swscale: error dithering for 16/9/10-bit to 8-bit.

Based on a somewhat similar idea in FFmpeg's swscale copy.

1 files changed, 36 insertions, 13 deletions

diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index db4d231e13..dd9f4a108f 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -182,6 +182,18 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
 { 77,  23,  60,  15,  72,  21,  56,  14, },
 };
 #endif
+DECLARE_ALIGNED(8, const uint8_t, dither_8x8_128)[8][8] = {
+{  36, 68, 60, 92, 34, 66, 58, 90,},
+{ 100,  4,124, 28, 98,  2,122, 26,},
+{  52, 84, 44, 76, 50, 82, 42, 74,},
+{ 116, 20,108, 12,114, 18,106, 10,},
+{  32, 64, 56, 88, 38, 70, 62, 94,},
+{  96,  0,120, 24,102,  6,126, 30,},
+{  48, 80, 40, 72, 54, 86, 46, 78,},
+{ 112, 16,104,  8,118, 22,110, 14,},
+};
+DECLARE_ALIGNED(8, const uint8_t, ff_sws_pb_64)[8] =
+{  64, 64, 64, 64, 64, 64, 64, 64 };
 
 static av_always_inline void
 yuv2yuvX16_c_template(const int16_t *lumFilter, const int32_t **lumSrc,
@@ -285,10 +297,11 @@ static void yuv2yuvX_c(SwsContext *c, const int16_t *lumFilter,
     uint8_t *yDest = dest[0], *uDest = dest[1], *vDest = dest[2],
             *aDest = CONFIG_SWSCALE_ALPHA ? dest[3] : NULL;
     int i;
+    const uint8_t *lumDither = c->lumDither8, *chrDither = c->chrDither8;
 
     //FIXME Optimize (just quickly written not optimized..)
     for (i=0; i<dstW; i++) {
-        int val=1<<18;
+        int val = lumDither[i & 7] << 12;
         int j;
         for (j=0; j<lumFilterSize; j++)
             val += lumSrc[j][i] * lumFilter[j];
@@ -298,8 +311,8 @@ static void yuv2yuvX_c(SwsContext *c, const int16_t *lumFilter,
 
     if (uDest)
         for (i=0; i<chrDstW; i++) {
-            int u=1<<18;
-            int v=1<<18;
+            int u = chrDither[i & 7] << 12;
+            int v = chrDither[(i + 3) & 7] << 12;
             int j;
             for (j=0; j<chrFilterSize; j++) {
                 u += chrUSrc[j][i] * chrFilter[j];
@@ -312,7 +325,7 @@ static void yuv2yuvX_c(SwsContext *c, const int16_t *lumFilter,
 
     if (CONFIG_SWSCALE_ALPHA && aDest)
         for (i=0; i<dstW; i++) {
-            int val=1<<18;
+            int val = lumDither[i & 7] << 12;
             int j;
             for (j=0; j<lumFilterSize; j++)
                 val += alpSrc[j][i] * lumFilter[j];
@@ -329,23 +342,24 @@ static void yuv2yuv1_c(SwsContext *c, const int16_t *lumSrc,
     uint8_t *yDest = dest[0], *uDest = dest[1], *vDest = dest[2],
             *aDest = CONFIG_SWSCALE_ALPHA ? dest[3] : NULL;
     int i;
+    const uint8_t *lumDither = c->lumDither8, *chrDither = c->chrDither8;
 
     for (i=0; i<dstW; i++) {
-        int val= (lumSrc[i]+64)>>7;
+        int val = (lumSrc[i]+  lumDither[i & 7]) >> 7;
         yDest[i]= av_clip_uint8(val);
     }
 
     if (uDest)
         for (i=0; i<chrDstW; i++) {
-            int u=(chrUSrc[i]+64)>>7;
-            int v=(chrVSrc[i]+64)>>7;
+            int u = (chrUSrc[i] + chrDither[i & 7])       >> 7;
+            int v = (chrVSrc[i] + chrDither[(i + 3) & 7]) >> 7;
             uDest[i]= av_clip_uint8(u);
             vDest[i]= av_clip_uint8(v);
         }
 
     if (CONFIG_SWSCALE_ALPHA && aDest)
         for (i=0; i<dstW; i++) {
-            int val= (alpSrc[i]+64)>>7;
+            int val = (alpSrc[i] + lumDither[i & 7]) >> 7;
             aDest[i]= av_clip_uint8(val);
         }
 }
@@ -359,11 +373,12 @@ static void yuv2nv12X_c(SwsContext *c, const int16_t *lumFilter,
 {
     uint8_t *yDest = dest[0], *uDest = dest[1];
     enum PixelFormat dstFormat = c->dstFormat;
+    const uint8_t *lumDither = c->lumDither8, *chrDither = c->chrDither8;
 
     //FIXME Optimize (just quickly written not optimized..)
     int i;
     for (i=0; i<dstW; i++) {
-        int val=1<<18;
+        int val = lumDither[i & 7] << 12;
         int j;
         for (j=0; j<lumFilterSize; j++)
             val += lumSrc[j][i] * lumFilter[j];
@@ -376,8 +391,8 @@ static void yuv2nv12X_c(SwsContext *c, const int16_t *lumFilter,
 
     if (dstFormat == PIX_FMT_NV12)
         for (i=0; i<chrDstW; i++) {
-            int u=1<<18;
-            int v=1<<18;
+            int u = chrDither[i & 7] << 12;
+            int v = chrDither[(i + 3) & 7] << 12;
             int j;
             for (j=0; j<chrFilterSize; j++) {
                 u += chrUSrc[j][i] * chrFilter[j];
@@ -389,8 +404,8 @@ static void yuv2nv12X_c(SwsContext *c, const int16_t *lumFilter,
         }
     else
         for (i=0; i<chrDstW; i++) {
-            int u=1<<18;
-            int v=1<<18;
+            int u = chrDither[i & 7] << 12;
+            int v = chrDither[(i + 3) & 7] << 12;
             int j;
             for (j=0; j<chrFilterSize; j++) {
                 u += chrUSrc[j][i] * chrFilter[j];
@@ -2352,6 +2367,7 @@ static int swScale(SwsContext *c, const uint8_t* src[],
     yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
     yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
     yuv2packedX_fn yuv2packedX = c->yuv2packedX;
+    int should_dither = is9_OR_10BPS(c->srcFormat) || is16BPS(c->srcFormat);
 
     /* vars which will change and which we need to store back in the context */
     int dstY= c->dstY;
@@ -2401,6 +2417,9 @@ static int swScale(SwsContext *c, const uint8_t* src[],
         lastInChrBuf= -1;
     }
 
+    if (!should_dither) {
+        c->chrDither8 = c->lumDither8 = ff_sws_pb_64;
+    }
     lastDstY= dstY;
 
     for (;dstY < dstH; dstY++) {
@@ -2490,6 +2509,10 @@ static int swScale(SwsContext *c, const uint8_t* src[],
 #if HAVE_MMX
         updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex, lastInLumBuf, lastInChrBuf);
 #endif
+        if (should_dither) {
+            c->chrDither8 = dither_8x8_128[chrDstY & 7];
+            c->lumDither8 = dither_8x8_128[dstY & 7];
+        }
         if (dstY >= dstH-2) {
             // hmm looks like we can't use MMX here without overwriting this array's tail
             find_c_packed_planar_out_funcs(c, &yuv2yuv1, &yuv2yuvX,