T_MUST_EVEX contains T_EVEX
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 5c5e281..07c6865 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -101,20 +101,20 @@
int type;
bool hasIMM;
} tbl[] = {
- { 0xC2, "vcmppd", T_0F | T_EVEX | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_YMM | T_66, true },
- { 0xC2, "vcmpps", T_0F | T_EVEX | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_YMM, true },
- { 0xC2, "vcmpsd", T_0F | T_EVEX | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_F2, true },
- { 0xC2, "vcmpss", T_0F | T_EVEX | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_F3, true },
+ { 0xC2, "vcmppd", T_0F | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_YMM | T_66, true },
+ { 0xC2, "vcmpps", T_0F | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_YMM, true },
+ { 0xC2, "vcmpsd", T_0F | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_F2, true },
+ { 0xC2, "vcmpss", T_0F | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_F3, true },
- { 0x74, "vpcmpeqb", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM, false },
- { 0x75, "vpcmpeqw", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM, false },
- { 0x76, "vpcmpeqd", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_B32, false },
- { 0x29, "vpcmpeqq", T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+ { 0x74, "vpcmpeqb", T_66 | T_0F | T_MUST_EVEX | T_YMM, false },
+ { 0x75, "vpcmpeqw", T_66 | T_0F | T_MUST_EVEX | T_YMM, false },
+ { 0x76, "vpcmpeqd", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_B32, false },
+ { 0x29, "vpcmpeqq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
- { 0x64, "vpcmpgtb", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM, false },
- { 0x65, "vpcmpgtw", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM, false },
- { 0x66, "vpcmpgtd", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_B32, false },
- { 0x37, "vpcmpgtq", T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+ { 0x64, "vpcmpgtb", T_66 | T_0F | T_MUST_EVEX | T_YMM, false },
+ { 0x65, "vpcmpgtw", T_66 | T_0F | T_MUST_EVEX | T_YMM, false },
+ { 0x66, "vpcmpgtd", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_B32, false },
+ { 0x37, "vpcmpgtq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -132,19 +132,19 @@
const char *name;
int type;
} tbl[] = {
- { 0x6F, "vmovdqa32", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x6F, "vmovdqa64", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x6F, "vmovdqu8", T_F2 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x6F, "vmovdqu16", T_F2 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x6F, "vmovdqu32", T_F3 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x6F, "vmovdqu64", T_F3 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x6F, "vmovdqa32", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x6F, "vmovdqa64", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x6F, "vmovdqu8", T_F2 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x6F, "vmovdqu16", T_F2 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x6F, "vmovdqu32", T_F3 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x6F, "vmovdqu64", T_F3 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
std::string type = type2String(p->type);
printf("void %s(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, %s, 0x%02X); }\n", p->name, type.c_str(), p->code);
}
- puts("void vpabsq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_EW1 | T_B64 | T_YMM, 0x1F); }");
+ puts("void vpabsq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_B64 | T_YMM, 0x1F); }");
}
void putM_X()
@@ -154,12 +154,12 @@
const char *name;
int type;
} tbl[] = {
- { 0x7F, "vmovdqa32", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x7F, "vmovdqa64", T_66 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x7F, "vmovdqu8", T_F2 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x7F, "vmovdqu16", T_F2 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x7F, "vmovdqu32", T_F3 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
- { 0x7F, "vmovdqu64", T_F3 | T_0F | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x7F, "vmovdqa32", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x7F, "vmovdqa64", T_66 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x7F, "vmovdqu8", T_F2 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x7F, "vmovdqu16", T_F2 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x7F, "vmovdqu32", T_F3 | T_0F | T_MUST_EVEX | T_YMM | T_EW0 | T_ER_X | T_ER_Y | T_ER_Z },
+ { 0x7F, "vmovdqu64", T_F3 | T_0F | T_MUST_EVEX | T_YMM | T_EW1 | T_ER_X | T_ER_Y | T_ER_Z },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
@@ -176,16 +176,16 @@
int type;
bool hasIMM;
} tbl[] = {
- { 0x03, "valignd", T_EVEX | T_MUST_EVEX | T_66 | T_0F3A | T_EW0 | T_YMM, true },
- { 0x03, "valignq", T_EVEX | T_MUST_EVEX | T_66 | T_0F3A | T_EW1 | T_YMM, true },
- { 0xDB, "vpandd", T_EVEX | T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW0 | T_B32, false },
- { 0xDB, "vpandq", T_EVEX | T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW1 | T_B64, false },
- { 0xDF, "vpandnd", T_EVEX | T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW0 | T_B32, false },
