diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 609ed8d..e530752 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -11,9 +11,10 @@ using namespace Xbyak;
 #endif
 
 enum {
-	PP_66 = 1 << 0,
-	PP_F3 = 1 << 1,
-	PP_F2 = 1 << 2,
+	PP_NONE = 1 << 0,
+	PP_66 = 1 << 1,
+	PP_F3 = 1 << 2,
+	PP_F2 = 1 << 3,
 	MM_0F = 1 << 5,
 	MM_0F38 = 1 << 6,
 	MM_0F3A = 1 << 7
@@ -1523,6 +1524,23 @@ void put()
 		printf("void vcvttsd2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, MM_0F | PP_F2, 0x2C, false, 1); }\n");
 		printf("#endif\n");
 	}
+	// haswell gpr(reg, reg, r/m)
+	{
+		const struct Tbl {
+			const char *name;
+			int type;
+			uint8 code;
+		} tbl[] = {
+			{ "andn", MM_0F38, 0xF2 },
+			{ "mulx", PP_F2 | MM_0F38, 0xF6 },
+			{ "pdep", PP_F2 | MM_0F38, 0xF5 },
+			{ "pext", PP_F3 | MM_0F38, 0xF5 },
+		};
+		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+			const Tbl& p = tbl[i];
+			printf("void %s(const Reg32e& r1, const Reg32e& r2, const Operand& op) { opGpr(r1, r2, op, %s, 0x%x); }\n", p.name, type2String(p.type).c_str(), p.code);
+		}
+	}
 }
 
 int main()
diff --git a/test/make_nm.cpp b/test/make_nm.cpp
index e60b771..4fc5da0 100644
--- a/test/make_nm.cpp
+++ b/test/make_nm.cpp
@@ -1843,10 +1843,29 @@ public:
 		printf(
 			"    }\n");
 	}
+	void putGpr()
+	{
+		const char *tbl1[] = {
+			"andn",
+			"mulx",
+			"pdep",
+			"pext",
+		};
+		for (size_t i = 0; i < NUM_OF_ARRAY(tbl1); i++) {
+			const char *name = tbl1[i];
+			put(name, REG32, REG32, REG32 | MEM);
+#ifdef XBYAK64
+			put(name, REG64, REG64, REG64 | MEM);
+#endif
+		}
+	}
 	void put()
 	{
 #ifdef USE_AVX
-#ifndef USE_YASM
+#ifdef USE_YASM
+		putFMA2();
+		putGpr();
+#else
 		putAVX1();
 		putAVX2();
 		putAVX_X_X_XM_omit();
@@ -1858,8 +1877,7 @@ public:
 		putAVX_X_X_IMM_omit();
 		putFMA();
 #endif
-		putFMA2();
-#else
+#else // USE_AVX
 		putJmp();
 #ifndef USE_YASM
 		putSIMPLE();
@@ -1897,7 +1915,7 @@ public:
 		putFpu();
 		putFpuFpu();
 		putCmp();
-#else
+#else // USE_YASM
 		putSSSE3();
 		putSSE4_1();
 		separateFunc();
@@ -1905,9 +1923,9 @@ public:
 		putMov64();
 #ifdef XBYAK64
 		putRip();
-#endif
-#endif
-#endif
+#endif // XBYAK64
+#endif // USE_YASM
+#endif // USE_AVX
 	}
 };
 
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index c179d12..3c4a061 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -1237,6 +1237,83 @@ private:
 	{
 		db(code1); db(code2 | reg.getIdx());
 	}
+	void opGpr(const Reg32e& r1, const Reg32e& r2, const Operand& op, int type, uint8 code)
+	{
+		const unsigned int bit = r1.getBit();
+		if (r2.getBit() != bit || (op.isREG() && op.getBit() != bit)) throw ERR_BAD_COMBINATION;
+		int w = bit == 64;
+		bool x, b;
+		if (op.isMEM()) {
+			const Address& addr = static_cast<const Address&>(op);
+			uint8 rex = addr.getRex();
+			x = (rex & 2) != 0;
+			b = (rex & 1) != 0;
+			if (BIT == 64 && addr.is32bit_) db(0x67);
+			if (BIT == 64 && w == -1) w = (rex & 4) ? 1 : 0;
