Use int for pixel value difference

This fixes overflow of unsigned int when the difference is negative.

Change-Id: I7a1ef2af84e59e675ec27f91de4e6d92a9c1d90c

test/dct16x16_test.cc

@@ -368,10 +368,10 @@ class Trans16x16TestBase {
 
       for (int j = 0; j < kNumCoeffs; ++j) {
 #if CONFIG_AOM_HIGHBITDEPTH
-        const uint32_t diff =
+        const int32_t diff =
             bit_depth_ == AOM_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
 #else
-        const uint32_t diff = dst[j] - src[j];
+        const int32_t diff = dst[j] - src[j];
 #endif
         const uint32_t error = diff * diff;
         if (max_error < error) max_error = error;

test/dct32x32_test.cc

@@ -154,10 +154,10 @@ TEST_P(Trans32x32Test, AccuracyCheck) {
 
     for (int j = 0; j < kNumCoeffs; ++j) {
 #if CONFIG_AOM_HIGHBITDEPTH
-      const uint32_t diff =
+      const int32_t diff =
           bit_depth_ == AOM_BITS_8 ? dst[j] - src[j] : dst16[j] - src16[j];
 #else
-      const uint32_t diff = dst[j] - src[j];
+      const int32_t diff = dst[j] - src[j];
 #endif
       const uint32_t error = diff * diff;
       if (max_error < error) max_error = error;