Update external/libyuv to r1837

Update external/libyuv to version r1837 (d53f1bee) from upstream.

This brings in some new functions that will be used by libavif.

Also update README.version with local modifications. The files/
subdirectory is a pristine copy of the upstream checkout except
for the local modifications that are listed in README.version.

Bug: b/241008246
Bug: b/228492909
Test: Builds. Media and Camera CTS tests pass.
Merged-In: I1bd041e475666e13fafd8c3792d7142a022d435a
Change-Id: I85df21181e582ff835aab1e9ef13ad3ac8421d30

21 files changed, 5587 insertions, 1330 deletions

diff --git a/files/unit_test/color_test.cc b/files/unit_test/color_test.cc
index 4bb448d5..01267ff1 100644
--- a/files/unit_test/color_test.cc
+++ b/files/unit_test/color_test.cc
@@ -20,20 +20,22 @@
 
 namespace libyuv {
 
-// TODO(fbarchard): Port high accuracy YUV to RGB to Neon.
-#if !defined(LIBYUV_DISABLE_NEON) && \
-    (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON))
-#define ERROR_R 1
-#define ERROR_G 1
-#define ERROR_B 3
-#define ERROR_FULL 6
-#define ERROR_J420 5
+// TODO(fbarchard): clang x86 has a higher accuracy YUV to RGB.
+// Port to Visual C and other CPUs
+#if !defined(LIBYUV_BIT_EXACT) && !defined(LIBYUV_DISABLE_X86) && \
+    (defined(__x86_64__) || defined(__i386__))
+#define ERROR_FULL 5
+#define ERROR_J420 4
 #else
+#define ERROR_FULL 6
+#define ERROR_J420 6
+#endif
 #define ERROR_R 1
 #define ERROR_G 1
-#define ERROR_B 3
-#define ERROR_FULL 5
-#define ERROR_J420 3
+#ifdef LIBYUV_UNLIMITED_DATA
+#define ERROR_B 1
+#else
+#define ERROR_B 18
 #endif
 
 #define TESTCS(TESTNAME, YUVTOARGB, ARGBTOYUV, HS1, HS, HN, DIFF)              \
@@ -187,6 +189,104 @@ static void YUVJToRGB(int y, int u, int v, int* r, int* g, int* b) {
   *r = orig_pixels[2];
 }
 
+static void YUVHToRGB(int y, int u, int v, int* r, int* g, int* b) {
+  const int kWidth = 16;
+  const int kHeight = 1;
+  const int kPixels = kWidth * kHeight;
+  const int kHalfPixels = ((kWidth + 1) / 2) * ((kHeight + 1) / 2);
+
+  SIMD_ALIGNED(uint8_t orig_y[16]);
+  SIMD_ALIGNED(uint8_t orig_u[8]);
+  SIMD_ALIGNED(uint8_t orig_v[8]);
+  SIMD_ALIGNED(uint8_t orig_pixels[16 * 4]);
+  memset(orig_y, y, kPixels);
+  memset(orig_u, u, kHalfPixels);
+  memset(orig_v, v, kHalfPixels);
+
+  /* YUV converted to ARGB. */
+  H422ToARGB(orig_y, kWidth, orig_u, (kWidth + 1) / 2, orig_v, (kWidth + 1) / 2,
+             orig_pixels, kWidth * 4, kWidth, kHeight);
+
+  *b = orig_pixels[0];
+  *g = orig_pixels[1];
+  *r = orig_pixels[2];
+}
+
+#define F422ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I422ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvF709Constants, i, j)
+
+static void YUVFToRGB(int y, int u, int v, int* r, int* g, int* b) {
+  const int kWidth = 16;
+  const int kHeight = 1;
+  const int kPixels = kWidth * kHeight;
+  const int kHalfPixels = ((kWidth + 1) / 2) * ((kHeight + 1) / 2);
+
+  SIMD_ALIGNED(uint8_t orig_y[16]);
+  SIMD_ALIGNED(uint8_t orig_u[8]);
+  SIMD_ALIGNED(uint8_t orig_v[8]);
+  SIMD_ALIGNED(uint8_t orig_pixels[16 * 4]);
+  memset(orig_y, y, kPixels);
+  memset(orig_u, u, kHalfPixels);
+  memset(orig_v, v, kHalfPixels);
+
+  /* YUV converted to ARGB. */
+  F422ToARGB(orig_y, kWidth, orig_u, (kWidth + 1) / 2, orig_v, (kWidth + 1) / 2,
+             orig_pixels, kWidth * 4, kWidth, kHeight);
+
+  *b = orig_pixels[0];
+  *g = orig_pixels[1];
+  *r = orig_pixels[2];
+}
+
+static void YUVUToRGB(int y, int u, int v, int* r, int* g, int* b) {
+  const int kWidth = 16;
+  const int kHeight = 1;
+  const int kPixels = kWidth * kHeight;
+  const int kHalfPixels = ((kWidth + 1) / 2) * ((kHeight + 1) / 2);
+
+  SIMD_ALIGNED(uint8_t orig_y[16]);
+  SIMD_ALIGNED(uint8_t orig_u[8]);
+  SIMD_ALIGNED(uint8_t orig_v[8]);
+  SIMD_ALIGNED(uint8_t orig_pixels[16 * 4]);
+  memset(orig_y, y, kPixels);
+  memset(orig_u, u, kHalfPixels);
+  memset(orig_v, v, kHalfPixels);
+
+  /* YUV converted to ARGB. */
+  U422ToARGB(orig_y, kWidth, orig_u, (kWidth + 1) / 2, orig_v, (kWidth + 1) / 2,
+             orig_pixels, kWidth * 4, kWidth, kHeight);
+
+  *b = orig_pixels[0];
+  *g = orig_pixels[1];
+  *r = orig_pixels[2];
+}
+
+#define V422ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I422ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvV2020Constants, i, j)
+
+static void YUVVToRGB(int y, int u, int v, int* r, int* g, int* b) {
+  const int kWidth = 16;
+  const int kHeight = 1;
+  const int kPixels = kWidth * kHeight;
+  const int kHalfPixels = ((kWidth + 1) / 2) * ((kHeight + 1) / 2);
+
+  SIMD_ALIGNED(uint8_t orig_y[16]);
+  SIMD_ALIGNED(uint8_t orig_u[8]);
+  SIMD_ALIGNED(uint8_t orig_v[8]);
+  SIMD_ALIGNED(uint8_t orig_pixels[16 * 4]);
+  memset(orig_y, y, kPixels);
+  memset(orig_u, u, kHalfPixels);
+  memset(orig_v, v, kHalfPixels);
+
+  /* YUV converted to ARGB. */
+  V422ToARGB(orig_y, kWidth, orig_u, (kWidth + 1) / 2, orig_v, (kWidth + 1) / 2,
+             orig_pixels, kWidth * 4, kWidth, kHeight);
+
+  *b = orig_pixels[0];
+  *g = orig_pixels[1];
+  *r = orig_pixels[2];
+}
+
 static void YToRGB(int y, int* r, int* g, int* b) {
   const int kWidth = 16;
   const int kHeight = 1;
@@ -335,18 +435,50 @@ TEST_F(LibYUVColorTest, TestRoundToByte) {
   EXPECT_LE(allb, 255);
 }
 
+// BT.601 limited range YUV to RGB reference
 static void YUVToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
   *r = RoundToByte((y - 16) * 1.164 - (v - 128) * -1.596);
   *g = RoundToByte((y - 16) * 1.164 - (u - 128) * 0.391 - (v - 128) * 0.813);
   *b = RoundToByte((y - 16) * 1.164 - (u - 128) * -2.018);
 }
 
+// BT.601 full range YUV to RGB reference (aka JPEG)
 static void YUVJToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
   *r = RoundToByte(y - (v - 128) * -1.40200);
   *g = RoundToByte(y - (u - 128) * 0.34414 - (v - 128) * 0.71414);
   *b = RoundToByte(y - (u - 128) * -1.77200);
 }
 
+// BT.709 limited range YUV to RGB reference
+// See also http://www.equasys.de/colorconversion.html
+static void YUVHToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
+  *r = RoundToByte((y - 16) * 1.164 - (v - 128) * -1.793);
+  *g = RoundToByte((y - 16) * 1.164 - (u - 128) * 0.213 - (v - 128) * 0.533);
+  *b = RoundToByte((y - 16) * 1.164 - (u - 128) * -2.112);
+}
+
+// BT.709 full range YUV to RGB reference
+static void YUVFToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
+  *r = RoundToByte(y - (v - 128) * -1.5748);
+  *g = RoundToByte(y - (u - 128) * 0.18732 - (v - 128) * 0.46812);
+  *b = RoundToByte(y - (u - 128) * -1.8556);
+}
+
+// BT.2020 limited range YUV to RGB reference
+static void YUVUToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
+  *r = RoundToByte((y - 16) * 1.164384 - (v - 128) * -1.67867);
+  *g = RoundToByte((y - 16) * 1.164384 - (u - 128) * 0.187326 -
+                   (v - 128) * 0.65042);
+  *b = RoundToByte((y - 16) * 1.164384 - (u - 128) * -2.14177);
+}
+
+// BT.2020 full range YUV to RGB reference
+static void YUVVToRGBReference(int y, int u, int v, int* r, int* g, int* b) {
+  *r = RoundToByte(y + (v - 128) * 1.474600);
+  *g = RoundToByte(y - (u - 128) * 0.164553 - (v - 128) * 0.571353);
+  *b = RoundToByte(y + (u - 128) * 1.881400);
+}
+
 TEST_F(LibYUVColorTest, TestYUV) {
   int r0, g0, b0, r1, g1, b1;
 
@@ -370,7 +502,11 @@ TEST_F(LibYUVColorTest, TestYUV) {
   YUVToRGB(240, 0, 0, &r1, &g1, &b1);
   EXPECT_EQ(57, r1);
   EXPECT_EQ(255, g1);
+#ifdef LIBYUV_UNLIMITED_DATA
+  EXPECT_EQ(3, b1);
+#else
   EXPECT_EQ(5, b1);
+#endif
 
   for (int i = 0; i < 256; ++i) {
     YUVToRGBReference(i, 128, 128, &r0, &g0, &b0);
@@ -444,28 +580,28 @@ TEST_F(LibYUVColorTest, TestGreyYUV) {
 
 static void PrintHistogram(int rh[256], int gh[256], int bh[256]) {
   int i;
-  printf("hist");
+  printf("hist ");
   for (i = 0; i < 256; ++i) {
     if (rh[i] || gh[i] || bh[i]) {
-      printf("\t%8d", i - 128);
+      printf(" %8d", i - 128);
     }
   }
-  printf("\nred");
+  printf("\nred  ");
   for (i = 0; i < 256; ++i) {
     if (rh[i] || gh[i] || bh[i]) {
-      printf("\t%8d", rh[i]);
+      printf(" %8d", rh[i]);
     }
   }
   printf("\ngreen");
   for (i = 0; i < 256; ++i) {
     if (rh[i] || gh[i] || bh[i]) {
-      printf("\t%8d", gh[i]);
+      printf(" %8d", gh[i]);
     }
   }
-  printf("\nblue");
+  printf("\nblue ");
   for (i = 0; i < 256; ++i) {
     if (rh[i] || gh[i] || bh[i]) {
-      printf("\t%8d", bh[i]);
+      printf(" %8d", bh[i]);
     }
   }
   printf("\n");
@@ -473,7 +609,13 @@ static void PrintHistogram(int rh[256], int gh[256], int bh[256]) {
 
 // Step by 5 on inner loop goes from 0 to 255 inclusive.
 // Set to 1 for better converage.  3, 5 or 17 for faster testing.
+#ifdef DISABLE_SLOW_TESTS
 #define FASTSTEP 5
+#else
+#define FASTSTEP 1
+#endif
+
+// BT.601 limited range.
 TEST_F(LibYUVColorTest, TestFullYUV) {
   int rh[256] = {
       0,
@@ -503,6 +645,7 @@ TEST_F(LibYUVColorTest, TestFullYUV) {
   PrintHistogram(rh, gh, bh);
 }
 
+// BT.601 full range.
 TEST_F(LibYUVColorTest, TestFullYUVJ) {
   int rh[256] = {
       0,
@@ -520,9 +663,129 @@ TEST_F(LibYUVColorTest, TestFullYUVJ) {
         int y = RANDOM256(y2);
         YUVJToRGBReference(y, u, v, &r0, &g0, &b0);
         YUVJToRGB(y, u, v, &r1, &g1, &b1);
-        EXPECT_NEAR(r0, r1, 1);
-        EXPECT_NEAR(g0, g1, 1);
-        EXPECT_NEAR(b0, b1, 1);
+        EXPECT_NEAR(r0, r1, ERROR_R);
+        EXPECT_NEAR(g0, g1, ERROR_G);
+        EXPECT_NEAR(b0, b1, ERROR_B);
+        ++rh[r1 - r0 + 128];
+        ++gh[g1 - g0 + 128];
+        ++bh[b1 - b0 + 128];
+      }
+    }
+  }
+  PrintHistogram(rh, gh, bh);
+}
+
+// BT.709 limited range.
+TEST_F(LibYUVColorTest, TestFullYUVH) {
+  int rh[256] = {
+      0,
+  };
+  int gh[256] = {
+      0,
+  };
+  int bh[256] = {
+      0,
+  };
+  for (int u = 0; u < 256; ++u) {
+    for (int v = 0; v < 256; ++v) {
+      for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
+        int r0, g0, b0, r1, g1, b1;
+        int y = RANDOM256(y2);
+        YUVHToRGBReference(y, u, v, &r0, &g0, &b0);
+        YUVHToRGB(y, u, v, &r1, &g1, &b1);
+        EXPECT_NEAR(r0, r1, ERROR_R);
+        EXPECT_NEAR(g0, g1, ERROR_G);
+        EXPECT_NEAR(b0, b1, ERROR_B);
+        ++rh[r1 - r0 + 128];
+        ++gh[g1 - g0 + 128];
+        ++bh[b1 - b0 + 128];
+      }
+    }
+  }
+  PrintHistogram(rh, gh, bh);
+}
+
+// BT.709 full range.
+TEST_F(LibYUVColorTest, TestFullYUVF) {
+  int rh[256] = {
+      0,
+  };
+  int gh[256] = {
+      0,
+  };
+  int bh[256] = {
+      0,
+  };
+  for (int u = 0; u < 256; ++u) {
+    for (int v = 0; v < 256; ++v) {
+      for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
+        int r0, g0, b0, r1, g1, b1;
+        int y = RANDOM256(y2);
+        YUVFToRGBReference(y, u, v, &r0, &g0, &b0);
+        YUVFToRGB(y, u, v, &r1, &g1, &b1);
+        EXPECT_NEAR(r0, r1, ERROR_R);
+        EXPECT_NEAR(g0, g1, ERROR_G);
+        EXPECT_NEAR(b0, b1, ERROR_B);
+        ++rh[r1 - r0 + 128];
+        ++gh[g1 - g0 + 128];
+        ++bh[b1 - b0 + 128];
+      }
+    }
+  }
+  PrintHistogram(rh, gh, bh);
+}
+
+// BT.2020 limited range.
+TEST_F(LibYUVColorTest, TestFullYUVU) {
+  int rh[256] = {
+      0,
+  };
+  int gh[256] = {
+      0,
+  };
+  int bh[256] = {
+      0,
+  };
+  for (int u = 0; u < 256; ++u) {
+    for (int v = 0; v < 256; ++v) {
+      for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
+        int r0, g0, b0, r1, g1, b1;
+        int y = RANDOM256(y2);
+        YUVUToRGBReference(y, u, v, &r0, &g0, &b0);
+        YUVUToRGB(y, u, v, &r1, &g1, &b1);
+        EXPECT_NEAR(r0, r1, ERROR_R);
+        EXPECT_NEAR(g0, g1, ERROR_G);
+        EXPECT_NEAR(b0, b1, ERROR_B);
+        ++rh[r1 - r0 + 128];
+        ++gh[g1 - g0 + 128];
+        ++bh[b1 - b0 + 128];
+      }
+    }
+  }
+  PrintHistogram(rh, gh, bh);
+}
+
+// BT.2020 full range.
+TEST_F(LibYUVColorTest, TestFullYUVV) {
+  int rh[256] = {
+      0,
+  };
+  int gh[256] = {
+      0,
+  };
+  int bh[256] = {
+      0,
+  };
+  for (int u = 0; u < 256; ++u) {
+    for (int v = 0; v < 256; ++v) {
+      for (int y2 = 0; y2 < 256; y2 += FASTSTEP) {
+        int r0, g0, b0, r1, g1, b1;
+        int y = RANDOM256(y2);
+        YUVVToRGBReference(y, u, v, &r0, &g0, &b0);
+        YUVVToRGB(y, u, v, &r1, &g1, &b1);
+        EXPECT_NEAR(r0, r1, ERROR_R);
+        EXPECT_NEAR(g0, g1, 2);
+        EXPECT_NEAR(b0, b1, ERROR_B);
         ++rh[r1 - r0 + 128];
         ++gh[g1 - g0 + 128];
         ++bh[b1 - b0 + 128];
diff --git a/files/unit_test/compare_test.cc b/files/unit_test/compare_test.cc
index 136254e1..c29562cb 100644
--- a/files/unit_test/compare_test.cc
+++ b/files/unit_test/compare_test.cc
@@ -15,10 +15,13 @@
 #include "../unit_test/unit_test.h"
 #include "libyuv/basic_types.h"
 #include "libyuv/compare.h"
-#include "libyuv/compare_row.h" /* For HammingDistance_C */
 #include "libyuv/cpu_id.h"
 #include "libyuv/video_common.h"
 
+#ifdef ENABLE_ROW_TESTS
+#include "libyuv/compare_row.h" /* For HammingDistance_C */
+#endif
+
 namespace libyuv {
 
 // hash seed of 5381 recommended.
@@ -206,6 +209,7 @@ TEST_F(LibYUVCompareTest, BenchmarkARGBDetect_Unaligned) {
   free_aligned_buffer_page_end(src_a);
 }
 
+#ifdef ENABLE_ROW_TESTS
 TEST_F(LibYUVCompareTest, BenchmarkHammingDistance_Opt) {
   const int kMaxWidth = 4096 * 3;
   align_buffer_page_end(src_a, kMaxWidth);
@@ -340,7 +344,7 @@ static const int kMaxOptCount = (1 << (32 - 3)) - 64;  // 536870848
 
 TEST_F(LibYUVCompareTest, TestHammingDistance_Opt) {
   uint32_t h1 = 0;
-  const int kMaxWidth = (benchmark_width_ * benchmark_height_ + 31) & ~31;
+  const int kMaxWidth = (benchmark_width_ * benchmark_height_ + 63) & ~63;
   align_buffer_page_end(src_a, kMaxWidth);
   align_buffer_page_end(src_b, kMaxWidth);
   memset(src_a, 255u, kMaxWidth);
@@ -403,6 +407,7 @@ TEST_F(LibYUVCompareTest, TestHammingDistance_Opt) {
   free_aligned_buffer_page_end(src_a);
   free_aligned_buffer_page_end(src_b);
 }
+#endif  // ENABLE_ROW_TESTS
 
 TEST_F(LibYUVCompareTest, TestHammingDistance) {
   align_buffer_page_end(src_a, benchmark_width_ * benchmark_height_);
diff --git a/files/unit_test/convert_test.cc b/files/unit_test/convert_test.cc
index 32a4cd1c..1f975825 100644
--- a/files/unit_test/convert_test.cc
+++ b/files/unit_test/convert_test.cc
@@ -12,8 +12,6 @@
 #include <stdlib.h>
 #include <time.h>
 
-#include "libyuv/row.h" /* For ARGBToAR30Row_AVX2 */
-
 #include "libyuv/basic_types.h"
 #include "libyuv/compare.h"
 #include "libyuv/convert.h"
@@ -29,12 +27,19 @@
 #include "libyuv/rotate.h"
 #include "libyuv/video_common.h"
 
-#if defined(__arm__) || defined(__aarch64__)
-// arm version subsamples by summing 4 pixels then multiplying by matrix with
-// 4x smaller coefficients which are rounded to nearest integer.
-#define ARM_YUV_ERROR 4
-#else
-#define ARM_YUV_ERROR 0
+#ifdef ENABLE_ROW_TESTS
+#include "libyuv/row.h" /* For ARGBToAR30Row_AVX2 */
+#endif
+
+// Some functions fail on big endian. Enable these tests on all cpus except
+// PowerPC, but they are not optimized so disabled by default.
+#if !defined(DISABLE_SLOW_TESTS) && !defined(__powerpc__)
+#define LITTLE_ENDIAN_ONLY_TEST 1
+#endif
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+// SLOW TESTS are those that are unoptimized C code.
+// FULL TESTS are optimized but test many variations of the same code.
+#define ENABLE_FULL_TESTS
 #endif
 
 namespace libyuv {
@@ -49,19 +54,20 @@ namespace libyuv {
 
 #define TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,         \
                        SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,             \
-                       DST_SUBSAMP_X, DST_SUBSAMP_Y, W1280, N, NEG, OFF)      \
+                       DST_SUBSAMP_X, DST_SUBSAMP_Y, W1280, N, NEG, OFF,      \
+                       SRC_DEPTH)                                             \
   TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) {              \
     static_assert(SRC_BPC == 1 || SRC_BPC == 2, "SRC BPC unsupported");       \
     static_assert(DST_BPC == 1 || DST_BPC == 2, "DST BPC unsupported");       \
     static_assert(SRC_SUBSAMP_X == 1 || SRC_SUBSAMP_X == 2,                   \
-                  "DST SRC_SUBSAMP_X unsupported");                           \
+                  "SRC_SUBSAMP_X unsupported");                               \
     static_assert(SRC_SUBSAMP_Y == 1 || SRC_SUBSAMP_Y == 2,                   \
-                  "DST SRC_SUBSAMP_Y unsupported");                           \
+                  "SRC_SUBSAMP_Y unsupported");                               \
     static_assert(DST_SUBSAMP_X == 1 || DST_SUBSAMP_X == 2,                   \
-                  "DST DST_SUBSAMP_X unsupported");                           \
+                  "DST_SUBSAMP_X unsupported");                               \
     static_assert(DST_SUBSAMP_Y == 1 || DST_SUBSAMP_Y == 2,                   \
-                  "DST DST_SUBSAMP_Y unsupported");                           \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+                  "DST_SUBSAMP_Y unsupported");                               \
+    const int kWidth = W1280;                                                 \
     const int kHeight = benchmark_height_;                                    \
     const int kSrcHalfWidth = SUBSAMPLE(kWidth, SRC_SUBSAMP_X);               \
     const int kSrcHalfHeight = SUBSAMPLE(kHeight, SRC_SUBSAMP_Y);             \
@@ -81,6 +87,16 @@ namespace libyuv {
     MemRandomize(src_y + OFF, kWidth * kHeight * SRC_BPC);                    \
     MemRandomize(src_u + OFF, kSrcHalfWidth * kSrcHalfHeight * SRC_BPC);      \
     MemRandomize(src_v + OFF, kSrcHalfWidth * kSrcHalfHeight * SRC_BPC);      \
+    SRC_T* src_y_p = reinterpret_cast<SRC_T*>(src_y + OFF);                   \
+    SRC_T* src_u_p = reinterpret_cast<SRC_T*>(src_u + OFF);                   \
+    SRC_T* src_v_p = reinterpret_cast<SRC_T*>(src_v + OFF);                   \
+    for (int i = 0; i < kWidth * kHeight; ++i) {                              \
+      src_y_p[i] = src_y_p[i] & ((1 << SRC_DEPTH) - 1);                       \
+    }                                                                         \
+    for (int i = 0; i < kSrcHalfWidth * kSrcHalfHeight; ++i) {                \
+      src_u_p[i] = src_u_p[i] & ((1 << SRC_DEPTH) - 1);                       \
+      src_v_p[i] = src_v_p[i] & ((1 << SRC_DEPTH) - 1);                       \
+    }                                                                         \
     memset(dst_y_c, 1, kWidth* kHeight* DST_BPC);                             \
     memset(dst_u_c, 2, kDstHalfWidth* kDstHalfHeight* DST_BPC);               \
     memset(dst_v_c, 3, kDstHalfWidth* kDstHalfHeight* DST_BPC);               \
@@ -89,9 +105,7 @@ namespace libyuv {
     memset(dst_v_opt, 103, kDstHalfWidth* kDstHalfHeight* DST_BPC);           \
     MaskCpuFlags(disable_cpu_flags_);                                         \
     SRC_FMT_PLANAR##To##FMT_PLANAR(                                           \
-        reinterpret_cast<SRC_T*>(src_y + OFF), kWidth,                        \
-        reinterpret_cast<SRC_T*>(src_u + OFF), kSrcHalfWidth,                 \
-        reinterpret_cast<SRC_T*>(src_v + OFF), kSrcHalfWidth,                 \
+        src_y_p, kWidth, src_u_p, kSrcHalfWidth, src_v_p, kSrcHalfWidth,      \
         reinterpret_cast<DST_T*>(dst_y_c), kWidth,                            \
         reinterpret_cast<DST_T*>(dst_u_c), kDstHalfWidth,                     \
         reinterpret_cast<DST_T*>(dst_v_c), kDstHalfWidth, kWidth,             \
@@ -99,9 +113,7 @@ namespace libyuv {
     MaskCpuFlags(benchmark_cpu_info_);                                        \
     for (int i = 0; i < benchmark_iterations_; ++i) {                         \
       SRC_FMT_PLANAR##To##FMT_PLANAR(                                         \
-          reinterpret_cast<SRC_T*>(src_y + OFF), kWidth,                      \
-          reinterpret_cast<SRC_T*>(src_u + OFF), kSrcHalfWidth,               \
-          reinterpret_cast<SRC_T*>(src_v + OFF), kSrcHalfWidth,               \
+          src_y_p, kWidth, src_u_p, kSrcHalfWidth, src_v_p, kSrcHalfWidth,    \
           reinterpret_cast<DST_T*>(dst_y_opt), kWidth,                        \
           reinterpret_cast<DST_T*>(dst_u_opt), kDstHalfWidth,                 \
           reinterpret_cast<DST_T*>(dst_v_opt), kDstHalfWidth, kWidth,         \
@@ -127,41 +139,58 @@ namespace libyuv {
 
 #define TESTPLANARTOP(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,           \
                       SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,               \
-                      DST_SUBSAMP_X, DST_SUBSAMP_Y)                            \
+                      DST_SUBSAMP_X, DST_SUBSAMP_Y, SRC_DEPTH)                 \
   TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
                  FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X, DST_SUBSAMP_Y,     \
-                 benchmark_width_ - 4, _Any, +, 0)                             \
+                 benchmark_width_ + 1, _Any, +, 0, SRC_DEPTH)                  \
   TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
                  FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X, DST_SUBSAMP_Y,     \
-                 benchmark_width_, _Unaligned, +, 1)                           \
+                 benchmark_width_, _Unaligned, +, 2, SRC_DEPTH)                \
   TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
                  FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X, DST_SUBSAMP_Y,     \
-                 benchmark_width_, _Invert, -, 0)                              \
+                 benchmark_width_, _Invert, -, 0, SRC_DEPTH)                   \
   TESTPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
                  FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X, DST_SUBSAMP_Y,     \
-                 benchmark_width_, _Opt, +, 0)
-
-TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I420, uint8_t, 1, 2, 2)
-TESTPLANARTOP(I422, uint8_t, 1, 2, 1, I420, uint8_t, 1, 2, 2)
-TESTPLANARTOP(I444, uint8_t, 1, 1, 1, I420, uint8_t, 1, 2, 2)
-TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I422, uint8_t, 1, 2, 1)
-TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I444, uint8_t, 1, 1, 1)
-TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I420Mirror, uint8_t, 1, 2, 2)
-TESTPLANARTOP(I422, uint8_t, 1, 2, 1, I422, uint8_t, 1, 2, 1)
-TESTPLANARTOP(I444, uint8_t, 1, 1, 1, I444, uint8_t, 1, 1, 1)
-TESTPLANARTOP(I010, uint16_t, 2, 2, 2, I010, uint16_t, 2, 2, 2)
-TESTPLANARTOP(I010, uint16_t, 2, 2, 2, I420, uint8_t, 1, 2, 2)
-TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I010, uint16_t, 2, 2, 2)
-TESTPLANARTOP(H010, uint16_t, 2, 2, 2, H010, uint16_t, 2, 2, 2)
-TESTPLANARTOP(H010, uint16_t, 2, 2, 2, H420, uint8_t, 1, 2, 2)
-TESTPLANARTOP(H420, uint8_t, 1, 2, 2, H010, uint16_t, 2, 2, 2)
+                 benchmark_width_, _Opt, +, 0, SRC_DEPTH)
+
+TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I420, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOP(I422, uint8_t, 1, 2, 1, I420, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOP(I444, uint8_t, 1, 1, 1, I420, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I422, uint8_t, 1, 2, 1, 8)
+TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I444, uint8_t, 1, 1, 1, 8)
+TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I420Mirror, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOP(I422, uint8_t, 1, 2, 1, I422, uint8_t, 1, 2, 1, 8)
+TESTPLANARTOP(I422, uint8_t, 1, 2, 1, I444, uint8_t, 1, 1, 1, 8)
+TESTPLANARTOP(I444, uint8_t, 1, 1, 1, I444, uint8_t, 1, 1, 1, 8)
+TESTPLANARTOP(I010, uint16_t, 2, 2, 2, I010, uint16_t, 2, 2, 2, 10)
+TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I010, uint16_t, 2, 2, 2, 8)
+TESTPLANARTOP(I420, uint8_t, 1, 2, 2, I012, uint16_t, 2, 2, 2, 8)
+TESTPLANARTOP(H010, uint16_t, 2, 2, 2, H010, uint16_t, 2, 2, 2, 10)
+TESTPLANARTOP(H010, uint16_t, 2, 2, 2, H420, uint8_t, 1, 2, 2, 10)
+TESTPLANARTOP(H420, uint8_t, 1, 2, 2, H010, uint16_t, 2, 2, 2, 8)
+TESTPLANARTOP(H420, uint8_t, 1, 2, 2, H012, uint16_t, 2, 2, 2, 8)
+TESTPLANARTOP(I010, uint16_t, 2, 2, 2, I410, uint16_t, 2, 1, 1, 10)
+TESTPLANARTOP(I210, uint16_t, 2, 2, 1, I410, uint16_t, 2, 1, 1, 10)
+TESTPLANARTOP(I012, uint16_t, 2, 2, 2, I412, uint16_t, 2, 1, 1, 12)
+TESTPLANARTOP(I212, uint16_t, 2, 2, 1, I412, uint16_t, 2, 1, 1, 12)
+TESTPLANARTOP(I410, uint16_t, 2, 1, 1, I010, uint16_t, 2, 2, 2, 10)
+TESTPLANARTOP(I210, uint16_t, 2, 2, 1, I010, uint16_t, 2, 2, 2, 10)
+TESTPLANARTOP(I412, uint16_t, 2, 1, 1, I012, uint16_t, 2, 2, 2, 12)
+TESTPLANARTOP(I212, uint16_t, 2, 2, 1, I012, uint16_t, 2, 2, 2, 12)
+TESTPLANARTOP(I010, uint16_t, 2, 2, 2, I420, uint8_t, 1, 2, 2, 10)
+TESTPLANARTOP(I210, uint16_t, 2, 2, 1, I420, uint8_t, 1, 2, 2, 10)
+TESTPLANARTOP(I210, uint16_t, 2, 2, 1, I422, uint8_t, 1, 2, 1, 10)
+TESTPLANARTOP(I410, uint16_t, 2, 1, 1, I444, uint8_t, 1, 1, 1, 10)
+TESTPLANARTOP(I012, uint16_t, 2, 2, 2, I420, uint8_t, 1, 2, 2, 12)
+TESTPLANARTOP(I212, uint16_t, 2, 2, 1, I422, uint8_t, 1, 2, 1, 12)
+TESTPLANARTOP(I412, uint16_t, 2, 1, 1, I444, uint8_t, 1, 1, 1, 12)
 
 // Test Android 420 to I420
 #define TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X,          \
                         SRC_SUBSAMP_Y, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
                         W1280, N, NEG, OFF, PN, OFF_U, OFF_V)                 \
-  TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##_##PN##N) {       \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+  TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##To##PN##N) {      \
+    const int kWidth = W1280;                                                 \
     const int kHeight = benchmark_height_;                                    \
     const int kSizeUV =                                                       \
         SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); \
@@ -216,41 +245,23 @@ TESTPLANARTOP(H420, uint8_t, 1, 2, 2, H010, uint16_t, 2, 2, 2)
           dst_y_opt, kWidth, dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X),         \
           dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight);      \
     }                                                                         \
-    int max_diff = 0;                                                         \
     for (int i = 0; i < kHeight; ++i) {                                       \
       for (int j = 0; j < kWidth; ++j) {                                      \
-        int abs_diff = abs(static_cast<int>(dst_y_c[i * kWidth + j]) -        \
-                           static_cast<int>(dst_y_opt[i * kWidth + j]));      \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
+        EXPECT_EQ(dst_y_c[i * kWidth + j], dst_y_opt[i * kWidth + j]);        \
       }                                                                       \
     }                                                                         \
-    EXPECT_EQ(0, max_diff);                                                   \
     for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                 \
       for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) {                \
-        int abs_diff = abs(                                                   \
-            static_cast<int>(dst_u_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) - \
-            static_cast<int>(                                                 \
-                dst_u_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]));            \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
+        EXPECT_EQ(dst_u_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j],              \
+                  dst_u_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]);           \
       }                                                                       \
     }                                                                         \
-    EXPECT_LE(max_diff, 3);                                                   \
     for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                 \
       for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) {                \
-        int abs_diff = abs(                                                   \
-            static_cast<int>(dst_v_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) - \
-            static_cast<int>(                                                 \
-                dst_v_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]));            \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
+        EXPECT_EQ(dst_v_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j],              \
+                  dst_v_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]);           \
       }                                                                       \
     }                                                                         \
-    EXPECT_LE(max_diff, 3);                                                   \
     free_aligned_buffer_page_end(dst_y_c);                                    \
     free_aligned_buffer_page_end(dst_u_c);                                    \
     free_aligned_buffer_page_end(dst_v_c);                                    \
@@ -265,11 +276,11 @@ TESTPLANARTOP(H420, uint8_t, 1, 2, 2, H010, uint16_t, 2, 2, 2)
                        SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, SUBSAMP_X,    \
                        SUBSAMP_Y)                                              \
   TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
-                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_ - 4,      \
+                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_ + 1,      \
                   _Any, +, 0, PN, OFF_U, OFF_V)                                \
   TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
                   FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_,          \
-                  _Unaligned, +, 1, PN, OFF_U, OFF_V)                          \
+                  _Unaligned, +, 2, PN, OFF_U, OFF_V)                          \
   TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
                   FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Invert, \
                   -, 0, PN, OFF_U, OFF_V)                                      \
@@ -280,6 +291,8 @@ TESTPLANARTOP(H420, uint8_t, 1, 2, 2, H010, uint16_t, 2, 2, 2)
 TESTAPLANARTOP(Android420, I420, 1, 0, 0, 2, 2, I420, 2, 2)
 TESTAPLANARTOP(Android420, NV12, 2, 0, 1, 2, 2, I420, 2, 2)
 TESTAPLANARTOP(Android420, NV21, 2, 1, 0, 2, 2, I420, 2, 2)
+#undef TESTAPLANARTOP
+#undef TESTAPLANARTOPI
 
 // wrapper to keep API the same
 int I400ToNV21(const uint8_t* src_y,
@@ -298,77 +311,75 @@ int I400ToNV21(const uint8_t* src_y,
                     dst_stride_vu, width, height);
 }
 
-#define TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,         \
-                        FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF) \
+#define TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,        \
+                        SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,            \
+                        DST_SUBSAMP_X, DST_SUBSAMP_Y, W1280, N, NEG, OFF,     \
+                        SRC_DEPTH)                                            \
   TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) {              \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+    static_assert(SRC_BPC == 1 || SRC_BPC == 2, "SRC BPC unsupported");       \
+    static_assert(DST_BPC == 1 || DST_BPC == 2, "DST BPC unsupported");       \
+    static_assert(SRC_SUBSAMP_X == 1 || SRC_SUBSAMP_X == 2,                   \
+                  "SRC_SUBSAMP_X unsupported");                               \
+    static_assert(SRC_SUBSAMP_Y == 1 || SRC_SUBSAMP_Y == 2,                   \
+                  "SRC_SUBSAMP_Y unsupported");                               \
+    static_assert(DST_SUBSAMP_X == 1 || DST_SUBSAMP_X == 2,                   \
+                  "DST_SUBSAMP_X unsupported");                               \
+    static_assert(DST_SUBSAMP_Y == 1 || DST_SUBSAMP_Y == 2,                   \
+                  "DST_SUBSAMP_Y unsupported");                               \
+    const int kWidth = W1280;                                                 \
     const int kHeight = benchmark_height_;                                    \
-    align_buffer_page_end(src_y, kWidth* kHeight + OFF);                      \
-    align_buffer_page_end(src_u, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) *           \
-                                         SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) +  \
-                                     OFF);                                    \
-    align_buffer_page_end(src_v, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) *           \
-                                         SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) +  \
-                                     OFF);                                    \
-    align_buffer_page_end(dst_y_c, kWidth* kHeight);                          \
-    align_buffer_page_end(dst_uv_c, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 *        \
-                                        SUBSAMPLE(kHeight, SUBSAMP_Y));       \
-    align_buffer_page_end(dst_y_opt, kWidth* kHeight);                        \
-    align_buffer_page_end(dst_uv_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 *      \
-                                          SUBSAMPLE(kHeight, SUBSAMP_Y));     \
-    for (int i = 0; i < kHeight; ++i)                                         \
-      for (int j = 0; j < kWidth; ++j)                                        \
-        src_y[i * kWidth + j + OFF] = (fastrand() & 0xff);                    \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) {             \
-      for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) {            \
-        src_u[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] =             \
-            (fastrand() & 0xff);                                              \
-        src_v[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] =             \
-            (fastrand() & 0xff);                                              \
-      }                                                                       \
+    const int kSrcHalfWidth = SUBSAMPLE(kWidth, SRC_SUBSAMP_X);               \
+    const int kSrcHalfHeight = SUBSAMPLE(kHeight, SRC_SUBSAMP_Y);             \
+    const int kDstHalfWidth = SUBSAMPLE(kWidth, DST_SUBSAMP_X);               \
+    const int kDstHalfHeight = SUBSAMPLE(kHeight, DST_SUBSAMP_Y);             \
+    align_buffer_page_end(src_y, kWidth* kHeight* SRC_BPC + OFF);             \
+    align_buffer_page_end(src_u,                                              \
+                          kSrcHalfWidth* kSrcHalfHeight* SRC_BPC + OFF);      \
+    align_buffer_page_end(src_v,                                              \
+                          kSrcHalfWidth* kSrcHalfHeight* SRC_BPC + OFF);      \
+    align_buffer_page_end(dst_y_c, kWidth* kHeight* DST_BPC);                 \
+    align_buffer_page_end(dst_uv_c,                                           \
+                          kDstHalfWidth* kDstHalfHeight* DST_BPC * 2);        \
+    align_buffer_page_end(dst_y_opt, kWidth* kHeight* DST_BPC);               \
+    align_buffer_page_end(dst_uv_opt,                                         \
+                          kDstHalfWidth* kDstHalfHeight* DST_BPC * 2);        \
+    MemRandomize(src_y + OFF, kWidth * kHeight * SRC_BPC);                    \
+    MemRandomize(src_u + OFF, kSrcHalfWidth * kSrcHalfHeight * SRC_BPC);      \
+    MemRandomize(src_v + OFF, kSrcHalfWidth * kSrcHalfHeight * SRC_BPC);      \
+    SRC_T* src_y_p = reinterpret_cast<SRC_T*>(src_y + OFF);                   \
+    SRC_T* src_u_p = reinterpret_cast<SRC_T*>(src_u + OFF);                   \
+    SRC_T* src_v_p = reinterpret_cast<SRC_T*>(src_v + OFF);                   \
+    for (int i = 0; i < kWidth * kHeight; ++i) {                              \
+      src_y_p[i] = src_y_p[i] & ((1 << SRC_DEPTH) - 1);                       \
     }                                                                         \
-    memset(dst_y_c, 1, kWidth* kHeight);                                      \
-    memset(dst_uv_c, 2,                                                       \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
-    memset(dst_y_opt, 101, kWidth* kHeight);                                  \
-    memset(dst_uv_opt, 102,                                                   \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
+    for (int i = 0; i < kSrcHalfWidth * kSrcHalfHeight; ++i) {                \
+      src_u_p[i] = src_u_p[i] & ((1 << SRC_DEPTH) - 1);                       \
+      src_v_p[i] = src_v_p[i] & ((1 << SRC_DEPTH) - 1);                       \
+    }                                                                         \
+    memset(dst_y_c, 1, kWidth* kHeight* DST_BPC);                             \
+    memset(dst_uv_c, 2, kDstHalfWidth* kDstHalfHeight* DST_BPC * 2);          \
+    memset(dst_y_opt, 101, kWidth* kHeight* DST_BPC);                         \
+    memset(dst_uv_opt, 102, kDstHalfWidth* kDstHalfHeight* DST_BPC * 2);      \
     MaskCpuFlags(disable_cpu_flags_);                                         \
-    SRC_FMT_PLANAR##To##FMT_PLANAR(                                           \
-        src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X),   \
-        src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_c, kWidth,       \
-        dst_uv_c, SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight);     \
+    SRC_FMT_PLANAR##To##FMT_PLANAR(src_y_p, kWidth, src_u_p, kSrcHalfWidth,   \
+                                   src_v_p, kSrcHalfWidth,                    \
+                                   reinterpret_cast<DST_T*>(dst_y_c), kWidth, \
+                                   reinterpret_cast<DST_T*>(dst_uv_c),        \
+                                   kDstHalfWidth * 2, kWidth, NEG kHeight);   \
     MaskCpuFlags(benchmark_cpu_info_);                                        \
     for (int i = 0; i < benchmark_iterations_; ++i) {                         \
       SRC_FMT_PLANAR##To##FMT_PLANAR(                                         \
-          src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \
-          src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_opt, kWidth,   \
-          dst_uv_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight); \
+          src_y_p, kWidth, src_u_p, kSrcHalfWidth, src_v_p, kSrcHalfWidth,    \
+          reinterpret_cast<DST_T*>(dst_y_opt), kWidth,                        \
+          reinterpret_cast<DST_T*>(dst_uv_opt), kDstHalfWidth * 2, kWidth,    \
+          NEG kHeight);                                                       \
     }                                                                         \
-    int max_diff = 0;                                                         \
-    for (int i = 0; i < kHeight; ++i) {                                       \
-      for (int j = 0; j < kWidth; ++j) {                                      \
-        int abs_diff = abs(static_cast<int>(dst_y_c[i * kWidth + j]) -        \
-                           static_cast<int>(dst_y_opt[i * kWidth + j]));      \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
-      }                                                                       \
+    for (int i = 0; i < kHeight * kWidth * DST_BPC; ++i) {                    \
+      EXPECT_EQ(dst_y_c[i], dst_y_opt[i]);                                    \
     }                                                                         \
-    EXPECT_LE(max_diff, 1);                                                   \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                 \
-      for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X) * 2; ++j) {            \
-        int abs_diff =                                                        \
-            abs(static_cast<int>(                                             \
-                    dst_uv_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j]) -     \
-                static_cast<int>(                                             \
-                    dst_uv_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j]));   \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
-      }                                                                       \
+    for (int i = 0; i < kDstHalfWidth * kDstHalfHeight * DST_BPC * 2; ++i) {  \
+      EXPECT_EQ(dst_uv_c[i], dst_uv_opt[i]);                                  \
     }                                                                         \
-    EXPECT_LE(max_diff, 1);                                                   \
     free_aligned_buffer_page_end(dst_y_c);                                    \
     free_aligned_buffer_page_end(dst_uv_c);                                   \
     free_aligned_buffer_page_end(dst_y_opt);                                  \
@@ -378,92 +389,110 @@ int I400ToNV21(const uint8_t* src_y,
     free_aligned_buffer_page_end(src_v);                                      \
   }
 
-#define TESTPLANARTOBP(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,        \
-                       FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y)                    \
-  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                  SUBSAMP_X, SUBSAMP_Y, benchmark_width_ - 4, _Any, +, 0)   \
-  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                  SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Unaligned, +, 1) \
-  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                  SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Invert, -, 0)    \
-  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                  SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, 0)
-
-TESTPLANARTOBP(I420, 2, 2, NV12, 2, 2)
-TESTPLANARTOBP(I420, 2, 2, NV21, 2, 2)
-TESTPLANARTOBP(I422, 2, 1, NV21, 2, 2)
-TESTPLANARTOBP(I444, 1, 1, NV21, 2, 2)
-TESTPLANARTOBP(I400, 2, 2, NV21, 2, 2)
-
-#define TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,        \
-                          FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG,     \
-                          OFF)                                                 \
+#define TESTPLANARTOBP(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,         \
+                       SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,             \
+                       DST_SUBSAMP_X, DST_SUBSAMP_Y, SRC_DEPTH)               \
+  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                  SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                  DST_SUBSAMP_Y, benchmark_width_ + 1, _Any, +, 0, SRC_DEPTH) \
+  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                  SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                  DST_SUBSAMP_Y, benchmark_width_, _Unaligned, +, 2,          \
+                  SRC_DEPTH)                                                  \
+  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                  SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                  DST_SUBSAMP_Y, benchmark_width_, _Invert, -, 0, SRC_DEPTH)  \
+  TESTPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                  SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                  DST_SUBSAMP_Y, benchmark_width_, _Opt, +, 0, SRC_DEPTH)
+
+TESTPLANARTOBP(I420, uint8_t, 1, 2, 2, NV12, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOBP(I420, uint8_t, 1, 2, 2, NV21, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOBP(I422, uint8_t, 1, 2, 1, NV21, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOBP(I444, uint8_t, 1, 1, 1, NV12, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOBP(I444, uint8_t, 1, 1, 1, NV21, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOBP(I400, uint8_t, 1, 2, 2, NV21, uint8_t, 1, 2, 2, 8)
+TESTPLANARTOBP(I010, uint16_t, 2, 2, 2, P010, uint16_t, 2, 2, 2, 10)
+TESTPLANARTOBP(I210, uint16_t, 2, 2, 1, P210, uint16_t, 2, 2, 1, 10)
+TESTPLANARTOBP(I012, uint16_t, 2, 2, 2, P012, uint16_t, 2, 2, 2, 12)
+TESTPLANARTOBP(I212, uint16_t, 2, 2, 1, P212, uint16_t, 2, 2, 1, 12)
+
+#define TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,       \
+                          SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,           \
+                          DST_SUBSAMP_X, DST_SUBSAMP_Y, W1280, N, NEG, OFF,    \
+                          DOY, SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)             \
   TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) {               \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+    static_assert(SRC_BPC == 1 || SRC_BPC == 2, "SRC BPC unsupported");        \
+    static_assert(DST_BPC == 1 || DST_BPC == 2, "DST BPC unsupported");        \
+    static_assert(SRC_SUBSAMP_X == 1 || SRC_SUBSAMP_X == 2,                    \
+                  "SRC_SUBSAMP_X unsupported");                                \
+    static_assert(SRC_SUBSAMP_Y == 1 || SRC_SUBSAMP_Y == 2,                    \
+                  "SRC_SUBSAMP_Y unsupported");                                \
+    static_assert(DST_SUBSAMP_X == 1 || DST_SUBSAMP_X == 2,                    \
+                  "DST_SUBSAMP_X unsupported");                                \
+    static_assert(DST_SUBSAMP_Y == 1 || DST_SUBSAMP_Y == 2,                    \
+                  "DST_SUBSAMP_Y unsupported");                                \
+    const int kWidth = W1280;                                                  \
     const int kHeight = benchmark_height_;                                     \
-    align_buffer_page_end(src_y, kWidth* kHeight + OFF);                       \
-    align_buffer_page_end(src_uv, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * 2 *       \
-                                          SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) +  \
-                                      OFF);                                    \
-    align_buffer_page_end(dst_y_c, kWidth* kHeight);                           \
-    align_buffer_page_end(dst_uv_c, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 *         \
-                                        SUBSAMPLE(kHeight, SUBSAMP_Y));        \
-    align_buffer_page_end(dst_y_opt, kWidth* kHeight);                         \
-    align_buffer_page_end(dst_uv_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 *       \
-                                          SUBSAMPLE(kHeight, SUBSAMP_Y));      \
-    for (int i = 0; i < kHeight; ++i)                                          \
-      for (int j = 0; j < kWidth; ++j)                                         \
-        src_y[i * kWidth + j + OFF] = (fastrand() & 0xff);                     \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) {              \
-      for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) {             \
-        src_uv[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) * 2 + j + 0 + OFF] =     \
-            (fastrand() & 0xff);                                               \
-        src_uv[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) * 2 + j + 1 + OFF] =     \
-            (fastrand() & 0xff);                                               \
-      }                                                                        \
+    const int kSrcHalfWidth = SUBSAMPLE(kWidth, SRC_SUBSAMP_X);                \
+    const int kDstHalfWidth = SUBSAMPLE(kWidth, DST_SUBSAMP_X);                \
+    const int kDstHalfHeight = SUBSAMPLE(kHeight, DST_SUBSAMP_Y);              \
+    const int kPaddedWidth = (kWidth + (TILE_WIDTH - 1)) & ~(TILE_WIDTH - 1);  \
+    const int kPaddedHeight =                                                  \
+        (kHeight + (TILE_HEIGHT - 1)) & ~(TILE_HEIGHT - 1);                    \
+    const int kSrcHalfPaddedWidth = SUBSAMPLE(kPaddedWidth, SRC_SUBSAMP_X);    \
+    const int kSrcHalfPaddedHeight = SUBSAMPLE(kPaddedHeight, SRC_SUBSAMP_Y);  \
+    align_buffer_page_end(src_y, kPaddedWidth* kPaddedHeight* SRC_BPC + OFF);  \
+    align_buffer_page_end(                                                     \
+        src_uv,                                                                \
+        2 * kSrcHalfPaddedWidth * kSrcHalfPaddedHeight * SRC_BPC + OFF);       \
+    align_buffer_page_end(dst_y_c, kWidth* kHeight* DST_BPC);                  \
+    align_buffer_page_end(dst_uv_c,                                            \
+                          2 * kDstHalfWidth * kDstHalfHeight * DST_BPC);       \
+    align_buffer_page_end(dst_y_opt, kWidth* kHeight* DST_BPC);                \
+    align_buffer_page_end(dst_uv_opt,                                          \
+                          2 * kDstHalfWidth * kDstHalfHeight * DST_BPC);       \
+    SRC_T* src_y_p = reinterpret_cast<SRC_T*>(src_y + OFF);                    \
+    SRC_T* src_uv_p = reinterpret_cast<SRC_T*>(src_uv + OFF);                  \
+    for (int i = 0; i < kPaddedWidth * kPaddedHeight; ++i) {                   \
+      src_y_p[i] =                                                             \
+          (fastrand() & (((SRC_T)(-1)) << ((8 * SRC_BPC) - SRC_DEPTH)));       \
     }                                                                          \
-    memset(dst_y_c, 1, kWidth* kHeight);                                       \
-    memset(dst_uv_c, 2,                                                        \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y));  \
-    memset(dst_y_opt, 101, kWidth* kHeight);                                   \
-    memset(dst_uv_opt, 102,                                                    \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y));  \
+    for (int i = 0; i < kSrcHalfPaddedWidth * kSrcHalfPaddedHeight * 2; ++i) { \
+      src_uv_p[i] =                                                            \
+          (fastrand() & (((SRC_T)(-1)) << ((8 * SRC_BPC) - SRC_DEPTH)));       \
+    }                                                                          \
+    memset(dst_y_c, 1, kWidth* kHeight* DST_BPC);                              \
+    memset(dst_uv_c, 2, 2 * kDstHalfWidth * kDstHalfHeight * DST_BPC);         \
+    memset(dst_y_opt, 101, kWidth* kHeight* DST_BPC);                          \
+    memset(dst_uv_opt, 102, 2 * kDstHalfWidth * kDstHalfHeight * DST_BPC);     \
     MaskCpuFlags(disable_cpu_flags_);                                          \
     SRC_FMT_PLANAR##To##FMT_PLANAR(                                            \
-        src_y + OFF, kWidth, src_uv + OFF,                                     \
-        SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * 2, dst_y_c, kWidth, dst_uv_c,       \
-        SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight);                \
+        src_y_p, kWidth, src_uv_p, 2 * kSrcHalfWidth,                          \
+        DOY ? reinterpret_cast<DST_T*>(dst_y_c) : NULL, kWidth,                \
+        reinterpret_cast<DST_T*>(dst_uv_c), 2 * kDstHalfWidth, kWidth,         \
+        NEG kHeight);                                                          \
     MaskCpuFlags(benchmark_cpu_info_);                                         \
     for (int i = 0; i < benchmark_iterations_; ++i) {                          \
       SRC_FMT_PLANAR##To##FMT_PLANAR(                                          \
-          src_y + OFF, kWidth, src_uv + OFF,                                   \
-          SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * 2, dst_y_opt, kWidth, dst_uv_opt, \
-          SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight);              \
+          src_y_p, kWidth, src_uv_p, 2 * kSrcHalfWidth,                        \
+          DOY ? reinterpret_cast<DST_T*>(dst_y_opt) : NULL, kWidth,            \
+          reinterpret_cast<DST_T*>(dst_uv_opt), 2 * kDstHalfWidth, kWidth,     \
+          NEG kHeight);                                                        \
     }                                                                          \
-    int max_diff = 0;                                                          \
-    for (int i = 0; i < kHeight; ++i) {                                        \
-      for (int j = 0; j < kWidth; ++j) {                                       \
-        int abs_diff = abs(static_cast<int>(dst_y_c[i * kWidth + j]) -         \
-                           static_cast<int>(dst_y_opt[i * kWidth + j]));       \
-        if (abs_diff > max_diff) {                                             \
-          max_diff = abs_diff;                                                 \
+    if (DOY) {                                                                 \
+      for (int i = 0; i < kHeight; ++i) {                                      \
+        for (int j = 0; j < kWidth; ++j) {                                     \
+          EXPECT_EQ(dst_y_c[i * kWidth + j], dst_y_opt[i * kWidth + j]);       \
         }                                                                      \
       }                                                                        \
     }                                                                          \
-    EXPECT_LE(max_diff, 1);                                                    \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                  \
-      for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X) * 2; ++j) {             \
-        int abs_diff =                                                         \
-            abs(static_cast<int>(                                              \
-                    dst_uv_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j]) -      \
-                static_cast<int>(                                              \
-                    dst_uv_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j]));    \
-        if (abs_diff > max_diff) {                                             \
-          max_diff = abs_diff;                                                 \
-        }                                                                      \
+    for (int i = 0; i < kDstHalfHeight; ++i) {                                 \
+      for (int j = 0; j < 2 * kDstHalfWidth; ++j) {                            \
+        EXPECT_EQ(dst_uv_c[i * 2 * kDstHalfWidth + j],                         \
+                  dst_uv_opt[i * 2 * kDstHalfWidth + j]);                      \
       }                                                                        \
     }                                                                          \
-    EXPECT_LE(max_diff, 1);                                                    \
     free_aligned_buffer_page_end(dst_y_c);                                     \
     free_aligned_buffer_page_end(dst_uv_c);                                    \
     free_aligned_buffer_page_end(dst_y_opt);                                   \
@@ -472,112 +501,116 @@ TESTPLANARTOBP(I400, 2, 2, NV21, 2, 2)
     free_aligned_buffer_page_end(src_uv);                                      \
   }
 