+		} else {
+			x = false;
+			b = static_cast<const Reg&>(op).isExtIdx();
+		}
+		if (w == -1) w = 0;
+		vex(r1.isExtIdx(), r2.getIdx(), false, type, x, b, w);
+		db(code);
+		if (op.isMEM()) {
+			const Address& addr = static_cast<const Address&>(op);
+			addr.updateRegField(static_cast<uint8>(r1.getIdx()));
+			db(addr.getCode(), static_cast<int>(addr.getSize()));
+		} else {
+			db(getModRM(3, r1.getIdx(), op.getIdx()));
+		}
+	}
+	// support (x, x, x/m), (y, y, y/m)
+	void opAVX_X_X_XM(const Xmm& x1, const Operand& op1, const Operand& op2, int type, int code0, bool supportYMM, int w = -1)
+	{
+		const Xmm *x2;
+		const Operand *op;
+		if (op2.isNone()) {
+			x2 = &x1;
+			op = &op1;
+		} else {
+			if (!(op1.isXMM() || (supportYMM && op1.isYMM()))) throw ERR_BAD_COMBINATION;
+			x2 = static_cast<const Xmm*>(&op1);
+			op = &op2;
+		}
+		// (x1, x2, op)
+		if (!((x1.isXMM() && x2->isXMM()) || (supportYMM && x1.isYMM() && x2->isYMM()))) throw ERR_BAD_COMBINATION;
+		bool x, b;
+		if (op->isMEM()) {
+			const Address& addr = *static_cast<const Address*>(op);
+			uint8 rex = addr.getRex();
+			x = (rex & 2) != 0;
+			b = (rex & 1) != 0;
+			if (BIT == 64 && addr.is32bit_) db(0x67);
+			if (BIT == 64 && w == -1) w = (rex & 4) ? 1 : 0;
+		} else {
+			x = false;
+			b = static_cast<const Reg*>(op)->isExtIdx();
+		}
+		if (w == -1) w = 0;
+		vex(x1.isExtIdx(), x2->getIdx(), x1.isYMM(), type, x, b, w);
+		db(code0);
+		if (op->isMEM()) {
+			const Address& addr = *static_cast<const Address*>(op);
+			addr.updateRegField(static_cast<uint8>(x1.getIdx()));
+			db(addr.getCode(), static_cast<int>(addr.getSize()));
+		} else {
+			db(getModRM(3, x1.getIdx(), op->getIdx()));
+		}
+	}
+	// if cvt then return pointer to Xmm(idx) (or Ymm(idx)), otherwise return op
+	void opAVX_X_X_XMcvt(const Xmm& x1, const Operand& op1, const Operand& op2, bool cvt, Operand::Kind kind, int type, int code0, bool supportYMM, int w = -1)
+	{
+		// use static_cast to avoid calling unintentional copy constructor on gcc
+		opAVX_X_X_XM(x1, op1, cvt ? kind == Operand::XMM ? static_cast<const Operand&>(Xmm(op2.getIdx())) : static_cast<const Operand&>(Ymm(op2.getIdx())) : op2, type, code0, supportYMM, w);
+	}
+	// support (x, x/m, imm), (y, y/m, imm)
+	void opAVX_X_XM_IMM(const Xmm& x, const Operand& op, int type, int code, bool supportYMM, int w = -1, int imm = NONE)
+	{
+		opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, op, type, code, supportYMM, w); if (imm != NONE) db((uint8)imm);
+	}
 public:
 	unsigned int getVersion() const { return VERSION; }
 	using CodeArray::db;
@@ -1608,57 +1685,7 @@ public:
 		db(0xF2);
 		opModRM(reg, op, op.isREG(), op.isMEM(), 0x0F, 0x38, 0xF0 | (op.isBit(8) ? 0 : 1));
 	}
-	// support (x, x, x/m), (y, y, y/m)
-	void opAVX_X_X_XM(const Xmm& x1, const Operand& op1, const Operand& op2, int type, int code0, bool supportYMM, int w = -1)
-	{
-		const Xmm *x2;
-		const Operand *op;
