From 2f9dea5cc355c266ad46d2f6397b141b99f78480 Mon Sep 17 00:00:00 2001
From: MerryMage <MerryMage@users.noreply.github.com>
Date: Sat, 27 Mar 2021 21:08:22 +0000
Subject: [PATCH] Squashed 'externals/xbyak/' changes from 0140eeff1..590c10e37

590c10e37 fix typo
396715a89 use github action
69a25fd92 change build status to travis-ci.com
a6a9dac91 meanings of the name
8d1e41b65 test_util.cpp supports OpenBSD
77ffe7173 Merge pull request #115 from Ryan-rsm-McKenzie/master
a1da3403a fix build interface include directory
e0136d4ef fix add_library call with INTERFACE
3071eee0c support slightly more modern cmake
e626d6209 v5.991
a49c4bc11 disable XBYAK_CONSTEXPR for g++-5 -std=c++-14
70777a699 Merge branch 'atafra-old_mac_fix' into dev
6b81678d0 fixed compile error on some older macOS versions
2c3b43f15 refactor util
91784e2b8 test_util checks AMX and AVX_VNNI
70b70c557 update to v5.99
284cc5bed refactor
6b3eb9c1e default encoding is always evex
f85b1100b refactor vnni
276d09bae Merge branch 'akharito-akharito/adl_support' into dev
50df86ce3 v5.98
97ce92d58 Merge branch 'akharito/adl_support' of https://github.com/akharito/xbyak into akharito-akharito/adl_support
1f119a04a support [scale * reg]
9ee1bef9a cpuid - check that GRT CPUID leaf 7 subleaf 0 should return EAX=1
be93adb2c add AVX VNNI instruction support
0c277240a add ADL CPUID
a9a5cc2e2 Add option to choose VEX or EVEX encoding
29bfd25ba fix indent
a0c49fa2e Merge branch 'atafra-mac_avx512_fix' into dev
ea388b3c6 fixed incorrect detection of AVX-512 on macOS
ed1b8186f Merge branch 'FEX-Emu-extended_features' into dev
3dacddfec Merge branch 'extended_features' of https://github.com/FEX-Emu/xbyak into FEX-Emu-extended_features
898f78ca3 Merge branch 'kariya-mitsuru-use-sh'
0b7f1411c Merge branch 'use-sh' of https://github.com/kariya-mitsuru/xbyak into kariya-mitsuru-use-sh
99e2b13b2 Fixes extended feature support checking
b0a43c7e5 Use sh instead of tcsh for test scripts
87e8f41ae remove warning of _MSC_VER

git-subtree-dir: externals/xbyak
git-subtree-split: 590c10e3746978dbfcf102d6da933ac2659e4544
---
 .github/workflows/main.yml | 11 ++++++
 .gitignore                 |  1 +
 .travis.yml                | 12 -------
 CMakeLists.txt             | 46 ++++++++++++++++++++++--
 cmake/config.cmake.in      |  1 +
 gen/gen_avx512.cpp         |  6 ----
 gen/gen_code.cpp           | 18 ++++++++++
 readme.md                  | 19 +++++++---
 readme.txt                 |  8 ++++-
 sample/test_util.cpp       | 13 +++++--
 test/make_nm.cpp           |  1 +
 test/misc.cpp              | 28 +++++++++++++++
 test/test_address.sh       |  2 +-
 test/test_avx.sh           | 51 +++++++++++++++------------
 test/test_avx512.sh        | 31 +++++++++--------
 test/test_nm.sh            | 71 +++++++++++++++++++++-----------------
 xbyak/xbyak.h              | 46 ++++++++++++++++++------
 xbyak/xbyak_mnemonic.h     | 10 +++---
 xbyak/xbyak_util.h         | 46 ++++++++++++++++++------
 19 files changed, 297 insertions(+), 124 deletions(-)
 create mode 100644 .github/workflows/main.yml
 create mode 100644 .gitignore
 delete mode 100644 .travis.yml
 create mode 100644 cmake/config.cmake.in

diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
new file mode 100644
index 00000000..3fad1422
--- /dev/null
+++ b/.github/workflows/main.yml
@@ -0,0 +1,11 @@
+name: test
+on: [push]
+
+jobs:
+  build:
+    name: test
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2
+    - run: sudo apt install nasm yasm g++-multilib tcsh
+    - run: make test
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 00000000..24b0b1de
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1 @@
+/build* # cmake
diff --git a/.travis.yml b/.travis.yml
deleted file mode 100644
index 4de8772d..00000000
--- a/.travis.yml
+++ /dev/null
@@ -1,12 +0,0 @@
-sudo: true
-dist: bionic
-language: cpp
-compiler:
-  - gcc
-  - clang
-addons:
-  apt:
-    packages:
-      - nasm yasm g++-multilib tcsh
-script:
-  - make test
diff --git a/CMakeLists.txt b/CMakeLists.txt
index be131b14..f2c54f14 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,6 +1,46 @@
-cmake_minimum_required(VERSION 2.6)
-project(xbyak)
+cmake_minimum_required(VERSION 2.6...3.0.2)
+
+project(xbyak CXX)
 
 file(GLOB headers xbyak/*.h)
-install(FILES ${headers} DESTINATION include/xbyak)
 
