]> git.sesse.net Git - ffmpeg/blobdiff - libavcodec/imgconvert.c
looks better IMHO
[ffmpeg] / libavcodec / imgconvert.c
index e120fc668d0f61357f23f364f671bae264a8a8d7..cfb21a4776d67c6937119a4497d7761e730ff1e2 100644 (file)
 /*
  * Misc image convertion routines
- * Copyright (c) 2001 Gerard Lantau.
+ * Copyright (c) 2001, 2002, 2003 Fabrice Bellard.
  *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
  *
- * This program is distributed in the hope that it will be useful,
+ * This library is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
  *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  */
+
+/**
+ * @file imgconvert.c
+ * Misc image convertion routines.
+ */
+
+
 #include "avcodec.h"
 #include "dsputil.h"
 
 #ifdef USE_FASTMEMCPY
 #include "fastmemcpy.h"
 #endif
+
+#ifdef HAVE_MMX
+#include "i386/mmx.h"
+#endif
+
+typedef struct PixFmtInfo {
+    const char *name;
+    uint8_t nb_components;     /* number of components in AVPicture array  */
+    uint8_t is_yuv : 1;    /* true if YUV instead of RGB color space */
+    uint8_t is_packed : 1; /* true if multiple components in same word */
+    uint8_t is_paletted : 1; /* true if paletted */
+    uint8_t is_alpha : 1;    /* true if alpha can be specified */
+    uint8_t is_gray : 1;     /* true if gray or monochrome format */
+    uint8_t x_chroma_shift; /* X chroma subsampling factor is 2 ^ shift */
+    uint8_t y_chroma_shift; /* Y chroma subsampling factor is 2 ^ shift */
+} PixFmtInfo;
+
+/* this table gives more information about formats */
+static PixFmtInfo pix_fmt_info[PIX_FMT_NB] = {
+    /* YUV formats */
+    [PIX_FMT_YUV420P] = {
+        .name = "yuv420p",
+        .nb_components = 3, .is_yuv = 1,
+        .x_chroma_shift = 1, .y_chroma_shift = 1, 
+    },
+    [PIX_FMT_YUV422P] = {
+        .name = "yuv422p",
+        .nb_components = 3, .is_yuv = 1,
+        .x_chroma_shift = 1, .y_chroma_shift = 0, 
+    },
+    [PIX_FMT_YUV444P] = {
+        .name = "yuv444p",
+        .nb_components = 3, .is_yuv = 1,
+        .x_chroma_shift = 0, .y_chroma_shift = 0, 
+    },
+    [PIX_FMT_YUV422] = {
+        .name = "yuv422",
+        .nb_components = 1, .is_yuv = 1, .is_packed = 1,
+        .x_chroma_shift = 1, .y_chroma_shift = 0,
+    },
+    [PIX_FMT_YUV410P] = {
+        .name = "yuv410p",
+        .nb_components = 3, .is_yuv = 1,
+        .x_chroma_shift = 2, .y_chroma_shift = 2,
+    },
+    [PIX_FMT_YUV411P] = {
+        .name = "yuv411p",
+        .nb_components = 3, .is_yuv = 1,
+        .x_chroma_shift = 2, .y_chroma_shift = 0,
+    },
+
+    /* RGB formats */
+    [PIX_FMT_RGB24] = {
+        .name = "rgb24",
+        .nb_components = 1, .is_packed = 1,
+    },
+    [PIX_FMT_BGR24] = {
+        .name = "bgr24",
+        .nb_components = 1, .is_packed = 1,
+    },
+    [PIX_FMT_RGBA32] = {
+        .name = "rgba32",
+        .nb_components = 1, .is_packed = 1, .is_alpha = 1,
+    },
+    [PIX_FMT_RGB565] = {
+        .name = "rgb565",
+        .nb_components = 1, .is_packed = 1,
+    },
+    [PIX_FMT_RGB555] = {
+        .name = "rgb555",
+        .nb_components = 1, .is_packed = 1, .is_alpha = 1,
+    },
+
+    /* gray / mono formats */
+    [PIX_FMT_GRAY8] = {
+        .name = "gray",
+        .nb_components = 1, .is_gray = 1,
+    },
+    [PIX_FMT_MONOWHITE] = {
+        .name = "monow",
+        .nb_components = 1, .is_packed = 1, .is_gray = 1,
+    },
+    [PIX_FMT_MONOBLACK] = {
+        .name = "monob",
+        .nb_components = 1, .is_packed = 1, .is_gray = 1,
+    },
+
+    /* paletted formats */
+    [PIX_FMT_PAL8] = {
+        .name = "pal8",
+        .nb_components = 1, .is_packed = 1, .is_paletted = 1,
+    },
+};
+
+void avcodec_get_chroma_sub_sample(int pix_fmt, int *h_shift, int *v_shift)
+{
+    if (pix_fmt_info[pix_fmt].is_yuv) {
+        *h_shift = pix_fmt_info[pix_fmt].x_chroma_shift;
+        *v_shift = pix_fmt_info[pix_fmt].y_chroma_shift;
+    } else {
+        *h_shift=0;
+        *v_shift=0;
+    }
+}
+
+const char *avcodec_get_pix_fmt_name(int pix_fmt)
+{
+    if (pix_fmt < 0 || pix_fmt >= PIX_FMT_NB)
+        return "???";
+    else
+        return pix_fmt_info[pix_fmt].name;
+}
+
+/* Picture field are filled with 'ptr' addresses. Also return size */
+int avpicture_fill(AVPicture *picture, uint8_t *ptr,
+                   int pix_fmt, int width, int height)
+{
+    int size, w2, h2, size2;
+    PixFmtInfo *pinfo;
+    
+    pinfo = &pix_fmt_info[pix_fmt];
+    size = width * height;
+    switch(pix_fmt) {
+    case PIX_FMT_YUV420P:
+    case PIX_FMT_YUV422P:
+    case PIX_FMT_YUV444P:
+    case PIX_FMT_YUV410P:
+    case PIX_FMT_YUV411P:
+        w2 = (width + (1 << pinfo->x_chroma_shift) - 1) >> pinfo->x_chroma_shift;
+        h2 = (height + (1 << pinfo->y_chroma_shift) - 1) >> pinfo->y_chroma_shift;
+        size2 = w2 * h2;
+        picture->data[0] = ptr;
+        picture->data[1] = picture->data[0] + size;
+        picture->data[2] = picture->data[1] + size2;
+        picture->linesize[0] = width;
+        picture->linesize[1] = w2;
+        picture->linesize[2] = w2;
+        return size + 2 * size2;
+    case PIX_FMT_RGB24:
+    case PIX_FMT_BGR24:
+        picture->data[0] = ptr;
+        picture->data[1] = NULL;
+        picture->data[2] = NULL;
+        picture->linesize[0] = width * 3;
+        return size * 3;
+    case PIX_FMT_RGBA32:
+        picture->data[0] = ptr;
+        picture->data[1] = NULL;
+        picture->data[2] = NULL;
+        picture->linesize[0] = width * 4;
+        return size * 4;
+    case PIX_FMT_RGB555:
+    case PIX_FMT_RGB565:
+    case PIX_FMT_YUV422:
+        picture->data[0] = ptr;
+        picture->data[1] = NULL;
+        picture->data[2] = NULL;
+        picture->linesize[0] = width * 2;
+        return size * 2;
+    case PIX_FMT_GRAY8:
+        picture->data[0] = ptr;
+        picture->data[1] = NULL;
+        picture->data[2] = NULL;
+        picture->linesize[0] = width;
+        return size;
+    case PIX_FMT_MONOWHITE:
+    case PIX_FMT_MONOBLACK:
+        picture->data[0] = ptr;
+        picture->data[1] = NULL;
+        picture->data[2] = NULL;
+        picture->linesize[0] = (width + 7) >> 3;
+        return picture->linesize[0] * height;
+    case PIX_FMT_PAL8:
+        size2 = (size + 3) & ~3;
+        picture->data[0] = ptr;
+        picture->data[1] = ptr + size2; /* palette is stored here as 256 32 bit words */
+        picture->data[2] = NULL;
+        picture->linesize[0] = width;
+        picture->linesize[1] = 4;
+        return size2 + 256 * 4;
+    default:
+        picture->data[0] = NULL;
+        picture->data[1] = NULL;
+        picture->data[2] = NULL;
+        picture->data[3] = NULL;
+        return -1;
+    }
+}
+
+int avpicture_get_size(int pix_fmt, int width, int height)
+{
+    AVPicture dummy_pict;
+    return avpicture_fill(&dummy_pict, NULL, pix_fmt, width, height);
+}
+
+
 /* XXX: totally non optimized */
 
