This is an automatic generated email to let you know that the following patch were queued at the http://git.linuxtv.org/cgit.cgi/v4l-utils.git tree:
Subject: v4lconvert: Fix decoding of jpeg data with no vertical sub-sampling Author: Hans de Goede <[email protected]> Date: Thu Nov 20 13:03:35 2014 +0100 Our YUV output is always 2x subsampled in both vertical and horizontal direction, but some cameras generate JPEG data which is only subsampled in the horizontal direction. Since averaging the extra UV data these JPEGs contains is seomwhat slow, and UV data is not all that important anyways, we simple take every other line. Or at least that is the intent. But before this commit the code was not doing this properly, for each 16 Y input lines 1 - 16 we also get 16 UV input lines 1 - 16, but we only need 8 output lines. so we should store input line 1 or 2 in output line 1, input line 3 or 4 in output line 2, etc. Instead we were storing input lines 9 - 16 into output lines 1 - 8, which leads to some unwanted color bleeding. More over this also leads for 1920x1080 JPEG to us writing (1080 / 8 + 1) / 2 * 8 = 544 UV output lines rather then 540, this means that the last 4 U output lines overwrite the first 4 V output lines, and worse that the last 4 V output lines overrun the output buffer. So far this only lead to some wrong colors in various places, but since that we dynamically allocate the output buffer to just the right size this actually causes a crash. This commit fixes both the crash, and the wrong colors. Signed-off-by: Hans de Goede <[email protected]> (cherry picked from commit 757d7910ddf43d9a9187dddae4f51a57fb723e8d) lib/libv4lconvert/jpeg.c | 37 ++++++++++++++++++++++++------------- 1 files changed, 24 insertions(+), 13 deletions(-) --- http://git.linuxtv.org/cgit.cgi/v4l-utils.git/commit/?id=c41d66e8d4b2c0314b8fb26308c630f58c221ef9 diff --git a/lib/libv4lconvert/jpeg.c b/lib/libv4lconvert/jpeg.c index aa9cace..0142d44 100644 --- a/lib/libv4lconvert/jpeg.c +++ b/lib/libv4lconvert/jpeg.c @@ -242,23 +242,34 @@ static int decode_libjpeg_h_samp2(struct v4lconvert_data *data, y_rows[y] = ydest; ydest += width; } - for (y = 0; y < 8; y++) { - u_rows[y] = udest; - v_rows[y] = vdest; - udest += width / 2; - vdest += width / 2; + /* + * For v_samp == 1 were going to get 1 set of uv values per + * line, but we need only 1 set per 2 lines since our output + * has v_samp == 2. We store every 2 sets in 1 line, + * effectively using the second set for each output line. + */ + if (v_samp == 1) { + for (y = 0; y < 8; y++) { + u_rows[y] = udest; + v_rows[y] = vdest; + y++; + u_rows[y] = udest; + v_rows[y] = vdest; + udest += width / 2; + vdest += width / 2; + } + } else { /* v_samp == 2 */ + for (y = 0; y < 8; y++) { + u_rows[y] = udest; + v_rows[y] = vdest; + udest += width / 2; + vdest += width / 2; + } } + y = jpeg_read_raw_data(cinfo, rows, 8 * v_samp); if (y != 8 * v_samp) return -1; - - /* For v_samp == 1 were going to get another set of uv values, - but we need only 1 set since our output has v_samp == 2, so - rewind u and vdest and overwrite the previous set. */ - if (cinfo->output_scanline % 16) { - udest -= width * 8 / 2; - vdest -= width * 8 / 2; - } } return 0; } _______________________________________________ linuxtv-commits mailing list [email protected] http://www.linuxtv.org/cgi-bin/mailman/listinfo/linuxtv-commits
