| Message ID | 20230802203334.23640-1-leo.izen@gmail.com |
|---|---|
| Headers | show |
| Series | JPEG XL Parser (and bug fixes) | expand |
On 8/2/23 16:33, Leo Izen wrote: > Changes from v6: > - Added dummy stub libavformat/jpegxl_parse.c to fix shared builds Bumping for review. I know the commit message in patch 3/5 is wrong, but I can fix that on merge, or v8 if a v8 is necessary. - Leo Izen
On 8/9/23 17:56, Leo Izen wrote: > On 8/2/23 16:33, Leo Izen wrote: >> Changes from v6: >> - Added dummy stub libavformat/jpegxl_parse.c to fix shared builds > > Bumping for review. I know the commit message in patch 3/5 is wrong, but > I can fix that on merge, or v8 if a v8 is necessary. > > - Leo Izen > Bumping again. This is a lot of code, so I don't want to threaten to merge it, so if you take a look and having nothing more to add then a LGTM would be appreciated, thanks. - Leo Izen
On 8/2/23 16:33, Leo Izen wrote: > Changes from v6: > - Added dummy stub libavformat/jpegxl_parse.c to fix shared builds > > Changes from v5: > - Attached an extra commit to fix existing bug with libjxldec > - Made various changes based on comments by Andreas Reinhardt > -- removed jpegxl_parse.c from avformat, and jpegxl_parse.o from avformat/Makefile/OBJS > -- checked for failure in init_vlc_lengths > -- used a macro to initialize VLCElem structs > -- used heap allocations for a buffer of size 250k, instead of stack > -- renamed "code" to "len" > - Fix demuxer to avoid using avio_size, breaking piped input > - Fix some parser bugs discovered during more extensive testing > > Changes from v4: > - Added an entropy decoder and full parser, which finds the > boundaries between files correctly > - Removed unnecessary logging in libjxldec > > Changes from v3: > > - Don't remove AV_CODEC_CAP_DR1 from libjxldec > - jpegxl_parse.o added to STLIBOBJS in avcodec/Makefile > - add pipe demuxer to avformat/Makefile's SHLIBOBJS > > Changes from v2: > > - Fix libjxldec to work with packets that are smaller than one frame > - Change how code is shared between libavcodec and libavformat to be more sensible. > - Fix the parser to work with large headers that proceed the codestream in a container format > (for example, if several-KB Exif boxes preceed the codestream.) > - Modify the parser to set width/height instead of avctx > Note: avctx->pix_fmt and s->format are both set, because otherwise the CLI tools won't print > the pixel format without libjxl enabled. > - Update the fate test based on the new parser's packetization > > This test relies on FATE samples that haven't been uploaded yet. To test, unzip > the following zipfile[1] in the FATE_SAMPLES directory, placing the test images in jxl/. > > [1]: https://buzo.us/y.zip > > sha256sum: 43a2eeb0dfdf471b47a9fdfb1653974fa156ceceb776891cc137569a8ebf0e75 > signature: https://buzo.us/R.asc > > Leo Izen (5): > avcodec/libjxldec: fix errors when decoding grayscale after rgb > avcodec/libjxldec: use internal AVFrame as buffered space > avcodec/jpegxl_parser: add JPEG XL parser > avformat/jpegxl: remove jpegxl_probe, instead call > avcodec/jpegxl_parse > fate/jpegxl_anim: add demuxer fate test for jpegxl_anim > > libavcodec/Makefile | 3 + > libavcodec/jpegxl.h | 94 ++ > libavcodec/jpegxl_parse.c | 520 ++++++ > libavcodec/jpegxl_parse.h | 72 + > libavcodec/jpegxl_parser.c | 1477 +++++++++++++++++ > libavcodec/libjxldec.c | 41 +- > libavcodec/parsers.c | 1 + > libavcodec/version.h | 2 +- > libavformat/Makefile | 6 +- > libavformat/img2dec.c | 4 +- > libavformat/jpegxl_anim_dec.c | 132 +- > .../{jpegxl_probe.h => jpegxl_parse.c} | 21 +- > libavformat/jpegxl_probe.c | 412 ----- > libavformat/version.h | 2 +- > tests/Makefile | 1 + > tests/fate/jxl.mak | 16 + > tests/ref/fate/jxl-anim-demux-belgium | 6 + > tests/ref/fate/jxl-anim-demux-icos4d | 6 + > tests/ref/fate/jxl-anim-demux-lenna256 | 6 + > tests/ref/fate/jxl-anim-demux-newton | 6 + > 20 files changed, 2273 insertions(+), 555 deletions(-) > create mode 100644 libavcodec/jpegxl.h > create mode 100644 libavcodec/jpegxl_parse.c > create mode 100644 libavcodec/jpegxl_parse.h > create mode 100644 libavcodec/jpegxl_parser.c > rename libavformat/{jpegxl_probe.h => jpegxl_parse.c} (55%) > delete mode 100644 libavformat/jpegxl_probe.c > create mode 100644 tests/fate/jxl.mak > create mode 100644 tests/ref/fate/jxl-anim-demux-belgium > create mode 100644 tests/ref/fate/jxl-anim-demux-icos4d > create mode 100644 tests/ref/fate/jxl-anim-demux-lenna256 > create mode 100644 tests/ref/fate/jxl-anim-demux-newton > Merging tomorrow as this has been through several review cycles. - Leo Izen
On 8/24/23 15:22, Leo Izen wrote: > On 8/2/23 16:33, Leo Izen wrote: >> Changes from v6: >> - Added dummy stub libavformat/jpegxl_parse.c to fix shared builds >> >> Changes from v5: >> - Attached an extra commit to fix existing bug with libjxldec >> - Made various changes based on comments by Andreas Reinhardt >> -- removed jpegxl_parse.c from avformat, and jpegxl_parse.o from >> avformat/Makefile/OBJS >> -- checked for failure in init_vlc_lengths >> -- used a macro to initialize VLCElem structs >> -- used heap allocations for a buffer of size 250k, instead of stack >> -- renamed "code" to "len" >> - Fix demuxer to avoid using avio_size, breaking piped input >> - Fix some parser bugs discovered during more extensive testing >> >> Changes from v4: >> - Added an entropy decoder and full parser, which finds the >> boundaries between files correctly >> - Removed unnecessary logging in libjxldec >> >> Changes from v3: >> >> - Don't remove AV_CODEC_CAP_DR1 from libjxldec >> - jpegxl_parse.o added to STLIBOBJS in avcodec/Makefile >> - add pipe demuxer to avformat/Makefile's SHLIBOBJS >> >> Changes from v2: >> >> - Fix libjxldec to work with packets that are smaller than one frame >> - Change how code is shared between libavcodec and libavformat to be >> more sensible. >> - Fix the parser to work with large headers that proceed the >> codestream in a container format >> (for example, if several-KB Exif boxes preceed the codestream.) >> - Modify the parser to set width/height instead of avctx >> Note: avctx->pix_fmt and s->format are both set, because otherwise >> the CLI tools won't print >> the pixel format without libjxl enabled. >> - Update the fate test based on the new parser's packetization >> >> This test relies on FATE samples that haven't been uploaded yet. To >> test, unzip >> the following zipfile[1] in the FATE_SAMPLES directory, placing the >> test images in jxl/. >> >> [1]: https://buzo.us/y.zip >> >> sha256sum: >> 43a2eeb0dfdf471b47a9fdfb1653974fa156ceceb776891cc137569a8ebf0e75 >> signature: https://buzo.us/R.asc >> >> Leo Izen (5): >> avcodec/libjxldec: fix errors when decoding grayscale after rgb >> avcodec/libjxldec: use internal AVFrame as buffered