-static void yuv422_to_yuv420p(UINT8 *lum, UINT8 *cb, UINT8 *cr,
-                              UINT8 *src, int width, int height)
+static void yuv422_to_yuv420p(AVPicture *dst, AVPicture *src,
+                              int width, int height)
 {
+    uint8_t *lum, *cb, *cr;
     int x, y;
-    UINT8 *p = src;
-
+    const uint8_t *p;
+    lum = dst->data[0];
+    cb = dst->data[1];
+    cr = dst->data[2];
+    p = src->data[0];
+   
     for(y=0;y<height;y+=2) {
         for(x=0;x<width;x+=2) {
             lum[0] = p[0];
@@ -54,140 +263,14 @@ static void yuv422_to_yuv420p(UINT8 *lum, UINT8 *cb, UINT8 *cr,
 #define ONE_HALF  (1 << (SCALEBITS - 1))
 #define FIX(x)         ((int) ((x) * (1L<<SCALEBITS) + 0.5))
 
-static void rgb24_to_yuv420p(UINT8 *lum, UINT8 *cb, UINT8 *cr,
-                              UINT8 *src, int width, int height)
-{
-    int wrap, wrap3, x, y;
-    int r, g, b, r1, g1, b1;
-    UINT8 *p;
-
-    wrap = width;
-    wrap3 = width * 3;
-    p = src;
-    for(y=0;y<height;y+=2) {
-        for(x=0;x<width;x+=2) {
-            r = p[0];
-            g = p[1];
-            b = p[2];
-            r1 = r;
-            g1 = g;
-            b1 = b;
-            lum[0] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            r = p[3];
-            g = p[4];
-            b = p[5];
-            r1 += r;
-            g1 += g;
-            b1 += b;
-            lum[1] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            p += wrap3;
-            lum += wrap;
-
-            r = p[0];
-            g = p[1];
-            b = p[2];
-            r1 += r;
-            g1 += g;
-            b1 += b;
-            lum[0] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            r = p[3];
-            g = p[4];
-            b = p[5];
-            r1 += r;
-            g1 += g;
-            b1 += b;
-            lum[1] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            
-            cb[0] = ((- FIX(0.16874) * r1 - FIX(0.33126) * g1 + 
-                      FIX(0.50000) * b1 + 4 * ONE_HALF - 1) >> (SCALEBITS + 2)) + 128;
-            cr[0] = ((FIX(0.50000) * r1 - FIX(0.41869) * g1 - 
-                     FIX(0.08131) * b1 + 4 * ONE_HALF - 1) >> (SCALEBITS + 2)) + 128;
-
-            cb++;
-            cr++;
-            p += -wrap3 + 2 * 3;
-            lum += -wrap + 2;
-        }
-        p += wrap3;
-        lum += wrap;
-    }
-}
-
-static void bgr24_to_yuv420p(UINT8 *lum, UINT8 *cb, UINT8 *cr,
-                              UINT8 *src, int width, int height)
-{
-    int wrap, wrap3, x, y;
-    int r, g, b, r1, g1, b1;
-    UINT8 *p;
-
-    wrap = width;
-    wrap3 = width * 3;
-    p = src;
-    for(y=0;y<height;y+=2) {
-        for(x=0;x<width;x+=2) {
-            b = p[0];
-            g = p[1];
-            r = p[2];
-            r1 = r;
-            g1 = g;
-            b1 = b;
-            lum[0] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            b = p[3];
-            g = p[4];
-            r = p[5];
-            r1 += r;
-            g1 += g;
-            b1 += b;
-            lum[1] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            p += wrap3;
-            lum += wrap;
-
-            b = p[0];
-            g = p[1];
-            r = p[2];
-            r1 += r;
-            g1 += g;
-            b1 += b;
-            lum[0] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            b = p[3];
-            g = p[4];
-            r = p[5];
-            r1 += r;
-            g1 += g;
-            b1 += b;
-            lum[1] = (FIX(0.29900) * r + FIX(0.58700) * g + 
-                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;
-            
-            cb[0] = ((- FIX(0.16874) * r1 - FIX(0.33126) * g1 + 
-                      FIX(0.50000) * b1 + 4 * ONE_HALF - 1) >> (SCALEBITS + 2)) + 128;
-            cr[0] = ((FIX(0.50000) * r1 - FIX(0.41869) * g1 - 
-                     FIX(0.08131) * b1 + 4 * ONE_HALF - 1) >> (SCALEBITS + 2)) + 128;
-
-            cb++;
-            cr++;
-            p += -wrap3 + 2 * 3;
-            lum += -wrap + 2;
-        }
-        p += wrap3;
-        lum += wrap;
-    }
-}
-
 /* XXX: use generic filter ? */
 /* 1x2 -> 1x1 */
-static void shrink2(UINT8 *dst, int dst_wrap, 
-                    UINT8 *src, int src_wrap,
+static void shrink2(uint8_t *dst, int dst_wrap, 
+                    uint8_t *src, int src_wrap,
                     int width, int height)
 {
     int w;
-    UINT8 *s1, *s2, *d;
+    uint8_t *s1, *s2, *d;
 
     for(;height > 0; height--) {
         s1 = src;
@@ -214,12 +297,12 @@ static void shrink2(UINT8 *dst, int dst_wrap,
 }
 
 /* 2x2 -> 1x1 */
-static void shrink22(UINT8 *dst, int dst_wrap, 
-                     UINT8 *src, int src_wrap,
+static void shrink22(uint8_t *dst, int dst_wrap, 
+                     uint8_t *src, int src_wrap,
                      int width, int height)
 {
     int w;
-    UINT8 *s1, *s2, *d;
+    uint8_t *s1, *s2, *d;
 
     for(;height > 0; height--) {
         s1 = src;
@@ -245,8 +328,61 @@ static void shrink22(UINT8 *dst, int dst_wrap,
     }
 }
 
-static void img_copy(UINT8 *dst, int dst_wrap, 
-                     UINT8 *src, int src_wrap,
+/* 1x1 -> 2x2 */
+static void grow22(uint8_t *dst, int dst_wrap,
+                     uint8_t *src, int src_wrap,
+                     int width, int height)
+{
+    int w;
+    uint8_t *s1, *d;
+
+    for(;height > 0; height--) {
+        s1 = src;
+        d = dst;
+        for(w = width;w >= 4; w-=4) {
+            d[1] = d[0] = s1[0];
+            d[3] = d[2] = s1[1];
+            s1 += 2;
+            d += 4;
+        }
+        for(;w > 0; w--) {
+            d[0] = s1[0];
+            s1 ++;
+            d++;
+        }
+        if (height%2)
+            src += src_wrap;
+        dst += dst_wrap;
+    }
+}
+
+/* 1x2 -> 2x1 */
+static void conv411(uint8_t *dst, int dst_wrap, 
+                    uint8_t *src, int src_wrap,
+                    int width, int height)
+{
+    int w, c;
+    uint8_t *s1, *s2, *d;
+
+    for(;height > 0; height--) {
+        s1 = src;
+        s2 = src + src_wrap;
+        d = dst;
+        for(w = width;w > 0; w--) {
+            c = (s1[0] + s2[0]) >> 1;
+            d[0] = c;
+            d[1] = c;
+            s1++;
+            s2++;
+            d += 2;
+        }
+        src += src_wrap * 2;
+        dst += dst_wrap;
+    }
+}
+
+static void img_copy(uint8_t *dst, int dst_wrap, 
+                     uint8_t *src, int src_wrap,
                      int width, int height)
 {
     for(;height > 0; height--) {
@@ -264,7 +400,7 @@ static void img_copy(UINT8 *dst, int dst_wrap,
 #define C_GU (13954 >> (16 - SCALE_BITS))
 #define C_GV (34903 >> (16 - SCALE_BITS))
 
-#define RGBOUT(r, g, b, y1)\
+#define YUV_TO_RGB2(r, g, b, y1)\
 {\
     y = (y1 - 16) * C_Y;\
     r = cm[(y + r_add) >> SCALE_BITS];\
@@ -273,249 +409,1196 @@ static void img_copy(UINT8 *dst, int dst_wrap,
 }
 