- { 0xDF, "vpandnq", T_EVEX | T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW1 | T_B64, false },
- { 0x3D, "vpmaxsq", T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
- { 0x3F, "vpmaxuq", T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
- { 0x39, "vpminsq", T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
- { 0x3B, "vpminuq", T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+ { 0x03, "valignd", T_MUST_EVEX | T_66 | T_0F3A | T_EW0 | T_YMM, true },
+ { 0x03, "valignq", T_MUST_EVEX | T_66 | T_0F3A | T_EW1 | T_YMM, true },
+ { 0xDB, "vpandd", T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW0 | T_B32, false },
+ { 0xDB, "vpandq", T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW1 | T_B64, false },
+ { 0xDF, "vpandnd", T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW0 | T_B32, false },
+ { 0xDF, "vpandnq", T_MUST_EVEX | T_YMM | T_66 | T_0F | T_EW1 | T_B64, false },
+ { 0x3D, "vpmaxsq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+ { 0x3F, "vpmaxuq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+ { 0x39, "vpminsq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
+ { 0x3B, "vpminuq", T_66 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1 | T_B64, false },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 0b3903e..3ef90c8 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1355,7 +1355,7 @@
T_W1 = 1 << 12,
T_EW0 = 1 << 13,
T_EW1 = 1 << 14,
- T_YMM = 1 << 15,
+ T_YMM = 1 << 15, // support YMM, ZMM
T_EVEX = 1 << 16,
T_ER_X = 1 << 17, // xmm{er}
T_ER_Y = 1 << 18, // ymm{er}
@@ -1363,7 +1363,7 @@
T_SAE_X = 1 << 20, // xmm{sae}
T_SAE_Y = 1 << 21, // ymm{sae}
T_SAE_Z = 1 << 22, // zmm{sae}
- T_MUST_EVEX = 1 << 23,
+ T_MUST_EVEX = 1 << 23, // contains T_EVEX
T_B32 = 1 << 24, // m32bcst
T_B64 = 1 << 25, // m64bcst
T_M_K = 1 << 26, // mem{k}
@@ -1406,7 +1406,7 @@
};
void evex(const Reg& reg, const Reg& base, const Operand *v, int type, int code, bool x = false, bool b = false, int aaa = 0)
{
- if (!(type & T_EVEX)) throw Error(ERR_EVEX_IS_INVALID);
+ if (!(type & (T_EVEX | T_MUST_EVEX))) throw Error(ERR_EVEX_IS_INVALID);
int w = (type & T_EW1) ? 1 : 0;
// bool is256 = (type & T_L1) ? true : (type & T_L0) ? false : reg.isYMM();
uint32 mm = (type & T_0F) ? 1 : (type & T_0F38) ? 2 : (type & T_0F3A) ? 3 : 0;
diff --git a/xbyak/xbyak_avx512.h b/xbyak/xbyak_avx512.h
index 2f35aea..e7b2cc2 100644
--- a/xbyak/xbyak_avx512.h
+++ b/xbyak/xbyak_avx512.h
@@ -64,39 +64,39 @@
void kmovb(const Reg32& r, const Opmask& k) { opVex(r, 0, k, T_L0 | T_0F | T_66 | T_W0, 0x93); }
void kmovd(const Opmask& k, const Reg32& r) { opVex(k, 0, r, T_L0 | T_0F | T_F2 | T_W0, 0x92); }
void kmovd(const Reg32& r, const Opmask& k) { opVex(r, 0, k, T_L0 | T_0F | T_F2 | T_W0, 0x93); }
-void vcmppd(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_EW1 | T_YMM | T_EVEX | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
-void vcmpps(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_EW0 | T_YMM | T_EVEX | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
-void vcmpsd(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_F2 | T_EW1 | T_EVEX | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
-void vcmpss(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_F3 | T_EW0 | T_EVEX | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
-void vpcmpeqb(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_EVEX | T_MUST_EVEX, 0x74); }
-void vpcmpeqw(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_EVEX | T_MUST_EVEX, 0x75); }
-void vpcmpeqd(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_EVEX | T_MUST_EVEX | T_B32, 0x76); }
-void vpcmpeqq(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0x29); }
-void vpcmpgtb(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_EVEX | T_MUST_EVEX, 0x64); }
-void vpcmpgtw(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_EVEX | T_MUST_EVEX, 0x65); }
-void vpcmpgtd(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_EW0 | T_YMM | T_EVEX | T_MUST_EVEX | T_B32, 0x66); }
-void vpcmpgtq(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0x37); }
-void vmovdqa32(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_66 | T_EW0 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
-void vmovdqa64(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_66 | T_EW1 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
-void vmovdqu8(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F2 | T_EW0 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
-void vmovdqu16(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F2 | T_EW1 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
-void vmovdqu32(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F3 | T_EW0 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
-void vmovdqu64(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F3 | T_EW1 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
-void vpabsq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EVEX | T_MUST_EVEX | T_EW1 | T_B64 | T_YMM, 0x1F); }
-void vmovdqa32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_66 | T_EW0 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
-void vmovdqa64(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_66 | T_EW1 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
