fix for yuv to rgb on arm64.

fill in aarch64 yuv constants to match how the code expects them.

TBR=harryjin@google.com
BUG=libyuv:502

Review URL: https://codereview.chromium.org/1396253004 .

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1506
+Version: 1507
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -417,18 +417,25 @@ typedef uint32 ulvec32[8];
 typedef uint8 ulvec8[32];
 #endif
 
-#if defined(__arm__) || defined(__aarch64__)
-
-// This struct is for Arm color conversion.
+#if defined(__aarch64__)
+// This struct is for Arm64 color conversion.
+struct YuvConstants {
+  uvec16 kUVToRB;
+  uvec16 kUVToRB2;
+  uvec16 kUVToG;
+  uvec16 kUVToG2;
+  vec16 kUVBiasBGR;
+  vec32 kYToRgb;
+};
+#elif defined(__arm__)
+// This struct is for ArmV7 color conversion.
 struct YuvConstants {
   uvec8 kUVToRB;
   uvec8 kUVToG;
   vec16 kUVBiasBGR;
   vec32 kYToRgb;
 };
-
 #else
-
 // This struct is for Intel color conversion.
 struct YuvConstants {
   lvec8 kUVToB;

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1506
+#define LIBYUV_VERSION 1507
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_common.cc

@@ -1017,13 +1017,17 @@ void J400ToARGBRow_C(const uint8* src_y, uint8* dst_argb, int width) {
 // BT.601 constants for YUV to RGB.
 // TODO(fbarchard): Unify these structures to be platform independent.
 // TODO(fbarchard): Generate SIMD structures from float matrix.
+
+// BT601 constants for YUV to RGB.
 #if defined(__aarch64__)
 const YuvConstants SIMD_ALIGNED(kYuvIConstants) = {
-  { -UB, 0, -UB, 0, -UB, 0, -UB, 0, -VR, 0, -VR, 0, -VR, 0, -VR, 0 },
-  { UG, 0, UG, 0, UG, 0, UG, 0, VG, 0, VG, 0, VG, 0, VG, 0 },
+  { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR },
+  { -UB, -VR, -UB, -VR, -UB, -VR, -UB, -VR },
+  { UG, VG, UG, VG, UG, VG, UG, VG },
+  { UG, VG, UG, VG, UG, VG, UG, VG },
   { BB, BG, BR, 0, 0, 0, 0, 0 },
   { 0x0101 * YG, 0, 0, 0 }
- };
+};
 
 #elif defined(__arm__)
 const YuvConstants SIMD_ALIGNED(kYuvIConstants) = {
@@ -1072,8 +1076,8 @@ static __inline void YuvPixel(uint8 y, uint8 u, uint8 v,
 #if defined(__aarch64__)
   int UB = -yuvconstants->kUVToRB[0];
   int UG = yuvconstants->kUVToG[0];
-  int VG = yuvconstants->kUVToG[8];
-  int VR = -yuvconstants->kUVToRB[8];
+  int VG = yuvconstants->kUVToG[1];
+  int VR = -yuvconstants->kUVToRB[1];
   int BB = yuvconstants->kUVBiasBGR[0];
   int BG = yuvconstants->kUVBiasBGR[1];
   int BR = yuvconstants->kUVBiasBGR[2];
@@ -1126,11 +1130,13 @@ static __inline void YuvPixel(uint8 y, uint8 u, uint8 v,
 // JPEG constants for YUV to RGB.
 #if defined(__aarch64__)
 const YuvConstants SIMD_ALIGNED(kYuvJConstants) = {
-  { -UBJ, 0, -UBJ, 0, -UBJ, 0, -UBJ, 0, -VRJ, 0, -VRJ, 0, -VRJ, 0, -VRJ, 0 },
-  { UGJ, 0, UGJ, 0, UGJ, 0, UGJ, 0, VGJ, 0, VGJ, 0, VGJ, 0, VGJ, 0 },
+  { -UBJ, -VRJ, -UBJ, -VRJ, -UBJ, -VRJ, -UBJ, -VRJ },
+  { -UBJ, -VRJ, -UBJ, -VRJ, -UBJ, -VRJ, -UBJ, -VRJ },
+  { UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, UGJ, VGJ },
+  { UGJ, VGJ, UGJ, VGJ, UGJ, VGJ, UGJ, VGJ },
   { BBJ, BGJ, BRJ, 0, 0, 0, 0, 0 },
   { 0x0101 * YGJ, 0, 0, 0 }
- };
+};
 
 #elif defined(__arm__)
 const YuvConstants SIMD_ALIGNED(kYuvJConstants) = {
@@ -1194,11 +1200,13 @@ const YuvConstants SIMD_ALIGNED(kYuvJConstants) = {
 // BT.709 constants for YUV to RGB.
 #if defined(__aarch64__)
 const YuvConstants SIMD_ALIGNED(kYuvHConstants) = {
-  { -UBH, 0, -UBH, 0, -UBH, 0, -UBH, 0, -VRH, 0, -VRH, 0, -VRH, 0, -VRH, 0 },
-  { UGH, 0, UGH, 0, UGH, 0, UGH, 0, VGH, 0, VGH, 0, VGH, 0, VGH, 0 },
+  { -UBH, -VRH, -UBH, -VRH, -UBH, -VRH, -UBH, -VRH },
+  { -UBH, -VRH, -UBH, -VRH, -UBH, -VRH, -UBH, -VRH },
+  { UGH, VGH, UGH, VGH, UGH, VGH, UGH, VGH },
+  { UGH, VGH, UGH, VGH, UGH, VGH, UGH, VGH },
   { BBH, BGH, BRH, 0, 0, 0, 0, 0 },
   { 0x0101 * YGH, 0, 0, 0 }
- };
+};
 
 #elif defined(__arm__)
 const YuvConstants SIMD_ALIGNED(kYuvHConstants) = {

source/row_neon64.cc

@@ -96,8 +96,8 @@ extern "C" {
     "ld1r       {v25.8h}, [%[kUVBiasBGR]], #2  \n"                             \
     "ld1r       {v26.8h}, [%[kUVBiasBGR]]      \n"                             \
     "ld1r       {v31.4s}, [%[kYToRgb]]         \n"                             \
-    "ld1        {v27.8h, v28.8h}, [%[kUVToRB]] \n"                             \
-    "ld1        {v29.8h, v30.8h}, [%[kUVToG]]  \n"
+    "ld2        {v27.8h, v28.8h}, [%[kUVToRB]] \n"                             \
+    "ld2        {v29.8h, v30.8h}, [%[kUVToG]]  \n"
 
 #define YUVTORGB(vR, vG, vB)                                                   \
     "uxtl       v0.8h, v0.8b                   \n" /* Extract Y    */          \