Optimize yuv alpha blend AVX2 code to do 32 pixels at time.

source/cpu_id.cc

 /*
  *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
  *
  *  Use of this source code is governed by a BSD-style license
  *  that can be found in the LICENSE file in the root of the source
  *  tree. An additional intellectual property rights grant can be found
  *  in the file PATENTS. All contributing project authors may
  *  be found in the AUTHORS file in the root of the source tree.
  */
 
@@ skipping 88 matching lines @@
 #endif
 
 // TODO(fbarchard): Enable xgetbv when validator supports it.
 #if (defined(_M_IX86) || defined(_M_X64) || \
     defined(__i386__) || defined(__x86_64__)) && \
     !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 GetXCR0() {
   uint32 xcr0 = 0u;
-#if (defined(_MSC_VER) && !defined(__clang__)) && (_MSC_FULL_VER >= 160040219)
+#if defined(_MSC_VER)
+//0 && (defined(_MSC_VER) && !defined(__clang__)) && (_MSC_FULL_VER >= 160040219)

[harryjin, 2015/12/08 07:48:58]

? Remove?

[fbarchard, 2015/12/08 19:24:04]

Done.

   xcr0 = (uint32)(_xgetbv(0));  // VS2010 SP1 required.
+  printf("xgetbv xcr0 %d\n", xcr0);

[harryjin, 2015/12/08 07:48:58]

Remove the printf?

[fbarchard, 2015/12/08 19:24:04]

Done.

 #elif defined(_M_IX86) && defined(_MSC_VER) && !defined(__clang__)
   __asm {
     xor        ecx, ecx    // xcr 0
     _asm _emit 0x0f _asm _emit 0x01 _asm _emit 0xd0  // For VS2010 and earlier.
     mov        xcr0, eax
   }
+  printf("asm xcr0 %d\n", xcr0);

[harryjin, 2015/12/08 07:48:58]

same here?

[fbarchard, 2015/12/08 19:24:04]

Done.

 #elif defined(__i386__) || defined(__x86_64__)
   asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx");
 #endif  // defined(__i386__) || defined(__x86_64__)
   return xcr0;
 }
 #endif  // defined(_M_IX86) || defined(_M_X64) ..
 
 // based on libvpx arm_cpudetect.c
 // For Arm, but public to allow testing on any CPU
 LIBYUV_API SAFEBUFFERS
@@ skipping 63 matching lines @@
   cpu_info = ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
              ((cpu_info1[2] & 0x00000200) ? kCpuHasSSSE3 : 0) |
              ((cpu_info1[2] & 0x00080000) ? kCpuHasSSE41 : 0) |
              ((cpu_info1[2] & 0x00100000) ? kCpuHasSSE42 : 0) |
              ((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0) |
              ((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
              kCpuHasX86;
 
 #ifdef HAS_XGETBV
   // AVX requires CPU has AVX, XSAVE and OSXSave for xgetbv
-  if ((cpu_info1[2] & 0x1c000000) == 0x1c000000 &&  // AVX and OSXSave
-      (GetXCR0() & 6) == 6) {  // Test OD saves YMM registers
+  if (((cpu_info1[2] & 0x1c000000) == 0x1c000000) &&  // AVX and OSXSave
+      ((GetXCR0() & 6) == 6)) {  // Test OS 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.
@@ skipping 73 matching lines @@
 // Note that use of this function is not thread safe.
 LIBYUV_API
 void MaskCpuFlags(int enable_flags) {
   cpu_info_ = InitCpuFlags() & enable_flags;
 }
 
 #ifdef __cplusplus
 }  // extern "C"
 }  // namespace libyuv
 #endif