[FFmpeg-devel] [PATCH] libopenjpeg: support decoding with bits per pixel greater than 8

Jean First jeanfirst at gmail.com
Thu Dec 22 22:26:21 CET 2011


> lets see what the code path would be for PIX_FMT_RGB24
> in check_image_attributes
> the compRatio should be anything but 0111111/0112121/0112222, which means
> non-equal component size rgb24 (haven't heard of one yet)
> 
> or if it's equal component size - precision is checked to be 8 to return
> rgb24:
>     switch (c0.prec) {
>     case 8:  return PIX_FMT_RGB24;
> 
> i'm fine with applying the patch, there's nothing wrong with it, but to me
> it's a bit of premature optimization.
> isnt it mostly a matter of taste?
> to my taste i'd just wait to see if someone finds an example where error is
> triggered.
> 
> On Thu, Dec 22, 2011 at 10:27 AM, Jean First <jeanfirst at gmail.com> wrote:
> 
> > On Wed Dec 21 21:26:45 2011, Alex Zhukov wrote:
> >
> >> from the top of my head: copyto8 expects the 8bit precision therefore no
> >> ">>adjust" should be needed
> >> can you provide an example video where this is not the case?
> >>
> >> On Wed, Dec 21, 2011 at 5:33 AM, Jean First<jeanfirst at gmail.com>  wrote:
> >>
> >>  Hi,
> >>>
> >>> attached patch fixes a regression introduced with commit eb511ef6.
> >>> maybe libopenjpeg_copyto8 and libopenjpeg_copyto16 need this fix too.
> >>>  Alex Zhukov, can you please check this ?
> >>>
> >>> Jean
> >>>
> >>> ---
> >>>
> >>
> > Hi Alex,
> > No atm not. I suggest to apply the patch until we have a sample that
> > triggers the error.
> > Jean

Well it's not an optimisation - it is a fix for a regression. We might wait
a long time for that sample - in the meantime this is fixed.

I added PIX_FMT_48 decoding, although the conversion from <16 bit to 16 bit
might not be the best solution. Everything above 8 bit will be treated as
16 bit.

And should I start top posting - or will you start bottom posting ;)

Jean

---
 libavcodec/libopenjpegdec.c |   32 +++++++++++++++++++++++++++++++-
 1 files changed, 31 insertions(+), 1 deletions(-)

diff --git a/libavcodec/libopenjpegdec.c b/libavcodec/libopenjpegdec.c
index baa516b..1f904da 100644
--- a/libavcodec/libopenjpegdec.c
+++ b/libavcodec/libopenjpegdec.c
@@ -54,7 +54,7 @@ static enum PixelFormat check_image_attributes(AVCodecContext *avctx, opj_image_
     case 0111111: goto libopenjpeg_yuv444_rgb;
     case 0112121: goto libopenjpeg_yuv422;
     case 0112222: goto libopenjpeg_yuv420;
-    default: return PIX_FMT_RGB24;
+    default: goto libopenjpeg_rgb;
     }
 
 libopenjpeg_yuv420:
@@ -80,6 +80,13 @@ libopenjpeg_yuv444_rgb:
     case 10: return PIX_FMT_YUV444P10;
     case 16: return PIX_FMT_YUV444P16;
     }
+
+libopenjpeg_rgb:
+    switch (c0.prec) {
+    case 8: return PIX_FMT_RGB24;
+    default: return PIX_FMT_RGB48;
+    }
+
     return PIX_FMT_RGB24;
 }
 
@@ -107,6 +114,24 @@ static inline void libopenjpeg_copy_to_packed8(AVFrame *picture, opj_image_t *im
     }
 }
 
+static inline void libopenjpeg_copy_to_packed16(AVFrame *picture, opj_image_t *image) {
+    uint16_t *img_ptr;
+    int index, x, y, c;
+    int adjust[4];
+    for (x = 0; x < image->numcomps; x++) {
+        adjust[x] = FFMAX(FFMIN(16 - image->comps[x].prec, 8), 0);
+    }
+    for (y = 0; y < picture->height; y++) {
+        index = y*picture->width;
+        img_ptr = (uint16_t*) (picture->data[0] + y*picture->linesize[0]);
+        for (x = 0; x < picture->width; x++, index++) {
+            for (c = 0; c < image->numcomps; c++) {
+                *img_ptr++ = image->comps[c].data[index] << adjust[c];
+            }
+        }
+    }
+}
+
 static inline void libopenjpeg_copyto8(AVFrame *picture, opj_image_t *image) {
     int *comp_data;
     uint8_t *img_ptr;
@@ -277,6 +302,11 @@ static int libopenjpeg_decode_frame(AVCodecContext *avctx,
             libopenjpeg_copy_to_packed8(picture, image);
         }
         break;
+    case 6:
+        if (ispacked) {
+            libopenjpeg_copy_to_packed16(picture, image);
+        }
+        break;
     default:
         av_log(avctx, AV_LOG_ERROR, "unsupported pixel size %d\n", pixel_size);
         goto done;
-- 
1.7.6.1



More information about the ffmpeg-devel mailing list