 /* XXX: no chroma interpolating is done */
-static void yuv420p_to_rgb24(AVPicture *dst, AVPicture *src, 
-                             int width, int height)
+#define RGB_FUNCTIONS(rgb_name)                                         \
+                                                                        \
+static void yuv420p_to_ ## rgb_name (AVPicture *dst, AVPicture *src,    \
+                                     int width, int height)             \
+{                                                                       \
+    uint8_t *y1_ptr, *y2_ptr, *cb_ptr, *cr_ptr, *d, *d1, *d2;             \
+    int w, y, cb, cr, r_add, g_add, b_add, width2;                      \
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;                                 \
+    unsigned int r, g, b;                                               \
+                                                                        \
+    d = dst->data[0];                                                   \
+    y1_ptr = src->data[0];                                              \
+    cb_ptr = src->data[1];                                              \
+    cr_ptr = src->data[2];                                              \
+    width2 = (width + 1) >> 1;                                          \
+    for(;height >= 2; height -= 2) {                                    \
+        d1 = d;                                                         \
+        d2 = d + dst->linesize[0];                                      \
+        y2_ptr = y1_ptr + src->linesize[0];                             \
+        for(w = width; w >= 2; w -= 2) {                                \
+            cb = cb_ptr[0] - 128;                                       \
+            cr = cr_ptr[0] - 128;                                       \
+            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));                \
+            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));  \
+            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));                \
+                                                                        \
+            /* output 4 pixels */                                       \
+            YUV_TO_RGB2(r, g, b, y1_ptr[0]);                            \
+            RGB_OUT(d1, r, g, b);                                       \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y1_ptr[1]);                            \
+            RGB_OUT(d1 + BPP, r, g, b);                                 \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y2_ptr[0]);                            \
+            RGB_OUT(d2, r, g, b);                                       \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y2_ptr[1]);                            \
+            RGB_OUT(d2 + BPP, r, g, b);                                 \
+                                                                        \
+            d1 += 2 * BPP;                                              \
+            d2 += 2 * BPP;                                              \
+                                                                        \
+            y1_ptr += 2;                                                \
+            y2_ptr += 2;                                                \
+            cb_ptr++;                                                   \
+            cr_ptr++;                                                   \
+        }                                                               \
+        /* handle odd width */                                          \
+        if (w) {                                                        \
+            cb = cb_ptr[0] - 128;                                       \
+            cr = cr_ptr[0] - 128;                                       \
+            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));                \
+            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));  \
+            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));                \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y1_ptr[0]);                            \
+            RGB_OUT(d1, r, g, b);                                       \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y2_ptr[0]);                            \
+            RGB_OUT(d2, r, g, b);                                       \
+            d1 += BPP;                                                  \
+            d2 += BPP;                                                  \
+            y1_ptr++;                                                   \
+            y2_ptr++;                                                   \
+            cb_ptr++;                                                   \
+            cr_ptr++;                                                   \
+        }                                                               \
+        d += 2 * dst->linesize[0];                                      \
+        y1_ptr += 2 * src->linesize[0] - width;                         \
+        cb_ptr += src->linesize[1] - width2;                            \
+        cr_ptr += src->linesize[2] - width2;                            \
+    }                                                                   \
+    /* handle odd height */                                             \
+    if (height) {                                                       \
+        d1 = d;                                                         \
+        for(w = width; w >= 2; w -= 2) {                                \
+            cb = cb_ptr[0] - 128;                                       \
+            cr = cr_ptr[0] - 128;                                       \
+            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));                \
+            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));  \
+            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));                \
+                                                                        \
+            /* output 2 pixels */                                       \
+            YUV_TO_RGB2(r, g, b, y1_ptr[0]);                            \
+            RGB_OUT(d1, r, g, b);                                       \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y1_ptr[1]);                            \
+            RGB_OUT(d1 + BPP, r, g, b);                                 \
+                                                                        \
+            d1 += 2 * BPP;                                              \
+                                                                        \
+            y1_ptr += 2;                                                \
+            cb_ptr++;                                                   \
+            cr_ptr++;                                                   \
+        }                                                               \
+        /* handle width */                                              \
+        if (w) {                                                        \
+            cb = cb_ptr[0] - 128;                                       \
+            cr = cr_ptr[0] - 128;                                       \
+            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));                \
+            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));  \
+            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));                \
+                                                                        \
+            /* output 2 pixels */                                       \
+            YUV_TO_RGB2(r, g, b, y1_ptr[0]);                            \
+            RGB_OUT(d1, r, g, b);                                       \
+            d1 += BPP;                                                  \
+                                                                        \
+            y1_ptr++;                                                   \
+            cb_ptr++;                                                   \
+            cr_ptr++;                                                   \
+        }                                                               \
+    }                                                                   \
+}                                                                       \
+                                                                        \
+/* XXX: no chroma interpolating is done */                              \
+static void yuv422p_to_ ## rgb_name (AVPicture *dst, AVPicture *src,    \
+                                    int width, int height)              \
+{                                                                       \
+    uint8_t *y1_ptr, *cb_ptr, *cr_ptr, *d, *d1;                           \
+    int w, y, cb, cr, r_add, g_add, b_add, width2;                      \
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;                                 \
+    unsigned int r, g, b;                                               \
+                                                                        \
+    d = dst->data[0];                                                   \
+    y1_ptr = src->data[0];                                              \
+    cb_ptr = src->data[1];                                              \
+    cr_ptr = src->data[2];                                              \
+    width2 = (width + 1) >> 1;                                          \
+    for(;height > 0; height --) {                                       \
+        d1 = d;                                                         \
+        for(w = width; w >= 2; w -= 2) {                                \
+            cb = cb_ptr[0] - 128;                                       \
+            cr = cr_ptr[0] - 128;                                       \
+            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));                \
+            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));  \
+            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));                \
+                                                                        \
+            /* output 2 pixels */                                       \
+            YUV_TO_RGB2(r, g, b, y1_ptr[0]);                            \
+            RGB_OUT(d1, r, g, b);                                       \
+                                                                        \
+            YUV_TO_RGB2(r, g, b, y1_ptr[1]);                            \
+            RGB_OUT(d1 + BPP, r, g, b);                                 \
+                                                                        \
+            d1 += 2 * BPP;                                              \
+                                                                        \
+            y1_ptr += 2;                                                \
+            cb_ptr++;                                                   \
+            cr_ptr++;                                                   \
+        }                                                               \
+        /* handle width */                                              \
+        if (w) {                                                        \
+            cb = cb_ptr[0] - 128;                                       \
+            cr = cr_ptr[0] - 128;                                       \
+            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));                \
+            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));  \
+            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));                \
+                                                                        \
+            /* output 2 pixels */                                       \
+            YUV_TO_RGB2(r, g, b, y1_ptr[0]);                            \
+            RGB_OUT(d1, r, g, b);                                       \
+            d1 += BPP;                                                  \
+                                                                        \
+            y1_ptr++;                                                   \
+            cb_ptr++;                                                   \
+            cr_ptr++;                                                   \
+        }                                                               \
+        d += dst->linesize[0];                                          \
+        y1_ptr += src->linesize[0] - width;                             \
+        cb_ptr += src->linesize[1] - width2;                            \
+        cr_ptr += src->linesize[2] - width2;                            \
+    }                                                                   \
+}                                                                       \
+                                                                        \
+static void rgb_name ## _to_yuv420p(AVPicture *dst, AVPicture *src,     \
+                                    int width, int height)              \
+{                                                                       \
+    int wrap, wrap3, x, y;                                              \
+    int r, g, b, r1, g1, b1;                                            \
+    uint8_t *lum, *cb, *cr;                                               \
+    const uint8_t *p;                                                     \
+                                                                        \
+    lum = dst->data[0];                                                 \
+    cb = dst->data[1];                                                  \
+    cr = dst->data[2];                                                  \
+                                                                        \
+    wrap = dst->linesize[0];                                            \
+    wrap3 = src->linesize[0];                                           \
+    p = src->data[0];                                                   \
+    for(y=0;y<height;y+=2) {                                            \
+        for(x=0;x<width;x+=2) {                                         \
+            RGB_IN(r, g, b, p);                                         \
+            r1 = r;                                                     \
+            g1 = g;                                                     \
+            b1 = b;                                                     \
+            lum[0] = (FIX(0.29900) * r + FIX(0.58700) * g +             \
+                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;        \
+            RGB_IN(r, g, b, p + BPP);                                   \
+            r1 += r;                                                    \
+            g1 += g;                                                    \
+            b1 += b;                                                    \
+            lum[1] = (FIX(0.29900) * r + FIX(0.58700) * g +             \
+                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;        \
+            p += wrap3;                                                 \
+            lum += wrap;                                                \
+                                                                        \
+            RGB_IN(r, g, b, p);                                         \
+            r1 += r;                                                    \
+            g1 += g;                                                    \
+            b1 += b;                                                    \
+            lum[0] = (FIX(0.29900) * r + FIX(0.58700) * g +             \
+                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;        \
+                                                                        \
+            RGB_IN(r, g, b, p + BPP);                                   \
+            r1 += r;                                                    \
+            g1 += g;                                                    \
+            b1 += b;                                                    \
+            lum[1] = (FIX(0.29900) * r + FIX(0.58700) * g +             \
+                      FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;        \
+                                                                        \
+            cb[0] = ((- FIX(0.16874) * r1 - FIX(0.33126) * g1 +         \
+                      FIX(0.50000) * b1 + 4 * ONE_HALF - 1) >>          \
+                     (SCALEBITS + 2)) + 128;                            \
+            cr[0] = ((FIX(0.50000) * r1 - FIX(0.41869) * g1 -           \
+                     FIX(0.08131) * b1 + 4 * ONE_HALF - 1) >>           \
+                     (SCALEBITS + 2)) + 128;                            \
+                                                                        \
+            cb++;                                                       \
+            cr++;                                                       \
+            p += -wrap3 + 2 * BPP;                                      \
+            lum += -wrap + 2;                                           \
+        }                                                               \
+        p += wrap3 + (wrap3 - width * BPP);                             \
+        lum += wrap + (wrap - width);                                   \
+        cb += dst->linesize[1] - width / 2;                             \
+        cr += dst->linesize[2] - width / 2;                             \
+    }                                                                   \
+}                                                                       \
+                                                                        \
+static void rgb_name ## _to_gray(AVPicture *dst, AVPicture *src,        \
+                                 int width, int height)                 \
+{                                                                       \
+    const unsigned char *p;                                             \
+    unsigned char *q;                                                   \
+    int r, g, b, dst_wrap, src_wrap;                                    \
+    int x, y;                                                           \
+                                                                        \
+    p = src->data[0];                                                   \
+    src_wrap = src->linesize[0] - BPP * width;                          \
+                                                                        \
+    q = dst->data[0];                                                   \
+    dst_wrap = dst->linesize[0] - width;                                \
+                                                                        \
+    for(y=0;y<height;y++) {                                             \
+        for(x=0;x<width;x++) {                                          \
+            RGB_IN(r, g, b, p);                                         \
+            q[0] = (FIX(0.29900) * r + FIX(0.58700) * g +               \
+                    FIX(0.11400) * b + ONE_HALF) >> SCALEBITS;          \
+            q++;                                                        \
+            p += BPP;                                                   \
+        }                                                               \
+        p += src_wrap;                                                  \
+        q += dst_wrap;                                                  \
+    }                                                                   \
+}                                                                       \
+                                                                        \
+static void gray_to_ ## rgb_name(AVPicture *dst, AVPicture *src,        \
+                                 int width, int height)                 \
+{                                                                       \
+    const unsigned char *p;                                             \
+    unsigned char *q;                                                   \
+    int r, dst_wrap, src_wrap;                                          \
+    int x, y;                                                           \
+                                                                        \
+    p = src->data[0];                                                   \
+    src_wrap = src->linesize[0] - width;                                \
+                                                                        \
+    q = dst->data[0];                                                   \
+    dst_wrap = dst->linesize[0] - BPP * width;                          \
+                                                                        \
+    for(y=0;y<height;y++) {                                             \
+        for(x=0;x<width;x++) {                                          \
+            r = p[0];                                                   \
+            RGB_OUT(q, r, r, r);                                        \
+            q += BPP;                                                   \
+            p ++;                                                       \
+        }                                                               \
+        p += src_wrap;                                                  \
+        q += dst_wrap;                                                  \
+    }                                                                   \
+}                                                                       \
+                                                                        \
+static void pal8_to_ ## rgb_name(AVPicture *dst, AVPicture *src,        \
+                                 int width, int height)                 \
+{                                                                       \
+    const unsigned char *p;                                             \
+    unsigned char *q;                                                   \
+    int r, g, b, dst_wrap, src_wrap;                                    \
+    int x, y;                                                           \
+    uint32_t v;\
+    const uint32_t *palette;\
+\
+    p = src->data[0];                                                   \
+    src_wrap = src->linesize[0] - width;                                \
+    palette = (uint32_t *)src->data[1];\
+                                                                        \
+    q = dst->data[0];                                                   \
+    dst_wrap = dst->linesize[0] - BPP * width;                          \
+                                                                        \
+    for(y=0;y<height;y++) {                                             \
+        for(x=0;x<width;x++) {                                          \
+            v = palette[p[0]];\
+            r = (v >> 16) & 0xff;\
+            g = (v >> 8) & 0xff;\
+            b = (v) & 0xff;\
+            RGB_OUT(q, r, g, b);                                        \
+            q += BPP;                                                   \
+            p ++;                                                       \
+        }                                                               \
+        p += src_wrap;                                                  \
+        q += dst_wrap;                                                  \
+    }                                                                   \
+}
+
+/* copy bit n to bits 0 ... n - 1 */
+static inline unsigned int bitcopy_n(unsigned int a, int n)
 {
-    UINT8 *y1_ptr, *y2_ptr, *cb_ptr, *cr_ptr, *d, *d1, *d2;
-    int w, y, cb, cr, r_add, g_add, b_add, width2;
-    UINT8 *cm = cropTbl + MAX_NEG_CROP;
+    int mask;
+    mask = (1 << n) - 1;
+    return (a & (0xff & ~mask)) | ((-((a >> n) & 1)) & mask);
+}
 