-#define TESTBIPLANARTOBP(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,        \
-                         FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y)                    \
-  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                    SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, 0)       \
-  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                    SUBSAMP_X, SUBSAMP_Y, benchmark_width_ - 4, _Any, +, 0)   \
-  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                    SUBSAMP_X, SUBSAMP_Y, benchmark_width, _Unaligned, +, 1)  \
-  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                    SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Invert, -, 0)    \
-  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
-                    SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, 0)
-
-// TODO(fbarchard): Fix msan on this unittest
-// TESTBIPLANARTOBP(NV21, 2, 2, NV12, 2, 2)
-
-#define TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,         \
-                         FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF, \
-                         DOY)                                                  \
+#define TESTBIPLANARTOBP(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,        \
+                         SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,            \
+                         DST_SUBSAMP_X, DST_SUBSAMP_Y, SRC_DEPTH, TILE_WIDTH,  \
+                         TILE_HEIGHT)                                          \
+  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,             \
+                    SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,  \
+                    DST_SUBSAMP_Y, benchmark_width_ + 1, _Any, +, 0, 1,        \
+                    SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)                        \
+  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,             \
+                    SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,  \
+                    DST_SUBSAMP_Y, benchmark_width_, _Unaligned, +, 2, 1,      \
+                    SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)                        \
+  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,             \
+                    SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,  \
+                    DST_SUBSAMP_Y, benchmark_width_, _Invert, -, 0, 1,         \
+                    SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)                        \
+  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,             \
+                    SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,  \
+                    DST_SUBSAMP_Y, benchmark_width_, _Opt, +, 0, 1, SRC_DEPTH, \
+                    TILE_WIDTH, TILE_HEIGHT)                                   \
+  TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,             \
+                    SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,  \
+                    DST_SUBSAMP_Y, benchmark_width_, _NullY, +, 0, 0,          \
+                    SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)
+
+TESTBIPLANARTOBP(NV21, uint8_t, 1, 2, 2, NV12, uint8_t, 1, 2, 2, 8, 1, 1)
+TESTBIPLANARTOBP(NV12, uint8_t, 1, 2, 2, NV12Mirror, uint8_t, 1, 2, 2, 8, 1, 1)
+TESTBIPLANARTOBP(NV12, uint8_t, 1, 2, 2, NV24, uint8_t, 1, 1, 1, 8, 1, 1)
+TESTBIPLANARTOBP(NV16, uint8_t, 1, 2, 1, NV24, uint8_t, 1, 1, 1, 8, 1, 1)
+TESTBIPLANARTOBP(P010, uint16_t, 2, 2, 2, P410, uint16_t, 2, 1, 1, 10, 1, 1)
+TESTBIPLANARTOBP(P210, uint16_t, 2, 2, 1, P410, uint16_t, 2, 1, 1, 10, 1, 1)
+TESTBIPLANARTOBP(P012, uint16_t, 2, 2, 2, P412, uint16_t, 2, 1, 1, 10, 1, 1)
+TESTBIPLANARTOBP(P212, uint16_t, 2, 2, 1, P412, uint16_t, 2, 1, 1, 12, 1, 1)
+TESTBIPLANARTOBP(P016, uint16_t, 2, 2, 2, P416, uint16_t, 2, 1, 1, 12, 1, 1)
+TESTBIPLANARTOBP(P216, uint16_t, 2, 2, 1, P416, uint16_t, 2, 1, 1, 12, 1, 1)
+TESTBIPLANARTOBP(MM21, uint8_t, 1, 2, 2, NV12, uint8_t, 1, 2, 2, 8, 16, 32)
+
+#define TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,        \
+                         SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,            \
+                         DST_SUBSAMP_X, DST_SUBSAMP_Y, W1280, N, NEG, OFF,     \
+                         SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)                   \
   TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) {               \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+    static_assert(SRC_BPC == 1 || SRC_BPC == 2, "SRC BPC unsupported");        \
+    static_assert(DST_BPC == 1 || DST_BPC == 2, "DST BPC unsupported");        \
+    static_assert(SRC_SUBSAMP_X == 1 || SRC_SUBSAMP_X == 2,                    \
+                  "SRC_SUBSAMP_X unsupported");                                \
+    static_assert(SRC_SUBSAMP_Y == 1 || SRC_SUBSAMP_Y == 2,                    \
+                  "SRC_SUBSAMP_Y unsupported");                                \
+    static_assert(DST_SUBSAMP_X == 1 || DST_SUBSAMP_X == 2,                    \
+                  "DST_SUBSAMP_X unsupported");                                \
+    static_assert(DST_SUBSAMP_Y == 1 || DST_SUBSAMP_Y == 2,                    \
+                  "DST_SUBSAMP_Y unsupported");                                \
+    const int kWidth = W1280;                                                  \
     const int kHeight = benchmark_height_;                                     \
-    align_buffer_page_end(src_y, kWidth* kHeight + OFF);                       \
-    align_buffer_page_end(src_uv, 2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X) *       \
-                                          SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) +  \
-                                      OFF);                                    \
-    align_buffer_page_end(dst_y_c, kWidth* kHeight);                           \
-    align_buffer_page_end(dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X) *              \
-                                       SUBSAMPLE(kHeight, SUBSAMP_Y));         \
-    align_buffer_page_end(dst_v_c, SUBSAMPLE(kWidth, SUBSAMP_X) *              \
-                                       SUBSAMPLE(kHeight, SUBSAMP_Y));         \
-    align_buffer_page_end(dst_y_opt, kWidth* kHeight);                         \
-    align_buffer_page_end(dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X) *            \
-                                         SUBSAMPLE(kHeight, SUBSAMP_Y));       \
-    align_buffer_page_end(dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X) *            \
-                                         SUBSAMPLE(kHeight, SUBSAMP_Y));       \
-    for (int i = 0; i < kHeight; ++i)                                          \
-      for (int j = 0; j < kWidth; ++j)                                         \
-        src_y[i * kWidth + j + OFF] = (fastrand() & 0xff);                     \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) {              \
-      for (int j = 0; j < 2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) {         \
-        src_uv[(i * 2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) + j + OFF] =         \
-            (fastrand() & 0xff);                                               \
-      }                                                                        \
+    const int kSrcHalfWidth = SUBSAMPLE(kWidth, SRC_SUBSAMP_X);                \
+    const int kDstHalfWidth = SUBSAMPLE(kWidth, DST_SUBSAMP_X);                \
+    const int kDstHalfHeight = SUBSAMPLE(kHeight, DST_SUBSAMP_Y);              \
+    const int kPaddedWidth = (kWidth + (TILE_WIDTH - 1)) & ~(TILE_WIDTH - 1);  \
+    const int kPaddedHeight =                                                  \
+        (kHeight + (TILE_HEIGHT - 1)) & ~(TILE_HEIGHT - 1);                    \
+    const int kSrcHalfPaddedWidth = SUBSAMPLE(kPaddedWidth, SRC_SUBSAMP_X);    \
+    const int kSrcHalfPaddedHeight = SUBSAMPLE(kPaddedHeight, SRC_SUBSAMP_Y);  \
+    align_buffer_page_end(src_y, kPaddedWidth* kPaddedHeight* SRC_BPC + OFF);  \
+    align_buffer_page_end(                                                     \
+        src_uv, kSrcHalfPaddedWidth* kSrcHalfPaddedHeight* SRC_BPC * 2 + OFF); \
+    align_buffer_page_end(dst_y_c, kWidth* kHeight* DST_BPC);                  \
+    align_buffer_page_end(dst_u_c, kDstHalfWidth* kDstHalfHeight* DST_BPC);    \
+    align_buffer_page_end(dst_v_c, kDstHalfWidth* kDstHalfHeight* DST_BPC);    \
+    align_buffer_page_end(dst_y_opt, kWidth* kHeight* DST_BPC);                \
+    align_buffer_page_end(dst_u_opt, kDstHalfWidth* kDstHalfHeight* DST_BPC);  \
+    align_buffer_page_end(dst_v_opt, kDstHalfWidth* kDstHalfHeight* DST_BPC);  \
+    SRC_T* src_y_p = reinterpret_cast<SRC_T*>(src_y + OFF);                    \
+    SRC_T* src_uv_p = reinterpret_cast<SRC_T*>(src_uv + OFF);                  \
+    for (int i = 0; i < kPaddedWidth * kPaddedHeight; ++i) {                   \
+      src_y_p[i] =                                                             \
+          (fastrand() & (((SRC_T)(-1)) << ((8 * SRC_BPC) - SRC_DEPTH)));       \
     }                                                                          \
-    memset(dst_y_c, 1, kWidth* kHeight);                                       \
-    memset(dst_u_c, 2,                                                         \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));      \
-    memset(dst_v_c, 3,                                                         \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));      \
-    memset(dst_y_opt, 101, kWidth* kHeight);                                   \
-    memset(dst_u_opt, 102,                                                     \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));      \
-    memset(dst_v_opt, 103,                                                     \
-           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));      \
+    for (int i = 0; i < kSrcHalfPaddedWidth * kSrcHalfPaddedHeight * 2; ++i) { \
+      src_uv_p[i] =                                                            \
+          (fastrand() & (((SRC_T)(-1)) << ((8 * SRC_BPC) - SRC_DEPTH)));       \
+    }                                                                          \
+    memset(dst_y_c, 1, kWidth* kHeight* DST_BPC);                              \
+    memset(dst_u_c, 2, kDstHalfWidth* kDstHalfHeight* DST_BPC);                \
+    memset(dst_v_c, 3, kDstHalfWidth* kDstHalfHeight* DST_BPC);                \
+    memset(dst_y_opt, 101, kWidth* kHeight* DST_BPC);                          \
+    memset(dst_u_opt, 102, kDstHalfWidth* kDstHalfHeight* DST_BPC);            \
+    memset(dst_v_opt, 103, kDstHalfWidth* kDstHalfHeight* DST_BPC);            \
     MaskCpuFlags(disable_cpu_flags_);                                          \
     SRC_FMT_PLANAR##To##FMT_PLANAR(                                            \
-        src_y + OFF, kWidth, src_uv + OFF,                                     \
-        2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X), DOY ? dst_y_c : NULL, kWidth,    \
-        dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_c,                        \
-        SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight);                    \
+        src_y_p, kWidth, src_uv_p, kSrcHalfWidth * 2,                          \
+        reinterpret_cast<DST_T*>(dst_y_c), kWidth,                             \
+        reinterpret_cast<DST_T*>(dst_u_c), kDstHalfWidth,                      \
+        reinterpret_cast<DST_T*>(dst_v_c), kDstHalfWidth, kWidth,              \
+        NEG kHeight);                                                          \
     MaskCpuFlags(benchmark_cpu_info_);                                         \
     for (int i = 0; i < benchmark_iterations_; ++i) {                          \
       SRC_FMT_PLANAR##To##FMT_PLANAR(                                          \
-          src_y + OFF, kWidth, src_uv + OFF,                                   \
-          2 * SUBSAMPLE(kWidth, SRC_SUBSAMP_X), DOY ? dst_y_opt : NULL,        \
-          kWidth, dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_opt,          \
-          SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight);                  \
-    }                                                                          \
-    int max_diff = 0;                                                          \
-    if (DOY) {                                                                 \
-      for (int i = 0; i < kHeight; ++i) {                                      \
-        for (int j = 0; j < kWidth; ++j) {                                     \
-          int abs_diff = abs(static_cast<int>(dst_y_c[i * kWidth + j]) -       \
-                             static_cast<int>(dst_y_opt[i * kWidth + j]));     \
-          if (abs_diff > max_diff) {                                           \
-            max_diff = abs_diff;                                               \
-          }                                                                    \
-        }                                                                      \
-      }                                                                        \
-      EXPECT_LE(max_diff, 1);                                                  \
+          src_y_p, kWidth, src_uv_p, kSrcHalfWidth * 2,                        \
+          reinterpret_cast<DST_T*>(dst_y_opt), kWidth,                         \
+          reinterpret_cast<DST_T*>(dst_u_opt), kDstHalfWidth,                  \
+          reinterpret_cast<DST_T*>(dst_v_opt), kDstHalfWidth, kWidth,          \
+          NEG kHeight);                                                        \
     }                                                                          \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                  \
-      for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) {                 \
-        int abs_diff = abs(                                                    \
-            static_cast<int>(dst_u_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) -  \
-            static_cast<int>(                                                  \
-                dst_u_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]));             \
-        if (abs_diff > max_diff) {                                             \
-          max_diff = abs_diff;                                                 \
-        }                                                                      \
-      }                                                                        \
+    for (int i = 0; i < kHeight * kWidth * DST_BPC; ++i) {                     \
+      EXPECT_EQ(dst_y_c[i], dst_y_opt[i]);                                     \
     }                                                                          \
-    EXPECT_LE(max_diff, 1);                                                    \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                  \
-      for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) {                 \
-        int abs_diff = abs(                                                    \
-            static_cast<int>(dst_v_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]) -  \
-            static_cast<int>(                                                  \
-                dst_v_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]));             \
-        if (abs_diff > max_diff) {                                             \
-          max_diff = abs_diff;                                                 \
-        }                                                                      \
-      }                                                                        \
+    for (int i = 0; i < kDstHalfWidth * kDstHalfHeight * DST_BPC; ++i) {       \
+      EXPECT_EQ(dst_u_c[i], dst_u_opt[i]);                                     \
+      EXPECT_EQ(dst_v_c[i], dst_v_opt[i]);                                     \
     }                                                                          \
-    EXPECT_LE(max_diff, 1);                                                    \
     free_aligned_buffer_page_end(dst_y_c);                                     \
     free_aligned_buffer_page_end(dst_u_c);                                     \
     free_aligned_buffer_page_end(dst_v_c);                                     \
@@ -588,29 +621,72 @@ TESTPLANARTOBP(I400, 2, 2, NV21, 2, 2)
     free_aligned_buffer_page_end(src_uv);                                      \
   }
 
-#define TESTBIPLANARTOP(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,         \
-                        FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y)                     \
-  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR,  \
-                   SUBSAMP_X, SUBSAMP_Y, benchmark_width_ - 4, _Any, +, 0, 1) \
-  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR,  \
-                   SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Unaligned, +, 1,  \
-                   1)                                                         \
-  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR,  \
-                   SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Invert, -, 0, 1)  \
-  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR,  \
-                   SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, 0, 1)     \
-  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR,  \
-                   SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _NullY, +, 0, 0)
-
-TESTBIPLANARTOP(NV12, 2, 2, I420, 2, 2)
-TESTBIPLANARTOP(NV21, 2, 2, I420, 2, 2)
+#define TESTBIPLANARTOP(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,         \
+                        SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC,             \
+                        DST_SUBSAMP_X, DST_SUBSAMP_Y, SRC_DEPTH, TILE_WIDTH,   \
+                        TILE_HEIGHT)                                           \
+  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                   SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                   DST_SUBSAMP_Y, benchmark_width_ + 1, _Any, +, 0, SRC_DEPTH, \
+                   TILE_WIDTH, TILE_HEIGHT)                                    \
+  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                   SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                   DST_SUBSAMP_Y, benchmark_width_, _Unaligned, +, 2,          \
+                   SRC_DEPTH, TILE_WIDTH, TILE_HEIGHT)                         \
+  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                   SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                   DST_SUBSAMP_Y, benchmark_width_, _Invert, -, 0, SRC_DEPTH,  \
+                   TILE_WIDTH, TILE_HEIGHT)                                    \
+  TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_T, SRC_BPC, SRC_SUBSAMP_X,              \
+                   SRC_SUBSAMP_Y, FMT_PLANAR, DST_T, DST_BPC, DST_SUBSAMP_X,   \
+                   DST_SUBSAMP_Y, benchmark_width_, _Opt, +, 0, SRC_DEPTH,     \
+                   TILE_WIDTH, TILE_HEIGHT)
+
+TESTBIPLANARTOP(NV12, uint8_t, 1, 2, 2, I420, uint8_t, 1, 2, 2, 8, 1, 1)
+TESTBIPLANARTOP(NV21, uint8_t, 1, 2, 2, I420, uint8_t, 1, 2, 2, 8, 1, 1)
+TESTBIPLANARTOP(MM21, uint8_t, 1, 2, 2, I420, uint8_t, 1, 2, 2, 8, 16, 32)
+
+// Provide matrix wrappers for full range bt.709
+#define F420ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I420ToARGBMatrix(a, b, e, f, c, d, g, h, &kYvuF709Constants, i, j)
+#define F420ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I420ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvF709Constants, i, j)
+#define F422ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I422ToARGBMatrix(a, b, e, f, c, d, g, h, &kYvuF709Constants, i, j)
+#define F422ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I422ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvF709Constants, i, j)
+#define F444ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I444ToARGBMatrix(a, b, e, f, c, d, g, h, &kYvuF709Constants, i, j)
+#define F444ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I444ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvF709Constants, i, j)
+
+// Provide matrix wrappers for full range bt.2020
+#define V420ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I420ToARGBMatrix(a, b, e, f, c, d, g, h, &kYvuV2020Constants, i, j)
+#define V420ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I420ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvV2020Constants, i, j)
+#define V422ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I422ToARGBMatrix(a, b, e, f, c, d, g, h, &kYvuV2020Constants, i, j)
+#define V422ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I422ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvV2020Constants, i, j)
+#define V444ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I444ToARGBMatrix(a, b, e, f, c, d, g, h, &kYvuV2020Constants, i, j)
+#define V444ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I444ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvV2020Constants, i, j)
+
+#define I420ToARGBFilter(a, b, c, d, e, f, g, h, i, j)                     \
+  I420ToARGBMatrixFilter(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j, \
+                         kFilterBilinear)
+#define I422ToARGBFilter(a, b, c, d, e, f, g, h, i, j)                     \
+  I422ToARGBMatrixFilter(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j, \
+                         kFilterBilinear)
 
 #define ALIGNINT(V, ALIGN) (((V) + (ALIGN)-1) / (ALIGN) * (ALIGN))
 
 #define TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
                        YALIGN, W1280, N, NEG, OFF)                            \
   TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                       \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+    const int kWidth = W1280;                                                 \
     const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                  \
     const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                     \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                       \
@@ -655,59 +731,123 @@ TESTBIPLANARTOP(NV21, 2, 2, I420, 2, 2)
     free_aligned_buffer_page_end(dst_argb_opt);                               \
   }
 
+#if defined(ENABLE_FULL_TESTS)
 #define TESTPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
                       YALIGN)                                                \
   TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                 YALIGN, benchmark_width_ - 4, _Any, +, 0)                   \
+                 YALIGN, benchmark_width_ + 1, _Any, +, 0)                   \
   TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                 YALIGN, benchmark_width_, _Unaligned, +, 1)                 \
+                 YALIGN, benchmark_width_, _Unaligned, +, 4)                 \
   TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
                  YALIGN, benchmark_width_, _Invert, -, 0)                    \
   TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
                  YALIGN, benchmark_width_, _Opt, +, 0)
+#else
+#define TESTPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
+                      YALIGN)                                                \
+  TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                 YALIGN, benchmark_width_ + 1, _Any, +, 0)                   \
+  TESTPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                 YALIGN, benchmark_width_, _Opt, +, 0)
+#endif
 
+#if defined(ENABLE_FULL_TESTS)
 TESTPLANARTOB(I420, 2, 2, ARGB, 4, 4, 1)
+TESTPLANARTOB(I420, 2, 2, ABGR, 4, 4, 1)
 TESTPLANARTOB(J420, 2, 2, ARGB, 4, 4, 1)
 TESTPLANARTOB(J420, 2, 2, ABGR, 4, 4, 1)
+TESTPLANARTOB(F420, 2, 2, ARGB, 4, 4, 1)
+TESTPLANARTOB(F420, 2, 2, ABGR, 4, 4, 1)
 TESTPLANARTOB(H420, 2, 2, ARGB, 4, 4, 1)
 TESTPLANARTOB(H420, 2, 2, ABGR, 4, 4, 1)
+TESTPLANARTOB(U420, 2, 2, ARGB, 4, 4, 1)
+TESTPLANARTOB(U420, 2, 2, ABGR, 4, 4, 1)
+TESTPLANARTOB(V420, 2, 2, ARGB, 4, 4, 1)
+TESTPLANARTOB(V420, 2, 2, ABGR, 4, 4, 1)
 TESTPLANARTOB(I420, 2, 2, BGRA, 4, 4, 1)
-TESTPLANARTOB(I420, 2, 2, ABGR, 4, 4, 1)
 TESTPLANARTOB(I420, 2, 2, RGBA, 4, 4, 1)
 TESTPLANARTOB(I420, 2, 2, RAW, 3, 3, 1)
 TESTPLANARTOB(I420, 2, 2, RGB24, 3, 3, 1)
+TESTPLANARTOB(J420, 2, 2, RAW, 3, 3, 1)
+TESTPLANARTOB(J420, 2, 2, RGB24, 3, 3, 1)
 TESTPLANARTOB(H420, 2, 2, RAW, 3, 3, 1)
 TESTPLANARTOB(H420, 2, 2, RGB24, 3, 3, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
 TESTPLANARTOB(I420, 2, 2, RGB565, 2, 2, 1)
 TESTPLANARTOB(J420, 2, 2, RGB565, 2, 2, 1)
 TESTPLANARTOB(H420, 2, 2, RGB565, 2, 2, 1)
 TESTPLANARTOB(I420, 2, 2, ARGB1555, 2, 2, 1)
 TESTPLANARTOB(I420, 2, 2, ARGB4444, 2, 2, 1)
-TESTPLANARTOB(I422, 2, 1, ARGB, 4, 4, 1)
 TESTPLANARTOB(I422, 2, 1, RGB565, 2, 2, 1)
+#endif
+TESTPLANARTOB(I422, 2, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, ABGR, 4, 4, 1)
 TESTPLANARTOB(J422, 2, 1, ARGB, 4, 4, 1)
 TESTPLANARTOB(J422, 2, 1, ABGR, 4, 4, 1)
 TESTPLANARTOB(H422, 2, 1, ARGB, 4, 4, 1)
 TESTPLANARTOB(H422, 2, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(U422, 2, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(U422, 2, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(V422, 2, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(V422, 2, 1, ABGR, 4, 4, 1)
 TESTPLANARTOB(I422, 2, 1, BGRA, 4, 4, 1)
-TESTPLANARTOB(I422, 2, 1, ABGR, 4, 4, 1)
 TESTPLANARTOB(I422, 2, 1, RGBA, 4, 4, 1)
 TESTPLANARTOB(I444, 1, 1, ARGB, 4, 4, 1)
-TESTPLANARTOB(J444, 1, 1, ARGB, 4, 4, 1)
 TESTPLANARTOB(I444, 1, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(J444, 1, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(J444, 1, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(H444, 1, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(H444, 1, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(U444, 1, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(U444, 1, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(V444, 1, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(V444, 1, 1, ABGR, 4, 4, 1)
 TESTPLANARTOB(I420, 2, 2, YUY2, 2, 4, 1)
 TESTPLANARTOB(I420, 2, 2, UYVY, 2, 4, 1)
 TESTPLANARTOB(I422, 2, 1, YUY2, 2, 4, 1)
 TESTPLANARTOB(I422, 2, 1, UYVY, 2, 4, 1)
 TESTPLANARTOB(I420, 2, 2, I400, 1, 1, 1)
 TESTPLANARTOB(J420, 2, 2, J400, 1, 1, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
 TESTPLANARTOB(I420, 2, 2, AR30, 4, 4, 1)
 TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
+TESTPLANARTOB(I420, 2, 2, AB30, 4, 4, 1)
+TESTPLANARTOB(H420, 2, 2, AB30, 4, 4, 1)
+#endif
+TESTPLANARTOB(I420, 2, 2, ARGBFilter, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, ARGBFilter, 4, 4, 1)
+#else
+TESTPLANARTOB(I420, 2, 2, ABGR, 4, 4, 1)
+TESTPLANARTOB(I420, 2, 2, ARGB, 4, 4, 1)
+TESTPLANARTOB(I420, 2, 2, BGRA, 4, 4, 1)
+TESTPLANARTOB(I420, 2, 2, RAW, 3, 3, 1)
+TESTPLANARTOB(I420, 2, 2, RGB24, 3, 3, 1)
+TESTPLANARTOB(I420, 2, 2, RGBA, 4, 4, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTPLANARTOB(I420, 2, 2, RGB565, 2, 2, 1)
+TESTPLANARTOB(I420, 2, 2, ARGB1555, 2, 2, 1)
+TESTPLANARTOB(I420, 2, 2, ARGB4444, 2, 2, 1)
+TESTPLANARTOB(I422, 2, 1, RGB565, 2, 2, 1)
+#endif
+TESTPLANARTOB(I420, 2, 2, I400, 1, 1, 1)
+TESTPLANARTOB(I420, 2, 2, UYVY, 2, 4, 1)
+TESTPLANARTOB(I420, 2, 2, YUY2, 2, 4, 1)
+TESTPLANARTOB(I420, 2, 2, ARGBFilter, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, ARGB, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, BGRA, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, RGBA, 4, 4, 1)
+TESTPLANARTOB(I422, 2, 1, UYVY, 2, 4, 1)
+TESTPLANARTOB(I422, 2, 1, YUY2, 2, 4, 1)
+TESTPLANARTOB(I422, 2, 1, ARGBFilter, 4, 4, 1)
+TESTPLANARTOB(I444, 1, 1, ABGR, 4, 4, 1)
+TESTPLANARTOB(I444, 1, 1, ARGB, 4, 4, 1)
+#endif
 
 #define TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
-                        YALIGN, W1280, DIFF, N, NEG, OFF, ATTEN)               \
+                        YALIGN, W1280, N, NEG, OFF, ATTEN)                     \
   TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                        \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+    const int kWidth = W1280;                                                  \
     const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                   \
     const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                      \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                        \
@@ -740,15 +880,9 @@ TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
                             dst_argb_opt + OFF, kStrideB, kWidth, NEG kHeight, \
                             ATTEN);                                            \
     }                                                                          \
-    int max_diff = 0;                                                          \
     for (int i = 0; i < kWidth * BPP_B * kHeight; ++i) {                       \
-      int abs_diff = abs(static_cast<int>(dst_argb_c[i + OFF]) -               \
-                         static_cast<int>(dst_argb_opt[i + OFF]));             \
-      if (abs_diff > max_diff) {                                               \
-        max_diff = abs_diff;                                                   \
-      }                                                                        \
+      EXPECT_EQ(dst_argb_c[i + OFF], dst_argb_opt[i + OFF]);                   \
     }                                                                          \
-    EXPECT_LE(max_diff, DIFF);                                                 \
     free_aligned_buffer_page_end(src_y);                                       \
     free_aligned_buffer_page_end(src_u);                                       \
     free_aligned_buffer_page_end(src_v);                                       \
@@ -757,26 +891,175 @@ TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
     free_aligned_buffer_page_end(dst_argb_opt);                                \
   }
 
+#if defined(ENABLE_FULL_TESTS)
 #define TESTQPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
-                       YALIGN, DIFF)                                          \
+                       YALIGN)                                                \
   TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                  YALIGN, benchmark_width_ - 4, DIFF, _Any, +, 0, 0)          \
+                  YALIGN, benchmark_width_ + 1, _Any, +, 0, 0)                \
   TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                  YALIGN, benchmark_width_, DIFF, _Unaligned, +, 1, 0)        \
+                  YALIGN, benchmark_width_, _Unaligned, +, 2, 0)              \
   TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                  YALIGN, benchmark_width_, DIFF, _Invert, -, 0, 0)           \
+                  YALIGN, benchmark_width_, _Invert, -, 0, 0)                 \
   TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                  YALIGN, benchmark_width_, DIFF, _Opt, +, 0, 0)              \
+                  YALIGN, benchmark_width_, _Opt, +, 0, 0)                    \
+  TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                  YALIGN, benchmark_width_, _Premult, +, 0, 1)
+#else
+#define TESTQPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
+                       YALIGN)                                                \
   TESTQPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                  YALIGN, benchmark_width_, DIFF, _Premult, +, 0, 1)
+                  YALIGN, benchmark_width_, _Opt, +, 0, 0)
+#endif
 
-TESTQPLANARTOB(I420Alpha, 2, 2, ARGB, 4, 4, 1, 2)
-TESTQPLANARTOB(I420Alpha, 2, 2, ABGR, 4, 4, 1, 2)
+#define J420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define J420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define F420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define F420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define H420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define H420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define U420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define U420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define V420AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I420AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define V420AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I420AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define J422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define J422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define F422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define F422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define H422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define H422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define U422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define U422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define V422AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I422AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define V422AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I422AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define J444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define J444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define F444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define F444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define H444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define H444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define U444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define U444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define V444AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I444AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define V444AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I444AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+
+#define I420AlphaToARGBFilter(a, b, c, d, e, f, g, h, i, j, k, l, m) \
+  I420AlphaToARGBMatrixFilter(a, b, c, d, e, f, g, h, i, j,          \
+                              &kYuvI601Constants, k, l, m, kFilterBilinear)
+#define I422AlphaToARGBFilter(a, b, c, d, e, f, g, h, i, j, k, l, m) \
+  I422AlphaToARGBMatrixFilter(a, b, c, d, e, f, g, h, i, j,          \
+                              &kYuvI601Constants, k, l, m, kFilterBilinear)
+
+#if defined(ENABLE_FULL_TESTS)
+TESTQPLANARTOB(I420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(I420Alpha, 2, 2, ABGR, 4, 4, 1)
+TESTQPLANARTOB(J420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(J420Alpha, 2, 2, ABGR, 4, 4, 1)
+TESTQPLANARTOB(H420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(H420Alpha, 2, 2, ABGR, 4, 4, 1)
+TESTQPLANARTOB(F420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(F420Alpha, 2, 2, ABGR, 4, 4, 1)
+TESTQPLANARTOB(U420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(U420Alpha, 2, 2, ABGR, 4, 4, 1)
+TESTQPLANARTOB(V420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(V420Alpha, 2, 2, ABGR, 4, 4, 1)
+TESTQPLANARTOB(I422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(I422Alpha, 2, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(J422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(J422Alpha, 2, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(H422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(H422Alpha, 2, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(F422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(F422Alpha, 2, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(U422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(U422Alpha, 2, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(V422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(V422Alpha, 2, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(I444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(I444Alpha, 1, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(J444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(J444Alpha, 1, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(H444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(H444Alpha, 1, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(F444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(F444Alpha, 1, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(U444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(U444Alpha, 1, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(V444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(V444Alpha, 1, 1, ABGR, 4, 4, 1)
+TESTQPLANARTOB(I420Alpha, 2, 2, ARGBFilter, 4, 4, 1)
+TESTQPLANARTOB(I422Alpha, 2, 1, ARGBFilter, 4, 4, 1)
+#else
+TESTQPLANARTOB(I420Alpha, 2, 2, ARGB, 4, 4, 1)
+TESTQPLANARTOB(I422Alpha, 2, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(I444Alpha, 1, 1, ARGB, 4, 4, 1)
+TESTQPLANARTOB(I420Alpha, 2, 2, ARGBFilter, 4, 4, 1)
+TESTQPLANARTOB(I422Alpha, 2, 1, ARGBFilter, 4, 4, 1)
+#endif
 
 #define TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C,       \
-                         BPP_B, W1280, DIFF, N, NEG, OFF)                      \
+                         BPP_B, W1280, N, NEG, OFF)                            \
   TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                        \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+    const int kWidth = W1280;                                                  \
     const int kHeight = benchmark_height_;                                     \
     const int kStrideB = kWidth * BPP_B;                                       \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                        \
@@ -813,18 +1096,12 @@ TESTQPLANARTOB(I420Alpha, 2, 2, ABGR, 4, 4, 1, 2)
                   kHeight);                                                    \
     FMT_C##ToARGB(dst_argb_opt, kStrideB, dst_argb32_opt, kWidth * 4, kWidth,  \
                   kHeight);                                                    \
-    int max_diff = 0;                                                          \
     for (int i = 0; i < kHeight; ++i) {                                        \
       for (int j = 0; j < kWidth * 4; ++j) {                                   \
-        int abs_diff =                                                         \
-            abs(static_cast<int>(dst_argb32_c[i * kWidth * 4 + j]) -           \
-                static_cast<int>(dst_argb32_opt[i * kWidth * 4 + j]));         \
-        if (abs_diff > max_diff) {                                             \
-          max_diff = abs_diff;                                                 \
-        }                                                                      \
+        EXPECT_EQ(dst_argb32_c[i * kWidth * 4 + j],                            \
+                  dst_argb32_opt[i * kWidth * 4 + j]);                         \
       }                                                                        \
     }                                                                          \
-    EXPECT_LE(max_diff, DIFF);                                                 \
     free_aligned_buffer_page_end(src_y);                                       \
     free_aligned_buffer_page_end(src_uv);                                      \
     free_aligned_buffer_page_end(dst_argb_c);                                  \
@@ -833,95 +1110,64 @@ TESTQPLANARTOB(I420Alpha, 2, 2, ABGR, 4, 4, 1, 2)
     free_aligned_buffer_page_end(dst_argb32_opt);                              \
   }
 
-#define TESTBIPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B, \
-                        DIFF)                                                  \
+#define TESTBIPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B) \
   TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B,      \
-                   benchmark_width_ - 4, DIFF, _Any, +, 0)                     \
+                   benchmark_width_ + 1, _Any, +, 0)                           \
   TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B,      \
-                   benchmark_width_, DIFF, _Unaligned, +, 1)                   \
+                   benchmark_width_, _Unaligned, +, 2)                         \
   TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B,      \
-                   benchmark_width_, DIFF, _Invert, -, 0)                      \
+                   benchmark_width_, _Invert, -, 0)                            \
   TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B,      \
-                   benchmark_width_, DIFF, _Opt, +, 0)
-
-TESTBIPLANARTOB(NV12, 2, 2, ARGB, ARGB, 4, 2)
-TESTBIPLANARTOB(NV21, 2, 2, ARGB, ARGB, 4, 2)
-TESTBIPLANARTOB(NV12, 2, 2, ABGR, ABGR, 4, 2)
-TESTBIPLANARTOB(NV21, 2, 2, ABGR, ABGR, 4, 2)
-TESTBIPLANARTOB(NV12, 2, 2, RGB24, RGB24, 3, 2)
-TESTBIPLANARTOB(NV21, 2, 2, RGB24, RGB24, 3, 2)
-TESTBIPLANARTOB(NV12, 2, 2, RAW, RAW, 3, 2)
-TESTBIPLANARTOB(NV21, 2, 2, RAW, RAW, 3, 2)
-TESTBIPLANARTOB(NV12, 2, 2, RGB565, RGB565, 2, 9)
-TESTBIPLANARTOB(NV21, 2, 2, YUV24, RAW, 3, 2)
-
-#ifdef DO_THREE_PLANES
-// Do 3 allocations for yuv.  conventional but slower.
-#define TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, \
-                       W1280, DIFF, N, NEG, OFF)                               \
-  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_PLANAR##N) {                        \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
-    const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                   \
-    const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                        \
-    const int kStride = (kStrideUV * SUBSAMP_X * 8 * BPP_A + 7) / 8;           \
-    align_buffer_page_end(src_argb, kStride* kHeight + OFF);                   \
-    align_buffer_page_end(dst_y_c, kWidth* kHeight);                           \
-    align_buffer_page_end(dst_u_c, kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));  \
-    align_buffer_page_end(dst_v_c, kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));  \
-    align_buffer_page_end(dst_y_opt, kWidth* kHeight);                         \
-    align_buffer_page_end(dst_u_opt,                                           \
-                          kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));           \
-    align_buffer_page_end(dst_v_opt,                                           \
-                          kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));           \
-    memset(dst_y_c, 1, kWidth* kHeight);                                       \
-    memset(dst_u_c, 2, kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));              \
-    memset(dst_v_c, 3, kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));              \
-    memset(dst_y_opt, 101, kWidth* kHeight);                                   \
-    memset(dst_u_opt, 102, kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));          \
-    memset(dst_v_opt, 103, kStrideUV* SUBSAMPLE(kHeight, SUBSAMP_Y));          \
-    for (int i = 0; i < kHeight; ++i)                                          \
-      for (int j = 0; j < kStride; ++j)                                        \
-        src_argb[(i * kStride) + j + OFF] = (fastrand() & 0xff);               \
-    MaskCpuFlags(disable_cpu_flags_);                                          \
-    FMT_A##To##FMT_PLANAR(src_argb + OFF, kStride, dst_y_c, kWidth, dst_u_c,   \
-                          kStrideUV, dst_v_c, kStrideUV, kWidth, NEG kHeight); \
-    MaskCpuFlags(benchmark_cpu_info_);                                         \
-    for (int i = 0; i < benchmark_iterations_; ++i) {                          \
-      FMT_A##To##FMT_PLANAR(src_argb + OFF, kStride, dst_y_opt, kWidth,        \
-                            dst_u_opt, kStrideUV, dst_v_opt, kStrideUV,        \
-                            kWidth, NEG kHeight);                              \
-    }                                                                          \
-    for (int i = 0; i < kHeight; ++i) {                                        \
-      for (int j = 0; j < kWidth; ++j) {                                       \
-        EXPECT_NEAR(static_cast<int>(dst_y_c[i * kWidth + j]),                 \
-                    static_cast<int>(dst_y_opt[i * kWidth + j]), DIFF);        \
-      }                                                                        \
-    }                                                                          \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                  \
-      for (int j = 0; j < kStrideUV; ++j) {                                    \
-        EXPECT_NEAR(static_cast<int>(dst_u_c[i * kStrideUV + j]),              \
-                    static_cast<int>(dst_u_opt[i * kStrideUV + j]), DIFF);     \
-      }                                                                        \
-    }                                                                          \
-    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                  \
-      for (int j = 0; j < kStrideUV; ++j) {                                    \
-        EXPECT_NEAR(static_cast<int>(dst_v_c[i * kStrideUV + j]),              \
-                    static_cast<int>(dst_v_opt[i * kStrideUV + j]), DIFF);     \
-      }                                                                        \
-    }                                                                          \
-    free_aligned_buffer_page_end(dst_y_c);                                     \
-    free_aligned_buffer_page_end(dst_u_c);                                     \
-    free_aligned_buffer_page_end(dst_v_c);                                     \
-    free_aligned_buffer_page_end(dst_y_opt);                                   \
-    free_aligned_buffer_page_end(dst_u_opt);                                   \
-    free_aligned_buffer_page_end(dst_v_opt);                                   \
-    free_aligned_buffer_page_end(src_argb);                                    \
-  }
-#else
+                   benchmark_width_, _Opt, +, 0)
+
+#define JNV12ToARGB(a, b, c, d, e, f, g, h) \
+  NV12ToARGBMatrix(a, b, c, d, e, f, &kYuvJPEGConstants, g, h)
+#define JNV21ToARGB(a, b, c, d, e, f, g, h) \
+  NV21ToARGBMatrix(a, b, c, d, e, f, &kYuvJPEGConstants, g, h)
+#define JNV12ToABGR(a, b, c, d, e, f, g, h) \
+  NV21ToARGBMatrix(a, b, c, d, e, f, &kYvuJPEGConstants, g, h)
+#define JNV21ToABGR(a, b, c, d, e, f, g, h) \
+  NV12ToARGBMatrix(a, b, c, d, e, f, &kYvuJPEGConstants, g, h)
+#define JNV12ToRGB24(a, b, c, d, e, f, g, h) \
+  NV12ToRGB24Matrix(a, b, c, d, e, f, &kYuvJPEGConstants, g, h)
+#define JNV21ToRGB24(a, b, c, d, e, f, g, h) \
+  NV21ToRGB24Matrix(a, b, c, d, e, f, &kYuvJPEGConstants, g, h)
+#define JNV12ToRAW(a, b, c, d, e, f, g, h) \
+  NV21ToRGB24Matrix(a, b, c, d, e, f, &kYvuJPEGConstants, g, h)
+#define JNV21ToRAW(a, b, c, d, e, f, g, h) \
+  NV12ToRGB24Matrix(a, b, c, d, e, f, &kYvuJPEGConstants, g, h)
+#define JNV12ToRGB565(a, b, c, d, e, f, g, h) \
+  NV12ToRGB565Matrix(a, b, c, d, e, f, &kYuvJPEGConstants, g, h)
+
+TESTBIPLANARTOB(JNV12, 2, 2, ARGB, ARGB, 4)
+TESTBIPLANARTOB(JNV21, 2, 2, ARGB, ARGB, 4)
+TESTBIPLANARTOB(JNV12, 2, 2, ABGR, ABGR, 4)
+TESTBIPLANARTOB(JNV21, 2, 2, ABGR, ABGR, 4)
+TESTBIPLANARTOB(JNV12, 2, 2, RGB24, RGB24, 3)
+TESTBIPLANARTOB(JNV21, 2, 2, RGB24, RGB24, 3)
+TESTBIPLANARTOB(JNV12, 2, 2, RAW, RAW, 3)
+TESTBIPLANARTOB(JNV21, 2, 2, RAW, RAW, 3)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTBIPLANARTOB(JNV12, 2, 2, RGB565, RGB565, 2)
+#endif
+
+TESTBIPLANARTOB(NV12, 2, 2, ARGB, ARGB, 4)
+TESTBIPLANARTOB(NV21, 2, 2, ARGB, ARGB, 4)
+TESTBIPLANARTOB(NV12, 2, 2, ABGR, ABGR, 4)
+TESTBIPLANARTOB(NV21, 2, 2, ABGR, ABGR, 4)
+TESTBIPLANARTOB(NV12, 2, 2, RGB24, RGB24, 3)
+TESTBIPLANARTOB(NV21, 2, 2, RGB24, RGB24, 3)
+TESTBIPLANARTOB(NV12, 2, 2, RAW, RAW, 3)
+TESTBIPLANARTOB(NV21, 2, 2, RAW, RAW, 3)
+TESTBIPLANARTOB(NV21, 2, 2, YUV24, RAW, 3)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTBIPLANARTOB(NV12, 2, 2, RGB565, RGB565, 2)
+#endif
+
 #define TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, \
-                       W1280, DIFF, N, NEG, OFF)                               \
+                       W1280, N, NEG, OFF)                                     \
   TEST_F(LibYUVConvertTest, FMT_A##To##FMT_PLANAR##N) {                        \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+    const int kWidth = W1280;                                                  \
     const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                   \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                        \
     const int kStride = (kStrideUV * SUBSAMP_X * 8 * BPP_A + 7) / 8;           \
@@ -951,14 +1197,12 @@ TESTBIPLANARTOB(NV21, 2, 2, YUV24, RAW, 3, 2)
     }                                                                          \
     for (int i = 0; i < kHeight; ++i) {                                        \
       for (int j = 0; j < kWidth; ++j) {                                       \
-        EXPECT_NEAR(static_cast<int>(dst_y_c[i * kWidth + j]),                 \
-                    static_cast<int>(dst_y_opt[i * kWidth + j]), DIFF);        \
+        EXPECT_EQ(dst_y_c[i * kWidth + j], dst_y_opt[i * kWidth + j]);         \
       }                                                                        \
     }                                                                          \
     for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y) * 2; ++i) {              \
       for (int j = 0; j < kStrideUV; ++j) {                                    \
-        EXPECT_NEAR(static_cast<int>(dst_uv_c[i * kStrideUV + j]),             \
-                    static_cast<int>(dst_uv_opt[i * kStrideUV + j]), DIFF);    \
+        EXPECT_EQ(dst_uv_c[i * kStrideUV + j], dst_uv_opt[i * kStrideUV + j]); \
       }                                                                        \
     }                                                                          \
     free_aligned_buffer_page_end(dst_y_c);                                     \
@@ -967,45 +1211,53 @@ TESTBIPLANARTOB(NV21, 2, 2, YUV24, RAW, 3, 2)
     free_aligned_buffer_page_end(dst_uv_opt);                                  \
     free_aligned_buffer_page_end(src_argb);                                    \
   }
-#endif
 
-#define TESTATOPLANAR(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, \
-                      DIFF)                                                   \
+#if defined(ENABLE_FULL_TESTS)
+#define TESTATOPLANAR(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y) \
   TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
-                 benchmark_width_ - 4, DIFF, _Any, +, 0)                      \
+                 benchmark_width_ + 1, _Any, +, 0)                            \
   TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
-                 benchmark_width_, DIFF, _Unaligned, +, 1)                    \
+                 benchmark_width_, _Unaligned, +, 2)                          \
   TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
-                 benchmark_width_, DIFF, _Invert, -, 0)                       \
+                 benchmark_width_, _Invert, -, 0)                             \
   TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
-                 benchmark_width_, DIFF, _Opt, +, 0)
-
-TESTATOPLANAR(ABGR, 4, 1, I420, 2, 2, 4)
-TESTATOPLANAR(ARGB, 4, 1, I420, 2, 2, 4)
-TESTATOPLANAR(ARGB, 4, 1, I422, 2, 1, 2)
-TESTATOPLANAR(ARGB, 4, 1, I444, 1, 1, 2)
-TESTATOPLANAR(ARGB, 4, 1, J420, 2, 2, ARM_YUV_ERROR)
-TESTATOPLANAR(ARGB, 4, 1, J422, 2, 1, ARM_YUV_ERROR)
-TESTATOPLANAR(ARGB1555, 2, 1, I420, 2, 2, 15)
-TESTATOPLANAR(ARGB4444, 2, 1, I420, 2, 2, 17)
-TESTATOPLANAR(BGRA, 4, 1, I420, 2, 2, 4)
-TESTATOPLANAR(I400, 1, 1, I420, 2, 2, 2)
-TESTATOPLANAR(J400, 1, 1, J420, 2, 2, 2)
-TESTATOPLANAR(RAW, 3, 1, I420, 2, 2, 4)
-TESTATOPLANAR(RGB24, 3, 1, I420, 2, 2, 4)
-// TODO(fbarchard): Investigate J420 error of 11 on Windows.
-TESTATOPLANAR(RGB24, 3, 1, J420, 2, 2, 11)
-TESTATOPLANAR(RGB565, 2, 1, I420, 2, 2, 5)
-TESTATOPLANAR(RGBA, 4, 1, I420, 2, 2, 4)
-TESTATOPLANAR(UYVY, 2, 1, I420, 2, 2, 2)
-TESTATOPLANAR(UYVY, 2, 1, I422, 2, 1, 2)
-TESTATOPLANAR(YUY2, 2, 1, I420, 2, 2, 2)
-TESTATOPLANAR(YUY2, 2, 1, I422, 2, 1, 2)
+                 benchmark_width_, _Opt, +, 0)
+#else
+#define TESTATOPLANAR(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y) \
+  TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
+                 benchmark_width_ + 1, _Any, +, 0)                            \
+  TESTATOPLANARI(FMT_A, BPP_A, YALIGN, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
+                 benchmark_width_, _Opt, +, 0)
+#endif
+
+TESTATOPLANAR(ABGR, 4, 1, I420, 2, 2)
+TESTATOPLANAR(ARGB, 4, 1, I420, 2, 2)
+TESTATOPLANAR(ARGB, 4, 1, I422, 2, 1)
+TESTATOPLANAR(ARGB, 4, 1, I444, 1, 1)
+TESTATOPLANAR(ARGB, 4, 1, J420, 2, 2)
+TESTATOPLANAR(ARGB, 4, 1, J422, 2, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOPLANAR(ARGB4444, 2, 1, I420, 2, 2)
+TESTATOPLANAR(RGB565, 2, 1, I420, 2, 2)
+TESTATOPLANAR(ARGB1555, 2, 1, I420, 2, 2)
+#endif
+TESTATOPLANAR(BGRA, 4, 1, I420, 2, 2)
+TESTATOPLANAR(I400, 1, 1, I420, 2, 2)
+TESTATOPLANAR(J400, 1, 1, J420, 2, 2)
+TESTATOPLANAR(RAW, 3, 1, I420, 2, 2)
+TESTATOPLANAR(RAW, 3, 1, J420, 2, 2)
+TESTATOPLANAR(RGB24, 3, 1, I420, 2, 2)
+TESTATOPLANAR(RGB24, 3, 1, J420, 2, 2)
+TESTATOPLANAR(RGBA, 4, 1, I420, 2, 2)
+TESTATOPLANAR(UYVY, 2, 1, I420, 2, 2)
+TESTATOPLANAR(UYVY, 2, 1, I422, 2, 1)
+TESTATOPLANAR(YUY2, 2, 1, I420, 2, 2)
+TESTATOPLANAR(YUY2, 2, 1, I422, 2, 1)
 
 #define TESTATOBIPLANARI(FMT_A, SUB_A, BPP_A, FMT_PLANAR, SUBSAMP_X,          \
                          SUBSAMP_Y, W1280, N, NEG, OFF)                       \
   TEST_F(LibYUVConvertTest, FMT_A##To##FMT_PLANAR##N) {                       \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+    const int kWidth = W1280;                                                 \
     const int kHeight = benchmark_height_;                                    \
     const int kStride = SUBSAMPLE(kWidth, SUB_A) * BPP_A;                     \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                       \
@@ -1031,28 +1283,17 @@ TESTATOPLANAR(YUY2, 2, 1, I422, 2, 1, 2)
       FMT_A##To##FMT_PLANAR(src_argb + OFF, kStride, dst_y_opt, kWidth,       \
                             dst_uv_opt, kStrideUV * 2, kWidth, NEG kHeight);  \
     }                                                                         \
-    int max_diff = 0;                                                         \
     for (int i = 0; i < kHeight; ++i) {                                       \
       for (int j = 0; j < kWidth; ++j) {                                      \
-        int abs_diff = abs(static_cast<int>(dst_y_c[i * kWidth + j]) -        \
-                           static_cast<int>(dst_y_opt[i * kWidth + j]));      \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
+        EXPECT_EQ(dst_y_c[i * kWidth + j], dst_y_opt[i * kWidth + j]);        \
       }                                                                       \
     }                                                                         \
-    EXPECT_LE(max_diff, 4);                                                   \
     for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                 \
       for (int j = 0; j < kStrideUV * 2; ++j) {                               \
-        int abs_diff =                                                        \
-            abs(static_cast<int>(dst_uv_c[i * kStrideUV * 2 + j]) -           \
-                static_cast<int>(dst_uv_opt[i * kStrideUV * 2 + j]));         \
-        if (abs_diff > max_diff) {                                            \
-          max_diff = abs_diff;                                                \
-        }                                                                     \
+        EXPECT_EQ(dst_uv_c[i * kStrideUV * 2 + j],                            \
+                  dst_uv_opt[i * kStrideUV * 2 + j]);                         \
       }                                                                       \
     }                                                                         \
-    EXPECT_LE(max_diff, 4);                                                   \
     free_aligned_buffer_page_end(dst_y_c);                                    \
     free_aligned_buffer_page_end(dst_uv_c);                                   \
     free_aligned_buffer_page_end(dst_y_opt);                                  \
@@ -1062,9 +1303,9 @@ TESTATOPLANAR(YUY2, 2, 1, I422, 2, 1, 2)
 
 #define TESTATOBIPLANAR(FMT_A, SUB_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y) \
   TESTATOBIPLANARI(FMT_A, SUB_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
-                   benchmark_width_ - 4, _Any, +, 0)                           \
+                   benchmark_width_ + 1, _Any, +, 0)                           \
   TESTATOBIPLANARI(FMT_A, SUB_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
-                   benchmark_width_, _Unaligned, +, 1)                         \
+                   benchmark_width_, _Unaligned, +, 2)                         \
   TESTATOBIPLANARI(FMT_A, SUB_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
                    benchmark_width_, _Invert, -, 0)                            \
   TESTATOBIPLANARI(FMT_A, SUB_A, BPP_A, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
@@ -1072,146 +1313,181 @@ TESTATOPLANAR(YUY2, 2, 1, I422, 2, 1, 2)
 
 TESTATOBIPLANAR(ARGB, 1, 4, NV12, 2, 2)
 TESTATOBIPLANAR(ARGB, 1, 4, NV21, 2, 2)
+TESTATOBIPLANAR(ABGR, 1, 4, NV12, 2, 2)
+TESTATOBIPLANAR(ABGR, 1, 4, NV21, 2, 2)
+TESTATOBIPLANAR(RAW, 1, 3, JNV21, 2, 2)
 TESTATOBIPLANAR(YUY2, 2, 4, NV12, 2, 2)
 TESTATOBIPLANAR(UYVY, 2, 4, NV12, 2, 2)
 TESTATOBIPLANAR(AYUV, 1, 4, NV12, 2, 2)
 TESTATOBIPLANAR(AYUV, 1, 4, NV21, 2, 2)
 
