diff --git a/CMakeLists.txt b/CMakeLists.txt
index 182a5de..5a710cf 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.5)
 
-project(xbyak LANGUAGES CXX VERSION 7.03)
+project(xbyak LANGUAGES CXX VERSION 7.04)
 
 file(GLOB headers xbyak/*.h)
 
diff --git a/doc/changelog.md b/doc/changelog.md
index df966c9..401bfb8 100644
--- a/doc/changelog.md
+++ b/doc/changelog.md
@@ -1,5 +1,6 @@
 # History
 
+* 2023/Dec/28 ver 7.04 rex2 supports two-byte opecode
 * 2023/Dec/26 ver 7.03 set the default value of dfv to 0
 * 2023/Dec/20 ver 7.02 SHA* support APX
 * 2023/Dec/19 ver 7.01 support AESKLE, WIDE_KL, KEYLOCKER, KEYLOCKER_WIDE, detection of APX10/APX
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index 84e5804..a174e6a 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1144,7 +1144,6 @@
 	// misc
 	{
 		puts("void lea(const Reg& reg, const Address& addr) { if (!reg.isBit(16 | i32e)) XBYAK_THROW(ERR_BAD_SIZE_OF_REGISTER) opMR(addr, reg, 0, 0x8D); }");
-		puts("void bswap(const Reg32e& reg) { opRR(Reg32(1), reg, 0, 0x0F); }");
 		puts("void ret(int imm = 0) { if (imm) { db(0xC2); dw(imm); } else { db(0xC3); } }");
 		puts("void retf(int imm = 0) { if (imm) { db(0xCA); dw(imm); } else { db(0xCB); } }");
 
diff --git a/meson.build b/meson.build
index abe46f8..b184932 100644
--- a/meson.build
+++ b/meson.build
@@ -5,7 +5,7 @@
 project(
 	'xbyak',
 	'cpp',
-	version: '7.03',
+	version: '7.04',
 	license: 'BSD-3-Clause',
 	default_options: 'b_ndebug=if-release'
 )
diff --git a/readme.md b/readme.md
index 42aedf6..b7b6fc8 100644
--- a/readme.md
+++ b/readme.md
@@ -1,5 +1,5 @@
 
-# Xbyak 7.03 [![Badge Build]][Build Status]
+# Xbyak 7.04 [![Badge Build]][Build Status]
 
 *A C++ JIT assembler for x86 (IA32), x64 (AMD64, x86-64)*
 
diff --git a/readme.txt b/readme.txt
index 6b5b4bb..507cce4 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 7.03
+    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 7.04
 
 -----------------------------------------------------------------------------
 ◎概要
@@ -404,6 +404,7 @@
 -----------------------------------------------------------------------------
 ◎履歴
 
+2023/12/28 ver 7.04 2バイトオペコードのrex2対応
 2023/12/26 ver 7.03 dfvのデフォルト値を0に設定
 2023/12/20 ver 7.02 SHA*のAPX対応
 2023/12/19 ver 7.01 AESKLE, WIDE_KL, KEYLOCKER, KEYLOCKER_WIDE対応 APX10/APX判定対応
diff --git a/test/apx.cpp b/test/apx.cpp
index 207389c..70b7d02 100644
--- a/test/apx.cpp
+++ b/test/apx.cpp
@@ -1074,6 +1074,13 @@
 			setb(r31b|T_zu);
 			setb(r15b|T_zu);
 			setb(ptr [r30]);
+
+			bswap(eax);
+			bswap(r8d);
+			bswap(r16d);
+			bswap(rcx);
+			bswap(r9);
+			bswap(r17);
 		}
 	} c;
 	const uint8_t tbl[] = {
@@ -1101,7 +1108,13 @@
 		0x62, 0xdc, 0x7f, 0x18, 0x42, 0xc7,
 		0x62, 0xd4, 0x7f, 0x18, 0x42, 0xc7,
 		0x62, 0xdc, 0x7f, 0x08, 0x42, 0x06,
-
+		// bswap
+		0x0f, 0xc8,
+		0x41, 0x0f, 0xc8,
+		0xd5, 0x90, 0xc8,
+		0x48, 0x0f, 0xc9,
+		0x49, 0x0f, 0xc9,
+		0xd5, 0x98, 0xc9,
 	};
 	const size_t n = sizeof(tbl);
 	CYBOZU_TEST_EQUAL(c.getSize(), n);
@@ -1898,3 +1911,28 @@
 	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
 }
 
