fix encoding of vpbroadcast{b,w,d,q}, vpinsr{b,w}, vpextr{b,w} for disp8N
diff --git a/gen/avx_type.hpp b/gen/avx_type.hpp
index e6fe5a3..261d809 100644
--- a/gen/avx_type.hpp
+++ b/gen/avx_type.hpp
@@ -24,13 +24,14 @@
T_B32 = 1 << 21, // m32bcst
T_B64 = 1 << 22, // m64bcst
T_M_K = 1 << 23, // mem{k}
- T_N2 = 1 << 24,
- T_N4 = 1 << 25,
- T_N8 = 1 << 26,
- T_N16 = 1 << 27,
- T_N32 = 1 << 28,
- T_N_VL = 1 << 29, // N * (1, 2, 4) for VL
- T_DUP = 1 << 30, // N = (8, 32, 64)
+ T_N1 = 1 << 24,
+ T_N2 = 1 << 25,
+ T_N4 = 1 << 26,
+ T_N8 = 1 << 27,
+ T_N16 = 1 << 28,
+ T_N32 = 1 << 29,
+ T_N_VL = 1 << 30, // N * (1, 2, 4) for VL
+ T_DUP = 1 << 31, // N = (8, 32, 64)
T_XXX
};
@@ -135,6 +136,10 @@
if (!str.empty()) str += " | ";
str += "T_M_K";
}
+ if (type & T_N1) {
+ if (!str.empty()) str += " | ";
+ str += "T_N1";
+ }
if (type & T_N2) {
if (!str.empty()) str += " | ";
str += "T_N2";
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index d29893f..1c8cf9b 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -437,10 +437,10 @@
int type;
int reg;
} tbl[] = {
- { 0x7A, "vpbroadcastb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, 8 },
- { 0x7B, "vpbroadcastw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, 16 },
- { 0x7C, "vpbroadcastd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_N4, 32 },
- { 0x7C, "vpbroadcastq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_N8, 64},
+ { 0x7A, "vpbroadcastb", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 8 },
+ { 0x7B, "vpbroadcastw", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 16 },
+ { 0x7C, "vpbroadcastd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 32 },
+ { 0x7C, "vpbroadcastq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 64},
};
for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
const Tbl& p = tbl[i];
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index b3a3863..b766ca0 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1467,8 +1467,8 @@
bool ew1;
} tbl[] = {
{ "vbroadcastss", 0x18, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
- { "vpbroadcastb", 0x78, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
- { "vpbroadcastw", 0x79, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
+ { "vpbroadcastb", 0x78, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N1 },
+ { "vpbroadcastw", 0x79, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N2 },
{ "vpbroadcastd", 0x58, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_N4 },
{ "vpbroadcastq", 0x59, T_0F38 | T_66 | T_W0 | T_YMM | T_EVEX | T_EW1 | T_N8 },
};
@@ -1491,13 +1491,13 @@
puts("void vstmxcsr(const Address& addr) { opAVX_X_X_XM(xm3, xm0, addr, T_0F, 0xAE); }");
puts("void vmaskmovdqu(const Xmm& x1, const Xmm& x2) { opAVX_X_X_XM(x1, xm0, x2, T_0F | T_66, 0xF7); }");
- puts("void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(8|16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N4, 0x14, imm); }");
- puts("void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N4, 0x15, imm); } }");
+ puts("void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(8|16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N1, 0x14, imm); }");
+ puts("void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N2, 0x15, imm); } }");
puts("void vpextrd(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(32) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x16, imm); }");
puts("void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(64) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }");
- puts("void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_EVEX | T_N4, 0x20, imm); }");