+if (DEFINED CMAKE_VERSION AND CMAKE_VERSION VERSION_GREATER_EQUAL 3.0.2)
+	include(GNUInstallDirs)
+	add_library(${PROJECT_NAME} INTERFACE)
+	add_library(${PROJECT_NAME}::${PROJECT_NAME} ALIAS ${PROJECT_NAME})
+	
+	target_include_directories(
+		${PROJECT_NAME} INTERFACE
+		"$<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}>"
+		"$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>"
+	)
+
+	install(
+		TARGETS ${PROJECT_NAME}
+		EXPORT ${PROJECT_NAME}-targets
+	)
+
+	configure_file(
+		cmake/config.cmake.in
+		${PROJECT_NAME}Config.cmake
+		@ONLY
+	)
+
+	install(
+		FILES ${CMAKE_CURRENT_BINARY_DIR}/${PROJECT_NAME}Config.cmake
+		DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/${PROJECT_NAME}
+	)
+
+	install(
+		EXPORT ${PROJECT_NAME}-targets
+		NAMESPACE ${PROJECT_NAME}::
+		DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/${PROJECT_NAME}
+	)
+elseif(NOT DEFINED CMAKE_INSTALL_INCLUDEDIR)
+	set(CMAKE_INSTALL_INCLUDEDIR "include")
+endif()
+
+install(
+	FILES ${headers}
+	DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/xbyak
+)
diff --git a/cmake/config.cmake.in b/cmake/config.cmake.in
new file mode 100644
index 00000000..f40ebfa8
--- /dev/null
+++ b/cmake/config.cmake.in
@@ -0,0 +1 @@
+include("${CMAKE_CURRENT_LIST_DIR}/@PROJECT_NAME@-targets.cmake")
diff --git a/gen/gen_avx512.cpp b/gen/gen_avx512.cpp
index 250c8d4b..b2b88c3f 100644
--- a/gen/gen_avx512.cpp
+++ b/gen/gen_avx512.cpp
@@ -363,12 +363,6 @@ void putX_X_XM_IMM()
 		{ 0x73, "vpshrdvd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
 		{ 0x73, "vpshrdvq", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW1 | T_SAE_Z | T_B64, false },
 
-		{ 0x50, "vpdpbusd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
-		{ 0x51, "vpdpbusds", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
-
-		{ 0x52, "vpdpwssd", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
-		{ 0x53, "vpdpwssds", T_66 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
-
 		{ 0x72, "vcvtne2ps2bf16", T_F2 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
 		{ 0x52, "vdpbf16ps", T_F3 | T_0F38 | T_YMM | T_MUST_EVEX | T_EW0 | T_SAE_Z | T_B32, false },
 	};
diff --git a/gen/gen_code.cpp b/gen/gen_code.cpp
index ba7dbf61..90f29672 100644
--- a/gen/gen_code.cpp
+++ b/gen/gen_code.cpp
@@ -1729,6 +1729,24 @@ void put()
 			printf("void %s(const Xmm& x1, const Address& addr, const Xmm& x2) { opGather(x1, addr, x2, T_0F38 | T_66 | T_YMM | T_VSIB | T_W%d, 0x%x, %d); }\n", p.name, p.w, p.code, p.mode);
 		}
 	}
+	// vnni
+	{
+		const struct Tbl {
+			uint8_t code;
+			const char *name;
+			int type;
+		} tbl[] = {
+			{ 0x50, "vpdpbusd", T_66 | T_0F38 | T_YMM | T_EW0 | T_SAE_Z | T_B32},
+			{ 0x51, "vpdpbusds", T_66 | T_0F38 | T_YMM | T_EW0 | T_SAE_Z | T_B32},
+			{ 0x52, "vpdpwssd", T_66 | T_0F38 | T_YMM | T_EW0 | T_SAE_Z | T_B32},
+			{ 0x53, "vpdpwssds", T_66 | T_0F38 | T_YMM | T_EW0 | T_SAE_Z | T_B32},
+		};
+		for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
+			const Tbl *p = &tbl[i];
+			std::string type = type2String(p->type);
+			printf("void %s(const Xmm& x1, const Xmm& x2, const Operand& op, PreferredEncoding encoding = DefaultEncoding) { opVnni(x1, x2, op, %s, 0x%02X, encoding); }\n", p->name, type.c_str(), p->code);
+		}
+	}
 }
 
 void put32()
diff --git a/readme.md b/readme.md
index 6caaa51a..d9aaa6ae 100644
--- a/readme.md
+++ b/readme.md
@@ -1,11 +1,14 @@
-[![Build Status](https://travis-ci.org/herumi/xbyak.png)](https://travis-ci.org/herumi/xbyak)
+[![Build Status](https://github.com/herumi/xbyak/actions/workflows/main.yml/badge.svg)](https://github.com/herumi/xbyak/actions/workflows/main.yml)
 