-#define TESTATOBI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,  \
-                  HEIGHT_B, W1280, DIFF, N, NEG, OFF)                        \
-  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##N) {                           \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                          \
-    const int kHeight = benchmark_height_;                                   \
-    const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A;     \
-    const int kHeightB = (kHeight + HEIGHT_B - 1) / HEIGHT_B * HEIGHT_B;     \
-    const int kStrideA =                                                     \
-        (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;               \
-    const int kStrideB =                                                     \
-        (kWidth * BPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B;               \
-    align_buffer_page_end(src_argb, kStrideA* kHeightA + OFF);               \
-    align_buffer_page_end(dst_argb_c, kStrideB* kHeightB);                   \
-    align_buffer_page_end(dst_argb_opt, kStrideB* kHeightB);                 \
-    for (int i = 0; i < kStrideA * kHeightA; ++i) {                          \
-      src_argb[i + OFF] = (fastrand() & 0xff);                               \
-    }                                                                        \
-    memset(dst_argb_c, 1, kStrideB* kHeightB);                               \
-    memset(dst_argb_opt, 101, kStrideB* kHeightB);                           \
-    MaskCpuFlags(disable_cpu_flags_);                                        \
-    FMT_A##To##FMT_B(src_argb + OFF, kStrideA, dst_argb_c, kStrideB, kWidth, \
-                     NEG kHeight);                                           \
-    MaskCpuFlags(benchmark_cpu_info_);                                       \
-    for (int i = 0; i < benchmark_iterations_; ++i) {                        \
-      FMT_A##To##FMT_B(src_argb + OFF, kStrideA, dst_argb_opt, kStrideB,     \
-                       kWidth, NEG kHeight);                                 \
-    }                                                                        \
-    int max_diff = 0;                                                        \
-    for (int i = 0; i < kStrideB * kHeightB; ++i) {                          \
-      int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -                   \
-                         static_cast<int>(dst_argb_opt[i]));                 \
-      if (abs_diff > max_diff) {                                             \
-        max_diff = abs_diff;                                                 \
-      }                                                                      \
-    }                                                                        \
-    EXPECT_LE(max_diff, DIFF);                                               \
-    free_aligned_buffer_page_end(src_argb);                                  \
-    free_aligned_buffer_page_end(dst_argb_c);                                \
-    free_aligned_buffer_page_end(dst_argb_opt);                              \
+#define TESTATOBI(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B,     \
+                  EPP_B, STRIDE_B, HEIGHT_B, W1280, N, NEG, OFF)               \
+  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##N) {                             \
+    const int kWidth = W1280;                                                  \
+    const int kHeight = benchmark_height_;                                     \
+    const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A;       \
+    const int kHeightB = (kHeight + HEIGHT_B - 1) / HEIGHT_B * HEIGHT_B;       \
+    const int kStrideA =                                                       \
+        (kWidth * EPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;                 \
+    const int kStrideB =                                                       \
+        (kWidth * EPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B;                 \
+    align_buffer_page_end(src_argb,                                            \
+                          kStrideA* kHeightA*(int)sizeof(TYPE_A) + OFF);       \
+    align_buffer_page_end(dst_argb_c, kStrideB* kHeightB*(int)sizeof(TYPE_B)); \
+    align_buffer_page_end(dst_argb_opt,                                        \
+                          kStrideB* kHeightB*(int)sizeof(TYPE_B));             \
+    for (int i = 0; i < kStrideA * kHeightA * (int)sizeof(TYPE_A); ++i) {      \
+      src_argb[i + OFF] = (fastrand() & 0xff);                                 \
+    }                                                                          \
+    memset(dst_argb_c, 1, kStrideB* kHeightB);                                 \
+    memset(dst_argb_opt, 101, kStrideB* kHeightB);                             \
+    MaskCpuFlags(disable_cpu_flags_);                                          \
+    FMT_A##To##FMT_B((TYPE_A*)(src_argb + OFF), kStrideA, (TYPE_B*)dst_argb_c, \
+                     kStrideB, kWidth, NEG kHeight);                           \
+    MaskCpuFlags(benchmark_cpu_info_);                                         \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                          \
+      FMT_A##To##FMT_B((TYPE_A*)(src_argb + OFF), kStrideA,                    \
+                       (TYPE_B*)dst_argb_opt, kStrideB, kWidth, NEG kHeight);  \
+    }                                                                          \
+    for (int i = 0; i < kStrideB * kHeightB * (int)sizeof(TYPE_B); ++i) {      \
+      EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]);                               \
+    }                                                                          \
+    free_aligned_buffer_page_end(src_argb);                                    \
+    free_aligned_buffer_page_end(dst_argb_c);                                  \
+    free_aligned_buffer_page_end(dst_argb_opt);                                \
   }
 
-#define TESTATOBRANDOM(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B,     \
-                       STRIDE_B, HEIGHT_B, DIFF)                           \
-  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##_Random) {                   \
-    for (int times = 0; times < benchmark_iterations_; ++times) {          \
-      const int kWidth = (fastrand() & 63) + 1;                            \
-      const int kHeight = (fastrand() & 31) + 1;                           \
-      const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A; \
-      const int kHeightB = (kHeight + HEIGHT_B - 1) / HEIGHT_B * HEIGHT_B; \
-      const int kStrideA =                                                 \
-          (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;           \
-      const int kStrideB =                                                 \
-          (kWidth * BPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B;           \
-      align_buffer_page_end(src_argb, kStrideA* kHeightA);                 \
-      align_buffer_page_end(dst_argb_c, kStrideB* kHeightB);               \
-      align_buffer_page_end(dst_argb_opt, kStrideB* kHeightB);             \
-      for (int i = 0; i < kStrideA * kHeightA; ++i) {                      \
-        src_argb[i] = (fastrand() & 0xff);                                 \
-      }                                                                    \
-      memset(dst_argb_c, 123, kStrideB* kHeightB);                         \
-      memset(dst_argb_opt, 123, kStrideB* kHeightB);                       \
-      MaskCpuFlags(disable_cpu_flags_);                                    \
-      FMT_A##To##FMT_B(src_argb, kStrideA, dst_argb_c, kStrideB, kWidth,   \
-                       kHeight);                                           \
-      MaskCpuFlags(benchmark_cpu_info_);                                   \
-      FMT_A##To##FMT_B(src_argb, kStrideA, dst_argb_opt, kStrideB, kWidth, \
-                       kHeight);                                           \
-      for (int i = 0; i < kStrideB * kHeightB; ++i) {                      \
-        EXPECT_NEAR(dst_argb_c[i], dst_argb_opt[i], DIFF);                 \
-      }                                                                    \
-      free_aligned_buffer_page_end(src_argb);                              \
-      free_aligned_buffer_page_end(dst_argb_c);                            \
-      free_aligned_buffer_page_end(dst_argb_opt);                          \
-    }                                                                      \
-  }
-
-#define TESTATOB(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
-                 HEIGHT_B, DIFF)                                           \
-  TESTATOBI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-            HEIGHT_B, benchmark_width_ - 4, DIFF, _Any, +, 0)              \
-  TESTATOBI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-            HEIGHT_B, benchmark_width_, DIFF, _Unaligned, +, 1)            \
-  TESTATOBI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-            HEIGHT_B, benchmark_width_, DIFF, _Invert, -, 0)               \
-  TESTATOBI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-            HEIGHT_B, benchmark_width_, DIFF, _Opt, +, 0)                  \
-  TESTATOBRANDOM(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
-                 HEIGHT_B, DIFF)
-
-// TODO(fbarchard): make ARM version of C code that matches NEON.
-TESTATOB(AB30, 4, 4, 1, ABGR, 4, 4, 1, 0)
-TESTATOB(AB30, 4, 4, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(ABGR, 4, 4, 1, AR30, 4, 4, 1, 0)
-TESTATOB(ABGR, 4, 4, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(AR30, 4, 4, 1, AB30, 4, 4, 1, 0)
-TESTATOB(AR30, 4, 4, 1, ABGR, 4, 4, 1, 0)
-TESTATOB(AR30, 4, 4, 1, AR30, 4, 4, 1, 0)
-TESTATOB(AR30, 4, 4, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, ABGR, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, AR30, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, ARGB1555, 2, 2, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, ARGB4444, 2, 2, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, ARGBMirror, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, BGRA, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, I400, 1, 1, 1, 2)
-TESTATOB(ARGB, 4, 4, 1, J400, 1, 1, 1, 2)
-TESTATOB(ARGB, 4, 4, 1, RAW, 3, 3, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, RGB24, 3, 3, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, RGB565, 2, 2, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, RGBA, 4, 4, 1, 0)
-TESTATOB(ARGB, 4, 4, 1, UYVY, 2, 4, 1, 4)
-TESTATOB(ARGB, 4, 4, 1, YUY2, 2, 4, 1, 4)
-TESTATOB(ARGB1555, 2, 2, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(ARGB4444, 2, 2, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(BGRA, 4, 4, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(I400, 1, 1, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(I400, 1, 1, 1, I400, 1, 1, 1, 0)
-TESTATOB(I400, 1, 1, 1, I400Mirror, 1, 1, 1, 0)
-TESTATOB(J400, 1, 1, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(J400, 1, 1, 1, J400, 1, 1, 1, 0)
-TESTATOB(RAW, 3, 3, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(RAW, 3, 3, 1, RGB24, 3, 3, 1, 0)
-TESTATOB(RGB24, 3, 3, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(RGB24, 3, 3, 1, J400, 1, 1, 1, 0)
-TESTATOB(RGB565, 2, 2, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(RGBA, 4, 4, 1, ARGB, 4, 4, 1, 0)
-TESTATOB(UYVY, 2, 4, 1, ARGB, 4, 4, 1, ARM_YUV_ERROR)
-TESTATOB(YUY2, 2, 4, 1, ARGB, 4, 4, 1, ARM_YUV_ERROR)
-TESTATOB(YUY2, 2, 4, 1, Y, 1, 1, 1, 0)
+#define TESTATOBRANDOM(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B,        \
+                       TYPE_B, EPP_B, STRIDE_B, HEIGHT_B)                      \
+  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##_Random) {                       \
+    for (int times = 0; times < benchmark_iterations_; ++times) {              \
+      const int kWidth = (fastrand() & 63) + 1;                                \
+      const int kHeight = (fastrand() & 31) + 1;                               \
+      const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A;     \
+      const int kHeightB = (kHeight + HEIGHT_B - 1) / HEIGHT_B * HEIGHT_B;     \
+      const int kStrideA =                                                     \
+          (kWidth * EPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;               \
+      const int kStrideB =                                                     \
+          (kWidth * EPP_B + STRIDE_B - 1) / STRIDE_B * STRIDE_B;               \
+      align_buffer_page_end(src_argb, kStrideA* kHeightA*(int)sizeof(TYPE_A)); \
+      align_buffer_page_end(dst_argb_c,                                        \
+                            kStrideB* kHeightB*(int)sizeof(TYPE_B));           \
+      align_buffer_page_end(dst_argb_opt,                                      \
+                            kStrideB* kHeightB*(int)sizeof(TYPE_B));           \
+      for (int i = 0; i < kStrideA * kHeightA * (int)sizeof(TYPE_A); ++i) {    \
+        src_argb[i] = 0xfe;                                                    \
+      }                                                                        \
+      memset(dst_argb_c, 123, kStrideB* kHeightB);                             \
+      memset(dst_argb_opt, 123, kStrideB* kHeightB);                           \
+      MaskCpuFlags(disable_cpu_flags_);                                        \
+      FMT_A##To##FMT_B((TYPE_A*)src_argb, kStrideA, (TYPE_B*)dst_argb_c,       \
+                       kStrideB, kWidth, kHeight);                             \
+      MaskCpuFlags(benchmark_cpu_info_);                                       \
+      FMT_A##To##FMT_B((TYPE_A*)src_argb, kStrideA, (TYPE_B*)dst_argb_opt,     \
+                       kStrideB, kWidth, kHeight);                             \
+      for (int i = 0; i < kStrideB * kHeightB * (int)sizeof(TYPE_B); ++i) {    \
+        EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]);                             \
+      }                                                                        \
+      free_aligned_buffer_page_end(src_argb);                                  \
+      free_aligned_buffer_page_end(dst_argb_c);                                \
+      free_aligned_buffer_page_end(dst_argb_opt);                              \
+    }                                                                          \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+#define TESTATOB(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B,   \
+                 EPP_B, STRIDE_B, HEIGHT_B)                                 \
+  TESTATOBI(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B, EPP_B, \
+            STRIDE_B, HEIGHT_B, benchmark_width_ + 1, _Any, +, 0)           \
+  TESTATOBI(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B, EPP_B, \
+            STRIDE_B, HEIGHT_B, benchmark_width_, _Unaligned, +, 4)         \
+  TESTATOBI(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B, EPP_B, \
+            STRIDE_B, HEIGHT_B, benchmark_width_, _Invert, -, 0)            \
+  TESTATOBI(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B, EPP_B, \
+            STRIDE_B, HEIGHT_B, benchmark_width_, _Opt, +, 0)               \
+  TESTATOBRANDOM(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B,   \
+                 EPP_B, STRIDE_B, HEIGHT_B)
+#else
+#define TESTATOB(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B,   \
+                 EPP_B, STRIDE_B, HEIGHT_B)                                 \
+  TESTATOBI(FMT_A, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, FMT_B, TYPE_B, EPP_B, \
+            STRIDE_B, HEIGHT_B, benchmark_width_, _Opt, +, 0)
+#endif
+
+TESTATOB(AB30, uint8_t, 4, 4, 1, ABGR, uint8_t, 4, 4, 1)
+TESTATOB(AB30, uint8_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOB(ABGR, uint8_t, 4, 4, 1, AR30, uint8_t, 4, 4, 1)
+#endif
+TESTATOB(ABGR, uint8_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOB(AR30, uint8_t, 4, 4, 1, AB30, uint8_t, 4, 4, 1)
+#endif
+TESTATOB(AR30, uint8_t, 4, 4, 1, ABGR, uint8_t, 4, 4, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOB(AR30, uint8_t, 4, 4, 1, AR30, uint8_t, 4, 4, 1)
+TESTATOB(AR30, uint8_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+#endif
+TESTATOB(ARGB, uint8_t, 4, 4, 1, ABGR, uint8_t, 4, 4, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOB(ARGB, uint8_t, 4, 4, 1, AR30, uint8_t, 4, 4, 1)
+#endif
+TESTATOB(ARGB, uint8_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, ARGB1555, uint8_t, 2, 2, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, ARGB4444, uint8_t, 2, 2, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, ARGBMirror, uint8_t, 4, 4, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, BGRA, uint8_t, 4, 4, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, I400, uint8_t, 1, 1, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, J400, uint8_t, 1, 1, 1)
+TESTATOB(RGBA, uint8_t, 4, 4, 1, J400, uint8_t, 1, 1, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, RAW, uint8_t, 3, 3, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, RGB24, uint8_t, 3, 3, 1)
+TESTATOB(ABGR, uint8_t, 4, 4, 1, RAW, uint8_t, 3, 3, 1)
+TESTATOB(ABGR, uint8_t, 4, 4, 1, RGB24, uint8_t, 3, 3, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOB(ARGB, uint8_t, 4, 4, 1, RGB565, uint8_t, 2, 2, 1)
+#endif
+TESTATOB(ARGB, uint8_t, 4, 4, 1, RGBA, uint8_t, 4, 4, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, UYVY, uint8_t, 2, 4, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, YUY2, uint8_t, 2, 4, 1)  // 4
+TESTATOB(ARGB1555, uint8_t, 2, 2, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(ARGB4444, uint8_t, 2, 2, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(BGRA, uint8_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(I400, uint8_t, 1, 1, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(I400, uint8_t, 1, 1, 1, I400, uint8_t, 1, 1, 1)
+TESTATOB(I400, uint8_t, 1, 1, 1, I400Mirror, uint8_t, 1, 1, 1)
+TESTATOB(J400, uint8_t, 1, 1, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(J400, uint8_t, 1, 1, 1, J400, uint8_t, 1, 1, 1)
+TESTATOB(RAW, uint8_t, 3, 3, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(RAW, uint8_t, 3, 3, 1, RGBA, uint8_t, 4, 4, 1)
+TESTATOB(RAW, uint8_t, 3, 3, 1, RGB24, uint8_t, 3, 3, 1)
+TESTATOB(RGB24, uint8_t, 3, 3, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(RGB24, uint8_t, 3, 3, 1, J400, uint8_t, 1, 1, 1)
+TESTATOB(RGB24, uint8_t, 3, 3, 1, RGB24Mirror, uint8_t, 3, 3, 1)
+TESTATOB(RAW, uint8_t, 3, 3, 1, J400, uint8_t, 1, 1, 1)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOB(RGB565, uint8_t, 2, 2, 1, ARGB, uint8_t, 4, 4, 1)
+#endif
+TESTATOB(RGBA, uint8_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(UYVY, uint8_t, 2, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(YUY2, uint8_t, 2, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(YUY2, uint8_t, 2, 4, 1, Y, uint8_t, 1, 1, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, AR64, uint16_t, 4, 4, 1)
+TESTATOB(ARGB, uint8_t, 4, 4, 1, AB64, uint16_t, 4, 4, 1)
+TESTATOB(ABGR, uint8_t, 4, 4, 1, AR64, uint16_t, 4, 4, 1)
+TESTATOB(ABGR, uint8_t, 4, 4, 1, AB64, uint16_t, 4, 4, 1)
+TESTATOB(AR64, uint16_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(AB64, uint16_t, 4, 4, 1, ARGB, uint8_t, 4, 4, 1)
+TESTATOB(AR64, uint16_t, 4, 4, 1, ABGR, uint8_t, 4, 4, 1)
+TESTATOB(AB64, uint16_t, 4, 4, 1, ABGR, uint8_t, 4, 4, 1)
+TESTATOB(AR64, uint16_t, 4, 4, 1, AB64, uint16_t, 4, 4, 1)
+TESTATOB(AB64, uint16_t, 4, 4, 1, AR64, uint16_t, 4, 4, 1)
 
 #define TESTATOBDI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
-                   HEIGHT_B, W1280, DIFF, N, NEG, OFF)                       \
+                   HEIGHT_B, W1280, N, NEG, OFF)                             \
   TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##Dither##N) {                   \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                          \
+    const int kWidth = W1280;                                                \
     const int kHeight = benchmark_height_;                                   \
     const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A;     \
     const int kHeightB = (kHeight + HEIGHT_B - 1) / HEIGHT_B * HEIGHT_B;     \
@@ -1235,22 +1511,16 @@ TESTATOB(YUY2, 2, 4, 1, Y, 1, 1, 1, 0)
       FMT_A##To##FMT_B##Dither(src_argb + OFF, kStrideA, dst_argb_opt,       \
                                kStrideB, NULL, kWidth, NEG kHeight);         \
     }                                                                        \
-    int max_diff = 0;                                                        \
     for (int i = 0; i < kStrideB * kHeightB; ++i) {                          \
-      int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -                   \
-                         static_cast<int>(dst_argb_opt[i]));                 \
-      if (abs_diff > max_diff) {                                             \
-        max_diff = abs_diff;                                                 \
-      }                                                                      \
+      EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]);                             \
     }                                                                        \
-    EXPECT_LE(max_diff, DIFF);                                               \
     free_aligned_buffer_page_end(src_argb);                                  \
     free_aligned_buffer_page_end(dst_argb_c);                                \
     free_aligned_buffer_page_end(dst_argb_opt);                              \
   }
 
 #define TESTATOBDRANDOM(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B,        \
-                        STRIDE_B, HEIGHT_B, DIFF)                              \
+                        STRIDE_B, HEIGHT_B)                                    \
   TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##Dither_Random) {                 \
     for (int times = 0; times < benchmark_iterations_; ++times) {              \
       const int kWidth = (fastrand() & 63) + 1;                                \
@@ -1275,15 +1545,9 @@ TESTATOB(YUY2, 2, 4, 1, Y, 1, 1, 1, 0)
       MaskCpuFlags(benchmark_cpu_info_);                                       \
       FMT_A##To##FMT_B##Dither(src_argb, kStrideA, dst_argb_opt, kStrideB,     \
                                NULL, kWidth, kHeight);                         \
-      int max_diff = 0;                                                        \
       for (int i = 0; i < kStrideB * kHeightB; ++i) {                          \
-        int abs_diff = abs(static_cast<int>(dst_argb_c[i]) -                   \
-                           static_cast<int>(dst_argb_opt[i]));                 \
-        if (abs_diff > max_diff) {                                             \
-          max_diff = abs_diff;                                                 \
-        }                                                                      \
+        EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]);                             \
       }                                                                        \
-      EXPECT_LE(max_diff, DIFF);                                               \
       free_aligned_buffer_page_end(src_argb);                                  \
       free_aligned_buffer_page_end(dst_argb_c);                                \
       free_aligned_buffer_page_end(dst_argb_opt);                              \
@@ -1291,49 +1555,57 @@ TESTATOB(YUY2, 2, 4, 1, Y, 1, 1, 1, 0)
   }
 
 #define TESTATOBD(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
-                  HEIGHT_B, DIFF)                                           \
+                  HEIGHT_B)                                                 \
   TESTATOBDI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-             HEIGHT_B, benchmark_width_ - 4, DIFF, _Any, +, 0)              \
+             HEIGHT_B, benchmark_width_ + 1, _Any, +, 0)                    \
   TESTATOBDI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-             HEIGHT_B, benchmark_width_, DIFF, _Unaligned, +, 1)            \
+             HEIGHT_B, benchmark_width_, _Unaligned, +, 2)                  \
   TESTATOBDI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-             HEIGHT_B, benchmark_width_, DIFF, _Invert, -, 0)               \
+             HEIGHT_B, benchmark_width_, _Invert, -, 0)                     \
   TESTATOBDI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B,      \
-             HEIGHT_B, benchmark_width_, DIFF, _Opt, +, 0)                  \
+             HEIGHT_B, benchmark_width_, _Opt, +, 0)                        \
   TESTATOBDRANDOM(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
-                  HEIGHT_B, DIFF)
+                  HEIGHT_B)
 
-TESTATOBD(ARGB, 4, 4, 1, RGB565, 2, 2, 1, 0)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTATOBD(ARGB, 4, 4, 1, RGB565, 2, 2, 1)
+#endif
 
-#define TESTSYMI(FMT_ATOB, BPP_A, STRIDE_A, HEIGHT_A, W1280, N, NEG, OFF)      \
-  TEST_F(LibYUVConvertTest, FMT_ATOB##_Symetric##N) {                          \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+// These conversions called twice, produce the original result.
+// e.g. endian swap twice.
+#define TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, W1280, N, NEG,   \
+                 OFF)                                                          \
+  TEST_F(LibYUVConvertTest, FMT_ATOB##_Endswap##N) {                           \
+    const int kWidth = W1280;                                                  \
     const int kHeight = benchmark_height_;                                     \
     const int kHeightA = (kHeight + HEIGHT_A - 1) / HEIGHT_A * HEIGHT_A;       \
     const int kStrideA =                                                       \
-        (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;                 \
-    align_buffer_page_end(src_argb, kStrideA* kHeightA + OFF);                 \
-    align_buffer_page_end(dst_argb_c, kStrideA* kHeightA);                     \
-    align_buffer_page_end(dst_argb_opt, kStrideA* kHeightA);                   \
-    for (int i = 0; i < kStrideA * kHeightA; ++i) {                            \
+        (kWidth * EPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;                 \
+    align_buffer_page_end(src_argb,                                            \
+                          kStrideA* kHeightA*(int)sizeof(TYPE_A) + OFF);       \
+    align_buffer_page_end(dst_argb_c, kStrideA* kHeightA*(int)sizeof(TYPE_A)); \
+    align_buffer_page_end(dst_argb_opt,                                        \
+                          kStrideA* kHeightA*(int)sizeof(TYPE_A));             \
+    for (int i = 0; i < kStrideA * kHeightA * (int)sizeof(TYPE_A); ++i) {      \
       src_argb[i + OFF] = (fastrand() & 0xff);                                 \
     }                                                                          \
     memset(dst_argb_c, 1, kStrideA* kHeightA);                                 \
     memset(dst_argb_opt, 101, kStrideA* kHeightA);                             \
     MaskCpuFlags(disable_cpu_flags_);                                          \
-    FMT_ATOB(src_argb + OFF, kStrideA, dst_argb_c, kStrideA, kWidth,           \
-             NEG kHeight);                                                     \
+    FMT_ATOB((TYPE_A*)(src_argb + OFF), kStrideA, (TYPE_A*)dst_argb_c,         \
+             kStrideA, kWidth, NEG kHeight);                                   \
     MaskCpuFlags(benchmark_cpu_info_);                                         \
     for (int i = 0; i < benchmark_iterations_; ++i) {                          \
-      FMT_ATOB(src_argb + OFF, kStrideA, dst_argb_opt, kStrideA, kWidth,       \
-               NEG kHeight);                                                   \
+      FMT_ATOB((TYPE_A*)(src_argb + OFF), kStrideA, (TYPE_A*)dst_argb_opt,     \
+               kStrideA, kWidth, NEG kHeight);                                 \
     }                                                                          \
     MaskCpuFlags(disable_cpu_flags_);                                          \
-    FMT_ATOB(dst_argb_c, kStrideA, dst_argb_c, kStrideA, kWidth, NEG kHeight); \
+    FMT_ATOB((TYPE_A*)dst_argb_c, kStrideA, (TYPE_A*)dst_argb_c, kStrideA,     \
+             kWidth, NEG kHeight);                                             \
     MaskCpuFlags(benchmark_cpu_info_);                                         \
-    FMT_ATOB(dst_argb_opt, kStrideA, dst_argb_opt, kStrideA, kWidth,           \
-             NEG kHeight);                                                     \
-    for (int i = 0; i < kStrideA * kHeightA; ++i) {                            \
+    FMT_ATOB((TYPE_A*)dst_argb_opt, kStrideA, (TYPE_A*)dst_argb_opt, kStrideA, \
+             kWidth, NEG kHeight);                                             \
+    for (int i = 0; i < kStrideA * kHeightA * (int)sizeof(TYPE_A); ++i) {      \
       EXPECT_EQ(src_argb[i + OFF], dst_argb_opt[i]);                           \
       EXPECT_EQ(dst_argb_c[i], dst_argb_opt[i]);                               \
     }                                                                          \
@@ -1342,32 +1614,25 @@ TESTATOBD(ARGB, 4, 4, 1, RGB565, 2, 2, 1, 0)
     free_aligned_buffer_page_end(dst_argb_opt);                                \
   }
 
-#define TESTSYM(FMT_ATOB, BPP_A, STRIDE_A, HEIGHT_A)                           \
-  TESTSYMI(FMT_ATOB, BPP_A, STRIDE_A, HEIGHT_A, benchmark_width_ - 4, _Any, +, \
-           0)                                                                  \
-  TESTSYMI(FMT_ATOB, BPP_A, STRIDE_A, HEIGHT_A, benchmark_width_, _Unaligned,  \
-           +, 1)                                                               \
-  TESTSYMI(FMT_ATOB, BPP_A, STRIDE_A, HEIGHT_A, benchmark_width_, _Opt, +, 0)
-
-TESTSYM(ARGBToARGB, 4, 4, 1)
-TESTSYM(ARGBToBGRA, 4, 4, 1)
-TESTSYM(ARGBToABGR, 4, 4, 1)
-TESTSYM(BGRAToARGB, 4, 4, 1)
-TESTSYM(ABGRToARGB, 4, 4, 1)
+#if defined(ENABLE_FULL_TESTS)
+#define TESTEND(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A)                  \
+  TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_ + 1, \
+           _Any, +, 0)                                                        \
+  TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_,     \
+           _Unaligned, +, 2)                                                  \
+  TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_,     \
+           _Opt, +, 0)
+#else
+#define TESTEND(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A)              \
+  TESTENDI(FMT_ATOB, TYPE_A, EPP_A, STRIDE_A, HEIGHT_A, benchmark_width_, \
+           _Opt, +, 0)
+#endif
 
-TEST_F(LibYUVConvertTest, Test565) {
-  SIMD_ALIGNED(uint8_t orig_pixels[256][4]);
-  SIMD_ALIGNED(uint8_t pixels565[256][2]);
-
-  for (int i = 0; i < 256; ++i) {
-    for (int j = 0; j < 4; ++j) {
-      orig_pixels[i][j] = i;
-    }
-  }
-  ARGBToRGB565(&orig_pixels[0][0], 0, &pixels565[0][0], 0, 256, 1);
-  uint32_t checksum = HashDjb2(&pixels565[0][0], sizeof(pixels565), 5381);
-  EXPECT_EQ(610919429u, checksum);
-}
+TESTEND(ARGBToBGRA, uint8_t, 4, 4, 1)
+TESTEND(ARGBToABGR, uint8_t, 4, 4, 1)
+TESTEND(BGRAToARGB, uint8_t, 4, 4, 1)
+TESTEND(ABGRToARGB, uint8_t, 4, 4, 1)
+TESTEND(AB64ToAR64, uint16_t, 4, 4, 1)
 
 #ifdef HAVE_JPEG
 TEST_F(LibYUVConvertTest, ValidateJpeg) {
@@ -1883,6 +2148,65 @@ TEST_F(LibYUVConvertTest, TestMJPGToI420_NV21) {
   free_aligned_buffer_page_end(dst_vu);
 }
 
+TEST_F(LibYUVConvertTest, TestMJPGToI420_NV12) {
+  int width = 0;
+  int height = 0;
+  int ret = MJPGSize(kTest2Jpg, kTest2JpgLen, &width, &height);
+  EXPECT_EQ(0, ret);
+
+  int half_width = (width + 1) / 2;
+  int half_height = (height + 1) / 2;
+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
+                             benchmark_height_ / (width * height);
+
+  // Convert to NV12
+  align_buffer_page_end(dst_y, width * height);
+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
+
+  for (int times = 0; times < benchmark_iterations; ++times) {
+    ret = MJPGToNV12(kTest2Jpg, kTest2JpgLen, dst_y, width, dst_uv,
+                     half_width * 2, width, height, width, height);
+  }
+  // Expect sucesss
+  EXPECT_EQ(0, ret);
+
+  // Convert to I420
+  align_buffer_page_end(dst2_y, width * height);
+  align_buffer_page_end(dst2_u, half_width * half_height);
+  align_buffer_page_end(dst2_v, half_width * half_height);
+  for (int times = 0; times < benchmark_iterations; ++times) {
+    ret = MJPGToI420(kTest2Jpg, kTest2JpgLen, dst2_y, width, dst2_u, half_width,
+                     dst2_v, half_width, width, height, width, height);
+  }
+  // Expect sucesss
+  EXPECT_EQ(0, ret);
+
+  // Convert I420 to NV12
+  align_buffer_page_end(dst3_y, width * height);
+  align_buffer_page_end(dst3_uv, half_width * half_height * 2);
+
+  I420ToNV12(dst2_y, width, dst2_u, half_width, dst2_v, half_width, dst3_y,
+             width, dst3_uv, half_width * 2, width, height);
+
+  for (int i = 0; i < width * height; ++i) {
+    EXPECT_EQ(dst_y[i], dst3_y[i]);
+  }
+  for (int i = 0; i < half_width * half_height * 2; ++i) {
+    EXPECT_EQ(dst_uv[i], dst3_uv[i]);
+    EXPECT_EQ(dst_uv[i], dst3_uv[i]);
+  }
+
+  free_aligned_buffer_page_end(dst3_y);
+  free_aligned_buffer_page_end(dst3_uv);
+
+  free_aligned_buffer_page_end(dst2_y);
+  free_aligned_buffer_page_end(dst2_u);
+  free_aligned_buffer_page_end(dst2_v);
+
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_uv);
+}
+
 TEST_F(LibYUVConvertTest, TestMJPGToNV21_420) {
   int width = 0;
   int height = 0;
@@ -1913,7 +2237,42 @@ TEST_F(LibYUVConvertTest, TestMJPGToNV21_420) {
   free_aligned_buffer_page_end(dst_uv);
 }
 
-TEST_F(LibYUVConvertTest, TestMJPGToNV21_422) {
+TEST_F(LibYUVConvertTest, TestMJPGToNV12_420) {
+  int width = 0;
+  int height = 0;
+  int ret = MJPGSize(kTest2Jpg, kTest2JpgLen, &width, &height);
+  EXPECT_EQ(0, ret);
+
+  int half_width = (width + 1) / 2;
+  int half_height = (height + 1) / 2;
+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
+                             benchmark_height_ / (width * height);
+
+  align_buffer_page_end(dst_y, width * height);
+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
+  for (int times = 0; times < benchmark_iterations; ++times) {
+    ret = MJPGToNV12(kTest2Jpg, kTest2JpgLen, dst_y, width, dst_uv,
+                     half_width * 2, width, height, width, height);
+  }
+  // Expect sucesss
+  EXPECT_EQ(0, ret);
+
+  // Test result matches known hash value. Hashes are for VU so flip the plane.
+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
+  align_buffer_page_end(dst_vu, half_width * half_height * 2);
+  SwapUVPlane(dst_uv, half_width * 2, dst_vu, half_width * 2, half_width,
+              half_height);
+  uint32_t dst_vu_hash = HashDjb2(dst_vu, half_width * half_height * 2, 5381);
+  EXPECT_EQ(dst_y_hash, 2682851208u);
+  EXPECT_EQ(dst_vu_hash, 1069662856u);
+
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_uv);
+  free_aligned_buffer_page_end(dst_vu);
+}
+
+// TODO(fbarchard): Improve test to compare against I422, not checksum
+TEST_F(LibYUVConvertTest, DISABLED_TestMJPGToNV21_422) {
   int width = 0;
   int height = 0;
   int ret = MJPGSize(kTest3Jpg, kTest3JpgLen, &width, &height);
@@ -1937,10 +2296,44 @@ TEST_F(LibYUVConvertTest, TestMJPGToNV21_422) {
   uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
   uint32_t dst_uv_hash = HashDjb2(dst_uv, half_width * half_height * 2, 5381);
   EXPECT_EQ(dst_y_hash, 2682851208u);
-  EXPECT_EQ(dst_uv_hash, 3543430771u);
+  EXPECT_EQ(dst_uv_hash, 493520167u);
+
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_uv);
+}
+
+TEST_F(LibYUVConvertTest, DISABLED_TestMJPGToNV12_422) {
+  int width = 0;
+  int height = 0;
+  int ret = MJPGSize(kTest3Jpg, kTest3JpgLen, &width, &height);
+  EXPECT_EQ(0, ret);
+
+  int half_width = (width + 1) / 2;
+  int half_height = (height + 1) / 2;
+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
+                             benchmark_height_ / (width * height);
+
+  align_buffer_page_end(dst_y, width * height);
+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
+  for (int times = 0; times < benchmark_iterations; ++times) {
+    ret = MJPGToNV12(kTest3Jpg, kTest3JpgLen, dst_y, width, dst_uv,
+                     half_width * 2, width, height, width, height);
+  }
+  // Expect sucesss
+  EXPECT_EQ(0, ret);
+
+  // Test result matches known hash value. Hashes are for VU so flip the plane.
+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
+  align_buffer_page_end(dst_vu, half_width * half_height * 2);
+  SwapUVPlane(dst_uv, half_width * 2, dst_vu, half_width * 2, half_width,
+              half_height);
+  uint32_t dst_vu_hash = HashDjb2(dst_vu, half_width * half_height * 2, 5381);
+  EXPECT_EQ(dst_y_hash, 2682851208u);
+  EXPECT_EQ(dst_vu_hash, 493520167u);
 
   free_aligned_buffer_page_end(dst_y);
   free_aligned_buffer_page_end(dst_uv);
+  free_aligned_buffer_page_end(dst_vu);
 }
 
 TEST_F(LibYUVConvertTest, TestMJPGToNV21_400) {
@@ -1973,6 +2366,40 @@ TEST_F(LibYUVConvertTest, TestMJPGToNV21_400) {
   free_aligned_buffer_page_end(dst_uv);
 }
 
+TEST_F(LibYUVConvertTest, TestMJPGToNV12_400) {
+  int width = 0;
+  int height = 0;
+  int ret = MJPGSize(kTest0Jpg, kTest0JpgLen, &width, &height);
+  EXPECT_EQ(0, ret);
+
+  int half_width = (width + 1) / 2;
+  int half_height = (height + 1) / 2;
+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
+                             benchmark_height_ / (width * height);
+
+  align_buffer_page_end(dst_y, width * height);
+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
+  for (int times = 0; times < benchmark_iterations; ++times) {
+    ret = MJPGToNV12(kTest0Jpg, kTest0JpgLen, dst_y, width, dst_uv,
+                     half_width * 2, width, height, width, height);
+  }
+  // Expect sucesss
+  EXPECT_EQ(0, ret);
+
+  // Test result matches known hash value. Hashes are for VU so flip the plane.
+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
+  align_buffer_page_end(dst_vu, half_width * half_height * 2);
+  SwapUVPlane(dst_uv, half_width * 2, dst_vu, half_width * 2, half_width,
+              half_height);
+  uint32_t dst_vu_hash = HashDjb2(dst_vu, half_width * half_height * 2, 5381);
+  EXPECT_EQ(dst_y_hash, 330644005u);
+  EXPECT_EQ(dst_vu_hash, 135214341u);
+
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_uv);
+  free_aligned_buffer_page_end(dst_vu);
+}
+
 TEST_F(LibYUVConvertTest, TestMJPGToNV21_444) {
   int width = 0;
   int height = 0;
@@ -2003,6 +2430,40 @@ TEST_F(LibYUVConvertTest, TestMJPGToNV21_444) {
   free_aligned_buffer_page_end(dst_uv);
 }
 
+TEST_F(LibYUVConvertTest, TestMJPGToNV12_444) {
+  int width = 0;
+  int height = 0;
+  int ret = MJPGSize(kTest1Jpg, kTest1JpgLen, &width, &height);
+  EXPECT_EQ(0, ret);
+
+  int half_width = (width + 1) / 2;
+  int half_height = (height + 1) / 2;
+  int benchmark_iterations = benchmark_iterations_ * benchmark_width_ *
+                             benchmark_height_ / (width * height);
+
+  align_buffer_page_end(dst_y, width * height);
+  align_buffer_page_end(dst_uv, half_width * half_height * 2);
+  for (int times = 0; times < benchmark_iterations; ++times) {
+    ret = MJPGToNV12(kTest1Jpg, kTest1JpgLen, dst_y, width, dst_uv,
+                     half_width * 2, width, height, width, height);
+  }
+  // Expect sucesss
+  EXPECT_EQ(0, ret);
+
+  // Test result matches known hash value. Hashes are for VU so flip the plane.
+  uint32_t dst_y_hash = HashDjb2(dst_y, width * height, 5381);
+  align_buffer_page_end(dst_vu, half_width * half_height * 2);
+  SwapUVPlane(dst_uv, half_width * 2, dst_vu, half_width * 2, half_width,
+              half_height);
+  uint32_t dst_vu_hash = HashDjb2(dst_vu, half_width * half_height * 2, 5381);
+  EXPECT_EQ(dst_y_hash, 2682851208u);
+  EXPECT_EQ(dst_vu_hash, 506143297u);
+
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_uv);
+  free_aligned_buffer_page_end(dst_vu);
+}
+
 TEST_F(LibYUVConvertTest, TestMJPGToARGB) {
   int width = 0;
   int height = 0;
@@ -2022,7 +2483,11 @@ TEST_F(LibYUVConvertTest, TestMJPGToARGB) {
 
   // Test result matches known hash value.
   uint32_t dst_argb_hash = HashDjb2(dst_argb, width * height, 5381);
+#ifdef LIBYUV_UNLIMITED_DATA
+  EXPECT_EQ(dst_argb_hash, 3900633302u);
+#else
   EXPECT_EQ(dst_argb_hash, 2355976473u);
+#endif
 
   free_aligned_buffer_page_end(dst_argb);
 }
@@ -2178,7 +2643,7 @@ TEST_F(LibYUVConvertTest, I420CropOddY) {
   const int SUBSAMP_Y = 2;
   const int kWidth = benchmark_width_;
   const int kHeight = benchmark_height_;
-  const int crop_y = 1;
+  const int crop_y = benchmark_height_ > 1 ? 1 : 0;
   const int kDestWidth = benchmark_width_;
   const int kDestHeight = benchmark_height_ - crop_y * 2;
   const int kStrideU = SUBSAMPLE(kWidth, SUBSAMP_X);
@@ -2329,9 +2794,9 @@ TEST_F(LibYUVConvertTest, TestDither) {
 }
 
 #define TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
-                        YALIGN, W1280, DIFF, N, NEG, OFF, FMT_C, BPP_C)        \
+                        YALIGN, W1280, N, NEG, OFF, FMT_C, BPP_C)              \
   TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##Dither##N) {                \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+    const int kWidth = W1280;                                                  \
     const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                   \
     const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                      \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                        \
@@ -2360,7 +2825,6 @@ TEST_F(LibYUVConvertTest, TestDither) {
           src_y + OFF, kWidth, src_u + OFF, kStrideUV, src_v + OFF, kStrideUV, \
           dst_argb_opt + OFF, kStrideB, NULL, kWidth, NEG kHeight);            \
     }                                                                          \
-    int max_diff = 0;                                                          \
     /* Convert to ARGB so 565 is expanded to bytes that can be compared. */    \
     align_buffer_page_end(dst_argb32_c, kWidth* BPP_C* kHeight);               \
     align_buffer_page_end(dst_argb32_opt, kWidth* BPP_C* kHeight);             \
@@ -2371,13 +2835,8 @@ TEST_F(LibYUVConvertTest, TestDither) {
     FMT_B##To##FMT_C(dst_argb_opt + OFF, kStrideB, dst_argb32_opt,             \
                      kWidth * BPP_C, kWidth, kHeight);                         \
     for (int i = 0; i < kWidth * BPP_C * kHeight; ++i) {                       \
-      int abs_diff = abs(static_cast<int>(dst_argb32_c[i]) -                   \
-                         static_cast<int>(dst_argb32_opt[i]));                 \
-      if (abs_diff > max_diff) {                                               \
-        max_diff = abs_diff;                                                   \
-      }                                                                        \
+      EXPECT_EQ(dst_argb32_c[i], dst_argb32_opt[i]);                           \
     }                                                                          \
-    EXPECT_LE(max_diff, DIFF);                                                 \
     free_aligned_buffer_page_end(src_y);                                       \
     free_aligned_buffer_page_end(src_u);                                       \
     free_aligned_buffer_page_end(src_v);                                       \
@@ -2387,20 +2846,20 @@ TEST_F(LibYUVConvertTest, TestDither) {
     free_aligned_buffer_page_end(dst_argb32_opt);                              \
   }
 
-#define TESTPLANARTOBD(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
-                       YALIGN, DIFF, FMT_C, BPP_C)                             \
-  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,       \
-                  YALIGN, benchmark_width_ - 4, DIFF, _Any, +, 0, FMT_C,       \
-                  BPP_C)                                                       \
-  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,       \
-                  YALIGN, benchmark_width_, DIFF, _Unaligned, +, 1, FMT_C,     \
-                  BPP_C)                                                       \
-  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,       \
-                  YALIGN, benchmark_width_, DIFF, _Invert, -, 0, FMT_C, BPP_C) \
-  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,       \
-                  YALIGN, benchmark_width_, DIFF, _Opt, +, 0, FMT_C, BPP_C)
-
-TESTPLANARTOBD(I420, 2, 2, RGB565, 2, 2, 1, 9, ARGB, 4)
+#define TESTPLANARTOBD(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
+                       YALIGN, FMT_C, BPP_C)                                  \
+  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                  YALIGN, benchmark_width_ + 1, _Any, +, 0, FMT_C, BPP_C)     \
+  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                  YALIGN, benchmark_width_, _Unaligned, +, 2, FMT_C, BPP_C)   \
+  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                  YALIGN, benchmark_width_, _Invert, -, 0, FMT_C, BPP_C)      \
+  TESTPLANARTOBID(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
+                  YALIGN, benchmark_width_, _Opt, +, 0, FMT_C, BPP_C)
+
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTPLANARTOBD(I420, 2, 2, RGB565, 2, 2, 1, ARGB, 4)
+#endif
 
 #define TESTPTOB(NAME, UYVYTOI420, UYVYTONV12)                                \
   TEST_F(LibYUVConvertTest, NAME) {                                           \
@@ -2462,12 +2921,12 @@ TESTPLANARTOBD(I420, 2, 2, RGB565, 2, 2, 1, 9, ARGB, 4)
 TESTPTOB(TestYUY2ToNV12, YUY2ToI420, YUY2ToNV12)
 TESTPTOB(TestUYVYToNV12, UYVYToI420, UYVYToNV12)
 
-// Transitive tests.  A to B to C is same as A to C.
-
+// Transitive test.  A to B to C is same as A to C.
+// Benchmarks A To B to C for comparison to 1 step, benchmarked elsewhere.
 #define TESTPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B, \
                        W1280, N, NEG, OFF, FMT_C, BPP_C)                      \
-  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##_##FMT_C##N) {             \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##To##FMT_C##N) {            \
+    const int kWidth = W1280;                                                 \
     const int kHeight = benchmark_height_;                                    \
     const int kStrideB = SUBSAMPLE(kWidth, SUB_B) * BPP_B;                    \
     const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                       \
@@ -2484,23 +2943,23 @@ TESTPTOB(TestUYVYToNV12, UYVYToI420, UYVYToNV12)
       src_v[i + OFF] = (fastrand() & 0xff);                                   \
     }                                                                         \
     memset(dst_argb_b + OFF, 1, kStrideB * kHeight);                          \
-    for (int i = 0; i < benchmark_iterations_; ++i) {                         \
-      FMT_PLANAR##To##FMT_B(src_y + OFF, kWidth, src_u + OFF, kStrideUV,      \
-                            src_v + OFF, kStrideUV, dst_argb_b + OFF,         \
-                            kStrideB, kWidth, NEG kHeight);                   \
-    }                                                                         \
+    FMT_PLANAR##To##FMT_B(src_y + OFF, kWidth, src_u + OFF, kStrideUV,        \
+                          src_v + OFF, kStrideUV, dst_argb_b + OFF, kStrideB, \
+                          kWidth, NEG kHeight);                               \
     /* Convert to a 3rd format in 1 step and 2 steps and compare  */          \
     const int kStrideC = kWidth * BPP_C;                                      \
     align_buffer_page_end(dst_argb_c, kStrideC* kHeight + OFF);               \
     align_buffer_page_end(dst_argb_bc, kStrideC* kHeight + OFF);              \
     memset(dst_argb_c + OFF, 2, kStrideC * kHeight);                          \
     memset(dst_argb_bc + OFF, 3, kStrideC * kHeight);                         \
-    FMT_PLANAR##To##FMT_C(src_y + OFF, kWidth, src_u + OFF, kStrideUV,        \
-                          src_v + OFF, kStrideUV, dst_argb_c + OFF, kStrideC, \
-                          kWidth, NEG kHeight);                               \
-    /* Convert B to C */                                                      \
-    FMT_B##To##FMT_C(dst_argb_b + OFF, kStrideB, dst_argb_bc + OFF, kStrideC, \
-                     kWidth, kHeight);                                        \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                         \
+      FMT_PLANAR##To##FMT_C(src_y + OFF, kWidth, src_u + OFF, kStrideUV,      \
+                            src_v + OFF, kStrideUV, dst_argb_c + OFF,         \
+                            kStrideC, kWidth, NEG kHeight);                   \
+      /* Convert B to C */                                                    \
+      FMT_B##To##FMT_C(dst_argb_b + OFF, kStrideB, dst_argb_bc + OFF,         \
+                       kStrideC, kWidth, kHeight);                            \
+    }                                                                         \
     for (int i = 0; i < kStrideC * kHeight; ++i) {                            \
       EXPECT_EQ(dst_argb_c[i + OFF], dst_argb_bc[i + OFF]);                   \
     }                                                                         \
@@ -2512,58 +2971,109 @@ TESTPTOB(TestUYVYToNV12, UYVYToI420, UYVYToNV12)
     free_aligned_buffer_page_end(dst_argb_bc);                                \
   }
 
+#if defined(ENABLE_FULL_TESTS)
 #define TESTPLANARTOE(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B, \
                       FMT_C, BPP_C)                                          \
   TESTPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
-                 benchmark_width_ - 4, _Any, +, 0, FMT_C, BPP_C)             \
+                 benchmark_width_ + 1, _Any, +, 0, FMT_C, BPP_C)             \
   TESTPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
-                 benchmark_width_, _Unaligned, +, 1, FMT_C, BPP_C)           \
+                 benchmark_width_, _Unaligned, +, 2, FMT_C, BPP_C)           \
   TESTPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
                  benchmark_width_, _Invert, -, 0, FMT_C, BPP_C)              \
   TESTPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
                  benchmark_width_, _Opt, +, 0, FMT_C, BPP_C)
+#else
+#define TESTPLANARTOE(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B, \
+                      FMT_C, BPP_C)                                          \
+  TESTPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
+                 benchmark_width_, _Opt, +, 0, FMT_C, BPP_C)
+#endif
 
+#if defined(ENABLE_FULL_TESTS)
+TESTPLANARTOE(I420, 2, 2, ABGR, 1, 4, ARGB, 4)
 TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, ABGR, 4)
-TESTPLANARTOE(J420, 2, 2, ARGB, 1, 4, ARGB, 4)
-TESTPLANARTOE(J420, 2, 2, ABGR, 1, 4, ARGB, 4)
-TESTPLANARTOE(H420, 2, 2, ARGB, 1, 4, ARGB, 4)
-TESTPLANARTOE(H420, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RAW, 3)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RGB24, 3)
 TESTPLANARTOE(I420, 2, 2, BGRA, 1, 4, ARGB, 4)
-TESTPLANARTOE(I420, 2, 2, ABGR, 1, 4, ARGB, 4)
-TESTPLANARTOE(I420, 2, 2, RGBA, 1, 4, ARGB, 4)
-TESTPLANARTOE(I420, 2, 2, RGB24, 1, 3, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, RAW, 1, 3, ARGB, 4)
 TESTPLANARTOE(I420, 2, 2, RAW, 1, 3, RGB24, 3)
+TESTPLANARTOE(I420, 2, 2, RGB24, 1, 3, ARGB, 4)
 TESTPLANARTOE(I420, 2, 2, RGB24, 1, 3, RAW, 3)
-TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RAW, 3)
-TESTPLANARTOE(I420, 2, 2, RAW, 1, 3, ARGB, 4)
-TESTPLANARTOE(H420, 2, 2, RGB24, 1, 3, ARGB, 4)
-TESTPLANARTOE(H420, 2, 2, RAW, 1, 3, RGB24, 3)
-TESTPLANARTOE(H420, 2, 2, RGB24, 1, 3, RAW, 3)
+TESTPLANARTOE(I420, 2, 2, RGBA, 1, 4, ARGB, 4)
+TESTPLANARTOE(H420, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(H420, 2, 2, ARGB, 1, 4, ABGR, 4)
 TESTPLANARTOE(H420, 2, 2, ARGB, 1, 4, RAW, 3)
+TESTPLANARTOE(H420, 2, 2, ARGB, 1, 4, RGB24, 3)
 TESTPLANARTOE(H420, 2, 2, RAW, 1, 3, ARGB, 4)
+TESTPLANARTOE(H420, 2, 2, RAW, 1, 3, RGB24, 3)
+TESTPLANARTOE(H420, 2, 2, RGB24, 1, 3, ARGB, 4)
+TESTPLANARTOE(H420, 2, 2, RGB24, 1, 3, RAW, 3)
+TESTPLANARTOE(J420, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(J420, 2, 2, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(U420, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(U420, 2, 2, ARGB, 1, 4, ARGB, 4)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
 TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RGB565, 2)
 TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, ARGB1555, 2)
 TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, ARGB4444, 2)
 TESTPLANARTOE(I422, 2, 1, ARGB, 1, 4, RGB565, 2)
+#endif
+TESTPLANARTOE(I422, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTPLANARTOE(I422, 2, 1, ABGR, 1, 4, ARGB, 4)
 TESTPLANARTOE(J422, 2, 1, ARGB, 1, 4, ARGB, 4)
 TESTPLANARTOE(J422, 2, 1, ABGR, 1, 4, ARGB, 4)
 TESTPLANARTOE(H422, 2, 1, ARGB, 1, 4, ARGB, 4)
 TESTPLANARTOE(H422, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(U422, 2, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(U422, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(V422, 2, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(V422, 2, 1, ABGR, 1, 4, ARGB, 4)
 TESTPLANARTOE(I422, 2, 1, BGRA, 1, 4, ARGB, 4)
-TESTPLANARTOE(I422, 2, 1, ABGR, 1, 4, ARGB, 4)
 TESTPLANARTOE(I422, 2, 1, RGBA, 1, 4, ARGB, 4)
-TESTPLANARTOE(I444, 1, 1, ARGB, 1, 4, ARGB, 4)
-TESTPLANARTOE(J444, 1, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(I444, 1, 1, ARGB, 1, 4, ABGR, 4)
 TESTPLANARTOE(I444, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(J444, 1, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(J444, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(H444, 1, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(H444, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(U444, 1, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(U444, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(V444, 1, 1, ARGB, 1, 4, ARGB, 4)
+TESTPLANARTOE(V444, 1, 1, ABGR, 1, 4, ARGB, 4)
 TESTPLANARTOE(I420, 2, 2, YUY2, 2, 4, ARGB, 4)
 TESTPLANARTOE(I420, 2, 2, UYVY, 2, 4, ARGB, 4)
 TESTPLANARTOE(I422, 2, 1, YUY2, 2, 4, ARGB, 4)
 TESTPLANARTOE(I422, 2, 1, UYVY, 2, 4, ARGB, 4)
+#else
+TESTPLANARTOE(I420, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, ARGB1555, 2)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, ARGB4444, 2)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RAW, 3)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RGB24, 3)
+TESTPLANARTOE(I420, 2, 2, ARGB, 1, 4, RGB565, 2)
+TESTPLANARTOE(I420, 2, 2, BGRA, 1, 4, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, RAW, 1, 3, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, RAW, 1, 3, RGB24, 3)
+TESTPLANARTOE(I420, 2, 2, RGB24, 1, 3, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, RGB24, 1, 3, RAW, 3)
+TESTPLANARTOE(I420, 2, 2, RGBA, 1, 4, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, UYVY, 2, 4, ARGB, 4)
+TESTPLANARTOE(I420, 2, 2, YUY2, 2, 4, ARGB, 4)
+TESTPLANARTOE(I422, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTPLANARTOE(I422, 2, 1, ARGB, 1, 4, RGB565, 2)
+TESTPLANARTOE(I422, 2, 1, BGRA, 1, 4, ARGB, 4)
+TESTPLANARTOE(I422, 2, 1, RGBA, 1, 4, ARGB, 4)
+TESTPLANARTOE(I422, 2, 1, UYVY, 2, 4, ARGB, 4)
+TESTPLANARTOE(I422, 2, 1, YUY2, 2, 4, ARGB, 4)
+TESTPLANARTOE(I444, 1, 1, ABGR, 1, 4, ARGB, 4)
+#endif
 