- puts("void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F | T_66 | T_EVEX | T_N4, 0xC4, imm); }");
+ puts("void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_EVEX | T_N1, 0x20, imm); }");
+ puts("void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F | T_66 | T_EVEX | T_N2, 0xC4, imm); }");
puts("void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x22, imm); }");
puts("void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(64) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }");
diff --git a/readme.md b/readme.md
index 87b201c..fa5ae8b 100644
--- a/readme.md
+++ b/readme.md
@@ -1,5 +1,5 @@
-Xbyak 5.32 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
+Xbyak 5.33 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
=============
Abstract
@@ -323,6 +323,7 @@
History
-------------
+* 2016/Dec/08 ver 5.33 fix encoding of vpbroadcast{b,w,d,q}, vpinsr{b,w}, vpextr{b,w} for disp8N
* 2016/Dec/01 ver 5.32 rename __xgetbv() to _xgetbv() to support clang for Visual Studio(thanks to freiro)
* 2016/Nov/27 ver 5.31 rename AVX512_4VNNI to AVX512_4VNNIW
* 2016/Nov/27 ver 5.30 add AVX512_4VNNI, AVX512_4FMAPS instructions(thanks to rsdubtso)
diff --git a/readme.txt b/readme.txt
index 9ca90c9..c6e0b7b 100644
--- a/readme.txt
+++ b/readme.txt
@@ -1,5 +1,5 @@
- C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.32
+ C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.33
-----------------------------------------------------------------------------
◎概要
@@ -333,6 +333,7 @@
-----------------------------------------------------------------------------
◎履歴
+2016/12/06 ver 5.33 disp8N時のvpbroadcast{b,w,d,q}, vpinsr{b,w}, vpextr{b,w}のバグ修正
2016/12/01 ver 5.32 clang for Visual Studioサポートのために__xgetbv()を_xgetbv()に変更(thanks to freiro)
2016/11/27 ver 5.31 AVX512_4VNNIをAVX512_4VNNIWに変更
2016/11/27 ver 5.30 AVX512_4VNNI, AVX512_4FMAPS命令の追加(thanks to rsdubtso)
diff --git a/test/make_512.cpp b/test/make_512.cpp
index e090cdb..60132a8 100644
--- a/test/make_512.cpp
+++ b/test/make_512.cpp
@@ -2089,6 +2089,23 @@
put("vextractf32x4", XMM_KZ, _YMM, IMM8);
#endif
}
+ void putDisp8N()
+ {
+ {
+ const int tbl[] = {
+ -129, -128, -127, 0, 1, 64, 65, 127, 128
+ };
+ for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+ char xs[128], ns[128];
+ int v = tbl[i];
+ CYBOZU_SNPRINTF(xs, sizeof(xs), "xmm0, ptr[eax%+d]", v);
+ CYBOZU_SNPRINTF(ns, sizeof(ns), "xmm0, [eax%+d]", v);
+ put("vpbroadcastb", xs, ns);
+ }
+ }
+ {
+ }
+ }
void putAVX512()
{
#ifdef MIN_TEST
@@ -2147,6 +2164,7 @@
putScatter();
separateFunc();
putClass();
+ putDisp8N();
#endif
}
};
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 322d820..26e0597 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -105,7 +105,7 @@
enum {
DEFAULT_MAX_CODE_SIZE = 4096,
- VERSION = 0x5320 /* 0xABCD = A.BC(D) */
+ VERSION = 0x5330 /* 0xABCD = A.BC(D) */
};
#ifndef MIE_INTEGER_TYPE_DEFINED
@@ -1373,13 +1373,14 @@
T_B32 = 1 << 21, // m32bcst
T_B64 = 1 << 22, // m64bcst
T_M_K = 1 << 23, // mem{k}
- T_N2 = 1 << 24,
- T_N4 = 1 << 25,
- T_N8 = 1 << 26,
- T_N16 = 1 << 27,
- T_N32 = 1 << 28,
- T_N_VL = 1 << 29, // N * (1, 2, 4) for VL
- T_DUP = 1 << 30, // N = (8, 32, 64)
+ T_N1 = 1 << 24,
+ T_N2 = 1 << 25,
+ T_N4 = 1 << 26,
+ T_N8 = 1 << 27,
+ T_N16 = 1 << 28,
+ T_N32 = 1 << 29,
+ T_N_VL = 1 << 30, // N * (1, 2, 4) for VL
+ T_DUP = 1 << 31, // N = (8, 32, 64)
T_XXX
};
void vex(const Reg& reg, const Reg& base, const Operand *v, int type, int code, bool x = false)
@@ -1457,11 +1458,11 @@
} else if (type & T_DUP) {
disp8N = VL == 128 ? 8 : VL == 256 ? 32 : 64;
} else {
- if ((type & (T_N2 | T_N4 | T_N8 | T_N16 | T_N32 | T_N_VL)) == 0) {
+ if ((type & (T_N1 | T_N2 | T_N4 | T_N8 | T_N16 | T_N32 | T_N_VL)) == 0) {