-# Xbyak 5.97 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
+# Xbyak 5.991 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
 
 ## Abstract
 
 Xbyak is a C++ header library that enables dynamically to assemble x86(IA32), x64(AMD64, x86-64) mnemonic.
 
+The pronunciation of Xbyak is `kəi-bja-k`.
+It is named from a Japanese word [開闢](https://translate.google.com/?hl=ja&sl=ja&tl=en&text=%E9%96%8B%E9%97%A2&op=translate), which means the beginning of the world.
+
 ## Feature
 * header file only
 * Intel/MASM like syntax
@@ -16,6 +19,7 @@ Use `and_()`, `or_()`, ... instead of `and()`, `or()`.
 If you want to use them, then specify `-fno-operator-names` option to gcc/clang.
 
 ### News
+- vnni instructions such as vpdpbusd supports vex encoding.
 - (break backward compatibility) `push(byte, imm)` (resp. `push(word, imm)`) forces to cast `imm` to 8(resp. 16) bit.
 - (Windows) `#include <winsock2.h>` has been removed from xbyak.h, so add it explicitly if you need it.
 - support exception-less mode see. [Exception-less mode](#exception-less-mode)
@@ -154,6 +158,10 @@ vcvtpd2dq xmm19, [eax+32]{1to4}         --> vcvtpd2dq(xmm19, yword_b [eax+32]);
 vfpclassps k5{k3}, zword [rax+64], 5    --> vfpclassps(k5|k3, zword [rax+64], 5); // specify m512
 vfpclasspd k5{k3}, [rax+64]{1to2}, 5    --> vfpclasspd(k5|k3, xword_b [rax+64], 5); // broadcast 64-bit to 128-bit
 vfpclassps k5{k3}, [rax+64]{1to4}, 5    --> vfpclassps(k5|k3, yword_b [rax+64], 5); // broadcast 64-bit to 256-bit
+
+vpdpbusd(xm0, xm1, xm2); // default encoding is EVEX
+vpdpbusd(xm0, xm1, xm2, EvexEncoding); // same as the above
+vpdpbusd(xm0, xm1, xm2, VexEncoding); // VEX encoding
 ```
 ### Remark
 * `k1`, ..., `k7` are opmask registers.
@@ -339,9 +347,9 @@ public:
 
 ## User allocated memory
 
-You can make jit code on prepaired memory.
+You can make jit code on prepared memory.
 
-Call `setProtectModeRE` yourself to change memory mode if using the prepaired memory.
+Call `setProtectModeRE` yourself to change memory mode if using the prepared memory.
 
 ```
 uint8_t alignas(4096) buf[8192]; // C++11 or later
@@ -438,6 +446,9 @@ modified new BSD License
 http://opensource.org/licenses/BSD-3-Clause
 
 ## History
+* 2020/Nov/16 ver 5.991 disable constexpr for gcc-5 with -std=c++-14
+* 2020/Oct/19 ver 5.99 support VNNI instructions(Thanks to akharito)
+* 2020/Oct/17 ver 5.98 support the form of [scale * reg]
 * 2020/Sep/08 ver 5.97 replace uint32 with uint32_t etc.
 * 2020/Aug/28 ver 5.95 some constructors of register classes support constexpr if C++14 or later
 * 2020/Aug/04 ver 5.941 `CodeGenerator::reset()` calls `ClearError()`.
diff --git a/readme.txt b/readme.txt
index dcbff9e7..d49bcccb 100644
--- a/readme.txt
+++ b/readme.txt
@@ -1,5 +1,5 @@
 
-    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.97
+    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 5.991
 
 -----------------------------------------------------------------------------
 ◎概要
@@ -163,6 +163,9 @@ vfpclassps k5{k3}, zword [rax+64], 5    --> vfpclassps(k5|k3, zword [rax+64], 5)
 vfpclasspd k5{k3}, [rax+64]{1to2}, 5    --> vfpclasspd(k5|k3, xword_b [rax+64], 5); // broadcast 64-bit to 128-bit
 vfpclassps k5{k3}, [rax+64]{1to4}, 5    --> vfpclassps(k5|k3, xword_b [rax+64], 5); // broadcast 64-bit to 256-bit
 