+// Transitive test: Compare 1 step vs 2 step conversion for YUVA to ARGB.
+// Benchmark 2 step conversion for comparison to 1 step conversion.
 #define TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B, \
                         W1280, N, NEG, OFF, FMT_C, BPP_C, ATTEN)               \
-  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##_##FMT_C##N) {              \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##To##FMT_C##N) {             \
+    const int kWidth = W1280;                                                  \
     const int kHeight = benchmark_height_;                                     \
     const int kStrideB = SUBSAMPLE(kWidth, SUB_B) * BPP_B;                     \
     const int kSizeUV =                                                        \
@@ -2573,6 +3083,12 @@ TESTPLANARTOE(I422, 2, 1, UYVY, 2, 4, ARGB, 4)
     align_buffer_page_end(src_v, kSizeUV + OFF);                               \
     align_buffer_page_end(src_a, kWidth* kHeight + OFF);                       \
     align_buffer_page_end(dst_argb_b, kStrideB* kHeight + OFF);                \
+    const int kStrideC = kWidth * BPP_C;                                       \
+    align_buffer_page_end(dst_argb_c, kStrideC* kHeight + OFF);                \
+    align_buffer_page_end(dst_argb_bc, kStrideC* kHeight + OFF);               \
+    memset(dst_argb_c + OFF, 2, kStrideC * kHeight);                           \
+    memset(dst_argb_b + OFF, 1, kStrideB * kHeight);                           \
+    memset(dst_argb_bc + OFF, 3, kStrideC * kHeight);                          \
     for (int i = 0; i < kWidth * kHeight; ++i) {                               \
       src_y[i + OFF] = (fastrand() & 0xff);                                    \
       src_a[i + OFF] = (fastrand() & 0xff);                                    \
@@ -2581,26 +3097,21 @@ TESTPLANARTOE(I422, 2, 1, UYVY, 2, 4, ARGB, 4)
       src_u[i + OFF] = (fastrand() & 0xff);                                    \
       src_v[i + OFF] = (fastrand() & 0xff);                                    \
     }                                                                          \
-    memset(dst_argb_b + OFF, 1, kStrideB * kHeight);                           \
     for (int i = 0; i < benchmark_iterations_; ++i) {                          \
+      /* Convert A to B */                                                     \
       FMT_PLANAR##To##FMT_B(                                                   \
           src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SUBSAMP_X),      \
           src_v + OFF, SUBSAMPLE(kWidth, SUBSAMP_X), src_a + OFF, kWidth,      \
           dst_argb_b + OFF, kStrideB, kWidth, NEG kHeight, ATTEN);             \
+      /* Convert B to C */                                                     \
+      FMT_B##To##FMT_C(dst_argb_b + OFF, kStrideB, dst_argb_bc + OFF,          \
+                       kStrideC, kWidth, kHeight);                             \
     }                                                                          \
-    /* Convert to a 3rd format in 1 step and 2 steps and compare  */           \
-    const int kStrideC = kWidth * BPP_C;                                       \
-    align_buffer_page_end(dst_argb_c, kStrideC* kHeight + OFF);                \
-    align_buffer_page_end(dst_argb_bc, kStrideC* kHeight + OFF);               \
-    memset(dst_argb_c + OFF, 2, kStrideC * kHeight);                           \
-    memset(dst_argb_bc + OFF, 3, kStrideC * kHeight);                          \
+    /* Convert A to C */                                                       \
     FMT_PLANAR##To##FMT_C(                                                     \
         src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SUBSAMP_X),        \
         src_v + OFF, SUBSAMPLE(kWidth, SUBSAMP_X), src_a + OFF, kWidth,        \
         dst_argb_c + OFF, kStrideC, kWidth, NEG kHeight, ATTEN);               \
-    /* Convert B to C */                                                       \
-    FMT_B##To##FMT_C(dst_argb_b + OFF, kStrideB, dst_argb_bc + OFF, kStrideC,  \
-                     kWidth, kHeight);                                         \
     for (int i = 0; i < kStrideC * kHeight; ++i) {                             \
       EXPECT_EQ(dst_argb_c[i + OFF], dst_argb_bc[i + OFF]);                    \
     }                                                                          \
@@ -2613,26 +3124,71 @@ TESTPLANARTOE(I422, 2, 1, UYVY, 2, 4, ARGB, 4)
     free_aligned_buffer_page_end(dst_argb_bc);                                 \
   }
 
+#if defined(ENABLE_FULL_TESTS)
 #define TESTQPLANARTOE(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B, \
                        FMT_C, BPP_C)                                          \
   TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
-                  benchmark_width_ - 4, _Any, +, 0, FMT_C, BPP_C, 0)          \
+                  benchmark_width_ + 1, _Any, +, 0, FMT_C, BPP_C, 0)          \
   TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
-                  benchmark_width_, _Unaligned, +, 1, FMT_C, BPP_C, 0)        \
+                  benchmark_width_, _Unaligned, +, 2, FMT_C, BPP_C, 0)        \
   TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
                   benchmark_width_, _Invert, -, 0, FMT_C, BPP_C, 0)           \
   TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
                   benchmark_width_, _Opt, +, 0, FMT_C, BPP_C, 0)              \
   TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
                   benchmark_width_, _Premult, +, 0, FMT_C, BPP_C, 1)
+#else
+#define TESTQPLANARTOE(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B, \
+                       FMT_C, BPP_C)                                          \
+  TESTQPLANARTOEI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, SUB_B, BPP_B,      \
+                  benchmark_width_, _Opt, +, 0, FMT_C, BPP_C, 0)
+#endif
 
+#if defined(ENABLE_FULL_TESTS)
 TESTQPLANARTOE(I420Alpha, 2, 2, ARGB, 1, 4, ABGR, 4)
 TESTQPLANARTOE(I420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(J420Alpha, 2, 2, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(J420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(H420Alpha, 2, 2, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(H420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(F420Alpha, 2, 2, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(F420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(U420Alpha, 2, 2, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(U420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(V420Alpha, 2, 2, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(V420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(I422Alpha, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(I422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(J422Alpha, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(J422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(F422Alpha, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(F422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(H422Alpha, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(H422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(U422Alpha, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(U422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(V422Alpha, 2, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(V422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(I444Alpha, 1, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(I444Alpha, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(J444Alpha, 1, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(J444Alpha, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(H444Alpha, 1, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(H444Alpha, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(U444Alpha, 1, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(U444Alpha, 1, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(V444Alpha, 1, 1, ARGB, 1, 4, ABGR, 4)
+TESTQPLANARTOE(V444Alpha, 1, 1, ABGR, 1, 4, ARGB, 4)
+#else
+TESTQPLANARTOE(I420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(I422Alpha, 2, 1, ABGR, 1, 4, ARGB, 4)
+TESTQPLANARTOE(I444Alpha, 1, 1, ABGR, 1, 4, ARGB, 4)
+#endif
 
 #define TESTPLANETOEI(FMT_A, SUB_A, BPP_A, FMT_B, SUB_B, BPP_B, W1280, N, NEG, \
                       OFF, FMT_C, BPP_C)                                       \
-  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##_##FMT_C##N) {                   \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                            \
+  TEST_F(LibYUVConvertTest, FMT_A##To##FMT_B##To##FMT_C##N) {                  \
+    const int kWidth = W1280;                                                  \
     const int kHeight = benchmark_height_;                                     \
     const int kStrideA = SUBSAMPLE(kWidth, SUB_A) * BPP_A;                     \
     const int kStrideB = SUBSAMPLE(kWidth, SUB_B) * BPP_B;                     \
@@ -2640,21 +3196,21 @@ TESTQPLANARTOE(I420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
     align_buffer_page_end(dst_argb_b, kStrideB* kHeight + OFF);                \
     MemRandomize(src_argb_a + OFF, kStrideA * kHeight);                        \
     memset(dst_argb_b + OFF, 1, kStrideB * kHeight);                           \
-    for (int i = 0; i < benchmark_iterations_; ++i) {                          \
-      FMT_A##To##FMT_B(src_argb_a + OFF, kStrideA, dst_argb_b + OFF, kStrideB, \
-                       kWidth, NEG kHeight);                                   \
-    }                                                                          \
+    FMT_A##To##FMT_B(src_argb_a + OFF, kStrideA, dst_argb_b + OFF, kStrideB,   \
+                     kWidth, NEG kHeight);                                     \
     /* Convert to a 3rd format in 1 step and 2 steps and compare  */           \
     const int kStrideC = kWidth * BPP_C;                                       \
     align_buffer_page_end(dst_argb_c, kStrideC* kHeight + OFF);                \
     align_buffer_page_end(dst_argb_bc, kStrideC* kHeight + OFF);               \
     memset(dst_argb_c + OFF, 2, kStrideC * kHeight);                           \
     memset(dst_argb_bc + OFF, 3, kStrideC * kHeight);                          \
-    FMT_A##To##FMT_C(src_argb_a + OFF, kStrideA, dst_argb_c + OFF, kStrideC,   \
-                     kWidth, NEG kHeight);                                     \
-    /* Convert B to C */                                                       \
-    FMT_B##To##FMT_C(dst_argb_b + OFF, kStrideB, dst_argb_bc + OFF, kStrideC,  \
-                     kWidth, kHeight);                                         \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                          \
+      FMT_A##To##FMT_C(src_argb_a + OFF, kStrideA, dst_argb_c + OFF, kStrideC, \
+                       kWidth, NEG kHeight);                                   \
+      /* Convert B to C */                                                     \
+      FMT_B##To##FMT_C(dst_argb_b + OFF, kStrideB, dst_argb_bc + OFF,          \
+                       kStrideC, kWidth, kHeight);                             \
+    }                                                                          \
     for (int i = 0; i < kStrideC * kHeight; i += 4) {                          \
       EXPECT_EQ(dst_argb_c[i + OFF + 0], dst_argb_bc[i + OFF + 0]);            \
       EXPECT_EQ(dst_argb_c[i + OFF + 1], dst_argb_bc[i + OFF + 1]);            \
@@ -2669,15 +3225,16 @@ TESTQPLANARTOE(I420Alpha, 2, 2, ABGR, 1, 4, ARGB, 4)
 
 #define TESTPLANETOE(FMT_A, SUB_A, BPP_A, FMT_B, SUB_B, BPP_B, FMT_C, BPP_C) \
   TESTPLANETOEI(FMT_A, SUB_A, BPP_A, FMT_B, SUB_B, BPP_B,                    \
-                benchmark_width_ - 4, _Any, +, 0, FMT_C, BPP_C)              \
+                benchmark_width_ + 1, _Any, +, 0, FMT_C, BPP_C)              \
   TESTPLANETOEI(FMT_A, SUB_A, BPP_A, FMT_B, SUB_B, BPP_B, benchmark_width_,  \
-                _Unaligned, +, 1, FMT_C, BPP_C)                              \
+                _Unaligned, +, 4, FMT_C, BPP_C)                              \
   TESTPLANETOEI(FMT_A, SUB_A, BPP_A, FMT_B, SUB_B, BPP_B, benchmark_width_,  \
                 _Invert, -, 0, FMT_C, BPP_C)                                 \
   TESTPLANETOEI(FMT_A, SUB_A, BPP_A, FMT_B, SUB_B, BPP_B, benchmark_width_,  \
                 _Opt, +, 0, FMT_C, BPP_C)
 
 // Caveat: Destination needs to be 4 bytes
+#ifdef LITTLE_ENDIAN_ONLY_TEST
 TESTPLANETOE(ARGB, 1, 4, AR30, 1, 4, ARGB, 4)
 TESTPLANETOE(ABGR, 1, 4, AR30, 1, 4, ABGR, 4)
 TESTPLANETOE(AR30, 1, 4, ARGB, 1, 4, ABGR, 4)
@@ -2686,6 +3243,7 @@ TESTPLANETOE(ARGB, 1, 4, AB30, 1, 4, ARGB, 4)
 TESTPLANETOE(ABGR, 1, 4, AB30, 1, 4, ABGR, 4)
 TESTPLANETOE(AB30, 1, 4, ARGB, 1, 4, ABGR, 4)
 TESTPLANETOE(AB30, 1, 4, ABGR, 1, 4, ARGB, 4)
+#endif
 
 TEST_F(LibYUVConvertTest, RotateWithARGBSource) {
   // 2x2 frames
@@ -2790,79 +3348,506 @@ TEST_F(LibYUVConvertTest, ABGRToAR30Row_Opt) {
 }
 #endif  // HAS_ABGRTOAR30ROW_AVX2
 
+// Provide matrix wrappers for 12 bit YUV
+#define I012ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I012ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j)
+#define I012ToAR30(a, b, c, d, e, f, g, h, i, j) \
+  I012ToAR30Matrix(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j)
+
+#define I410ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I410ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j)
+#define I410ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I410ToABGRMatrix(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j)
+#define H410ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I410ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuvH709Constants, i, j)
+#define H410ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I410ToABGRMatrix(a, b, c, d, e, f, g, h, &kYuvH709Constants, i, j)
+#define U410ToARGB(a, b, c, d, e, f, g, h, i, j) \
+  I410ToARGBMatrix(a, b, c, d, e, f, g, h, &kYuv2020Constants, i, j)
+#define U410ToABGR(a, b, c, d, e, f, g, h, i, j) \
+  I410ToABGRMatrix(a, b, c, d, e, f, g, h, &kYuv2020Constants, i, j)
+#define I410ToAR30(a, b, c, d, e, f, g, h, i, j) \
+  I410ToAR30Matrix(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j)
+#define I410ToAB30(a, b, c, d, e, f, g, h, i, j) \
+  I410ToAB30Matrix(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j)
+#define H410ToAR30(a, b, c, d, e, f, g, h, i, j) \
+  I410ToAR30Matrix(a, b, c, d, e, f, g, h, &kYuvH709Constants, i, j)
+#define H410ToAB30(a, b, c, d, e, f, g, h, i, j) \
+  I410ToAB30Matrix(a, b, c, d, e, f, g, h, &kYuvH709Constants, i, j)
+#define U410ToAR30(a, b, c, d, e, f, g, h, i, j) \
+  I410ToAR30Matrix(a, b, c, d, e, f, g, h, &kYuv2020Constants, i, j)
+#define U410ToAB30(a, b, c, d, e, f, g, h, i, j) \
+  I410ToAB30Matrix(a, b, c, d, e, f, g, h, &kYuv2020Constants, i, j)
+
+#define I010ToARGBFilter(a, b, c, d, e, f, g, h, i, j)                     \
+  I010ToARGBMatrixFilter(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j, \
+                         kFilterBilinear)
+#define I010ToAR30Filter(a, b, c, d, e, f, g, h, i, j)                     \
+  I010ToAR30MatrixFilter(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j, \
+                         kFilterBilinear)
+#define I210ToARGBFilter(a, b, c, d, e, f, g, h, i, j)                     \
+  I210ToARGBMatrixFilter(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j, \
+                         kFilterBilinear)
+#define I210ToAR30Filter(a, b, c, d, e, f, g, h, i, j)                     \
+  I210ToAR30MatrixFilter(a, b, c, d, e, f, g, h, &kYuvI601Constants, i, j, \
+                         kFilterBilinear)
+
 // TODO(fbarchard): Fix clamping issue affected by U channel.
-#define TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B,   \
-                         ALIGN, YALIGN, W1280, DIFF, N, NEG, SOFF, DOFF)   \
-  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                    \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                        \
-    const int kHeight = ALIGNINT(benchmark_height_, YALIGN);               \
-    const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                  \
-    const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                    \
-    const int kSizeUV = kStrideUV * SUBSAMPLE(kHeight, SUBSAMP_Y);         \
-    const int kBpc = 2;                                                    \
-    align_buffer_page_end(src_y, kWidth* kHeight* kBpc + SOFF);            \
-    align_buffer_page_end(src_u, kSizeUV* kBpc + SOFF);                    \
-    align_buffer_page_end(src_v, kSizeUV* kBpc + SOFF);                    \
-    align_buffer_page_end(dst_argb_c, kStrideB* kHeight + DOFF);           \
-    align_buffer_page_end(dst_argb_opt, kStrideB* kHeight + DOFF);         \
-    for (int i = 0; i < kWidth * kHeight; ++i) {                           \
-      reinterpret_cast<uint16_t*>(src_y + SOFF)[i] = (fastrand() & 0x3ff); \
-    }                                                                      \
-    for (int i = 0; i < kSizeUV; ++i) {                                    \
-      reinterpret_cast<uint16_t*>(src_u + SOFF)[i] = (fastrand() & 0x3ff); \
-      reinterpret_cast<uint16_t*>(src_v + SOFF)[i] = (fastrand() & 0x3ff); \
-    }                                                                      \
-    memset(dst_argb_c + DOFF, 1, kStrideB * kHeight);                      \
-    memset(dst_argb_opt + DOFF, 101, kStrideB * kHeight);                  \
-    MaskCpuFlags(disable_cpu_flags_);                                      \
-    FMT_PLANAR##To##FMT_B(                                                 \
-        reinterpret_cast<uint16_t*>(src_y + SOFF), kWidth,                 \
-        reinterpret_cast<uint16_t*>(src_u + SOFF), kStrideUV,              \
-        reinterpret_cast<uint16_t*>(src_v + SOFF), kStrideUV,              \
-        dst_argb_c + DOFF, kStrideB, kWidth, NEG kHeight);                 \
-    MaskCpuFlags(benchmark_cpu_info_);                                     \
-    for (int i = 0; i < benchmark_iterations_; ++i) {                      \
-      FMT_PLANAR##To##FMT_B(                                               \
-          reinterpret_cast<uint16_t*>(src_y + SOFF), kWidth,               \
-          reinterpret_cast<uint16_t*>(src_u + SOFF), kStrideUV,            \
-          reinterpret_cast<uint16_t*>(src_v + SOFF), kStrideUV,            \
-          dst_argb_opt + DOFF, kStrideB, kWidth, NEG kHeight);             \
-    }                                                                      \
-    int max_diff = 0;                                                      \
-    for (int i = 0; i < kWidth * BPP_B * kHeight; ++i) {                   \
-      int abs_diff = abs(static_cast<int>(dst_argb_c[i + DOFF]) -          \
-                         static_cast<int>(dst_argb_opt[i + DOFF]));        \
-      if (abs_diff > max_diff) {                                           \
-        max_diff = abs_diff;                                               \
-      }                                                                    \
-    }                                                                      \
-    EXPECT_LE(max_diff, DIFF);                                             \
-    free_aligned_buffer_page_end(src_y);                                   \
-    free_aligned_buffer_page_end(src_u);                                   \
-    free_aligned_buffer_page_end(src_v);                                   \
-    free_aligned_buffer_page_end(dst_argb_c);                              \
-    free_aligned_buffer_page_end(dst_argb_opt);                            \
-  }
-
-#define TESTPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
-                        YALIGN, DIFF)                                          \
-  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                   YALIGN, benchmark_width_ - 4, DIFF, _Any, +, 0, 0)          \
-  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                   YALIGN, benchmark_width_, DIFF, _Unaligned, +, 1, 1)        \
-  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                   YALIGN, benchmark_width_, DIFF, _Invert, -, 0, 0)           \
-  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,      \
-                   YALIGN, benchmark_width_, DIFF, _Opt, +, 0, 0)
-
-TESTPLANAR16TOB(I010, 2, 2, ARGB, 4, 4, 1, 2)
-TESTPLANAR16TOB(I010, 2, 2, ABGR, 4, 4, 1, 2)
-TESTPLANAR16TOB(I010, 2, 2, AR30, 4, 4, 1, 2)
-TESTPLANAR16TOB(I010, 2, 2, AB30, 4, 4, 1, 2)
-TESTPLANAR16TOB(H010, 2, 2, ARGB, 4, 4, 1, 2)
-TESTPLANAR16TOB(H010, 2, 2, ABGR, 4, 4, 1, 2)
-TESTPLANAR16TOB(H010, 2, 2, AR30, 4, 4, 1, 2)
-TESTPLANAR16TOB(H010, 2, 2, AB30, 4, 4, 1, 2)
+#define TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_MASK, FMT_B,   \
+                         BPP_B, ALIGN, YALIGN, W1280, N, NEG, SOFF, DOFF)     \
+  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                       \
+    const int kWidth = W1280;                                                 \
+    const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                  \
+    const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                     \
+    const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                       \
+    const int kSizeUV = kStrideUV * SUBSAMPLE(kHeight, SUBSAMP_Y);            \
+    const int kBpc = 2;                                                       \
+    align_buffer_page_end(src_y, kWidth* kHeight* kBpc + SOFF);               \
+    align_buffer_page_end(src_u, kSizeUV* kBpc + SOFF);                       \
+    align_buffer_page_end(src_v, kSizeUV* kBpc + SOFF);                       \
+    align_buffer_page_end(dst_argb_c, kStrideB* kHeight + DOFF);              \
+    align_buffer_page_end(dst_argb_opt, kStrideB* kHeight + DOFF);            \
+    for (int i = 0; i < kWidth * kHeight; ++i) {                              \
+      reinterpret_cast<uint16_t*>(src_y + SOFF)[i] = (fastrand() & FMT_MASK); \
+    }                                                                         \
+    for (int i = 0; i < kSizeUV; ++i) {                                       \
+      reinterpret_cast<uint16_t*>(src_u + SOFF)[i] = (fastrand() & FMT_MASK); \
+      reinterpret_cast<uint16_t*>(src_v + SOFF)[i] = (fastrand() & FMT_MASK); \
+    }                                                                         \
+    memset(dst_argb_c + DOFF, 1, kStrideB * kHeight);                         \
+    memset(dst_argb_opt + DOFF, 101, kStrideB * kHeight);                     \
+    MaskCpuFlags(disable_cpu_flags_);                                         \
+    FMT_PLANAR##To##FMT_B(                                                    \
+        reinterpret_cast<uint16_t*>(src_y + SOFF), kWidth,                    \
+        reinterpret_cast<uint16_t*>(src_u + SOFF), kStrideUV,                 \
+        reinterpret_cast<uint16_t*>(src_v + SOFF), kStrideUV,                 \
+        dst_argb_c + DOFF, kStrideB, kWidth, NEG kHeight);                    \
+    MaskCpuFlags(benchmark_cpu_info_);                                        \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                         \
+      FMT_PLANAR##To##FMT_B(                                                  \
+          reinterpret_cast<uint16_t*>(src_y + SOFF), kWidth,                  \
+          reinterpret_cast<uint16_t*>(src_u + SOFF), kStrideUV,               \
+          reinterpret_cast<uint16_t*>(src_v + SOFF), kStrideUV,               \
+          dst_argb_opt + DOFF, kStrideB, kWidth, NEG kHeight);                \
+    }                                                                         \
+    for (int i = 0; i < kWidth * BPP_B * kHeight; ++i) {                      \
+      EXPECT_EQ(dst_argb_c[i + DOFF], dst_argb_opt[i + DOFF]);                \
+    }                                                                         \
+    free_aligned_buffer_page_end(src_y);                                      \
+    free_aligned_buffer_page_end(src_u);                                      \
+    free_aligned_buffer_page_end(src_v);                                      \
+    free_aligned_buffer_page_end(dst_argb_c);                                 \
+    free_aligned_buffer_page_end(dst_argb_opt);                               \
+  }
+
+#define TESTPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_MASK, FMT_B,   \
+                        BPP_B, ALIGN, YALIGN)                                \
+  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_MASK, FMT_B, BPP_B, \
+                   ALIGN, YALIGN, benchmark_width_ + 1, _Any, +, 0, 0)       \
+  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_MASK, FMT_B, BPP_B, \
+                   ALIGN, YALIGN, benchmark_width_, _Unaligned, +, 4, 4)     \
+  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_MASK, FMT_B, BPP_B, \
+                   ALIGN, YALIGN, benchmark_width_, _Invert, -, 0, 0)        \
+  TESTPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_MASK, FMT_B, BPP_B, \
+                   ALIGN, YALIGN, benchmark_width_, _Opt, +, 0, 0)
+
+// These conversions are only optimized for x86
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+TESTPLANAR16TOB(I010, 2, 2, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(I010, 2, 2, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(H010, 2, 2, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(H010, 2, 2, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(U010, 2, 2, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(U010, 2, 2, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(I210, 2, 1, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(I210, 2, 1, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(H210, 2, 1, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(H210, 2, 1, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(U210, 2, 1, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(U210, 2, 1, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(I410, 1, 1, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(I410, 1, 1, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(H410, 1, 1, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(H410, 1, 1, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(U410, 1, 1, 0x3ff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(U410, 1, 1, 0x3ff, ABGR, 4, 4, 1)
+TESTPLANAR16TOB(I012, 2, 2, 0xfff, ARGB, 4, 4, 1)
+TESTPLANAR16TOB(I010, 2, 2, 0x3ff, ARGBFilter, 4, 4, 1)
+TESTPLANAR16TOB(I210, 2, 1, 0x3ff, ARGBFilter, 4, 4, 1)
+
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTPLANAR16TOB(I010, 2, 2, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(I010, 2, 2, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(H010, 2, 2, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(H010, 2, 2, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(U010, 2, 2, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(U010, 2, 2, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(I210, 2, 1, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(I210, 2, 1, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(H210, 2, 1, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(H210, 2, 1, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(U210, 2, 1, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(U210, 2, 1, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(I410, 1, 1, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(I410, 1, 1, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(H410, 1, 1, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(H410, 1, 1, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(U410, 1, 1, 0x3ff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(U410, 1, 1, 0x3ff, AB30, 4, 4, 1)
+TESTPLANAR16TOB(I012, 2, 2, 0xfff, AR30, 4, 4, 1)
+TESTPLANAR16TOB(I010, 2, 2, 0x3ff, AR30Filter, 4, 4, 1)
+TESTPLANAR16TOB(I210, 2, 1, 0x3ff, AR30Filter, 4, 4, 1)
+#endif  // LITTLE_ENDIAN_ONLY_TEST
+#endif  // DISABLE_SLOW_TESTS
+
+#define TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B,      \
+                          ALIGN, YALIGN, W1280, N, NEG, OFF, ATTEN, S_DEPTH)   \
+  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                        \
+    const int kWidth = W1280;                                                  \
+    const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                   \
+    const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                      \
+    const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X);                        \
+    const int kSizeUV = kStrideUV * SUBSAMPLE(kHeight, SUBSAMP_Y);             \
+    const int kBpc = 2;                                                        \
+    align_buffer_page_end(src_y, kWidth* kHeight* kBpc + OFF);                 \
+    align_buffer_page_end(src_u, kSizeUV* kBpc + OFF);                         \
+    align_buffer_page_end(src_v, kSizeUV* kBpc + OFF);                         \
+    align_buffer_page_end(src_a, kWidth* kHeight* kBpc + OFF);                 \
+    align_buffer_page_end(dst_argb_c, kStrideB* kHeight + OFF);                \
+    align_buffer_page_end(dst_argb_opt, kStrideB* kHeight + OFF);              \
+    for (int i = 0; i < kWidth * kHeight; ++i) {                               \
+      reinterpret_cast<uint16_t*>(src_y + OFF)[i] =                            \
+          (fastrand() & ((1 << S_DEPTH) - 1));                                 \
+      reinterpret_cast<uint16_t*>(src_a + OFF)[i] =                            \
+          (fastrand() & ((1 << S_DEPTH) - 1));                                 \
+    }                                                                          \
+    for (int i = 0; i < kSizeUV; ++i) {                                        \
+      reinterpret_cast<uint16_t*>(src_u + OFF)[i] =                            \
+          (fastrand() & ((1 << S_DEPTH) - 1));                                 \
+      reinterpret_cast<uint16_t*>(src_v + OFF)[i] =                            \
+          (fastrand() & ((1 << S_DEPTH) - 1));                                 \
+    }                                                                          \
+    memset(dst_argb_c + OFF, 1, kStrideB * kHeight);                           \
+    memset(dst_argb_opt + OFF, 101, kStrideB * kHeight);                       \
+    MaskCpuFlags(disable_cpu_flags_);                                          \
+    FMT_PLANAR##To##FMT_B(reinterpret_cast<uint16_t*>(src_y + OFF), kWidth,    \
+                          reinterpret_cast<uint16_t*>(src_u + OFF), kStrideUV, \
+                          reinterpret_cast<uint16_t*>(src_v + OFF), kStrideUV, \
+                          reinterpret_cast<uint16_t*>(src_a + OFF), kWidth,    \
+                          dst_argb_c + OFF, kStrideB, kWidth, NEG kHeight,     \
+                          ATTEN);                                              \
+    MaskCpuFlags(benchmark_cpu_info_);                                         \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                          \
+      FMT_PLANAR##To##FMT_B(                                                   \
+          reinterpret_cast<uint16_t*>(src_y + OFF), kWidth,                    \
+          reinterpret_cast<uint16_t*>(src_u + OFF), kStrideUV,                 \
+          reinterpret_cast<uint16_t*>(src_v + OFF), kStrideUV,                 \
+          reinterpret_cast<uint16_t*>(src_a + OFF), kWidth,                    \
+          dst_argb_opt + OFF, kStrideB, kWidth, NEG kHeight, ATTEN);           \
+    }                                                                          \
+    for (int i = 0; i < kWidth * BPP_B * kHeight; ++i) {                       \
+      EXPECT_EQ(dst_argb_c[i + OFF], dst_argb_opt[i + OFF]);                   \
+    }                                                                          \
+    free_aligned_buffer_page_end(src_y);                                       \
+    free_aligned_buffer_page_end(src_u);                                       \
+    free_aligned_buffer_page_end(src_v);                                       \
+    free_aligned_buffer_page_end(src_a);                                       \
+    free_aligned_buffer_page_end(dst_argb_c);                                  \
+    free_aligned_buffer_page_end(dst_argb_opt);                                \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+#define TESTQPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B,    \
+                         ALIGN, YALIGN, S_DEPTH)                            \
+  TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                    YALIGN, benchmark_width_ + 1, _Any, +, 0, 0, S_DEPTH)   \
+  TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                    YALIGN, benchmark_width_, _Unaligned, +, 2, 0, S_DEPTH) \
+  TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                    YALIGN, benchmark_width_, _Invert, -, 0, 0, S_DEPTH)    \
+  TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                    YALIGN, benchmark_width_, _Opt, +, 0, 0, S_DEPTH)       \
+  TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                    YALIGN, benchmark_width_, _Premult, +, 0, 1, S_DEPTH)
+#else
+#define TESTQPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B,   \
+                         ALIGN, YALIGN, S_DEPTH)                           \
+  TESTQPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN, \
+                    YALIGN, benchmark_width_, _Opt, +, 0, 0, S_DEPTH)
+#endif
+
+#define I010AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvI601Constants, k, \
+                        l, m)
+#define I010AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvI601Constants, k, \
+                        l, m)
+#define J010AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define J010AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define F010AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define F010AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define H010AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define H010AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define U010AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define U010AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I010AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define V010AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I010AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define V010AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I010AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define I210AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvI601Constants, k, \
+                        l, m)
+#define I210AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvI601Constants, k, \
+                        l, m)
+#define J210AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define J210AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define F210AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define F210AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define H210AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define H210AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define U210AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define U210AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I210AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define V210AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I210AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define V210AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I210AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define I410AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvI601Constants, k, \
+                        l, m)
+#define I410AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvI601Constants, k, \
+                        l, m)
+#define J410AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define J410AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvJPEGConstants, k, \
+                        l, m)
+#define F410AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define F410AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvF709Constants, k, \
+                        l, m)
+#define H410AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define H410AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvH709Constants, k, \
+                        l, m)
+#define U410AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define U410AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)               \
+  I410AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuv2020Constants, k, \
+                        l, m)
+#define V410AlphaToARGB(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I410AlphaToARGBMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define V410AlphaToABGR(a, b, c, d, e, f, g, h, i, j, k, l, m)                \
+  I410AlphaToABGRMatrix(a, b, c, d, e, f, g, h, i, j, &kYuvV2020Constants, k, \
+                        l, m)
+#define I010AlphaToARGBFilter(a, b, c, d, e, f, g, h, i, j, k, l, m) \
+  I010AlphaToARGBMatrixFilter(a, b, c, d, e, f, g, h, i, j,          \
+                              &kYuvI601Constants, k, l, m, kFilterBilinear)
+#define I210AlphaToARGBFilter(a, b, c, d, e, f, g, h, i, j, k, l, m) \
+  I010AlphaToARGBMatrixFilter(a, b, c, d, e, f, g, h, i, j,          \
+                              &kYuvI601Constants, k, l, m, kFilterBilinear)
+
+// These conversions are only optimized for x86
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+TESTQPLANAR16TOB(I010Alpha, 2, 2, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I010Alpha, 2, 2, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(J010Alpha, 2, 2, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(J010Alpha, 2, 2, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(H010Alpha, 2, 2, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(H010Alpha, 2, 2, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(F010Alpha, 2, 2, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(F010Alpha, 2, 2, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(U010Alpha, 2, 2, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(U010Alpha, 2, 2, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(V010Alpha, 2, 2, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(V010Alpha, 2, 2, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I210Alpha, 2, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I210Alpha, 2, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(J210Alpha, 2, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(J210Alpha, 2, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(H210Alpha, 2, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(H210Alpha, 2, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(F210Alpha, 2, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(F210Alpha, 2, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(U210Alpha, 2, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(U210Alpha, 2, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(V210Alpha, 2, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(V210Alpha, 2, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I410Alpha, 1, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I410Alpha, 1, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(J410Alpha, 1, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(J410Alpha, 1, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(H410Alpha, 1, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(H410Alpha, 1, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(F410Alpha, 1, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(F410Alpha, 1, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(U410Alpha, 1, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(U410Alpha, 1, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(V410Alpha, 1, 1, ARGB, 4, 4, 1, 10)
+TESTQPLANAR16TOB(V410Alpha, 1, 1, ABGR, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I010Alpha, 2, 2, ARGBFilter, 4, 4, 1, 10)
+TESTQPLANAR16TOB(I210Alpha, 2, 1, ARGBFilter, 4, 4, 1, 10)
+#endif  // DISABLE_SLOW_TESTS
+
+#define TESTBIPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B,     \
+                           ALIGN, YALIGN, W1280, N, NEG, SOFF, DOFF, S_DEPTH)  \
+  TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) {                        \
+    const int kWidth = W1280;                                                  \
+    const int kHeight = ALIGNINT(benchmark_height_, YALIGN);                   \
+    const int kStrideB = ALIGNINT(kWidth * BPP_B, ALIGN);                      \
+    const int kStrideUV = SUBSAMPLE(kWidth, SUBSAMP_X) * 2;                    \
+    const int kSizeUV = kStrideUV * SUBSAMPLE(kHeight, SUBSAMP_Y) * 2;         \
+    const int kBpc = 2;                                                        \
+    align_buffer_page_end(src_y, kWidth* kHeight* kBpc + SOFF);                \
+    align_buffer_page_end(src_uv, kSizeUV* kBpc + SOFF);                       \
+    align_buffer_page_end(dst_argb_c, kStrideB* kHeight + DOFF);               \
+    align_buffer_page_end(dst_argb_opt, kStrideB* kHeight + DOFF);             \
+    for (int i = 0; i < kWidth * kHeight; ++i) {                               \
+      reinterpret_cast<uint16_t*>(src_y + SOFF)[i] =                           \
+          (fastrand() & (((uint16_t)(-1)) << (16 - S_DEPTH)));                 \
+    }                                                                          \
+    for (int i = 0; i < kSizeUV; ++i) {                                        \
+      reinterpret_cast<uint16_t*>(src_uv + SOFF)[i] =                          \
+          (fastrand() & (((uint16_t)(-1)) << (16 - S_DEPTH)));                 \
+    }                                                                          \
+    memset(dst_argb_c + DOFF, 1, kStrideB * kHeight);                          \
+    memset(dst_argb_opt + DOFF, 101, kStrideB * kHeight);                      \
+    MaskCpuFlags(disable_cpu_flags_);                                          \
+    FMT_PLANAR##To##FMT_B(reinterpret_cast<uint16_t*>(src_y + SOFF), kWidth,   \
+                          reinterpret_cast<uint16_t*>(src_uv + SOFF),          \
+                          kStrideUV, dst_argb_c + DOFF, kStrideB, kWidth,      \
+                          NEG kHeight);                                        \
+    MaskCpuFlags(benchmark_cpu_info_);                                         \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                          \
+      FMT_PLANAR##To##FMT_B(reinterpret_cast<uint16_t*>(src_y + SOFF), kWidth, \
+                            reinterpret_cast<uint16_t*>(src_uv + SOFF),        \
+                            kStrideUV, dst_argb_opt + DOFF, kStrideB, kWidth,  \
+                            NEG kHeight);                                      \
+    }                                                                          \
+    for (int i = 0; i < kWidth * BPP_B * kHeight; ++i) {                       \
+      EXPECT_EQ(dst_argb_c[i + DOFF], dst_argb_opt[i + DOFF]);                 \
+    }                                                                          \
+    free_aligned_buffer_page_end(src_y);                                       \
+    free_aligned_buffer_page_end(src_uv);                                      \
+    free_aligned_buffer_page_end(dst_argb_c);                                  \
+    free_aligned_buffer_page_end(dst_argb_opt);                                \
+  }
+
+#define TESTBIPLANAR16TOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B,    \
+                          ALIGN, YALIGN, S_DEPTH)                            \
+  TESTBIPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                     YALIGN, benchmark_width_ + 1, _Any, +, 0, 0, S_DEPTH)   \
+  TESTBIPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                     YALIGN, benchmark_width_, _Unaligned, +, 4, 4, S_DEPTH) \
+  TESTBIPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                     YALIGN, benchmark_width_, _Invert, -, 0, 0, S_DEPTH)    \
+  TESTBIPLANAR16TOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, BPP_B, ALIGN,  \
+                     YALIGN, benchmark_width_, _Opt, +, 0, 0, S_DEPTH)
+
+#define P010ToARGB(a, b, c, d, e, f, g, h) \
+  P010ToARGBMatrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P210ToARGB(a, b, c, d, e, f, g, h) \
+  P210ToARGBMatrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P010ToAR30(a, b, c, d, e, f, g, h) \
+  P010ToAR30Matrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P210ToAR30(a, b, c, d, e, f, g, h) \
+  P210ToAR30Matrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+
+#define P012ToARGB(a, b, c, d, e, f, g, h) \
+  P012ToARGBMatrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P212ToARGB(a, b, c, d, e, f, g, h) \
+  P212ToARGBMatrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P012ToAR30(a, b, c, d, e, f, g, h) \
+  P012ToAR30Matrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P212ToAR30(a, b, c, d, e, f, g, h) \
+  P212ToAR30Matrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+
+#define P016ToARGB(a, b, c, d, e, f, g, h) \
+  P016ToARGBMatrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P216ToARGB(a, b, c, d, e, f, g, h) \
+  P216ToARGBMatrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P016ToAR30(a, b, c, d, e, f, g, h) \
+  P016ToAR30Matrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+#define P216ToAR30(a, b, c, d, e, f, g, h) \
+  P216ToAR30Matrix(a, b, c, d, e, f, &kYuvH709Constants, g, h)
+
+#define P010ToARGBFilter(a, b, c, d, e, f, g, h)                     \
+  P010ToARGBMatrixFilter(a, b, c, d, e, f, &kYuvH709Constants, g, h, \
+                         kFilterBilinear)
+#define P210ToARGBFilter(a, b, c, d, e, f, g, h)                     \
+  P210ToARGBMatrixFilter(a, b, c, d, e, f, &kYuvH709Constants, g, h, \
+                         kFilterBilinear)
+#define P010ToAR30Filter(a, b, c, d, e, f, g, h)                     \
+  P010ToAR30MatrixFilter(a, b, c, d, e, f, &kYuvH709Constants, g, h, \
+                         kFilterBilinear)
+#define P210ToAR30Filter(a, b, c, d, e, f, g, h)                     \
+  P210ToAR30MatrixFilter(a, b, c, d, e, f, &kYuvH709Constants, g, h, \
+                         kFilterBilinear)
+
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+TESTBIPLANAR16TOB(P010, 2, 2, ARGB, 4, 4, 1, 10)
+TESTBIPLANAR16TOB(P210, 2, 1, ARGB, 4, 4, 1, 10)
+TESTBIPLANAR16TOB(P012, 2, 2, ARGB, 4, 4, 1, 12)
+TESTBIPLANAR16TOB(P212, 2, 1, ARGB, 4, 4, 1, 12)
+TESTBIPLANAR16TOB(P016, 2, 2, ARGB, 4, 4, 1, 16)
+TESTBIPLANAR16TOB(P216, 2, 1, ARGB, 4, 4, 1, 16)
+TESTBIPLANAR16TOB(P010, 2, 2, ARGBFilter, 4, 4, 1, 10)
+TESTBIPLANAR16TOB(P210, 2, 1, ARGBFilter, 4, 4, 1, 10)
+#ifdef LITTLE_ENDIAN_ONLY_TEST
+TESTBIPLANAR16TOB(P010, 2, 2, AR30, 4, 4, 1, 10)
+TESTBIPLANAR16TOB(P210, 2, 1, AR30, 4, 4, 1, 10)
+TESTBIPLANAR16TOB(P012, 2, 2, AR30, 4, 4, 1, 12)
+TESTBIPLANAR16TOB(P212, 2, 1, AR30, 4, 4, 1, 12)
+TESTBIPLANAR16TOB(P016, 2, 2, AR30, 4, 4, 1, 16)
+TESTBIPLANAR16TOB(P216, 2, 1, AR30, 4, 4, 1, 16)
+TESTBIPLANAR16TOB(P010, 2, 2, AR30Filter, 4, 4, 1, 10)
+TESTBIPLANAR16TOB(P210, 2, 1, AR30Filter, 4, 4, 1, 10)
+#endif  // LITTLE_ENDIAN_ONLY_TEST
+#endif  // DISABLE_SLOW_TESTS
 
 static int Clamp(int y) {
   if (y < 0) {
@@ -2918,10 +3903,11 @@ TEST_F(LibYUVConvertTest, TestH420ToARGB) {
     ++histogram_b[b];
     ++histogram_g[g];
     ++histogram_r[r];
-    int expected_y = Clamp(static_cast<int>((i - 16) * 1.164f));
-    EXPECT_NEAR(b, expected_y, 1);
-    EXPECT_NEAR(g, expected_y, 1);
-    EXPECT_NEAR(r, expected_y, 1);
+    // Reference formula for Y channel contribution in YUV to RGB conversions:
+    int expected_y = Clamp(static_cast<int>((i - 16) * 1.164f + 0.5f));
+    EXPECT_EQ(b, expected_y);
+    EXPECT_EQ(g, expected_y);
+    EXPECT_EQ(r, expected_y);
     EXPECT_EQ(a, 255);
   }
 
@@ -3043,7 +4029,7 @@ TEST_F(LibYUVConvertTest, TestH010ToAR30) {
     ++histogram_b[b10];
     ++histogram_g[g10];
     ++histogram_r[r10];
-    int expected_y = Clamp10(static_cast<int>((i - 64) * 1.164f));
+    int expected_y = Clamp10(static_cast<int>((i - 64) * 1.164f + 0.5));
     EXPECT_NEAR(b10, expected_y, 4);
     EXPECT_NEAR(g10, expected_y, 4);
     EXPECT_NEAR(r10, expected_y, 4);
@@ -3196,6 +4182,68 @@ TEST_F(LibYUVConvertTest, TestH420ToAR30) {
   free_aligned_buffer_page_end(ar30_pixels);
 }
 
+// Test I400 with jpeg matrix is same as J400
+TEST_F(LibYUVConvertTest, TestI400) {
+  const int kSize = 256;
+  align_buffer_page_end(orig_i400, kSize);
+  align_buffer_page_end(argb_pixels_i400, kSize * 4);
+  align_buffer_page_end(argb_pixels_j400, kSize * 4);
+  align_buffer_page_end(argb_pixels_jpeg_i400, kSize * 4);
+  align_buffer_page_end(argb_pixels_h709_i400, kSize * 4);
+  align_buffer_page_end(argb_pixels_2020_i400, kSize * 4);
+
+  // Test grey scale
+  for (int i = 0; i < kSize; ++i) {
+    orig_i400[i] = i;
+  }
+
+  J400ToARGB(orig_i400, 0, argb_pixels_j400, 0, kSize, 1);
+  I400ToARGB(orig_i400, 0, argb_pixels_i400, 0, kSize, 1);
+  I400ToARGBMatrix(orig_i400, 0, argb_pixels_jpeg_i400, 0, &kYuvJPEGConstants,
+                   kSize, 1);
+  I400ToARGBMatrix(orig_i400, 0, argb_pixels_h709_i400, 0, &kYuvH709Constants,
+                   kSize, 1);
+  I400ToARGBMatrix(orig_i400, 0, argb_pixels_2020_i400, 0, &kYuv2020Constants,
+                   kSize, 1);
+
+  EXPECT_EQ(0, argb_pixels_i400[0]);
+  EXPECT_EQ(0, argb_pixels_j400[0]);
+  EXPECT_EQ(0, argb_pixels_jpeg_i400[0]);
+  EXPECT_EQ(0, argb_pixels_h709_i400[0]);
+  EXPECT_EQ(0, argb_pixels_2020_i400[0]);
+  EXPECT_EQ(0, argb_pixels_i400[16 * 4]);
+  EXPECT_EQ(16, argb_pixels_j400[16 * 4]);
+  EXPECT_EQ(16, argb_pixels_jpeg_i400[16 * 4]);
+  EXPECT_EQ(0, argb_pixels_h709_i400[16 * 4]);
+  EXPECT_EQ(0, argb_pixels_2020_i400[16 * 4]);
+  EXPECT_EQ(130, argb_pixels_i400[128 * 4]);
+  EXPECT_EQ(128, argb_pixels_j400[128 * 4]);
+  EXPECT_EQ(128, argb_pixels_jpeg_i400[128 * 4]);
+  EXPECT_EQ(130, argb_pixels_h709_i400[128 * 4]);
+  EXPECT_EQ(130, argb_pixels_2020_i400[128 * 4]);
+  EXPECT_EQ(255, argb_pixels_i400[255 * 4]);
+  EXPECT_EQ(255, argb_pixels_j400[255 * 4]);
+  EXPECT_EQ(255, argb_pixels_jpeg_i400[255 * 4]);
+  EXPECT_EQ(255, argb_pixels_h709_i400[255 * 4]);
+  EXPECT_EQ(255, argb_pixels_2020_i400[255 * 4]);
+
+  for (int i = 0; i < kSize * 4; ++i) {
+    if ((i & 3) == 3) {
+      EXPECT_EQ(255, argb_pixels_j400[i]);
+    } else {
+      EXPECT_EQ(i / 4, argb_pixels_j400[i]);
+    }
+    EXPECT_EQ(argb_pixels_jpeg_i400[i], argb_pixels_j400[i]);
+  }
+
+  free_aligned_buffer_page_end(orig_i400);
+  free_aligned_buffer_page_end(argb_pixels_i400);
+  free_aligned_buffer_page_end(argb_pixels_j400);
+  free_aligned_buffer_page_end(argb_pixels_jpeg_i400);
+  free_aligned_buffer_page_end(argb_pixels_h709_i400);
+  free_aligned_buffer_page_end(argb_pixels_2020_i400);
+}
+
 // Test RGB24 to ARGB and back to RGB24
 TEST_F(LibYUVConvertTest, TestARGBToRGB24) {
   const int kSize = 256;
@@ -3220,4 +4268,74 @@ TEST_F(LibYUVConvertTest, TestARGBToRGB24) {
   free_aligned_buffer_page_end(dest_rgb24);
 }
 
+TEST_F(LibYUVConvertTest, Test565) {
+  SIMD_ALIGNED(uint8_t orig_pixels[256][4]);
+  SIMD_ALIGNED(uint8_t pixels565[256][2]);
+
+  for (int i = 0; i < 256; ++i) {
+    for (int j = 0; j < 4; ++j) {
+      orig_pixels[i][j] = i;
+    }
+  }
+  ARGBToRGB565(&orig_pixels[0][0], 0, &pixels565[0][0], 0, 256, 1);
+  uint32_t checksum = HashDjb2(&pixels565[0][0], sizeof(pixels565), 5381);
+  EXPECT_EQ(610919429u, checksum);
+}
+
+// Test RGB24 to J420 is exact
+#if defined(LIBYUV_BIT_EXACT)
+TEST_F(LibYUVConvertTest, TestRGB24ToJ420) {
+  const int kSize = 256;
+  align_buffer_page_end(orig_rgb24, kSize * 3 * 2);  // 2 rows of RGB24
+  align_buffer_page_end(dest_j420, kSize * 3 / 2 * 2);
+  int iterations256 = (benchmark_width_ * benchmark_height_ + (kSize * 2 - 1)) /
+                      (kSize * 2) * benchmark_iterations_;
+
+  for (int i = 0; i < kSize * 3 * 2; ++i) {
+    orig_rgb24[i] = i;
+  }
+
+  for (int i = 0; i < iterations256; ++i) {
+    RGB24ToJ420(orig_rgb24, kSize * 3, dest_j420, kSize,  // Y plane
+                dest_j420 + kSize * 2, kSize / 2,         // U plane
+                dest_j420 + kSize * 5 / 2, kSize / 2,     // V plane
+                kSize, 2);
+  }
+
+  uint32_t checksum = HashDjb2(dest_j420, kSize * 3 / 2 * 2, 5381);
+  EXPECT_EQ(2755440272u, checksum);
+
+  free_aligned_buffer_page_end(orig_rgb24);
+  free_aligned_buffer_page_end(dest_j420);
+}
+#endif
+
+// Test RGB24 to I420 is exact
+#if defined(LIBYUV_BIT_EXACT)
+TEST_F(LibYUVConvertTest, TestRGB24ToI420) {
+  const int kSize = 256;
+  align_buffer_page_end(orig_rgb24, kSize * 3 * 2);  // 2 rows of RGB24
+  align_buffer_page_end(dest_i420, kSize * 3 / 2 * 2);
+  int iterations256 = (benchmark_width_ * benchmark_height_ + (kSize * 2 - 1)) /
+                      (kSize * 2) * benchmark_iterations_;
+
+  for (int i = 0; i < kSize * 3 * 2; ++i) {
+    orig_rgb24[i] = i;
+  }
+
+  for (int i = 0; i < iterations256; ++i) {
+    RGB24ToI420(orig_rgb24, kSize * 3, dest_i420, kSize,  // Y plane
+                dest_i420 + kSize * 2, kSize / 2,         // U plane
+                dest_i420 + kSize * 5 / 2, kSize / 2,     // V plane
+                kSize, 2);
+  }
+
+  uint32_t checksum = HashDjb2(dest_i420, kSize * 3 / 2 * 2, 5381);
+  EXPECT_EQ(1526656597u, checksum);
+
+  free_aligned_buffer_page_end(orig_rgb24);
+  free_aligned_buffer_page_end(dest_i420);
+}
+#endif
+
 }  // namespace libyuv
diff --git a/files/unit_test/cpu_test.cc b/files/unit_test/cpu_test.cc
index a7991d2b..080778f5 100644
--- a/files/unit_test/cpu_test.cc
+++ b/files/unit_test/cpu_test.cc
@@ -40,6 +40,7 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
   int has_gfni = TestCpuFlag(kCpuHasGFNI);
   int has_avx512bw = TestCpuFlag(kCpuHasAVX512BW);
   int has_avx512vl = TestCpuFlag(kCpuHasAVX512VL);
+  int has_avx512vnni = TestCpuFlag(kCpuHasAVX512VNNI);
   int has_avx512vbmi = TestCpuFlag(kCpuHasAVX512VBMI);
   int has_avx512vbmi2 = TestCpuFlag(kCpuHasAVX512VBMI2);
   int has_avx512vbitalg = TestCpuFlag(kCpuHasAVX512VBITALG);
@@ -57,6 +58,7 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
   printf("Has GFNI %d\n", has_gfni);
   printf("Has AVX512BW %d\n", has_avx512bw);
   printf("Has AVX512VL %d\n", has_avx512vl);
+  printf("Has AVX512VNNI %d\n", has_avx512vnni);
   printf("Has AVX512VBMI %d\n", has_avx512vbmi);
   printf("Has AVX512VBMI2 %d\n", has_avx512vbmi2);
   printf("Has AVX512VBITALG %d\n", has_avx512vbitalg);
@@ -67,31 +69,113 @@ TEST_F(LibYUVBaseTest, TestCpuHas) {
   printf("Has MIPS %d\n", has_mips);
   int has_msa = TestCpuFlag(kCpuHasMSA);
   printf("Has MSA %d\n", has_msa);
-  int has_mmi = TestCpuFlag(kCpuHasMMI);
-  printf("Has MMI %d\n", has_mmi);
+#endif
+
+#if defined(__loongarch__)
+  int has_loongarch = TestCpuFlag(kCpuHasLOONGARCH);
+  printf("Has LOONGARCH %d\n", has_loongarch);
+  int has_lsx = TestCpuFlag(kCpuHasLSX);
+  printf("Has LSX %d\n", has_lsx);
+  int has_lasx = TestCpuFlag(kCpuHasLASX);
+  printf("Has LASX %d\n", has_lasx);
 #endif
 }
 
