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpsrldq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym3 : xm3, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpsrldq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym3 : xm3, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpsllw(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym6 : xm6, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x71, imm); }
-void vpsllw(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym6 : xm6, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x71, imm); }
-void vpslld(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym6 : xm6, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x72, imm); }
-void vpslld(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym6 : xm6, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x72, imm); }
-void vpsllq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym6 : xm6, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpsllq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym6 : xm6, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpsraw(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym4 : xm4, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x71, imm); }
-void vpsraw(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym4 : xm4, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x71, imm); }
-void vpsrad(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym4 : xm4, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x72, imm); }
-void vpsrad(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym4 : xm4, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x72, imm); }
-void vpsrlw(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym2 : xm2, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x71, imm); }
-void vpsrlw(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym2 : xm2, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x71, imm); }
-void vpsrld(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym2 : xm2, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x72, imm); }
-void vpsrld(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym2 : xm2, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x72, imm); }
-void vpsrlq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym2 : xm2, x1, x2, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vpsrlq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym2 : xm2, x, x, T_0F | T_66 | T_SUPPORT_YMM, 0x73, imm); }
-void vblendvpd(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x4B, x4.getIdx() << 4); }
+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& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym7 : xm7, x1, x2, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpslldq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym7 : xm7, x, x, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpsrldq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym3 : xm3, x1, x2, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpsrldq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym3 : xm3, x, x, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpsllw(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym6 : xm6, x1, x2, T_0F | T_66 | T_YMM, 0x71, imm); }
+void vpsllw(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym6 : xm6, x, x, T_0F | T_66 | T_YMM, 0x71, imm); }
+void vpslld(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym6 : xm6, x1, x2, T_0F | T_66 | T_YMM, 0x72, imm); }
+void vpslld(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym6 : xm6, x, x, T_0F | T_66 | T_YMM, 0x72, imm); }
+void vpsllq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym6 : xm6, x1, x2, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpsllq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym6 : xm6, x, x, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpsraw(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym4 : xm4, x1, x2, T_0F | T_66 | T_YMM, 0x71, imm); }
+void vpsraw(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym4 : xm4, x, x, T_0F | T_66 | T_YMM, 0x71, imm); }
+void vpsrad(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym4 : xm4, x1, x2, T_0F | T_66 | T_YMM, 0x72, imm); }
+void vpsrad(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym4 : xm4, x, x, T_0F | T_66 | T_YMM, 0x72, imm); }
+void vpsrlw(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym2 : xm2, x1, x2, T_0F | T_66 | T_YMM, 0x71, imm); }
+void vpsrlw(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym2 : xm2, x, x, T_0F | T_66 | T_YMM, 0x71, imm); }
+void vpsrld(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym2 : xm2, x1, x2, T_0F | T_66 | T_YMM, 0x72, imm); }
+void vpsrld(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym2 : xm2, x, x, T_0F | T_66 | T_YMM, 0x72, imm); }
+void vpsrlq(const Xmm& x1, const Xmm& x2, uint8 imm) { opAVX_X_X_XM(x1.isYMM() ? ym2 : xm2, x1, x2, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vpsrlq(const Xmm& x, uint8 imm) { opAVX_X_X_XM(x.isYMM() ? ym2 : xm2, x, x, T_0F | T_66 | T_YMM, 0x73, imm); }
+void vblendvpd(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_YMM, 0x4B, x4.getIdx() << 4); }
void vblendvpd(const Xmm& x1, const Operand& op, const Xmm& x4) { vblendvpd(x1, x1, op, x4); }