-    d = dst->data[0];
-    y1_ptr = src->data[0];
-    cb_ptr = src->data[1];
-    cr_ptr = src->data[2];
-    width2 = width >> 1;
-    for(;height > 0; height -= 2) {
-        d1 = d;
-        d2 = d + dst->linesize[0];
-        y2_ptr = y1_ptr + src->linesize[0];
-        for(w = width2; w > 0; w --) {
-            cb = cb_ptr[0] - 128;
-            cr = cr_ptr[0] - 128;
-            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));
-            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));
-            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));
-            
-            /* output 4 pixels */
-            RGBOUT(d1[0], d1[1], d1[2], y1_ptr[0]);
-            RGBOUT(d1[3], d1[4], d1[5], y1_ptr[1]);
-            RGBOUT(d2[0], d2[1], d2[2], y2_ptr[0]);
-            RGBOUT(d2[3], d2[4], d2[5], y2_ptr[1]);
-
-            d1 += 6;
-            d2 += 6;
-            y1_ptr += 2;
-            y2_ptr += 2;
-            cb_ptr++;
-            cr_ptr++;
+/* rgb555 handling */
+
+#define RGB_IN(r, g, b, s)\
+{\
+    unsigned int v = ((const uint16_t *)(s))[0];\
+    r = bitcopy_n(v >> (10 - 3), 3);\
+    g = bitcopy_n(v >> (5 - 3), 3);\
+    b = bitcopy_n(v << 3, 3);\
+}
+
+#define RGB_OUT(d, r, g, b)\
+{\
+    ((uint16_t *)(d))[0] = ((r >> 3) << 10) | ((g >> 3) << 5) | (b >> 3) | 0x8000;\
+}
+
+#define BPP 2
+
+RGB_FUNCTIONS(rgb555)
+
+#undef RGB_IN
+#undef RGB_OUT
+#undef BPP
+
+/* rgb565 handling */
+
+#define RGB_IN(r, g, b, s)\
+{\
+    unsigned int v = ((const uint16_t *)(s))[0];\
+    r = bitcopy_n(v >> (11 - 3), 3);\
+    g = bitcopy_n(v >> (5 - 2), 2);\
+    b = bitcopy_n(v << 3, 3);\
+}
+
+#define RGB_OUT(d, r, g, b)\
+{\
+    ((uint16_t *)(d))[0] = ((r >> 3) << 11) | ((g >> 2) << 5) | (b >> 3);\
+}
+
+#define BPP 2
+
+RGB_FUNCTIONS(rgb565)
+
+#undef RGB_IN
+#undef RGB_OUT
+#undef BPP
+
+/* bgr24 handling */
+
+#define RGB_IN(r, g, b, s)\
+{\
+    b = (s)[0];\
+    g = (s)[1];\
+    r = (s)[2];\
+}
+
+#define RGB_OUT(d, r, g, b)\
+{\
+    (d)[0] = b;\
+    (d)[1] = g;\
+    (d)[2] = r;\
+}
+
+#define BPP 3
+
+RGB_FUNCTIONS(bgr24)
+
+#undef RGB_IN
+#undef RGB_OUT
+#undef BPP
+
+/* rgb24 handling */
+
+#define RGB_IN(r, g, b, s)\
+{\
+    r = (s)[0];\
+    g = (s)[1];\
+    b = (s)[2];\
+}
+
+#define RGB_OUT(d, r, g, b)\
+{\
+    (d)[0] = r;\
+    (d)[1] = g;\
+    (d)[2] = b;\
+}
+
+#define BPP 3
+
+RGB_FUNCTIONS(rgb24)
+
+#undef RGB_IN
+#undef RGB_OUT
+#undef BPP
+
+/* rgba32 handling */
+
+#define RGB_IN(r, g, b, s)\
+{\
+    unsigned int v = ((const uint32_t *)(s))[0];\
+    r = (v >> 16) & 0xff;\
+    g = (v >> 8) & 0xff;\
+    b = v & 0xff;\
+}
+
+#define RGB_OUT(d, r, g, b)\
+{\
+    ((uint32_t *)(d))[0] = (0xff << 24) | (r << 16) | (g << 8) | b;\
+}
+
+#define BPP 4
+
+RGB_FUNCTIONS(rgba32)
+
+#undef RGB_IN
+#undef RGB_OUT
+#undef BPP
+
+
+static void rgb24_to_rgb565(AVPicture *dst, AVPicture *src,
+                            int width, int height)
+{
+    const unsigned char *p;
+    unsigned char *q;
+    int r, g, b, dst_wrap, src_wrap;
+    int x, y;
+
+    p = src->data[0];
+    src_wrap = src->linesize[0] - 3 * width;
+
+    q = dst->data[0];
+    dst_wrap = dst->linesize[0] - 2 * width;
+
+    for(y=0;y<height;y++) {
+        for(x=0;x<width;x++) {
+            r = p[0];
+            g = p[1];
+            b = p[2];
+
+            ((unsigned short *)q)[0] = 
+                ((r >> 3) << 11) | ((g >> 2) << 5) | (b >> 3);
+            q += 2;
+            p += 3;
         }
-        d += 2 * dst->linesize[0];
-        y1_ptr += 2 * src->linesize[0] - width;
-        cb_ptr += src->linesize[1] - width2;
-        cr_ptr += src->linesize[2] - width2;
+        p += src_wrap;
+        q += dst_wrap;
     }
 }
 
