use __popcnt for visual c HammingDistance_X86

BUG=libyuv:701
TEST=HammingDistance unittest performance is comparable to x64
R=wangcheng@google.com

Change-Id: I8abe861e086e0162ba4c7ba6f1ef7d1c006cd9d4
Reviewed-on: https://chromium-review.googlesource.com/505454Reviewed-by: Frank Barchard <fbarchard@google.com>
Commit-Queue: Frank Barchard <fbarchard@google.com>

include/libyuv/compare_row.h

@@ -52,10 +52,6 @@ extern "C" {
     (defined(__x86_64__) || defined(__i386__) || defined(_M_IX86))
 #define HAS_HASHDJB2_SSE41
 #define HAS_SUMSQUAREERROR_SSE2
-#endif
-
-// The following are available for GCC:
-#if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__))
 #define HAS_HAMMINGDISTANCE_X86
 #endif
 

source/compare_win.cc

@@ -13,6 +13,10 @@
 #include "libyuv/compare_row.h"
 #include "libyuv/row.h"
 
+#if defined(_MSC_VER)
+#include <intrin.h>   // For __popcnt
+#endif
+
 #ifdef __cplusplus
 namespace libyuv {
 extern "C" {
@@ -21,6 +25,19 @@ extern "C" {
 // This module is for 32 bit Visual C x86 and clangcl
 #if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
 
+uint32 HammingDistance_X86(const uint8* src_a, const uint8* src_b, int count) {
+  uint32 diff = 0u;
+
+  int i;
+  for (i = 0; i < count - 3; i += 4) {
+    uint32 x = *((uint32*)src_a) ^ *((uint32*)src_b);
+    src_a += 4;
+    src_b += 4;
+    diff += __popcnt(x);
+  }
+  return diff;
+}
+
 __declspec(naked) uint32
     SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
   __asm {

unit_test/compare_test.cc

@@ -227,10 +227,9 @@ TEST_F(LibYUVBaseTest, BenchmarkHammingDistance_Opt) {
       benchmark_iterations_ *
       ((benchmark_width_ * benchmark_height_ + kMaxWidth - 1) / kMaxWidth);
   for (int i = 0; i < count; ++i) {
-#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
+#if defined(HAS_HAMMINGDISTANCE_NEON)
     h1 = HammingDistance_NEON(src_a, src_b, kMaxWidth);
-#elif !defined(LIBYUV_DISABLE_X86) && \
-    (defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
+#elif defined(HAS_HAMMINGDISTANCE_X86)
     h1 = HammingDistance_X86(src_a, src_b, kMaxWidth);
 #else
     h1 = HammingDistance_C(src_a, src_b, kMaxWidth);