put512_X_X_XM() is ok
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 4909da5..c05e46a 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1165,11 +1165,11 @@
{ 0x51, "sqrtsd", T_0F | T_F2 | T_EVEX | T_EW1 | T_ER_X | T_N8, false, true },
{ 0x51, "sqrtss", T_0F | T_F3 | T_EVEX | T_EW0 | T_ER_X | T_N4, false, true },
- { 0x15, "unpckhpd", T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64, false, true },
- { 0x15, "unpckhps", T_0F | T_YMM | T_EVEX | T_EW0 | T_B32, false, true },
+ { 0x15, "unpckhpd", T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_N16 | T_N_VL, false, true },
+ { 0x15, "unpckhps", T_0F | T_YMM | T_EVEX | T_EW0 | T_B32 | T_N16 | T_N_VL, false, true },
- { 0x14, "unpcklpd", T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64, false, true },
- { 0x14, "unpcklps", T_0F | T_YMM | T_EVEX | T_EW0 | T_B32, false, true },
+ { 0x14, "unpcklpd", T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_N16 | T_N_VL, false, true },
+ { 0x14, "unpcklps", T_0F | T_YMM | T_EVEX | T_EW0 | T_B32 | T_N16 | T_N_VL, false, true },
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl *p = &tbl[i];
diff --git a/test/make_512.cpp b/test/make_512.cpp
index e57b909..3854012 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -1435,14 +1435,14 @@
separateFunc();
put_vmov();
separateFunc();
-#endif
put512_X_XM();
separateFunc();
-#if 0
put512_X_X_XM();
separateFunc();
+#endif
put512_X3();
separateFunc();
+#if 0
put512_X3_I();
separateFunc();
put512_FMA();
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 013d0ca..1390ed8 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -954,13 +954,13 @@
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_F3 | T_0F | T_EW0 | T_EVEX | T_ER_X | T_N4, 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_66 | T_0F | T_EW1 | T_YMM | T_EVEX | T_B64, 0x15); }
+void vunpckhpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_EW1 | T_YMM | T_EVEX | T_B64 | T_N16 | T_N_VL, 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_EW0 | T_YMM | T_EVEX | T_B32, 0x15); }
+void vunpckhps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_EW0 | T_YMM | T_EVEX | T_B32 | T_N16 | T_N_VL, 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_66 | T_0F | T_EW1 | T_YMM | T_EVEX | T_B64, 0x14); }
+void vunpcklpd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_EW1 | T_YMM | T_EVEX | T_B64 | T_N16 | T_N_VL, 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_EW0 | T_YMM | T_EVEX | T_B32, 0x14); }
+void vunpcklps(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F | T_EW0 | T_YMM | T_EVEX | T_B32 | T_N16 | T_N_VL, 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_66 | T_0F3A, 0xDF, imm); }
void vroundpd(const Xmm& xm, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(xm, op, T_66 | T_0F3A | T_YMM, 0x09, imm); }