-TEST_F(LibYUVBaseTest, TestCpuCompilerEnabled) {
-#if defined(__aarch64__)
-  printf("Arm64 build\n");
+TEST_F(LibYUVBaseTest, TestCompilerMacros) {
+  // Tests all macros used in public headers.
+#ifdef __ATOMIC_RELAXED
+  printf("__ATOMIC_RELAXED %d\n", __ATOMIC_RELAXED);
+#endif
+#ifdef __cplusplus
+  printf("__cplusplus %ld\n", __cplusplus);
+#endif
+#ifdef __clang_major__
+  printf("__clang_major__ %d\n", __clang_major__);
+#endif
+#ifdef __clang_minor__
+  printf("__clang_minor__ %d\n", __clang_minor__);
 #endif
-#if defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON)
-  printf("Neon build enabled\n");
+#ifdef __GNUC__
+  printf("__GNUC__ %d\n", __GNUC__);
 #endif
-#if defined(__x86_64__) || defined(_M_X64)
-  printf("x64 build\n");
+#ifdef __GNUC_MINOR__
+  printf("__GNUC_MINOR__ %d\n", __GNUC_MINOR__);
+#endif
+#ifdef __i386__
+  printf("__i386__ %d\n", __i386__);
+#endif
+#ifdef __mips
+  printf("__mips %d\n", __mips);
+#endif
+#ifdef __mips_isa_rev
+  printf("__mips_isa_rev %d\n", __mips_isa_rev);
+#endif
+#ifdef __x86_64__
+  printf("__x86_64__ %d\n", __x86_64__);
 #endif
 #ifdef _MSC_VER
   printf("_MSC_VER %d\n", _MSC_VER);
 #endif
-#if !defined(LIBYUV_DISABLE_X86) &&                      \
-    (defined(GCC_HAS_AVX2) || defined(CLANG_HAS_AVX2) || \
-     defined(VISUALC_HAS_AVX2))
-  printf("Has AVX2 1\n");
-#else
-  printf("Has AVX2 0\n");
-// If compiler does not support AVX2, the following function not expected:
+#ifdef __aarch64__
+  printf("__aarch64__ %d\n", __aarch64__);
+#endif
+#ifdef __APPLE__
+  printf("__APPLE__ %d\n", __APPLE__);
+#endif
+#ifdef __arm__
+  printf("__arm__ %d\n", __arm__);
+#endif
+#ifdef __clang__
+  printf("__clang__ %d\n", __clang__);
+#endif
+#ifdef __CLR_VER
+  printf("__CLR_VER %d\n", __CLR_VER);
+#endif
+#ifdef __CYGWIN__
+  printf("__CYGWIN__ %d\n", __CYGWIN__);
+#endif
+#ifdef __llvm__
+  printf("__llvm__ %d\n", __llvm__);
+#endif
+#ifdef __mips_msa
+  printf("__mips_msa %d\n", __mips_msa);
+#endif
+#ifdef __native_client__
+  printf("__native_client__ %d\n", __native_client__);
+#endif
+#ifdef __pic__
+  printf("__pic__ %d\n", __pic__);
+#endif
+#ifdef __pnacl__
+  printf("__pnacl__ %d\n", __pnacl__);
+#endif
+#ifdef _M_IX86
+  printf("_M_IX86 %d\n", _M_IX86);
+#endif
+#ifdef _M_X64
+  printf("_M_X64 %d\n", _M_X64);
+#endif
+#ifdef _MIPS_ARCH_LOONGSON3A
+  printf("_MIPS_ARCH_LOONGSON3A %d\n", _MIPS_ARCH_LOONGSON3A);
+#endif
+#ifdef __loongarch__
+  printf("__loongarch__ %d\n", __loongarch__);
+#endif
+#ifdef _WIN32
+  printf("_WIN32 %d\n", _WIN32);
+#endif
+#ifdef GG_LONGLONG
+  printf("GG_LONGLONG %d\n", GG_LONGLONG);
+#endif
+#ifdef INT_TYPES_DEFINED
+  printf("INT_TYPES_DEFINED\n");
+#endif
+#ifdef __has_feature
+  printf("__has_feature\n");
+#if __has_feature(memory_sanitizer)
+  printf("__has_feature(memory_sanitizer) %d\n",
+         __has_feature(memory_sanitizer));
+#endif
 #endif
 }
 
@@ -156,11 +240,36 @@ TEST_F(LibYUVBaseTest, TestLinuxNeon) {
     printf("WARNING: unable to load \"../../unit_test/testdata/arm_v7.txt\"\n");
   }
 #if defined(__linux__) && defined(__ARM_NEON__)
-  EXPECT_EQ(kCpuHasNEON, ArmCpuCaps("/proc/cpuinfo"));
+  if (FileExists("/proc/cpuinfo")) {
+    if (kCpuHasNEON != ArmCpuCaps("/proc/cpuinfo")) {
+      // This can happen on ARM emulator but /proc/cpuinfo is from host.
+      printf("WARNING: Neon build enabled but CPU does not have NEON\n");
+    }
+  } else {
+    printf("WARNING: unable to load \"/proc/cpuinfo\"\n");
+  }
 #endif
 }
 
+TEST_F(LibYUVBaseTest, TestLinuxMipsMsa) {
+  if (FileExists("../../unit_test/testdata/mips.txt")) {
+    printf("Note: testing to load \"../../unit_test/testdata/mips.txt\"\n");
+
+    EXPECT_EQ(0, MipsCpuCaps("../../unit_test/testdata/mips.txt"));
+    EXPECT_EQ(kCpuHasMSA, MipsCpuCaps("../../unit_test/testdata/mips_msa.txt"));
+    EXPECT_EQ(kCpuHasMSA,
+              MipsCpuCaps("../../unit_test/testdata/mips_loongson2k.txt"));
+  } else {
+    printf("WARNING: unable to load \"../../unit_test/testdata/mips.txt\"\n");
+  }
+}
+
+// TODO(fbarchard): Fix clangcl test of cpuflags.
+#ifdef _MSC_VER
+TEST_F(LibYUVBaseTest, DISABLED_TestSetCpuFlags) {
+#else
 TEST_F(LibYUVBaseTest, TestSetCpuFlags) {
+#endif
   // Reset any masked flags that may have been set so auto init is enabled.
   MaskCpuFlags(0);
 
diff --git a/files/unit_test/cpu_thread_test.cc b/files/unit_test/cpu_thread_test.cc
index 59061b98..69aab74e 100644
--- a/files/unit_test/cpu_thread_test.cc
+++ b/files/unit_test/cpu_thread_test.cc
@@ -12,7 +12,7 @@
 
 #include "libyuv/cpu_id.h"
 
-#if defined(__clang__)
+#if defined(__clang__) && !defined(__wasm__)
 #if __has_include(<pthread.h>)
 #define LIBYUV_HAVE_PTHREAD 1
 #endif
@@ -30,7 +30,7 @@ namespace libyuv {
 void* ThreadMain(void* arg) {
   int* flags = static_cast<int*>(arg);
 
-  *flags = TestCpuFlag(kCpuHasSSSE3);
+  *flags = TestCpuFlag(kCpuInitialized);
   return nullptr;
 }
 #endif  // LIBYUV_HAVE_PTHREAD
diff --git a/files/unit_test/math_test.cc b/files/unit_test/math_test.cc
index 0abbad51..a1544c12 100644
--- a/files/unit_test/math_test.cc
+++ b/files/unit_test/math_test.cc
@@ -16,10 +16,14 @@
 #include "libyuv/basic_types.h"
 #include "libyuv/cpu_id.h"
 #include "libyuv/scale.h"
+
+#ifdef ENABLE_ROW_TESTS
 #include "libyuv/scale_row.h"
+#endif
 
 namespace libyuv {
 
+#ifdef ENABLE_ROW_TESTS
 TEST_F(LibYUVBaseTest, TestFixedDiv) {
   int num[1280];
   int div[1280];
@@ -151,5 +155,6 @@ TEST_F(LibYUVBaseTest, TestFixedDiv1_Opt) {
     EXPECT_NEAR(result_c[j], result_opt[j], 1);
   }
 }
+#endif  // ENABLE_ROW_TESTS
 
 }  // namespace libyuv
diff --git a/files/unit_test/planar_test.cc b/files/unit_test/planar_test.cc
index 70f8966e..3a8c470b 100644
--- a/files/unit_test/planar_test.cc
+++ b/files/unit_test/planar_test.cc
@@ -12,9 +12,6 @@
 #include <stdlib.h>
 #include <time.h>
 
-// row.h defines SIMD_ALIGNED, overriding unit_test.h
-#include "libyuv/row.h" /* For ScaleSumSamples_Neon */
-
 #include "../unit_test/unit_test.h"
 #include "libyuv/compare.h"
 #include "libyuv/convert.h"
@@ -24,6 +21,19 @@
 #include "libyuv/cpu_id.h"
 #include "libyuv/planar_functions.h"
 #include "libyuv/rotate.h"
+#include "libyuv/scale.h"
+
+#ifdef ENABLE_ROW_TESTS
+// row.h defines SIMD_ALIGNED, overriding unit_test.h
+// TODO(fbarchard): Remove row.h from unittests.  Test public functions.
+#include "libyuv/row.h" /* For ScaleSumSamples_Neon */
+#endif
+
+#if defined(LIBYUV_BIT_EXACT)
+#define EXPECTED_ATTENUATE_DIFF 0
+#else
+#define EXPECTED_ATTENUATE_DIFF 2
+#endif
 
 namespace libyuv {
 
@@ -96,9 +106,9 @@ TEST_F(LibYUVPlanarTest, TestAttenuate) {
   EXPECT_EQ(32, atten_pixels[128 * 4 + 1]);
   EXPECT_EQ(21, atten_pixels[128 * 4 + 2]);
   EXPECT_EQ(128, atten_pixels[128 * 4 + 3]);
-  EXPECT_NEAR(255, atten_pixels[255 * 4 + 0], 1);
-  EXPECT_NEAR(127, atten_pixels[255 * 4 + 1], 1);
-  EXPECT_NEAR(85, atten_pixels[255 * 4 + 2], 1);
+  EXPECT_NEAR(254, atten_pixels[255 * 4 + 0], EXPECTED_ATTENUATE_DIFF);
+  EXPECT_NEAR(127, atten_pixels[255 * 4 + 1], EXPECTED_ATTENUATE_DIFF);
+  EXPECT_NEAR(85, atten_pixels[255 * 4 + 2], EXPECTED_ATTENUATE_DIFF);
   EXPECT_EQ(255, atten_pixels[255 * 4 + 3]);
 
   free_aligned_buffer_page_end(atten2_pixels);
@@ -151,31 +161,32 @@ static int TestAttenuateI(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBAttenuate_Any) {
-  int max_diff = TestAttenuateI(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestAttenuateI(benchmark_width_ + 1, benchmark_height_,
                                 benchmark_iterations_, disable_cpu_flags_,
                                 benchmark_cpu_info_, +1, 0);
-  EXPECT_LE(max_diff, 2);
+
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBAttenuate_Unaligned) {
   int max_diff =
       TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
                      disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBAttenuate_Invert) {
   int max_diff =
       TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
                      disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBAttenuate_Opt) {
   int max_diff =
       TestAttenuateI(benchmark_width_, benchmark_height_, benchmark_iterations_,
                      disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 static int TestUnattenuateI(int width,
@@ -224,31 +235,31 @@ static int TestUnattenuateI(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Any) {
-  int max_diff = TestUnattenuateI(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestUnattenuateI(benchmark_width_ + 1, benchmark_height_,
                                   benchmark_iterations_, disable_cpu_flags_,
                                   benchmark_cpu_info_, +1, 0);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Unaligned) {
   int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
                                   benchmark_iterations_, disable_cpu_flags_,
                                   benchmark_cpu_info_, +1, 1);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Invert) {
   int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
                                   benchmark_iterations_, disable_cpu_flags_,
                                   benchmark_cpu_info_, -1, 0);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBUnattenuate_Opt) {
   int max_diff = TestUnattenuateI(benchmark_width_, benchmark_height_,
                                   benchmark_iterations_, disable_cpu_flags_,
                                   benchmark_cpu_info_, +1, 0);
-  EXPECT_LE(max_diff, 2);
+  EXPECT_LE(max_diff, EXPECTED_ATTENUATE_DIFF);
 }
 
 TEST_F(LibYUVPlanarTest, TestARGBComputeCumulativeSum) {
@@ -277,6 +288,7 @@ TEST_F(LibYUVPlanarTest, TestARGBComputeCumulativeSum) {
   }
 }
 
+// near is for legacy platforms.
 TEST_F(LibYUVPlanarTest, TestARGBGray) {
   SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
   memset(orig_pixels, 0, sizeof(orig_pixels));
@@ -313,17 +325,17 @@ TEST_F(LibYUVPlanarTest, TestARGBGray) {
   orig_pixels[5][3] = 224u;
   // Do 16 to test asm version.
   ARGBGray(&orig_pixels[0][0], 0, 0, 0, 16, 1);
-  EXPECT_EQ(30u, orig_pixels[0][0]);
-  EXPECT_EQ(30u, orig_pixels[0][1]);
-  EXPECT_EQ(30u, orig_pixels[0][2]);
+  EXPECT_NEAR(29u, orig_pixels[0][0], 1);
+  EXPECT_NEAR(29u, orig_pixels[0][1], 1);
+  EXPECT_NEAR(29u, orig_pixels[0][2], 1);
   EXPECT_EQ(128u, orig_pixels[0][3]);
   EXPECT_EQ(149u, orig_pixels[1][0]);
   EXPECT_EQ(149u, orig_pixels[1][1]);
   EXPECT_EQ(149u, orig_pixels[1][2]);
   EXPECT_EQ(0u, orig_pixels[1][3]);
-  EXPECT_EQ(76u, orig_pixels[2][0]);
-  EXPECT_EQ(76u, orig_pixels[2][1]);
-  EXPECT_EQ(76u, orig_pixels[2][2]);
+  EXPECT_NEAR(77u, orig_pixels[2][0], 1);
+  EXPECT_NEAR(77u, orig_pixels[2][1], 1);
+  EXPECT_NEAR(77u, orig_pixels[2][2], 1);
   EXPECT_EQ(255u, orig_pixels[2][3]);
   EXPECT_EQ(0u, orig_pixels[3][0]);
   EXPECT_EQ(0u, orig_pixels[3][1]);
@@ -333,9 +345,9 @@ TEST_F(LibYUVPlanarTest, TestARGBGray) {
   EXPECT_EQ(255u, orig_pixels[4][1]);
   EXPECT_EQ(255u, orig_pixels[4][2]);
   EXPECT_EQ(255u, orig_pixels[4][3]);
-  EXPECT_EQ(96u, orig_pixels[5][0]);
-  EXPECT_EQ(96u, orig_pixels[5][1]);
-  EXPECT_EQ(96u, orig_pixels[5][2]);
+  EXPECT_NEAR(97u, orig_pixels[5][0], 1);
+  EXPECT_NEAR(97u, orig_pixels[5][1], 1);
+  EXPECT_NEAR(97u, orig_pixels[5][2], 1);
   EXPECT_EQ(224u, orig_pixels[5][3]);
   for (int i = 0; i < 1280; ++i) {
     orig_pixels[i][0] = i;
@@ -385,30 +397,30 @@ TEST_F(LibYUVPlanarTest, TestARGBGrayTo) {
   orig_pixels[5][3] = 224u;
   // Do 16 to test asm version.
   ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 16, 1);
-  EXPECT_EQ(30u, gray_pixels[0][0]);
-  EXPECT_EQ(30u, gray_pixels[0][1]);
-  EXPECT_EQ(30u, gray_pixels[0][2]);
-  EXPECT_EQ(128u, gray_pixels[0][3]);
-  EXPECT_EQ(149u, gray_pixels[1][0]);
-  EXPECT_EQ(149u, gray_pixels[1][1]);
-  EXPECT_EQ(149u, gray_pixels[1][2]);
-  EXPECT_EQ(0u, gray_pixels[1][3]);
-  EXPECT_EQ(76u, gray_pixels[2][0]);
-  EXPECT_EQ(76u, gray_pixels[2][1]);
-  EXPECT_EQ(76u, gray_pixels[2][2]);
-  EXPECT_EQ(255u, gray_pixels[2][3]);
-  EXPECT_EQ(0u, gray_pixels[3][0]);
-  EXPECT_EQ(0u, gray_pixels[3][1]);
-  EXPECT_EQ(0u, gray_pixels[3][2]);
-  EXPECT_EQ(255u, gray_pixels[3][3]);
-  EXPECT_EQ(255u, gray_pixels[4][0]);
-  EXPECT_EQ(255u, gray_pixels[4][1]);
-  EXPECT_EQ(255u, gray_pixels[4][2]);
-  EXPECT_EQ(255u, gray_pixels[4][3]);
-  EXPECT_EQ(96u, gray_pixels[5][0]);
-  EXPECT_EQ(96u, gray_pixels[5][1]);
-  EXPECT_EQ(96u, gray_pixels[5][2]);
-  EXPECT_EQ(224u, gray_pixels[5][3]);
+  EXPECT_NEAR(30u, gray_pixels[0][0], 1);
+  EXPECT_NEAR(30u, gray_pixels[0][1], 1);
+  EXPECT_NEAR(30u, gray_pixels[0][2], 1);
+  EXPECT_NEAR(128u, gray_pixels[0][3], 1);
+  EXPECT_NEAR(149u, gray_pixels[1][0], 1);
+  EXPECT_NEAR(149u, gray_pixels[1][1], 1);
+  EXPECT_NEAR(149u, gray_pixels[1][2], 1);
+  EXPECT_NEAR(0u, gray_pixels[1][3], 1);
+  EXPECT_NEAR(76u, gray_pixels[2][0], 1);
+  EXPECT_NEAR(76u, gray_pixels[2][1], 1);
+  EXPECT_NEAR(76u, gray_pixels[2][2], 1);
+  EXPECT_NEAR(255u, gray_pixels[2][3], 1);
+  EXPECT_NEAR(0u, gray_pixels[3][0], 1);
+  EXPECT_NEAR(0u, gray_pixels[3][1], 1);
+  EXPECT_NEAR(0u, gray_pixels[3][2], 1);
+  EXPECT_NEAR(255u, gray_pixels[3][3], 1);
+  EXPECT_NEAR(255u, gray_pixels[4][0], 1);
+  EXPECT_NEAR(255u, gray_pixels[4][1], 1);
+  EXPECT_NEAR(255u, gray_pixels[4][2], 1);
+  EXPECT_NEAR(255u, gray_pixels[4][3], 1);
+  EXPECT_NEAR(96u, gray_pixels[5][0], 1);
+  EXPECT_NEAR(96u, gray_pixels[5][1], 1);
+  EXPECT_NEAR(96u, gray_pixels[5][2], 1);
+  EXPECT_NEAR(224u, gray_pixels[5][3], 1);
   for (int i = 0; i < 1280; ++i) {
     orig_pixels[i][0] = i;
     orig_pixels[i][1] = i / 2;
@@ -418,6 +430,20 @@ TEST_F(LibYUVPlanarTest, TestARGBGrayTo) {
   for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
     ARGBGrayTo(&orig_pixels[0][0], 0, &gray_pixels[0][0], 0, 1280, 1);
   }
+
+  for (int i = 0; i < 256; ++i) {
+    orig_pixels[i][0] = i;
+    orig_pixels[i][1] = i;
+    orig_pixels[i][2] = i;
+    orig_pixels[i][3] = i;
+  }
+  ARGBGray(&orig_pixels[0][0], 0, 0, 0, 256, 1);
+  for (int i = 0; i < 256; ++i) {
+    EXPECT_EQ(i, orig_pixels[i][0]);
+    EXPECT_EQ(i, orig_pixels[i][1]);
+    EXPECT_EQ(i, orig_pixels[i][2]);
+    EXPECT_EQ(i, orig_pixels[i][3]);
+  }
 }
 
 TEST_F(LibYUVPlanarTest, TestARGBSepia) {
@@ -763,27 +789,75 @@ TEST_F(LibYUVPlanarTest, TestARGBQuantize) {
   }
 }
 
-TEST_F(LibYUVPlanarTest, TestARGBMirror) {
-  SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
-  SIMD_ALIGNED(uint8_t dst_pixels[1280][4]);
+TEST_F(LibYUVPlanarTest, ARGBMirror_Opt) {
+  align_buffer_page_end(src_pixels, benchmark_width_ * benchmark_height_ * 4);
+  align_buffer_page_end(dst_pixels_opt,
+                        benchmark_width_ * benchmark_height_ * 4);
+  align_buffer_page_end(dst_pixels_c, benchmark_width_ * benchmark_height_ * 4);
 
-  for (int i = 0; i < 1280; ++i) {
-    orig_pixels[i][0] = i;
-    orig_pixels[i][1] = i / 2;
-    orig_pixels[i][2] = i / 3;
-    orig_pixels[i][3] = i / 4;
+  MemRandomize(src_pixels, benchmark_width_ * benchmark_height_ * 4);
+  MaskCpuFlags(disable_cpu_flags_);
+  ARGBMirror(src_pixels, benchmark_width_ * 4, dst_pixels_c,
+             benchmark_width_ * 4, benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    ARGBMirror(src_pixels, benchmark_width_ * 4, dst_pixels_opt,
+               benchmark_width_ * 4, benchmark_width_, benchmark_height_);
   }
-  ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 1280, 1);
+  for (int i = 0; i < benchmark_width_ * benchmark_height_ * 4; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
 
-  for (int i = 0; i < 1280; ++i) {
-    EXPECT_EQ(i & 255, dst_pixels[1280 - 1 - i][0]);
-    EXPECT_EQ((i / 2) & 255, dst_pixels[1280 - 1 - i][1]);
-    EXPECT_EQ((i / 3) & 255, dst_pixels[1280 - 1 - i][2]);
-    EXPECT_EQ((i / 4) & 255, dst_pixels[1280 - 1 - i][3]);
+TEST_F(LibYUVPlanarTest, MirrorPlane_Opt) {
+  align_buffer_page_end(src_pixels, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(dst_pixels_opt, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(dst_pixels_c, benchmark_width_ * benchmark_height_);
+
+  MemRandomize(src_pixels, benchmark_width_ * benchmark_height_);
+  MaskCpuFlags(disable_cpu_flags_);
+  MirrorPlane(src_pixels, benchmark_width_, dst_pixels_c, benchmark_width_,
+              benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    MirrorPlane(src_pixels, benchmark_width_, dst_pixels_opt, benchmark_width_,
+                benchmark_width_, benchmark_height_);
   }
-  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
-    ARGBMirror(&orig_pixels[0][0], 0, &dst_pixels[0][0], 0, 1280, 1);
+  for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
+
+TEST_F(LibYUVPlanarTest, MirrorUVPlane_Opt) {
+  align_buffer_page_end(src_pixels, benchmark_width_ * benchmark_height_ * 2);
+  align_buffer_page_end(dst_pixels_opt,
+                        benchmark_width_ * benchmark_height_ * 2);
+  align_buffer_page_end(dst_pixels_c, benchmark_width_ * benchmark_height_ * 2);
+
+  MemRandomize(src_pixels, benchmark_width_ * benchmark_height_ * 2);
+  MaskCpuFlags(disable_cpu_flags_);
+  MirrorUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_c,
+                benchmark_width_ * 2, benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    MirrorUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_opt,
+                  benchmark_width_ * 2, benchmark_width_, benchmark_height_);
   }
+  for (int i = 0; i < benchmark_width_ * benchmark_height_ * 2; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
 }
 
 TEST_F(LibYUVPlanarTest, TestShade) {
@@ -1006,10 +1080,91 @@ TEST_F(LibYUVPlanarTest, TestInterpolatePlane) {
   }
 }
 
+TEST_F(LibYUVPlanarTest, TestInterpolatePlane_16) {
+  SIMD_ALIGNED(uint16_t orig_pixels_0[1280]);
+  SIMD_ALIGNED(uint16_t orig_pixels_1[1280]);
+  SIMD_ALIGNED(uint16_t interpolate_pixels[1280]);
+  memset(orig_pixels_0, 0, sizeof(orig_pixels_0));
+  memset(orig_pixels_1, 0, sizeof(orig_pixels_1));
+
+  orig_pixels_0[0] = 16u;
+  orig_pixels_0[1] = 32u;
+  orig_pixels_0[2] = 64u;
+  orig_pixels_0[3] = 128u;
+  orig_pixels_0[4] = 0u;
+  orig_pixels_0[5] = 0u;
+  orig_pixels_0[6] = 0u;
+  orig_pixels_0[7] = 255u;
+  orig_pixels_0[8] = 0u;
+  orig_pixels_0[9] = 0u;
+  orig_pixels_0[10] = 0u;
+  orig_pixels_0[11] = 0u;
+  orig_pixels_0[12] = 0u;
+  orig_pixels_0[13] = 0u;
+  orig_pixels_0[14] = 0u;
+  orig_pixels_0[15] = 0u;
+
+  orig_pixels_1[0] = 0u;
+  orig_pixels_1[1] = 0u;
+  orig_pixels_1[2] = 0u;
+  orig_pixels_1[3] = 0u;
+  orig_pixels_1[4] = 0u;
+  orig_pixels_1[5] = 0u;
+  orig_pixels_1[6] = 0u;
+  orig_pixels_1[7] = 0u;
+  orig_pixels_1[8] = 0u;
+  orig_pixels_1[9] = 0u;
+  orig_pixels_1[10] = 0u;
+  orig_pixels_1[11] = 0u;
+  orig_pixels_1[12] = 255u;
+  orig_pixels_1[13] = 255u;
+  orig_pixels_1[14] = 255u;
+  orig_pixels_1[15] = 255u;
+
+  InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
+                      &interpolate_pixels[0], 0, 16, 1, 128);
+  EXPECT_EQ(8u, interpolate_pixels[0]);
+  EXPECT_EQ(16u, interpolate_pixels[1]);
+  EXPECT_EQ(32u, interpolate_pixels[2]);
+  EXPECT_EQ(64u, interpolate_pixels[3]);
+  EXPECT_EQ(0u, interpolate_pixels[4]);
+  EXPECT_EQ(0u, interpolate_pixels[5]);
+  EXPECT_EQ(0u, interpolate_pixels[6]);
+  EXPECT_EQ(128u, interpolate_pixels[7]);
+  EXPECT_EQ(0u, interpolate_pixels[8]);
+  EXPECT_EQ(0u, interpolate_pixels[9]);
+  EXPECT_EQ(0u, interpolate_pixels[10]);
+  EXPECT_EQ(0u, interpolate_pixels[11]);
+  EXPECT_EQ(128u, interpolate_pixels[12]);
+  EXPECT_EQ(128u, interpolate_pixels[13]);
+  EXPECT_EQ(128u, interpolate_pixels[14]);
+  EXPECT_EQ(128u, interpolate_pixels[15]);
+
+  InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
+                      &interpolate_pixels[0], 0, 16, 1, 0);
+  EXPECT_EQ(16u, interpolate_pixels[0]);
+  EXPECT_EQ(32u, interpolate_pixels[1]);
+  EXPECT_EQ(64u, interpolate_pixels[2]);
+  EXPECT_EQ(128u, interpolate_pixels[3]);
+
+  InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
+                      &interpolate_pixels[0], 0, 16, 1, 192);
+
+  EXPECT_EQ(4u, interpolate_pixels[0]);
+  EXPECT_EQ(8u, interpolate_pixels[1]);
+  EXPECT_EQ(16u, interpolate_pixels[2]);
+  EXPECT_EQ(32u, interpolate_pixels[3]);
+
+  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
+    InterpolatePlane_16(&orig_pixels_0[0], 0, &orig_pixels_1[0], 0,
+                        &interpolate_pixels[0], 0, 1280, 1, 123);
+  }
+}
+
 #define TESTTERP(FMT_A, BPP_A, STRIDE_A, FMT_B, BPP_B, STRIDE_B, W1280, TERP, \
                  N, NEG, OFF)                                                 \
   TEST_F(LibYUVPlanarTest, ARGBInterpolate##TERP##N) {                        \
-    const int kWidth = ((W1280) > 0) ? (W1280) : 1;                           \
+    const int kWidth = W1280;                                                 \
     const int kHeight = benchmark_height_;                                    \
     const int kStrideA =                                                      \
         (kWidth * BPP_A + STRIDE_A - 1) / STRIDE_A * STRIDE_A;                \
@@ -1041,7 +1196,7 @@ TEST_F(LibYUVPlanarTest, TestInterpolatePlane) {
   }
 
 #define TESTINTERPOLATE(TERP)                                                \
-  TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_ - 1, TERP, _Any, +, 0)   \
+  TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_ + 1, TERP, _Any, +, 0)   \
   TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Unaligned, +, 1) \
   TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Invert, -, 0)    \
   TESTTERP(ARGB, 4, 1, ARGB, 4, 1, benchmark_width_, TERP, _Opt, +, 0)
@@ -1058,7 +1213,8 @@ static int TestBlend(int width,
                      int disable_cpu_flags,
                      int benchmark_cpu_info,
                      int invert,
-                     int off) {
+                     int off,
+                     int attenuate) {
   if (width < 1) {
     width = 1;
   }
@@ -1072,10 +1228,12 @@ static int TestBlend(int width,
     src_argb_a[i + off] = (fastrand() & 0xff);
     src_argb_b[i + off] = (fastrand() & 0xff);
   }
-  ARGBAttenuate(src_argb_a + off, kStride, src_argb_a + off, kStride, width,
-                height);
-  ARGBAttenuate(src_argb_b + off, kStride, src_argb_b + off, kStride, width,
-                height);
+  MemRandomize(src_argb_a, kStride * height + off);
+  MemRandomize(src_argb_b, kStride * height + off);
+  if (attenuate) {
+    ARGBAttenuate(src_argb_a + off, kStride, src_argb_a + off, kStride, width,
+                  height);
+  }
   memset(dst_argb_c, 255, kStride * height);
   memset(dst_argb_opt, 255, kStride * height);
 
@@ -1104,29 +1262,36 @@ static int TestBlend(int width,
 
 TEST_F(LibYUVPlanarTest, ARGBBlend_Any) {
   int max_diff =
-      TestBlend(benchmark_width_ - 4, benchmark_height_, benchmark_iterations_,
-                disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
+      TestBlend(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
+                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 1);
   EXPECT_LE(max_diff, 1);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBBlend_Unaligned) {
   int max_diff =
       TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
-                disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
+                disable_cpu_flags_, benchmark_cpu_info_, +1, 1, 1);
   EXPECT_LE(max_diff, 1);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBBlend_Invert) {
   int max_diff =
       TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
-                disable_cpu_flags_, benchmark_cpu_info_, -1, 0);
+                disable_cpu_flags_, benchmark_cpu_info_, -1, 0, 1);
+  EXPECT_LE(max_diff, 1);
+}
+
+TEST_F(LibYUVPlanarTest, ARGBBlend_Unattenuated) {
+  int max_diff =
+      TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
+                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 0);
   EXPECT_LE(max_diff, 1);
 }
 
 TEST_F(LibYUVPlanarTest, ARGBBlend_Opt) {
   int max_diff =
       TestBlend(benchmark_width_, benchmark_height_, benchmark_iterations_,
-                disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
+                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, 1);
   EXPECT_LE(max_diff, 1);
 }
 
@@ -1203,7 +1368,7 @@ TEST_F(LibYUVPlanarTest, BlendPlane_Unaligned) {
                  disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
 }
 TEST_F(LibYUVPlanarTest, BlendPlane_Any) {
-  TestBlendPlane(benchmark_width_ - 4, benchmark_height_, benchmark_iterations_,
+  TestBlendPlane(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                  disable_cpu_flags_, benchmark_cpu_info_, +1, 1);
 }
 TEST_F(LibYUVPlanarTest, BlendPlane_Invert) {
@@ -1298,7 +1463,7 @@ TEST_F(LibYUVPlanarTest, I420Blend_Unaligned) {
 
 // TODO(fbarchard): DISABLED because _Any uses C.  Avoid C and re-enable.
 TEST_F(LibYUVPlanarTest, DISABLED_I420Blend_Any) {
-  TestI420Blend(benchmark_width_ - 4, benchmark_height_, benchmark_iterations_,
+  TestI420Blend(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                 disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
 }
 TEST_F(LibYUVPlanarTest, I420Blend_Invert) {
@@ -1400,6 +1565,212 @@ TEST_F(LibYUVPlanarTest, TestCopyPlane) {
   EXPECT_EQ(0, err);
 }
 
+TEST_F(LibYUVPlanarTest, CopyPlane_Opt) {
+  int i;
+  int y_plane_size = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(orig_y, y_plane_size);
+  align_buffer_page_end(dst_c, y_plane_size);
+  align_buffer_page_end(dst_opt, y_plane_size);
+
+  MemRandomize(orig_y, y_plane_size);
+  memset(dst_c, 1, y_plane_size);
+  memset(dst_opt, 2, y_plane_size);
+
+  // Disable all optimizations.
+  MaskCpuFlags(disable_cpu_flags_);
+  for (i = 0; i < benchmark_iterations_; i++) {
+    CopyPlane(orig_y, benchmark_width_, dst_c, benchmark_width_,
+              benchmark_width_, benchmark_height_);
+  }
+
+  // Enable optimizations.
+  MaskCpuFlags(benchmark_cpu_info_);
+  for (i = 0; i < benchmark_iterations_; i++) {
+    CopyPlane(orig_y, benchmark_width_, dst_opt, benchmark_width_,
+              benchmark_width_, benchmark_height_);
+  }
+
+  for (i = 0; i < y_plane_size; ++i) {
+    EXPECT_EQ(dst_c[i], dst_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(orig_y);
+  free_aligned_buffer_page_end(dst_c);
+  free_aligned_buffer_page_end(dst_opt);
+}
+
+TEST_F(LibYUVPlanarTest, TestCopyPlaneZero) {
+  // Test to verify copying a rect with a zero height or width does
+  // not touch destination memory.
+  uint8_t src = 42;
+  uint8_t dst = 0;
+
+  // Disable all optimizations.
+  MaskCpuFlags(disable_cpu_flags_);
+  CopyPlane(&src, 0, &dst, 0, 0, 0);
+  EXPECT_EQ(src, 42);
+  EXPECT_EQ(dst, 0);
+
+  CopyPlane(&src, 1, &dst, 1, 1, 0);
+  EXPECT_EQ(src, 42);
+  EXPECT_EQ(dst, 0);
+
+  CopyPlane(&src, 1, &dst, 1, 0, 1);
+  EXPECT_EQ(src, 42);
+  EXPECT_EQ(dst, 0);
+
+  // Enable optimizations.
+  MaskCpuFlags(benchmark_cpu_info_);
+  CopyPlane(&src, 0, &dst, 0, 0, 0);
+  EXPECT_EQ(src, 42);
+  EXPECT_EQ(dst, 0);
+
+  CopyPlane(&src, 1, &dst, 1, 1, 0);
+  EXPECT_EQ(src, 42);
+  EXPECT_EQ(dst, 0);
+
+  CopyPlane(&src, 1, &dst, 1, 0, 1);
+  EXPECT_EQ(src, 42);
+  EXPECT_EQ(dst, 0);
+}
+
+TEST_F(LibYUVPlanarTest, TestDetilePlane) {
+  int i, j;
+
+  // orig is tiled.  Allocate enough memory for tiles.
+  int orig_width = (benchmark_width_ + 15) & ~15;
+  int orig_height = (benchmark_height_ + 15) & ~15;
+  int orig_plane_size = orig_width * orig_height;
+  int y_plane_size = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(orig_y, orig_plane_size);
+  align_buffer_page_end(dst_c, y_plane_size);
+  align_buffer_page_end(dst_opt, y_plane_size);
+
+  MemRandomize(orig_y, orig_plane_size);
+  memset(dst_c, 0, y_plane_size);
+  memset(dst_opt, 0, y_plane_size);
+
+  // Disable all optimizations.
+  MaskCpuFlags(disable_cpu_flags_);
+  for (j = 0; j < benchmark_iterations_; j++) {
+    DetilePlane(orig_y, orig_width, dst_c, benchmark_width_, benchmark_width_,
+                benchmark_height_, 16);
+  }
+
+  // Enable optimizations.
+  MaskCpuFlags(benchmark_cpu_info_);
+  for (j = 0; j < benchmark_iterations_; j++) {
+    DetilePlane(orig_y, orig_width, dst_opt, benchmark_width_, benchmark_width_,
+                benchmark_height_, 16);
+  }
+
+  for (i = 0; i < y_plane_size; ++i) {
+    EXPECT_EQ(dst_c[i], dst_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(orig_y);
+  free_aligned_buffer_page_end(dst_c);
+  free_aligned_buffer_page_end(dst_opt);
+}
+
+// Compares DetileSplitUV to 2 step Detile + SplitUV
+TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Correctness) {
+  int i, j;
+
+  // orig is tiled.  Allocate enough memory for tiles.
+  int orig_width = (benchmark_width_ + 15) & ~15;
+  int orig_height = (benchmark_height_ + 15) & ~15;
+  int orig_plane_size = orig_width * orig_height;
+  int uv_plane_size = ((benchmark_width_ + 1) / 2) * benchmark_height_;
+  align_buffer_page_end(orig_uv, orig_plane_size);
+  align_buffer_page_end(detiled_uv, orig_plane_size);
+  align_buffer_page_end(dst_u_two_stage, uv_plane_size);
+  align_buffer_page_end(dst_u_opt, uv_plane_size);
+  align_buffer_page_end(dst_v_two_stage, uv_plane_size);
+  align_buffer_page_end(dst_v_opt, uv_plane_size);
+
+  MemRandomize(orig_uv, orig_plane_size);
+  memset(detiled_uv, 0, orig_plane_size);
+  memset(dst_u_two_stage, 0, uv_plane_size);
+  memset(dst_u_opt, 0, uv_plane_size);
+  memset(dst_v_two_stage, 0, uv_plane_size);
+  memset(dst_v_opt, 0, uv_plane_size);
+
+  DetileSplitUVPlane(orig_uv, orig_width, dst_u_opt, (benchmark_width_ + 1) / 2,
+                     dst_v_opt, (benchmark_width_ + 1) / 2, benchmark_width_,
+                     benchmark_height_, 16);
+
+  // Benchmark 2 step conversion for comparison.
+  for (j = 0; j < benchmark_iterations_; j++) {
+    DetilePlane(orig_uv, orig_width, detiled_uv, benchmark_width_,
+                benchmark_width_, benchmark_height_, 16);
+    SplitUVPlane(detiled_uv, orig_width, dst_u_two_stage,
+                 (benchmark_width_ + 1) / 2, dst_v_two_stage,
+                 (benchmark_width_ + 1) / 2, (benchmark_width_ + 1) / 2,
+                 benchmark_height_);
+  }
+
+  for (i = 0; i < uv_plane_size; ++i) {
+    EXPECT_EQ(dst_u_two_stage[i], dst_u_opt[i]);
+    EXPECT_EQ(dst_v_two_stage[i], dst_v_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(orig_uv);
+  free_aligned_buffer_page_end(detiled_uv);
+  free_aligned_buffer_page_end(dst_u_two_stage);
+  free_aligned_buffer_page_end(dst_u_opt);
+  free_aligned_buffer_page_end(dst_v_two_stage);
+  free_aligned_buffer_page_end(dst_v_opt);
+}
+
+TEST_F(LibYUVPlanarTest, TestDetileSplitUVPlane_Benchmark) {
+  int i, j;
+
+  // orig is tiled.  Allocate enough memory for tiles.
+  int orig_width = (benchmark_width_ + 15) & ~15;
+  int orig_height = (benchmark_height_ + 15) & ~15;
+  int orig_plane_size = orig_width * orig_height;
+  int uv_plane_size = ((benchmark_width_ + 1) / 2) * benchmark_height_;
+  align_buffer_page_end(orig_uv, orig_plane_size);
+  align_buffer_page_end(dst_u_c, uv_plane_size);
+  align_buffer_page_end(dst_u_opt, uv_plane_size);
+  align_buffer_page_end(dst_v_c, uv_plane_size);
+  align_buffer_page_end(dst_v_opt, uv_plane_size);
+
+  MemRandomize(orig_uv, orig_plane_size);
+  memset(dst_u_c, 0, uv_plane_size);
+  memset(dst_u_opt, 0, uv_plane_size);
+  memset(dst_v_c, 0, uv_plane_size);
+  memset(dst_v_opt, 0, uv_plane_size);
+
+  // Disable all optimizations.
+  MaskCpuFlags(disable_cpu_flags_);
+
+  DetileSplitUVPlane(orig_uv, orig_width, dst_u_c, (benchmark_width_ + 1) / 2,
+                     dst_v_c, (benchmark_width_ + 1) / 2, benchmark_width_,
+                     benchmark_height_, 16);
+
+  // Enable optimizations.
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (j = 0; j < benchmark_iterations_; j++) {
+    DetileSplitUVPlane(
+        orig_uv, orig_width, dst_u_opt, (benchmark_width_ + 1) / 2, dst_v_opt,
+        (benchmark_width_ + 1) / 2, benchmark_width_, benchmark_height_, 16);
+  }
+
+  for (i = 0; i < uv_plane_size; ++i) {
+    EXPECT_EQ(dst_u_c[i], dst_u_opt[i]);
+    EXPECT_EQ(dst_v_c[i], dst_v_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(orig_uv);
+  free_aligned_buffer_page_end(dst_u_c);
+  free_aligned_buffer_page_end(dst_u_opt);
+  free_aligned_buffer_page_end(dst_v_c);
+  free_aligned_buffer_page_end(dst_v_opt);
+}
+
 static int TestMultiply(int width,
                         int height,
                         int benchmark_iterations,
@@ -1447,7 +1818,7 @@ static int TestMultiply(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBMultiply_Any) {
-  int max_diff = TestMultiply(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestMultiply(benchmark_width_ + 1, benchmark_height_,
                               benchmark_iterations_, disable_cpu_flags_,
                               benchmark_cpu_info_, +1, 0);
   EXPECT_LE(max_diff, 1);
@@ -1522,7 +1893,7 @@ static int TestAdd(int width,
 
 TEST_F(LibYUVPlanarTest, ARGBAdd_Any) {
   int max_diff =
-      TestAdd(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_,
+      TestAdd(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
               disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
   EXPECT_LE(max_diff, 1);
 }
@@ -1595,7 +1966,7 @@ static int TestSubtract(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBSubtract_Any) {
-  int max_diff = TestSubtract(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestSubtract(benchmark_width_ + 1, benchmark_height_,
                               benchmark_iterations_, disable_cpu_flags_,
                               benchmark_cpu_info_, +1, 0);
   EXPECT_LE(max_diff, 1);
@@ -1668,7 +2039,7 @@ static int TestSobel(int width,
 
 TEST_F(LibYUVPlanarTest, ARGBSobel_Any) {
   int max_diff =
-      TestSobel(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_,
+      TestSobel(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                 disable_cpu_flags_, benchmark_cpu_info_, +1, 0);
   EXPECT_EQ(0, max_diff);
 }
@@ -1741,7 +2112,7 @@ static int TestSobelToPlane(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBSobelToPlane_Any) {
-  int max_diff = TestSobelToPlane(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestSobelToPlane(benchmark_width_ + 1, benchmark_height_,
                                   benchmark_iterations_, disable_cpu_flags_,
                                   benchmark_cpu_info_, +1, 0);
   EXPECT_EQ(0, max_diff);
@@ -1813,7 +2184,7 @@ static int TestSobelXY(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBSobelXY_Any) {
-  int max_diff = TestSobelXY(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestSobelXY(benchmark_width_ + 1, benchmark_height_,
                              benchmark_iterations_, disable_cpu_flags_,
                              benchmark_cpu_info_, +1, 0);
   EXPECT_EQ(0, max_diff);
@@ -1889,29 +2260,35 @@ static int TestBlur(int width,
   return max_diff;
 }
 
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+#define DISABLED_ARM(name) name
+#else
+#define DISABLED_ARM(name) DISABLED_##name
+#endif
+
 static const int kBlurSize = 55;
-TEST_F(LibYUVPlanarTest, ARGBBlur_Any) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Any)) {
   int max_diff =
-      TestBlur(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_,
+      TestBlur(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, ARGBBlur_Unaligned) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Unaligned)) {
   int max_diff =
       TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 1, kBlurSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, ARGBBlur_Invert) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Invert)) {
   int max_diff =
       TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, -1, 0, kBlurSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, ARGBBlur_Opt) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlur_Opt)) {
   int max_diff =
       TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSize);
@@ -1919,35 +2296,35 @@ TEST_F(LibYUVPlanarTest, ARGBBlur_Opt) {
 }
 
 static const int kBlurSmallSize = 5;
-TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Any) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Any)) {
   int max_diff =
-      TestBlur(benchmark_width_ - 1, benchmark_height_, benchmark_iterations_,
+      TestBlur(benchmark_width_ + 1, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSmallSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Unaligned) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Unaligned)) {
   int max_diff =
       TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 1, kBlurSmallSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Invert) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Invert)) {
   int max_diff =
       TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, -1, 0, kBlurSmallSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, ARGBBlurSmall_Opt) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(ARGBBlurSmall_Opt)) {
   int max_diff =
       TestBlur(benchmark_width_, benchmark_height_, benchmark_iterations_,
                disable_cpu_flags_, benchmark_cpu_info_, +1, 0, kBlurSmallSize);
   EXPECT_LE(max_diff, 1);
 }
 
-TEST_F(LibYUVPlanarTest, TestARGBPolynomial) {
+TEST_F(LibYUVPlanarTest, DISABLED_ARM(TestARGBPolynomial)) {
   SIMD_ALIGNED(uint8_t orig_pixels[1280][4]);
   SIMD_ALIGNED(uint8_t dst_pixels_opt[1280][4]);
   SIMD_ALIGNED(uint8_t dst_pixels_c[1280][4]);
@@ -2426,7 +2803,7 @@ static int TestARGBRect(int width,
 }
 
 TEST_F(LibYUVPlanarTest, ARGBRect_Any) {
-  int max_diff = TestARGBRect(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestARGBRect(benchmark_width_ + 1, benchmark_height_,
                               benchmark_iterations_, disable_cpu_flags_,
                               benchmark_cpu_info_, +1, 0, 4);
   EXPECT_EQ(0, max_diff);
@@ -2454,7 +2831,7 @@ TEST_F(LibYUVPlanarTest, ARGBRect_Opt) {
 }
 
 TEST_F(LibYUVPlanarTest, SetPlane_Any) {
-  int max_diff = TestARGBRect(benchmark_width_ - 1, benchmark_height_,
+  int max_diff = TestARGBRect(benchmark_width_ + 1, benchmark_height_,
                               benchmark_iterations_, disable_cpu_flags_,
                               benchmark_cpu_info_, +1, 0, 1);
   EXPECT_EQ(0, max_diff);
@@ -2483,33 +2860,24 @@ TEST_F(LibYUVPlanarTest, SetPlane_Opt) {
 
 TEST_F(LibYUVPlanarTest, MergeUVPlane_Opt) {
   const int kPixels = benchmark_width_ * benchmark_height_;
-  align_buffer_page_end(src_pixels, kPixels * 2);
-  align_buffer_page_end(tmp_pixels_u, kPixels);
-  align_buffer_page_end(tmp_pixels_v, kPixels);
+  align_buffer_page_end(src_pixels_u, kPixels);
+  align_buffer_page_end(src_pixels_v, kPixels);
   align_buffer_page_end(dst_pixels_opt, kPixels * 2);
   align_buffer_page_end(dst_pixels_c, kPixels * 2);
 
-  MemRandomize(src_pixels, kPixels * 2);
-  MemRandomize(tmp_pixels_u, kPixels);
-  MemRandomize(tmp_pixels_v, kPixels);
+  MemRandomize(src_pixels_u, kPixels);
+  MemRandomize(src_pixels_v, kPixels);
   MemRandomize(dst_pixels_opt, kPixels * 2);
   MemRandomize(dst_pixels_c, kPixels * 2);
 
   MaskCpuFlags(disable_cpu_flags_);
-  SplitUVPlane(src_pixels, benchmark_width_ * 2, tmp_pixels_u, benchmark_width_,
-               tmp_pixels_v, benchmark_width_, benchmark_width_,
-               benchmark_height_);
-  MergeUVPlane(tmp_pixels_u, benchmark_width_, tmp_pixels_v, benchmark_width_,
+  MergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v, benchmark_width_,
                dst_pixels_c, benchmark_width_ * 2, benchmark_width_,
                benchmark_height_);
   MaskCpuFlags(benchmark_cpu_info_);
 
-  SplitUVPlane(src_pixels, benchmark_width_ * 2, tmp_pixels_u, benchmark_width_,
-               tmp_pixels_v, benchmark_width_, benchmark_width_,
-               benchmark_height_);
-
   for (int i = 0; i < benchmark_iterations_; ++i) {
-    MergeUVPlane(tmp_pixels_u, benchmark_width_, tmp_pixels_v, benchmark_width_,
+    MergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v, benchmark_width_,
                  dst_pixels_opt, benchmark_width_ * 2, benchmark_width_,
                  benchmark_height_);
   }
@@ -2518,9 +2886,43 @@ TEST_F(LibYUVPlanarTest, MergeUVPlane_Opt) {
     EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
   }
 
-  free_aligned_buffer_page_end(src_pixels);
-  free_aligned_buffer_page_end(tmp_pixels_u);
-  free_aligned_buffer_page_end(tmp_pixels_v);
+  free_aligned_buffer_page_end(src_pixels_u);
+  free_aligned_buffer_page_end(src_pixels_v);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
+
+// 16 bit channel split and merge
+TEST_F(LibYUVPlanarTest, MergeUVPlane_16_Opt) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels_u, kPixels * 2);
+  align_buffer_page_end(src_pixels_v, kPixels * 2);
+  align_buffer_page_end(dst_pixels_opt, kPixels * 2 * 2);
+  align_buffer_page_end(dst_pixels_c, kPixels * 2 * 2);
+  MemRandomize(src_pixels_u, kPixels * 2);
+  MemRandomize(src_pixels_v, kPixels * 2);
+  MemRandomize(dst_pixels_opt, kPixels * 2 * 2);
+  MemRandomize(dst_pixels_c, kPixels * 2 * 2);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  MergeUVPlane_16((const uint16_t*)src_pixels_u, benchmark_width_,
+                  (const uint16_t*)src_pixels_v, benchmark_width_,
+                  (uint16_t*)dst_pixels_c, benchmark_width_ * 2,
+                  benchmark_width_, benchmark_height_, 12);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    MergeUVPlane_16((const uint16_t*)src_pixels_u, benchmark_width_,
+                    (const uint16_t*)src_pixels_v, benchmark_width_,
+                    (uint16_t*)dst_pixels_opt, benchmark_width_ * 2,
+                    benchmark_width_, benchmark_height_, 12);
+  }
+
+  for (int i = 0; i < kPixels * 2 * 2; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+  free_aligned_buffer_page_end(src_pixels_u);
+  free_aligned_buffer_page_end(src_pixels_v);
   free_aligned_buffer_page_end(dst_pixels_opt);
   free_aligned_buffer_page_end(dst_pixels_c);
 }
@@ -2528,47 +2930,112 @@ TEST_F(LibYUVPlanarTest, MergeUVPlane_Opt) {
 TEST_F(LibYUVPlanarTest, SplitUVPlane_Opt) {
   const int kPixels = benchmark_width_ * benchmark_height_;
   align_buffer_page_end(src_pixels, kPixels * 2);
-  align_buffer_page_end(tmp_pixels_u, kPixels);
-  align_buffer_page_end(tmp_pixels_v, kPixels);
+  align_buffer_page_end(dst_pixels_u_c, kPixels);
+  align_buffer_page_end(dst_pixels_v_c, kPixels);
+  align_buffer_page_end(dst_pixels_u_opt, kPixels);
+  align_buffer_page_end(dst_pixels_v_opt, kPixels);
+
+  MemRandomize(src_pixels, kPixels * 2);
+  MemRandomize(dst_pixels_u_c, kPixels);
+  MemRandomize(dst_pixels_v_c, kPixels);
+  MemRandomize(dst_pixels_u_opt, kPixels);
+  MemRandomize(dst_pixels_v_opt, kPixels);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  SplitUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_u_c,
+               benchmark_width_, dst_pixels_v_c, benchmark_width_,
+               benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    SplitUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_u_opt,
+                 benchmark_width_, dst_pixels_v_opt, benchmark_width_,
+                 benchmark_width_, benchmark_height_);
+  }
+
+  for (int i = 0; i < kPixels; ++i) {
+    EXPECT_EQ(dst_pixels_u_c[i], dst_pixels_u_opt[i]);
+    EXPECT_EQ(dst_pixels_v_c[i], dst_pixels_v_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(dst_pixels_u_c);
+  free_aligned_buffer_page_end(dst_pixels_v_c);
+  free_aligned_buffer_page_end(dst_pixels_u_opt);
+  free_aligned_buffer_page_end(dst_pixels_v_opt);
+}
+
+// 16 bit channel split
+TEST_F(LibYUVPlanarTest, SplitUVPlane_16_Opt) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels, kPixels * 2 * 2);
+  align_buffer_page_end(dst_pixels_u_c, kPixels * 2);
+  align_buffer_page_end(dst_pixels_v_c, kPixels * 2);
+  align_buffer_page_end(dst_pixels_u_opt, kPixels * 2);
+  align_buffer_page_end(dst_pixels_v_opt, kPixels * 2);
+  MemRandomize(src_pixels, kPixels * 2 * 2);
+  MemRandomize(dst_pixels_u_c, kPixels * 2);
+  MemRandomize(dst_pixels_v_c, kPixels * 2);
+  MemRandomize(dst_pixels_u_opt, kPixels * 2);
+  MemRandomize(dst_pixels_v_opt, kPixels * 2);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  SplitUVPlane_16((const uint16_t*)src_pixels, benchmark_width_ * 2,
+                  (uint16_t*)dst_pixels_u_c, benchmark_width_,
+                  (uint16_t*)dst_pixels_v_c, benchmark_width_, benchmark_width_,
+                  benchmark_height_, 10);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    SplitUVPlane_16((const uint16_t*)src_pixels, benchmark_width_ * 2,
+                    (uint16_t*)dst_pixels_u_opt, benchmark_width_,
+                    (uint16_t*)dst_pixels_v_opt, benchmark_width_,
+                    benchmark_width_, benchmark_height_, 10);
+  }
+
+  for (int i = 0; i < kPixels * 2; ++i) {
+    EXPECT_EQ(dst_pixels_u_c[i], dst_pixels_u_opt[i]);
+    EXPECT_EQ(dst_pixels_v_c[i], dst_pixels_v_opt[i]);
+  }
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(dst_pixels_u_c);
+  free_aligned_buffer_page_end(dst_pixels_v_c);
+  free_aligned_buffer_page_end(dst_pixels_u_opt);
+  free_aligned_buffer_page_end(dst_pixels_v_opt);
+}
+
+TEST_F(LibYUVPlanarTest, SwapUVPlane_Opt) {
+  // Round count up to multiple of 16
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels, kPixels * 2);
   align_buffer_page_end(dst_pixels_opt, kPixels * 2);
   align_buffer_page_end(dst_pixels_c, kPixels * 2);
 
