swscale: fix 16-bit scaling when output is 8-bits.

We would use the second half of the U plane buffer, rather than the
V plane buffer, to output the V plane pixels.

libswscale/utils.c

@@ -1052,7 +1052,7 @@ int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
     for (i=0; i<c->vChrBufSize; i++) {
         FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+1, fail);
         c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
-        c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride>>1);
+        c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1);
     }
     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
         for (i=0; i<c->vLumBufSize; i++) {

libswscale/x86/swscale_template.c

@@ -83,7 +83,7 @@ static void RENAME(yuv2yuvX)(SwsContext *c, const int16_t *lumFilter,
             *aDest = CONFIG_SWSCALE_ALPHA ? dest[3] : NULL;
 
     if (uDest) {
-        x86_reg uv_off = c->uv_off;
+        x86_reg uv_off = c->uv_offx2 >> 1;
         for(i=0; i<8; i++) c->dither16[i] = chrDither[i]>>4;
         YSCALEYUV2YV12X(CHR_MMX_FILTER_OFFSET, uDest, chrDstW, 0)
         for(i=0; i<8; i++) c->dither16[i] = chrDither[(i+3)&7]>>4;
@@ -170,7 +170,7 @@ static void RENAME(yuv2yuvX_ar)(SwsContext *c, const int16_t *lumFilter,
             *aDest = CONFIG_SWSCALE_ALPHA ? dest[3] : NULL;
 
     if (uDest) {
-        x86_reg uv_off = c->uv_off;
+        x86_reg uv_off = c->uv_offx2 >> 1;
         for(i=0; i<8; i++) c->dither32[i] = chrDither[i]<<12;
         YSCALEYUV2YV12X_ACCURATE(CHR_MMX_FILTER_OFFSET, uDest, chrDstW, 0)
         for(i=0; i<8; i++) c->dither32[i] = chrDither[(i+3)&7]<<12;
@@ -485,7 +485,7 @@ static void RENAME(yuv2rgb32_X_ar)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
         YSCALEYUV2PACKEDX_ACCURATE
@@ -518,7 +518,7 @@ static void RENAME(yuv2rgb32_X)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
         YSCALEYUV2PACKEDX
@@ -575,7 +575,7 @@ static void RENAME(yuv2rgb565_X_ar)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX_ACCURATE
     YSCALEYUV2RGBX
@@ -599,7 +599,7 @@ static void RENAME(yuv2rgb565_X)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX
     YSCALEYUV2RGBX
@@ -652,7 +652,7 @@ static void RENAME(yuv2rgb555_X_ar)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX_ACCURATE
     YSCALEYUV2RGBX
@@ -676,7 +676,7 @@ static void RENAME(yuv2rgb555_X)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX
     YSCALEYUV2RGBX
@@ -809,7 +809,7 @@ static void RENAME(yuv2bgr24_X_ar)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX_ACCURATE
     YSCALEYUV2RGBX
@@ -833,7 +833,7 @@ static void RENAME(yuv2bgr24_X)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX
     YSCALEYUV2RGBX
@@ -874,7 +874,7 @@ static void RENAME(yuv2yuyv422_X_ar)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX_ACCURATE
     /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */
@@ -895,7 +895,7 @@ static void RENAME(yuv2yuyv422_X)(SwsContext *c, const int16_t *lumFilter,
 {
     x86_reg dummy=0;
     x86_reg dstW_reg = dstW;
-    x86_reg uv_off = c->uv_off << 1;
+    x86_reg uv_off = c->uv_offx2;
 
     YSCALEYUV2PACKEDX
     /* mm2=B, %%mm4=G, %%mm5=R, %%mm7=0 */