diff --git a/3rdparty/xbyak/xbyak/xbyak.h b/3rdparty/xbyak/xbyak/xbyak.h
index 41b5cc0684..01fddf7505 100644
--- a/3rdparty/xbyak/xbyak/xbyak.h
+++ b/3rdparty/xbyak/xbyak/xbyak.h
@@ -77,7 +77,11 @@
 	#endif
 	#include <windows.h>
 	#include <malloc.h>
-	#define XBYAK_TLS __declspec(thread)
+	#ifdef _MSC_VER
+		#define XBYAK_TLS __declspec(thread)
+	#else
+		#define XBYAK_TLS __thread
+	#endif
 #elif defined(__GNUC__)
 	#include <unistd.h>
 	#include <sys/mman.h>
@@ -95,7 +99,9 @@
 	#include <stdint.h>
 #endif
 
-#if !defined(MFD_CLOEXEC) // defined only linux 3.17 or later
+// MFD_CLOEXEC defined only linux 3.17 or later.
+// Android wraps the memfd_create syscall from API version 30.
+#if !defined(MFD_CLOEXEC) || (defined(__ANDROID__) && __ANDROID_API__ < 30)
 	#undef XBYAK_USE_MEMFD
 #endif
 
@@ -112,7 +118,7 @@
 	#endif
 #endif
 
-#if (__cplusplus >= 201103) || (defined(_MSC_VER) && _MSC_VER >= 1800)
+#if (__cplusplus >= 201103) || (defined(_MSC_VER) && _MSC_VER >= 1900)
 	#undef XBYAK_TLS
 	#define XBYAK_TLS thread_local
 	#define XBYAK_VARIADIC_TEMPLATE
@@ -138,11 +144,18 @@
 	#pragma warning(disable : 4127) /* constant expresison */
 #endif
 
+// disable -Warray-bounds because it may be a bug of gcc. https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104603
+#if defined(__GNUC__) && !defined(__clang__)
+	#define XBYAK_DISABLE_WARNING_ARRAY_BOUNDS
+	#pragma GCC diagnostic push
+	#pragma GCC diagnostic ignored "-Warray-bounds"
+#endif
+
 namespace Xbyak {
 
 enum {
 	DEFAULT_MAX_CODE_SIZE = 4096,
-	VERSION = 0x6000 /* 0xABCD = A.BC(D) */
+	VERSION = 0x6730 /* 0xABCD = A.BC(.D) */
 };
 
 #ifndef MIE_INTEGER_TYPE_DEFINED
@@ -291,10 +304,10 @@ inline void SetError(int err) {
 inline void ClearError() {
 	local::GetErrorRef() = 0;
 }
-inline int GetError() { return local::GetErrorRef(); }
+inline int GetError() { return Xbyak::local::GetErrorRef(); }
 
-#define XBYAK_THROW(err) { local::SetError(err); return; }
-#define XBYAK_THROW_RET(err, r) { local::SetError(err); return r; }
+#define XBYAK_THROW(err) { Xbyak::local::SetError(err); return; }
+#define XBYAK_THROW_RET(err, r) { Xbyak::local::SetError(err); return r; }
 
 #else
 class Error : public std::exception {
@@ -358,14 +371,36 @@ inline const To CastTo(From p) XBYAK_NOEXCEPT
 }
 namespace inner {
 
-static const size_t ALIGN_PAGE_SIZE = 4096;
+#ifdef _WIN32
+struct SystemInfo {
+	SYSTEM_INFO info;
+	SystemInfo()
+	{
+		GetSystemInfo(&info);
+	}
+};
+#endif
+//static const size_t ALIGN_PAGE_SIZE = 4096;
+inline size_t getPageSize()
+{
+#ifdef _WIN32
+	static const SystemInfo si;
+	return si.info.dwPageSize;
+#elif defined(__GNUC__)
+	static const long pageSize = sysconf(_SC_PAGESIZE);
+	if (pageSize > 0) {
+		return (size_t)pageSize;
+	}
+#endif
+	return 4096;
+}
 
 inline bool IsInDisp8(uint32_t x) { return 0xFFFFFF80 <= x || x <= 0x7F; }
 inline bool IsInInt32(uint64_t x) { return ~uint64_t(0x7fffffffu) <= x || x <= 0x7FFFFFFFU; }
 
 inline uint32_t VerifyInInt32(uint64_t x)
 {
-#ifdef XBYAK64
+#if defined(XBYAK64) && !defined(__ILP32__)
 	if (!IsInInt32(x)) XBYAK_THROW_RET(ERR_OFFSET_IS_TOO_BIG, 0)
 #endif
 	return static_cast<uint32_t>(x);
@@ -383,7 +418,8 @@ enum LabelMode {
 	custom allocator
 */
 struct Allocator {
-	virtual uint8_t *alloc(size_t size) { return reinterpret_cast<uint8_t*>(AlignedMalloc(size, inner::ALIGN_PAGE_SIZE)); }
+	explicit Allocator(const std::string& = "") {} // same interface with MmapAllocator
+	virtual uint8_t *alloc(size_t size) { return reinterpret_cast<uint8_t*>(AlignedMalloc(size, inner::getPageSize())); }
 	virtual void free(uint8_t *p) { AlignedFree(p); }
 	virtual ~Allocator() {}
 	/* override to return false if you call protect() manually */
@@ -414,13 +450,24 @@ inline int getMacOsVersion()
 
 } // util
 #endif
-class MmapAllocator : Allocator {
-	typedef XBYAK_STD_UNORDERED_MAP<uintptr_t, size_t> SizeList;
-	SizeList sizeList_;
+class MmapAllocator : public Allocator {
+	struct Allocation {
+		size_t size;
+#if defined(XBYAK_USE_MEMFD)
+		// fd_ is only used with XBYAK_USE_MEMFD. We keep the file open
+		// during the lifetime of each allocation in order to support
+		// checkpoint/restore by unprivileged users.
+		int fd;
+#endif
+	};
+	const std::string name_; // only used with XBYAK_USE_MEMFD
+	typedef XBYAK_STD_UNORDERED_MAP<uintptr_t, Allocation> AllocationList;
+	AllocationList allocList_;
 public:
+	explicit MmapAllocator(const std::string& name = "xbyak") : name_(name) {}
 	uint8_t *alloc(size_t size)
 	{
-		const size_t alignedSizeM1 = inner::ALIGN_PAGE_SIZE - 1;
+		const size_t alignedSizeM1 = inner::getPageSize() - 1;
 		size = (size + alignedSizeM1) & ~alignedSizeM1;
 #if defined(MAP_ANONYMOUS)
 		int mode = MAP_PRIVATE | MAP_ANONYMOUS;
@@ -435,30 +482,42 @@ public:
 #endif
 		int fd = -1;
 #if defined(XBYAK_USE_MEMFD)
-		fd = memfd_create("xbyak", MFD_CLOEXEC);
+		fd = memfd_create(name_.c_str(), MFD_CLOEXEC);
 		if (fd != -1) {
 			mode = MAP_SHARED;
-			if (ftruncate(fd, size) != 0) XBYAK_THROW_RET(ERR_CANT_ALLOC, 0)
+			if (ftruncate(fd, size) != 0) {
+				close(fd);
+				XBYAK_THROW_RET(ERR_CANT_ALLOC, 0)
+			}
 		}
 #endif
 		void *p = mmap(NULL, size, PROT_READ | PROT_WRITE, mode, fd, 0);
-#if defined(XBYAK_USE_MEMFD)
-		if (fd != -1) close(fd);
-#endif
-		if (p == MAP_FAILED) XBYAK_THROW_RET(ERR_CANT_ALLOC, 0)
+		if (p == MAP_FAILED) {
+			if (fd != -1) close(fd);
+			XBYAK_THROW_RET(ERR_CANT_ALLOC, 0)
+		}
 		assert(p);
-		sizeList_[(uintptr_t)p] = size;
+		Allocation &alloc = allocList_[(uintptr_t)p];
+		alloc.size = size;
+#if defined(XBYAK_USE_MEMFD)
+		alloc.fd = fd;
+#endif
 		return (uint8_t*)p;
 	}
 	void free(uint8_t *p)
 	{
 		if (p == 0) return;
-		SizeList::iterator i = sizeList_.find((uintptr_t)p);
-		if (i == sizeList_.end()) XBYAK_THROW(ERR_BAD_PARAMETER)
-		if (munmap((void*)i->first, i->second) < 0) XBYAK_THROW(ERR_MUNMAP)
-		sizeList_.erase(i);
+		AllocationList::iterator i = allocList_.find((uintptr_t)p);
+		if (i == allocList_.end()) XBYAK_THROW(ERR_BAD_PARAMETER)
+		if (munmap((void*)i->first, i->second.size) < 0) XBYAK_THROW(ERR_MUNMAP)
+#if defined(XBYAK_USE_MEMFD)
+		if (i->second.fd != -1) close(i->second.fd);
+#endif
+		allocList_.erase(i);
 	}
 };
+#else
+typedef Allocator MmapAllocator;
 #endif
 
 class Address;
@@ -1176,9 +1235,6 @@ public:
 		size_t pageSize = sysconf(_SC_PAGESIZE);
 		size_t iaddr = reinterpret_cast<size_t>(addr);
 		size_t roundAddr = iaddr & ~(pageSize - static_cast<size_t>(1));
-#ifndef NDEBUG
-		if (pageSize != 4096) fprintf(stderr, "large page(%zd) is used. not tested enough.\n", pageSize);
-#endif
 		return mprotect(reinterpret_cast<void*>(roundAddr), size + (iaddr - roundAddr), mode) == 0;
 #else
 		return true;
@@ -1448,7 +1504,6 @@ public:
 		clabelDefList_.clear();
 		clabelUndefList_.clear();
 		resetLabelPtrList();
-		ClearError();
 	}
 	void enterLocal()
 	{
@@ -1574,6 +1629,7 @@ public:
 	enum LabelType {
 		T_SHORT,
 		T_NEAR,
+		T_FAR, // far jump
 		T_AUTO // T_SHORT if possible
 	};
 private:
@@ -1622,6 +1678,11 @@ private:
 	{
 		return op1.isREG(i32e) && ((op2.isREG(i32e) && op1.getBit() == op2.getBit()) || op2.isMEM());
 	}
+	static inline bool isValidSSE(const Operand& op1)
+	{
+		// SSE instructions do not support XMM16 - XMM31
+		return !(op1.isXMM() && op1.getIdx() >= 16);
+	}
 	void rex(const Operand& op1, const Operand& op2 = Operand())
 	{
 		uint8_t rex = 0;
@@ -1784,9 +1845,16 @@ private:
 	void setSIB(const RegExp& e, int reg, int disp8N = 0)
 	{
 		uint64_t disp64 = e.getDisp();
-#ifdef XBYAK64
+#if defined(XBYAK64) && !defined(__ILP32__)
+#ifdef XBYAK_OLD_DISP_CHECK
+		// treat 0xffffffff as 0xffffffffffffffff
 		uint64_t high = disp64 >> 32;
 		if (high != 0 && high != 0xFFFFFFFF) XBYAK_THROW(ERR_OFFSET_IS_TOO_BIG)
+#else
+		// displacement should be a signed 32-bit value, so also check sign bit
+		uint64_t high = disp64 >> 31;
+		if (high != 0 && high != 0x1FFFFFFFF) XBYAK_THROW(ERR_OFFSET_IS_TOO_BIG)
+#endif
 #endif
 		uint32_t disp = static_cast<uint32_t>(disp64);
 		const Reg& base = e.getBase();
@@ -1887,6 +1955,7 @@ private:
 	template<class T>
 	void opJmp(T& label, LabelType type, uint8_t shortCode, uint8_t longCode, uint8_t longPref)
 	{
+		if (type == T_FAR) XBYAK_THROW(ERR_NOT_SUPPORTED)
 		if (isAutoGrow() && size_ + 16 >= maxSize_) growMemory(); /* avoid splitting code of jmp */
 		size_t offset = 0;
 		if (labelMgr_.getOffset(&offset, label)) { /* label exists */
@@ -1907,6 +1976,7 @@ private:
 	}
 	void opJmpAbs(const void *addr, LabelType type, uint8_t shortCode, uint8_t longCode, uint8_t longPref = 0)
 	{
+		if (type == T_FAR) XBYAK_THROW(ERR_NOT_SUPPORTED)
 		if (isAutoGrow()) {
 			if (!isNEAR(type)) XBYAK_THROW(ERR_ONLY_T_NEAR_IS_SUPPORTED_IN_AUTO_GROW)
 			if (size_ + 16 >= maxSize_) growMemory();
@@ -1919,6 +1989,16 @@ private:
 		}
 