   MemRandomize(src_pixels, kPixels * 2);
-  MemRandomize(tmp_pixels_u, kPixels);
-  MemRandomize(tmp_pixels_v, kPixels);
   MemRandomize(dst_pixels_opt, kPixels * 2);
   MemRandomize(dst_pixels_c, kPixels * 2);
 
   MaskCpuFlags(disable_cpu_flags_);
-  SplitUVPlane(src_pixels, benchmark_width_ * 2, tmp_pixels_u, benchmark_width_,
-               tmp_pixels_v, benchmark_width_, benchmark_width_,
-               benchmark_height_);
-  MergeUVPlane(tmp_pixels_u, benchmark_width_, tmp_pixels_v, benchmark_width_,
-               dst_pixels_c, benchmark_width_ * 2, benchmark_width_,
-               benchmark_height_);
+  SwapUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_c,
+              benchmark_width_ * 2, benchmark_width_, benchmark_height_);
   MaskCpuFlags(benchmark_cpu_info_);
 
   for (int i = 0; i < benchmark_iterations_; ++i) {
-    SplitUVPlane(src_pixels, benchmark_width_ * 2, tmp_pixels_u,
-                 benchmark_width_, tmp_pixels_v, benchmark_width_,
-                 benchmark_width_, benchmark_height_);
+    SwapUVPlane(src_pixels, benchmark_width_ * 2, dst_pixels_opt,
+                benchmark_width_ * 2, benchmark_width_, benchmark_height_);
   }
-  MergeUVPlane(tmp_pixels_u, benchmark_width_, tmp_pixels_v, benchmark_width_,
-               dst_pixels_opt, benchmark_width_ * 2, benchmark_width_,
-               benchmark_height_);
 
   for (int i = 0; i < kPixels * 2; ++i) {
     EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
   }
 
   free_aligned_buffer_page_end(src_pixels);
-  free_aligned_buffer_page_end(tmp_pixels_u);
-  free_aligned_buffer_page_end(tmp_pixels_v);
   free_aligned_buffer_page_end(dst_pixels_opt);
   free_aligned_buffer_page_end(dst_pixels_c);
 }
 
 TEST_F(LibYUVPlanarTest, MergeRGBPlane_Opt) {
+  // Round count up to multiple of 16
   const int kPixels = benchmark_width_ * benchmark_height_;
   align_buffer_page_end(src_pixels, kPixels * 3);
   align_buffer_page_end(tmp_pixels_r, kPixels);
@@ -2617,6 +3084,7 @@ TEST_F(LibYUVPlanarTest, MergeRGBPlane_Opt) {
 }
 
 TEST_F(LibYUVPlanarTest, SplitRGBPlane_Opt) {
+  // Round count up to multiple of 16
   const int kPixels = benchmark_width_ * benchmark_height_;
   align_buffer_page_end(src_pixels, kPixels * 3);
   align_buffer_page_end(tmp_pixels_r, kPixels);
@@ -2663,10 +3131,373 @@ TEST_F(LibYUVPlanarTest, SplitRGBPlane_Opt) {
   free_aligned_buffer_page_end(dst_pixels_c);
 }
 
+TEST_F(LibYUVPlanarTest, MergeARGBPlane_Opt) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels, kPixels * 4);
+  align_buffer_page_end(tmp_pixels_r, kPixels);
+  align_buffer_page_end(tmp_pixels_g, kPixels);
+  align_buffer_page_end(tmp_pixels_b, kPixels);
+  align_buffer_page_end(tmp_pixels_a, kPixels);
+  align_buffer_page_end(dst_pixels_opt, kPixels * 4);
+  align_buffer_page_end(dst_pixels_c, kPixels * 4);
+
+  MemRandomize(src_pixels, kPixels * 4);
+  MemRandomize(tmp_pixels_r, kPixels);
+  MemRandomize(tmp_pixels_g, kPixels);
+  MemRandomize(tmp_pixels_b, kPixels);
+  MemRandomize(tmp_pixels_a, kPixels);
+  MemRandomize(dst_pixels_opt, kPixels * 4);
+  MemRandomize(dst_pixels_c, kPixels * 4);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                 benchmark_width_, tmp_pixels_g, benchmark_width_, tmp_pixels_b,
+                 benchmark_width_, tmp_pixels_a, benchmark_width_,
+                 benchmark_width_, benchmark_height_);
+  MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g, benchmark_width_,
+                 tmp_pixels_b, benchmark_width_, tmp_pixels_a, benchmark_width_,
+                 dst_pixels_c, benchmark_width_ * 4, benchmark_width_,
+                 benchmark_height_);
+
+  MaskCpuFlags(benchmark_cpu_info_);
+  SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                 benchmark_width_, tmp_pixels_g, benchmark_width_, tmp_pixels_b,
+                 benchmark_width_, tmp_pixels_a, benchmark_width_,
+                 benchmark_width_, benchmark_height_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g,
+                   benchmark_width_, tmp_pixels_b, benchmark_width_,
+                   tmp_pixels_a, benchmark_width_, dst_pixels_opt,
+                   benchmark_width_ * 4, benchmark_width_, benchmark_height_);
+  }
+
+  for (int i = 0; i < kPixels * 4; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(tmp_pixels_r);
+  free_aligned_buffer_page_end(tmp_pixels_g);
+  free_aligned_buffer_page_end(tmp_pixels_b);
+  free_aligned_buffer_page_end(tmp_pixels_a);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
+
+TEST_F(LibYUVPlanarTest, SplitARGBPlane_Opt) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels, kPixels * 4);
+  align_buffer_page_end(tmp_pixels_r, kPixels);
+  align_buffer_page_end(tmp_pixels_g, kPixels);
+  align_buffer_page_end(tmp_pixels_b, kPixels);
+  align_buffer_page_end(tmp_pixels_a, kPixels);
+  align_buffer_page_end(dst_pixels_opt, kPixels * 4);
+  align_buffer_page_end(dst_pixels_c, kPixels * 4);
+
+  MemRandomize(src_pixels, kPixels * 4);
+  MemRandomize(tmp_pixels_r, kPixels);
+  MemRandomize(tmp_pixels_g, kPixels);
+  MemRandomize(tmp_pixels_b, kPixels);
+  MemRandomize(tmp_pixels_a, kPixels);
+  MemRandomize(dst_pixels_opt, kPixels * 4);
+  MemRandomize(dst_pixels_c, kPixels * 4);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                 benchmark_width_, tmp_pixels_g, benchmark_width_, tmp_pixels_b,
+                 benchmark_width_, tmp_pixels_a, benchmark_width_,
+                 benchmark_width_, benchmark_height_);
+  MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g, benchmark_width_,
+                 tmp_pixels_b, benchmark_width_, tmp_pixels_a, benchmark_width_,
+                 dst_pixels_c, benchmark_width_ * 4, benchmark_width_,
+                 benchmark_height_);
+
+  MaskCpuFlags(benchmark_cpu_info_);
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                   benchmark_width_, tmp_pixels_g, benchmark_width_,
+                   tmp_pixels_b, benchmark_width_, tmp_pixels_a,
+                   benchmark_width_, benchmark_width_, benchmark_height_);
+  }
+
+  MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g, benchmark_width_,
+                 tmp_pixels_b, benchmark_width_, tmp_pixels_a, benchmark_width_,
+                 dst_pixels_opt, benchmark_width_ * 4, benchmark_width_,
+                 benchmark_height_);
+
+  for (int i = 0; i < kPixels * 4; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(tmp_pixels_r);
+  free_aligned_buffer_page_end(tmp_pixels_g);
+  free_aligned_buffer_page_end(tmp_pixels_b);
+  free_aligned_buffer_page_end(tmp_pixels_a);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
+
+TEST_F(LibYUVPlanarTest, MergeXRGBPlane_Opt) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels, kPixels * 4);
+  align_buffer_page_end(tmp_pixels_r, kPixels);
+  align_buffer_page_end(tmp_pixels_g, kPixels);
+  align_buffer_page_end(tmp_pixels_b, kPixels);
+  align_buffer_page_end(dst_pixels_opt, kPixels * 4);
+  align_buffer_page_end(dst_pixels_c, kPixels * 4);
+
+  MemRandomize(src_pixels, kPixels * 4);
+  MemRandomize(tmp_pixels_r, kPixels);
+  MemRandomize(tmp_pixels_g, kPixels);
+  MemRandomize(tmp_pixels_b, kPixels);
+  MemRandomize(dst_pixels_opt, kPixels * 4);
+  MemRandomize(dst_pixels_c, kPixels * 4);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                 benchmark_width_, tmp_pixels_g, benchmark_width_, tmp_pixels_b,
+                 benchmark_width_, NULL, 0, benchmark_width_,
+                 benchmark_height_);
+  MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g, benchmark_width_,
+                 tmp_pixels_b, benchmark_width_, NULL, 0, dst_pixels_c,
+                 benchmark_width_ * 4, benchmark_width_, benchmark_height_);
+
+  MaskCpuFlags(benchmark_cpu_info_);
+  SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                 benchmark_width_, tmp_pixels_g, benchmark_width_, tmp_pixels_b,
+                 benchmark_width_, NULL, 0, benchmark_width_,
+                 benchmark_height_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g,
+                   benchmark_width_, tmp_pixels_b, benchmark_width_, NULL, 0,
+                   dst_pixels_opt, benchmark_width_ * 4, benchmark_width_,
+                   benchmark_height_);
+  }
+
+  for (int i = 0; i < kPixels * 4; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(tmp_pixels_r);
+  free_aligned_buffer_page_end(tmp_pixels_g);
+  free_aligned_buffer_page_end(tmp_pixels_b);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
+
+TEST_F(LibYUVPlanarTest, SplitXRGBPlane_Opt) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels, kPixels * 4);
+  align_buffer_page_end(tmp_pixels_r, kPixels);
+  align_buffer_page_end(tmp_pixels_g, kPixels);
+  align_buffer_page_end(tmp_pixels_b, kPixels);
+  align_buffer_page_end(dst_pixels_opt, kPixels * 4);
+  align_buffer_page_end(dst_pixels_c, kPixels * 4);
+
+  MemRandomize(src_pixels, kPixels * 4);
+  MemRandomize(tmp_pixels_r, kPixels);
+  MemRandomize(tmp_pixels_g, kPixels);
+  MemRandomize(tmp_pixels_b, kPixels);
+  MemRandomize(dst_pixels_opt, kPixels * 4);
+  MemRandomize(dst_pixels_c, kPixels * 4);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                 benchmark_width_, tmp_pixels_g, benchmark_width_, tmp_pixels_b,
+                 benchmark_width_, NULL, 0, benchmark_width_,
+                 benchmark_height_);
+  MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g, benchmark_width_,
+                 tmp_pixels_b, benchmark_width_, NULL, 0, dst_pixels_c,
+                 benchmark_width_ * 4, benchmark_width_, benchmark_height_);
+
+  MaskCpuFlags(benchmark_cpu_info_);
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    SplitARGBPlane(src_pixels, benchmark_width_ * 4, tmp_pixels_r,
+                   benchmark_width_, tmp_pixels_g, benchmark_width_,
+                   tmp_pixels_b, benchmark_width_, NULL, 0, benchmark_width_,
+                   benchmark_height_);
+  }
+
+  MergeARGBPlane(tmp_pixels_r, benchmark_width_, tmp_pixels_g, benchmark_width_,
+                 tmp_pixels_b, benchmark_width_, NULL, 0, dst_pixels_opt,
+                 benchmark_width_ * 4, benchmark_width_, benchmark_height_);
+
+  for (int i = 0; i < kPixels * 4; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels);
+  free_aligned_buffer_page_end(tmp_pixels_r);
+  free_aligned_buffer_page_end(tmp_pixels_g);
+  free_aligned_buffer_page_end(tmp_pixels_b);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(dst_pixels_c);
+}
+
+// Merge 4 channels
+#define TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, W1280, N, NEG, OFF)      \
+  TEST_F(LibYUVPlanarTest, FUNC##Plane_##DEPTH##N) {                        \
+    const int kWidth = W1280;                                               \
+    const int kPixels = kWidth * benchmark_height_;                         \
+    align_buffer_page_end(src_memory_r, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_g, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_b, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_a, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(dst_memory_c, kPixels * 4 * sizeof(DTYPE));       \
+    align_buffer_page_end(dst_memory_opt, kPixels * 4 * sizeof(DTYPE));     \
+    MemRandomize(src_memory_r, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_g, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_b, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_a, kPixels * sizeof(STYPE) + OFF);              \
+    memset(dst_memory_c, 0, kPixels * 4 * sizeof(DTYPE));                   \
+    memset(dst_memory_opt, 0, kPixels * 4 * sizeof(DTYPE));                 \
+    STYPE* src_pixels_r = reinterpret_cast<STYPE*>(src_memory_r + OFF);     \
+    STYPE* src_pixels_g = reinterpret_cast<STYPE*>(src_memory_g + OFF);     \
+    STYPE* src_pixels_b = reinterpret_cast<STYPE*>(src_memory_b + OFF);     \
+    STYPE* src_pixels_a = reinterpret_cast<STYPE*>(src_memory_a + OFF);     \
+    DTYPE* dst_pixels_c = reinterpret_cast<DTYPE*>(dst_memory_c);           \
+    DTYPE* dst_pixels_opt = reinterpret_cast<DTYPE*>(dst_memory_opt);       \
+    MaskCpuFlags(disable_cpu_flags_);                                       \
+    FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b,   \
+                kWidth, src_pixels_a, kWidth, dst_pixels_c, kWidth * 4,     \
+                kWidth, NEG benchmark_height_, DEPTH);                      \
+    MaskCpuFlags(benchmark_cpu_info_);                                      \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                       \
+      FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \
+                  kWidth, src_pixels_a, kWidth, dst_pixels_opt, kWidth * 4, \
+                  kWidth, NEG benchmark_height_, DEPTH);                    \
+    }                                                                       \
+    for (int i = 0; i < kPixels * 4; ++i) {                                 \
+      EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);                        \
+    }                                                                       \
+    free_aligned_buffer_page_end(src_memory_r);                             \
+    free_aligned_buffer_page_end(src_memory_g);                             \
+    free_aligned_buffer_page_end(src_memory_b);                             \
+    free_aligned_buffer_page_end(src_memory_a);                             \
+    free_aligned_buffer_page_end(dst_memory_c);                             \
+    free_aligned_buffer_page_end(dst_memory_opt);                           \
+  }
+
+// Merge 3 channel RGB into 4 channel XRGB with opaque alpha
+#define TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, W1280, N, NEG, OFF)     \
+  TEST_F(LibYUVPlanarTest, FUNC##Plane_Opaque_##DEPTH##N) {                 \
+    const int kWidth = W1280;                                               \
+    const int kPixels = kWidth * benchmark_height_;                         \
+    align_buffer_page_end(src_memory_r, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_g, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_b, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(dst_memory_c, kPixels * 4 * sizeof(DTYPE));       \
+    align_buffer_page_end(dst_memory_opt, kPixels * 4 * sizeof(DTYPE));     \
+    MemRandomize(src_memory_r, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_g, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_b, kPixels * sizeof(STYPE) + OFF);              \
+    memset(dst_memory_c, 0, kPixels * 4 * sizeof(DTYPE));                   \
+    memset(dst_memory_opt, 0, kPixels * 4 * sizeof(DTYPE));                 \
+    STYPE* src_pixels_r = reinterpret_cast<STYPE*>(src_memory_r + OFF);     \
+    STYPE* src_pixels_g = reinterpret_cast<STYPE*>(src_memory_g + OFF);     \
+    STYPE* src_pixels_b = reinterpret_cast<STYPE*>(src_memory_b + OFF);     \
+    DTYPE* dst_pixels_c = reinterpret_cast<DTYPE*>(dst_memory_c);           \
+    DTYPE* dst_pixels_opt = reinterpret_cast<DTYPE*>(dst_memory_opt);       \
+    MaskCpuFlags(disable_cpu_flags_);                                       \
+    FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b,   \
+                kWidth, NULL, 0, dst_pixels_c, kWidth * 4, kWidth,          \
+                NEG benchmark_height_, DEPTH);                              \
+    MaskCpuFlags(benchmark_cpu_info_);                                      \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                       \
+      FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \
+                  kWidth, NULL, 0, dst_pixels_opt, kWidth * 4, kWidth,      \
+                  NEG benchmark_height_, DEPTH);                            \
+    }                                                                       \
+    for (int i = 0; i < kPixels * 4; ++i) {                                 \
+      EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);                        \
+    }                                                                       \
+    free_aligned_buffer_page_end(src_memory_r);                             \
+    free_aligned_buffer_page_end(src_memory_g);                             \
+    free_aligned_buffer_page_end(src_memory_b);                             \
+    free_aligned_buffer_page_end(dst_memory_c);                             \
+    free_aligned_buffer_page_end(dst_memory_opt);                           \
+  }
+
+#define TESTQPLANARTOP(FUNC, STYPE, DTYPE, DEPTH)                              \
+  TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_ + 1, _Any, +, 0) \
+  TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Unaligned, +,  \
+                  2)                                                           \
+  TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Invert, -, 0)  \
+  TESTQPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Opt, +, 0)     \
+  TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_ + 1, _Any, +,   \
+                   0)                                                          \
+  TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Unaligned, +, \
+                   2)                                                          \
+  TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Invert, -, 0) \
+  TESTQPLANAROTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Opt, +, 0)
+
+TESTQPLANARTOP(MergeAR64, uint16_t, uint16_t, 10)
+TESTQPLANARTOP(MergeAR64, uint16_t, uint16_t, 12)
+TESTQPLANARTOP(MergeAR64, uint16_t, uint16_t, 16)
+TESTQPLANARTOP(MergeARGB16To8, uint16_t, uint8_t, 10)
+TESTQPLANARTOP(MergeARGB16To8, uint16_t, uint8_t, 12)
+TESTQPLANARTOP(MergeARGB16To8, uint16_t, uint8_t, 16)
+
+#define TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, W1280, N, NEG, OFF)      \
+  TEST_F(LibYUVPlanarTest, FUNC##Plane_##DEPTH##N) {                        \
+    const int kWidth = W1280;                                               \
+    const int kPixels = kWidth * benchmark_height_;                         \
+    align_buffer_page_end(src_memory_r, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_g, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(src_memory_b, kPixels * sizeof(STYPE) + OFF);     \
+    align_buffer_page_end(dst_memory_c, kPixels * 4 * sizeof(DTYPE));       \
+    align_buffer_page_end(dst_memory_opt, kPixels * 4 * sizeof(DTYPE));     \
+    MemRandomize(src_memory_r, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_g, kPixels * sizeof(STYPE) + OFF);              \
+    MemRandomize(src_memory_b, kPixels * sizeof(STYPE) + OFF);              \
+    STYPE* src_pixels_r = reinterpret_cast<STYPE*>(src_memory_r + OFF);     \
+    STYPE* src_pixels_g = reinterpret_cast<STYPE*>(src_memory_g + OFF);     \
+    STYPE* src_pixels_b = reinterpret_cast<STYPE*>(src_memory_b + OFF);     \
+    DTYPE* dst_pixels_c = reinterpret_cast<DTYPE*>(dst_memory_c);           \
+    DTYPE* dst_pixels_opt = reinterpret_cast<DTYPE*>(dst_memory_opt);       \
+    memset(dst_pixels_c, 1, kPixels * 4 * sizeof(DTYPE));                   \
+    memset(dst_pixels_opt, 2, kPixels * 4 * sizeof(DTYPE));                 \
+    MaskCpuFlags(disable_cpu_flags_);                                       \
+    FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b,   \
+                kWidth, dst_pixels_c, kWidth * 4, kWidth,                   \
+                NEG benchmark_height_, DEPTH);                              \
+    MaskCpuFlags(benchmark_cpu_info_);                                      \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                       \
+      FUNC##Plane(src_pixels_r, kWidth, src_pixels_g, kWidth, src_pixels_b, \
+                  kWidth, dst_pixels_opt, kWidth * 4, kWidth,               \
+                  NEG benchmark_height_, DEPTH);                            \
+    }                                                                       \
+    for (int i = 0; i < kPixels * 4; ++i) {                                 \
+      EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);                        \
+    }                                                                       \
+    free_aligned_buffer_page_end(src_memory_r);                             \
+    free_aligned_buffer_page_end(src_memory_g);                             \
+    free_aligned_buffer_page_end(src_memory_b);                             \
+    free_aligned_buffer_page_end(dst_memory_c);                             \
+    free_aligned_buffer_page_end(dst_memory_opt);                           \
+  }
+
+#define TESTTPLANARTOP(FUNC, STYPE, DTYPE, DEPTH)                              \
+  TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_ + 1, _Any, +, 0) \
+  TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Unaligned, +,  \
+                  2)                                                           \
+  TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Invert, -, 0)  \
+  TESTTPLANARTOPI(FUNC, STYPE, DTYPE, DEPTH, benchmark_width_, _Opt, +, 0)
+
+TESTTPLANARTOP(MergeXR30, uint16_t, uint8_t, 10)
+TESTTPLANARTOP(MergeXR30, uint16_t, uint8_t, 12)
+TESTTPLANARTOP(MergeXR30, uint16_t, uint8_t, 16)
+
 // TODO(fbarchard): improve test for platforms and cpu detect
 #ifdef HAS_MERGEUVROW_16_AVX2
 TEST_F(LibYUVPlanarTest, MergeUVRow_16_Opt) {
-  const int kPixels = benchmark_width_ * benchmark_height_;
+  // Round count up to multiple of 16
+  const int kPixels = (benchmark_width_ * benchmark_height_ + 15) & ~15;
+
   align_buffer_page_end(src_pixels_u, kPixels * 2);
   align_buffer_page_end(src_pixels_v, kPixels * 2);
   align_buffer_page_end(dst_pixels_uv_opt, kPixels * 2 * 2);
@@ -2679,19 +3510,19 @@ TEST_F(LibYUVPlanarTest, MergeUVRow_16_Opt) {
 
   MergeUVRow_16_C(reinterpret_cast<const uint16_t*>(src_pixels_u),
                   reinterpret_cast<const uint16_t*>(src_pixels_v),
-                  reinterpret_cast<uint16_t*>(dst_pixels_uv_c), 64, kPixels);
+                  reinterpret_cast<uint16_t*>(dst_pixels_uv_c), 16, kPixels);
 
   int has_avx2 = TestCpuFlag(kCpuHasAVX2);
   for (int i = 0; i < benchmark_iterations_; ++i) {
     if (has_avx2) {
       MergeUVRow_16_AVX2(reinterpret_cast<const uint16_t*>(src_pixels_u),
                          reinterpret_cast<const uint16_t*>(src_pixels_v),
-                         reinterpret_cast<uint16_t*>(dst_pixels_uv_opt), 64,
+                         reinterpret_cast<uint16_t*>(dst_pixels_uv_opt), 16,
                          kPixels);
     } else {
       MergeUVRow_16_C(reinterpret_cast<const uint16_t*>(src_pixels_u),
                       reinterpret_cast<const uint16_t*>(src_pixels_v),
-                      reinterpret_cast<uint16_t*>(dst_pixels_uv_opt), 64,
+                      reinterpret_cast<uint16_t*>(dst_pixels_uv_opt), 16,
                       kPixels);
     }
   }
@@ -2710,7 +3541,9 @@ TEST_F(LibYUVPlanarTest, MergeUVRow_16_Opt) {
 // TODO(fbarchard): Improve test for more platforms.
 #ifdef HAS_MULTIPLYROW_16_AVX2
 TEST_F(LibYUVPlanarTest, MultiplyRow_16_Opt) {
-  const int kPixels = benchmark_width_ * benchmark_height_;
+  // Round count up to multiple of 32
+  const int kPixels = (benchmark_width_ * benchmark_height_ + 31) & ~31;
+
   align_buffer_page_end(src_pixels_y, kPixels * 2);
   align_buffer_page_end(dst_pixels_y_opt, kPixels * 2);
   align_buffer_page_end(dst_pixels_y_c, kPixels * 2);
@@ -2776,6 +3609,65 @@ TEST_F(LibYUVPlanarTest, Convert16To8Plane) {
   free_aligned_buffer_page_end(dst_pixels_y_c);
 }
 
+TEST_F(LibYUVPlanarTest, YUY2ToY) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels_y, kPixels * 2);
+  align_buffer_page_end(dst_pixels_y_opt, kPixels);
+  align_buffer_page_end(dst_pixels_y_c, kPixels);
+
+  MemRandomize(src_pixels_y, kPixels * 2);
+  memset(dst_pixels_y_opt, 0, kPixels);
+  memset(dst_pixels_y_c, 1, kPixels);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  YUY2ToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_c, benchmark_width_,
+          benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    YUY2ToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_opt,
+            benchmark_width_, benchmark_width_, benchmark_height_);
+  }
+
+  for (int i = 0; i < kPixels; ++i) {
+    EXPECT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels_y);
+  free_aligned_buffer_page_end(dst_pixels_y_opt);
+  free_aligned_buffer_page_end(dst_pixels_y_c);
+}
+
+TEST_F(LibYUVPlanarTest, UYVYToY) {
+  const int kPixels = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(src_pixels_y, kPixels * 2);
+  align_buffer_page_end(dst_pixels_y_opt, kPixels);
+  align_buffer_page_end(dst_pixels_y_c, kPixels);
+
+  MemRandomize(src_pixels_y, kPixels * 2);
+  memset(dst_pixels_y_opt, 0, kPixels);
+  memset(dst_pixels_y_c, 1, kPixels);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  UYVYToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_c, benchmark_width_,
+          benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    UYVYToY(src_pixels_y, benchmark_width_ * 2, dst_pixels_y_opt,
+            benchmark_width_, benchmark_width_, benchmark_height_);
+  }
+
+  for (int i = 0; i < kPixels; ++i) {
+    EXPECT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels_y);
+  free_aligned_buffer_page_end(dst_pixels_y_opt);
+  free_aligned_buffer_page_end(dst_pixels_y_c);
+}
+
+#ifdef ENABLE_ROW_TESTS
 // TODO(fbarchard): Improve test for more platforms.
 #ifdef HAS_CONVERT16TO8ROW_AVX2
 TEST_F(LibYUVPlanarTest, Convert16To8Row_Opt) {
@@ -2822,6 +3714,36 @@ TEST_F(LibYUVPlanarTest, Convert16To8Row_Opt) {
 }
 #endif  // HAS_CONVERT16TO8ROW_AVX2
 
+#ifdef HAS_UYVYTOYROW_NEON
+TEST_F(LibYUVPlanarTest, UYVYToYRow_Opt) {
+  // NEON does multiple of 16, so round count up
+  const int kPixels = (benchmark_width_ * benchmark_height_ + 15) & ~15;
+  align_buffer_page_end(src_pixels_y, kPixels * 2);
+  align_buffer_page_end(dst_pixels_y_opt, kPixels);
+  align_buffer_page_end(dst_pixels_y_c, kPixels);
+
+  MemRandomize(src_pixels_y, kPixels * 2);
+  memset(dst_pixels_y_opt, 0, kPixels);
+  memset(dst_pixels_y_c, 1, kPixels);
+
+  UYVYToYRow_C(src_pixels_y, dst_pixels_y_c, kPixels);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    UYVYToYRow_NEON(src_pixels_y, dst_pixels_y_opt, kPixels);
+  }
+
+  for (int i = 0; i < kPixels; ++i) {
+    EXPECT_EQ(dst_pixels_y_opt[i], dst_pixels_y_c[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels_y);
+  free_aligned_buffer_page_end(dst_pixels_y_opt);
+  free_aligned_buffer_page_end(dst_pixels_y_c);
+}
+#endif  // HAS_UYVYTOYROW_NEON
+
+#endif  // ENABLE_ROW_TESTS
+
 TEST_F(LibYUVPlanarTest, Convert8To16Plane) {
   const int kPixels = benchmark_width_ * benchmark_height_;
   align_buffer_page_end(src_pixels_y, kPixels);
@@ -2855,6 +3777,7 @@ TEST_F(LibYUVPlanarTest, Convert8To16Plane) {
   free_aligned_buffer_page_end(dst_pixels_y_c);
 }
 
+#ifdef ENABLE_ROW_TESTS
 // TODO(fbarchard): Improve test for more platforms.
 #ifdef HAS_CONVERT8TO16ROW_AVX2
 TEST_F(LibYUVPlanarTest, Convert8To16Row_Opt) {
@@ -3173,33 +4096,33 @@ extern "C" void GaussRow_NEON(const uint32_t* src, uint16_t* dst, int width);
 extern "C" void GaussRow_C(const uint32_t* src, uint16_t* dst, int width);
 
 TEST_F(LibYUVPlanarTest, TestGaussRow_Opt) {
-  SIMD_ALIGNED(uint32_t orig_pixels[640 + 4]);
-  SIMD_ALIGNED(uint16_t dst_pixels_c[640]);
-  SIMD_ALIGNED(uint16_t dst_pixels_opt[640]);
+  SIMD_ALIGNED(uint32_t orig_pixels[1280 + 8]);
+  SIMD_ALIGNED(uint16_t dst_pixels_c[1280]);
+  SIMD_ALIGNED(uint16_t dst_pixels_opt[1280]);
 
   memset(orig_pixels, 0, sizeof(orig_pixels));
   memset(dst_pixels_c, 1, sizeof(dst_pixels_c));
   memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt));
 
-  for (int i = 0; i < 640 + 4; ++i) {
+  for (int i = 0; i < 1280 + 8; ++i) {
     orig_pixels[i] = i * 256;
   }
-  GaussRow_C(&orig_pixels[0], &dst_pixels_c[0], 640);
-  for (int i = 0; i < benchmark_pixels_div1280_ * 2; ++i) {
+  GaussRow_C(&orig_pixels[0], &dst_pixels_c[0], 1280);
+  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
 #if !defined(LIBYUV_DISABLE_NEON) && \
     (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON))
     int has_neon = TestCpuFlag(kCpuHasNEON);
     if (has_neon) {
-      GaussRow_NEON(&orig_pixels[0], &dst_pixels_opt[0], 640);
+      GaussRow_NEON(&orig_pixels[0], &dst_pixels_opt[0], 1280);
     } else {
-      GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 640);
+      GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 1280);
     }
 #else
-    GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 640);
+    GaussRow_C(&orig_pixels[0], &dst_pixels_opt[0], 1280);
 #endif
   }
 
-  for (int i = 0; i < 640; ++i) {
+  for (int i = 0; i < 1280; ++i) {
     EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
   }
 
@@ -3225,141 +4148,139 @@ extern "C" void GaussCol_C(const uint16_t* src0,
                            int width);
 
 TEST_F(LibYUVPlanarTest, TestGaussCol_Opt) {
-  SIMD_ALIGNED(uint16_t orig_pixels[640 * 5]);
-  SIMD_ALIGNED(uint32_t dst_pixels_c[640]);
-  SIMD_ALIGNED(uint32_t dst_pixels_opt[640]);
+  SIMD_ALIGNED(uint16_t orig_pixels[1280 * 5]);
+  SIMD_ALIGNED(uint32_t dst_pixels_c[1280]);
+  SIMD_ALIGNED(uint32_t dst_pixels_opt[1280]);
 
   memset(orig_pixels, 0, sizeof(orig_pixels));
   memset(dst_pixels_c, 1, sizeof(dst_pixels_c));
   memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt));
 
-  for (int i = 0; i < 640 * 5; ++i) {
-    orig_pixels[i] = i;
+  for (int i = 0; i < 1280 * 5; ++i) {
+    orig_pixels[i] = static_cast<float>(i);
   }
-  GaussCol_C(&orig_pixels[0], &orig_pixels[640], &orig_pixels[640 * 2],
-             &orig_pixels[640 * 3], &orig_pixels[640 * 4], &dst_pixels_c[0],
-             640);
-  for (int i = 0; i < benchmark_pixels_div1280_ * 2; ++i) {
+  GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2],
+             &orig_pixels[1280 * 3], &orig_pixels[1280 * 4], &dst_pixels_c[0],
+             1280);
+  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
 #if !defined(LIBYUV_DISABLE_NEON) && \
     (defined(__aarch64__) || defined(__ARM_NEON__) || defined(LIBYUV_NEON))
     int has_neon = TestCpuFlag(kCpuHasNEON);
     if (has_neon) {
-      GaussCol_NEON(&orig_pixels[0], &orig_pixels[640], &orig_pixels[640 * 2],
-                    &orig_pixels[640 * 3], &orig_pixels[640 * 4],
-                    &dst_pixels_opt[0], 640);
+      GaussCol_NEON(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2],
+                    &orig_pixels[1280 * 3], &orig_pixels[1280 * 4],
+                    &dst_pixels_opt[0], 1280);
     } else {
-      GaussCol_C(&orig_pixels[0], &orig_pixels[640], &orig_pixels[640 * 2],
-                 &orig_pixels[640 * 3], &orig_pixels[640 * 4],
-                 &dst_pixels_opt[0], 640);
+      GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2],
+                 &orig_pixels[1280 * 3], &orig_pixels[1280 * 4],
+                 &dst_pixels_opt[0], 1280);
     }
 #else
-    GaussCol_C(&orig_pixels[0], &orig_pixels[640], &orig_pixels[640 * 2],
-               &orig_pixels[640 * 3], &orig_pixels[640 * 4], &dst_pixels_opt[0],
-               640);
+    GaussCol_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2],
+               &orig_pixels[1280 * 3], &orig_pixels[1280 * 4],
+               &dst_pixels_opt[0], 1280);
 #endif
   }
 
-  for (int i = 0; i < 640; ++i) {
+  for (int i = 0; i < 1280; ++i) {
     EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
   }
-
-  EXPECT_EQ(dst_pixels_c[0],
-            static_cast<uint32_t>(0 * 1 + 640 * 4 + 640 * 2 * 6 + 640 * 3 * 4 +
-                                  640 * 4 * 1));
-  EXPECT_EQ(dst_pixels_c[639], static_cast<uint32_t>(30704));
 }
 
-float TestFloatDivToByte(int benchmark_width,
-                         int benchmark_height,
-                         int benchmark_iterations,
-                         float scale,
-                         bool opt) {
-  int i, j;
-  // NEON does multiple of 8, so round count up
-  const int kPixels = (benchmark_width * benchmark_height + 7) & ~7;
-  align_buffer_page_end(src_weights, kPixels * 4);
-  align_buffer_page_end(src_values, kPixels * 4);
-  align_buffer_page_end(dst_out_c, kPixels);
-  align_buffer_page_end(dst_out_opt, kPixels);
-  align_buffer_page_end(dst_mask_c, kPixels);
-  align_buffer_page_end(dst_mask_opt, kPixels);
-
-  // Randomize works but may contain some denormals affecting performance.
-  // MemRandomize(orig_y, kPixels * 4);
-  // large values are problematic.  audio is really -1 to 1.
-  for (i = 0; i < kPixels; ++i) {
-    (reinterpret_cast<float*>(src_weights))[i] = scale;
-    (reinterpret_cast<float*>(src_values))[i] =
-        sinf(static_cast<float>(i) * 0.1f);
-  }
-  memset(dst_out_c, 0, kPixels);
-  memset(dst_out_opt, 1, kPixels);
-  memset(dst_mask_c, 2, kPixels);
-  memset(dst_mask_opt, 3, kPixels);
+TEST_F(LibYUVPlanarTest, TestGaussRow_F32_Opt) {
+  SIMD_ALIGNED(float orig_pixels[1280 + 4]);
+  SIMD_ALIGNED(float dst_pixels_c[1280]);
+  SIMD_ALIGNED(float dst_pixels_opt[1280]);
 
-  FloatDivToByteRow_C(reinterpret_cast<float*>(src_weights),
-                      reinterpret_cast<float*>(src_values), dst_out_c,
-                      dst_mask_c, kPixels);
+  memset(orig_pixels, 0, sizeof(orig_pixels));
+  memset(dst_pixels_c, 1, sizeof(dst_pixels_c));
+  memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt));
 
-  for (j = 0; j < benchmark_iterations; j++) {
-    if (opt) {
-#ifdef HAS_FLOATDIVTOBYTEROW_NEON
-      FloatDivToByteRow_NEON(reinterpret_cast<float*>(src_weights),
-                             reinterpret_cast<float*>(src_values), dst_out_opt,
-                             dst_mask_opt, kPixels);
-#else
-      FloatDivToByteRow_C(reinterpret_cast<float*>(src_weights),
-                          reinterpret_cast<float*>(src_values), dst_out_opt,
-                          dst_mask_opt, kPixels);
-#endif
+  for (int i = 0; i < 1280 + 4; ++i) {
+    orig_pixels[i] = static_cast<float>(i);
+  }
+  GaussRow_F32_C(&orig_pixels[0], &dst_pixels_c[0], 1280);
+  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
+#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
+    int has_neon = TestCpuFlag(kCpuHasNEON);
+    if (has_neon) {
+      GaussRow_F32_NEON(&orig_pixels[0], &dst_pixels_opt[0], 1280);
     } else {
-      FloatDivToByteRow_C(reinterpret_cast<float*>(src_weights),
-                          reinterpret_cast<float*>(src_values), dst_out_opt,
-                          dst_mask_opt, kPixels);
+      GaussRow_F32_C(&orig_pixels[0], &dst_pixels_opt[0], 1280);
     }
+#else
+    GaussRow_F32_C(&orig_pixels[0], &dst_pixels_opt[0], 1280);
+#endif
   }
 
-  uint8_t max_diff = 0;
-  for (i = 0; i < kPixels; ++i) {
-    uint8_t abs_diff = abs(dst_out_c[i] - dst_out_opt[i]) +
-                       abs(dst_mask_c[i] - dst_mask_opt[i]);
-    if (abs_diff > max_diff) {
-      max_diff = abs_diff;
-    }
+  for (int i = 0; i < 1280; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
   }
+}
 
-  free_aligned_buffer_page_end(src_weights);
-  free_aligned_buffer_page_end(src_values);
-  free_aligned_buffer_page_end(dst_out_c);
-  free_aligned_buffer_page_end(dst_out_opt);
-  free_aligned_buffer_page_end(dst_mask_c);
-  free_aligned_buffer_page_end(dst_mask_opt);
+TEST_F(LibYUVPlanarTest, TestGaussCol_F32_Opt) {
+  SIMD_ALIGNED(float dst_pixels_c[1280]);
+  SIMD_ALIGNED(float dst_pixels_opt[1280]);
+  align_buffer_page_end(orig_pixels_buf, 1280 * 5 * 4);  // 5 rows
+  float* orig_pixels = reinterpret_cast<float*>(orig_pixels_buf);
 
-  return max_diff;
-}
+  memset(orig_pixels, 0, 1280 * 5 * 4);
+  memset(dst_pixels_c, 1, sizeof(dst_pixels_c));
+  memset(dst_pixels_opt, 2, sizeof(dst_pixels_opt));
 
-TEST_F(LibYUVPlanarTest, TestFloatDivToByte_C) {
-  float diff = TestFloatDivToByte(benchmark_width_, benchmark_height_,
-                                  benchmark_iterations_, 1.2f, false);
-  EXPECT_EQ(0, diff);
-}
+  for (int i = 0; i < 1280 * 5; ++i) {
+    orig_pixels[i] = static_cast<float>(i);
+  }
+  GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2],
+                 &orig_pixels[1280 * 3], &orig_pixels[1280 * 4],
+                 &dst_pixels_c[0], 1280);
+  for (int i = 0; i < benchmark_pixels_div1280_; ++i) {
+#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
+    int has_neon = TestCpuFlag(kCpuHasNEON);
+    if (has_neon) {
+      GaussCol_F32_NEON(&orig_pixels[0], &orig_pixels[1280],
+                        &orig_pixels[1280 * 2], &orig_pixels[1280 * 3],
+                        &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280);
+    } else {
+      GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280],
+                     &orig_pixels[1280 * 2], &orig_pixels[1280 * 3],
+                     &orig_pixels[1280 * 4], &dst_pixels_opt[0], 1280);
+    }
+#else
+    GaussCol_F32_C(&orig_pixels[0], &orig_pixels[1280], &orig_pixels[1280 * 2],
+                   &orig_pixels[1280 * 3], &orig_pixels[1280 * 4],
+                   &dst_pixels_opt[0], 1280);
+#endif
+  }
 
-TEST_F(LibYUVPlanarTest, TestFloatDivToByte_Opt) {
-  float diff = TestFloatDivToByte(benchmark_width_, benchmark_height_,
-                                  benchmark_iterations_, 1.2f, true);
-  EXPECT_EQ(0, diff);
+  for (int i = 0; i < 1280; ++i) {
+    EXPECT_EQ(dst_pixels_c[i], dst_pixels_opt[i]);
+  }
+  free_aligned_buffer_page_end(orig_pixels_buf);
 }
 
-TEST_F(LibYUVPlanarTest, UVToVURow) {
+TEST_F(LibYUVPlanarTest, SwapUVRow) {
   const int kPixels = benchmark_width_ * benchmark_height_;
+  void (*SwapUVRow)(const uint8_t* src_uv, uint8_t* dst_vu, int width) =
+      SwapUVRow_C;
+
   align_buffer_page_end(src_pixels_vu, kPixels * 2);
   align_buffer_page_end(dst_pixels_uv, kPixels * 2);
-
   MemRandomize(src_pixels_vu, kPixels * 2);
   memset(dst_pixels_uv, 1, kPixels * 2);
 
-  UVToVURow_C(src_pixels_vu, dst_pixels_uv, kPixels);
+#if defined(HAS_SWAPUVROW_NEON)
+  if (TestCpuFlag(kCpuHasNEON)) {
+    SwapUVRow = SwapUVRow_Any_NEON;
+    if (IS_ALIGNED(kPixels, 16)) {
+      SwapUVRow = SwapUVRow_NEON;
+    }
+  }
+#endif
 
+  for (int j = 0; j < benchmark_iterations_; j++) {
+    SwapUVRow(src_pixels_vu, dst_pixels_uv, kPixels);
+  }
   for (int i = 0; i < kPixels; ++i) {
     EXPECT_EQ(dst_pixels_uv[i * 2 + 0], src_pixels_vu[i * 2 + 1]);
     EXPECT_EQ(dst_pixels_uv[i * 2 + 1], src_pixels_vu[i * 2 + 0]);
@@ -3368,5 +4289,144 @@ TEST_F(LibYUVPlanarTest, UVToVURow) {
   free_aligned_buffer_page_end(src_pixels_vu);
   free_aligned_buffer_page_end(dst_pixels_uv);
 }
+#endif  // ENABLE_ROW_TESTS
+
+TEST_F(LibYUVPlanarTest, TestGaussPlane_F32) {
+  const int kSize = benchmark_width_ * benchmark_height_ * 4;
+  align_buffer_page_end(orig_pixels, kSize);
+  align_buffer_page_end(dst_pixels_opt, kSize);
+  align_buffer_page_end(dst_pixels_c, kSize);
+
+  for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) {
+    ((float*)(orig_pixels))[i] = (i & 1023) * 3.14f;
+  }
+  memset(dst_pixels_opt, 1, kSize);
+  memset(dst_pixels_c, 2, kSize);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  GaussPlane_F32((const float*)(orig_pixels), benchmark_width_,
+                 (float*)(dst_pixels_c), benchmark_width_, benchmark_width_,
+                 benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    GaussPlane_F32((const float*)(orig_pixels), benchmark_width_,
+                   (float*)(dst_pixels_opt), benchmark_width_, benchmark_width_,
+                   benchmark_height_);
+  }
+  for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) {
+    EXPECT_NEAR(((float*)(dst_pixels_c))[i], ((float*)(dst_pixels_opt))[i], 1.f)
+        << i;
+  }
+
+  free_aligned_buffer_page_end(dst_pixels_c);
+  free_aligned_buffer_page_end(dst_pixels_opt);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVPlanarTest, HalfMergeUVPlane_Opt) {
+  int dst_width = (benchmark_width_ + 1) / 2;
+  int dst_height = (benchmark_height_ + 1) / 2;
+  align_buffer_page_end(src_pixels_u, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(src_pixels_v, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(tmp_pixels_u, dst_width * dst_height);
+  align_buffer_page_end(tmp_pixels_v, dst_width * dst_height);
+  align_buffer_page_end(dst_pixels_uv_opt, dst_width * 2 * dst_height);
+  align_buffer_page_end(dst_pixels_uv_c, dst_width * 2 * dst_height);
+
+  MemRandomize(src_pixels_u, benchmark_width_ * benchmark_height_);
+  MemRandomize(src_pixels_v, benchmark_width_ * benchmark_height_);
+  MemRandomize(tmp_pixels_u, dst_width * dst_height);
+  MemRandomize(tmp_pixels_v, dst_width * dst_height);
+  MemRandomize(dst_pixels_uv_opt, dst_width * 2 * dst_height);
+  MemRandomize(dst_pixels_uv_c, dst_width * 2 * dst_height);
+
+  MaskCpuFlags(disable_cpu_flags_);
+  HalfMergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v,
+                   benchmark_width_, dst_pixels_uv_c, dst_width * 2,
+                   benchmark_width_, benchmark_height_);
+  MaskCpuFlags(benchmark_cpu_info_);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    HalfMergeUVPlane(src_pixels_u, benchmark_width_, src_pixels_v,
+                     benchmark_width_, dst_pixels_uv_opt, dst_width * 2,
+                     benchmark_width_, benchmark_height_);
+  }
+
+  for (int i = 0; i < dst_width * 2 * dst_height; ++i) {
+    EXPECT_EQ(dst_pixels_uv_c[i], dst_pixels_uv_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(src_pixels_u);
+  free_aligned_buffer_page_end(src_pixels_v);
+  free_aligned_buffer_page_end(tmp_pixels_u);
+  free_aligned_buffer_page_end(tmp_pixels_v);
+  free_aligned_buffer_page_end(dst_pixels_uv_opt);
+  free_aligned_buffer_page_end(dst_pixels_uv_c);
+}
+
+TEST_F(LibYUVPlanarTest, NV12Copy) {
+  const int halfwidth = (benchmark_width_ + 1) >> 1;
+  const int halfheight = (benchmark_height_ + 1) >> 1;
+  align_buffer_page_end(src_y, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(src_uv, halfwidth * 2 * halfheight);
+  align_buffer_page_end(dst_y, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(dst_uv, halfwidth * 2 * halfheight);
+
+  MemRandomize(src_y, benchmark_width_ * benchmark_height_);
+  MemRandomize(src_uv, halfwidth * 2 * halfheight);
+  MemRandomize(dst_y, benchmark_width_ * benchmark_height_);
+  MemRandomize(dst_uv, halfwidth * 2 * halfheight);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    NV12Copy(src_y, benchmark_width_, src_uv, halfwidth * 2, dst_y,
+             benchmark_width_, dst_uv, halfwidth * 2, benchmark_width_,
+             benchmark_height_);
+  }
+
+  for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) {
+    EXPECT_EQ(src_y[i], dst_y[i]);
+  }
+  for (int i = 0; i < halfwidth * 2 * halfheight; ++i) {
+    EXPECT_EQ(src_uv[i], dst_uv[i]);
+  }
+
+  free_aligned_buffer_page_end(src_y);
+  free_aligned_buffer_page_end(src_uv);
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_uv);
+}
+
+TEST_F(LibYUVPlanarTest, NV21Copy) {
+  const int halfwidth = (benchmark_width_ + 1) >> 1;
+  const int halfheight = (benchmark_height_ + 1) >> 1;
+  align_buffer_page_end(src_y, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(src_vu, halfwidth * 2 * halfheight);
+  align_buffer_page_end(dst_y, benchmark_width_ * benchmark_height_);
+  align_buffer_page_end(dst_vu, halfwidth * 2 * halfheight);
+
+  MemRandomize(src_y, benchmark_width_ * benchmark_height_);
+  MemRandomize(src_vu, halfwidth * 2 * halfheight);
+  MemRandomize(dst_y, benchmark_width_ * benchmark_height_);
+  MemRandomize(dst_vu, halfwidth * 2 * halfheight);
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    NV21Copy(src_y, benchmark_width_, src_vu, halfwidth * 2, dst_y,
+             benchmark_width_, dst_vu, halfwidth * 2, benchmark_width_,
+             benchmark_height_);
+  }
+
+  for (int i = 0; i < benchmark_width_ * benchmark_height_; ++i) {
+    EXPECT_EQ(src_y[i], dst_y[i]);
+  }
+  for (int i = 0; i < halfwidth * 2 * halfheight; ++i) {
+    EXPECT_EQ(src_vu[i], dst_vu[i]);
+  }
+
+  free_aligned_buffer_page_end(src_y);
+  free_aligned_buffer_page_end(src_vu);
+  free_aligned_buffer_page_end(dst_y);
+  free_aligned_buffer_page_end(dst_vu);
+}
 
 }  // namespace libyuv
diff --git a/files/unit_test/rotate_argb_test.cc b/files/unit_test/rotate_argb_test.cc
index d2003895..01ed69ca 100644
--- a/files/unit_test/rotate_argb_test.cc
+++ b/files/unit_test/rotate_argb_test.cc
@@ -156,31 +156,73 @@ TEST_F(LibYUVRotateTest, RotatePlane270_Opt) {
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_RotatePlane0_Odd) {
-  TestRotatePlane(benchmark_width_ - 3, benchmark_height_ - 1,
-                  benchmark_width_ - 3, benchmark_height_ - 1, kRotate0,
+  TestRotatePlane(benchmark_width_ + 1, benchmark_height_ + 1,
+                  benchmark_width_ + 1, benchmark_height_ + 1, kRotate0,
                   benchmark_iterations_, disable_cpu_flags_,
                   benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_RotatePlane90_Odd) {
-  TestRotatePlane(benchmark_width_ - 3, benchmark_height_ - 1,
-                  benchmark_height_ - 1, benchmark_width_ - 3, kRotate90,
+  TestRotatePlane(benchmark_width_ + 1, benchmark_height_ + 1,
+                  benchmark_height_ + 1, benchmark_width_ + 1, kRotate90,
                   benchmark_iterations_, disable_cpu_flags_,
                   benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_RotatePlane180_Odd) {
-  TestRotatePlane(benchmark_width_ - 3, benchmark_height_ - 1,
-                  benchmark_width_ - 3, benchmark_height_ - 1, kRotate180,
+  TestRotatePlane(benchmark_width_ + 1, benchmark_height_ + 1,
+                  benchmark_width_ + 1, benchmark_height_ + 1, kRotate180,
                   benchmark_iterations_, disable_cpu_flags_,
                   benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_RotatePlane270_Odd) {
-  TestRotatePlane(benchmark_width_ - 3, benchmark_height_ - 1,
-                  benchmark_height_ - 1, benchmark_width_ - 3, kRotate270,
+  TestRotatePlane(benchmark_width_ + 1, benchmark_height_ + 1,
+                  benchmark_height_ + 1, benchmark_width_ + 1, kRotate270,
                   benchmark_iterations_, disable_cpu_flags_,
                   benchmark_cpu_info_);
 }
 
