sws: Support error diffusion dither for mono output

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

Changelog

@@ -4,6 +4,7 @@ releases are sorted from youngest to oldest.
 version <next>:
 - VDPAU hardware acceleration through normal hwaccel
 - SRTP support
+- Error diffusion dither in Swscale
 
 
 version 1.1:

libswscale/output.c

@@ -333,6 +333,7 @@ yuv2mono_X_c_template(SwsContext *c, const int16_t *lumFilter,
     const uint8_t * const d128=dither_8x8_220[y&7];
     int i;
     unsigned acc = 0;
+    int err = 0;
 
     for (i = 0; i < dstW; i += 2) {
         int j;
@@ -349,12 +350,25 @@ yuv2mono_X_c_template(SwsContext *c, const int16_t *lumFilter,
             Y1 = av_clip_uint8(Y1);
             Y2 = av_clip_uint8(Y2);
         }
-        accumulate_bit(acc, Y1 + d128[(i + 0) & 7]);
-        accumulate_bit(acc, Y2 + d128[(i + 1) & 7]);
+        if (c->flags & SWS_ERROR_DIFFUSION) {
+            Y1 += (7*err + 1*c->dither_error[0][i] + 5*c->dither_error[0][i+1] + 3*c->dither_error[0][i+2])>>4;
+            c->dither_error[0][i] = err;
+            acc = 2*acc + (Y1 >= 128);
+            Y1 -= 220*(acc&1);
+
+            err = Y2 + ((7*Y1 + 1*c->dither_error[0][i+1] + 5*c->dither_error[0][i+2] + 3*c->dither_error[0][i+3])>>4);
+            c->dither_error[0][i+1] = Y1;
+            acc = 2*acc + (err >= 128);
+            err -= 220*(acc&1);
+        } else {
+            accumulate_bit(acc, Y1 + d128[(i + 0) & 7]);
+            accumulate_bit(acc, Y2 + d128[(i + 1) & 7]);
+        }
         if ((i & 7) == 6) {
             output_pixel(*dest++, acc);
         }
     }
+    c->dither_error[0][i] = err;
 
     if (i & 6) {
         output_pixel(*dest, acc);
@@ -373,6 +387,29 @@ yuv2mono_2_c_template(SwsContext *c, const int16_t *buf[2],
     int  yalpha1 = 4096 - yalpha;
     int i;
 
+    if (c->flags & SWS_ERROR_DIFFUSION) {
+        int err = 0;
+        int acc = 0;
+        for (i = 0; i < dstW; i +=2) {
+            int Y;
+
+            Y = (buf0[i + 0] * yalpha1 + buf1[i + 0] * yalpha) >> 19;
+            Y += (7*err + 1*c->dither_error[0][i] + 5*c->dither_error[0][i+1] + 3*c->dither_error[0][i+2])>>4;
+            c->dither_error[0][i] = err;
+            acc = 2*acc + (Y >= 128);
+            Y -= 220*(acc&1);
+
+            err = (buf0[i + 1] * yalpha1 + buf1[i + 1] * yalpha) >> 19;
+            err += (7*Y + 1*c->dither_error[0][i+1] + 5*c->dither_error[0][i+2] + 3*c->dither_error[0][i+3])>>4;
+            c->dither_error[0][i+1] = Y;
+            acc = 2*acc + (err >= 128);
+            err -= 220*(acc&1);
+
+            if ((i & 7) == 6)
+                output_pixel(*dest++, acc);
+        }
+        c->dither_error[0][i] = err;
+    } else {
     for (i = 0; i < dstW; i += 8) {
         int Y, acc = 0;
 
@@ -395,6 +432,7 @@ yuv2mono_2_c_template(SwsContext *c, const int16_t *buf[2],
 
         output_pixel(*dest++, acc);
     }
+    }
 }
 
 static av_always_inline void
@@ -406,9 +444,31 @@ yuv2mono_1_c_template(SwsContext *c, const int16_t *buf0,
     const uint8_t * const d128 = dither_8x8_220[y & 7];
     int i;
 