+vpdpbusd(xm0, xm1, xm2); // default encoding is EVEX
+vpdpbusd(xm0, xm1, xm2, EvexEncoding); // same as the above
+vpdpbusd(xm0, xm1, xm2, VexEncoding); // VEX encoding
 
 注意
 * k1, ..., k7 は新しいopmaskレジスタです。
@@ -379,6 +382,9 @@ sample/{echo,hello}.bfは http://www.kmonos.net/alang/etc/brainfuck.php から
 -----------------------------------------------------------------------------
 ◎履歴
 
+2020/11/16 ver 5.991 g++-5のC++14でconstexpr機能の抑制
+2020/10/19 ver 5.99 VNNI命令サポート(Thanks to akharito)
+2020/10/17 ver 5.98 [scale * reg]のサポート
 2020/09/08 ver 5.97 uint32などをuint32_tに置換
 2020/08/28 ver 5.95 レジスタクラスのコンストラクタがconstexprに対応(C++14以降)
 2020/08/04 ver 5.941 `CodeGenerator::reset()`が`ClearError()`を呼ぶように変更
diff --git a/sample/test_util.cpp b/sample/test_util.cpp
index afb6e5a4..18dcce50 100644
--- a/sample/test_util.cpp
+++ b/sample/test_util.cpp
@@ -1,12 +1,13 @@
 #include <stdio.h>
-#define XBYAK_NO_OP_NAMES
 #include "xbyak/xbyak_util.h"
 
 #define NUM_OF_ARRAY(x) (sizeof(x) / sizeof(x[0]))
 
 struct PopCountTest : public Xbyak::CodeGenerator {
 	PopCountTest(int n)
+		: Xbyak::CodeGenerator(4096, Xbyak::DontSetProtectRWE)
 	{
+ret();
 		mov(eax, n);
 		popcnt(eax, eax);
 		ret();
@@ -80,6 +81,10 @@ void putCPUinfo()
 		{ Cpu::tAVX512_VPOPCNTDQ, "avx512_vpopcntdq" },
 		{ Cpu::tAVX512_BF16, "avx512_bf16" },
 		{ Cpu::tAVX512_VP2INTERSECT, "avx512_vp2intersect" },
+		{ Cpu::tAMX_TILE, "amx(tile)" },
+		{ Cpu::tAMX_INT8, "amx(int8)" },
+		{ Cpu::tAMX_BF16, "amx(bf16)" },
+		{ Cpu::tAVX_VNNI, "avx_vnni" },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		if (cpu.has(tbl[i].type)) printf(" %s", tbl[i].str);
@@ -88,12 +93,16 @@ void putCPUinfo()
 	if (cpu.has(Cpu::tPOPCNT)) {
 		const int n = 0x12345678; // bitcount = 13
 		const int ok = 13;
-		int r = PopCountTest(n).getCode<int (*)()>()();
+		PopCountTest code(n);
+		code.setProtectModeRE();
+		int (*f)() = code.getCode<int (*)()>();
+		int r = f();
 		if (r == ok) {
 			puts("popcnt ok");
 		} else {
 			printf("popcnt ng %d %d\n", r, ok);
 		}
+		code.setProtectModeRW();
 	}
 	/*
 		                displayFamily displayModel
diff --git a/test/make_nm.cpp b/test/make_nm.cpp
index 47eb0237..494d8ac5 100644
--- a/test/make_nm.cpp
+++ b/test/make_nm.cpp
@@ -643,6 +643,7 @@ class Test {
 		puts(isXbyak_ ? "out_(dx, al); dump();" : "out dx, al");
 		puts(isXbyak_ ? "out_(dx, ax); dump();" : "out dx, ax");
 		puts(isXbyak_ ? "out_(dx, eax); dump();" : "out dx, eax");
+		puts(isXbyak_ ? "lea(eax, ptr[edi + 4 * eax]); dump();" : "lea eax, [edi + 4 * eax]");
 	}
 	void putJmp() const
 	{
diff --git a/test/misc.cpp b/test/misc.cpp
index 2a55ec2e..2f0413ee 100644
--- a/test/misc.cpp
+++ b/test/misc.cpp
@@ -815,4 +815,32 @@ CYBOZU_TEST_AUTO(tileloadd)
 	CYBOZU_TEST_EXCEPTION(c.notSupported(), std::exception);
 	CYBOZU_TEST_EXCEPTION(c.notSupported2(), std::exception);
 }
+
+CYBOZU_TEST_AUTO(vnni)
+{
+	struct Code : Xbyak::CodeGenerator {
+		Code()
+		{
+			// default encoding is EVEX
+			vpdpbusd(xm0, xm1, xm2);
+			vpdpbusd(xm0, xm1, xm2, EvexEncoding); // EVEX
+			vpdpbusd(xm0, xm1, xm2, VexEncoding); // VEX
+		}
+		void badVex()
+		{
+			vpdpbusd(xm0, xm1, xm31, VexEncoding);
+		}
+	} c;
+	const uint8_t tbl[] = {
+		0x62, 0xF2, 0x75, 0x08, 0x50, 0xC2,
+		0x62, 0xF2, 0x75, 0x08, 0x50, 0xC2,
+		0xC4, 0xE2, 0x71, 0x50, 0xC2,
+	};
+	const size_t n = sizeof(tbl) / sizeof(tbl[0]);
+	CYBOZU_TEST_EQUAL(c.getSize(), n);
+	CYBOZU_TEST_EQUAL_ARRAY(c.getCode(), tbl, n);
+
+	CYBOZU_TEST_EXCEPTION(c.badVex(), std::exception);
+}
+
 #endif
diff --git a/test/test_address.sh b/test/test_address.sh
index d63a4ef5..d283a5f3 100755
--- a/test/test_address.sh
+++ b/test/test_address.sh
@@ -12,7 +12,7 @@ g++ $CFLAGS address.cpp -o address
 ./address $1 > a.asm
 echo "asm"
 $EXE -f$OPT3 a.asm -l a.lst
-awk '{if (index($3, "-")) { conti=substr($3, 0, length($3) - 1) } else { conti = conti $3; print conti; conti = "" }} ' < a.lst | $FILTER > ok.lst
+awk '{printf "%s", sub(/-$/, "", $3) ? $3 : $3 ORS}' a.lst | $FILTER > ok.lst
 