 	}
+	void opJmpOp(const Operand& op, LabelType type, int ext)
+	{
+		const int bit = 16|i32e;
+		if (type == T_FAR) {
+			if (!op.isMEM(bit)) XBYAK_THROW(ERR_NOT_SUPPORTED)
+			opR_ModM(op, bit, ext + 1, 0xFF, NONE, NONE, false);
+		} else {
+			opR_ModM(op, bit, ext, 0xFF, NONE, NONE, true);
+		}
+	}
 	// reg is reg field of ModRM
 	// immSize is the size for immediate value
 	// disp8N = 0(normal), disp8N = 1(force disp32), disp8N = {2, 4, 8} ; compressed displacement
@@ -1945,6 +2025,7 @@ private:
 	void opGen(const Operand& reg, const Operand& op, int code, int pref, bool isValid(const Operand&, const Operand&), int imm8 = NONE, int preCode = NONE)
 	{
 		if (isValid && !isValid(reg, op)) XBYAK_THROW(ERR_BAD_COMBINATION)
+		if (!isValidSSE(reg) || !isValidSSE(op)) XBYAK_THROW(ERR_NOT_SUPPORTED)
 		if (pref != NONE) db(pref);
 		if (op.isMEM()) {
 			opModM(op.getAddress(), reg.getReg(), 0x0F, preCode, code, (imm8 != NONE) ? 1 : 0);
@@ -1955,6 +2036,7 @@ private:
 	}
 	void opMMX_IMM(const Mmx& mmx, int imm8, int code, int ext)
 	{
+		if (!isValidSSE(mmx)) XBYAK_THROW(ERR_NOT_SUPPORTED)
 		if (mmx.isXMM()) db(0x66);
 		opModR(Reg32(ext), mmx, 0x0F, code);
 		db(imm8);
@@ -1965,6 +2047,7 @@ private:
 	}
 	void opMovXMM(const Operand& op1, const Operand& op2, int code, int pref)
 	{
+		if (!isValidSSE(op1) || !isValidSSE(op2)) XBYAK_THROW(ERR_NOT_SUPPORTED)
 		if (pref != NONE) db(pref);
 		if (op1.isXMM() && op2.isMEM()) {
 			opModM(op2.getAddress(), op1.getReg(), 0x0F, code);
@@ -1976,6 +2059,7 @@ private:
 	}
 	void opExt(const Operand& op, const Mmx& mmx, int code, int imm, bool hasMMX2 = false)
 	{
+		if (!isValidSSE(op) || !isValidSSE(mmx)) XBYAK_THROW(ERR_NOT_SUPPORTED)
 		if (hasMMX2 && op.isREG(i32e)) { /* pextrw is special */
 			if (mmx.isXMM()) db(0x66);
 			opModR(op.getReg(), mmx, 0x0F, 0xC5); db(imm);
@@ -2132,9 +2216,6 @@ private:
 	{
 		if (op.isBit(32)) XBYAK_THROW(ERR_BAD_COMBINATION)
 		int w = op.isBit(16);
-#ifdef XBYAK64
-		if (op.isHigh8bit()) XBYAK_THROW(ERR_BAD_COMBINATION)
-#endif
 		bool cond = reg.isREG() && (reg.getBit() > op.getBit());
 		opModRM(reg, op, cond && op.isREG(), cond && op.isMEM(), 0x0F, code | w);
 	}
@@ -2356,18 +2437,21 @@ private:
 		if (addr.getRegExp().getIndex().getKind() != kind) XBYAK_THROW(ERR_BAD_VSIB_ADDRESSING)
 		opVex(x, 0, addr, type, code);
 	}
-	void opVnni(const Xmm& x1, const Xmm& x2, const Operand& op, int type, int code0, PreferredEncoding encoding)
+	void opEncoding(const Xmm& x1, const Xmm& x2, const Operand& op, int type, int code0, PreferredEncoding encoding)
 	{
+		opAVX_X_X_XM(x1, x2, op, type | orEvexIf(encoding), code0);
+	}
+	int orEvexIf(PreferredEncoding encoding) {
 		if (encoding == DefaultEncoding) {
-			encoding = EvexEncoding;
+			encoding = defaultEncoding_;
 		}
 		if (encoding == EvexEncoding) {
 #ifdef XBYAK_DISABLE_AVX512
 			XBYAK_THROW(ERR_EVEX_IS_INVALID)
 #endif
-			type |= T_MUST_EVEX;
+			return T_MUST_EVEX;
 		}
-		opAVX_X_X_XM(x1, x2, op, type, code0);
+		return 0;
 	}
 	void opInOut(const Reg& a, const Reg& d, uint8_t code)
 	{
@@ -2452,6 +2536,7 @@ public:
 #endif
 private:
 	bool isDefaultJmpNEAR_;
+	PreferredEncoding defaultEncoding_;
 public:
 	void L(const std::string& label) { labelMgr_.defineSlabel(label); }
 	void L(Label& label) { labelMgr_.defineClabel(label); }
@@ -2474,13 +2559,13 @@ public:
 
 	// set default type of `jmp` of undefined label to T_NEAR
 	void setDefaultJmpNEAR(bool isNear) { isDefaultJmpNEAR_ = isNear; }
-	void jmp(const Operand& op) { opR_ModM(op, BIT, 4, 0xFF, NONE, NONE, true); }
+	void jmp(const Operand& op, LabelType type = T_AUTO) { opJmpOp(op, type, 4); }
 	void jmp(std::string label, LabelType type = T_AUTO) { opJmp(label, type, 0xEB, 0xE9, 0); }
 	void jmp(const char *label, LabelType type = T_AUTO) { jmp(std::string(label), type); }
 	void jmp(const Label& label, LabelType type = T_AUTO) { opJmp(label, type, 0xEB, 0xE9, 0); }
 	void jmp(const void *addr, LabelType type = T_AUTO) { opJmpAbs(addr, type, 0xEB, 0xE9); }
 
-	void call(const Operand& op) { opR_ModM(op, 16 | i32e, 2, 0xFF, NONE, NONE, true); }
+	void call(const Operand& op, LabelType type = T_AUTO) { opJmpOp(op, type, 2); }
 	// call(string label), not const std::string&
 	void call(std::string label) { opJmp(label, T_NEAR, 0, 0xE8, 0); }
 	void call(const char *label) { call(std::string(label)); }
@@ -2731,11 +2816,13 @@ public:
 		, es(Segment::es), cs(Segment::cs), ss(Segment::ss), ds(Segment::ds), fs(Segment::fs), gs(Segment::gs)
 #endif
 		, isDefaultJmpNEAR_(false)
+		, defaultEncoding_(EvexEncoding)
 	{
 		labelMgr_.set(this);
 	}
 	void reset()
 	{
+		ClearError();
 		resetSize();
 		labelMgr_.reset();
 		labelMgr_.set(this);
@@ -2767,6 +2854,9 @@ public:
 	#undef jnl
 #endif
 
+	// set default encoding to select Vex or Evex
+	void setDefaultEncoding(PreferredEncoding encoding) { defaultEncoding_ = encoding; }
+
 	/*
 		use single byte nop if useMultiByteNop = false
 	*/
@@ -2813,7 +2903,7 @@ public:
 	{
 		if (x == 1) return;
 		if (x < 1 || (x & (x - 1))) XBYAK_THROW(ERR_BAD_ALIGN)
-		if (isAutoGrow() && x > inner::ALIGN_PAGE_SIZE) fprintf(stderr, "warning:autoGrow mode does not support %d align\n", (int)x);
+		if (isAutoGrow()) XBYAK_THROW(ERR_BAD_ALIGN)
 		size_t remain = size_t(getCurr()) % x;
 		if (remain) {
 			nop(x - remain, useMultiByteNop);
@@ -2871,6 +2961,10 @@ static const XBYAK_CONSTEXPR Segment es(Segment::es), cs(Segment::cs), ss(Segmen
 	#pragma warning(pop)
 #endif
 
+#if defined(__GNUC__) && !defined(__clang__)
+	#pragma GCC diagnostic pop
+#endif
+
 } // end of namespace
 