-/* XXX: no chroma interpolating is done */
-static void yuv422p_to_rgb24(AVPicture *dst, AVPicture *src, 
-                             int width, int height)
+/* NOTE: we also add a dummy alpha bit */
+static void rgb24_to_rgb555(AVPicture *dst, AVPicture *src,
+                            int width, int height)
+{
+    const unsigned char *p;
+    unsigned char *q;
+    int r, g, b, dst_wrap, src_wrap;
+    int x, y;
+
+    p = src->data[0];
+    src_wrap = src->linesize[0] - 3 * width;
+
+    q = dst->data[0];
+    dst_wrap = dst->linesize[0] - 2 * width;
+
+    for(y=0;y<height;y++) {
+        for(x=0;x<width;x++) {
+            r = p[0];
+            g = p[1];
+            b = p[2];
+
+            ((unsigned short *)q)[0] = 
+                ((r >> 3) << 10) | ((g >> 3) << 5) | (b >> 3) | 0x8000;
+            q += 2;
+            p += 3;
+        }
+        p += src_wrap;
+        q += dst_wrap;
+    }
+}
+
+static void mono_to_gray(AVPicture *dst, AVPicture *src,
+                         int width, int height, int xor_mask)
+{
+    const unsigned char *p;
+    unsigned char *q;
+    int v, dst_wrap, src_wrap;
+    int y, w;
+
+    p = src->data[0];
+    src_wrap = src->linesize[0] - ((width + 7) >> 3);
+
+    q = dst->data[0];
+    dst_wrap = dst->linesize[0] - width;
+    for(y=0;y<height;y++) {
+        w = width; 
+        while (w >= 8) {
+            v = *p++ ^ xor_mask;
+            q[0] = -(v >> 7);
+            q[1] = -((v >> 6) & 1);
+            q[2] = -((v >> 5) & 1);
+            q[3] = -((v >> 4) & 1);
+            q[4] = -((v >> 3) & 1);
+            q[5] = -((v >> 2) & 1);
+            q[6] = -((v >> 1) & 1);
+            q[7] = -((v >> 0) & 1);
+            w -= 8;
+            q += 8;
+        }
+        if (w > 0) {
+            v = *p++ ^ xor_mask;
+            do {
+                q[0] = -((v >> 7) & 1);
+                q++;
+                v <<= 1;
+            } while (--w);
+        }
+        p += src_wrap;
+        q += dst_wrap;
+    }
+}
+
+static void monowhite_to_gray(AVPicture *dst, AVPicture *src,
+                               int width, int height)
 {
-    UINT8 *y1_ptr, *cb_ptr, *cr_ptr, *d, *d1;
-    int w, y, cb, cr, r_add, g_add, b_add, width2;
-    UINT8 *cm = cropTbl + MAX_NEG_CROP;
+    mono_to_gray(dst, src, width, height, 0xff);
+}
+
+static void monoblack_to_gray(AVPicture *dst, AVPicture *src,
+                               int width, int height)
+{
+    mono_to_gray(dst, src, width, height, 0x00);
+}
+
+static void gray_to_mono(AVPicture *dst, AVPicture *src,
+                         int width, int height, int xor_mask)
+{
+    int n;
+    const uint8_t *s;
+    uint8_t *d;
+    int j, b, v, n1, src_wrap, dst_wrap, y;
+
+    s = src->data[0];
+    src_wrap = src->linesize[0] - width;
 
     d = dst->data[0];
-    y1_ptr = src->data[0];
-    cb_ptr = src->data[1];
-    cr_ptr = src->data[2];
-    width2 = width >> 1;
-    for(;height > 0; height --) {
-        d1 = d;
-        for(w = width2; w > 0; w --) {
-            cb = cb_ptr[0] - 128;
-            cr = cr_ptr[0] - 128;
-            r_add = C_RV * cr + (1 << (SCALE_BITS - 1));
-            g_add = - C_GU * cb - C_GV * cr + (1 << (SCALE_BITS - 1));
-            b_add = C_BU * cb + (1 << (SCALE_BITS - 1));
-            
-            /* output 2 pixels */
-            RGBOUT(d1[0], d1[1], d1[2], y1_ptr[0]);
-            RGBOUT(d1[3], d1[4], d1[5], y1_ptr[1]);
-
-            d1 += 6;
-            y1_ptr += 2;
-            cb_ptr++;
-            cr_ptr++;
+    dst_wrap = dst->linesize[0] - ((width + 7) >> 3);
+    printf("%d %d\n", width, height);
+
+    for(y=0;y<height;y++) {
+        n = width;
+        while (n >= 8) {
+            v = 0;
+            for(j=0;j<8;j++) {
+                b = s[0];
+                s++;
+                v = (v << 1) | (b >> 7);
+            }
+            d[0] = v ^ xor_mask;
+            d++;
+            n -= 8;
         }
-        d += dst->linesize[0];
-        y1_ptr += src->linesize[0] - width;
-        cb_ptr += src->linesize[1] - width2;
-        cr_ptr += src->linesize[2] - width2;
+        if (n > 0) {
+            n1 = n;
+            v = 0;
+            while (n > 0) {
+                b = s[0];
+                s++;
+                v = (v << 1) | (b >> 7);
+                n--;
+            }
+            d[0] = (v << (8 - (n1 & 7))) ^ xor_mask;
+            d++;
+        }
+        s += src_wrap;
+        d += dst_wrap;
     }
 }
 
-/* XXX: always use linesize. Return -1 if not supported */
-int img_convert(AVPicture *dst, int dst_pix_fmt,
-                AVPicture *src, int pix_fmt, 
-                int width, int height)
+static void gray_to_monowhite(AVPicture *dst, AVPicture *src,
+                              int width, int height)
 {
-    int i;
+    gray_to_mono(dst, src, width, height, 0xff);
+}
+
+static void gray_to_monoblack(AVPicture *dst, AVPicture *src,
+                              int width, int height)
+{
+    gray_to_mono(dst, src, width, height, 0x00);
+}
+
+/* this is maybe slow, but allows for extensions */
+static inline unsigned char gif_clut_index(uint8_t r, uint8_t g, uint8_t b)
+{
+    return ((((r)/47)%6)*6*6+(((g)/47)%6)*6+(((b)/47)%6));
+}
+
+/* XXX: put jpeg quantize code instead */
+static void rgb24_to_pal8(AVPicture *dst, AVPicture *src,
+                          int width, int height)
+{
+    const unsigned char *p;
+    unsigned char *q;
+    int r, g, b, dst_wrap, src_wrap;
+    int x, y, i;
+    static const uint8_t pal_value[6] = { 0x00, 0x33, 0x66, 0x99, 0xcc, 0xff };
+    uint32_t *pal;
+
+    p = src->data[0];
+    src_wrap = src->linesize[0] - 3 * width;
+
+    q = dst->data[0];
+    dst_wrap = dst->linesize[0] - width;
+
+    for(y=0;y<height;y++) {
+        for(x=0;x<width;x++) {
+            r = p[0];
+            g = p[1];
+            b = p[2];
+
+            q[0] = gif_clut_index(r, g, b);
+            q++;
+            p += 3;
+        }
+        p += src_wrap;
+        q += dst_wrap;
+    }
 
