add detection of avx-512

6 files changed, 44 insertions, 6 deletions

diff --git a/readme.md b/readme.md
index 4c01c3c..6cf2408 100644
--- a/readme.md
+++ b/readme.md
@@ -1,5 +1,5 @@
 
-Xbyak 4.901 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
+Xbyak 4.91 ; JIT assembler for x86(IA32), x64(AMD64, x86-64) by C++
 =============
 
 Abstract
@@ -285,7 +285,8 @@ The header files under xbyak/ are independent of cybozulib.
 
 History
 -------------
-* 2016/May/14 ver 4.901 comment to ready() function(thanks to skmp)
+* 2016/May/05 ver 4.91 add detection of AVX-512 to Xbyak::util::Cpu
+* 2016/Mar/14 ver 4.901 comment to ready() function(thanks to skmp)
 * 2016/Feb/04 ver 4.90 add jcc(const void *addr);
 * 2016/Jan/30 ver 4.89 vpblendvb supports ymm reg(thanks to John Funnell)
 * 2016/Jan/24 ver 4.88 lea, cmov supports 16-bit register(thanks to whyisthisfieldhere)
diff --git a/readme.txt b/readme.txt
index fb569c9..dff3478 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 4.901

+    C++用x86(IA-32), x64(AMD64, x86-64) JITアセンブラ Xbyak 4.91

 

 -----------------------------------------------------------------------------

 ◎概要

@@ -301,7 +301,8 @@ cybozulibは単体テストでのみ利用されていて、xbyak/ディレク�
 -----------------------------------------------------------------------------

 ◎履歴

 

-2016/03/14 ver 4.90 ready()関数にコメント加筆(thanks to skmp)

+2016/05/05 ver 4.91 AVX-512命令の検出サポート

+2016/03/14 ver 4.901 ready()関数にコメント加筆(thanks to skmp)

 2016/02/04 ver 4.90 条件分岐命令にjcc(const void *addr);のタイプを追加

 2016/01/30 ver 4.89 vpblendvbがymmレジスタをサポートしていなかった(thanks to John Funnell)

 2016/01/24 ver 4.88 lea, cmovの16bitレジスタ対応(thanks to whyisthisfieldhere)

diff --git a/sample/test_util.cpp b/sample/test_util.cpp
index b6f4472..7d5af42 100644
--- a/sample/test_util.cpp
+++ b/sample/test_util.cpp
@@ -55,6 +55,15 @@ void putCPUinfo()
 		{ Cpu::tRTM, "rtm" },
 		{ Cpu::tF16C, "f16c" },
 		{ Cpu::tMOVBE, "movbe" },
+		{ Cpu::tAVX512F, "avx512f" },
+		{ Cpu::tAVX512DQ, "avx512dq" },
+		{ Cpu::tAVX512IFMA, "avx512ifma" },
+		{ Cpu::tAVX512PF, "avx512pf" },
+		{ Cpu::tAVX512ER, "avx512er" },
+		{ Cpu::tAVX512CD, "avx512cd" },
+		{ Cpu::tAVX512BW, "avx512bw" },
+		{ Cpu::tAVX512VL, "avx512vl" },
+		{ Cpu::tAVX512VBMI, "avx512vbmi" },
 	};
 	for (size_t i = 0; i < NUM_OF_ARRAY(tbl); i++) {
 		if (cpu.has(tbl[i].type)) printf(" %s", tbl[i].str);
diff --git a/xbyak/xbyak.h b/xbyak/xbyak.h
index 3483935..351de28 100644
--- a/xbyak/xbyak.h
+++ b/xbyak/xbyak.h
@@ -101,7 +101,7 @@ namespace Xbyak {
 
 enum {
 	DEFAULT_MAX_CODE_SIZE = 4096,
-	VERSION = 0x4901 /* 0xABCD = A.BC(D) */
+	VERSION = 0x4910 /* 0xABCD = A.BC(D) */
 };
 
 #ifndef MIE_INTEGER_TYPE_DEFINED
diff --git a/xbyak/xbyak_mnemonic.h b/xbyak/xbyak_mnemonic.h
index fb18c5f..d1f2502 100644
--- a/xbyak/xbyak_mnemonic.h
+++ b/xbyak/xbyak_mnemonic.h
@@ -1,4 +1,4 @@
-const char *getVersionString() const { return "4.901"; }
+const char *getVersionString() const { return "4.91"; }
 void packssdw(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x6B); }
 void packsswb(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x63); }
 void packuswb(const Mmx& mmx, const Operand& op) { opMMX(mmx, op, 0x67); }
diff --git a/xbyak/xbyak_util.h b/xbyak/xbyak_util.h
index 3a7c2c2..67fba08 100644
--- a/xbyak/xbyak_util.h
+++ b/xbyak/xbyak_util.h
@@ -96,6 +96,9 @@ public:
 	int extFamily;
 	int displayFamily; // family + extFamily
 	int displayModel; // model + extModel
+	/*
+		data[] = { eax, ebx, ecx, edx }
+	*/
 	static inline void getCpuid(unsigned int eaxIn, unsigned int data[4])
 	{
 #ifdef _MSC_VER
@@ -164,6 +167,15 @@ public:
 	static const Type tRTM = uint64(1) << 32; // xbegin, xend, xabort
 	static const Type tF16C = uint64(1) << 33; // vcvtph2ps, vcvtps2ph
 	static const Type tMOVBE = uint64(1) << 34; // mobve
+	static const Type tAVX512F = uint64(1) << 35;
+	static const Type tAVX512DQ = uint64(1) << 36;
+	static const Type tAVX512IFMA = uint64(1) << 37;
+	static const Type tAVX512PF = uint64(1) << 38;
+	static const Type tAVX512ER = uint64(1) << 39;
+	static const Type tAVX512CD = uint64(1) << 40;
+	static const Type tAVX512BW = uint64(1) << 41;
+	static const Type tAVX512VL = uint64(1) << 42;
+	static const Type tAVX512VBMI = uint64(1) << 43;
 
 	Cpu()
 		: type_(NONE)
@@ -212,6 +224,21 @@ public:
 			if ((bv & 6) == 6) {
 				if (data[2] & (1U << 28)) type_ |= tAVX;
 				if (data[2] & (1U << 12)) type_ |= tFMA;
+				if (((bv >> 5) & 7) == 7) {
+					getCpuid(7, data);
+					if (data[1] & (1U << 16)) type_ |= tAVX512F;
+					if (type_ & tAVX512F) {
+						getCpuidEx(7, 0, data);
+						if (data[1] & (1U << 17)) type_ |= tAVX512DQ;
+						if (data[1] & (1U << 21)) type_ |= tAVX512IFMA;
+						if (data[1] & (1U << 26)) type_ |= tAVX512PF;
+						if (data[1] & (1U << 27)) type_ |= tAVX512ER;
+						if (data[1] & (1U << 28)) type_ |= tAVX512CD;
+						if (data[1] & (1U << 30)) type_ |= tAVX512BW;
+						if (data[1] & (1U << 31)) type_ |= tAVX512VL;
+						if (data[2] & (1U << 1)) type_ |= tAVX512VBMI;
+					}
+				}
 			}
 		}
 		if (maxNum >= 7) {