Speed up rounding to byte test

unit_test/color_test.cc

Index: unit_test/color_test.cc
diff --git a/unit_test/color_test.cc b/unit_test/color_test.cc
index d3fda0d4dcd7faa0b82b74c70997fad0f59f9dff..81189083dc18690d0afb5555301e4e02b9889386 100644
--- a/unit_test/color_test.cc
+++ b/unit_test/color_test.cc
@@ -229,10 +229,10 @@ static void YJToRGB(int y, int* r, int* g, int* b) {
 //  #define CLAMPMETHOD_MASK 1
 
 // Pick a method for rounding.
-#define ROUND(f) static_cast<int>(f + 0.5)
-//  #define ROUND(f) lrintf(f)
-//  #define ROUND(f) static_cast<int>(round(f))
-//  #define ROUND(f) _mm_cvt_ss2si(_mm_load_ss(&f))
+#define ROUND(f) lrintf(f)

[brucedawson, 2015/10/02 18:48:07]

But, lrintf is ~376 times slower on VC++
(https://connect.microsoft.com/VisualStudio/feedback/details/1179496/cq-intrin...)

[fbarchard, 2015/10/02 21:44:53]

Done.

+//#define ROUND(f) static_cast<int>(f + 0.5)
+//#define ROUND(f) static_cast<int>(round(f))
+//#define ROUND(f) _mm_cvt_ss2si(_mm_load_ss(&f))
 
 #if defined(CLAMPMETHOD_IF)
 static int RoundToByte(float f) {
@@ -314,16 +314,11 @@ static int RoundToByte(float f) {
 
 TEST_F(libyuvTest, TestRoundToByte) {
   int allb = 0;
-  for (int i = 0; i < benchmark_iterations_; ++i) {
-    for (int u2 = 0; u2 < 256; ++u2) {
-      for (int v2 = 0; v2 < 256; ++v2) {
-        for (int y2 = 0; y2 < 256; ++y2) {
-          int y = RANDOM256(y2);
-          int b = RoundToByte(y * 810.33 - 257);
-          allb |= b;
-        }
-      }
-    }
+  int count = benchmark_width_ * benchmark_height_ * benchmark_iterations_;
+  for (int i = 0; i < count; ++i) {
+    int y = RANDOM256(i);
+    int b = RoundToByte(y * 810.33 - 257);
+    allb |= b;
   }
   EXPECT_GE(allb, 0);
   EXPECT_LE(allb, 255);