-    assert(pix_fmt != PIX_FMT_ANY && dst_pix_fmt != PIX_FMT_ANY);
-
-    if (dst_pix_fmt == pix_fmt) {
-        switch(pix_fmt) {
-        case PIX_FMT_YUV420P:
-            for(i=0;i<3;i++) {
-                if (i == 1) {
-                    width >>= 1;
-                    height >>= 1;
-                }
-                img_copy(dst->data[i], dst->linesize[i],
-                         src->data[i], src->linesize[i],
-                         width, height);
+    /* build palette */
+    pal = (uint32_t *)dst->data[1];
+    i = 0;
+    for(r = 0; r < 6; r++) {
+        for(g = 0; g < 6; g++) {
+            for(b = 0; b < 6; b++) {
+                pal[i++] = (0xff << 24) | (pal_value[r] << 16) | 
+                    (pal_value[g] << 8) | pal_value[b];
             }
-            break;
-        default:
-            return -1;
         }
-    } else if (dst_pix_fmt == PIX_FMT_YUV420P) {
+    }
+    while (i < 256)
+        pal[i++] = 0;
+}
         
-        switch(pix_fmt) {
-        case PIX_FMT_YUV420P:
-            for(i=0;i<3;i++) {
-                img_copy(dst->data[i], dst->linesize[i],
-                         src->data[i], src->linesize[i],
-                         width, height);
-            }
-            break;
-        case PIX_FMT_YUV422P:
-            img_copy(dst->data[0], dst->linesize[0],
-                     src->data[0], src->linesize[0],
-                     width, height);
-            width >>= 1;
-            height >>= 1;
-            for(i=1;i<3;i++) {
-                shrink2(dst->data[i], dst->linesize[i],
-                        src->data[i], src->linesize[i],
-                        width, height);
+typedef struct ConvertEntry {
+    void (*convert)(AVPicture *dst, AVPicture *src, int width, int height);
+} ConvertEntry;
+
+/* add each new convertion function in this table */
+/* constraints;
+   - all non YUV modes must convert at least to and from PIX_FMT_RGB24
+*/
+static ConvertEntry convert_table[PIX_FMT_NB][PIX_FMT_NB] = {
+    [PIX_FMT_YUV420P] = {
+        [PIX_FMT_RGB555] = { 
+            .convert = yuv420p_to_rgb555
+        },
+        [PIX_FMT_RGB565] = { 
+            .convert = yuv420p_to_rgb565
+        },
+        [PIX_FMT_BGR24] = { 
+            .convert = yuv420p_to_bgr24
+        },
+        [PIX_FMT_RGB24] = { 
+            .convert = yuv420p_to_rgb24
+        },
+        [PIX_FMT_RGBA32] = { 
+            .convert = yuv420p_to_rgba32
+        },
+    },
+    [PIX_FMT_YUV422P] = {
+        [PIX_FMT_RGB555] = { 
+            .convert = yuv422p_to_rgb555
+        },
+        [PIX_FMT_RGB565] = { 
+            .convert = yuv422p_to_rgb565
+        },
+        [PIX_FMT_BGR24] = { 
+            .convert = yuv422p_to_bgr24
+        },
+        [PIX_FMT_RGB24] = { 
+            .convert = yuv422p_to_rgb24
+        },
+        [PIX_FMT_RGBA32] = { 
+            .convert = yuv422p_to_rgba32
+        },
+    },
+    [PIX_FMT_YUV422] = { 
+        [PIX_FMT_YUV420P] = { 
+            .convert = yuv422_to_yuv420p,
+        },
+    },
+
+    [PIX_FMT_RGB24] = {
+        [PIX_FMT_YUV420P] = { 
+            .convert = rgb24_to_yuv420p
+        },
+        [PIX_FMT_RGB565] = { 
+            .convert = rgb24_to_rgb565
+        },
+        [PIX_FMT_RGB555] = { 
+            .convert = rgb24_to_rgb555
+        },
+        [PIX_FMT_GRAY8] = { 
+            .convert = rgb24_to_gray
+        },
+        [PIX_FMT_PAL8] = { 
+            .convert = rgb24_to_pal8
+        },
+    },
+    [PIX_FMT_RGBA32] = {
+        [PIX_FMT_YUV420P] = { 
+            .convert = rgba32_to_yuv420p
+        },
+        [PIX_FMT_GRAY8] = { 
+            .convert = rgba32_to_gray
+        },
+    },
+    [PIX_FMT_BGR24] = {
+        [PIX_FMT_YUV420P] = { 
+            .convert = bgr24_to_yuv420p
+        },
+        [PIX_FMT_GRAY8] = { 
+            .convert = bgr24_to_gray
+        },
+    },
+    [PIX_FMT_RGB555] = {
+        [PIX_FMT_YUV420P] = { 
+            .convert = rgb555_to_yuv420p
+        },
+        [PIX_FMT_GRAY8] = { 
+            .convert = rgb555_to_gray
+        },
+    },
+    [PIX_FMT_RGB565] = {
+        [PIX_FMT_YUV420P] = { 
+            .convert = rgb565_to_yuv420p
+        },
+        [PIX_FMT_GRAY8] = { 
+            .convert = rgb565_to_gray
+        },
+    },
+    [PIX_FMT_GRAY8] = {
+        [PIX_FMT_RGB555] = { 
+            .convert = gray_to_rgb555
+        },
+        [PIX_FMT_RGB565] = { 
+            .convert = gray_to_rgb565
+        },
+        [PIX_FMT_RGB24] = { 
+            .convert = gray_to_rgb24
+        },
+        [PIX_FMT_BGR24] = { 
+            .convert = gray_to_bgr24
+        },
+        [PIX_FMT_RGBA32] = { 
+            .convert = gray_to_rgba32
+        },
+        [PIX_FMT_MONOWHITE] = { 
+            .convert = gray_to_monowhite
+        },
+        [PIX_FMT_MONOBLACK] = { 
+            .convert = gray_to_monoblack
+        },
+    },
+    [PIX_FMT_MONOWHITE] = {
+        [PIX_FMT_GRAY8] = { 
+            .convert = monowhite_to_gray
+        },
+    },
+    [PIX_FMT_MONOBLACK] = {
+        [PIX_FMT_GRAY8] = { 
+            .convert = monoblack_to_gray
+        },
+    },
+    [PIX_FMT_PAL8] = {
+        [PIX_FMT_RGB555] = { 
+            .convert = pal8_to_rgb555
+        },
+        [PIX_FMT_RGB565] = { 
+            .convert = pal8_to_rgb565
+        },
+        [PIX_FMT_BGR24] = { 
+            .convert = pal8_to_bgr24
+        },
+        [PIX_FMT_RGB24] = { 
+            .convert = pal8_to_rgb24
+        },
+        [PIX_FMT_RGBA32] = { 
+            .convert = pal8_to_rgba32
+        },
+    },
+};
+
+static int avpicture_alloc(AVPicture *picture,
+                           int pix_fmt, int width, int height)
+{
+    unsigned int size;
+    void *ptr;
+
+    size = avpicture_get_size(pix_fmt, width, height);
+    if (size < 0)
+        goto fail;
+    ptr = av_malloc(size);
+    if (!ptr)
+        goto fail;
+    avpicture_fill(picture, ptr, pix_fmt, width, height);
+    return 0;
+ fail:
+    memset(picture, 0, sizeof(AVPicture));
+    return -1;
+}
+
+static void avpicture_free(AVPicture *picture)
+{
+    av_free(picture->data[0]);
+}
+
+/* XXX: always use linesize. Return -1 if not supported */
+int img_convert(AVPicture *dst, int dst_pix_fmt,
+                AVPicture *src, int src_pix_fmt, 
+                int src_width, int src_height)
+{
+    int i, ret, dst_width, dst_height, int_pix_fmt;
+    PixFmtInfo *src_pix, *dst_pix;
+    ConvertEntry *ce;
+    AVPicture tmp1, *tmp = &tmp1;
+
+    if (src_pix_fmt < 0 || src_pix_fmt >= PIX_FMT_NB ||
+        dst_pix_fmt < 0 || dst_pix_fmt >= PIX_FMT_NB)
+        return -1;
+    if (src_width <= 0 || src_height <= 0)
+        return 0;
+
+    dst_width = src_width;
+    dst_height = src_height;
+
+    dst_pix = &pix_fmt_info[dst_pix_fmt];
+    src_pix = &pix_fmt_info[src_pix_fmt];
+    if (src_pix_fmt == dst_pix_fmt) {
+        /* XXX: incorrect */
+        /* same format: just copy */
+        for(i = 0; i < dst_pix->nb_components; i++) {
+            int w, h;
+            w = dst_width;
+            h = dst_height;
+            if (dst_pix->is_yuv && (i == 1 || i == 2)) {
+                w >>= dst_pix->x_chroma_shift;
+                h >>= dst_pix->y_chroma_shift;
             }
-            break;
-        case PIX_FMT_YUV444P:
-            img_copy(dst->data[0], dst->linesize[0],
-                     src->data[0], src->linesize[0],
-                     width, height);
-            width >>= 1;
-            height >>= 1;
-            for(i=1;i<3;i++) {
-                shrink22(dst->data[i], dst->linesize[i],
-                         src->data[i], src->linesize[i],
-                         width, height);
+            img_copy(dst->data[i], dst->linesize[i],
+                     src->data[i], src->linesize[i],
+                     w, h);
+        }
+        return 0;
+    }
+
+    ce = &convert_table[src_pix_fmt][dst_pix_fmt];
+    if (ce->convert) {
+        /* specific convertion routine */
+        ce->convert(dst, src, dst_width, dst_height);
+        return 0;
+    }
+
+    /* gray to YUV */
+    if (dst_pix->is_yuv && src_pix_fmt == PIX_FMT_GRAY8) {
+        int w, h, y;
+        uint8_t *d;
+
+        img_copy(dst->data[0], dst->linesize[0],
+                 src->data[0], src->linesize[0],
+                 dst_width, dst_height);
+        /* fill U and V with 128 */
+        w = dst_width;
+        h = dst_height;
+        w >>= dst_pix->x_chroma_shift;
+        h >>= dst_pix->y_chroma_shift;
+        for(i = 1; i <= 2; i++) {
+            d = dst->data[i];
+            for(y = 0; y< h; y++) {
+                memset(d, 128, w);
+                d += dst->linesize[i];
             }
-            break;
-        case PIX_FMT_YUV422:
-            yuv422_to_yuv420p(dst->data[0], dst->data[1], dst->data[2], 
-                              src->data[0], width, height);
-            break;
-        case PIX_FMT_RGB24:
-            rgb24_to_yuv420p(dst->data[0], dst->data[1], dst->data[2], 
-                             src->data[0], width, height);
-            break;
-        case PIX_FMT_BGR24:
-            bgr24_to_yuv420p(dst->data[0], dst->data[1], dst->data[2], 
-                             src->data[0], width, height);
-            break;
-        default:
-            return -1;
         }
-    } else if (dst_pix_fmt == PIX_FMT_RGB24) {
-        switch(pix_fmt) {
-        case PIX_FMT_YUV420P:
-            yuv420p_to_rgb24(dst, src, width, height);
-            break;
-        case PIX_FMT_YUV422P:
-            yuv422p_to_rgb24(dst, src, width, height);
-            break;
-        default:
+        return 0;
+    }
+
+    /* YUV to gray */
+    if (src_pix->is_yuv && dst_pix_fmt == PIX_FMT_GRAY8) {
+        img_copy(dst->data[0], dst->linesize[0],
+                 src->data[0], src->linesize[0],
+                 dst_width, dst_height);
+        return 0;
+    }
+
+    /* YUV to YUV */
+    if (dst_pix->is_yuv && src_pix->is_yuv) {
+        int x_shift, y_shift, w, h;
+        void (*resize_func)(uint8_t *dst, int dst_wrap, 
+                            uint8_t *src, int src_wrap,
+                            int width, int height);
+
+        /* compute chroma size of the smallest dimensions */
+        w = dst_width;
+        h = dst_height;
+        if (dst_pix->x_chroma_shift >= src_pix->x_chroma_shift)
+            w >>= dst_pix->x_chroma_shift;
+        else
+            w >>= src_pix->x_chroma_shift;
+        if (dst_pix->y_chroma_shift >= src_pix->y_chroma_shift)
+            h >>= dst_pix->y_chroma_shift;
+        else
+            h >>= src_pix->y_chroma_shift;
+
+        x_shift = (dst_pix->x_chroma_shift - src_pix->x_chroma_shift);
+        y_shift = (dst_pix->y_chroma_shift - src_pix->y_chroma_shift);
+        if (x_shift == 0 && y_shift == 0) {
+            resize_func = img_copy; /* should never happen */
+        } else if (x_shift == 0 && y_shift == 1) {
+            resize_func = shrink2;
+        } else if (x_shift == 1 && y_shift == 1) {
+            resize_func = shrink22;
+        } else if (x_shift == -1 && y_shift == -1) {
+            resize_func = grow22;
+        } else if (x_shift == -1 && y_shift == 1) {
+            resize_func = conv411;
+        } else {
+            /* currently not handled */
             return -1;
         }
+
+        img_copy(dst->data[0], dst->linesize[0],
+                 src->data[0], src->linesize[0],
+                 dst_width, dst_height);
+
+        for(i = 1;i <= 2; i++)
+            resize_func(dst->data[i], dst->linesize[i],
+                        src->data[i], src->linesize[i],
+                        dst_width>>dst_pix->x_chroma_shift, dst_height>>dst_pix->y_chroma_shift);
+       return 0;
+    }
+
+    /* try to use an intermediate format */
+    if (src_pix_fmt == PIX_FMT_MONOWHITE ||
+        src_pix_fmt == PIX_FMT_MONOBLACK ||
+        dst_pix_fmt == PIX_FMT_MONOWHITE ||
+        dst_pix_fmt == PIX_FMT_MONOBLACK) {
+        int_pix_fmt = PIX_FMT_GRAY8;
     } else {
-        return -1;
+        int_pix_fmt = PIX_FMT_RGB24;
     }
-    return 0;
+    if (avpicture_alloc(tmp, int_pix_fmt, dst_width, dst_height) < 0)
+        return -1;
+    ret = -1;
+    if (img_convert(tmp, int_pix_fmt,
+                    src, src_pix_fmt, src_width, src_height) < 0)
+        goto fail1;
+    if (img_convert(dst, dst_pix_fmt,
+                    tmp, int_pix_fmt, dst_width, dst_height) < 0)
+        goto fail1;
+    ret = 0;
+ fail1:
+    avpicture_free(tmp);
+    return ret;
 }
 