+CYBOZU_TEST_AUTO(0x0f_rex2)
+{
+	struct Code : Xbyak::CodeGenerator {
+		Code()
+		{
+			addps(xmm3, ptr [r30+r20*4+0x4]);
+			movups(xmm5, ptr [r16]);
+			movq(r31, xmm5);
+			cvtsd2si(r20, ptr [r30]);
+			bsr(r20, r30);
+		}
+	} c;
+	const uint8_t tbl[] = {
+		0xd5, 0xb1, 0x58, 0x5c, 0xa6, 0x04,
+		0xd5, 0x90, 0x10, 0x28, 0x66,
+		0xd5, 0x99, 0x7e, 0xef, 0xf2,
+		0xd5, 0xd9, 0x2d, 0x26,
+		0xd5, 0xd9, 0xbd, 0xe6,
+
+	};
+	const size_t n = sizeof(tbl);
+	CYBOZU_TEST_EQUAL(c.getSize(), n);
+	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
+}
+
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 0a4337e..990728d 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -155,7 +155,7 @@
 
 enum {
 	DEFAULT_MAX_CODE_SIZE = 4096,
-	VERSION = 0x7030 /* 0xABCD = A.BC(.D) */
+	VERSION = 0x7040 /* 0xABCD = A.BC(.D) */
 };
 