-void vblendvps(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x4A, x4.getIdx() << 4); }
+void vblendvps(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_YMM, 0x4A, x4.getIdx() << 4); }
void vblendvps(const Xmm& x1, const Operand& op, const Xmm& x4) { vblendvps(x1, x1, op, x4); }
-void vpblendvb(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_SUPPORT_YMM, 0x4C, x4.getIdx() << 4); }
+void vpblendvb(const Xmm& x1, const Xmm& x2, const Operand& op, const Xmm& x4) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_YMM, 0x4C, x4.getIdx() << 4); }
void vpblendvb(const Xmm& x1, const Operand& op, const Xmm& x4) { vpblendvb(x1, x1, op, x4); }
void vmovd(const Xmm& x, const Reg32& reg) { opAVX_X_X_XM(x, xm0, Xmm(reg.getIdx()), T_0F | T_66 | T_W0, 0x6E); }
void vmovd(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, xm0, addr, T_0F | T_66 | T_W0, 0x6E); }
@@ -1463,12 +1463,12 @@
void vmovq(const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x1, xm0, x2, T_0F | T_F3, 0x7E); }
void vmovhlps(const Xmm& x1, const Xmm& x2, const Operand& op = Operand()) { if (!op.isNone() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x1, x2, op, T_0F, 0x12); }
void vmovlhps(const Xmm& x1, const Xmm& x2, const Operand& op = Operand()) { if (!op.isNone() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x1, x2, op, T_0F, 0x16); }
-void vmovmskpd(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_66 | T_W0 | T_SUPPORT_YMM, 0x50); }
-void vmovmskps(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_W0 | T_SUPPORT_YMM, 0x50); }
-void vmovntdq(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_SUPPORT_YMM, 0xE7); }
-void vmovntpd(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_SUPPORT_YMM, 0x2B); }
-void vmovntps(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_SUPPORT_YMM, 0x2B); }
-void vmovntdqa(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ymm0, addr, T_0F38 | T_66 | T_SUPPORT_YMM, 0x2A); }
+void vmovmskpd(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_66 | T_W0 | T_YMM, 0x50); }
+void vmovmskps(const Reg& r, const Xmm& x) { if (!r.isBit(i32e)) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x.isXMM() ? Xmm(r.getIdx()) : Ymm(r.getIdx()), x.isXMM() ? xm0 : ym0, x, T_0F | T_W0 | T_YMM, 0x50); }
+void vmovntdq(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_YMM, 0xE7); }
+void vmovntpd(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_66 | T_YMM, 0x2B); }
+void vmovntps(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, addr, T_0F | T_YMM, 0x2B); }
+void vmovntdqa(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, x.isXMM() ? xm0 : ymm0, addr, T_0F38 | T_66 | T_YMM, 0x2A); }
void vmovsd(const Xmm& x1, const Xmm& x2, const Operand& op = Operand()) { if (!op.isNone() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(x1, x2, op, T_0F | T_F2, 0x10); }
void vmovsd(const Xmm& x, const Address& addr) { opAVX_X_X_XM(x, xm0, addr, T_0F | T_F2, 0x10); }
void vmovsd(const Address& addr, const Xmm& x) { opAVX_X_X_XM(x, xm0, addr, T_0F | T_F2, 0x11); }
@@ -1481,11 +1481,11 @@
void vcvttsd2si(const Reg32& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F2 | T_W0, 0x2C); }
void vcvtsi2ss(const Xmm& x, const Operand& op1, const Operand& op2 = Operand()) { if (!op2.isNone() && !(op2.isREG(i32e) || op2.isMEM())) throw Error(ERR_BAD_COMBINATION); int type = T_0F | T_F3; if (!op1.isMEM() && !op2.isMEM()) type |= (op1.isREG(32) || op2.isREG(32)) ? T_W0 : T_W1; opAVX_X_X_XMcvt(x, op1, op2, op2.isREG(), Operand::XMM, type, 0x2A); }
void vcvtsi2sd(const Xmm& x, const Operand& op1, const Operand& op2 = Operand()) { if (!op2.isNone() && !(op2.isREG(i32e) || op2.isMEM())) throw Error(ERR_BAD_COMBINATION); int type = T_0F | T_F2; if (!op1.isMEM() && !op2.isMEM()) type |= (op1.isREG(32) || op2.isREG(32)) ? T_W0 : T_W1; opAVX_X_X_XMcvt(x, op1, op2, op2.isREG(), Operand::XMM, type, 0x2A); }
-void vcvtps2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_SUPPORT_YMM, 0x5A); }
-void vcvtdq2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_F3 | T_SUPPORT_YMM, 0xE6); }
-void vcvtpd2ps(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_SUPPORT_YMM, 0x5A); }
-void vcvtpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_F2 | T_SUPPORT_YMM, 0xE6); }
-void vcvttpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_SUPPORT_YMM, 0xE6); }
+void vcvtps2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_YMM, 0x5A); }
+void vcvtdq2pd(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XMcvt(x, x.isXMM() ? xm0 : ym0, op, !op.isMEM(), x.isXMM() ? Operand::XMM : Operand::YMM, T_0F | T_F3 | T_YMM, 0xE6); }
+void vcvtpd2ps(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_YMM, 0x5A); }
+void vcvtpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_F2 | T_YMM, 0xE6); }
+void vcvttpd2dq(const Xmm& x, const Operand& op) { if (x.isYMM()) throw Error(ERR_BAD_COMBINATION); opAVX_X_X_XM(op.isYMM() ? Ymm(x.getIdx()) : x, op.isYMM() ? ym0 : xm0, op, T_0F | T_66 | T_YMM, 0xE6); }
void vcvtph2ps(const Xmm& x, const Operand& op) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F38 | T_66 | T_W0, 0x13); }
void vcvtps2ph(const Operand& op, const Xmm& x, uint8 imm) { if (!op.isMEM() && !op.isXMM()) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W0, 0x1d, imm); }
#ifdef XBYAK64
