Add avx512bw detection code

R=harryjin@google.com
BUG=libyuv:514

Review URL: https://codereview.chromium.org/1413463004 .

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1522
+Version: 1523
 License: BSD
 License File: LICENSE
 

include/libyuv/cpu_id.h

@@ -36,7 +36,7 @@ static const int kCpuHasAVX = 0x200;
 static const int kCpuHasAVX2 = 0x400;
 static const int kCpuHasERMS = 0x800;
 static const int kCpuHasFMA3 = 0x1000;
-static const int kCpuHasAVX3 = 0x2000;  // TODO(fbarchard): implement detect.
+static const int kCpuHasAVX3 = 0x2000;
 // 0x2000, 0x4000, 0x8000 reserved for future X86 flags.
 
 // These flags are only valid on MIPS processors.

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1522
+#define LIBYUV_VERSION 1523
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/cpu_id.cc

@@ -104,7 +104,7 @@ void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
     !defined(__pnacl__) && !defined(__CLR_VER) && !defined(__native_client__)
 #define HAS_XGETBV
 // X86 CPUs have xgetbv to detect OS saves high parts of ymm registers.
-int TestOsSaveYmm() {
+int GetXCR0() {
   uint32 xcr0 = 0u;
 #if (defined(_MSC_VER) && !defined(__clang__)) && (_MSC_FULL_VER >= 160040219)
   xcr0 = (uint32)(_xgetbv(0));  // VS2010 SP1 required.
@@ -117,7 +117,7 @@ int TestOsSaveYmm() {
 #elif defined(__i386__) || defined(__x86_64__)
   asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx");
 #endif  // defined(__i386__) || defined(__x86_64__)
-  return((xcr0 & 6) == 6);  // Is ymm saved?
+  return xcr0;
 }
 #endif  // defined(_M_IX86) || defined(_M_X64) ..
 
@@ -219,10 +219,16 @@ int InitCpuFlags(void) {
 #ifdef HAS_XGETBV
   // AVX requires CPU has AVX, XSAVE and OSXSave for xgetbv
   if ((cpu_info1[2] & 0x1c000000) == 0x1c000000 &&  // AVX and OSXSave
-      !TestEnv("LIBYUV_DISABLE_AVX") && TestOsSaveYmm()) {  // Saves YMM.
+      (GetXCR0() & 6) == 6) {  // Test OD saves YMM registers
     cpu_info |= ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) | kCpuHasAVX;
+
+    // Detect AVX512bw
+    if ((GetXCR0() & 0xe0) == 0xe0) {
+      cpu_info |= (cpu_info7[1] & 0x40000000) ? kCpuHasAVX3 : 0;
+    }
   }
 #endif
+
   // Environment variable overrides for testing.
   if (TestEnv("LIBYUV_DISABLE_X86")) {
     cpu_info &= ~kCpuHasX86;
@@ -239,6 +245,9 @@ int InitCpuFlags(void) {
   if (TestEnv("LIBYUV_DISABLE_SSE42")) {
     cpu_info &= ~kCpuHasSSE42;
   }
+  if (TestEnv("LIBYUV_DISABLE_AVX")) {
+    cpu_info &= ~kCpuHasAVX;
+  }
   if (TestEnv("LIBYUV_DISABLE_AVX2")) {
     cpu_info &= ~kCpuHasAVX2;
   }
@@ -248,6 +257,9 @@ int InitCpuFlags(void) {
   if (TestEnv("LIBYUV_DISABLE_FMA3")) {
     cpu_info &= ~kCpuHasFMA3;
   }
+  if (TestEnv("LIBYUV_DISABLE_AVX3")) {
+    cpu_info &= ~kCpuHasAVX3;
+  }
 #endif
 #if defined(__mips__) && defined(__linux__)
   // Linux mips parse text file for dsp detect.

unit_test/cpu_test.cc

@@ -44,6 +44,8 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
   printf("Has ERMS %x\n", has_erms);
   int has_fma3 = TestCpuFlag(kCpuHasFMA3);
   printf("Has FMA3 %x\n", has_fma3);
+  int has_avx3 = TestCpuFlag(kCpuHasAVX3);
+  printf("Has AVX3 %x\n", has_avx3);
   int has_mips = TestCpuFlag(kCpuHasMIPS);
   printf("Has MIPS %x\n", has_mips);
   int has_mips_dsp = TestCpuFlag(kCpuHasMIPS_DSP);