 #endif // XBYAK_XBYAK_H_
diff --git a/3rdparty/xbyak/xbyak/xbyak_bin2hex.h b/3rdparty/xbyak/xbyak/xbyak_bin2hex.h
new file mode 100644
index 0000000000..69ecdbfedc
--- /dev/null
+++ b/3rdparty/xbyak/xbyak/xbyak_bin2hex.h
@@ -0,0 +1,258 @@
+enum {
+	B00000000= 0,
+	B00000001= 1,
+	B00000010= 2,
+	B00000011= 3,
+	B00000100= 4,
+	B00000101= 5,
+	B00000110= 6,
+	B00000111= 7,
+	B00001000= 8,
+	B00001001= 9,
+	B00001010= 10,
+	B00001011= 11,
+	B00001100= 12,
+	B00001101= 13,
+	B00001110= 14,
+	B00001111= 15,
+	B00010000= 16,
+	B00010001= 17,
+	B00010010= 18,
+	B00010011= 19,
+	B00010100= 20,
+	B00010101= 21,
+	B00010110= 22,
+	B00010111= 23,
+	B00011000= 24,
+	B00011001= 25,
+	B00011010= 26,
+	B00011011= 27,
+	B00011100= 28,
+	B00011101= 29,
+	B00011110= 30,
+	B00011111= 31,
+	B00100000= 32,
+	B00100001= 33,
+	B00100010= 34,
+	B00100011= 35,
+	B00100100= 36,
+	B00100101= 37,
+	B00100110= 38,
+	B00100111= 39,
+	B00101000= 40,
+	B00101001= 41,
+	B00101010= 42,
+	B00101011= 43,
+	B00101100= 44,
+	B00101101= 45,
+	B00101110= 46,
+	B00101111= 47,
+	B00110000= 48,
+	B00110001= 49,
+	B00110010= 50,
+	B00110011= 51,
+	B00110100= 52,
+	B00110101= 53,
+	B00110110= 54,
+	B00110111= 55,
+	B00111000= 56,
+	B00111001= 57,
+	B00111010= 58,
+	B00111011= 59,
+	B00111100= 60,
+	B00111101= 61,
+	B00111110= 62,
+	B00111111= 63,
+	B01000000= 64,
+	B01000001= 65,
+	B01000010= 66,
+	B01000011= 67,
+	B01000100= 68,
+	B01000101= 69,
+	B01000110= 70,
+	B01000111= 71,
+	B01001000= 72,
+	B01001001= 73,
+	B01001010= 74,
+	B01001011= 75,
+	B01001100= 76,
+	B01001101= 77,
+	B01001110= 78,
+	B01001111= 79,
+	B01010000= 80,
+	B01010001= 81,
+	B01010010= 82,
+	B01010011= 83,
+	B01010100= 84,
+	B01010101= 85,
+	B01010110= 86,
+	B01010111= 87,
+	B01011000= 88,
+	B01011001= 89,
+	B01011010= 90,
+	B01011011= 91,
+	B01011100= 92,
+	B01011101= 93,
+	B01011110= 94,
+	B01011111= 95,
+	B01100000= 96,
+	B01100001= 97,
+	B01100010= 98,
+	B01100011= 99,
+	B01100100= 100,
+	B01100101= 101,
+	B01100110= 102,
+	B01100111= 103,
+	B01101000= 104,
+	B01101001= 105,
+	B01101010= 106,
+	B01101011= 107,
+	B01101100= 108,
+	B01101101= 109,
+	B01101110= 110,
+	B01101111= 111,
+	B01110000= 112,
+	B01110001= 113,
+	B01110010= 114,
+	B01110011= 115,
+	B01110100= 116,
+	B01110101= 117,
+	B01110110= 118,
+	B01110111= 119,
+	B01111000= 120,
+	B01111001= 121,
+	B01111010= 122,
+	B01111011= 123,
+	B01111100= 124,
+	B01111101= 125,
+	B01111110= 126,
+	B01111111= 127,
+	B10000000= 128,
+	B10000001= 129,
+	B10000010= 130,
+	B10000011= 131,
+	B10000100= 132,
+	B10000101= 133,
+	B10000110= 134,
+	B10000111= 135,
+	B10001000= 136,
+	B10001001= 137,
+	B10001010= 138,
+	B10001011= 139,
+	B10001100= 140,
+	B10001101= 141,
+	B10001110= 142,
+	B10001111= 143,
+	B10010000= 144,
+	B10010001= 145,
+	B10010010= 146,
+	B10010011= 147,
+	B10010100= 148,
+	B10010101= 149,
+	B10010110= 150,
+	B10010111= 151,
+	B10011000= 152,
+	B10011001= 153,
+	B10011010= 154,
+	B10011011= 155,
+	B10011100= 156,
+	B10011101= 157,
+	B10011110= 158,
+	B10011111= 159,
+	B10100000= 160,
+	B10100001= 161,
+	B10100010= 162,
+	B10100011= 163,
+	B10100100= 164,
+	B10100101= 165,
+	B10100110= 166,
+	B10100111= 167,
+	B10101000= 168,
+	B10101001= 169,
+	B10101010= 170,
+	B10101011= 171,
+	B10101100= 172,
+	B10101101= 173,
+	B10101110= 174,
+	B10101111= 175,
+	B10110000= 176,
+	B10110001= 177,
+	B10110010= 178,
+	B10110011= 179,
+	B10110100= 180,
+	B10110101= 181,
+	B10110110= 182,
+	B10110111= 183,
+	B10111000= 184,
+	B10111001= 185,
+	B10111010= 186,
+	B10111011= 187,
+	B10111100= 188,
+	B10111101= 189,
+	B10111110= 190,
+	B10111111= 191,
+	B11000000= 192,
+	B11000001= 193,
+	B11000010= 194,
+	B11000011= 195,
+	B11000100= 196,
+	B11000101= 197,
+	B11000110= 198,
+	B11000111= 199,
+	B11001000= 200,
+	B11001001= 201,
+	B11001010= 202,
+	B11001011= 203,
+	B11001100= 204,
+	B11001101= 205,
+	B11001110= 206,
+	B11001111= 207,
+	B11010000= 208,
+	B11010001= 209,
+	B11010010= 210,
+	B11010011= 211,
+	B11010100= 212,
+	B11010101= 213,
+	B11010110= 214,
+	B11010111= 215,
+	B11011000= 216,
+	B11011001= 217,
+	B11011010= 218,
+	B11011011= 219,
+	B11011100= 220,
+	B11011101= 221,
+	B11011110= 222,
+	B11011111= 223,
+	B11100000= 224,
+	B11100001= 225,
+	B11100010= 226,
+	B11100011= 227,
+	B11100100= 228,
+	B11100101= 229,
+	B11100110= 230,
+	B11100111= 231,
+	B11101000= 232,
+	B11101001= 233,
+	B11101010= 234,
+	B11101011= 235,
+	B11101100= 236,
+	B11101101= 237,
+	B11101110= 238,
+	B11101111= 239,
+	B11110000= 240,
+	B11110001= 241,
+	B11110010= 242,
+	B11110011= 243,
+	B11110100= 244,
+	B11110101= 245,
+	B11110110= 246,
+	B11110111= 247,
+	B11111000= 248,
+	B11111001= 249,
+	B11111010= 250,
+	B11111011= 251,
+	B11111100= 252,
+	B11111101= 253,
+	B11111110= 254,
+	B11111111= 255
+};
diff --git a/3rdparty/xbyak/xbyak/xbyak_mnemonic.h b/3rdparty/xbyak/xbyak/xbyak_mnemonic.h
index 72bcb22ee5..edc76c5bb8 100644
--- a/3rdparty/xbyak/xbyak/xbyak_mnemonic.h
+++ b/3rdparty/xbyak/xbyak/xbyak_mnemonic.h
@@ -1,4 +1,6 @@
-const char *getVersionString() const { return "6.00"; }
+const char *getVersionString() const { return "6.73"; }
+void aadd(const Address& addr, const Reg32e &reg) { opModM(addr, reg, 0x0F, 0x38, 0x0FC); }
+void aand(const Address& addr, const Reg32e &reg) { db(0x66); opModM(addr, reg, 0x0F, 0x38, 0x0FC); }
 void adc(const Operand& op, uint32_t 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); }
@@ -24,6 +26,8 @@ void andnpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x55, 0x66, isXM
 void andnps(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x55, 0x100, isXMM_XMMorMEM); }
 void andpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x54, 0x66, isXMM_XMMorMEM); }
 void andps(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x54, 0x100, isXMM_XMMorMEM); }
+void aor(const Address& addr, const Reg32e &reg) { db(0xF2); opModM(addr, reg, 0x0F, 0x38, 0x0FC); }
+void axor(const Address& addr, const Reg32e &reg) { db(0xF3); opModM(addr, reg, 0x0F, 0x38, 0x0FC); }
 void bextr(const Reg32e& r1, const Operand& op, const Reg32e& r2) { opGpr(r1, op, r2, T_0F38, 0xf7, false); }
 void blendpd(const Xmm& xmm, const Operand& op, int imm) { opGen(xmm, op, 0x0D, 0x66, isXMM_XMMorMEM, static_cast<uint8_t>(imm), 0x3A); }
 void blendps(const Xmm& xmm, const Operand& op, int imm) { opGen(xmm, op, 0x0C, 0x66, isXMM_XMMorMEM, static_cast<uint8_t>(imm), 0x3A); }
@@ -57,9 +61,11 @@ void cbw() { db(0x66); db(0x98); }
 void cdq() { db(0x99); }
 void clc() { db(0xF8); }
 void cld() { db(0xFC); }
+void cldemote(const Address& addr) { opMIB(addr, eax, 0x0F, 0x1C); }
 void clflush(const Address& addr) { opModM(addr, Reg32(7), 0x0F, 0xAE); }
 void clflushopt(const Address& addr) { db(0x66); opModM(addr, Reg32(7), 0x0F, 0xAE); }
 void cli() { db(0xFA); }
+void clwb(const Address& addr) { db(0x66); opMIB(addr, esi, 0x0F, 0xAE); }
 void clzero() { db(0x0F); db(0x01); db(0xFC); }
 void cmc() { db(0xF5); }
 void cmova(const Reg& reg, const Operand& op) { opModRM(reg, op, op.isREG(16 | i32e), op.isMEM(), 0x0F, 0x40 | 7); }//-V524
@@ -323,6 +329,7 @@ void gf2p8affineqb(const Xmm& xmm, const Operand& op, int imm) { opGen(xmm, op,
 void gf2p8mulb(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0xCF, 0x66, isXMM_XMMorMEM, NONE, 0x38); }
 void haddpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x7C, 0x66, isXMM_XMMorMEM); }
 void haddps(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x7C, 0xF2, isXMM_XMMorMEM); }
+void hlt() { db(0xF4); }
 void hsubpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x7D, 0x66, isXMM_XMMorMEM); }
 void hsubps(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x7D, 0xF2, isXMM_XMMorMEM); }
 void idiv(const Operand& op) { opR_ModM(op, 0, 7, 0xF6); }
@@ -500,6 +507,8 @@ void movd(const Mmx& mmx, const Address& addr) { if (mmx.isXMM()) db(0x66); opMo
 void movd(const Mmx& mmx, const Reg32& reg) { if (mmx.isXMM()) db(0x66); opModR(mmx, reg, 0x0F, 0x6E); }
 void movd(const Reg32& reg, const Mmx& mmx) { if (mmx.isXMM()) db(0x66); opModR(mmx, reg, 0x0F, 0x7E); }
 void movddup(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x12, 0xF2, isXMM_XMMorMEM, NONE, NONE); }
+void movdir64b(const Reg& reg, const Address& addr) { db(0x66); opModM(addr, reg.cvt32(), 0x0F, 0x38, 0xF8); }
+void movdiri(const Address& addr, const Reg32e& reg) { opModM(addr, reg, 0x0F, 0x38, 0xF9); }
 void movdq2q(const Mmx& mmx, const Xmm& xmm) { db(0xF2); opModR(mmx, xmm, 0x0F, 0xD6); }
 void movdqa(const Address& addr, const Xmm& xmm) { db(0x66); opModM(addr, xmm, 0x0F, 0x7F); }
 void movdqa(const Xmm& xmm, const Operand& op) { opMMX(xmm, op, 0x6F, 0x66); }
@@ -580,9 +589,9 @@ void pavgb(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0xE0); }
 void pavgw(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0xE3); }
 void pblendvb(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x10, 0x66, isXMM_XMMorMEM, NONE, 0x38); }
 void pblendw(const Xmm& xmm, const Operand& op, int imm) { opGen(xmm, op, 0x0E, 0x66, isXMM_XMMorMEM, static_cast<uint8_t>(imm), 0x3A); }
-void pclmulhqhdq(const Xmm& xmm, const Operand& op) { pclmulqdq(xmm, op, 0x11); }
+void pclmulhqhqdq(const Xmm& xmm, const Operand& op) { pclmulqdq(xmm, op, 0x11); }
 void pclmulhqlqdq(const Xmm& xmm, const Operand& op) { pclmulqdq(xmm, op, 0x01); }
-void pclmullqhdq(const Xmm& xmm, const Operand& op) { pclmulqdq(xmm, op, 0x10); }
+void pclmullqhqdq(const Xmm& xmm, const Operand& op) { pclmulqdq(xmm, op, 0x10); }
 void pclmullqlqdq(const Xmm& xmm, const Operand& op) { pclmulqdq(xmm, op, 0x00); }
 void pclmulqdq(const Xmm& xmm, const Operand& op, int imm) { opGen(xmm, op, 0x44, 0x66, isXMM_XMMorMEM, static_cast<uint8_t>(imm), 0x3A); }
 void pcmpeqb(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x74); }
@@ -649,6 +658,8 @@ void pmuludq(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0xF4); }
 void popcnt(const Reg&reg, const Operand& op) { opSp1(reg, op, 0xF3, 0x0F, 0xB8); }
 void popf() { db(0x9D); }
 void por(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0xEB); }
+void prefetchit0(const Address& addr) { opModM(addr, Reg32(7), 0x0F, 0x18); }
+void prefetchit1(const Address& addr) { opModM(addr, Reg32(6), 0x0F, 0x18); }
 void prefetchnta(const Address& addr) { opModM(addr, Reg32(0), 0x0F, 0x18); }
 void prefetcht0(const Address& addr) { opModM(addr, Reg32(1), 0x0F, 0x18); }
 void prefetcht1(const Address& addr) { opModM(addr, Reg32(2), 0x0F, 0x18); }
@@ -719,6 +730,7 @@ void repne() { db(0xF2); }
 void repnz() { db(0xF2); }
 void repz() { db(0xF3); }
 void ret(int imm = 0) { if (imm) { db(0xC2); dw(imm); } else { db(0xC3); } }
+void retf(int imm = 0) { if (imm) { db(0xCA); dw(imm); } else { db(0xCB); } }
 void rol(const Operand& op, const Reg8& _cl) { opShift(op, _cl, 0); }
 void rol(const Operand& op, int imm) { opShift(op, imm, 0); }
 void ror(const Operand& op, const Reg8& _cl) { opShift(op, _cl, 1); }
@@ -741,6 +753,7 @@ void sbb(const Operand& op1, const Operand& op2) { opRM_RM(op1, op2, 0x18); }
 void scasb() { db(0xAE); }
 void scasd() { db(0xAF); }
 void scasw() { db(0x66); db(0xAF); }
+void serialize() { db(0x0F); db(0x01); db(0xE8); }
 void seta(const Operand& op) { opR_ModM(op, 8, 0, 0x0F, 0x90 | 7); }//-V524
 void setae(const Operand& op) { opR_ModM(op, 8, 0, 0x0F, 0x90 | 3); }//-V524
 void setb(const Operand& op) { opR_ModM(op, 8, 0, 0x0F, 0x90 | 2); }//-V524
@@ -811,10 +824,13 @@ void subsd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x5C, 0xF2, isXMM
 void subss(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x5C, 0xF3, isXMM_XMMorMEM); }
 void sysenter() { db(0x0F); db(0x34); }
 void sysexit() { db(0x0F); db(0x35); }
+void tpause(const Reg32& r) { int idx = r.getIdx(); if (idx > 7) XBYAK_THROW(ERR_BAD_PARAMETER) db(0x66); db(0x0F); db(0xAE); setModRM(3, 6, idx); }
 void tzcnt(const Reg&reg, const Operand& op) { opSp1(reg, op, 0xF3, 0x0F, 0xBC); }
 void ucomisd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x2E, 0x66, isXMM_XMMorMEM); }
 void ucomiss(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x2E, 0x100, isXMM_XMMorMEM); }
 void ud2() { db(0x0F); db(0x0B); }
+void umonitor(const Reg& r) { int idx = r.getIdx(); if (idx > 7) XBYAK_THROW(ERR_BAD_PARAMETER) int bit = r.getBit(); if (BIT != bit) { if ((BIT == 32 && bit == 16) || (BIT == 64 && bit == 32)) { db(0x67); } else { XBYAK_THROW(ERR_BAD_SIZE_OF_REGISTER) } } db(0xF3); db(0x0F); db(0xAE); setModRM(3, 6, idx); }
+void umwait(const Reg32& r) { int idx = r.getIdx(); if (idx > 7) XBYAK_THROW(ERR_BAD_PARAMETER) db(0xF2); db(0x0F); db(0xAE); setModRM(3, 6, idx); }
 void unpckhpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x15, 0x66, isXMM_XMMorMEM); }
 void unpckhps(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x15, 0x100, isXMM_XMMorMEM); }
 void unpcklpd(const Xmm& xmm, const Operand& op) { opGen(xmm, op, 0x14, 0x66, isXMM_XMMorMEM); }
@@ -835,6 +851,8 @@ void vandnpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand())
 void vandnps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM | T_EVEX | T_ER_Z | T_B32, 0x55); }
 void vandpd(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_66 | T_EW1 | T_YMM | T_EVEX | T_ER_Z | T_B64, 0x54); }
 void vandps(const Xmm& xmm, const Operand& op1, const Operand& op2 = Operand()) { opAVX_X_X_XM(xmm, op1, op2, T_0F | T_EW0 | T_YMM | T_EVEX | T_ER_Z | T_B32, 0x54); }
+void vbcstnebf162ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_F3 | T_0F38 | T_W0 | T_YMM | T_B16, 0xB1); }
+void vbcstnesh2ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_66 | T_0F38 | T_W0 | T_YMM | T_B16, 0xB1); }
 void vblendpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0 | T_YMM, 0x0D, imm); }
 void vblendps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0 | T_YMM, 0x0C, imm); }
 void vblendvpd(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, 0x4B, x4.getIdx() << 4); }
@@ -979,6 +997,11 @@ void vcomisd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_N8 | T
 void vcomiss(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_N4 | T_0F | T_EW0 | T_EVEX | T_SAE_X, 0x2F); }
 void vcvtdq2pd(const Xmm& x, const Operand& op) { checkCvt1(x, op); opVex(x, 0, op, T_0F | T_F3 | T_YMM | T_EVEX | T_EW0 | T_B32 | T_N8 | T_N_VL, 0xE6); }
 void vcvtdq2ps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_EW0 | T_YMM | T_EVEX | T_ER_Z | T_B32, 0x5B); }
+void vcvtneebf162ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_F3 | T_0F38 | T_W0 | T_YMM, 0xB0); }
+void vcvtneeph2ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_66 | T_0F38 | T_W0 | T_YMM, 0xB0); }
+void vcvtneobf162ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_F2 | T_0F38 | T_W0 | T_YMM, 0xB0); }
+void vcvtneoph2ps(const Xmm& x, const Address& addr) { opVex(x, 0, addr, T_0F38 | T_W0 | T_YMM, 0xB0); }
+void vcvtneps2bf16(const Xmm& x, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opCvt2(x, op, T_F3 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32 | orEvexIf(encoding), 0x72); }
 void vcvtpd2dq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_F2 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0xE6); }
 void vcvtpd2ps(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_66 | T_YMM | T_EVEX | T_EW1 | T_B64 | T_ER_Z, 0x5A); }
 void vcvtph2ps(const Xmm& x, const Operand& op) { checkCvt1(x, op); opVex(x, 0, op, T_0F38 | T_66 | T_W0 | T_EVEX | T_EW0 | T_N8 | T_N_VL | T_SAE_Y, 0x13); }
@@ -1169,6 +1192,10 @@ void vpbroadcastb(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isME
 void vpbroadcastd(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_XM_IMM(x, op, T_N4 | T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX, 0x58); }
 void vpbroadcastq(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_XM_IMM(x, op, T_N8 | T_66 | T_0F38 | T_W0 | T_EW1 | T_YMM | T_EVEX, 0x59); }
 void vpbroadcastw(const Xmm& x, const Operand& op) { if (!(op.isXMM() || op.isMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opAVX_X_XM_IMM(x, op, T_N2 | T_66 | T_0F38 | T_W0 | T_YMM | T_EVEX, 0x79); }
+void vpclmulhqhqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { vpclmulqdq(x1, x2, op, 0x11); }
+void vpclmulhqlqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { vpclmulqdq(x1, x2, op, 0x01); }
+void vpclmullqhqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { vpclmulqdq(x1, x2, op, 0x10); }
+void vpclmullqlqdq(const Xmm& x1, const Xmm& x2, const Operand& op) { vpclmulqdq(x1, x2, op, 0x00); }
 void vpclmulqdq(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0 | T_YMM | T_EVEX, 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); }
@@ -1182,10 +1209,22 @@ void vpcmpgtq(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1
 void vpcmpgtw(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_YMM, 0x65); }
 void vpcmpistri(const Xmm& xm, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(xm, op, T_66 | T_0F3A, 0x63, imm); }
 void vpcmpistrm(const Xmm& xm, const Operand& op, uint8_t imm) { opAVX_X_XM_IMM(xm, op, T_66 | T_0F3A, 0x62, imm); }
-void vpdpbusd(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVnni(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x50, encoding); }
-void vpdpbusds(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVnni(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x51, encoding); }
-void vpdpwssd(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVnni(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x52, encoding); }