 echo "xbyak"
 ./address $1 jit > nm.cpp
diff --git a/test/test_avx.sh b/test/test_avx.sh
index 236f7aa5..34dc1e55 100755
--- a/test/test_avx.sh
+++ b/test/test_avx.sh
@@ -1,39 +1,44 @@
-#!/bin/tcsh
+#!/bin/sh
 
-set FILTER="grep -v warning"
+FILTER="grep -v warning"
 
-if ($1 == "Y") then
+case $1 in
+Y)
 	echo "yasm(32bit)"
-	set EXE=yasm
-	set OPT2="-DUSE_YASM -DXBYAK32"
-	set OPT3=win32
-else if ($1 == "64") then
+	EXE=yasm
+	OPT2="-DUSE_YASM -DXBYAK32"
+	OPT3=win32
+	;;
+64)
 	echo "nasm(64bit)"
-	set EXE=nasm
-	set OPT2=-DXBYAK64
-	set OPT3=win64
-	set FILTER=./normalize_prefix
-else if ($1 == "Y64") then
+	EXE=nasm
+	OPT2=-DXBYAK64
+	OPT3=win64
+	FILTER=./normalize_prefix
+	;;
+Y64)
 	echo "yasm(64bit)"
-	set EXE=yasm
-	set OPT2="-DUSE_YASM -DXBYAK64"
-	set OPT3=win64
-	set FILTER=./normalize_prefix
-else
+	EXE=yasm
+	OPT2="-DUSE_YASM -DXBYAK64"
+	OPT3=win64
+	FILTER=./normalize_prefix
+	;;
+*)
 	echo "nasm(32bit)"
-	set EXE=nasm
-	set OPT2=-DXBYAK32
-	set OPT3=win32
-endif
+	EXE=nasm
+	OPT2=-DXBYAK32
+	OPT3=win32
+	;;
+esac
 
-set CFLAGS="-Wall -fno-operator-names -I../ $OPT2 -DUSE_AVX"
+CFLAGS="-Wall -fno-operator-names -I../ $OPT2 -DUSE_AVX"
 echo "compile make_nm.cpp"
 g++ $CFLAGS make_nm.cpp -o make_nm
 
 ./make_nm > a.asm
 echo "asm"
 $EXE -f$OPT3 a.asm -l a.lst
-awk '{if (index($3, "-")) { conti=substr($3, 0, length($3) - 1) } else { conti = conti $3; print conti; conti = "" }} ' < a.lst | $FILTER | grep -v "1+1" > ok.lst
+awk '$3 != "1+1" {printf "%s", sub(/-$/, "", $3) ? $3 : $3 ORS}' a.lst | $FILTER > ok.lst
 
 echo "xbyak"
 ./make_nm jit > nm.cpp
diff --git a/test/test_avx512.sh b/test/test_avx512.sh
index cce5de0c..17edfeec 100755
--- a/test/test_avx512.sh
+++ b/test/test_avx512.sh
@@ -1,28 +1,31 @@
-#!/bin/tcsh
+#!/bin/sh
 
-set FILTER="grep -v warning"
+FILTER="grep -v warning"
 