+    if (c->flags & SWS_ERROR_DIFFUSION) {
+        int err = 0;
+        int acc = 0;
+        for (i = 0; i < dstW; i +=2) {
+            int Y;
+
+            Y = ((buf0[i + 0] + 64) >> 7);
+            Y += (7*err + 1*c->dither_error[0][i] + 5*c->dither_error[0][i+1] + 3*c->dither_error[0][i+2])>>4;
+            c->dither_error[0][i] = err;
+            acc = 2*acc + (Y >= 128);
+            Y -= 220*(acc&1);
+
+            err = ((buf0[i + 1] + 64) >> 7);
+            err += (7*Y + 1*c->dither_error[0][i+1] + 5*c->dither_error[0][i+2] + 3*c->dither_error[0][i+3])>>4;
+            c->dither_error[0][i+1] = Y;
+            acc = 2*acc + (err >= 128);
+            err -= 220*(acc&1);
+
+            if ((i & 7) == 6)
+                output_pixel(*dest++, acc);
+        }
+        c->dither_error[0][i] = err;
+    } else {
     for (i = 0; i < dstW; i += 8) {
         int acc = 0;
-
         accumulate_bit(acc, ((buf0[i + 0] + 64) >> 7) + d128[0]);
         accumulate_bit(acc, ((buf0[i + 1] + 64) >> 7) + d128[1]);
         accumulate_bit(acc, ((buf0[i + 2] + 64) >> 7) + d128[2]);
@@ -420,6 +480,7 @@ yuv2mono_1_c_template(SwsContext *c, const int16_t *buf0,
 
         output_pixel(*dest++, acc);
     }
+    }
 }
 
 #undef output_pixel

libswscale/swscale.c

@@ -910,6 +910,11 @@ int attribute_align_arg sws_scale(struct SwsContext *c,
         src2[0] = base;
     }
 
+    if (!srcSliceY && (c->flags & SWS_ERROR_DIFFUSION))
+        for (i = 0; i < 4; i++)
+            memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2));
+
+
     // copy strides, so they can safely be modified
     if (c->sliceDir == 1) {
         // slices go from top to bottom

libswscale/swscale.h

@@ -80,6 +80,7 @@ const char *swscale_license(void);
 //input subsampling info
 #define SWS_FULL_CHR_H_INP    0x4000
 #define SWS_DIRECT_BGR        0x8000
+#define SWS_ERROR_DIFFUSION   0x20000
 #define SWS_ACCURATE_RND      0x40000
 #define SWS_BITEXACT          0x80000
 

libswscale/swscale_internal.h

@@ -327,6 +327,8 @@ typedef struct SwsContext {
     int table_gV[256 + 2*YUVRGB_TABLE_HEADROOM];
     uint8_t *table_bU[256 + 2*YUVRGB_TABLE_HEADROOM];
 
+    int *dither_error[4];
+
     //Colorspace stuff
     int contrast, brightness, saturation;    // for sws_getColorspaceDetails
     int srcColorspaceTable[4];

libswscale/swscale_unscaled.c

@@ -936,7 +936,7 @@ void ff_get_unscaled_swscale(SwsContext *c)
     /* yuv2bgr */
     if ((srcFormat == AV_PIX_FMT_YUV420P || srcFormat == AV_PIX_FMT_YUV422P ||
          srcFormat == AV_PIX_FMT_YUVA420P) && isAnyRGB(dstFormat) &&
-        !(flags & SWS_ACCURATE_RND) && !(dstH & 1)) {
+        !(flags & (SWS_ACCURATE_RND|SWS_ERROR_DIFFUSION)) && !(dstH & 1)) {
         c->swScale = ff_yuv2rgb_get_func_ptr(c);
     }
 

libswscale/utils.c

@@ -1247,6 +1247,9 @@ av_cold int sws_init_context(SwsContext *c, SwsFilter *srcFilter,
                              c->vChrFilterPos[chrI];
     }
 
+    for (i = 0; i < 4; i++)
+        FF_ALLOCZ_OR_GOTO(c, c->dither_error[i], (c->dstW+2) * sizeof(int), fail);
+
     /* Allocate pixbufs (we use dynamic allocation because otherwise we would
      * need to allocate several megabytes to handle all possible cases) */
     FF_ALLOC_OR_GOTO(c, c->lumPixBuf,  c->vLumBufSize * 3 * sizeof(int16_t *), fail);
@@ -1740,6 +1743,9 @@ void sws_freeContext(SwsContext *c)
         av_freep(&c->alpPixBuf);
     }
 
+    for (i = 0; i < 4; i++)
+        av_freep(&c->dither_error[i]);
+
     av_freep(&c->vLumFilter);
     av_freep(&c->vChrFilter);
     av_freep(&c->hLumFilter);