-void vpdpwssds(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVnni(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x53, encoding); }
+void vpdpbssd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F2 | T_0F38 | T_W0 | T_YMM, 0x50); }
+void vpdpbssds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F2 | T_0F38 | T_W0 | T_YMM, 0x51); }
+void vpdpbsud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_YMM, 0x50); }
+void vpdpbsuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_YMM, 0x51); }
+void vpdpbusd(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x50, encoding); }
+void vpdpbusds(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x51, encoding); }
+void vpdpbuud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0x50); }
+void vpdpbuuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0x51); }
+void vpdpwssd(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x52, encoding); }
+void vpdpwssds(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_B32, 0x53, encoding); }
+void vpdpwsud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_YMM, 0xD2); }
+void vpdpwsuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_YMM, 0xD3); }
+void vpdpwusd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_YMM, 0xD2); }
+void vpdpwusds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_YMM, 0xD3); }
+void vpdpwuud(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0xD2); }
+void vpdpwuuds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_YMM, 0xD3); }
 void vperm2f128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { if (!(y1.isYMM() && y2.isYMM() && op.isYMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(y1, &y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x06, imm); }
 void vperm2i128(const Ymm& y1, const Ymm& y2, const Operand& op, uint8_t imm) { if (!(y1.isYMM() && y2.isYMM() && op.isYMEM())) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(y1, &y2, op, T_0F3A | T_66 | T_W0 | T_YMM, 0x46, imm); }
 void vpermd(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, 0x36); }
@@ -1217,6 +1256,8 @@ void vpinsrb(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { if
 void vpinsrd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) XBYAK_THROW(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_t imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(64) || op.isMEM()))) XBYAK_THROW(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_t imm) { if (!(x1.isXMM() && x2.isXMM() && (op.isREG(32) || op.isMEM()))) XBYAK_THROW(ERR_BAD_COMBINATION) opVex(x1, &x2, op, T_0F | T_66 | T_EVEX | T_N2, 0xC4, imm); }
+void vpmadd52huq(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_B64, 0xB5, encoding); }
+void vpmadd52luq(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opEncoding(x1, x2, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_B64, 0xB4, encoding); }
 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); }
@@ -1313,8 +1354,16 @@ void vroundsd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { op
 void vroundss(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0, 0x0A, imm); }
 void vrsqrtps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_YMM, 0x52); }
 void vrsqrtss(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F, 0x52); }
+void vsha512msg1(const Ymm& y, const Xmm& x) { if (!(y.isYMM() && x.isXMM())) XBYAK_THROW(ERR_BAD_PARAMETER) opVex(y, 0, x, T_F2 | T_0F38 | T_W0 | T_YMM, 0xCC); }
+void vsha512msg2(const Ymm& y1, const Ymm& y2) { if (!(y1.isYMM() && y2.isYMM())) XBYAK_THROW(ERR_BAD_PARAMETER) opVex(y1, 0, y2, T_F2 | T_0F38 | T_W0 | T_YMM, 0xCD); }
+void vsha512rnds2(const Ymm& y1, const Ymm& y2, const Xmm& x) { if (!(y1.isYMM() && y2.isYMM() && x.isXMM())) XBYAK_THROW(ERR_BAD_PARAMETER) opVex(y1, &y2, x, T_F2 | T_0F38 | T_W0 | T_YMM, 0xCB); }
 void vshufpd(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F | T_EW1 | T_YMM | T_EVEX | T_B64, 0xC6, imm); }
 void vshufps(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_0F | T_EW0 | T_YMM | T_EVEX | T_B32, 0xC6, imm); }
+void vsm3msg1(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_0F38 | T_W0 | T_EW0 | T_EVEX, 0xDA); }
+void vsm3msg2(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_W0 | T_EW0 | T_EVEX, 0xDA); }
+void vsm3rnds2(const Xmm& x1, const Xmm& x2, const Operand& op, uint8_t imm) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F3A | T_W0 | T_EW0 | T_EVEX, 0xDE, imm); }
+void vsm4key4(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F3 | T_0F38 | T_W0 | T_EW0 | T_EVEX, 0xDA); }
+void vsm4rnds4(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F2 | T_0F38 | T_W0 | T_EW0 | T_EVEX, 0xDA); }
 void vsqrtpd(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_66 | T_0F | T_EW1 | T_YMM | T_EVEX | T_ER_Z | T_B64, 0x51); }
 void vsqrtps(const Xmm& xm, const Operand& op) { opAVX_X_XM_IMM(xm, op, T_0F | T_EW0 | T_YMM | T_EVEX | T_ER_Z | T_B32, 0x51); }
 void vsqrtsd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_N8 | T_F2 | T_0F | T_EW1 | T_EVEX | T_ER_X, 0x51); }
@@ -1339,7 +1388,10 @@ void vzeroupper() { db(0xC5); db(0xF8); db(0x77); }
 void wait() { db(0x9B); }
 void wbinvd() { db(0x0F); db(0x09); }
 void wrmsr() { db(0x0F); db(0x30); }
+void xabort(uint8_t imm) { db(0xC6); db(0xF8); db(imm); }
 void xadd(const Operand& op, const Reg& reg) { opModRM(reg, op, (op.isREG() && reg.isREG() && op.getBit() == reg.getBit()), op.isMEM(), 0x0F, 0xC0 | (reg.isBit(8) ? 0 : 1)); }
+void xbegin(uint32_t rel) { db(0xC7); db(0xF8); dd(rel); }
+void xend() { db(0x0F); db(0x01); db(0xD5); }
 void xgetbv() { db(0x0F); db(0x01); db(0xD0); }
 void xlatb() { db(0xD7); }
 void xor_(const Operand& op, uint32_t imm) { opRM_I(op, imm, 0x30, 6); }
@@ -1620,6 +1672,10 @@ void scasq() { db(0x48); db(0xAF); }
 void stosq() { db(0x48); db(0xAB); }
 void syscall() { db(0x0F); db(0x05); }
 void sysret() { db(0x0F); db(0x07); }
+void clui() { db(0xF3); db(0x0F); db(0x01); db(0xEE); }
+void stui() { db(0xF3); db(0x0F); db(0x01); db(0xEF); }
+void testui() { db(0xF3); db(0x0F); db(0x01); db(0xED); }
+void uiret() { db(0xF3); db(0x0F); db(0x01); db(0xEC); }
 void cmpxchg16b(const Address& addr) { opModM(addr, Reg64(1), 0x0F, 0xC7); }
 void fxrstor64(const Address& addr) { opModM(addr, Reg64(1), 0x0F, 0xAE); }
 void movq(const Reg64& reg, const Mmx& mmx) { if (mmx.isXMM()) db(0x66); opModR(mmx, reg, 0x0F, 0x7E); }
@@ -1627,12 +1683,29 @@ void movq(const Mmx& mmx, const Reg64& reg) { if (mmx.isXMM()) db(0x66); opModR(
 void movsxd(const Reg64& reg, const Operand& op) { if (!op.isBit(32)) XBYAK_THROW(ERR_BAD_COMBINATION) opModRM(reg, op, op.isREG(), op.isMEM(), 0x63); }
 void pextrq(const Operand& op, const Xmm& xmm, uint8_t imm) { if (!op.isREG(64) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opGen(Reg64(xmm.getIdx()), op, 0x16, 0x66, 0, imm, 0x3A); }
 void pinsrq(const Xmm& xmm, const Operand& op, uint8_t imm) { if (!op.isREG(64) && !op.isMEM()) XBYAK_THROW(ERR_BAD_COMBINATION) opGen(Reg64(xmm.getIdx()), op, 0x22, 0x66, 0, imm, 0x3A); }
+void senduipi(const Reg64& r) { db(0xF3); opModR(Reg32(6), r.cvt32(), 0x0F, 0xC7); }
 void vcvtss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1 | T_EVEX | T_EW1 | T_ER_X | T_N8, 0x2D); }
 void vcvttss2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F3 | T_W1 | T_EVEX | T_EW1 | T_SAE_X | T_N8, 0x2C); }
 void vcvtsd2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F2 | T_W1 | T_EVEX | T_EW1 | T_N4 | T_ER_X, 0x2D); }
 void vcvttsd2si(const Reg64& r, const Operand& op) { opAVX_X_X_XM(Xmm(r.getIdx()), xm0, op, T_0F | T_F2 | T_W1 | T_EVEX | T_EW1 | T_N4 | T_SAE_X, 0x2C); }
 void vmovq(const Xmm& x, const Reg64& r) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x6E); }
 void vmovq(const Reg64& r, const Xmm& x) { opAVX_X_X_XM(x, xm0, Xmm(r.getIdx()), T_66 | T_0F | T_W1 | T_EVEX | T_EW1, 0x7E); }
+void cmpbexadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE6, false); }
+void cmpbxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE2, false); }
+void cmplexadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xEE, false); }
+void cmplxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xEC, false); }
+void cmpnbexadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE7, false); }
+void cmpnbxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE3, false); }
+void cmpnlexadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xEF, false); }
+void cmpnlxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xED, false); }
+void cmpnoxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE1, false); }
+void cmpnpxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xEB, false); }
+void cmpnsxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE9, false); }
+void cmpnzxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE5, false); }
+void cmpoxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE0, false); }
+void cmppxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xEA, false); }
+void cmpsxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE8, false); }
+void cmpzxadd(const Address& addr, const Reg32e& r1, const Reg32e& r2) { opGpr(r1, addr, r2, T_66 | T_0F38, 0xE4, false); }
 void ldtilecfg(const Address& addr) { opVex(tmm0, &tmm0, addr, T_0F38 | T_W0, 0x49); }
 void sttilecfg(const Address& addr) { opVex(tmm0, &tmm0, addr, T_66 | T_0F38 | T_W0, 0x49); }
 void tileloadd(const Tmm& tm, const Address& addr) { opAMX(tm, addr, T_F2 | T_0F38 | T_W0, 0x4b); }
@@ -1644,6 +1717,7 @@ void tdpbssd(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T
 void tdpbsud(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T_F3 | T_0F38 | T_W0, 0x5e); }
 void tdpbusd(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T_66 | T_0F38 | T_W0, 0x5e); }
 void tdpbuud(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T_0F38 | T_W0, 0x5e); }
+void tdpfp16ps(const Tmm &x1, const Tmm &x2, const Tmm &x3) { opVex(x1, &x3, x2, T_F2 | T_0F38 | T_W0, 0x5c); }
 void tdpbf16ps(const Tmm& x1, const Tmm& x2, const Tmm& x3) { opVex(x1, &x3, x2, T_F3 | T_0F38 | T_W0, 0x5c); }
 #else
 void jcxz(std::string label) { db(0x67); opJmp(label, T_SHORT, 0xe3, 0, 0); }
@@ -1898,7 +1972,6 @@ void vcompressps(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_N4 |
 void vcompressw(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_N2 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x63); }
 void vcvtdq2ph(const Xmm& x, const Operand& op) { checkCvt4(x, op); opCvt(x, op, T_N16 | T_N_VL | T_MAP5 | T_EW0 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B32, 0x5B); }
 void vcvtne2ps2bf16(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_F2 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x72); }
-void vcvtneps2bf16(const Xmm& x, const Operand& op) { opCvt2(x, op, T_F3 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x72); }
 void vcvtpd2ph(const Xmm& x, const Operand& op) { opCvt5(x, op, T_N16 | T_N_VL | T_66 | T_MAP5 | T_EW1 | T_ER_Z | T_MUST_EVEX | T_B64, 0x5A); }
 void vcvtpd2qq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F | T_EW1 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B64, 0x7B); }
 void vcvtpd2udq(const Xmm& x, const Operand& op) { opCvt2(x, op, T_0F | T_EW1 | T_YMM | T_ER_Z | T_MUST_EVEX | T_B64, 0x79); }
@@ -2132,38 +2205,36 @@ void vpgatherqd(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N4 | T
 void vpgatherqq(const Xmm& x, const Address& addr) { opGather2(x, addr, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_VSIB, 0x91, 0); }
 void vplzcntd(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x44); }
 void vplzcntq(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX | T_B64, 0x44); }
-void vpmadd52huq(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, 0xB5); }
-void vpmadd52luq(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, 0xB4); }
 void vpmaxsq(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, 0x3D); }
 void vpmaxuq(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, 0x3F); }
 void vpminsq(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, 0x39); }
 void vpminuq(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, 0x3B); }
 void vpmovb2m(const Opmask& k, const Xmm& x) { opVex(k, 0, x, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0, 0x29); }
 void vpmovd2m(const Opmask& k, const Xmm& x) { opVex(k, 0, x, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0, 0x39); }
-void vpmovdb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x31, false); }
-void vpmovdw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x33, true); }
+void vpmovdb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x31, false); }
+void vpmovdw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x33, true); }
 void vpmovm2b(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0, 0x28); }
 void vpmovm2d(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW0, 0x38); }
 void vpmovm2q(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, 0x38); }
 void vpmovm2w(const Xmm& x, const Opmask& k) { opVex(x, 0, k, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, 0x28); }
 void vpmovq2m(const Opmask& k, const Xmm& x) { opVex(k, 0, x, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, 0x39); }
-void vpmovqb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N2 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x32, false); }
-void vpmovqd(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x35, true); }
-void vpmovqw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x34, false); }
-void vpmovsdb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x21, false); }
-void vpmovsdw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x23, true); }
-void vpmovsqb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N2 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x22, false); }
-void vpmovsqd(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x25, true); }
-void vpmovsqw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x24, false); }
-void vpmovswb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x20, true); }
-void vpmovusdb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x11, false); }
-void vpmovusdw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x13, true); }
-void vpmovusqb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N2 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x12, false); }
-void vpmovusqd(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x15, true); }
-void vpmovusqw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x14, false); }
-void vpmovuswb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x10, true); }
+void vpmovqb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N2 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x32, false); }
+void vpmovqd(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x35, true); }
+void vpmovqw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x34, false); }
+void vpmovsdb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x21, false); }
+void vpmovsdw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x23, true); }
+void vpmovsqb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N2 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x22, false); }
+void vpmovsqd(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x25, true); }
+void vpmovsqw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x24, false); }
+void vpmovswb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x20, true); }
+void vpmovusdb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x11, false); }
+void vpmovusdw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x13, true); }
+void vpmovusqb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N2 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x12, false); }
+void vpmovusqd(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x15, true); }
+void vpmovusqw(const Operand& op, const Xmm& x) { opVmov(op, x, T_N4 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x14, false); }
+void vpmovuswb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x10, true); }
 void vpmovw2m(const Opmask& k, const Xmm& x) { opVex(k, 0, x, T_F3 | T_0F38 | T_MUST_EVEX | T_YMM | T_EW1, 0x29); }
-void vpmovwb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x30, true); }
+void vpmovwb(const Operand& op, const Xmm& x) { opVmov(op, x, T_N8 | T_N_VL | T_F3 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_M_K, 0x30, true); }
 void vpmullq(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, 0x40); }
 void vpmultishiftqb(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, 0x83); }
 void vpopcntb(const Xmm& x, const Operand& op) { opAVX_X_XM_IMM(x, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX, 0x54); }
diff --git a/3rdparty/xbyak/xbyak/xbyak_util.h b/3rdparty/xbyak/xbyak/xbyak_util.h
index f2f651f266..0bcc134994 100644
--- a/3rdparty/xbyak/xbyak/xbyak_util.h
+++ b/3rdparty/xbyak/xbyak/xbyak_util.h
@@ -4,12 +4,18 @@
 #ifdef XBYAK_ONLY_CLASS_CPU
 #include <stdint.h>
 #include <stdlib.h>
-#include <algorithm>
 #include <assert.h>
 #ifndef XBYAK_THROW
 	#define XBYAK_THROW(x) ;
 	#define XBYAK_THROW_RET(x, y) return y;
 #endif
+#ifndef XBYAK_CONSTEXPR
+#if ((__cplusplus >= 201402L) && !(!defined(__clang__) && defined(__GNUC__) && (__GNUC__ <= 5))) || (defined(_MSC_VER) && _MSC_VER >= 1910)
+	#define XBYAK_CONSTEXPR constexpr
+#else
+	#define XBYAK_CONSTEXPR
+#endif
+#endif
 #else
 #include <string.h>
 