+
+#ifdef HAVE_MMX
+#define DEINT_INPLACE_LINE_LUM \
+                    movd_m2r(lum_m4[0],mm0);\
+                    movd_m2r(lum_m3[0],mm1);\
+                    movd_m2r(lum_m2[0],mm2);\
+                    movd_m2r(lum_m1[0],mm3);\
+                    movd_m2r(lum[0],mm4);\
+                    punpcklbw_r2r(mm7,mm0);\
+                    movd_r2m(mm2,lum_m4[0]);\
+                    punpcklbw_r2r(mm7,mm1);\
+                    punpcklbw_r2r(mm7,mm2);\
+                    punpcklbw_r2r(mm7,mm3);\
+                    punpcklbw_r2r(mm7,mm4);\
+                    paddw_r2r(mm3,mm1);\
+                    psllw_i2r(1,mm2);\
+                    paddw_r2r(mm4,mm0);\
+                    psllw_i2r(2,mm1);\
+                    paddw_r2r(mm6,mm2);\
+                    paddw_r2r(mm2,mm1);\
+                    psubusw_r2r(mm0,mm1);\
+                    psrlw_i2r(3,mm1);\
+                    packuswb_r2r(mm7,mm1);\
+                    movd_r2m(mm1,lum_m2[0]);
+
+#define DEINT_LINE_LUM \
+                    movd_m2r(lum_m4[0],mm0);\
+                    movd_m2r(lum_m3[0],mm1);\
+                    movd_m2r(lum_m2[0],mm2);\
+                    movd_m2r(lum_m1[0],mm3);\
+                    movd_m2r(lum[0],mm4);\
+                    punpcklbw_r2r(mm7,mm0);\
+                    punpcklbw_r2r(mm7,mm1);\
+                    punpcklbw_r2r(mm7,mm2);\
+                    punpcklbw_r2r(mm7,mm3);\
+                    punpcklbw_r2r(mm7,mm4);\
+                    paddw_r2r(mm3,mm1);\
+                    psllw_i2r(1,mm2);\
+                    paddw_r2r(mm4,mm0);\
+                    psllw_i2r(2,mm1);\
+                    paddw_r2r(mm6,mm2);\
+                    paddw_r2r(mm2,mm1);\
+                    psubusw_r2r(mm0,mm1);\
+                    psrlw_i2r(3,mm1);\
+                    packuswb_r2r(mm7,mm1);\
+                    movd_r2m(mm1,dst[0]);
+#endif
+
 /* filter parameters: [-1 4 2 4 -1] // 8 */
-static void deinterlace_line(UINT8 *dst, UINT8 *src, int src_wrap,
-                             int size)
+static void deinterlace_line(uint8_t *dst, uint8_t *lum_m4, uint8_t *lum_m3, uint8_t *lum_m2, uint8_t *lum_m1, uint8_t *lum,
+                                int size)
 {
-    UINT8 *cm = cropTbl + MAX_NEG_CROP;
+#ifndef HAVE_MMX
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;
     int sum;
-    UINT8 *s;
 