-if ($1 == "64") then
+case $1 in
+64)
 	echo "nasm(64bit)"
-	set EXE=nasm
-	set OPT2=-DXBYAK64
-	set OPT3=win64
-	set FILTER=./normalize_prefix
-else
+	EXE=nasm
+	OPT2=-DXBYAK64
+	OPT3=win64
+	FILTER=./normalize_prefix
+	;;
+*)
 	echo "nasm(32bit)"
-	set EXE=nasm
-	set OPT2=-DXBYAK32
-	set OPT3=win32
-endif
+	EXE=nasm
+	OPT2=-DXBYAK32
+	OPT3=win32
+	;;
+esac
 
-set CFLAGS="-Wall -fno-operator-names -I../ $OPT2 -DUSE_AVX512"
+CFLAGS="-Wall -fno-operator-names -I../ $OPT2 -DUSE_AVX512"
 echo "compile make_512.cpp"
 g++ $CFLAGS make_512.cpp -o make_512
 
 ./make_512 > a.asm
 echo "asm"
 $EXE -f$OPT3 a.asm -l a.lst
-awk '{if (index($3, "-")) { conti=substr($3, 0, length($3) - 1) } else { conti = conti $3; print conti; conti = "" }} ' < a.lst | $FILTER > ok.lst
+awk '{printf "%s", sub(/-$/, "", $3) ? $3 : $3 ORS}' a.lst | $FILTER > ok.lst
 
 echo "xbyak"
 ./make_512 jit > nm.cpp
diff --git a/test/test_nm.sh b/test/test_nm.sh
index 3e328012..afa2b1eb 100755
--- a/test/test_nm.sh
+++ b/test/test_nm.sh
@@ -1,50 +1,57 @@
-#!/bin/tcsh
+#!/bin/sh
 
-set FILTER=cat
+FILTER=cat
 
-if ($1 == "Y") then
+case $1 in
+Y)
 	echo "yasm(32bit)"
-	set EXE=yasm
-	set OPT2="-DUSE_YASM -DXBYAK32"
-	set OPT3=win32
-else if ($1 == "64") then
+	EXE=yasm
+	OPT2="-DUSE_YASM -DXBYAK32"
+	OPT3=win32
+	;;
+64)
 	echo "nasm(64bit)"
-	set EXE=nasm
-	set OPT2=-DXBYAK64
-	set OPT3=win64
-	set FILTER=./normalize_prefix
-else if ($1 == "Y64") then
+	EXE=nasm
+	OPT2=-DXBYAK64
+	OPT3=win64
+	FILTER=./normalize_prefix
+	;;
+Y64)
 	echo "yasm(64bit)"
-	set EXE=yasm
-	set OPT2="-DUSE_YASM -DXBYAK64"
-	set OPT3=win64
-	set FILTER=./normalize_prefix
-else if ($1 == "avx512") then
+	EXE=yasm
+	OPT2="-DUSE_YASM -DXBYAK64"
+	OPT3=win64
+	FILTER=./normalize_prefix
+	;;
+avx512)
 	echo "nasm(64bit) + avx512"
-	set EXE=nasm
-	set OPT2="-DXBYAK64 -DUSE_AVX512"
-	set OPT3=win64
-	set FILTER=./normalize_prefix
-else if ($1 == "noexcept") then
+	EXE=nasm
+	OPT2="-DXBYAK64 -DUSE_AVX512"
+	OPT3=win64
+	FILTER=./normalize_prefix
+	;;
+noexcept)
 	echo "nasm(32bit) without exception"
-	set EXE=nasm
-	set OPT2="-DXBYAK32 -DXBYAK_NO_EXCEPTION"
-	set OPT3=win32
-else
+	EXE=nasm
+	OPT2="-DXBYAK32 -DXBYAK_NO_EXCEPTION"
+	OPT3=win32
+	;;
+*)
 	echo "nasm(32bit)"
-	set EXE=nasm
-	set OPT2=-DXBYAK32
-	set OPT3=win32
-endif
+	EXE=nasm
+	OPT2=-DXBYAK32
+	OPT3=win32
+	;;
+esac
 
-set CFLAGS="-Wall -fno-operator-names -I../ $OPT2"
+CFLAGS="-Wall -fno-operator-names -I../ $OPT2"
 echo "compile make_nm.cpp with $CFLAGS"
 g++ $CFLAGS make_nm.cpp -o make_nm
 
 ./make_nm > a.asm
 echo "asm"
 $EXE -f$OPT3 a.asm -l a.lst
-awk '{if (index($3, "-")) { conti=substr($3, 0, length($3) - 1) } else { conti = conti $3; print conti; conti = "" }} ' < a.lst | $FILTER | grep -v "1+1" > ok.lst
+awk '$3 != "1+1" {printf "%s", sub(/-$/, "", $3) ? $3 : $3 ORS}' a.lst | $FILTER > ok.lst
 
 echo "xbyak"
 ./make_nm jit > nm.cpp
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 4310455b..bc0d71f5 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -108,7 +108,7 @@
 	#endif
 #endif
 