@@ -17,7 +23,6 @@
 	utility class and functions for Xbyak
 	Xbyak::util::Clock ; rdtsc timer
 	Xbyak::util::Cpu ; detect CPU
-	@note this header is UNDER CONSTRUCTION!
 */
 #include "xbyak.h"
 #endif // XBYAK_ONLY_CLASS_CPU
@@ -27,8 +32,8 @@
 #endif
 
 #ifdef XBYAK_INTEL_CPU_SPECIFIC
-#ifdef _MSC_VER
-	#if (_MSC_VER < 1400) && defined(XBYAK32)
+#ifdef _WIN32
+	#if defined(_MSC_VER) && (_MSC_VER < 1400) && defined(XBYAK32)
 		static inline __declspec(naked) void __cpuid(int[4], int)
 		{
 			__asm {
@@ -88,32 +93,69 @@ typedef enum {
    CoreLevel = 2
 } IntelCpuTopologyLevel;
 
+namespace local {
+
+template<uint64_t L, uint64_t H = 0>
+struct TypeT {
+};
+
+template<uint64_t L1, uint64_t H1, uint64_t L2, uint64_t H2>
+XBYAK_CONSTEXPR TypeT<L1 | L2, H1 | H2> operator|(TypeT<L1, H1>, TypeT<L2, H2>) { return TypeT<L1 | L2, H1 | H2>(); }
+
+template<typename T>
+inline T max_(T x, T y) { return x >= y ? x : y; }
+template<typename T>
+inline T min_(T x, T y) { return x < y ? x : y; }
+
+} // local
+
 /**
 	CPU detection class
+	@note static inline const member is supported by c++17 or later, so use template hack
 */
 class Cpu {
-	uint64_t type_;
+public:
+	class Type {
+		uint64_t L;
+		uint64_t H;
+	public:
+		Type(uint64_t L = 0, uint64_t H = 0) : L(L), H(H) { }
+		template<uint64_t L_, uint64_t H_>
+		Type(local::TypeT<L_, H_>) : L(L_), H(H_) {}
+		Type& operator&=(const Type& rhs) { L &= rhs.L; H &= rhs.H; return *this; }
+		Type& operator|=(const Type& rhs) { L |= rhs.L; H |= rhs.H; return *this; }
+		Type operator&(const Type& rhs) const { Type t = *this; t &= rhs; return t; }
+		Type operator|(const Type& rhs) const { Type t = *this; t |= rhs; return t; }
+		bool operator==(const Type& rhs) const { return H == rhs.H && L == rhs.L; }
+		bool operator!=(const Type& rhs) const { return !operator==(rhs); }
+		// without explicit because backward compatilibity
+		operator bool() const { return (H | L) != 0; }
+		uint64_t getL() const { return L; }
+		uint64_t getH() const { return H; }
+	};
+private:
+	Type type_;
 	//system topology
 	bool x2APIC_supported_;
 	static const size_t maxTopologyLevels = 2;
-	unsigned int numCores_[maxTopologyLevels];
+	uint32_t numCores_[maxTopologyLevels];
 
