libswscale: Extend the unaccelerated path of the unscaled yuv2rgb special

            converter with support for rgb444 output format.
Patch by Janusz Krzysztofik jkrzyszt chez tis icnet pl


git-svn-id: file:///var/local/repositories/mplayer/trunk/libswscale@30841 b3059339-0415-0410-9bf9-f77b7e298cf2

swscale.c

@@ -27,7 +27,7 @@
   {BGR,RGB}{1,4,8,15,16} support dithering
 
   unscaled special converters (YV12=I420=IYUV, Y800=Y8)
-  YV12 -> {BGR,RGB}{1,4,8,15,16,24,32}
+  YV12 -> {BGR,RGB}{1,4,8,12,15,16,24,32}
   x -> x
   YUV9 -> YV12
   YUV9/YV12 -> Y800
@@ -198,6 +198,13 @@ DECLARE_ALIGNED(8, static const uint8_t, dither_2x2_8)[2][8]={
 {  0,   4,   0,   4,   0,   4,   0,   4, },
 };
 
+DECLARE_ALIGNED(8, const uint8_t, dither_4x4_16)[4][8]={
+{  8,   4,  11,   7,   8,   4,  11,   7, },
+{  2,  14,   1,  13,   2,  14,   1,  13, },
+{ 10,   6,   9,   5,  10,   6,   9,   5, },
+{  0,  12,   3,  15,   0,  12,   3,  15, },
+};
+
 DECLARE_ALIGNED(8, const uint8_t, dither_8x8_32)[8][8]={
 { 17,   9,  23,  15,  16,   8,  22,  14, },
 {  5,  29,   3,  27,   4,  28,   2,  26, },

swscale_internal.h

@@ -395,6 +395,7 @@ const char *sws_format_name(enum PixelFormat format);
         || (x)==PIX_FMT_RGB565LE    \
         || (x)==PIX_FMT_RGB555BE    \
         || (x)==PIX_FMT_RGB555LE    \
+        || (x)==PIX_FMT_RGB444      \
         || (x)==PIX_FMT_RGB8        \
         || (x)==PIX_FMT_RGB4        \
         || (x)==PIX_FMT_RGB4_BYTE   \
@@ -409,6 +410,7 @@ const char *sws_format_name(enum PixelFormat format);
         || (x)==PIX_FMT_BGR565LE    \
         || (x)==PIX_FMT_BGR555BE    \
         || (x)==PIX_FMT_BGR555LE    \
+        || (x)==PIX_FMT_BGR444      \
         || (x)==PIX_FMT_BGR8        \
         || (x)==PIX_FMT_BGR4        \
         || (x)==PIX_FMT_BGR4_BYTE   \

yuv2rgb.c

@@ -35,6 +35,7 @@
 #include "libavutil/x86_cpu.h"
 #include "libavutil/bswap.h"
 
+extern const uint8_t dither_4x4_16[4][8];
 extern const uint8_t dither_8x8_32[8][8];
 extern const uint8_t dither_8x8_73[8][8];
 extern const uint8_t dither_8x8_220[8][8];
@@ -351,6 +352,32 @@ YUV2RGBFUNC(yuv2rgb_c_8, uint8_t, 0)
 CLOSEYUV2RGBFUNC(8)
 #endif
 
+// r, g, b, dst_1, dst_2
+YUV2RGBFUNC(yuv2rgb_c_12_ordered_dither, uint16_t, 0)
+    const uint8_t *d16 = dither_4x4_16[y&3];
+#define PUTRGB12(dst,src,i,o)                                   \
+    Y = src[2*i];                                               \
+    dst[2*i]   = r[Y+d16[0+o]] + g[Y+d16[0+o]] + b[Y+d16[0+o]]; \
+    Y = src[2*i+1];                                             \
+    dst[2*i+1] = r[Y+d16[1+o]] + g[Y+d16[1+o]] + b[Y+d16[1+o]];
+
+    LOADCHROMA(0);
+    PUTRGB12(dst_1,py_1,0,0);
+    PUTRGB12(dst_2,py_2,0,0+8);
+
+    LOADCHROMA(1);
+    PUTRGB12(dst_2,py_2,1,2+8);
+    PUTRGB12(dst_1,py_1,1,2);
+
+    LOADCHROMA(2);
+    PUTRGB12(dst_1,py_1,2,4);
+    PUTRGB12(dst_2,py_2,2,4+8);
+
+    LOADCHROMA(3);
+    PUTRGB12(dst_2,py_2,3,6+8);
+    PUTRGB12(dst_1,py_1,3,6);
+CLOSEYUV2RGBFUNC(8)
+
 // r, g, b, dst_1, dst_2
 YUV2RGBFUNC(yuv2rgb_c_8_ordered_dither, uint8_t, 0)
     const uint8_t *d32 = dither_8x8_32[y&7];
@@ -553,6 +580,8 @@ SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c)
     case PIX_FMT_BGR565:
     case PIX_FMT_RGB555:
     case PIX_FMT_BGR555:     return yuv2rgb_c_16;
+    case PIX_FMT_RGB444:
+    case PIX_FMT_BGR444:     return yuv2rgb_c_12_ordered_dither;
     case PIX_FMT_RGB8:
     case PIX_FMT_BGR8:       return yuv2rgb_c_8_ordered_dither;
     case PIX_FMT_RGB4:
@@ -601,6 +630,7 @@ av_cold int ff_yuv2rgb_c_init_tables(SwsContext *c, const int inv_table[4], int
                         || c->dstFormat==PIX_FMT_RGB565LE
                         || c->dstFormat==PIX_FMT_RGB555BE
                         || c->dstFormat==PIX_FMT_RGB555LE
+                        || c->dstFormat==PIX_FMT_RGB444
                         || c->dstFormat==PIX_FMT_RGB8
                         || c->dstFormat==PIX_FMT_RGB4
                         || c->dstFormat==PIX_FMT_RGB4_BYTE
@@ -701,6 +731,25 @@ av_cold int ff_yuv2rgb_c_init_tables(SwsContext *c, const int inv_table[4], int
         fill_table(c->table_bU, 1, cbu, y_table + yoffs + 2048);
         fill_gv_table(c->table_gV, 1, cgv);
         break;
+    case 12:
+        rbase = isRgb ? 8 : 0;
+        gbase = 4;
+        bbase = isRgb ? 0 : 8;
+        c->yuvTable = av_malloc(1024*3*2);
+        y_table16 = c->yuvTable;
+        yb = -(384<<16) - oy;
+        for (i = 0; i < 1024; i++) {
+            uint8_t yval = av_clip_uint8((yb + 0x8000) >> 16);
+            y_table16[i     ] = (yval >> 4)          << rbase;
+            y_table16[i+1024] = (yval >> 4) << gbase;
+            y_table16[i+2048] = (yval >> 4)          << bbase;
+            yb += cy;
+        }
+        fill_table(c->table_rV, 2, crv, y_table16 + yoffs);
+        fill_table(c->table_gU, 2, cgu, y_table16 + yoffs + 1024);
+        fill_table(c->table_bU, 2, cbu, y_table16 + yoffs + 2048);
+        fill_gv_table(c->table_gV, 2, cgv);
+        break;
     case 15:
     case 16:
         rbase = isRgb ? bpp - 5 : 0;