+TEST_F(LibYUVRotateTest, RotatePlane90_TestStride) {
+  int argb_plane_size = benchmark_width_ * 4 * abs(benchmark_height_);
+
+  align_buffer_page_end(src_argb, argb_plane_size);
+  align_buffer_page_end(dst_argb, argb_plane_size);
+
+  EXPECT_EQ(0, ARGBRotate(src_argb, benchmark_width_ * 4, dst_argb,
+                          benchmark_width_ * 4, benchmark_width_,
+                          benchmark_height_, kRotate0));
+
+  EXPECT_EQ(0, ARGBRotate(src_argb, benchmark_width_ * 4 - 1, dst_argb,
+                          benchmark_width_ * 4 - 1, benchmark_width_ - 1,
+                          benchmark_height_, kRotate0));
+
+  EXPECT_EQ(0, ARGBRotate(src_argb, benchmark_width_ * 4, dst_argb,
+                          benchmark_width_ * 4, benchmark_width_,
+                          benchmark_height_, kRotate180));
+
+  EXPECT_EQ(0, ARGBRotate(src_argb, benchmark_width_ * 4 - 1, dst_argb,
+                          benchmark_width_ * 4 - 1, benchmark_width_ - 1,
+                          benchmark_height_, kRotate180));
+
+  EXPECT_EQ(0, ARGBRotate(src_argb, benchmark_width_ * 4, dst_argb,
+                          abs(benchmark_height_) * 4, benchmark_width_,
+                          benchmark_height_, kRotate90));
+
+  EXPECT_EQ(-1, ARGBRotate(src_argb, benchmark_width_ * 4 - 1, dst_argb,
+                           abs(benchmark_height_) * 4, benchmark_width_ - 1,
+                           benchmark_height_, kRotate90));
+
+  EXPECT_EQ(0, ARGBRotate(src_argb, benchmark_width_ * 4, dst_argb,
+                          abs(benchmark_height_) * 4, benchmark_width_,
+                          benchmark_height_, kRotate270));
+
+  EXPECT_EQ(-1, ARGBRotate(src_argb, benchmark_width_ * 4 - 1, dst_argb,
+                           abs(benchmark_height_) * 4, benchmark_width_ - 1,
+                           benchmark_height_, kRotate270));
+
+  free_aligned_buffer_page_end(dst_argb);
+  free_aligned_buffer_page_end(src_argb);
+}
+
 }  // namespace libyuv
diff --git a/files/unit_test/rotate_test.cc b/files/unit_test/rotate_test.cc
index 61941e63..d3887414 100644
--- a/files/unit_test/rotate_test.cc
+++ b/files/unit_test/rotate_test.cc
@@ -16,6 +16,8 @@
 
 namespace libyuv {
 
+#define SUBSAMPLE(v, a) ((((v) + (a)-1)) / (a))
+
 static void I420TestRotate(int src_width,
                            int src_height,
                            int dst_width,
@@ -108,33 +110,121 @@ TEST_F(LibYUVRotateTest, I420Rotate270_Opt) {
 // Odd width tests work but disabled because they use C code and can be
 // tested by passing an odd width command line or environment variable.
 TEST_F(LibYUVRotateTest, DISABLED_I420Rotate0_Odd) {
-  I420TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_width_ - 3, benchmark_height_ - 1, kRotate0,
+  I420TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_width_ + 1, benchmark_height_ + 1, kRotate0,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_I420Rotate90_Odd) {
-  I420TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_height_ - 1, benchmark_width_ - 3, kRotate90,
+  I420TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_height_ + 1, benchmark_width_ + 1, kRotate90,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_I420Rotate180_Odd) {
-  I420TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_width_ - 3, benchmark_height_ - 1, kRotate180,
+  I420TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_width_ + 1, benchmark_height_ + 1, kRotate180,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_I420Rotate270_Odd) {
-  I420TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_height_ - 1, benchmark_width_ - 3, kRotate270,
+  I420TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_height_ + 1, benchmark_width_ + 1, kRotate270,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
+static void I422TestRotate(int src_width,
+                           int src_height,
+                           int dst_width,
+                           int dst_height,
+                           libyuv::RotationMode mode,
+                           int benchmark_iterations,
+                           int disable_cpu_flags,
+                           int benchmark_cpu_info) {
+  if (src_width < 1) {
+    src_width = 1;
+  }
+  if (src_height == 0) {
+    src_height = 1;
+  }
+  if (dst_width < 1) {
+    dst_width = 1;
+  }
+  if (dst_height < 1) {
+    dst_height = 1;
+  }
+  int src_i422_y_size = src_width * Abs(src_height);
+  int src_i422_uv_size = ((src_width + 1) / 2) * Abs(src_height);
+  int src_i422_size = src_i422_y_size + src_i422_uv_size * 2;
+  align_buffer_page_end(src_i422, src_i422_size);
+  for (int i = 0; i < src_i422_size; ++i) {
+    src_i422[i] = fastrand() & 0xff;
+  }
+
+  int dst_i422_y_size = dst_width * dst_height;
+  int dst_i422_uv_size = ((dst_width + 1) / 2) * dst_height;
+  int dst_i422_size = dst_i422_y_size + dst_i422_uv_size * 2;
+  align_buffer_page_end(dst_i422_c, dst_i422_size);
+  align_buffer_page_end(dst_i422_opt, dst_i422_size);
+  memset(dst_i422_c, 2, dst_i422_size);
+  memset(dst_i422_opt, 3, dst_i422_size);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  I422Rotate(src_i422, src_width, src_i422 + src_i422_y_size,
+             (src_width + 1) / 2, src_i422 + src_i422_y_size + src_i422_uv_size,
+             (src_width + 1) / 2, dst_i422_c, dst_width,
+             dst_i422_c + dst_i422_y_size, (dst_width + 1) / 2,
+             dst_i422_c + dst_i422_y_size + dst_i422_uv_size,
+             (dst_width + 1) / 2, src_width, src_height, mode);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (int i = 0; i < benchmark_iterations; ++i) {
+    I422Rotate(
+        src_i422, src_width, src_i422 + src_i422_y_size, (src_width + 1) / 2,
+        src_i422 + src_i422_y_size + src_i422_uv_size, (src_width + 1) / 2,
+        dst_i422_opt, dst_width, dst_i422_opt + dst_i422_y_size,
+        (dst_width + 1) / 2, dst_i422_opt + dst_i422_y_size + dst_i422_uv_size,
+        (dst_width + 1) / 2, src_width, src_height, mode);
+  }
+
+  // Rotation should be exact.
+  for (int i = 0; i < dst_i422_size; ++i) {
+    EXPECT_EQ(dst_i422_c[i], dst_i422_opt[i]);
+  }
+
+  free_aligned_buffer_page_end(dst_i422_c);
+  free_aligned_buffer_page_end(dst_i422_opt);
+  free_aligned_buffer_page_end(src_i422);
+}
+
+TEST_F(LibYUVRotateTest, I422Rotate0_Opt) {
+  I422TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate0, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I422Rotate90_Opt) {
+  I422TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate90, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I422Rotate180_Opt) {
+  I422TestRotate(benchmark_width_, benchmark_height_, benchmark_width_,
+                 benchmark_height_, kRotate180, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
+TEST_F(LibYUVRotateTest, I422Rotate270_Opt) {
+  I422TestRotate(benchmark_width_, benchmark_height_, benchmark_height_,
+                 benchmark_width_, kRotate270, benchmark_iterations_,
+                 disable_cpu_flags_, benchmark_cpu_info_);
+}
+
 static void I444TestRotate(int src_width,
                            int src_height,
                            int dst_width,
@@ -225,29 +315,29 @@ TEST_F(LibYUVRotateTest, I444Rotate270_Opt) {
 // Odd width tests work but disabled because they use C code and can be
 // tested by passing an odd width command line or environment variable.
 TEST_F(LibYUVRotateTest, DISABLED_I444Rotate0_Odd) {
-  I444TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_width_ - 3, benchmark_height_ - 1, kRotate0,
+  I444TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_width_ + 1, benchmark_height_ + 1, kRotate0,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_I444Rotate90_Odd) {
-  I444TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_height_ - 1, benchmark_width_ - 3, kRotate90,
+  I444TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_height_ + 1, benchmark_width_ + 1, kRotate90,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_I444Rotate180_Odd) {
-  I444TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_width_ - 3, benchmark_height_ - 1, kRotate180,
+  I444TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_width_ + 1, benchmark_height_ + 1, kRotate180,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_I444Rotate270_Odd) {
-  I444TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_height_ - 1, benchmark_width_ - 3, kRotate270,
+  I444TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_height_ + 1, benchmark_width_ + 1, kRotate270,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
@@ -340,29 +430,29 @@ TEST_F(LibYUVRotateTest, NV12Rotate270_Opt) {
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_NV12Rotate0_Odd) {
-  NV12TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_width_ - 3, benchmark_height_ - 1, kRotate0,
+  NV12TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_width_ + 1, benchmark_height_ + 1, kRotate0,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_NV12Rotate90_Odd) {
-  NV12TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_height_ - 1, benchmark_width_ - 3, kRotate90,
+  NV12TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_height_ + 1, benchmark_width_ + 1, kRotate90,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_NV12Rotate180_Odd) {
-  NV12TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_width_ - 3, benchmark_height_ - 1, kRotate180,
+  NV12TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_width_ + 1, benchmark_height_ + 1, kRotate180,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
 
 TEST_F(LibYUVRotateTest, DISABLED_NV12Rotate270_Odd) {
-  NV12TestRotate(benchmark_width_ - 3, benchmark_height_ - 1,
-                 benchmark_height_ - 1, benchmark_width_ - 3, kRotate270,
+  NV12TestRotate(benchmark_width_ + 1, benchmark_height_ + 1,
+                 benchmark_height_ + 1, benchmark_width_ + 1, kRotate270,
                  benchmark_iterations_, disable_cpu_flags_,
                  benchmark_cpu_info_);
 }
@@ -391,4 +481,119 @@ TEST_F(LibYUVRotateTest, NV12Rotate270_Invert) {
                  disable_cpu_flags_, benchmark_cpu_info_);
 }
 
+// Test Android 420 to I420 Rotate
+#define TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X,          \
+                        SRC_SUBSAMP_Y, FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y,      \
+                        W1280, N, NEG, OFF, PN, OFF_U, OFF_V, ROT)            \
+  TEST_F(LibYUVRotateTest,                                                    \
+         SRC_FMT_PLANAR##To##FMT_PLANAR##Rotate##ROT##To##PN##N) {            \
+    const int kWidth = W1280;                                                 \
+    const int kHeight = benchmark_height_;                                    \
+    const int kSizeUV =                                                       \
+        SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); \
+    align_buffer_page_end(src_y, kWidth* kHeight + OFF);                      \
+    align_buffer_page_end(src_uv,                                             \
+                          kSizeUV*((PIXEL_STRIDE == 3) ? 3 : 2) + OFF);       \
+    align_buffer_page_end(dst_y_c, kWidth* kHeight);                          \
+    align_buffer_page_end(dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X) *             \
+                                       SUBSAMPLE(kHeight, SUBSAMP_Y));        \
+    align_buffer_page_end(dst_v_c, SUBSAMPLE(kWidth, SUBSAMP_X) *             \
+                                       SUBSAMPLE(kHeight, SUBSAMP_Y));        \
+    align_buffer_page_end(dst_y_opt, kWidth* kHeight);                        \
+    align_buffer_page_end(dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X) *           \
+                                         SUBSAMPLE(kHeight, SUBSAMP_Y));      \
+    align_buffer_page_end(dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X) *           \
+                                         SUBSAMPLE(kHeight, SUBSAMP_Y));      \
+    uint8_t* src_u = src_uv + OFF_U;                                          \
+    uint8_t* src_v = src_uv + (PIXEL_STRIDE == 1 ? kSizeUV : OFF_V);          \
+    int src_stride_uv = SUBSAMPLE(kWidth, SUBSAMP_X) * PIXEL_STRIDE;          \
+    for (int i = 0; i < kHeight; ++i)                                         \
+      for (int j = 0; j < kWidth; ++j)                                        \
+        src_y[i * kWidth + j + OFF] = (fastrand() & 0xff);                    \
+    for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) {             \
+      for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) {            \
+        src_u[(i * src_stride_uv) + j * PIXEL_STRIDE + OFF] =                 \
+            (fastrand() & 0xff);                                              \
+        src_v[(i * src_stride_uv) + j * PIXEL_STRIDE + OFF] =                 \
+            (fastrand() & 0xff);                                              \
+      }                                                                       \
+    }                                                                         \
+    memset(dst_y_c, 1, kWidth* kHeight);                                      \
+    memset(dst_u_c, 2,                                                        \
+           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));     \
+    memset(dst_v_c, 3,                                                        \
+           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));     \
+    memset(dst_y_opt, 101, kWidth* kHeight);                                  \
+    memset(dst_u_opt, 102,                                                    \
+           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));     \
+    memset(dst_v_opt, 103,                                                    \
+           SUBSAMPLE(kWidth, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y));     \
+    MaskCpuFlags(disable_cpu_flags_);                                         \
+    SRC_FMT_PLANAR##To##FMT_PLANAR##Rotate(                                   \
+        src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X),   \
+        src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), PIXEL_STRIDE, dst_y_c, \
+        kWidth, dst_u_c, SUBSAMPLE(kWidth, SUBSAMP_X), dst_v_c,               \
+        SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight,                    \
+        (libyuv::RotationMode)ROT);                                           \
+    MaskCpuFlags(benchmark_cpu_info_);                                        \
+    for (int i = 0; i < benchmark_iterations_; ++i) {                         \
+      SRC_FMT_PLANAR##To##FMT_PLANAR##Rotate(                                 \
+          src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \
+          src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), PIXEL_STRIDE,        \
+          dst_y_opt, kWidth, dst_u_opt, SUBSAMPLE(kWidth, SUBSAMP_X),         \
+          dst_v_opt, SUBSAMPLE(kWidth, SUBSAMP_X), kWidth, NEG kHeight,       \
+          (libyuv::RotationMode)ROT);                                         \
+    }                                                                         \
+    for (int i = 0; i < kHeight; ++i) {                                       \
+      for (int j = 0; j < kWidth; ++j) {                                      \
+        EXPECT_EQ(dst_y_c[i * kWidth + j], dst_y_opt[i * kWidth + j]);        \
+      }                                                                       \
+    }                                                                         \
+    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                 \
+      for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) {                \
+        EXPECT_EQ(dst_u_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j],              \
+                  dst_u_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]);           \
+      }                                                                       \
+    }                                                                         \
+    for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) {                 \
+      for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X); ++j) {                \
+        EXPECT_EQ(dst_v_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j],              \
+                  dst_v_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) + j]);           \
+      }                                                                       \
+    }                                                                         \
+    free_aligned_buffer_page_end(dst_y_c);                                    \
+    free_aligned_buffer_page_end(dst_u_c);                                    \
+    free_aligned_buffer_page_end(dst_v_c);                                    \
+    free_aligned_buffer_page_end(dst_y_opt);                                  \
+    free_aligned_buffer_page_end(dst_u_opt);                                  \
+    free_aligned_buffer_page_end(dst_v_opt);                                  \
+    free_aligned_buffer_page_end(src_y);                                      \
+    free_aligned_buffer_page_end(src_uv);                                     \
+  }
+
+#define TESTAPLANARTOP(SRC_FMT_PLANAR, PN, PIXEL_STRIDE, OFF_U, OFF_V,         \
+                       SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, SUBSAMP_X,    \
+                       SUBSAMP_Y)                                              \
+  TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
+                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_ + 1,      \
+                  _Any, +, 0, PN, OFF_U, OFF_V, 0)                             \
+  TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
+                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_,          \
+                  _Unaligned, +, 2, PN, OFF_U, OFF_V, 0)                       \
+  TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
+                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Invert, \
+                  -, 0, PN, OFF_U, OFF_V, 0)                                   \
+  TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
+                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, \
+                  0, PN, OFF_U, OFF_V, 0)                                      \
+  TESTAPLANARTOPI(SRC_FMT_PLANAR, PIXEL_STRIDE, SRC_SUBSAMP_X, SRC_SUBSAMP_Y,  \
+                  FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, \
+                  0, PN, OFF_U, OFF_V, 180)
+
+TESTAPLANARTOP(Android420, I420, 1, 0, 0, 2, 2, I420, 2, 2)
+TESTAPLANARTOP(Android420, NV12, 2, 0, 1, 2, 2, I420, 2, 2)
+TESTAPLANARTOP(Android420, NV21, 2, 1, 0, 2, 2, I420, 2, 2)
+#undef TESTAPLANARTOP
+#undef TESTAPLANARTOPI
+
 }  // namespace libyuv
diff --git a/files/unit_test/scale_argb_test.cc b/files/unit_test/scale_argb_test.cc
index 94aef60e..f54a68f1 100644
--- a/files/unit_test/scale_argb_test.cc
+++ b/files/unit_test/scale_argb_test.cc
@@ -22,6 +22,12 @@ namespace libyuv {
 #define STRINGIZE(line) #line
 #define FILELINESTR(file, line) file ":" STRINGIZE(line)
 
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+// SLOW TESTS are those that are unoptimized C code.
+// FULL TESTS are optimized but test many variations of the same code.
+#define ENABLE_FULL_TESTS
+#endif
+
 // Test scaling with C vs Opt and return maximum pixel difference. 0 = exact.
 static int ARGBTestFilter(int src_width,
                           int src_height,
@@ -114,8 +120,8 @@ static int ARGBTestFilter(int src_width,
   return max_diff;
 }
 
-static const int kTileX = 8;
-static const int kTileY = 8;
+static const int kTileX = 64;
+static const int kTileY = 64;
 
 static int TileARGBScale(const uint8_t* src_argb,
                          int src_stride_argb,
@@ -232,7 +238,7 @@ static int ARGBClipTestFilter(int src_width,
 #define DX(x, nom, denom) static_cast<int>((Abs(x) / nom) * nom)
 #define SX(x, nom, denom) static_cast<int>((x / nom) * denom)
 
-#define TEST_FACTOR1(name, filter, nom, denom, max_diff)                     \
+#define TEST_FACTOR1(DISABLED_, name, filter, nom, denom, max_diff)          \
   TEST_F(LibYUVScaleTest, ARGBScaleDownBy##name##_##filter) {                \
     int diff = ARGBTestFilter(                                               \
         SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
@@ -241,7 +247,7 @@ static int ARGBClipTestFilter(int src_width,
         benchmark_cpu_info_);                                                \
     EXPECT_LE(diff, max_diff);                                               \
   }                                                                          \
-  TEST_F(LibYUVScaleTest, ARGBScaleDownClipBy##name##_##filter) {            \
+  TEST_F(LibYUVScaleTest, DISABLED_##ARGBScaleDownClipBy##name##_##filter) { \
     int diff = ARGBClipTestFilter(                                           \
         SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
         DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
@@ -251,15 +257,30 @@ static int ARGBClipTestFilter(int src_width,
 
 // Test a scale factor with all 4 filters.  Expect unfiltered to be exact, but
 // filtering is different fixed point implementations for SSSE3, Neon and C.
-#define TEST_FACTOR(name, nom, denom)         \
-  TEST_FACTOR1(name, None, nom, denom, 0)     \
-  TEST_FACTOR1(name, Linear, nom, denom, 3)   \
-  TEST_FACTOR1(name, Bilinear, nom, denom, 3) \
-  TEST_FACTOR1(name, Box, nom, denom, 3)
+#ifndef DISABLE_SLOW_TESTS
+#define TEST_FACTOR(name, nom, denom)           \
+  TEST_FACTOR1(, name, None, nom, denom, 0)     \
+  TEST_FACTOR1(, name, Linear, nom, denom, 3)   \
+  TEST_FACTOR1(, name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(, name, Box, nom, denom, 3)
+#else
+#if defined(ENABLE_FULL_TESTS)
+#define TEST_FACTOR(name, nom, denom)                    \
+  TEST_FACTOR1(DISABLED_, name, None, nom, denom, 0)     \
+  TEST_FACTOR1(DISABLED_, name, Linear, nom, denom, 3)   \
+  TEST_FACTOR1(DISABLED_, name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(DISABLED_, name, Box, nom, denom, 3)
+#else
+#define TEST_FACTOR(name, nom, denom) \
+  TEST_FACTOR1(DISABLED_, name, Bilinear, nom, denom, 3)
+#endif
+#endif
 
 TEST_FACTOR(2, 1, 2)
 TEST_FACTOR(4, 1, 4)
+#ifndef DISABLE_SLOW_TESTS
 TEST_FACTOR(8, 1, 8)
+#endif
 TEST_FACTOR(3by4, 3, 4)
 TEST_FACTOR(3by8, 3, 8)
 TEST_FACTOR(3, 1, 3)
@@ -268,7 +289,7 @@ TEST_FACTOR(3, 1, 3)
 #undef SX
 #undef DX
 
-#define TEST_SCALETO1(name, width, height, filter, max_diff)                   \
+#define TEST_SCALETO1(DISABLED_, name, width, height, filter, max_diff)        \
   TEST_F(LibYUVScaleTest, name##To##width##x##height##_##filter) {             \
     int diff = ARGBTestFilter(benchmark_width_, benchmark_height_, width,      \
                               height, kFilter##filter, benchmark_iterations_,  \
@@ -282,34 +303,70 @@ TEST_FACTOR(3, 1, 3)
                               benchmark_cpu_info_);                            \
     EXPECT_LE(diff, max_diff);                                                 \
   }                                                                            \
-  TEST_F(LibYUVScaleTest, name##ClipTo##width##x##height##_##filter) {         \
+  TEST_F(LibYUVScaleTest,                                                      \
+         DISABLED_##name##ClipTo##width##x##height##_##filter) {               \
     int diff =                                                                 \
         ARGBClipTestFilter(benchmark_width_, benchmark_height_, width, height, \
                            kFilter##filter, benchmark_iterations_);            \
     EXPECT_LE(diff, max_diff);                                                 \
   }                                                                            \
-  TEST_F(LibYUVScaleTest, name##ClipFrom##width##x##height##_##filter) {       \
+  TEST_F(LibYUVScaleTest,                                                      \
+         DISABLED_##name##ClipFrom##width##x##height##_##filter) {             \
     int diff = ARGBClipTestFilter(width, height, Abs(benchmark_width_),        \
                                   Abs(benchmark_height_), kFilter##filter,     \
                                   benchmark_iterations_);                      \
     EXPECT_LE(diff, max_diff);                                                 \
   }
 
-/// Test scale to a specified size with all 4 filters.
-#define TEST_SCALETO(name, width, height)       \
-  TEST_SCALETO1(name, width, height, None, 0)   \
-  TEST_SCALETO1(name, width, height, Linear, 3) \
-  TEST_SCALETO1(name, width, height, Bilinear, 3)
+#ifndef DISABLE_SLOW_TESTS
+// Test scale to a specified size with all 4 filters.
+#define TEST_SCALETO(name, width, height)         \
+  TEST_SCALETO1(, name, width, height, None, 0)   \
+  TEST_SCALETO1(, name, width, height, Linear, 3) \
+  TEST_SCALETO1(, name, width, height, Bilinear, 3)
+#else
+#if defined(ENABLE_FULL_TESTS)
+#define TEST_SCALETO(name, width, height)                  \
+  TEST_SCALETO1(DISABLED_, name, width, height, None, 0)   \
+  TEST_SCALETO1(DISABLED_, name, width, height, Linear, 3) \
+  TEST_SCALETO1(DISABLED_, name, width, height, Bilinear, 3)
+#else
+#define TEST_SCALETO(name, width, height) \
+  TEST_SCALETO1(DISABLED_, name, width, height, Bilinear, 3)
+#endif
+#endif
 
 TEST_SCALETO(ARGBScale, 1, 1)
-TEST_SCALETO(ARGBScale, 320, 240)
 TEST_SCALETO(ARGBScale, 569, 480)
 TEST_SCALETO(ARGBScale, 640, 360)
+#ifndef DISABLE_SLOW_TESTS
+TEST_SCALETO(ARGBScale, 256, 144) /* 128x72 * 2 */
+TEST_SCALETO(ARGBScale, 320, 240)
 TEST_SCALETO(ARGBScale, 1280, 720)
 TEST_SCALETO(ARGBScale, 1920, 1080)
+#endif  // DISABLE_SLOW_TESTS
 #undef TEST_SCALETO1
 #undef TEST_SCALETO
 
+#define TEST_SCALESWAPXY1(name, filter, max_diff)                       \
+  TEST_F(LibYUVScaleTest, name##SwapXY_##filter) {                      \
+    int diff = ARGBTestFilter(benchmark_width_, benchmark_height_,      \
+                              benchmark_height_, benchmark_width_,      \
+                              kFilter##filter, benchmark_iterations_,   \
+                              disable_cpu_flags_, benchmark_cpu_info_); \
+    EXPECT_LE(diff, max_diff);                                          \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+// Test scale with swapped width and height with all 3 filters.
+TEST_SCALESWAPXY1(ARGBScale, None, 0)
+TEST_SCALESWAPXY1(ARGBScale, Linear, 0)
+TEST_SCALESWAPXY1(ARGBScale, Bilinear, 0)
+#else
+TEST_SCALESWAPXY1(ARGBScale, Bilinear, 0)
+#endif
+#undef TEST_SCALESWAPXY1
+
 // Scale with YUV conversion to ARGB and clipping.
 // TODO(fbarchard): Add fourcc support.  All 4 ARGB formats is easy to support.
 LIBYUV_API
@@ -454,4 +511,78 @@ TEST_F(LibYUVScaleTest, YUVToRGBScaleDown) {
   EXPECT_LE(diff, 10);
 }
 
+TEST_F(LibYUVScaleTest, ARGBTest3x) {
+  const int kSrcStride = 480 * 4;
+  const int kDstStride = 160 * 4;
+  const int kSize = kSrcStride * 3;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 480 * 3; ++i) {
+    orig_pixels[i * 4 + 0] = i;
+    orig_pixels[i * 4 + 1] = 255 - i;
+    orig_pixels[i * 4 + 2] = i + 1;
+    orig_pixels[i * 4 + 3] = i + 10;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    ARGBScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+              kFilterBilinear);
+  }
+
+  EXPECT_EQ(225, dest_pixels[0]);
+  EXPECT_EQ(255 - 225, dest_pixels[1]);
+  EXPECT_EQ(226, dest_pixels[2]);
+  EXPECT_EQ(235, dest_pixels[3]);
+
+  ARGBScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+            kFilterNone);
+
+  EXPECT_EQ(225, dest_pixels[0]);
+  EXPECT_EQ(255 - 225, dest_pixels[1]);
+  EXPECT_EQ(226, dest_pixels[2]);
+  EXPECT_EQ(235, dest_pixels[3]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, ARGBTest4x) {
+  const int kSrcStride = 640 * 4;
+  const int kDstStride = 160 * 4;
+  const int kSize = kSrcStride * 4;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 640 * 4; ++i) {
+    orig_pixels[i * 4 + 0] = i;
+    orig_pixels[i * 4 + 1] = 255 - i;
+    orig_pixels[i * 4 + 2] = i + 1;
+    orig_pixels[i * 4 + 3] = i + 10;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    ARGBScale(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
+              kFilterBilinear);
+  }
+
+  EXPECT_NEAR(66, dest_pixels[0], 4);
+  EXPECT_NEAR(255 - 66, dest_pixels[1], 4);
+  EXPECT_NEAR(67, dest_pixels[2], 4);
+  EXPECT_NEAR(76, dest_pixels[3], 4);
+
+  ARGBScale(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
+            kFilterNone);
+
+  EXPECT_EQ(2, dest_pixels[0]);
+  EXPECT_EQ(255 - 2, dest_pixels[1]);
+  EXPECT_EQ(3, dest_pixels[2]);
+  EXPECT_EQ(12, dest_pixels[3]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
 }  // namespace libyuv
diff --git a/files/unit_test/scale_rgb_test.cc b/files/unit_test/scale_rgb_test.cc
new file mode 100644
index 00000000..8296abe3
--- /dev/null
+++ b/files/unit_test/scale_rgb_test.cc
@@ -0,0 +1,280 @@
+/*
+ *  Copyright 2022 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.
+ */
+
+#include <stdlib.h>
+#include <time.h>
+
+#include "../unit_test/unit_test.h"
+#include "libyuv/cpu_id.h"
+#include "libyuv/scale_rgb.h"
+
+namespace libyuv {
+
+#define STRINGIZE(line) #line
+#define FILELINESTR(file, line) file ":" STRINGIZE(line)
+
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+// SLOW TESTS are those that are unoptimized C code.
+// FULL TESTS are optimized but test many variations of the same code.
+#define ENABLE_FULL_TESTS
+#endif
+
+// Test scaling with C vs Opt and return maximum pixel difference. 0 = exact.
+static int RGBTestFilter(int src_width,
+                         int src_height,
+                         int dst_width,
+                         int dst_height,
+                         FilterMode f,
+                         int benchmark_iterations,
+                         int disable_cpu_flags,
+                         int benchmark_cpu_info) {
+  if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
+    return 0;
+  }
+
+  int i, j;
+  const int b = 0;  // 128 to test for padding/stride.
+  int64_t src_rgb_plane_size =
+      (Abs(src_width) + b * 3) * (Abs(src_height) + b * 3) * 3LL;
+  int src_stride_rgb = (b * 3 + Abs(src_width)) * 3;
+
+  align_buffer_page_end(src_rgb, src_rgb_plane_size);
+  if (!src_rgb) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  MemRandomize(src_rgb, src_rgb_plane_size);
+
+  int64_t dst_rgb_plane_size = (dst_width + b * 3) * (dst_height + b * 3) * 3LL;
+  int dst_stride_rgb = (b * 3 + dst_width) * 3;
+
+  align_buffer_page_end(dst_rgb_c, dst_rgb_plane_size);
+  align_buffer_page_end(dst_rgb_opt, dst_rgb_plane_size);
+  if (!dst_rgb_c || !dst_rgb_opt) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  memset(dst_rgb_c, 2, dst_rgb_plane_size);
+  memset(dst_rgb_opt, 3, dst_rgb_plane_size);
+
+  // Warm up both versions for consistent benchmarks.
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  RGBScale(src_rgb + (src_stride_rgb * b) + b * 3, src_stride_rgb, src_width,
+           src_height, dst_rgb_c + (dst_stride_rgb * b) + b * 3, dst_stride_rgb,
+           dst_width, dst_height, f);
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  RGBScale(src_rgb + (src_stride_rgb * b) + b * 3, src_stride_rgb, src_width,
+           src_height, dst_rgb_opt + (dst_stride_rgb * b) + b * 3,
+           dst_stride_rgb, dst_width, dst_height, f);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  double c_time = get_time();
+  RGBScale(src_rgb + (src_stride_rgb * b) + b * 3, src_stride_rgb, src_width,
+           src_height, dst_rgb_c + (dst_stride_rgb * b) + b * 3, dst_stride_rgb,
+           dst_width, dst_height, f);
+
+  c_time = (get_time() - c_time);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  double opt_time = get_time();
+  for (i = 0; i < benchmark_iterations; ++i) {
+    RGBScale(src_rgb + (src_stride_rgb * b) + b * 3, src_stride_rgb, src_width,
+             src_height, dst_rgb_opt + (dst_stride_rgb * b) + b * 3,
+             dst_stride_rgb, dst_width, dst_height, f);
+  }
+  opt_time = (get_time() - opt_time) / benchmark_iterations;
+
+  // Report performance of C vs OPT
+  printf("filter %d - %8d us C - %8d us OPT\n", f,
+         static_cast<int>(c_time * 1e6), static_cast<int>(opt_time * 1e6));
+
+  // C version may be a little off from the optimized. Order of
+  //  operations may introduce rounding somewhere. So do a difference
+  //  of the buffers and look to see that the max difference isn't
+  //  over 2.
+  int max_diff = 0;
+  for (i = b; i < (dst_height + b); ++i) {
+    for (j = b * 3; j < (dst_width + b) * 3; ++j) {
+      int abs_diff = Abs(dst_rgb_c[(i * dst_stride_rgb) + j] -
+                         dst_rgb_opt[(i * dst_stride_rgb) + j]);
+      if (abs_diff > max_diff) {
+        max_diff = abs_diff;
+      }
+    }
+  }
+
+  free_aligned_buffer_page_end(dst_rgb_c);
+  free_aligned_buffer_page_end(dst_rgb_opt);
+  free_aligned_buffer_page_end(src_rgb);
+  return max_diff;
+}
+
+// The following adjustments in dimensions ensure the scale factor will be
+// exactly achieved.
+#define DX(x, nom, denom) static_cast<int>((Abs(x) / nom) * nom)
+#define SX(x, nom, denom) static_cast<int>((x / nom) * denom)
+
+#define TEST_FACTOR1(name, filter, nom, denom, max_diff)                     \
+  TEST_F(LibYUVScaleTest, RGBScaleDownBy##name##_##filter) {                 \
+    int diff = RGBTestFilter(                                                \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
+        benchmark_cpu_info_);                                                \
+    EXPECT_LE(diff, max_diff);                                               \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+// Test a scale factor with all 4 filters.  Expect unfiltered to be exact, but
+// filtering is different fixed point implementations for SSSE3, Neon and C.
+#define TEST_FACTOR(name, nom, denom)         \
+  TEST_FACTOR1(name, None, nom, denom, 0)     \
+  TEST_FACTOR1(name, Linear, nom, denom, 3)   \
+  TEST_FACTOR1(name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(name, Box, nom, denom, 3)
+#else
+// Test a scale factor with Bilinear.
+#define TEST_FACTOR(name, nom, denom) \
+  TEST_FACTOR1(name, Bilinear, nom, denom, 3)
+#endif
+
+TEST_FACTOR(2, 1, 2)
+#ifndef DISABLE_SLOW_TESTS
+TEST_FACTOR(4, 1, 4)
+// TEST_FACTOR(8, 1, 8)  Disable for benchmark performance.
+TEST_FACTOR(3by4, 3, 4)
+TEST_FACTOR(3by8, 3, 8)
+TEST_FACTOR(3, 1, 3)
+#endif
+#undef TEST_FACTOR1
+#undef TEST_FACTOR
+#undef SX
+#undef DX
+
+#define TEST_SCALETO1(name, width, height, filter, max_diff)                 \
+  TEST_F(LibYUVScaleTest, name##To##width##x##height##_##filter) {           \
+    int diff = RGBTestFilter(benchmark_width_, benchmark_height_, width,     \
+                             height, kFilter##filter, benchmark_iterations_, \
+                             disable_cpu_flags_, benchmark_cpu_info_);       \
+    EXPECT_LE(diff, max_diff);                                               \
+  }                                                                          \
+  TEST_F(LibYUVScaleTest, name##From##width##x##height##_##filter) {         \
+    int diff = RGBTestFilter(width, height, Abs(benchmark_width_),           \
+                             Abs(benchmark_height_), kFilter##filter,        \
+                             benchmark_iterations_, disable_cpu_flags_,      \
+                             benchmark_cpu_info_);                           \
+    EXPECT_LE(diff, max_diff);                                               \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+/// Test scale to a specified size with all 4 filters.
+#define TEST_SCALETO(name, width, height)       \
+  TEST_SCALETO1(name, width, height, None, 0)   \
+  TEST_SCALETO1(name, width, height, Linear, 3) \
+  TEST_SCALETO1(name, width, height, Bilinear, 3)
+#else
+#define TEST_SCALETO(name, width, height) \
+  TEST_SCALETO1(name, width, height, Bilinear, 3)
+#endif
+
+TEST_SCALETO(RGBScale, 640, 360)
+#ifndef DISABLE_SLOW_TESTS
+TEST_SCALETO(RGBScale, 1, 1)
+TEST_SCALETO(RGBScale, 256, 144) /* 128x72 * 3 */
+TEST_SCALETO(RGBScale, 320, 240)
+TEST_SCALETO(RGBScale, 569, 480)
+TEST_SCALETO(RGBScale, 1280, 720)
+TEST_SCALETO(RGBScale, 1920, 1080)
+#endif  // DISABLE_SLOW_TESTS
+#undef TEST_SCALETO1
+#undef TEST_SCALETO
+
+#define TEST_SCALESWAPXY1(name, filter, max_diff)                      \
+  TEST_F(LibYUVScaleTest, name##SwapXY_##filter) {                     \
+    int diff = RGBTestFilter(benchmark_width_, benchmark_height_,      \
+                             benchmark_height_, benchmark_width_,      \
+                             kFilter##filter, benchmark_iterations_,   \
+                             disable_cpu_flags_, benchmark_cpu_info_); \
+    EXPECT_LE(diff, max_diff);                                         \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+// Test scale with swapped width and height with all 3 filters.
+TEST_SCALESWAPXY1(RGBScale, None, 0)
+TEST_SCALESWAPXY1(RGBScale, Linear, 0)
+TEST_SCALESWAPXY1(RGBScale, Bilinear, 0)
+#else
+TEST_SCALESWAPXY1(RGBScale, Bilinear, 0)
+#endif
+#undef TEST_SCALESWAPXY1
+
+TEST_F(LibYUVScaleTest, RGBTest3x) {
+  const int kSrcStride = 480 * 3;
+  const int kDstStride = 160 * 3;
+  const int kSize = kSrcStride * 3;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 480 * 3; ++i) {
+    orig_pixels[i * 3 + 0] = i;
+    orig_pixels[i * 3 + 1] = 255 - i;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    RGBScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+             kFilterBilinear);
+  }
+
+  EXPECT_EQ(225, dest_pixels[0]);
+  EXPECT_EQ(255 - 225, dest_pixels[1]);
+
+  RGBScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+           kFilterNone);
+
+  EXPECT_EQ(225, dest_pixels[0]);
+  EXPECT_EQ(255 - 225, dest_pixels[1]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, RGBTest4x) {
+  const int kSrcStride = 640 * 3;
+  const int kDstStride = 160 * 3;
+  const int kSize = kSrcStride * 4;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 640 * 4; ++i) {
+    orig_pixels[i * 3 + 0] = i;
+    orig_pixels[i * 3 + 1] = 255 - i;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    RGBScale(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
+             kFilterBilinear);
+  }
+
+  EXPECT_EQ(66, dest_pixels[0]);
+  EXPECT_EQ(190, dest_pixels[1]);
+
+  RGBScale(orig_pixels, kSrcStride, 64, 4, dest_pixels, kDstStride, 16, 1,
+           kFilterNone);
+
+  EXPECT_EQ(2, dest_pixels[0]);  // expect the 3rd pixel of the 3rd row
+  EXPECT_EQ(255 - 2, dest_pixels[1]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+}  // namespace libyuv
diff --git a/files/unit_test/scale_test.cc b/files/unit_test/scale_test.cc
index 811b2d04..a8c95268 100644
--- a/files/unit_test/scale_test.cc
+++ b/files/unit_test/scale_test.cc
@@ -14,11 +14,20 @@
 #include "../unit_test/unit_test.h"
 #include "libyuv/cpu_id.h"
 #include "libyuv/scale.h"
+
+#ifdef ENABLE_ROW_TESTS
 #include "libyuv/scale_row.h"  // For ScaleRowDown2Box_Odd_C
+#endif
 
 #define STRINGIZE(line) #line
 #define FILELINESTR(file, line) file ":" STRINGIZE(line)
 
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+// SLOW TESTS are those that are unoptimized C code.
+// FULL TESTS are optimized but test many variations of the same code.
+#define ENABLE_FULL_TESTS
+#endif
+
 namespace libyuv {
 
 // Test scaling with C vs Opt and return maximum pixel difference. 0 = exact.
@@ -139,6 +148,123 @@ static int I420TestFilter(int src_width,
   return max_diff;
 }
 
+// Test scaling with 8 bit C vs 12 bit C and return maximum pixel difference.
+// 0 = exact.
+static int I420TestFilter_12(int src_width,
+                             int src_height,
+                             int dst_width,
+                             int dst_height,
+                             FilterMode f,
+                             int benchmark_iterations,
+                             int disable_cpu_flags,
+                             int benchmark_cpu_info) {
+  if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
+    return 0;
+  }
+
+  int i;
+  int src_width_uv = (Abs(src_width) + 1) >> 1;
+  int src_height_uv = (Abs(src_height) + 1) >> 1;
+
+  int64_t src_y_plane_size = (Abs(src_width)) * (Abs(src_height));
+  int64_t src_uv_plane_size = (src_width_uv) * (src_height_uv);
+
+  int src_stride_y = Abs(src_width);
+  int src_stride_uv = src_width_uv;
+
+  align_buffer_page_end(src_y, src_y_plane_size);
+  align_buffer_page_end(src_u, src_uv_plane_size);
+  align_buffer_page_end(src_v, src_uv_plane_size);
+  align_buffer_page_end(src_y_12, src_y_plane_size * 2);
+  align_buffer_page_end(src_u_12, src_uv_plane_size * 2);
+  align_buffer_page_end(src_v_12, src_uv_plane_size * 2);
+  if (!src_y || !src_u || !src_v || !src_y_12 || !src_u_12 || !src_v_12) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  uint16_t* p_src_y_12 = reinterpret_cast<uint16_t*>(src_y_12);
+  uint16_t* p_src_u_12 = reinterpret_cast<uint16_t*>(src_u_12);
+  uint16_t* p_src_v_12 = reinterpret_cast<uint16_t*>(src_v_12);
+
+  MemRandomize(src_y, src_y_plane_size);
+  MemRandomize(src_u, src_uv_plane_size);
+  MemRandomize(src_v, src_uv_plane_size);
+
+  for (i = 0; i < src_y_plane_size; ++i) {
+    p_src_y_12[i] = src_y[i];
+  }
+  for (i = 0; i < src_uv_plane_size; ++i) {
+    p_src_u_12[i] = src_u[i];
+    p_src_v_12[i] = src_v[i];
+  }
+
+  int dst_width_uv = (dst_width + 1) >> 1;
+  int dst_height_uv = (dst_height + 1) >> 1;
+
+  int dst_y_plane_size = (dst_width) * (dst_height);
+  int dst_uv_plane_size = (dst_width_uv) * (dst_height_uv);
+
+  int dst_stride_y = dst_width;
+  int dst_stride_uv = dst_width_uv;
+
+  align_buffer_page_end(dst_y_8, dst_y_plane_size);
+  align_buffer_page_end(dst_u_8, dst_uv_plane_size);
+  align_buffer_page_end(dst_v_8, dst_uv_plane_size);
+  align_buffer_page_end(dst_y_12, dst_y_plane_size * 2);
+  align_buffer_page_end(dst_u_12, dst_uv_plane_size * 2);
+  align_buffer_page_end(dst_v_12, dst_uv_plane_size * 2);
+
+  uint16_t* p_dst_y_12 = reinterpret_cast<uint16_t*>(dst_y_12);
+  uint16_t* p_dst_u_12 = reinterpret_cast<uint16_t*>(dst_u_12);
+  uint16_t* p_dst_v_12 = reinterpret_cast<uint16_t*>(dst_v_12);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  I420Scale(src_y, src_stride_y, src_u, src_stride_uv, src_v, src_stride_uv,
+            src_width, src_height, dst_y_8, dst_stride_y, dst_u_8,
+            dst_stride_uv, dst_v_8, dst_stride_uv, dst_width, dst_height, f);
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (i = 0; i < benchmark_iterations; ++i) {
+    I420Scale_12(p_src_y_12, src_stride_y, p_src_u_12, src_stride_uv,
+                 p_src_v_12, src_stride_uv, src_width, src_height, p_dst_y_12,
+                 dst_stride_y, p_dst_u_12, dst_stride_uv, p_dst_v_12,
+                 dst_stride_uv, dst_width, dst_height, f);
+  }
+
+  // Expect an exact match.
+  int max_diff = 0;
+  for (i = 0; i < dst_y_plane_size; ++i) {
+    int abs_diff = Abs(dst_y_8[i] - p_dst_y_12[i]);
+    if (abs_diff > max_diff) {
+      max_diff = abs_diff;
+    }
+  }
+  for (i = 0; i < dst_uv_plane_size; ++i) {
+    int abs_diff = Abs(dst_u_8[i] - p_dst_u_12[i]);
+    if (abs_diff > max_diff) {
+      max_diff = abs_diff;
+    }
+    abs_diff = Abs(dst_v_8[i] - p_dst_v_12[i]);
+    if (abs_diff > max_diff) {
+      max_diff = abs_diff;
+    }
+  }
+
+  free_aligned_buffer_page_end(dst_y_8);
+  free_aligned_buffer_page_end(dst_u_8);
+  free_aligned_buffer_page_end(dst_v_8);
+  free_aligned_buffer_page_end(dst_y_12);
+  free_aligned_buffer_page_end(dst_u_12);
+  free_aligned_buffer_page_end(dst_v_12);
+  free_aligned_buffer_page_end(src_y);
+  free_aligned_buffer_page_end(src_u);
+  free_aligned_buffer_page_end(src_v);
+  free_aligned_buffer_page_end(src_y_12);
+  free_aligned_buffer_page_end(src_u_12);
+  free_aligned_buffer_page_end(src_v_12);
+
+  return max_diff;
+}
+
 // Test scaling with 8 bit C vs 16 bit C and return maximum pixel difference.
 // 0 = exact.
 static int I420TestFilter_16(int src_width,
@@ -374,6 +500,123 @@ static int I444TestFilter(int src_width,
   return max_diff;
 }
 
+// Test scaling with 8 bit C vs 12 bit C and return maximum pixel difference.
+// 0 = exact.
+static int I444TestFilter_12(int src_width,
+                             int src_height,
+                             int dst_width,
+                             int dst_height,
+                             FilterMode f,
+                             int benchmark_iterations,
+                             int disable_cpu_flags,
+                             int benchmark_cpu_info) {
+  if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
+    return 0;
+  }
+
+  int i;
+  int src_width_uv = Abs(src_width);
+  int src_height_uv = Abs(src_height);
+
+  int64_t src_y_plane_size = (Abs(src_width)) * (Abs(src_height));
+  int64_t src_uv_plane_size = (src_width_uv) * (src_height_uv);
+
+  int src_stride_y = Abs(src_width);
+  int src_stride_uv = src_width_uv;
+
+  align_buffer_page_end(src_y, src_y_plane_size);
+  align_buffer_page_end(src_u, src_uv_plane_size);
+  align_buffer_page_end(src_v, src_uv_plane_size);
+  align_buffer_page_end(src_y_12, src_y_plane_size * 2);
+  align_buffer_page_end(src_u_12, src_uv_plane_size * 2);
+  align_buffer_page_end(src_v_12, src_uv_plane_size * 2);
+  if (!src_y || !src_u || !src_v || !src_y_12 || !src_u_12 || !src_v_12) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  uint16_t* p_src_y_12 = reinterpret_cast<uint16_t*>(src_y_12);
+  uint16_t* p_src_u_12 = reinterpret_cast<uint16_t*>(src_u_12);
+  uint16_t* p_src_v_12 = reinterpret_cast<uint16_t*>(src_v_12);
+
+  MemRandomize(src_y, src_y_plane_size);
+  MemRandomize(src_u, src_uv_plane_size);
+  MemRandomize(src_v, src_uv_plane_size);
+
+  for (i = 0; i < src_y_plane_size; ++i) {
+    p_src_y_12[i] = src_y[i];
+  }
+  for (i = 0; i < src_uv_plane_size; ++i) {
+    p_src_u_12[i] = src_u[i];
+    p_src_v_12[i] = src_v[i];
+  }
+
+  int dst_width_uv = dst_width;
+  int dst_height_uv = dst_height;
+
+  int dst_y_plane_size = (dst_width) * (dst_height);
+  int dst_uv_plane_size = (dst_width_uv) * (dst_height_uv);
+
+  int dst_stride_y = dst_width;
+  int dst_stride_uv = dst_width_uv;
+
+  align_buffer_page_end(dst_y_8, dst_y_plane_size);
+  align_buffer_page_end(dst_u_8, dst_uv_plane_size);
+  align_buffer_page_end(dst_v_8, dst_uv_plane_size);
+  align_buffer_page_end(dst_y_12, dst_y_plane_size * 2);
+  align_buffer_page_end(dst_u_12, dst_uv_plane_size * 2);
+  align_buffer_page_end(dst_v_12, dst_uv_plane_size * 2);
+
+  uint16_t* p_dst_y_12 = reinterpret_cast<uint16_t*>(dst_y_12);
+  uint16_t* p_dst_u_12 = reinterpret_cast<uint16_t*>(dst_u_12);
+  uint16_t* p_dst_v_12 = reinterpret_cast<uint16_t*>(dst_v_12);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  I444Scale(src_y, src_stride_y, src_u, src_stride_uv, src_v, src_stride_uv,
+            src_width, src_height, dst_y_8, dst_stride_y, dst_u_8,
+            dst_stride_uv, dst_v_8, dst_stride_uv, dst_width, dst_height, f);
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  for (i = 0; i < benchmark_iterations; ++i) {
+    I444Scale_12(p_src_y_12, src_stride_y, p_src_u_12, src_stride_uv,
+                 p_src_v_12, src_stride_uv, src_width, src_height, p_dst_y_12,
+                 dst_stride_y, p_dst_u_12, dst_stride_uv, p_dst_v_12,
+                 dst_stride_uv, dst_width, dst_height, f);
+  }
+
+  // Expect an exact match.
+  int max_diff = 0;
+  for (i = 0; i < dst_y_plane_size; ++i) {
+    int abs_diff = Abs(dst_y_8[i] - p_dst_y_12[i]);
+    if (abs_diff > max_diff) {
+      max_diff = abs_diff;
+    }
+  }
+  for (i = 0; i < dst_uv_plane_size; ++i) {
+    int abs_diff = Abs(dst_u_8[i] - p_dst_u_12[i]);
+    if (abs_diff > max_diff) {
+      max_diff = abs_diff;
+    }
+    abs_diff = Abs(dst_v_8[i] - p_dst_v_12[i]);
+    if (abs_diff > max_diff) {
+      max_diff = abs_diff;
+    }
+  }
+
+  free_aligned_buffer_page_end(dst_y_8);
+  free_aligned_buffer_page_end(dst_u_8);
+  free_aligned_buffer_page_end(dst_v_8);
+  free_aligned_buffer_page_end(dst_y_12);
+  free_aligned_buffer_page_end(dst_u_12);
+  free_aligned_buffer_page_end(dst_v_12);
+  free_aligned_buffer_page_end(src_y);
+  free_aligned_buffer_page_end(src_u);
+  free_aligned_buffer_page_end(src_v);
+  free_aligned_buffer_page_end(src_y_12);
+  free_aligned_buffer_page_end(src_u_12);
+  free_aligned_buffer_page_end(src_v_12);
+
+  return max_diff;
+}
+
 // Test scaling with 8 bit C vs 16 bit C and return maximum pixel difference.
 // 0 = exact.
 static int I444TestFilter_16(int src_width,
@@ -491,57 +734,185 @@ static int I444TestFilter_16(int src_width,
   return max_diff;
 }
 
+// Test scaling with C vs Opt and return maximum pixel difference. 0 = exact.
+static int NV12TestFilter(int src_width,
+                          int src_height,
+                          int dst_width,
+                          int dst_height,
+                          FilterMode f,
+                          int benchmark_iterations,
+                          int disable_cpu_flags,
+                          int benchmark_cpu_info) {
+  if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
+    return 0;
+  }
+
+  int i, j;
+  int src_width_uv = (Abs(src_width) + 1) >> 1;
+  int src_height_uv = (Abs(src_height) + 1) >> 1;
+
+  int64_t src_y_plane_size = (Abs(src_width)) * (Abs(src_height));
+  int64_t src_uv_plane_size = (src_width_uv) * (src_height_uv)*2;
+
+  int src_stride_y = Abs(src_width);
+  int src_stride_uv = src_width_uv * 2;
+
+  align_buffer_page_end(src_y, src_y_plane_size);
+  align_buffer_page_end(src_uv, src_uv_plane_size);
+  if (!src_y || !src_uv) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  MemRandomize(src_y, src_y_plane_size);
+  MemRandomize(src_uv, src_uv_plane_size);
+
+  int dst_width_uv = (dst_width + 1) >> 1;
+  int dst_height_uv = (dst_height + 1) >> 1;
+
+  int64_t dst_y_plane_size = (dst_width) * (dst_height);
+  int64_t dst_uv_plane_size = (dst_width_uv) * (dst_height_uv)*2;
+
+  int dst_stride_y = dst_width;
+  int dst_stride_uv = dst_width_uv * 2;
+
+  align_buffer_page_end(dst_y_c, dst_y_plane_size);
+  align_buffer_page_end(dst_uv_c, dst_uv_plane_size);
+  align_buffer_page_end(dst_y_opt, dst_y_plane_size);
+  align_buffer_page_end(dst_uv_opt, dst_uv_plane_size);
+  if (!dst_y_c || !dst_uv_c || !dst_y_opt || !dst_uv_opt) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  double c_time = get_time();
+  NV12Scale(src_y, src_stride_y, src_uv, src_stride_uv, src_width, src_height,
+            dst_y_c, dst_stride_y, dst_uv_c, dst_stride_uv, dst_width,
+            dst_height, f);
+  c_time = (get_time() - c_time);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  double opt_time = get_time();
+  for (i = 0; i < benchmark_iterations; ++i) {
+    NV12Scale(src_y, src_stride_y, src_uv, src_stride_uv, src_width, src_height,
+              dst_y_opt, dst_stride_y, dst_uv_opt, dst_stride_uv, dst_width,
+              dst_height, f);
+  }
+  opt_time = (get_time() - opt_time) / benchmark_iterations;
+  // Report performance of C vs OPT.
+  printf("filter %d - %8d us C - %8d us OPT\n", f,
+         static_cast<int>(c_time * 1e6), static_cast<int>(opt_time * 1e6));
+
+  // C version may be a little off from the optimized. Order of
+  //  operations may introduce rounding somewhere. So do a difference
+  //  of the buffers and look to see that the max difference is not
+  //  over 3.
+  int max_diff = 0;
+  for (i = 0; i < (dst_height); ++i) {
+    for (j = 0; j < (dst_width); ++j) {
+      int abs_diff = Abs(dst_y_c[(i * dst_stride_y) + j] -
+                         dst_y_opt[(i * dst_stride_y) + j]);
+      if (abs_diff > max_diff) {
+        max_diff = abs_diff;
+      }
+    }
+  }
+
+  for (i = 0; i < (dst_height_uv); ++i) {
+    for (j = 0; j < (dst_width_uv * 2); ++j) {
+      int abs_diff = Abs(dst_uv_c[(i * dst_stride_uv) + j] -
+                         dst_uv_opt[(i * dst_stride_uv) + j]);
+      if (abs_diff > max_diff) {
+        max_diff = abs_diff;
+      }
+    }
+  }
+
+  free_aligned_buffer_page_end(dst_y_c);
+  free_aligned_buffer_page_end(dst_uv_c);
+  free_aligned_buffer_page_end(dst_y_opt);
+  free_aligned_buffer_page_end(dst_uv_opt);
+  free_aligned_buffer_page_end(src_y);
+  free_aligned_buffer_page_end(src_uv);
+
+  return max_diff;
+}
+
 // The following adjustments in dimensions ensure the scale factor will be
 // exactly achieved.
 // 2 is chroma subsample.
 #define DX(x, nom, denom) static_cast<int>(((Abs(x) / nom + 1) / 2) * nom * 2)
 #define SX(x, nom, denom) static_cast<int>(((x / nom + 1) / 2) * denom * 2)
 