-	static const unsigned int maxNumberCacheLevels = 10;
-	unsigned int dataCacheSize_[maxNumberCacheLevels];
-	unsigned int coresSharignDataCache_[maxNumberCacheLevels];
-	unsigned int dataCacheLevels_;
+	static const uint32_t maxNumberCacheLevels = 10;
+	uint32_t dataCacheSize_[maxNumberCacheLevels];
+	uint32_t coresSharignDataCache_[maxNumberCacheLevels];
+	uint32_t dataCacheLevels_;
 
-	unsigned int get32bitAsBE(const char *x) const
+	uint32_t get32bitAsBE(const char *x) const
 	{
 		return x[0] | (x[1] << 8) | (x[2] << 16) | (x[3] << 24);
 	}
-	unsigned int mask(int n) const
+	uint32_t mask(int n) const
 	{
 		return (1U << n) - 1;
 	}
 	void setFamily()
 	{
-		unsigned int data[4] = {};
+		uint32_t data[4] = {};
 		getCpuid(1, data);
 		stepping = data[0] & mask(4);
 		model = (data[0] >> 4) & mask(4);
@@ -132,17 +174,15 @@ class Cpu {
 			displayModel = model;
 		}
 	}
-	unsigned int extractBit(unsigned int val, unsigned int base, unsigned int end)
+	uint32_t extractBit(uint32_t val, uint32_t base, uint32_t end)
 	{
 		return (val >> base) & ((1u << (end - base)) - 1);
 	}
 	void setNumCores()
 	{
-		if ((type_ & tINTEL) == 0) return;
+		if (!has(tINTEL) && !has(tAMD)) return;
 
-		unsigned int data[4] = {};
-
-		 /* CAUTION: These numbers are configuration as shipped by Intel. */
+		uint32_t data[4] = {};
 		getCpuidEx(0x0, 0, data);
 		if (data[0] >= 0xB) {
 			 /*
@@ -152,7 +192,7 @@ class Cpu {
 				leaf 0xB can be zeroed-out by a hypervisor
 			*/
 			x2APIC_supported_ = true;
-			for (unsigned int i = 0; i < maxTopologyLevels; i++) {
+			for (uint32_t i = 0; i < maxTopologyLevels; i++) {
 				getCpuidEx(0xB, i, data);
 				IntelCpuTopologyLevel level = (IntelCpuTopologyLevel)extractBit(data[2], 8, 15);
 				if (level == SmtLevel || level == CoreLevel) {
@@ -162,8 +202,8 @@ class Cpu {
 			/*
 				Fallback values in case a hypervisor has 0xB leaf zeroed-out.
 			*/
-			numCores_[SmtLevel - 1] = (std::max)(1u, numCores_[SmtLevel - 1]);
-			numCores_[CoreLevel - 1] = (std::max)(numCores_[SmtLevel - 1], numCores_[CoreLevel - 1]);
+			numCores_[SmtLevel - 1] = local::max_(1u, numCores_[SmtLevel - 1]);
+			numCores_[CoreLevel - 1] = local::max_(numCores_[SmtLevel - 1], numCores_[CoreLevel - 1]);
 		} else {
 			/*
 				Failed to deremine num of cores without x2APIC support.
@@ -176,14 +216,55 @@ class Cpu {
 	}
 	void setCacheHierarchy()
 	{
-		if ((type_ & tINTEL) == 0) return;
-		const unsigned int NO_CACHE = 0;
-		const unsigned int DATA_CACHE = 1;
-//		const unsigned int INSTRUCTION_CACHE = 2;
-		const unsigned int UNIFIED_CACHE = 3;
-		unsigned int smt_width = 0;
-		unsigned int logical_cores = 0;
-		unsigned int data[4] = {};
+		if (!has(tINTEL) && !has(tAMD)) return;
+
+		// https://github.com/amd/ZenDNN/blob/a08bf9a9efc160a69147cdecfb61cc85cc0d4928/src/cpu/x64/xbyak/xbyak_util.h#L236-L288
+		if (has(tAMD)) {
+			// There are 3 Data Cache Levels (L1, L2, L3)
+			dataCacheLevels_ = 3;
+			const uint32_t leaf = 0x8000001D; // for modern AMD CPus
+			// Sub leaf value ranges from 0 to 3
+			// Sub leaf value 0 refers to L1 Data Cache
+			// Sub leaf value 1 refers to L1 Instruction Cache
+			// Sub leaf value 2 refers to L2 Cache
+			// Sub leaf value 3 refers to L3 Cache
+			// For legacy AMD CPU, use leaf 0x80000005 for L1 cache
+			// and 0x80000006 for L2 and L3 cache
+			int cache_index = 0;
+			for (uint32_t sub_leaf = 0; sub_leaf <= dataCacheLevels_; sub_leaf++) {
+				// Skip sub_leaf = 1 as it refers to
+				// L1 Instruction Cache (not required)
+				if (sub_leaf == 1) {
+					continue;
+				}
+				uint32_t data[4] = {};
+				getCpuidEx(leaf, sub_leaf, data);
+				// Cache Size = Line Size * Partitions * Associativity * Cache Sets
+				dataCacheSize_[cache_index] =
+					(extractBit(data[1], 22, 31) + 1) // Associativity-1
+					* (extractBit(data[1], 12, 21) + 1) // Partitions-1
+					* (extractBit(data[1], 0, 11) + 1) // Line Size
+					* (data[2] + 1);
+				// Calculate the number of cores sharing the current data cache
+				int smt_width = numCores_[0];
+				int logical_cores = numCores_[1];
+				int actual_logical_cores = extractBit(data[0], 14, 25) /* # of cores * # of threads */ + 1;
+				if (logical_cores != 0) {
+					actual_logical_cores = local::min_(actual_logical_cores, logical_cores);
+				}
+				coresSharignDataCache_[cache_index] = local::max_(actual_logical_cores / smt_width, 1);
+				++cache_index;
+			}
+			return;
+		}
+		// intel
+		const uint32_t NO_CACHE = 0;
+		const uint32_t DATA_CACHE = 1;
+//		const uint32_t INSTRUCTION_CACHE = 2;
+		const uint32_t UNIFIED_CACHE = 3;
+		uint32_t smt_width = 0;
+		uint32_t logical_cores = 0;
+		uint32_t data[4] = {};
 
 		if (x2APIC_supported_) {
 			smt_width = numCores_[0];
@@ -201,12 +282,12 @@ class Cpu {
 		*/
 		for (int i = 0; dataCacheLevels_ < maxNumberCacheLevels; i++) {
 			getCpuidEx(0x4, i, data);
-			unsigned int cacheType = extractBit(data[0], 0, 4);
+			uint32_t cacheType = extractBit(data[0], 0, 4);
 			if (cacheType == NO_CACHE) break;
 			if (cacheType == DATA_CACHE || cacheType == UNIFIED_CACHE) {
-				unsigned int actual_logical_cores = extractBit(data[0], 14, 25) + 1;
+				uint32_t actual_logical_cores = extractBit(data[0], 14, 25) + 1;
 				if (logical_cores != 0) { // true only if leaf 0xB is supported and valid
-					actual_logical_cores = (std::min)(actual_logical_cores, logical_cores);
+					actual_logical_cores = local::min_(actual_logical_cores, logical_cores);
 				}
 				assert(actual_logical_cores != 0);
 				dataCacheSize_[dataCacheLevels_] =
@@ -216,7 +297,7 @@ class Cpu {
 					* (data[2] + 1);
 				if (cacheType == DATA_CACHE && smt_width == 0) smt_width = actual_logical_cores;
 				assert(smt_width != 0);
-				coresSharignDataCache_[dataCacheLevels_] = (std::max)(actual_logical_cores / smt_width, 1u);
+				coresSharignDataCache_[dataCacheLevels_] = local::max_(actual_logical_cores / smt_width, 1u);
 				dataCacheLevels_++;
 			}
 		}
@@ -231,7 +312,7 @@ public:
 	int displayFamily; // family + extFamily
 	int displayModel; // model + extModel
 
-	unsigned int getNumCores(IntelCpuTopologyLevel level) const {
+	uint32_t getNumCores(IntelCpuTopologyLevel level) const {
 		if (!x2APIC_supported_) XBYAK_THROW_RET(ERR_X2APIC_IS_NOT_SUPPORTED, 0)
 		switch (level) {
 		case SmtLevel: return numCores_[level - 1];
@@ -240,13 +321,13 @@ public:
 		}
 	}
 
-	unsigned int getDataCacheLevels() const { return dataCacheLevels_; }
-	unsigned int getCoresSharingDataCache(unsigned int i) const
+	uint32_t getDataCacheLevels() const { return dataCacheLevels_; }
+	uint32_t getCoresSharingDataCache(uint32_t i) const
 	{
 		if (i >= dataCacheLevels_) XBYAK_THROW_RET(ERR_BAD_PARAMETER, 0)
 		return coresSharignDataCache_[i];
 	}
-	unsigned int getDataCacheSize(unsigned int i) const
+	uint32_t getDataCacheSize(uint32_t i) const
 	{
 		if (i >= dataCacheLevels_) XBYAK_THROW_RET(ERR_BAD_PARAMETER, 0)
 		return dataCacheSize_[i];
@@ -255,10 +336,10 @@ public:
 	/*
 		data[] = { eax, ebx, ecx, edx }
 	*/
-	static inline void getCpuid(unsigned int eaxIn, unsigned int data[4])
+	static inline void getCpuid(uint32_t eaxIn, uint32_t data[4])
 	{
 #ifdef XBYAK_INTEL_CPU_SPECIFIC
-	#ifdef _MSC_VER
+	#ifdef _WIN32
 		__cpuid(reinterpret_cast<int*>(data), eaxIn);
 	#else
 		__cpuid(eaxIn, data[0], data[1], data[2], data[3]);
@@ -268,10 +349,10 @@ public:
 		(void)data;
 #endif
 	}
-	static inline void getCpuidEx(unsigned int eaxIn, unsigned int ecxIn, unsigned int data[4])
+	static inline void getCpuidEx(uint32_t eaxIn, uint32_t ecxIn, uint32_t data[4])
 	{
 #ifdef XBYAK_INTEL_CPU_SPECIFIC
-	#ifdef _MSC_VER
+	#ifdef _WIN32
 		__cpuidex(reinterpret_cast<int*>(data), eaxIn, ecxIn);
 	#else
 		__cpuid_count(eaxIn, ecxIn, data[0], data[1], data[2], data[3]);
@@ -288,7 +369,7 @@ public:
 	#ifdef _MSC_VER
 		return _xgetbv(0);
 	#else
-		unsigned int eax, edx;
+		uint32_t eax, edx;
 		// xgetvb is not support on gcc 4.2
 //		__asm__ volatile("xgetbv" : "=a"(eax), "=d"(edx) : "c"(0));
 		__asm__ volatile(".byte 0x0f, 0x01, 0xd0" : "=a"(eax), "=d"(edx) : "c"(0));
@@ -298,93 +379,116 @@ public:
 		return 0;
 #endif
 	}
-	typedef uint64_t Type;
 