-#if (__cplusplus >= 201103) || (_MSC_VER >= 1800)
+#if (__cplusplus >= 201103) || (defined(_MSC_VER) && _MSC_VER >= 1800)
 	#undef XBYAK_TLS
 	#define XBYAK_TLS thread_local
 	#define XBYAK_VARIADIC_TEMPLATE
@@ -117,8 +117,11 @@
 	#define XBYAK_NOEXCEPT throw()
 #endif
 
-#if (__cplusplus >= 201402L) || (_MSC_VER >= 1910) // Visual Studio 2017 version 15.0
-	#define XBYAK_CONSTEXPR constexpr // require c++14 or later
+// require c++14 or later
+// Visual Studio 2017 version 15.0 or later
+// g++-6 or later
+#if ((__cplusplus >= 201402L) && !(!defined(__clang__) && defined(__GNUC__) && (__GNUC__ <= 5))) || (defined(_MSC_VER) && _MSC_VER >= 1910)
+	#define XBYAK_CONSTEXPR constexpr
 #else
 	#define XBYAK_CONSTEXPR
 #endif
@@ -135,7 +138,7 @@ namespace Xbyak {
 
 enum {
 	DEFAULT_MAX_CODE_SIZE = 4096,
-	VERSION = 0x5970 /* 0xABCD = A.BC(D) */
+	VERSION = 0x5991 /* 0xABCD = A.BC(D) */
 };
 
 #ifndef MIE_INTEGER_TYPE_DEFINED
@@ -413,16 +416,16 @@ public:
 	{
 		const size_t alignedSizeM1 = inner::ALIGN_PAGE_SIZE - 1;
 		size = (size + alignedSizeM1) & ~alignedSizeM1;
-#if defined(XBYAK_USE_MAP_JIT)
+#if defined(MAP_ANONYMOUS)
 		int mode = MAP_PRIVATE | MAP_ANONYMOUS;
-		const int mojaveVersion = 18;
-		if (util::getMacOsVersion() >= mojaveVersion) mode |= MAP_JIT;
-#elif defined(MAP_ANONYMOUS)
-		const int mode = MAP_PRIVATE | MAP_ANONYMOUS;
 #elif defined(MAP_ANON)
-		const int mode = MAP_PRIVATE | MAP_ANON;
+		int mode = MAP_PRIVATE | MAP_ANON;
 #else
 		#error "not supported"
+#endif
+#if defined(XBYAK_USE_MAP_JIT)
+		const int mojaveVersion = 18;
+		if (util::getMacOsVersion() >= mojaveVersion) mode |= MAP_JIT;
 #endif
 		void *p = mmap(NULL, size, PROT_READ | PROT_WRITE, mode, -1, 0);
 		if (p == MAP_FAILED) XBYAK_THROW_RET(ERR_CANT_ALLOC, 0)
@@ -924,6 +927,10 @@ inline RegExp operator*(const Reg& r, int scale)
 {
 	return RegExp(r, scale);
 }
+inline RegExp operator*(int scale, const Reg& r)
+{
+	return r * scale;
+}
 inline RegExp operator-(const RegExp& e, size_t disp)
 {
 	RegExp ret = e;
@@ -1539,6 +1546,12 @@ inline const uint8_t* Label::getAddress() const
 	return mgr->getCode() + offset;
 }
 
+typedef enum {
+	DefaultEncoding,
+	VexEncoding,
+	EvexEncoding
+} PreferredEncoding;
+
 class CodeGenerator : public CodeArray {
 public:
 	enum LabelType {
@@ -2293,6 +2306,19 @@ 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)
+	{
+		if (encoding == DefaultEncoding) {
+			encoding = EvexEncoding;
+		}
+		if (encoding == EvexEncoding) {
+#ifdef XBYAK_DISABLE_AVX512
+			XBYAK_THROW(ERR_EVEX_IS_INVALID)
+#endif
+			type |= T_MUST_EVEX;
+		}
+		opAVX_X_X_XM(x1, x2, op, type, code0);
+	}
 	void opInOut(const Reg& a, const Reg& d, uint8_t code)
 	{
 		if (a.getIdx() == Operand::AL && d.getIdx() == Operand::DX && d.getBit() == 16) {
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index 85e8bed5..5c1ecffb 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "5.97"; }
+const char *getVersionString() const { return "5.991"; }
 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); }
@@ -1180,6 +1180,10 @@ 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 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); }
@@ -1883,10 +1887,6 @@ void vpcompressd(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_N4 |
 void vpcompressq(const Operand& op, const Xmm& x) { opAVX_X_XM_IMM(x, op, T_N8 | T_66 | T_0F38 | T_EW1 | T_YMM | T_MUST_EVEX, 0x8B); }
 void vpconflictd(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, 0xC4); }
 void vpconflictq(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, 0xC4); }
-void vpdpbusd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x50); }
-void vpdpbusds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x51); }
-void vpdpwssd(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x52); }
-void vpdpwssds(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_SAE_Z | T_MUST_EVEX | T_B32, 0x53); }
 void vpermb(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x8D); }
 void vpermi2b(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX, 0x75); }
 void vpermi2d(const Xmm& x1, const Xmm& x2, const Operand& op) { opAVX_X_X_XM(x1, x2, op, T_66 | T_0F38 | T_EW0 | T_YMM | T_MUST_EVEX | T_B32, 0x76); }
diff --git a/xbyak/xbyak_util.h b/xbyak/xbyak_util.h
index 1516fc33..6ff9aab8 100644
--- a/xbyak/xbyak_util.h
+++ b/xbyak/xbyak_util.h
@@ -368,6 +368,7 @@ public:
 	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;
 