-		if (op2.isNone()) {
-			x2 = &x1;
-			op = &op1;
-		} else {
-			if (!(op1.isXMM() || (supportYMM && op1.isYMM()))) throw ERR_BAD_COMBINATION;
-			x2 = static_cast<const Xmm*>(&op1);
-			op = &op2;
-		}
-		// (x1, x2, op)
-		if (!((x1.isXMM() && x2->isXMM()) || (supportYMM && x1.isYMM() && x2->isYMM()))) throw ERR_BAD_COMBINATION;
-		bool x, b;
-		if (op->isMEM()) {
-			const Address& addr = *static_cast<const Address*>(op);
-			uint8 rex = addr.getRex();
-			x = (rex & 2) != 0;
-			b = (rex & 1) != 0;
-			if (BIT == 64 && addr.is32bit_) db(0x67);
-			if (BIT == 64 && w == -1) w = (rex & 4) ? 1 : 0;
-		} else {
-			x = false;
-			b = static_cast<const Reg*>(op)->isExtIdx();
-		}
-		if (w == -1) w = 0;
-		vex(x1.isExtIdx(), x2->getIdx(), x1.isYMM(), type, x, b, w);
-		db(code0);
-		if (op->isMEM()) {
-			const Address& addr = *static_cast<const Address*>(op);
-			addr.updateRegField(static_cast<uint8>(x1.getIdx()));
-			db(addr.getCode(), static_cast<int>(addr.getSize()));
-		} else {
-			db(getModRM(3, x1.getIdx(), op->getIdx()));
-		}
-	}
-	// if cvt then return pointer to Xmm(idx) (or Ymm(idx)), otherwise return op
-	void opAVX_X_X_XMcvt(const Xmm& x1, const Operand& op1, const Operand& op2, bool cvt, Operand::Kind kind, int type, int code0, bool supportYMM, int w = -1)
-	{
-		// use static_cast to avoid calling unintentional copy constructor on gcc
-		opAVX_X_X_XM(x1, op1, cvt ? kind == Operand::XMM ? static_cast<const Operand&>(Xmm(op2.getIdx())) : static_cast<const Operand&>(Ymm(op2.getIdx())) : op2, type, code0, supportYMM, w);
-	}
-	// support (x, x/m, imm), (y, y/m, imm)
-	void opAVX_X_XM_IMM(const Xmm& x, const Operand& op, int type, int code, bool supportYMM, int w = -1, int imm = NONE)
-	{
-		opAVX_X_X_XM(x, x.isXMM() ? xm0 : ym0, op, type, code, supportYMM, w); if (imm != NONE) db((uint8)imm);
-	}
 	enum { NONE = 256 };
-public:
 	CodeGenerator(size_t maxSize = DEFAULT_MAX_CODE_SIZE, void *userPtr = 0, Allocator *allocator = 0)
 		: CodeArray(maxSize, userPtr, allocator)
 		, mm0(0), mm1(1), mm2(2), mm3(3), mm4(4), mm5(5), mm6(6), mm7(7)
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 0506eb2..0edbaf3 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1368,3 +1368,7 @@ void vcvttss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()
 void vcvtsd2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, MM_0F | PP_F2, 0x2D, false, 1); }
 void vcvttsd2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, MM_0F | PP_F2, 0x2C, false, 1); }
 #endif
+void andn(const Reg32e& r1, const Reg32e& r2, const Operand& op) { opGpr(r1, r2, op, MM_0F38, 0xf2); }
+void mulx(const Reg32e& r1, const Reg32e& r2, const Operand& op) { opGpr(r1, r2, op, MM_0F38 | PP_F2, 0xf6); }
+void pdep(const Reg32e& r1, const Reg32e& r2, const Operand& op) { opGpr(r1, r2, op, MM_0F38 | PP_F2, 0xf5); }
+void pext(const Reg32e& r1, const Reg32e& r2, const Operand& op) { opGpr(r1, r2, op, MM_0F38 | PP_F3, 0xf5); }