-	static const Type NONE = 0;
-	static const Type tMMX = 1 << 0;
-	static const Type tMMX2 = 1 << 1;
-	static const Type tCMOV = 1 << 2;
-	static const Type tSSE = 1 << 3;
-	static const Type tSSE2 = 1 << 4;
-	static const Type tSSE3 = 1 << 5;
-	static const Type tSSSE3 = 1 << 6;
-	static const Type tSSE41 = 1 << 7;
-	static const Type tSSE42 = 1 << 8;
-	static const Type tPOPCNT = 1 << 9;
-	static const Type tAESNI = 1 << 10;
-	static const Type tOSXSAVE = 1 << 12;
-	static const Type tPCLMULQDQ = 1 << 13;
-	static const Type tAVX = 1 << 14;
-	static const Type tFMA = 1 << 15;
+#define XBYAK_SPLIT_ID(id) ((0 <= id && id < 64) ? (1ull << (id % 64)) : 0), (id >= 64 ? (1ull << (id % 64)) : 0)
+#if (__cplusplus >= 201103) || (defined(_MSC_VER) && (_MSC_VER >= 1700)) /* VS2012 */
+	#define XBYAK_DEFINE_TYPE(id, NAME) static const constexpr local::TypeT<XBYAK_SPLIT_ID(id)> NAME{}
+#else
+	#define XBYAK_DEFINE_TYPE(id, NAME) static const local::TypeT<XBYAK_SPLIT_ID(id)> NAME
+#endif
+	XBYAK_DEFINE_TYPE(0, tMMX);
+	XBYAK_DEFINE_TYPE(1, tMMX2);
+	XBYAK_DEFINE_TYPE(2, tCMOV);
+	XBYAK_DEFINE_TYPE(3, tSSE);
+	XBYAK_DEFINE_TYPE(4, tSSE2);
+	XBYAK_DEFINE_TYPE(5, tSSE3);
+	XBYAK_DEFINE_TYPE(6, tSSSE3);
+	XBYAK_DEFINE_TYPE(7, tSSE41);
+	XBYAK_DEFINE_TYPE(8, tSSE42);
+	XBYAK_DEFINE_TYPE(9, tPOPCNT);
+	XBYAK_DEFINE_TYPE(10, tAESNI);
+	XBYAK_DEFINE_TYPE(11, tAVX512_FP16);
+	XBYAK_DEFINE_TYPE(12, tOSXSAVE);
+	XBYAK_DEFINE_TYPE(13, tPCLMULQDQ);
+	XBYAK_DEFINE_TYPE(14, tAVX);
+	XBYAK_DEFINE_TYPE(15, tFMA);
+	XBYAK_DEFINE_TYPE(16, t3DN);
+	XBYAK_DEFINE_TYPE(17, tE3DN);
+	XBYAK_DEFINE_TYPE(18, tWAITPKG);
+	XBYAK_DEFINE_TYPE(19, tRDTSCP);
+	XBYAK_DEFINE_TYPE(20, tAVX2);
+	XBYAK_DEFINE_TYPE(21, tBMI1); // andn, bextr, blsi, blsmsk, blsr, tzcnt
+	XBYAK_DEFINE_TYPE(22, tBMI2); // bzhi, mulx, pdep, pext, rorx, sarx, shlx, shrx
+	XBYAK_DEFINE_TYPE(23, tLZCNT);
+	XBYAK_DEFINE_TYPE(24, tINTEL);
+	XBYAK_DEFINE_TYPE(25, tAMD);
+	XBYAK_DEFINE_TYPE(26, tENHANCED_REP); // enhanced rep movsb/stosb
+	XBYAK_DEFINE_TYPE(27, tRDRAND);
+	XBYAK_DEFINE_TYPE(28, tADX); // adcx, adox
+	XBYAK_DEFINE_TYPE(29, tRDSEED); // rdseed
+	XBYAK_DEFINE_TYPE(30, tSMAP); // stac
+	XBYAK_DEFINE_TYPE(31, tHLE); // xacquire, xrelease, xtest
+	XBYAK_DEFINE_TYPE(32, tRTM); // xbegin, xend, xabort
+	XBYAK_DEFINE_TYPE(33, tF16C); // vcvtph2ps, vcvtps2ph
+	XBYAK_DEFINE_TYPE(34, tMOVBE); // mobve
+	XBYAK_DEFINE_TYPE(35, tAVX512F);
+	XBYAK_DEFINE_TYPE(36, tAVX512DQ);
+	XBYAK_DEFINE_TYPE(37, tAVX512_IFMA);
+	XBYAK_DEFINE_TYPE(37, tAVX512IFMA);// = tAVX512_IFMA;
+	XBYAK_DEFINE_TYPE(38, tAVX512PF);
+	XBYAK_DEFINE_TYPE(39, tAVX512ER);
+	XBYAK_DEFINE_TYPE(40, tAVX512CD);
+	XBYAK_DEFINE_TYPE(41, tAVX512BW);
+	XBYAK_DEFINE_TYPE(42, tAVX512VL);
+	XBYAK_DEFINE_TYPE(43, tAVX512_VBMI);
+	XBYAK_DEFINE_TYPE(43, tAVX512VBMI); // = tAVX512_VBMI; // changed by Intel's manual
+	XBYAK_DEFINE_TYPE(44, tAVX512_4VNNIW);
+	XBYAK_DEFINE_TYPE(45, tAVX512_4FMAPS);
+	XBYAK_DEFINE_TYPE(46, tPREFETCHWT1);
+	XBYAK_DEFINE_TYPE(47, tPREFETCHW);
+	XBYAK_DEFINE_TYPE(48, tSHA);
+	XBYAK_DEFINE_TYPE(49, tMPX);
+	XBYAK_DEFINE_TYPE(50, tAVX512_VBMI2);
+	XBYAK_DEFINE_TYPE(51, tGFNI);
+	XBYAK_DEFINE_TYPE(52, tVAES);
+	XBYAK_DEFINE_TYPE(53, tVPCLMULQDQ);
+	XBYAK_DEFINE_TYPE(54, tAVX512_VNNI);
+	XBYAK_DEFINE_TYPE(55, tAVX512_BITALG);
+	XBYAK_DEFINE_TYPE(56, tAVX512_VPOPCNTDQ);
+	XBYAK_DEFINE_TYPE(57, tAVX512_BF16);
+	XBYAK_DEFINE_TYPE(58, tAVX512_VP2INTERSECT);
+	XBYAK_DEFINE_TYPE(59, tAMX_TILE);
+	XBYAK_DEFINE_TYPE(60, tAMX_INT8);
+	XBYAK_DEFINE_TYPE(61, tAMX_BF16);
+	XBYAK_DEFINE_TYPE(62, tAVX_VNNI);
+	XBYAK_DEFINE_TYPE(63, tCLFLUSHOPT);
+	XBYAK_DEFINE_TYPE(64, tCLDEMOTE);
+	XBYAK_DEFINE_TYPE(65, tMOVDIRI);
+	XBYAK_DEFINE_TYPE(66, tMOVDIR64B);
+	XBYAK_DEFINE_TYPE(67, tCLZERO); // AMD Zen
+	XBYAK_DEFINE_TYPE(68, tAMX_FP16);
+	XBYAK_DEFINE_TYPE(69, tAVX_VNNI_INT8);
+	XBYAK_DEFINE_TYPE(70, tAVX_NE_CONVERT);
+	XBYAK_DEFINE_TYPE(71, tAVX_IFMA);
+	XBYAK_DEFINE_TYPE(72, tRAO_INT);
+	XBYAK_DEFINE_TYPE(73, tCMPCCXADD);
+	XBYAK_DEFINE_TYPE(74, tPREFETCHITI);
+	XBYAK_DEFINE_TYPE(75, tSERIALIZE);
+	XBYAK_DEFINE_TYPE(76, tUINTR);
+	XBYAK_DEFINE_TYPE(77, tXSAVE);
+	XBYAK_DEFINE_TYPE(78, tSHA512);
+	XBYAK_DEFINE_TYPE(79, tSM3);
+	XBYAK_DEFINE_TYPE(80, tSM4);
+	XBYAK_DEFINE_TYPE(81, tAVX_VNNI_INT16);
 
-	static const Type t3DN = 1 << 16;
-	static const Type tE3DN = 1 << 17;
-	static const Type tRDTSCP = 1 << 19;
-	static const Type tAVX2 = 1 << 20;
-	static const Type tBMI1 = 1 << 21; // andn, bextr, blsi, blsmsk, blsr, tzcnt
-	static const Type tBMI2 = 1 << 22; // bzhi, mulx, pdep, pext, rorx, sarx, shlx, shrx
-	static const Type tLZCNT = 1 << 23;
-
-	static const Type tINTEL = 1 << 24;
-	static const Type tAMD = 1 << 25;
-
-	static const Type tENHANCED_REP = 1 << 26; // enhanced rep movsb/stosb
-	static const Type tRDRAND = 1 << 27;
-	static const Type tADX = 1 << 28; // adcx, adox
-	static const Type tRDSEED = 1 << 29; // rdseed
-	static const Type tSMAP = 1 << 30; // stac
-	static const Type tHLE = uint64_t(1) << 31; // xacquire, xrelease, xtest
-	static const Type tRTM = uint64_t(1) << 32; // xbegin, xend, xabort
-	static const Type tF16C = uint64_t(1) << 33; // vcvtph2ps, vcvtps2ph
-	static const Type tMOVBE = uint64_t(1) << 34; // mobve
-	static const Type tAVX512F = uint64_t(1) << 35;
-	static const Type tAVX512DQ = uint64_t(1) << 36;
-	static const Type tAVX512_IFMA = uint64_t(1) << 37;
-	static const Type tAVX512IFMA = tAVX512_IFMA;
-	static const Type tAVX512PF = uint64_t(1) << 38;
-	static const Type tAVX512ER = uint64_t(1) << 39;
-	static const Type tAVX512CD = uint64_t(1) << 40;
-	static const Type tAVX512BW = uint64_t(1) << 41;
-	static const Type tAVX512VL = uint64_t(1) << 42;
-	static const Type tAVX512_VBMI = uint64_t(1) << 43;
-	static const Type tAVX512VBMI = tAVX512_VBMI; // changed by Intel's manual
-	static const Type tAVX512_4VNNIW = uint64_t(1) << 44;
-	static const Type tAVX512_4FMAPS = uint64_t(1) << 45;
-	static const Type tPREFETCHWT1 = uint64_t(1) << 46;
-	static const Type tPREFETCHW = uint64_t(1) << 47;
-	static const Type tSHA = uint64_t(1) << 48;
-	static const Type tMPX = uint64_t(1) << 49;
-	static const Type tAVX512_VBMI2 = uint64_t(1) << 50;
-	static const Type tGFNI = uint64_t(1) << 51;
-	static const Type tVAES = uint64_t(1) << 52;
-	static const Type tVPCLMULQDQ = uint64_t(1) << 53;
-	static const Type tAVX512_VNNI = uint64_t(1) << 54;
-	static const Type tAVX512_BITALG = uint64_t(1) << 55;
-	static const Type tAVX512_VPOPCNTDQ = uint64_t(1) << 56;
-	static const Type tAVX512_BF16 = uint64_t(1) << 57;
-	static const Type tAVX512_VP2INTERSECT = uint64_t(1) << 58;
-	static const Type tAMX_TILE = uint64_t(1) << 59;
-	static const Type tAMX_INT8 = uint64_t(1) << 60;
-	static const Type tAMX_BF16 = uint64_t(1) << 61;
-	static const Type tAVX_VNNI = uint64_t(1) << 62;
-	static const Type tAVX512_FP16 = uint64_t(1) << 11;
-	// 18, 63
+#undef XBYAK_SPLIT_ID
+#undef XBYAK_DEFINE_TYPE
 
 	Cpu()
-		: type_(NONE)
+		: type_()
 		, x2APIC_supported_(false)
 		, numCores_()
 		, dataCacheSize_()
 		, coresSharignDataCache_()
 		, dataCacheLevels_(0)
 	{
-		unsigned int data[4] = {};
-		const unsigned int& EAX = data[0];
-		const unsigned int& EBX = data[1];
-		const unsigned int& ECX = data[2];
-		const unsigned int& EDX = data[3];
+		uint32_t data[4] = {};
+		const uint32_t& EAX = data[0];
+		const uint32_t& EBX = data[1];
+		const uint32_t& ECX = data[2];
+		const uint32_t& EDX = data[3];
 		getCpuid(0, data);
-		const unsigned int maxNum = EAX;
+		const uint32_t maxNum = EAX;
 		static const char intel[] = "ntel";
 		static const char amd[] = "cAMD";
 		if (ECX == get32bitAsBE(amd)) {
@@ -407,7 +511,8 @@ public:
 