type |= T_N16 | T_N_VL; // default
}
- if (type & (T_N2 | T_N4 | T_N8 | T_N16 | T_N32)) {
- disp8N = (type & T_N2) ? 2 : (type & T_N4) ? 4 : (type & T_N8) ? 8 : (type & T_N16) ? 16 : 32;
+ if (type & (T_N1 | T_N2 | T_N4 | T_N8 | T_N16 | T_N32)) {
+ disp8N = (type & T_N1) ? 1 : (type & T_N2) ? 2 : (type & T_N4) ? 4 : (type & T_N8) ? 8 : (type & T_N16) ? 16 : 32;
if (type & T_N_VL) disp8N *= (VL == 512 ? 4 : VL == 256 ? 2 : 1);
}
}
@@ -1504,8 +1505,8 @@
if (inner::IsInDisp8(disp)) {
mod = mod01;
}
- } else if (disp8N > 1) {
- uint32_t t = disp / disp8N;
+ } else {
+ uint32 t = disp / disp8N;
if (t * disp8N == disp && inner::IsInDisp8(t)) {
disp = t;
mod = mod01;
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 9bc3a84..053493d 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "5.32"; }
+const char *getVersionString() const { return "5.33"; }
void adc(const Operand& op, uint32 imm) { opRM_I(op, imm, 0x10, 2); }
void adc(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x10); }
void adcx(const Reg32e& reg, const Operand& op) { opGen(reg, op, 0xF6, 0x66, isREG32_REG32orMEM, NONE, 0x38); }
@@ -1075,10 +1075,10 @@
void vpblendd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0 | T_YMM, 0x02, imm); }
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 vpblendw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0 | T_YMM, 0x0E, imm); }
-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_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x78); }
+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_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N1, 0x78); }
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_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 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_66 | T_0F38 | T_W0 | T_EW1 | T_YMM | T_EVEX | T_N8, 0x59); }
-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_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N4, 0x79); }
+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_66 | T_0F38 | T_W0 | T_YMM | T_EVEX | T_N2, 0x79); }
void vpclmulqdq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0, 0x44, imm); }
void vpcmpeqb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_YMM, 0x74); }
void vpcmpeqd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_YMM, 0x76); }
@@ -1104,10 +1104,10 @@
void vpermps(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_YMM | T_EVEX | T_B32, 0x16); }
void vpermq(const Ymm& y, const Operand& op, uint8 imm) { opAVX_X_XM_IMM(y, op, T_66 | T_0F3A | T_W1 | T_EW1 | T_YMM | T_EVEX | T_B64, 0x00, imm); }
void vpermq(const Ymm& y1, const Ymm& y2, const Operand& op) { opAVX_X_X_XM(y1, y2, op, T_66 | T_0F38 | T_W0 | T_EW1 | T_YMM | T_EVEX | T_B64, 0x36); }
-void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(8|16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N4, 0x14, imm); }
+void vpextrb(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(8|16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N1, 0x14, imm); }
void vpextrd(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(32) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x16, imm); }
void vpextrq(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(64) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); opVex(x, 0, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x16, imm); }
-void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N4, 0x15, imm); } }
+void vpextrw(const Operand& op, const Xmm& x, uint8 imm) { if (!((op.isREG(16|i32e) || op.isMEM()) && x.isXMM())) throw Error(ERR_BAD_COMBINATION); if (op.isREG() && x.getIdx() < 16) { opAVX_X_X_XM(Xmm(op.getIdx()), xm0, x, T_0F | T_66, 0xC5, imm); } else { opVex(x, 0, op, T_0F3A | T_66 | T_EVEX | T_N2, 0x15, imm); } }