-#define TEST_FACTOR1(name, filter, nom, denom, max_diff)                     \
-  TEST_F(LibYUVScaleTest, I420ScaleDownBy##name##_##filter) {                \
-    int diff = I420TestFilter(                                               \
-        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
-        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
-        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
-        benchmark_cpu_info_);                                                \
-    EXPECT_LE(diff, max_diff);                                               \
-  }                                                                          \
-  TEST_F(LibYUVScaleTest, I444ScaleDownBy##name##_##filter) {                \
-    int diff = I444TestFilter(                                               \
-        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
-        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
-        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
-        benchmark_cpu_info_);                                                \
-    EXPECT_LE(diff, max_diff);                                               \
-  }                                                                          \
-  TEST_F(LibYUVScaleTest, I420ScaleDownBy##name##_##filter##_16) {           \
-    int diff = I420TestFilter_16(                                            \
-        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
-        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
-        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
-        benchmark_cpu_info_);                                                \
-    EXPECT_LE(diff, max_diff);                                               \
-  }                                                                          \
-  TEST_F(LibYUVScaleTest, I444ScaleDownBy##name##_##filter##_16) {           \
-    int diff = I444TestFilter_16(                                            \
-        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
-        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
-        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
-        benchmark_cpu_info_);                                                \
-    EXPECT_LE(diff, max_diff);                                               \
+#define TEST_FACTOR1(DISABLED_, name, filter, nom, denom, max_diff)           \
+  TEST_F(LibYUVScaleTest, I420ScaleDownBy##name##_##filter) {                 \
+    int diff = I420TestFilter(                                                \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom),  \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom),  \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,           \
+        benchmark_cpu_info_);                                                 \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest, I444ScaleDownBy##name##_##filter) {                 \
+    int diff = I444TestFilter(                                                \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom),  \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom),  \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,           \
+        benchmark_cpu_info_);                                                 \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest, DISABLED_##I420ScaleDownBy##name##_##filter##_12) { \
+    int diff = I420TestFilter_12(                                             \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom),  \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom),  \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,           \
+        benchmark_cpu_info_);                                                 \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest, DISABLED_##I444ScaleDownBy##name##_##filter##_12) { \
+    int diff = I444TestFilter_12(                                             \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom),  \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom),  \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,           \
+        benchmark_cpu_info_);                                                 \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest, NV12ScaleDownBy##name##_##filter) {                 \
+    int diff = NV12TestFilter(                                                \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom),  \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom),  \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,           \
+        benchmark_cpu_info_);                                                 \
+    EXPECT_LE(diff, max_diff);                                                \
   }
 
 // Test a scale factor with all 4 filters.  Expect unfiltered to be exact, but
 // filtering is different fixed point implementations for SSSE3, Neon and C.
-#define TEST_FACTOR(name, nom, denom, boxdiff) \
-  TEST_FACTOR1(name, None, nom, denom, 0)      \
-  TEST_FACTOR1(name, Linear, nom, denom, 3)    \
-  TEST_FACTOR1(name, Bilinear, nom, denom, 3)  \
-  TEST_FACTOR1(name, Box, nom, denom, boxdiff)
+#ifndef DISABLE_SLOW_TESTS
+#define TEST_FACTOR(name, nom, denom, boxdiff)  \
+  TEST_FACTOR1(, name, None, nom, denom, 0)     \
+  TEST_FACTOR1(, name, Linear, nom, denom, 3)   \
+  TEST_FACTOR1(, name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(, name, Box, nom, denom, boxdiff)
+#else
+#if defined(ENABLE_FULL_TESTS)
+#define TEST_FACTOR(name, nom, denom, boxdiff)           \
+  TEST_FACTOR1(DISABLED_, name, None, nom, denom, 0)     \
+  TEST_FACTOR1(DISABLED_, name, Linear, nom, denom, 3)   \
+  TEST_FACTOR1(DISABLED_, name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(DISABLED_, name, Box, nom, denom, boxdiff)
+#else
+#define TEST_FACTOR(name, nom, denom, boxdiff)           \
+  TEST_FACTOR1(DISABLED_, name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(DISABLED_, name, Box, nom, denom, boxdiff)
+#endif
+#endif
 
 TEST_FACTOR(2, 1, 2, 0)
 TEST_FACTOR(4, 1, 4, 0)
+#ifndef DISABLE_SLOW_TESTS
 TEST_FACTOR(8, 1, 8, 0)
+#endif
 TEST_FACTOR(3by4, 3, 4, 1)
 TEST_FACTOR(3by8, 3, 8, 1)
 TEST_FACTOR(3, 1, 3, 0)
@@ -550,7 +921,7 @@ TEST_FACTOR(3, 1, 3, 0)
 #undef SX
 #undef DX
 
-#define TEST_SCALETO1(name, width, height, filter, max_diff)                  \
+#define TEST_SCALETO1(DISABLED_, name, width, height, filter, max_diff)       \
   TEST_F(LibYUVScaleTest, I420##name##To##width##x##height##_##filter) {      \
     int diff = I420TestFilter(benchmark_width_, benchmark_height_, width,     \
                               height, kFilter##filter, benchmark_iterations_, \
@@ -563,18 +934,40 @@ TEST_FACTOR(3, 1, 3, 0)
                               disable_cpu_flags_, benchmark_cpu_info_);       \
     EXPECT_LE(diff, max_diff);                                                \
   }                                                                           \
-  TEST_F(LibYUVScaleTest, I420##name##To##width##x##height##_##filter##_16) { \
+  TEST_F(LibYUVScaleTest,                                                     \
+         DISABLED_##I420##name##To##width##x##height##_##filter##_12) {       \
+    int diff = I420TestFilter_12(                                             \
+        benchmark_width_, benchmark_height_, width, height, kFilter##filter,  \
+        benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_);      \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest,                                                     \
+         DISABLED_##I444##name##To##width##x##height##_##filter##_12) {       \
+    int diff = I444TestFilter_12(                                             \
+        benchmark_width_, benchmark_height_, width, height, kFilter##filter,  \
+        benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_);      \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest,                                                     \
+         DISABLED_##I420##name##To##width##x##height##_##filter##_16) {       \
     int diff = I420TestFilter_16(                                             \
         benchmark_width_, benchmark_height_, width, height, kFilter##filter,  \
         benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_);      \
     EXPECT_LE(diff, max_diff);                                                \
   }                                                                           \
-  TEST_F(LibYUVScaleTest, I444##name##To##width##x##height##_##filter##_16) { \
+  TEST_F(LibYUVScaleTest,                                                     \
+         DISABLED_##I444##name##To##width##x##height##_##filter##_16) {       \
     int diff = I444TestFilter_16(                                             \
         benchmark_width_, benchmark_height_, width, height, kFilter##filter,  \
         benchmark_iterations_, disable_cpu_flags_, benchmark_cpu_info_);      \
     EXPECT_LE(diff, max_diff);                                                \
   }                                                                           \
+  TEST_F(LibYUVScaleTest, NV12##name##To##width##x##height##_##filter) {      \
+    int diff = NV12TestFilter(benchmark_width_, benchmark_height_, width,     \
+                              height, kFilter##filter, benchmark_iterations_, \
+                              disable_cpu_flags_, benchmark_cpu_info_);       \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
   TEST_F(LibYUVScaleTest, I420##name##From##width##x##height##_##filter) {    \
     int diff = I420TestFilter(width, height, Abs(benchmark_width_),           \
                               Abs(benchmark_height_), kFilter##filter,        \
@@ -590,7 +983,23 @@ TEST_FACTOR(3, 1, 3, 0)
     EXPECT_LE(diff, max_diff);                                                \
   }                                                                           \
   TEST_F(LibYUVScaleTest,                                                     \
-         I420##name##From##width##x##height##_##filter##_16) {                \
+         DISABLED_##I420##name##From##width##x##height##_##filter##_12) {     \
+    int diff = I420TestFilter_12(width, height, Abs(benchmark_width_),        \
+                                 Abs(benchmark_height_), kFilter##filter,     \
+                                 benchmark_iterations_, disable_cpu_flags_,   \
+                                 benchmark_cpu_info_);                        \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest,                                                     \
+         DISABLED_##I444##name##From##width##x##height##_##filter##_12) {     \
+    int diff = I444TestFilter_12(width, height, Abs(benchmark_width_),        \
+                                 Abs(benchmark_height_), kFilter##filter,     \
+                                 benchmark_iterations_, disable_cpu_flags_,   \
+                                 benchmark_cpu_info_);                        \
+    EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest,                                                     \
+         DISABLED_##I420##name##From##width##x##height##_##filter##_16) {     \
     int diff = I420TestFilter_16(width, height, Abs(benchmark_width_),        \
                                  Abs(benchmark_height_), kFilter##filter,     \
                                  benchmark_iterations_, disable_cpu_flags_,   \
@@ -598,30 +1007,126 @@ TEST_FACTOR(3, 1, 3, 0)
     EXPECT_LE(diff, max_diff);                                                \
   }                                                                           \
   TEST_F(LibYUVScaleTest,                                                     \
-         I444##name##From##width##x##height##_##filter##_16) {                \
+         DISABLED_##I444##name##From##width##x##height##_##filter##_16) {     \
     int diff = I444TestFilter_16(width, height, Abs(benchmark_width_),        \
                                  Abs(benchmark_height_), kFilter##filter,     \
                                  benchmark_iterations_, disable_cpu_flags_,   \
                                  benchmark_cpu_info_);                        \
     EXPECT_LE(diff, max_diff);                                                \
+  }                                                                           \
+  TEST_F(LibYUVScaleTest, NV12##name##From##width##x##height##_##filter) {    \
+    int diff = NV12TestFilter(width, height, Abs(benchmark_width_),           \
+                              Abs(benchmark_height_), kFilter##filter,        \
+                              benchmark_iterations_, disable_cpu_flags_,      \
+                              benchmark_cpu_info_);                           \
+    EXPECT_LE(diff, max_diff);                                                \
   }
 
+#ifndef DISABLE_SLOW_TESTS
 // Test scale to a specified size with all 4 filters.
-#define TEST_SCALETO(name, width, height)         \
-  TEST_SCALETO1(name, width, height, None, 0)     \
-  TEST_SCALETO1(name, width, height, Linear, 3)   \
-  TEST_SCALETO1(name, width, height, Bilinear, 3) \
-  TEST_SCALETO1(name, width, height, Box, 3)
+#define TEST_SCALETO(name, width, height)           \
+  TEST_SCALETO1(, name, width, height, None, 0)     \
+  TEST_SCALETO1(, name, width, height, Linear, 3)   \
+  TEST_SCALETO1(, name, width, height, Bilinear, 3) \
+  TEST_SCALETO1(, name, width, height, Box, 3)
+#else
+#if defined(ENABLE_FULL_TESTS)
+#define TEST_SCALETO(name, width, height)                    \
+  TEST_SCALETO1(DISABLED_, name, width, height, None, 0)     \
+  TEST_SCALETO1(DISABLED_, name, width, height, Linear, 3)   \
+  TEST_SCALETO1(DISABLED_, name, width, height, Bilinear, 3) \
+  TEST_SCALETO1(DISABLED_, name, width, height, Box, 3)
+#else
+#define TEST_SCALETO(name, width, height)                    \
+  TEST_SCALETO1(DISABLED_, name, width, height, Bilinear, 3) \
+  TEST_SCALETO1(DISABLED_, name, width, height, Box, 3)
+#endif
+#endif
 
 TEST_SCALETO(Scale, 1, 1)
-TEST_SCALETO(Scale, 320, 240)
 TEST_SCALETO(Scale, 569, 480)
 TEST_SCALETO(Scale, 640, 360)
+#ifndef DISABLE_SLOW_TESTS
+TEST_SCALETO(Scale, 256, 144) /* 128x72 * 2 */
+TEST_SCALETO(Scale, 320, 240)
 TEST_SCALETO(Scale, 1280, 720)
 TEST_SCALETO(Scale, 1920, 1080)
+#endif  // DISABLE_SLOW_TESTS
 #undef TEST_SCALETO1
 #undef TEST_SCALETO
 
+#define TEST_SCALESWAPXY1(DISABLED_, name, filter, max_diff)               \
+  TEST_F(LibYUVScaleTest, I420##name##SwapXY_##filter) {                   \
+    int diff = I420TestFilter(benchmark_width_, benchmark_height_,         \
+                              benchmark_height_, benchmark_width_,         \
+                              kFilter##filter, benchmark_iterations_,      \
+                              disable_cpu_flags_, benchmark_cpu_info_);    \
+    EXPECT_LE(diff, max_diff);                                             \
+  }                                                                        \
+  TEST_F(LibYUVScaleTest, I444##name##SwapXY_##filter) {                   \
+    int diff = I444TestFilter(benchmark_width_, benchmark_height_,         \
+                              benchmark_height_, benchmark_width_,         \
+                              kFilter##filter, benchmark_iterations_,      \
+                              disable_cpu_flags_, benchmark_cpu_info_);    \
+    EXPECT_LE(diff, max_diff);                                             \
+  }                                                                        \
+  TEST_F(LibYUVScaleTest, DISABLED_##I420##name##SwapXY_##filter##_12) {   \
+    int diff = I420TestFilter_12(benchmark_width_, benchmark_height_,      \
+                                 benchmark_height_, benchmark_width_,      \
+                                 kFilter##filter, benchmark_iterations_,   \
+                                 disable_cpu_flags_, benchmark_cpu_info_); \
+    EXPECT_LE(diff, max_diff);                                             \
+  }                                                                        \
+  TEST_F(LibYUVScaleTest, DISABLED_##I444##name##SwapXY_##filter##_12) {   \
+    int diff = I444TestFilter_12(benchmark_width_, benchmark_height_,      \
+                                 benchmark_height_, benchmark_width_,      \
+                                 kFilter##filter, benchmark_iterations_,   \
+                                 disable_cpu_flags_, benchmark_cpu_info_); \
+    EXPECT_LE(diff, max_diff);                                             \
+  }                                                                        \
+  TEST_F(LibYUVScaleTest, DISABLED_##I420##name##SwapXY_##filter##_16) {   \
+    int diff = I420TestFilter_16(benchmark_width_, benchmark_height_,      \
+                                 benchmark_height_, benchmark_width_,      \
+                                 kFilter##filter, benchmark_iterations_,   \
+                                 disable_cpu_flags_, benchmark_cpu_info_); \
+    EXPECT_LE(diff, max_diff);                                             \
+  }                                                                        \
+  TEST_F(LibYUVScaleTest, DISABLED_##I444##name##SwapXY_##filter##_16) {   \
+    int diff = I444TestFilter_16(benchmark_width_, benchmark_height_,      \
+                                 benchmark_height_, benchmark_width_,      \
+                                 kFilter##filter, benchmark_iterations_,   \
+                                 disable_cpu_flags_, benchmark_cpu_info_); \
+    EXPECT_LE(diff, max_diff);                                             \
+  }                                                                        \
+  TEST_F(LibYUVScaleTest, NV12##name##SwapXY_##filter) {                   \
+    int diff = NV12TestFilter(benchmark_width_, benchmark_height_,         \
+                              benchmark_height_, benchmark_width_,         \
+                              kFilter##filter, benchmark_iterations_,      \
+                              disable_cpu_flags_, benchmark_cpu_info_);    \
+    EXPECT_LE(diff, max_diff);                                             \
+  }
+
+// Test scale to a specified size with all 4 filters.
+#ifndef DISABLE_SLOW_TESTS
+TEST_SCALESWAPXY1(, Scale, None, 0)
+TEST_SCALESWAPXY1(, Scale, Linear, 3)
+TEST_SCALESWAPXY1(, Scale, Bilinear, 3)
+TEST_SCALESWAPXY1(, Scale, Box, 3)
+#else
+#if defined(ENABLE_FULL_TESTS)
+TEST_SCALESWAPXY1(DISABLED_, Scale, None, 0)
+TEST_SCALESWAPXY1(DISABLED_, Scale, Linear, 3)
+TEST_SCALESWAPXY1(DISABLED_, Scale, Bilinear, 3)
+TEST_SCALESWAPXY1(DISABLED_, Scale, Box, 3)
+#else
+TEST_SCALESWAPXY1(DISABLED_, Scale, Bilinear, 3)
+TEST_SCALESWAPXY1(DISABLED_, Scale, Box, 3)
+#endif
+#endif
+
+#undef TEST_SCALESWAPXY1
+
+#ifdef ENABLE_ROW_TESTS
 #ifdef HAS_SCALEROWDOWN2_SSSE3
 TEST_F(LibYUVScaleTest, TestScaleRowDown2Box_Odd_SSSE3) {
   SIMD_ALIGNED(uint8_t orig_pixels[128 * 2]);
@@ -716,10 +1221,6 @@ extern "C" void ScaleRowUp2_16_NEON(const uint16_t* src_ptr,
                                     ptrdiff_t src_stride,
                                     uint16_t* dst,
                                     int dst_width);
-extern "C" void ScaleRowUp2_16_MMI(const uint16_t* src_ptr,
-                                   ptrdiff_t src_stride,
-                                   uint16_t* dst,
-                                   int dst_width);
 extern "C" void ScaleRowUp2_16_C(const uint16_t* src_ptr,
                                  ptrdiff_t src_stride,
                                  uint16_t* dst,
@@ -746,13 +1247,6 @@ TEST_F(LibYUVScaleTest, TestScaleRowUp2_16) {
     } else {
       ScaleRowUp2_16_C(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
     }
-#elif !defined(LIBYUV_DISABLE_MMI) && defined(_MIPS_ARCH_LOONGSON3A)
-    int has_mmi = TestCpuFlag(kCpuHasMMI);
-    if (has_mmi) {
-      ScaleRowUp2_16_MMI(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
-    } else {
-      ScaleRowUp2_16_C(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
-    }
 #else
     ScaleRowUp2_16_C(&orig_pixels[0], 640, &dst_pixels_opt[0], 1280);
 #endif
@@ -803,8 +1297,9 @@ TEST_F(LibYUVScaleTest, TestScaleRowDown2Box_16) {
   EXPECT_EQ(dst_pixels_c[0], (0 + 1 + 2560 + 2561 + 2) / 4);
   EXPECT_EQ(dst_pixels_c[1279], 3839);
 }
+#endif  // ENABLE_ROW_TESTS
 
-// Test scaling plane with 8 bit C vs 16 bit C and return maximum pixel
+// Test scaling plane with 8 bit C vs 12 bit C and return maximum pixel
 // difference.
 // 0 = exact.
 static int TestPlaneFilter_16(int src_width,
@@ -873,14 +1368,14 @@ static int TestPlaneFilter_16(int src_width,
 #define DX(x, nom, denom) static_cast<int>(((Abs(x) / nom + 1) / 2) * nom * 2)
 #define SX(x, nom, denom) static_cast<int>(((x / nom + 1) / 2) * denom * 2)
 
-#define TEST_FACTOR1(name, filter, nom, denom, max_diff)                     \
-  TEST_F(LibYUVScaleTest, ScalePlaneDownBy##name##_##filter##_16) {          \
-    int diff = TestPlaneFilter_16(                                           \
-        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
-        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
-        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
-        benchmark_cpu_info_);                                                \
-    EXPECT_LE(diff, max_diff);                                               \
+#define TEST_FACTOR1(name, filter, nom, denom, max_diff)                       \
+  TEST_F(LibYUVScaleTest, DISABLED_##ScalePlaneDownBy##name##_##filter##_16) { \
+    int diff = TestPlaneFilter_16(                                             \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom),   \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom),   \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,            \
+        benchmark_cpu_info_);                                                  \
+    EXPECT_LE(diff, max_diff);                                                 \
   }
 
 // Test a scale factor with all 4 filters.  Expect unfiltered to be exact, but
@@ -893,7 +1388,7 @@ static int TestPlaneFilter_16(int src_width,
 
 TEST_FACTOR(2, 1, 2, 0)
 TEST_FACTOR(4, 1, 4, 0)
-TEST_FACTOR(8, 1, 8, 0)
+// TEST_FACTOR(8, 1, 8, 0) Disable for benchmark performance.  Takes 90 seconds.
 TEST_FACTOR(3by4, 3, 4, 1)
 TEST_FACTOR(3by8, 3, 8, 1)
 TEST_FACTOR(3, 1, 3, 0)
@@ -901,4 +1396,206 @@ TEST_FACTOR(3, 1, 3, 0)
 #undef TEST_FACTOR
 #undef SX
 #undef DX
+
+TEST_F(LibYUVScaleTest, PlaneTest3x) {
+  const int kSrcStride = 480;
+  const int kDstStride = 160;
+  const int kSize = kSrcStride * 3;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 480 * 3; ++i) {
+    orig_pixels[i] = i;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    ScalePlane(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+               kFilterBilinear);
+  }
+
+  EXPECT_EQ(225, dest_pixels[0]);
+
+  ScalePlane(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+             kFilterNone);
+
+  EXPECT_EQ(225, dest_pixels[0]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, PlaneTest4x) {
+  const int kSrcStride = 640;
+  const int kDstStride = 160;
+  const int kSize = kSrcStride * 4;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 640 * 4; ++i) {
+    orig_pixels[i] = i;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    ScalePlane(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
+               kFilterBilinear);
+  }
+
+  EXPECT_EQ(66, dest_pixels[0]);
+
+  ScalePlane(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
+             kFilterNone);
+
+  EXPECT_EQ(2, dest_pixels[0]);  // expect the 3rd pixel of the 3rd row
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+// Intent is to test 200x50 to 50x200 but width and height can be parameters.
+TEST_F(LibYUVScaleTest, PlaneTestRotate_None) {
+  const int kSize = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < kSize; ++i) {
+    orig_pixels[i] = i;
+  }
+  align_buffer_page_end(dest_opt_pixels, kSize);
+  align_buffer_page_end(dest_c_pixels, kSize);
+
+  MaskCpuFlags(disable_cpu_flags_);  // Disable all CPU optimization.
+  ScalePlane(orig_pixels, benchmark_width_, benchmark_width_, benchmark_height_,
+             dest_c_pixels, benchmark_height_, benchmark_height_,
+             benchmark_width_, kFilterNone);
+  MaskCpuFlags(benchmark_cpu_info_);  // Enable all CPU optimization.
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    ScalePlane(orig_pixels, benchmark_width_, benchmark_width_,
+               benchmark_height_, dest_opt_pixels, benchmark_height_,
+               benchmark_height_, benchmark_width_, kFilterNone);
+  }
+
+  for (int i = 0; i < kSize; ++i) {
+    EXPECT_EQ(dest_c_pixels[i], dest_opt_pixels[i]);
+  }
+
+  free_aligned_buffer_page_end(dest_c_pixels);
+  free_aligned_buffer_page_end(dest_opt_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, PlaneTestRotate_Bilinear) {
+  const int kSize = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < kSize; ++i) {
+    orig_pixels[i] = i;
+  }
+  align_buffer_page_end(dest_opt_pixels, kSize);
+  align_buffer_page_end(dest_c_pixels, kSize);
+
+  MaskCpuFlags(disable_cpu_flags_);  // Disable all CPU optimization.
+  ScalePlane(orig_pixels, benchmark_width_, benchmark_width_, benchmark_height_,
+             dest_c_pixels, benchmark_height_, benchmark_height_,
+             benchmark_width_, kFilterBilinear);
+  MaskCpuFlags(benchmark_cpu_info_);  // Enable all CPU optimization.
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    ScalePlane(orig_pixels, benchmark_width_, benchmark_width_,
+               benchmark_height_, dest_opt_pixels, benchmark_height_,
+               benchmark_height_, benchmark_width_, kFilterBilinear);
+  }
+
+  for (int i = 0; i < kSize; ++i) {
+    EXPECT_EQ(dest_c_pixels[i], dest_opt_pixels[i]);
+  }
+
+  free_aligned_buffer_page_end(dest_c_pixels);
+  free_aligned_buffer_page_end(dest_opt_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+// Intent is to test 200x50 to 50x200 but width and height can be parameters.
+TEST_F(LibYUVScaleTest, PlaneTestRotate_Box) {
+  const int kSize = benchmark_width_ * benchmark_height_;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < kSize; ++i) {
+    orig_pixels[i] = i;
+  }
+  align_buffer_page_end(dest_opt_pixels, kSize);
+  align_buffer_page_end(dest_c_pixels, kSize);
+
+  MaskCpuFlags(disable_cpu_flags_);  // Disable all CPU optimization.
+  ScalePlane(orig_pixels, benchmark_width_, benchmark_width_, benchmark_height_,
+             dest_c_pixels, benchmark_height_, benchmark_height_,
+             benchmark_width_, kFilterBox);
+  MaskCpuFlags(benchmark_cpu_info_);  // Enable all CPU optimization.
+
+  for (int i = 0; i < benchmark_iterations_; ++i) {
+    ScalePlane(orig_pixels, benchmark_width_, benchmark_width_,
+               benchmark_height_, dest_opt_pixels, benchmark_height_,
+               benchmark_height_, benchmark_width_, kFilterBox);
+  }
+
+  for (int i = 0; i < kSize; ++i) {
+    EXPECT_EQ(dest_c_pixels[i], dest_opt_pixels[i]);
+  }
+
+  free_aligned_buffer_page_end(dest_c_pixels);
+  free_aligned_buffer_page_end(dest_opt_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, PlaneTest1_Box) {
+  align_buffer_page_end(orig_pixels, 3);
+  align_buffer_page_end(dst_pixels, 3);
+
+  // Pad the 1x1 byte image with invalid values before and after in case libyuv
+  // reads outside the memory boundaries.
+  orig_pixels[0] = 0;
+  orig_pixels[1] = 1;  // scale this pixel
+  orig_pixels[2] = 2;
+  dst_pixels[0] = 3;
+  dst_pixels[1] = 3;
+  dst_pixels[2] = 3;
+
+  libyuv::ScalePlane(orig_pixels + 1, /* src_stride= */ 1, /* src_width= */ 1,
+                     /* src_height= */ 1, dst_pixels, /* dst_stride= */ 1,
+                     /* dst_width= */ 1, /* dst_height= */ 2,
+                     libyuv::kFilterBox);
+
+  EXPECT_EQ(dst_pixels[0], 1);
+  EXPECT_EQ(dst_pixels[1], 1);
+  EXPECT_EQ(dst_pixels[2], 3);
+
+  free_aligned_buffer_page_end(dst_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, PlaneTest1_16_Box) {
+  align_buffer_page_end(orig_pixels_alloc, 3 * 2);
+  align_buffer_page_end(dst_pixels_alloc, 3 * 2);
+  uint16_t* orig_pixels = (uint16_t*)orig_pixels_alloc;
+  uint16_t* dst_pixels = (uint16_t*)dst_pixels_alloc;
+
+  // Pad the 1x1 byte image with invalid values before and after in case libyuv
+  // reads outside the memory boundaries.
+  orig_pixels[0] = 0;
+  orig_pixels[1] = 1;  // scale this pixel
+  orig_pixels[2] = 2;
+  dst_pixels[0] = 3;
+  dst_pixels[1] = 3;
+  dst_pixels[2] = 3;
+
+  libyuv::ScalePlane_16(
+      orig_pixels + 1, /* src_stride= */ 1, /* src_width= */ 1,
+      /* src_height= */ 1, dst_pixels, /* dst_stride= */ 1,
+      /* dst_width= */ 1, /* dst_height= */ 2, libyuv::kFilterNone);
+
+  EXPECT_EQ(dst_pixels[0], 1);
+  EXPECT_EQ(dst_pixels[1], 1);
+  EXPECT_EQ(dst_pixels[2], 3);
+
+  free_aligned_buffer_page_end(dst_pixels_alloc);
+  free_aligned_buffer_page_end(orig_pixels_alloc);
+}
 }  // namespace libyuv
diff --git a/files/unit_test/scale_uv_test.cc b/files/unit_test/scale_uv_test.cc
new file mode 100644
index 00000000..3d524bef
--- /dev/null
+++ b/files/unit_test/scale_uv_test.cc
@@ -0,0 +1,278 @@
+/*
+ *  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.
+ */
+
+#include <stdlib.h>
+#include <time.h>
+
+#include "../unit_test/unit_test.h"
+#include "libyuv/cpu_id.h"
+#include "libyuv/scale_uv.h"
+
+namespace libyuv {
+
+#define STRINGIZE(line) #line
+#define FILELINESTR(file, line) file ":" STRINGIZE(line)
+
+#if !defined(DISABLE_SLOW_TESTS) || defined(__x86_64__) || defined(__i386__)
+// SLOW TESTS are those that are unoptimized C code.
+// FULL TESTS are optimized but test many variations of the same code.
+#define ENABLE_FULL_TESTS
+#endif
+
+// Test scaling with C vs Opt and return maximum pixel difference. 0 = exact.
+static int UVTestFilter(int src_width,
+                        int src_height,
+                        int dst_width,
+                        int dst_height,
+                        FilterMode f,
+                        int benchmark_iterations,
+                        int disable_cpu_flags,
+                        int benchmark_cpu_info) {
+  if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
+    return 0;
+  }
+
+  int i, j;
+  const int b = 0;  // 128 to test for padding/stride.
+  int64_t src_uv_plane_size =
+      (Abs(src_width) + b * 2) * (Abs(src_height) + b * 2) * 2LL;
+  int src_stride_uv = (b * 2 + Abs(src_width)) * 2;
+
+  align_buffer_page_end(src_uv, src_uv_plane_size);
+  if (!src_uv) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  MemRandomize(src_uv, src_uv_plane_size);
+
+  int64_t dst_uv_plane_size = (dst_width + b * 2) * (dst_height + b * 2) * 2LL;
+  int dst_stride_uv = (b * 2 + dst_width) * 2;
+
+  align_buffer_page_end(dst_uv_c, dst_uv_plane_size);
+  align_buffer_page_end(dst_uv_opt, dst_uv_plane_size);
+  if (!dst_uv_c || !dst_uv_opt) {
+    printf("Skipped.  Alloc failed " FILELINESTR(__FILE__, __LINE__) "\n");
+    return 0;
+  }
+  memset(dst_uv_c, 2, dst_uv_plane_size);
+  memset(dst_uv_opt, 3, dst_uv_plane_size);
+
+  // Warm up both versions for consistent benchmarks.
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
+          src_height, dst_uv_c + (dst_stride_uv * b) + b * 2, dst_stride_uv,
+          dst_width, dst_height, f);
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
+          src_height, dst_uv_opt + (dst_stride_uv * b) + b * 2, dst_stride_uv,
+          dst_width, dst_height, f);
+
+  MaskCpuFlags(disable_cpu_flags);  // Disable all CPU optimization.
+  double c_time = get_time();
+  UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
+          src_height, dst_uv_c + (dst_stride_uv * b) + b * 2, dst_stride_uv,
+          dst_width, dst_height, f);
+
+  c_time = (get_time() - c_time);
+
+  MaskCpuFlags(benchmark_cpu_info);  // Enable all CPU optimization.
+  double opt_time = get_time();
+  for (i = 0; i < benchmark_iterations; ++i) {
+    UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
+            src_height, dst_uv_opt + (dst_stride_uv * b) + b * 2, dst_stride_uv,
+            dst_width, dst_height, f);
+  }
+  opt_time = (get_time() - opt_time) / benchmark_iterations;
+
+  // Report performance of C vs OPT
+  printf("filter %d - %8d us C - %8d us OPT\n", f,
+         static_cast<int>(c_time * 1e6), static_cast<int>(opt_time * 1e6));
+
+  // C version may be a little off from the optimized. Order of
+  //  operations may introduce rounding somewhere. So do a difference
+  //  of the buffers and look to see that the max difference isn't
+  //  over 2.
+  int max_diff = 0;
+  for (i = b; i < (dst_height + b); ++i) {
+    for (j = b * 2; j < (dst_width + b) * 2; ++j) {
+      int abs_diff = Abs(dst_uv_c[(i * dst_stride_uv) + j] -
+                         dst_uv_opt[(i * dst_stride_uv) + j]);
+      if (abs_diff > max_diff) {
+        max_diff = abs_diff;
+      }
+    }
+  }
+
+  free_aligned_buffer_page_end(dst_uv_c);
+  free_aligned_buffer_page_end(dst_uv_opt);
+  free_aligned_buffer_page_end(src_uv);
+  return max_diff;
+}
+
+// The following adjustments in dimensions ensure the scale factor will be
+// exactly achieved.
+#define DX(x, nom, denom) static_cast<int>((Abs(x) / nom) * nom)
+#define SX(x, nom, denom) static_cast<int>((x / nom) * denom)
+
+#define TEST_FACTOR1(name, filter, nom, denom, max_diff)                     \
+  TEST_F(LibYUVScaleTest, UVScaleDownBy##name##_##filter) {                  \
+    int diff = UVTestFilter(                                                 \
+        SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
+        DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
+        kFilter##filter, benchmark_iterations_, disable_cpu_flags_,          \
+        benchmark_cpu_info_);                                                \
+    EXPECT_LE(diff, max_diff);                                               \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+// Test a scale factor with all 4 filters.  Expect unfiltered to be exact, but
+// filtering is different fixed point implementations for SSSE3, Neon and C.
+#define TEST_FACTOR(name, nom, denom)         \
+  TEST_FACTOR1(name, None, nom, denom, 0)     \
+  TEST_FACTOR1(name, Linear, nom, denom, 3)   \
+  TEST_FACTOR1(name, Bilinear, nom, denom, 3) \
+  TEST_FACTOR1(name, Box, nom, denom, 3)
+#else
+// Test a scale factor with Bilinear.
+#define TEST_FACTOR(name, nom, denom) \
+  TEST_FACTOR1(name, Bilinear, nom, denom, 3)
+#endif
+
+TEST_FACTOR(2, 1, 2)
+TEST_FACTOR(4, 1, 4)
+// TEST_FACTOR(8, 1, 8)  Disable for benchmark performance.
+TEST_FACTOR(3by4, 3, 4)
+TEST_FACTOR(3by8, 3, 8)
+TEST_FACTOR(3, 1, 3)
+#undef TEST_FACTOR1
+#undef TEST_FACTOR
+#undef SX
+#undef DX
+
+#define TEST_SCALETO1(name, width, height, filter, max_diff)                \
+  TEST_F(LibYUVScaleTest, name##To##width##x##height##_##filter) {          \
+    int diff = UVTestFilter(benchmark_width_, benchmark_height_, width,     \
+                            height, kFilter##filter, benchmark_iterations_, \
+                            disable_cpu_flags_, benchmark_cpu_info_);       \
+    EXPECT_LE(diff, max_diff);                                              \
+  }                                                                         \
+  TEST_F(LibYUVScaleTest, name##From##width##x##height##_##filter) {        \
+    int diff = UVTestFilter(width, height, Abs(benchmark_width_),           \
+                            Abs(benchmark_height_), kFilter##filter,        \
+                            benchmark_iterations_, disable_cpu_flags_,      \
+                            benchmark_cpu_info_);                           \
+    EXPECT_LE(diff, max_diff);                                              \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+/// Test scale to a specified size with all 4 filters.
+#define TEST_SCALETO(name, width, height)       \
+  TEST_SCALETO1(name, width, height, None, 0)   \
+  TEST_SCALETO1(name, width, height, Linear, 3) \
+  TEST_SCALETO1(name, width, height, Bilinear, 3)
+#else
+#define TEST_SCALETO(name, width, height) \
+  TEST_SCALETO1(name, width, height, Bilinear, 3)
+#endif
+
+TEST_SCALETO(UVScale, 1, 1)
+TEST_SCALETO(UVScale, 569, 480)
+TEST_SCALETO(UVScale, 640, 360)
+#ifndef DISABLE_SLOW_TESTS
+TEST_SCALETO(UVScale, 256, 144) /* 128x72 * 2 */
+TEST_SCALETO(UVScale, 320, 240)
+TEST_SCALETO(UVScale, 1280, 720)
+TEST_SCALETO(UVScale, 1920, 1080)
+#endif  // DISABLE_SLOW_TESTS
+#undef TEST_SCALETO1
+#undef TEST_SCALETO
+
+#define TEST_SCALESWAPXY1(name, filter, max_diff)                              \
+  TEST_F(LibYUVScaleTest, name##SwapXY_##filter) {                             \
+    int diff =                                                                 \
+        UVTestFilter(benchmark_width_, benchmark_height_, benchmark_height_,   \
+                     benchmark_width_, kFilter##filter, benchmark_iterations_, \
+                     disable_cpu_flags_, benchmark_cpu_info_);                 \
+    EXPECT_LE(diff, max_diff);                                                 \
+  }
+
+#if defined(ENABLE_FULL_TESTS)
+// Test scale with swapped width and height with all 3 filters.
+TEST_SCALESWAPXY1(UVScale, None, 0)
+TEST_SCALESWAPXY1(UVScale, Linear, 0)
+TEST_SCALESWAPXY1(UVScale, Bilinear, 0)
+#else
+TEST_SCALESWAPXY1(UVScale, Bilinear, 0)
+#endif
+#undef TEST_SCALESWAPXY1
+
+TEST_F(LibYUVScaleTest, UVTest3x) {
+  const int kSrcStride = 480 * 2;
+  const int kDstStride = 160 * 2;
+  const int kSize = kSrcStride * 3;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 480 * 3; ++i) {
+    orig_pixels[i * 2 + 0] = i;
+    orig_pixels[i * 2 + 1] = 255 - i;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    UVScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+            kFilterBilinear);
+  }
+
+  EXPECT_EQ(225, dest_pixels[0]);
+  EXPECT_EQ(255 - 225, dest_pixels[1]);
+
+  UVScale(orig_pixels, kSrcStride, 480, 3, dest_pixels, kDstStride, 160, 1,
+          kFilterNone);
+
+  EXPECT_EQ(225, dest_pixels[0]);
+  EXPECT_EQ(255 - 225, dest_pixels[1]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+TEST_F(LibYUVScaleTest, UVTest4x) {
+  const int kSrcStride = 640 * 2;
+  const int kDstStride = 160 * 2;
+  const int kSize = kSrcStride * 4;
+  align_buffer_page_end(orig_pixels, kSize);
+  for (int i = 0; i < 640 * 4; ++i) {
+    orig_pixels[i * 2 + 0] = i;
+    orig_pixels[i * 2 + 1] = 255 - i;
+  }
+  align_buffer_page_end(dest_pixels, kDstStride);
+
+  int iterations160 = (benchmark_width_ * benchmark_height_ + (160 - 1)) / 160 *
+                      benchmark_iterations_;
+  for (int i = 0; i < iterations160; ++i) {
+    UVScale(orig_pixels, kSrcStride, 640, 4, dest_pixels, kDstStride, 160, 1,
+            kFilterBilinear);
+  }
+
+  EXPECT_EQ(66, dest_pixels[0]);
+  EXPECT_EQ(190, dest_pixels[1]);
+
+  UVScale(orig_pixels, kSrcStride, 64, 4, dest_pixels, kDstStride, 16, 1,
+          kFilterNone);
+
+  EXPECT_EQ(2, dest_pixels[0]);  // expect the 3rd pixel of the 3rd row
+  EXPECT_EQ(255 - 2, dest_pixels[1]);
+
+  free_aligned_buffer_page_end(dest_pixels);
+  free_aligned_buffer_page_end(orig_pixels);
+}
+
+}  // namespace libyuv
diff --git a/files/unit_test/testdata/mips.txt b/files/unit_test/testdata/mips.txt
new file mode 100644
index 00000000..d9f28cbf
--- /dev/null
+++ b/files/unit_test/testdata/mips.txt
@@ -0,0 +1,7 @@
+system type             : generic-loongson-machine
+machine                 : loongson,generic
+processor               : 0
+
+isa                     : mips1 mips2 mips3 mips4 mips5 mips32r1 mips32r2 mips64r1 mips64r2
+ASEs implemented        : vz
+shadow register sets    : 1
diff --git a/files/unit_test/testdata/mips_loongson2k.txt b/files/unit_test/testdata/mips_loongson2k.txt
new file mode 100644
index 00000000..8a88d38f
--- /dev/null
+++ b/files/unit_test/testdata/mips_loongson2k.txt
@@ -0,0 +1,5 @@
+system type		: Loongson2K-SBC
+machine			: loongson,LS2k1000-EVP
+processor		: 0
+cpu model		: Loongson-2K V0.3  FPU V0.1
+BogoMIPS		: 1980.41
diff --git a/files/unit_test/testdata/mips_loongson3.txt b/files/unit_test/testdata/mips_loongson3.txt
new file mode 100644
index 00000000..1f540b12
--- /dev/null
+++ b/files/unit_test/testdata/mips_loongson3.txt
@@ -0,0 +1,10 @@
+system type		: generic-loongson-machine
+machine			: Unknown
+processor		: 0
+cpu model		: ICT Loongson-3 V0.9  FPU V0.1
+model name		: ICT Loongson-3A R3 (Loongson-3A3000) @ 1500MHz
+BogoMIPS		: 2990.15
+
+isa			: mips1 mips2 mips3 mips4 mips5 mips32r1 mips32r2 mips64r1 mips64r2
+ASEs implemented	: dsp dsp2 vz
+shadow register sets	: 1
diff --git a/files/unit_test/testdata/mips_loongson_mmi.txt b/files/unit_test/testdata/mips_loongson_mmi.txt
new file mode 100644
index 00000000..0f10b8bb
--- /dev/null
+++ b/files/unit_test/testdata/mips_loongson_mmi.txt
@@ -0,0 +1,7 @@
+system type             : generic-loongson-machine
+machine                 : loongson,generic
+processor               : 0
+
+isa                     : mips1 mips2 mips3 mips4 mips5 mips32r1 mips32r2 mips64r1 mips64r2
+ASEs implemented        : vz loongson-mmi loongson-ext
+shadow register sets    : 1
diff --git a/files/unit_test/testdata/mips_msa.txt b/files/unit_test/testdata/mips_msa.txt
new file mode 100644
index 00000000..ac930615
--- /dev/null
+++ b/files/unit_test/testdata/mips_msa.txt
@@ -0,0 +1,7 @@
+system type             : generic-loongson-machine
+machine                 : loongson,generic
+processor               : 0
+
+isa                     : mips1 mips2 mips3 mips4 mips5 mips32r1 mips32r2 mips64r1 mips64r2
+ASEs implemented        : vz msa
+shadow register sets    : 1
diff --git a/files/unit_test/unit_test.cc b/files/unit_test/unit_test.cc
index a1ae7ea3..61145a46 100644
--- a/files/unit_test/unit_test.cc
+++ b/files/unit_test/unit_test.cc
@@ -14,23 +14,28 @@
 
 #include <cstring>
 
-#ifdef LIBYUV_USE_GFLAGS
-#include "gflags/gflags.h"
+#ifdef LIBYUV_USE_ABSL_FLAGS
+#include "absl/flags/flag.h"
+#include "absl/flags/parse.h"
 #endif
 #include "libyuv/cpu_id.h"
 
 unsigned int fastrand_seed = 0xfb;
 
-#ifdef LIBYUV_USE_GFLAGS
-DEFINE_int32(libyuv_width, 0, "width of test image.");
-DEFINE_int32(libyuv_height, 0, "height of test image.");
-DEFINE_int32(libyuv_repeat, 0, "number of times to repeat test.");
-DEFINE_int32(libyuv_flags, 0, "cpu flags for reference code. 1 = C, -1 = SIMD");
-DEFINE_int32(libyuv_cpu_info,
-             0,
-             "cpu flags for benchmark code. 1 = C, -1 = SIMD");
+#ifdef LIBYUV_USE_ABSL_FLAGS
+ABSL_FLAG(int32_t, libyuv_width, 0, "width of test image.");
+ABSL_FLAG(int32_t, libyuv_height, 0, "height of test image.");
+ABSL_FLAG(int32_t, libyuv_repeat, 0, "number of times to repeat test.");
+ABSL_FLAG(int32_t,
+          libyuv_flags,
+          0,
+          "cpu flags for reference code. 1 = C, -1 = SIMD");
+ABSL_FLAG(int32_t,
+          libyuv_cpu_info,
+          0,
+          "cpu flags for benchmark code. 1 = C, -1 = SIMD");
 #else
-// Disable command line parameters if gflags disabled.
+// Disable command line parameters if absl/flags disabled.
 static const int32_t FLAGS_libyuv_width = 0;
 static const int32_t FLAGS_libyuv_height = 0;
 static const int32_t FLAGS_libyuv_repeat = 0;
@@ -38,6 +43,12 @@ static const int32_t FLAGS_libyuv_flags = 0;
 static const int32_t FLAGS_libyuv_cpu_info = 0;
 #endif
 
+#ifdef LIBYUV_USE_ABSL_FLAGS
+#define LIBYUV_GET_FLAG(f) absl::GetFlag(f)
+#else
+#define LIBYUV_GET_FLAG(f) f
+#endif
+
 // Test environment variable for disabling CPU features. Any non-zero value
 // to disable. Zero ignored to make it easy to set the variable on/off.
 #if !defined(__native_client__) && !defined(_M_ARM)
@@ -66,8 +77,15 @@ int TestCpuEnv(int cpu_info) {
   if (TestEnv("LIBYUV_DISABLE_MSA")) {
     cpu_info &= ~libyuv::kCpuHasMSA;
   }
-  if (TestEnv("LIBYUV_DISABLE_MMI")) {
-    cpu_info &= ~libyuv::kCpuHasMMI;
+#endif
+#if defined(__longarch__) && defined(__linux__)
+  if (TestEnv("LIBYUV_DISABLE_LSX")) {
+    cpu_info &= ~libyuv::kCpuHasLSX;
+  }
+#endif
+#if defined(__longarch__) && defined(__linux__)
+  if (TestEnv("LIBYUV_DISABLE_LASX")) {
+    cpu_info &= ~libyuv::kCpuHasLASX;
   }
 #endif
 #if !defined(__pnacl__) && !defined(__CLR_VER) &&                   \
@@ -109,6 +127,9 @@ int TestCpuEnv(int cpu_info) {
   if (TestEnv("LIBYUV_DISABLE_AVX512VL")) {
     cpu_info &= ~libyuv::kCpuHasAVX512VL;
   }
+  if (TestEnv("LIBYUV_DISABLE_AVX512VNNI")) {
+    cpu_info &= ~libyuv::kCpuHasAVX512VNNI;
+  }
   if (TestEnv("LIBYUV_DISABLE_AVX512VBMI")) {
     cpu_info &= ~libyuv::kCpuHasAVX512VBMI;
   }
@@ -145,8 +166,8 @@ LibYUVConvertTest::LibYUVConvertTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -156,29 +177,29 @@ LibYUVConvertTest::LibYUVConvertTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -201,8 +222,8 @@ LibYUVColorTest::LibYUVColorTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -212,29 +233,29 @@ LibYUVColorTest::LibYUVColorTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -257,8 +278,8 @@ LibYUVScaleTest::LibYUVScaleTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -268,29 +289,29 @@ LibYUVScaleTest::LibYUVScaleTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -313,8 +334,8 @@ LibYUVRotateTest::LibYUVRotateTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -324,29 +345,29 @@ LibYUVRotateTest::LibYUVRotateTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -369,8 +390,8 @@ LibYUVPlanarTest::LibYUVPlanarTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -380,29 +401,29 @@ LibYUVPlanarTest::LibYUVPlanarTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -425,8 +446,8 @@ LibYUVBaseTest::LibYUVBaseTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -436,29 +457,29 @@ LibYUVBaseTest::LibYUVBaseTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -481,8 +502,8 @@ LibYUVCompareTest::LibYUVCompareTest()
   if (repeat) {
     benchmark_iterations_ = atoi(repeat);  // NOLINT
   }
-  if (FLAGS_libyuv_repeat) {
-    benchmark_iterations_ = FLAGS_libyuv_repeat;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_repeat)) {
+    benchmark_iterations_ = LIBYUV_GET_FLAG(FLAGS_libyuv_repeat);
   }
   if (benchmark_iterations_ > 1) {
     benchmark_width_ = 1280;
@@ -492,29 +513,29 @@ LibYUVCompareTest::LibYUVCompareTest()
   if (width) {
     benchmark_width_ = atoi(width);  // NOLINT
   }
-  if (FLAGS_libyuv_width) {
-    benchmark_width_ = FLAGS_libyuv_width;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_width)) {
+    benchmark_width_ = LIBYUV_GET_FLAG(FLAGS_libyuv_width);
   }
   const char* height = getenv("LIBYUV_HEIGHT");
   if (height) {
     benchmark_height_ = atoi(height);  // NOLINT
   }
-  if (FLAGS_libyuv_height) {
-    benchmark_height_ = FLAGS_libyuv_height;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_height)) {
+    benchmark_height_ = LIBYUV_GET_FLAG(FLAGS_libyuv_height);
   }
   const char* cpu_flags = getenv("LIBYUV_FLAGS");
   if (cpu_flags) {
     disable_cpu_flags_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_flags) {
-    disable_cpu_flags_ = FLAGS_libyuv_flags;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_flags)) {
+    disable_cpu_flags_ = LIBYUV_GET_FLAG(FLAGS_libyuv_flags);
   }
   const char* cpu_info = getenv("LIBYUV_CPU_INFO");
   if (cpu_info) {
     benchmark_cpu_info_ = atoi(cpu_flags);  // NOLINT
   }
-  if (FLAGS_libyuv_cpu_info) {
-    benchmark_cpu_info_ = FLAGS_libyuv_cpu_info;
+  if (LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info)) {
+    benchmark_cpu_info_ = LIBYUV_GET_FLAG(FLAGS_libyuv_cpu_info);
   }
   disable_cpu_flags_ = TestCpuEnv(disable_cpu_flags_);
   benchmark_cpu_info_ = TestCpuEnv(benchmark_cpu_info_);
@@ -529,11 +550,8 @@ LibYUVCompareTest::LibYUVCompareTest()
 
 int main(int argc, char** argv) {
   ::testing::InitGoogleTest(&argc, argv);
-#ifdef LIBYUV_USE_GFLAGS
-  // AllowCommandLineParsing allows us to ignore flags passed on to us by
-  // Chromium build bots without having to explicitly disable them.
-  google::AllowCommandLineReparsing();
-  google::ParseCommandLineFlags(&argc, &argv, true);
+#ifdef LIBYUV_USE_ABSL_FLAGS
+  absl::ParseCommandLine(argc, argv);
 #endif
   return RUN_ALL_TESTS();
 }
diff --git a/files/unit_test/unit_test.h b/files/unit_test/unit_test.h
index 87907fa1..0a8df4d2 100644
--- a/files/unit_test/unit_test.h
+++ b/files/unit_test/unit_test.h
@@ -11,7 +11,7 @@
 #ifndef UNIT_TEST_UNIT_TEST_H_  // NOLINT
 #define UNIT_TEST_UNIT_TEST_H_
 
-#ifdef WIN32
+#ifdef _WIN32
 #include <windows.h>
 #else
 #include <sys/resource.h>
@@ -111,10 +111,13 @@ inline int fastrand() {
   return static_cast<int>((fastrand_seed >> 16) & 0xffff);
 }
 
+// ubsan fails if dst is unaligned unless we use uint8
 static inline void MemRandomize(uint8_t* dst, int64_t len) {
   int64_t i;
   for (i = 0; i < len - 1; i += 2) {
-    *reinterpret_cast<uint16_t*>(dst) = fastrand();
+    int r = fastrand();
+    dst[0] = static_cast<uint8_t>(r);
+    dst[1] = static_cast<uint8_t>(r >> 8);
     dst += 2;
   }
   for (; i < len; ++i) {
diff --git a/files/unit_test/video_common_test.cc b/files/unit_test/video_common_test.cc
index a84206a2..36728ea9 100644
--- a/files/unit_test/video_common_test.cc
+++ b/files/unit_test/video_common_test.cc
@@ -29,7 +29,7 @@ static bool TestValidFourCC(uint32_t fourcc, int bpp) {
       !TestValidChar((fourcc >> 24) & 0xff)) {
     return false;
   }
-  if (bpp < 0 || bpp > 32) {
+  if (bpp < 0 || bpp > 64) {
     return false;
   }
   return true;
@@ -65,13 +65,15 @@ TEST_F(LibYUVBaseTest, TestFourCC) {
   EXPECT_TRUE(TestValidFourCC(FOURCC_NV12, FOURCC_BPP_NV12));
   EXPECT_TRUE(TestValidFourCC(FOURCC_YUY2, FOURCC_BPP_YUY2));
   EXPECT_TRUE(TestValidFourCC(FOURCC_UYVY, FOURCC_BPP_UYVY));
-  EXPECT_TRUE(TestValidFourCC(FOURCC_M420, FOURCC_BPP_M420));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_M420, FOURCC_BPP_M420));  // deprecated.
   EXPECT_TRUE(TestValidFourCC(FOURCC_Q420, FOURCC_BPP_Q420));  // deprecated.
   EXPECT_TRUE(TestValidFourCC(FOURCC_ARGB, FOURCC_BPP_ARGB));
   EXPECT_TRUE(TestValidFourCC(FOURCC_BGRA, FOURCC_BPP_BGRA));
   EXPECT_TRUE(TestValidFourCC(FOURCC_ABGR, FOURCC_BPP_ABGR));
   EXPECT_TRUE(TestValidFourCC(FOURCC_AR30, FOURCC_BPP_AR30));
   EXPECT_TRUE(TestValidFourCC(FOURCC_AB30, FOURCC_BPP_AB30));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_AR64, FOURCC_BPP_AR64));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_AB64, FOURCC_BPP_AB64));
   EXPECT_TRUE(TestValidFourCC(FOURCC_24BG, FOURCC_BPP_24BG));
   EXPECT_TRUE(TestValidFourCC(FOURCC_RAW, FOURCC_BPP_RAW));
   EXPECT_TRUE(TestValidFourCC(FOURCC_RGBA, FOURCC_BPP_RGBA));
@@ -81,6 +83,11 @@ TEST_F(LibYUVBaseTest, TestFourCC) {
   EXPECT_TRUE(TestValidFourCC(FOURCC_H420, FOURCC_BPP_H420));
   EXPECT_TRUE(TestValidFourCC(FOURCC_H422, FOURCC_BPP_H422));
   EXPECT_TRUE(TestValidFourCC(FOURCC_H010, FOURCC_BPP_H010));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_H210, FOURCC_BPP_H210));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_I010, FOURCC_BPP_I010));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_I210, FOURCC_BPP_I210));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_P010, FOURCC_BPP_P010));
+  EXPECT_TRUE(TestValidFourCC(FOURCC_P210, FOURCC_BPP_P210));
   EXPECT_TRUE(TestValidFourCC(FOURCC_MJPG, FOURCC_BPP_MJPG));
   EXPECT_TRUE(TestValidFourCC(FOURCC_YV12, FOURCC_BPP_YV12));
   EXPECT_TRUE(TestValidFourCC(FOURCC_YV16, FOURCC_BPP_YV16));