     for(;size > 0;size--) {
-        s = src;
-        sum = -s[0];
-        s += src_wrap;
-        sum += s[0] << 2;
-        s += src_wrap;
-        sum += s[0] << 1;
-        s += src_wrap;
-        sum += s[0] << 2;
-        s += src_wrap;
-        sum += -s[0];
+        sum = -lum_m4[0];
+        sum += lum_m3[0] << 2;
+        sum += lum_m2[0] << 1;
+        sum += lum_m1[0] << 2;
+        sum += -lum[0];
         dst[0] = cm[(sum + 4) >> 3];
+        lum_m4++;
+        lum_m3++;
+        lum_m2++;
+        lum_m1++;
+        lum++;
         dst++;
-        src++;
     }
+#else
+
+    {
+        mmx_t rounder;
+        rounder.uw[0]=4;
+        rounder.uw[1]=4;
+        rounder.uw[2]=4;
+        rounder.uw[3]=4;
+        pxor_r2r(mm7,mm7);
+        movq_m2r(rounder,mm6);
+    }
+    for (;size > 3; size-=4) {
+        DEINT_LINE_LUM
+        lum_m4+=4;
+        lum_m3+=4;
+        lum_m2+=4;
+        lum_m1+=4;
+        lum+=4;
+        dst+=4;
+    }
+#endif
+}
+static void deinterlace_line_inplace(uint8_t *lum_m4, uint8_t *lum_m3, uint8_t *lum_m2, uint8_t *lum_m1, uint8_t *lum,
+                             int size)
+{
+#ifndef HAVE_MMX
+    uint8_t *cm = cropTbl + MAX_NEG_CROP;
+    int sum;
+
+    for(;size > 0;size--) {
+        sum = -lum_m4[0];
+        sum += lum_m3[0] << 2;
+        sum += lum_m2[0] << 1;
+        lum_m4[0]=lum_m2[0];
+        sum += lum_m1[0] << 2;
+        sum += -lum[0];
+        lum_m2[0] = cm[(sum + 4) >> 3];
+        lum_m4++;
+        lum_m3++;
+        lum_m2++;
+        lum_m1++;
+        lum++;
+    }
+#else
+
+    {
+        mmx_t rounder;
+        rounder.uw[0]=4;
+        rounder.uw[1]=4;
+        rounder.uw[2]=4;
+        rounder.uw[3]=4;
+        pxor_r2r(mm7,mm7);
+        movq_m2r(rounder,mm6);
+    }
+    for (;size > 3; size-=4) {
+        DEINT_INPLACE_LINE_LUM
+        lum_m4+=4;
+        lum_m3+=4;
+        lum_m2+=4;
+        lum_m1+=4;
+        lum+=4;
+    }
+#endif
 }
 
 /* deinterlacing : 2 temporal taps, 3 spatial taps linear filter. The
    top field is copied as is, but the bottom field is deinterlaced
    against the top field. */
-static void deinterlace_bottom_field(UINT8 *dst, int dst_wrap,
-                                     UINT8 *src1, int src_wrap,
-                                     int width, int height)
+static void deinterlace_bottom_field(uint8_t *dst, int dst_wrap,
+                                    uint8_t *src1, int src_wrap,
+                                    int width, int height)
 {
-    UINT8 *src, *ptr;
-    int y, y1, i;
-    UINT8 *buf;
+    uint8_t *src_m2, *src_m1, *src_0, *src_p1, *src_p2;
+    int y;
 
-    buf = (UINT8*)av_malloc(5 * width);
-
-    src = src1;
-    for(y=0;y<height;y+=2) {
-        /* copy top field line */
-        memcpy(dst, src, width);
+    src_m2 = src1;
+    src_m1 = src1;
+    src_0=&src_m1[src_wrap];
+    src_p1=&src_0[src_wrap];
+    src_p2=&src_p1[src_wrap];
+    for(y=0;y<(height-2);y+=2) {
+        memcpy(dst,src_m1,width);
         dst += dst_wrap;
-        src += (1 - 2) * src_wrap;
-        y1 = y - 2;
-        if (y1 >= 0 && (y1 + 4) < height) {
-            /* fast case : no edges */
-            deinterlace_line(dst, src, src_wrap, width);
-        } else {
-            /* in order to use the same function, we use an intermediate buffer */
-            ptr = buf;
-            for(i=0;i<5;i++) {
-                if (y1 < 0)
-                    memcpy(ptr, src1, width);
-                else if (y1 >= height) 
-                    memcpy(ptr, src1 + (height - 1) * src_wrap, width);
-                else
-                    memcpy(ptr, src1 + y1 * src_wrap, width);
-                y1++;
-                ptr += width;
-            }
-            deinterlace_line(dst, buf, width, width);
-        }
+        deinterlace_line(dst,src_m2,src_m1,src_0,src_p1,src_p2,width);
+        src_m2 = src_0;
+        src_m1 = src_p1;
+        src_0 = src_p2;
+        src_p1 += 2*src_wrap;
+        src_p2 += 2*src_wrap;
         dst += dst_wrap;
-        src += (2 + 1) * src_wrap;
     }
+    memcpy(dst,src_m1,width);
+    dst += dst_wrap;
+    /* do last line */
+    deinterlace_line(dst,src_m2,src_m1,src_0,src_0,src_0,width);
+}
+
+static void deinterlace_bottom_field_inplace(uint8_t *src1, int src_wrap,
+                                     int width, int height)
+{
+    uint8_t *src_m1, *src_0, *src_p1, *src_p2;
+    int y;
+    uint8_t *buf;
+    buf = (uint8_t*)av_malloc(width);
+
+    src_m1 = src1;
+    memcpy(buf,src_m1,width);
+    src_0=&src_m1[src_wrap];
+    src_p1=&src_0[src_wrap];
+    src_p2=&src_p1[src_wrap];
+    for(y=0;y<(height-2);y+=2) {
+        deinterlace_line_inplace(buf,src_m1,src_0,src_p1,src_p2,width);
+        src_m1 = src_p1;
+        src_0 = src_p2;
+        src_p1 += 2*src_wrap;
+        src_p2 += 2*src_wrap;
+    }
+    /* do last line */
+    deinterlace_line_inplace(buf,src_m1,src_0,src_0,src_0,width);
     av_free(buf);
 }
 
 
-/* deinterlace, return -1 if format not handled */
+/* deinterlace - if not supported return -1 */
 int avpicture_deinterlace(AVPicture *dst, AVPicture *src,
                           int pix_fmt, int width, int height)
 {
@@ -525,9 +1608,9 @@ int avpicture_deinterlace(AVPicture *dst, AVPicture *src,
         pix_fmt != PIX_FMT_YUV422P &&
         pix_fmt != PIX_FMT_YUV444P)
         return -1;
-    if ((width & 1) != 0 || (height & 3) != 0)
+    if ((width & 3) != 0 || (height & 3) != 0)
         return -1;
-    
+
     for(i=0;i<3;i++) {
         if (i == 1) {
             switch(pix_fmt) {
@@ -542,9 +1625,19 @@ int avpicture_deinterlace(AVPicture *dst, AVPicture *src,
                 break;
             }
         }
-        deinterlace_bottom_field(dst->data[i], dst->linesize[i],
-                                 src->data[i], src->linesize[i],
+        if (src == dst) {
+            deinterlace_bottom_field_inplace(src->data[i], src->linesize[i],
                                  width, height);
+        } else {
+            deinterlace_bottom_field(dst->data[i],dst->linesize[i],
+                                        src->data[i], src->linesize[i],
+                                        width, height);
+        }
     }
+#ifdef HAVE_MMX
+    emms();
+#endif
     return 0;
 }
+
+#undef FIX