From 78070bc1ba0978000268f919f1138bede28df436 Mon Sep 17 00:00:00 2001
From: "fbarchard@google.com"
 <fbarchard@google.com@16f28f9a-4ce2-e073-06de-1de4eb20be90>
Date: Tue, 25 Sep 2012 19:13:04 +0000
Subject: [PATCH] Added CPUID change to detect AVX2. BUG=98 TEST=none Review
 URL: https://webrtc-codereview.appspot.com/839004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@373 16f28f9a-4ce2-e073-06de-1de4eb20be90
---
 README.chromium          | 18 +++++++++---------
 include/libyuv/cpu_id.h  |  2 +-
 include/libyuv/version.h |  2 +-
 source/cpu_id.cc         | 30 ++++++++++++++++++++++++++++--
 unit_test/cpu_test.cc    |  2 ++
 5 files changed, 41 insertions(+), 13 deletions(-)

diff --git a/README.chromium b/README.chromium
index 9dc6f330e..0412b0daf 100644
--- a/README.chromium
+++ b/README.chromium
@@ -1,9 +1,9 @@
-Name: libyuv
-URL: http://code.google.com/p/libyuv/
-Version: 372
-License: BSD
-License File: LICENSE
-
-Description:
-libyuv is an open source project that includes
-YUV conversion and scaling functionality.
+Name: libyuv
+URL: http://code.google.com/p/libyuv/
+Version: 373
+License: BSD
+License File: LICENSE
+
+Description:
+libyuv is an open source project that includes
+YUV conversion and scaling functionality.
diff --git a/include/libyuv/cpu_id.h b/include/libyuv/cpu_id.h
index e96438388..019329d2d 100644
--- a/include/libyuv/cpu_id.h
+++ b/include/libyuv/cpu_id.h
@@ -31,7 +31,7 @@ static const int kCpuHasSSSE3 = 0x40;
 static const int kCpuHasSSE41 = 0x80;
 static const int kCpuHasSSE42 = 0x100;
 static const int kCpuHasAVX = 0x200;
-// 0x400 reserved for AVX2.
+static const int kCpuHasAVX2 = 0x400;
 
 // Detect CPU has SSE2 etc.
 // Test_flag parameter should be one of kCpuHas constants above.
diff --git a/include/libyuv/version.h b/include/libyuv/version.h
index 090bde592..b8fe8d8dd 100644
--- a/include/libyuv/version.h
+++ b/include/libyuv/version.h
@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 372
+#define LIBYUV_VERSION 373
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT
diff --git a/source/cpu_id.cc b/source/cpu_id.cc
index fb307f045..61c2f93de 100644
--- a/source/cpu_id.cc
+++ b/source/cpu_id.cc
@@ -50,11 +50,25 @@ extern "C" {
 #if !defined(__CLR_VER) && (defined(_M_IX86) || defined(_M_X64) || \
     defined(__i386__) || defined(__x86_64__))
 void CpuId(int cpu_info[4], int info_type) {
-    __cpuid(cpu_info, info_type);
+  __cpuid(cpu_info, info_type);
 }
 #else
 void CpuId(int cpu_info[4], int) {
-    cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
+  cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
+}
+#endif
+
+// Low level cpuid for X86.  Returns zeros on other CPUs.
+#if !defined(__CLR_VER) && defined(_M_IX86)
+// TODO(fbarchard): Port to GCC and 64 bit Visual C.
+#define HAS_XGETBV
+// Return low 32 bits of BV - OS support for register saving.
+__declspec(naked)
+static uint32 XGetBV32(void) {
+  _asm _emit 0x0f
+  _asm _emit 0x01
+  _asm _emit 0xd0  // xgetbv
+  _asm ret
 }
 #endif
 
@@ -91,8 +105,17 @@ int InitCpuFlags() {
               ((cpu_info[2] & 0x00000200) ? kCpuHasSSSE3 : 0) |
               ((cpu_info[2] & 0x00080000) ? kCpuHasSSE41 : 0) |
               ((cpu_info[2] & 0x00100000) ? kCpuHasSSE42 : 0) |
+              // TODO(fbarchard): AVX test BV same as AVX2.
               (((cpu_info[2] & 0x18000000) == 0x18000000) ? kCpuHasAVX : 0) |
               kCpuInitialized | kCpuHasX86;
+#ifdef HAS_XGETBV
+  if (cpu_info_ & kCpuHasAVX) {
+    __cpuid(cpu_info, 7);
+    if ((cpu_info[1] & 0x00000020) && ((XGetBV32() & 0x06) == 0x06)) {
+      cpu_info_ |= kCpuHasAVX2;
+    }
+  }
+#endif
 
   // environment variable overrides for testing.
   if (getenv("LIBYUV_DISABLE_X86")) {
@@ -113,6 +136,9 @@ int InitCpuFlags() {
   if (getenv("LIBYUV_DISABLE_AVX")) {
     cpu_info_ &= ~kCpuHasAVX;
   }
+  if (getenv("LIBYUV_DISABLE_AVX2")) {
+    cpu_info_ &= ~kCpuHasAVX2;
+  }
   if (getenv("LIBYUV_DISABLE_ASM")) {
     cpu_info_ = kCpuInitialized;
   }
diff --git a/unit_test/cpu_test.cc b/unit_test/cpu_test.cc
index 5be4e0c21..fbf6da92a 100644
--- a/unit_test/cpu_test.cc
+++ b/unit_test/cpu_test.cc
@@ -37,6 +37,8 @@ TEST_F(libyuvTest, TestCpuHas) {
   printf("Has SSE4.2 %x\n", has_sse42);
   int has_avx = TestCpuFlag(kCpuHasAVX);
   printf("Has AVX %x\n", has_avx);
+  int has_avx2 = TestCpuFlag(kCpuHasAVX2);
+  printf("Has AVX2 %x\n", has_avx2);
 }
 
 #if defined(__i386__) || defined(__x86_64__) || \