 #ifndef MIE_INTEGER_TYPE_DEFINED
@@ -1736,14 +1736,15 @@
 		db(0xD5);
 		db((rexRXB(4, bit3, r, b, x) << 4) | rex4bit);
 	}
-	void rex(const Operand& op1, const Operand& op2 = Operand(), uint64_t type = 0)
+	// return true if rex2 is selected
+	bool rex(const Operand& op1, const Operand& op2 = Operand(), uint64_t type = 0)
 	{
-		if (op1.getNF() | op2.getNF()) XBYAK_THROW(ERR_INVALID_NF)
-		if (op1.getZU() | op2.getZU()) XBYAK_THROW(ERR_INVALID_ZU)
+		if (op1.getNF() | op2.getNF()) XBYAK_THROW_RET(ERR_INVALID_NF, false)
+		if (op1.getZU() | op2.getZU()) XBYAK_THROW_RET(ERR_INVALID_ZU, false)
 		uint8_t rex = 0;
 		const Operand *p1 = &op1, *p2 = &op2;
 		if (p1->isMEM()) std::swap(p1, p2);
-		if (p1->isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION)
+		if (p1->isMEM()) XBYAK_THROW_RET(ERR_BAD_COMBINATION, false)
 		// except movsx(16bit, 32/64bit)
 		bool p66 = (op1.isBit(16) && !op2.isBit(i32e)) || (op2.isBit(16) && !op1.isBit(i32e));
 		if ((type & T_66) || p66) db(0x66);
@@ -1753,6 +1754,7 @@
 		if (type & T_F3) {
 			db(0xF3);
 		}
+		bool is0F = type & T_0F;
 		if (p2->isMEM()) {
 			const Reg& r = *static_cast<const Reg*>(p1);
 			const Address& addr = p2->getAddress();
@@ -1762,9 +1764,9 @@
 			if (BIT == 64 && addr.is32bit()) db(0x67);
 			rex = rexRXB(3, r.isREG(64), r, base, idx);
 			if (r.hasRex2() || addr.hasRex2()) {
-				if (type & (T_0F|T_0F38|T_0F3A)) XBYAK_THROW(ERR_CANT_USE_REX2)
-				rex2(0, rex, r, base, idx);
-				return;
+				if (type & (T_0F38|T_0F3A)) XBYAK_THROW_RET(ERR_CANT_USE_REX2, false)
+				rex2(is0F, rex, r, base, idx);
+				return true;
 			}
 			if (rex || r.isExt8bit()) rex |= 0x40;
 		} else {
@@ -1773,13 +1775,14 @@
 			// ModRM(reg, base);
 			rex = rexRXB(3, r1.isREG(64) || r2.isREG(64), r2, r1);
 			if (r1.hasRex2() || r2.hasRex2()) {
-				if (type & (T_0F|T_0F38|T_0F3A)) XBYAK_THROW(ERR_CANT_USE_REX2)
-				rex2(0, rex, r2, r1);
-				return;
+				if (type & (T_0F38|T_0F3A)) XBYAK_THROW_RET(ERR_CANT_USE_REX2, 0)
+				rex2(is0F, rex, r2, r1);
+				return true;
 			}
 			if (rex || r1.isExt8bit() || r2.isExt8bit()) rex |= 0x40;
 		}
 		if (rex) db(rex);
+		return false;
 	}
 	// @@@begin of avx_type_def.h
 	static const uint64_t T_NONE = 0ull;
@@ -2022,9 +2025,9 @@
 	}
 	LabelManager labelMgr_;
 	bool isInDisp16(uint32_t x) const { return 0xFFFF8000 <= x || x <= 0x7FFF; }
-	void writeCode(uint64_t type, const Reg& r, int code)
+	void writeCode(uint64_t type, const Reg& r, int code, bool rex2 = false)
 	{
-		if (!(type & T_APX)) {
+		if (!(type&T_APX || rex2)) {
 			if (type & T_0F) {
 				db(0x0F);
 			} else if (type & T_0F38) {
@@ -2037,15 +2040,15 @@
 	}
 	void opRR(const Reg& reg1, const Reg& reg2, uint64_t type, int code)
 	{
-		rex(reg2, reg1, type);
-		writeCode(type, reg1, code);
+		bool rex2 = rex(reg2, reg1, type);
+		writeCode(type, reg1, code, rex2);
 		setModRM(3, reg1.getIdx(), reg2.getIdx());
 	}
 	void opMR(const Address& addr, const Reg& r, uint64_t type, int code, int immSize = 0)
 	{
 		if (addr.is64bitDisp()) XBYAK_THROW(ERR_CANT_USE_64BIT_DISP)
-		rex(addr, r, type);
-		writeCode(type, r, code);
+		bool rex2 = rex(addr, r, type);
+		writeCode(type, r, code, rex2);
 		opAddr(addr, r.getIdx(), immSize);
 	}
 	void opLoadSeg(const Address& addr, const Reg& reg, uint64_t type, int code)
@@ -3133,6 +3136,18 @@
 	{
 		opROO(Reg(), op, x, T_MUST_EVEX, 0xD9);
 	}
+	void bswap(const Reg32e& r)
+	{
+		int idx = r.getIdx();
+		uint8_t rex = (r.isREG(64) ? 8 : 0) | ((idx & 8) ? 1 : 0);
+		if (idx >= 16) {
+			db(0xD5); db((1<<7) | (idx & 16) | rex);
+		} else {
+			if (rex) db(0x40 | rex);
+			db(0x0F);
+		}
+		db(0xC8 + (idx & 7));
+	}
 	/*
 		use single byte nop if useMultiByteNop = false
 	*/
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index ff7900f..7771417 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "7.03"; }
+const char *getVersionString() const { return "7.04"; }
 void aadd(const Address& addr, const Reg32e &reg) { opMR(addr, reg, T_0F38, 0x0FC); }
 void aand(const Address& addr, const Reg32e &reg) { opMR(addr, reg, T_0F38 | T_66, 0x0FC); }
 void adc(const Operand& op, uint32_t imm) { opOI(op, imm, 0x10, 2); }
@@ -55,7 +55,6 @@
 void bndstx(const Address& addr, const BoundsReg& bnd) { opMIB(addr, bnd, T_0F, 0x1B); }
 void bsf(const Reg&reg, const Operand& op) { opRO(reg, op, T_0F, 0xBC, op.isREG(16|i32e)); }
 void bsr(const Reg&reg, const Operand& op) { opRO(reg, op, T_0F, 0xBD, op.isREG(16|i32e)); }
-void bswap(const Reg32e& reg) { opRR(Reg32(1), reg, 0, 0x0F); }
 void bt(const Operand& op, const Reg& reg) { opRO(reg, op, T_0F, 0xA3, op.isREG(16|i32e) && op.getBit() == reg.getBit()); }
 void bt(const Operand& op, uint8_t imm) { opRext(op, 16|i32e, 4, T_0F, 0xba, false, 1); db(imm); }
 void btc(const Operand& op, const Reg& reg) { opRO(reg, op, T_0F, 0xBB, op.isREG(16|i32e) && op.getBit() == reg.getBit()); }