void vpgatherdd(const Xmm& x1, const Address& addr, const Xmm& x2) { opGather(x1, addr, x2, T_0F38 | T_66 | T_W0, 0x90, 1); }
void vpgatherdq(const Xmm& x1, const Address& addr, const Xmm& x2) { opGather(x1, addr, x2, T_0F38 | T_66 | T_W1, 0x90, 0); }
void vpgatherqd(const Xmm& x1, const Address& addr, const Xmm& x2) { opGather(x1, addr, x2, T_0F38 | T_66 | T_W0, 0x91, 2); }
@@ -1119,10 +1119,10 @@
void vphsubd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_YMM, 0x06); }
void vphsubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_YMM, 0x07); }
void vphsubw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_YMM, 0x05); }
-void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_EVEX | T_N4, 0x20, imm); }
+void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_EVEX | T_N1, 0x20, imm); }
void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_W0 | T_EVEX | T_EW0 | T_N4, 0x22, imm); }
void vpinsrq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(64) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F3A | T_66 | T_W1 | T_EVEX | T_EW1 | T_N8, 0x22, imm); }
-void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F | T_66 | T_EVEX | T_N4, 0xC4, imm); }
+void vpinsrw(const Xmm& x1, const Xmm& x2, const Operand& op, uint8 imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) throw Error(ERR_BAD_COMBINATION); opVex(x1, &x2, op, T_0F | T_66 | T_EVEX | T_N2, 0xC4, imm); }
void vpmaddubsw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_YMM | T_EVEX, 0x04); }
void vpmaddwd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_YMM | T_EVEX, 0xF5); }
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); }
@@ -1738,11 +1738,11 @@
void vpblendmd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x64); }
void vpblendmq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x64); }
void vpblendmw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x66); }
-void vpbroadcastb(const Xmm& x, const Reg8& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x7A); }
-void vpbroadcastd(const Xmm& x, const Reg32& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x7C); }
+void vpbroadcastb(const Xmm& x, const Reg8& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x7A); }
+void vpbroadcastd(const Xmm& x, const Reg32& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x7C); }
void vpbroadcastmb2q(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1, 0x2A); }
void vpbroadcastmw2d(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0, 0x3A); }
-void vpbroadcastw(const Xmm& x, const Reg16& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_N4, 0x7B); }
+void vpbroadcastw(const Xmm& x, const Reg16& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x7B); }
void vpcmpb(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX, 0x3F, imm); }
void vpcmpd(const Opmask& k, const Xmm& x, const Operand& op, uint8 imm) { opAVX_K_X_XM(k, x, op, T_66 | T_0F3A | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x1F, imm); }
void vpcmpeqb(const Opmask& k, const Xmm& x, const Operand& op) { opAVX_K_X_XM(k, x, op, T_66 | T_0F | T_YMM | T_MUST_EVEX, 0x74); }
@@ -1902,6 +1902,6 @@
#ifdef XBYAK64
void kmovq(const Opmask& k, const Reg64& r) { opVex(k, 0, r, T_L0 | T_0F | T_F2 | T_W1, 0x92); }
void kmovq(const Reg64& r, const Opmask& k) { opVex(r, 0, k, T_L0 | T_0F | T_F2 | T_W1, 0x93); }
-void vpbroadcastq(const Xmm& x, const Reg64& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_N8, 0x7C); }
+void vpbroadcastq(const Xmm& x, const Reg64& r) { opVex(x, 0, r, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x7C); }
#endif
#endif