 	Cpu()
 		: type_(NONE)
@@ -389,19 +390,35 @@ public:
 		if (ECX == get32bitAsBE(amd)) {
 			type_ |= tAMD;
 			getCpuid(0x80000001, data);
-			if (EDX & (1U << 31)) type_ |= t3DN;
-			if (EDX & (1U << 15)) type_ |= tCMOV;
-			if (EDX & (1U << 30)) type_ |= tE3DN;
-			if (EDX & (1U << 22)) type_ |= tMMX2;
-			if (EDX & (1U << 27)) type_ |= tRDTSCP;
+			if (EDX & (1U << 31)) {
+				type_ |= t3DN;
+				// 3DNow! implies support for PREFETCHW on AMD
+				type_ |= tPREFETCHW;
+			}
+
+			if (EDX & (1U << 29)) {
+				// Long mode implies support for PREFETCHW on AMD
+				type_ |= tPREFETCHW;
+			}
 		}
 		if (ECX == get32bitAsBE(intel)) {
 			type_ |= tINTEL;
+		}
+
+		// Extended flags information
+		getCpuid(0x80000000, data);
+		if (EAX >= 0x80000001) {
 			getCpuid(0x80000001, data);
+
+			if (EDX & (1U << 31)) type_ |= t3DN;
+			if (EDX & (1U << 30)) type_ |= tE3DN;
 			if (EDX & (1U << 27)) type_ |= tRDTSCP;
+			if (EDX & (1U << 22)) type_ |= tMMX2;
+			if (EDX & (1U << 15)) type_ |= tCMOV;
 			if (ECX & (1U << 5)) type_ |= tLZCNT;
 			if (ECX & (1U << 8)) type_ |= tPREFETCHW;
 		}
+
 		getCpuid(1, data);
 		if (ECX & (1U << 0)) type_ |= tSSE3;
 		if (ECX & (1U << 9)) type_ |= tSSSE3;
@@ -426,7 +443,11 @@ public:
 			if ((bv & 6) == 6) {
 				if (ECX & (1U << 28)) type_ |= tAVX;
 				if (ECX & (1U << 12)) type_ |= tFMA;
-				if (((bv >> 5) & 7) == 7) {
+				// do *not* check AVX-512 state on macOS because it has on-demand AVX-512 support
+#if !defined(__APPLE__)
+				if (((bv >> 5) & 7) == 7)
+#endif
+				{
 					getCpuidEx(7, 0, data);
 					if (EBX & (1U << 16)) type_ |= tAVX512F;
 					if (type_ & tAVX512F) {
@@ -449,16 +470,12 @@ public:
 						if (EDX & (1U << 3)) type_ |= tAVX512_4FMAPS;
 						if (EDX & (1U << 8)) type_ |= tAVX512_VP2INTERSECT;
 					}
-					// EAX=07H, ECX=1
-					getCpuidEx(7, 1, data);
-					if (type_ & tAVX512F) {
-						if (EAX & (1U << 5)) type_ |= tAVX512_BF16;
-					}
 				}
 			}
 		}
 		if (maxNum >= 7) {
 			getCpuidEx(7, 0, data);
+			const uint32_t maxNumSubLeaves = EAX;
 			if (type_ & tAVX && (EBX & (1U << 5))) type_ |= tAVX2;
 			if (EBX & (1U << 3)) type_ |= tBMI1;
 			if (EBX & (1U << 8)) type_ |= tBMI2;
@@ -474,6 +491,13 @@ public:
 			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 << 4)) type_ |= tAVX_VNNI;
+				if (type_ & tAVX512F) {
+					if (EAX & (1U << 5)) type_ |= tAVX512_BF16;
+				}
+			}
 		}
 		setFamily();
 		setNumCores();