 		// Extended flags information
 		getCpuid(0x80000000, data);
-		if (EAX >= 0x80000001) {
+		const uint32_t maxExtendedNum = EAX;
+		if (maxExtendedNum >= 0x80000001) {
 			getCpuid(0x80000001, data);
 
 			if (EDX & (1U << 31)) type_ |= t3DN;
@@ -419,15 +524,21 @@ public:
 			if (ECX & (1U << 8)) type_ |= tPREFETCHW;
 		}
 
+		if (maxExtendedNum >= 0x80000008) {
+			getCpuid(0x80000008, data);
+			if (EBX & (1U << 0)) type_ |= tCLZERO;
+		}
+
 		getCpuid(1, data);
 		if (ECX & (1U << 0)) type_ |= tSSE3;
+		if (ECX & (1U << 1)) type_ |= tPCLMULQDQ;
 		if (ECX & (1U << 9)) type_ |= tSSSE3;
 		if (ECX & (1U << 19)) type_ |= tSSE41;
 		if (ECX & (1U << 20)) type_ |= tSSE42;
 		if (ECX & (1U << 22)) type_ |= tMOVBE;
 		if (ECX & (1U << 23)) type_ |= tPOPCNT;
 		if (ECX & (1U << 25)) type_ |= tAESNI;
-		if (ECX & (1U << 1)) type_ |= tPCLMULQDQ;
+		if (ECX & (1U << 26)) type_ |= tXSAVE;
 		if (ECX & (1U << 27)) type_ |= tOSXSAVE;
 		if (ECX & (1U << 30)) type_ |= tRDRAND;
 		if (ECX & (1U << 29)) type_ |= tF16C;
@@ -460,9 +571,6 @@ public:
 						if (EBX & (1U << 31)) type_ |= tAVX512VL;
 						if (ECX & (1U << 1)) type_ |= tAVX512_VBMI;
 						if (ECX & (1U << 6)) type_ |= tAVX512_VBMI2;
-						if (ECX & (1U << 8)) type_ |= tGFNI;
-						if (ECX & (1U << 9)) type_ |= tVAES;
-						if (ECX & (1U << 10)) type_ |= tVPCLMULQDQ;
 						if (ECX & (1U << 11)) type_ |= tAVX512_VNNI;
 						if (ECX & (1U << 12)) type_ |= tAVX512_BITALG;
 						if (ECX & (1U << 14)) type_ |= tAVX512_VPOPCNTDQ;
@@ -484,20 +592,41 @@ public:
 			if (EBX & (1U << 18)) type_ |= tRDSEED;
 			if (EBX & (1U << 19)) type_ |= tADX;
 			if (EBX & (1U << 20)) type_ |= tSMAP;
+			if (EBX & (1U << 23)) type_ |= tCLFLUSHOPT;
 			if (EBX & (1U << 4)) type_ |= tHLE;
 			if (EBX & (1U << 11)) type_ |= tRTM;
 			if (EBX & (1U << 14)) type_ |= tMPX;
 			if (EBX & (1U << 29)) type_ |= tSHA;
 			if (ECX & (1U << 0)) type_ |= tPREFETCHWT1;
+			if (ECX & (1U << 5)) type_ |= tWAITPKG;
+			if (ECX & (1U << 8)) type_ |= tGFNI;
+			if (ECX & (1U << 9)) type_ |= tVAES;
+			if (ECX & (1U << 10)) type_ |= tVPCLMULQDQ;
+			if (ECX & (1U << 25)) type_ |= tCLDEMOTE;
+			if (ECX & (1U << 27)) type_ |= tMOVDIRI;
+			if (ECX & (1U << 28)) type_ |= tMOVDIR64B;
+			if (EDX & (1U << 5)) type_ |= tUINTR;
+			if (EDX & (1U << 14)) type_ |= tSERIALIZE;
+			if (EDX & (1U << 22)) type_ |= tAMX_BF16;
 			if (EDX & (1U << 24)) type_ |= tAMX_TILE;
 			if (EDX & (1U << 25)) type_ |= tAMX_INT8;
-			if (EDX & (1U << 22)) type_ |= tAMX_BF16;
 			if (maxNumSubLeaves >= 1) {
 				getCpuidEx(7, 1, data);
+				if (EAX & (1U << 0)) type_ |= tSHA512;
+				if (EAX & (1U << 1)) type_ |= tSM3;
+				if (EAX & (1U << 2)) type_ |= tSM4;
+				if (EAX & (1U << 3)) type_ |= tRAO_INT;
 				if (EAX & (1U << 4)) type_ |= tAVX_VNNI;
 				if (type_ & tAVX512F) {
 					if (EAX & (1U << 5)) type_ |= tAVX512_BF16;
 				}
+				if (EAX & (1U << 7)) type_ |= tCMPCCXADD;
+				if (EAX & (1U << 21)) type_ |= tAMX_FP16;
+				if (EAX & (1U << 23)) type_ |= tAVX_IFMA;
+				if (EDX & (1U << 4)) type_ |= tAVX_VNNI_INT8;
+				if (EDX & (1U << 5)) type_ |= tAVX_NE_CONVERT;
+				if (EDX & (1U << 10)) type_ |= tAVX_VNNI_INT16;
+				if (EDX & (1U << 14)) type_ |= tPREFETCHITI;
 			}
 		}
 		setFamily();
@@ -512,9 +641,9 @@ public:
 		printf("display:family=%X, model=%X\n", displayFamily, displayModel);
 #endif
 	}
-	bool has(Type type) const
+	bool has(const Type& type) const
 	{
-		return (type & type_) != 0;
+		return (type & type_) == type;
 	}
 };
 
@@ -527,7 +656,7 @@ public:
 	#ifdef _MSC_VER
 		return __rdtsc();
 	#else
-		unsigned int eax, edx;
+		uint32_t eax, edx;
 		__asm__ volatile("rdtsc" : "=a"(eax), "=d"(edx));
 		return ((uint64_t)edx << 32) | eax;
 	#endif
@@ -564,7 +693,7 @@ const int UseRDX = 1 << 7;
 
 class Pack {
 	static const size_t maxTblNum = 15;
-	const Xbyak::Reg64 *tbl_[maxTblNum];
+	Xbyak::Reg64 tbl_[maxTblNum];
 	size_t n_;
 public:
 	Pack() : tbl_(), n_(0) {}
@@ -581,32 +710,36 @@ public:
 		return *this;
 	}
 	Pack(const Xbyak::Reg64& t0)
-	{ n_ = 1; tbl_[0] = &t0; }
+	{ n_ = 1; tbl_[0] = t0; }
 	Pack(const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 2; tbl_[0] = &t0; tbl_[1] = &t1; }
+	{ n_ = 2; tbl_[0] = t0; tbl_[1] = t1; }
 	Pack(const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 3; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; }
+	{ n_ = 3; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; }
 	Pack(const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 4; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; }
+	{ n_ = 4; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; }
 	Pack(const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 5; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; tbl_[4] = &t4; }
+	{ n_ = 5; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; }
 	Pack(const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 6; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; tbl_[4] = &t4; tbl_[5] = &t5; }
+	{ n_ = 6; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; }
 	Pack(const Xbyak::Reg64& t6, const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 7; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; tbl_[4] = &t4; tbl_[5] = &t5; tbl_[6] = &t6; }
+	{ n_ = 7; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; tbl_[6] = t6; }
 	Pack(const Xbyak::Reg64& t7, const Xbyak::Reg64& t6, const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 8; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; tbl_[4] = &t4; tbl_[5] = &t5; tbl_[6] = &t6; tbl_[7] = &t7; }
+	{ n_ = 8; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; tbl_[6] = t6; tbl_[7] = t7; }
 	Pack(const Xbyak::Reg64& t8, const Xbyak::Reg64& t7, const Xbyak::Reg64& t6, const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 9; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; tbl_[4] = &t4; tbl_[5] = &t5; tbl_[6] = &t6; tbl_[7] = &t7; tbl_[8] = &t8; }
+	{ n_ = 9; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; tbl_[6] = t6; tbl_[7] = t7; tbl_[8] = t8; }
 	Pack(const Xbyak::Reg64& t9, const Xbyak::Reg64& t8, const Xbyak::Reg64& t7, const Xbyak::Reg64& t6, const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
-	{ n_ = 10; tbl_[0] = &t0; tbl_[1] = &t1; tbl_[2] = &t2; tbl_[3] = &t3; tbl_[4] = &t4; tbl_[5] = &t5; tbl_[6] = &t6; tbl_[7] = &t7; tbl_[8] = &t8; tbl_[9] = &t9; }
+	{ n_ = 10; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; tbl_[6] = t6; tbl_[7] = t7; tbl_[8] = t8; tbl_[9] = t9; }
+	Pack(const Xbyak::Reg64& ta, const Xbyak::Reg64& t9, const Xbyak::Reg64& t8, const Xbyak::Reg64& t7, const Xbyak::Reg64& t6, const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
+	{ n_ = 11; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; tbl_[6] = t6; tbl_[7] = t7; tbl_[8] = t8; tbl_[9] = t9; tbl_[10] = ta; }
+	Pack(const Xbyak::Reg64& tb, const Xbyak::Reg64& ta, const Xbyak::Reg64& t9, const Xbyak::Reg64& t8, const Xbyak::Reg64& t7, const Xbyak::Reg64& t6, const Xbyak::Reg64& t5, const Xbyak::Reg64& t4, const Xbyak::Reg64& t3, const Xbyak::Reg64& t2, const Xbyak::Reg64& t1, const Xbyak::Reg64& t0)
+	{ n_ = 12; tbl_[0] = t0; tbl_[1] = t1; tbl_[2] = t2; tbl_[3] = t3; tbl_[4] = t4; tbl_[5] = t5; tbl_[6] = t6; tbl_[7] = t7; tbl_[8] = t8; tbl_[9] = t9; tbl_[10] = ta; tbl_[11] = tb; }
 	Pack& append(const Xbyak::Reg64& t)
 	{
 		if (n_ == maxTblNum) {
 			fprintf(stderr, "ERR Pack::can't append\n");
 			XBYAK_THROW_RET(ERR_BAD_PARAMETER, *this)
 		}
-		tbl_[n_++] = &t;
+		tbl_[n_++] = t;
 		return *this;
 	}
 	void init(const Xbyak::Reg64 *tbl, size_t n)
@@ -617,7 +750,7 @@ public:
 		}
 		n_ = n;
 		for (size_t i = 0; i < n; i++) {
-			tbl_[i] = &tbl[i];
+			tbl_[i] = tbl[i];
 		}
 	}
 	const Xbyak::Reg64& operator[](size_t n) const
@@ -626,7 +759,7 @@ public:
 			fprintf(stderr, "ERR Pack bad n=%d(%d)\n", (int)n, (int)n_);
 			XBYAK_THROW_RET(ERR_BAD_PARAMETER, rax)
 		}
-		return *tbl_[n];
+		return tbl_[n];
 	}
 	size_t size() const { return n_; }
 	/*
@@ -649,7 +782,7 @@ public:
 	void put() const
 	{
 		for (size_t i = 0; i < n_; i++) {
-			printf("%s ", tbl_[i]->toString());
+			printf("%s ", tbl_[i].toString());
 		}
 		printf("\n");
 	}
@@ -716,7 +849,7 @@ public:
 		const int allRegNum = pNum + tNum_ + (useRcx_ ? 1 : 0) + (useRdx_ ? 1 : 0);
 		if (tNum_ < 0 || allRegNum > maxRegNum) XBYAK_THROW(ERR_BAD_TNUM)
 		const Reg64& _rsp = code->rsp;
-		saveNum_ = (std::max)(0, allRegNum - noSaveNum);
+		saveNum_ = local::max_(0, allRegNum - noSaveNum);
 		const int *tbl = getOrderTbl() + noSaveNum;
 		for (int i = 0; i < saveNum_; i++) {
 			code->push(Reg64(tbl[i]));