-void vmovdqu8(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F2 | T_EW0 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
-void vmovdqu16(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F2 | T_EW1 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
-void vmovdqu32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F3 | T_EW0 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
-void vmovdqu64(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F3 | T_EW1 | T_YMM | T_EVEX | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
-void valignd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_EW0 | T_YMM | T_EVEX | T_MUST_EVEX, 0x03, imm); }
-void valignq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX, 0x03, imm); }
-void vpandd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW0 | T_YMM | T_EVEX | T_MUST_EVEX | T_B32, 0xDB); }
-void vpandq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0xDB); }
-void vpandnd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW0 | T_YMM | T_EVEX | T_MUST_EVEX | T_B32, 0xDF); }
-void vpandnq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0xDF); }
-void vpmaxsq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0x3D); }
-void vpmaxuq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0x3F); }
-void vpminsq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0x39); }
-void vpminuq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_EVEX | T_MUST_EVEX | T_B64, 0x3B); }
+void vcmppd(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_EW1 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
+void vcmpps(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
+void vcmpsd(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_F2 | T_EW1 | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
+void vcmpss(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_0F | T_F3 | T_EW0 | T_SAE_Z | T_MUST_EVEX, 0xC2, imm); }
+void vpcmpeqb(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_MUST_EVEX, 0x74); }
+void vpcmpeqw(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_MUST_EVEX, 0x75); }
+void vpcmpeqd(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_MUST_EVEX | T_B32, 0x76); }
+void vpcmpeqq(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x29); }
+void vpcmpgtb(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_MUST_EVEX, 0x64); }
+void vpcmpgtw(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_YMM | T_MUST_EVEX, 0x65); }
+void vpcmpgtd(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F | T_66 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x66); }
+void vpcmpgtq(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x37); }
+void vmovdqa32(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_66 | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vmovdqa64(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_66 | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vmovdqu8(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F2 | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vmovdqu16(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F2 | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vmovdqu32(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F3 | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vmovdqu64(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_0F | T_F3 | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x6F); }
+void vpabsq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_MUST_EVEX | T_EW1 | T_B64 | T_YMM, 0x1F); }
+void vmovdqa32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_66 | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void vmovdqa64(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_66 | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void vmovdqu8(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F2 | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void vmovdqu16(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F2 | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void vmovdqu32(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F3 | T_EW0 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void vmovdqu64(const Address& addr, const Xmm& x) { opAVX_X_XM_IMM(x, addr, T_0F | T_F3 | T_EW1 | T_YMM | T_ER_X | T_ER_Y | T_ER_Z | T_MUST_EVEX, 0x7F); }
+void valignd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_EW0 | T_YMM | T_MUST_EVEX, 0x03, imm); }
+void valignq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_0F3A | T_66 | T_EW1 | T_YMM | T_MUST_EVEX, 0x03, imm); }
+void vpandd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0xDB); }
+void vpandq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0xDB); }
+void vpandnd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0xDF); }
+void vpandnq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0xDF); }
+void vpmaxsq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x3D); }
+void vpmaxuq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x3F); }
+void vpminsq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x39); }
+void vpminuq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_66 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x3B); }
#endif