space >> avcodec/jpegxl_parser: add JPEG XL parser >> avformat/jpegxl: remove jpegxl_probe, instead call >> avcodec/jpegxl_parse >> fate/jpegxl_anim: add demuxer fate test for jpegxl_anim >> >> libavcodec/Makefile | 3 + >> libavcodec/jpegxl.h | 94 ++ >> libavcodec/jpegxl_parse.c | 520 ++++++ >> libavcodec/jpegxl_parse.h | 72 + >> libavcodec/jpegxl_parser.c | 1477 +++++++++++++++++ >> libavcodec/libjxldec.c | 41 +- >> libavcodec/parsers.c | 1 + >> libavcodec/version.h | 2 +- >> libavformat/Makefile | 6 +- >> libavformat/img2dec.c | 4 +- >> libavformat/jpegxl_anim_dec.c | 132 +- >> .../{jpegxl_probe.h => jpegxl_parse.c} | 21 +- >> libavformat/jpegxl_probe.c | 412 ----- >> libavformat/version.h | 2 +- >> tests/Makefile | 1 + >> tests/fate/jxl.mak | 16 + >> tests/ref/fate/jxl-anim-demux-belgium | 6 + >> tests/ref/fate/jxl-anim-demux-icos4d | 6 + >> tests/ref/fate/jxl-anim-demux-lenna256 | 6 + >> tests/ref/fate/jxl-anim-demux-newton | 6 + >> 20 files changed, 2273 insertions(+), 555 deletions(-) >> create mode 100644 libavcodec/jpegxl.h >> create mode 100644 libavcodec/jpegxl_parse.c >> create mode 100644 libavcodec/jpegxl_parse.h >> create mode 100644 libavcodec/jpegxl_parser.c >> rename libavformat/{jpegxl_probe.h => jpegxl_parse.c} (55%) >> delete mode 100644 libavformat/jpegxl_probe.c >> create mode 100644 tests/fate/jxl.mak >> create mode 100644 tests/ref/fate/jxl-anim-demux-belgium >> create mode 100644 tests/ref/fate/jxl-anim-demux-icos4d >> create mode 100644 tests/ref/fate/jxl-anim-demux-lenna256 >> create mode 100644 tests/ref/fate/jxl-anim-demux-newton >> > > Merging tomorrow as this has been through several review cycles. > > - Leo Izen Pushed all except the last patch as the samples haven't been uploaded yet, so FATE will fail if I do that. - Leo Izen (Traneptora)
Changes from v6: - Added dummy stub libavformat/jpegxl_parse.c to fix shared builds Changes from v5: - Attached an extra commit to fix existing bug with libjxldec - Made various changes based on comments by Andreas Reinhardt -- removed jpegxl_parse.c from avformat, and jpegxl_parse.o from avformat/Makefile/OBJS -- checked for failure in init_vlc_lengths -- used a macro to initialize VLCElem structs -- used heap allocations for a buffer of size 250k, instead of stack -- renamed "code" to "len" - Fix demuxer to avoid using avio_size, breaking piped input - Fix some parser bugs discovered during more extensive testing Changes from v4: - Added an entropy decoder and full parser, which finds the boundaries between files correctly - Removed unnecessary logging in libjxldec Changes from v3: - Don't remove AV_CODEC_CAP_DR1 from libjxldec - jpegxl_parse.o added to STLIBOBJS in avcodec/Makefile - add pipe demuxer to avformat/Makefile's SHLIBOBJS Changes from v2: - Fix libjxldec to work with packets that are smaller than one frame - Change how code is shared between libavcodec and libavformat to be more sensible. - Fix the parser to work with large headers that proceed the codestream in a container format (for example, if several-KB Exif boxes preceed the codestream.) - Modify the parser to set width/height instead of avctx Note: avctx->pix_fmt and s->format are both set, because otherwise the CLI tools won't print the pixel format without libjxl enabled. - Update the fate test based on the new parser's packetization This test relies on FATE samples that haven't been uploaded yet. To test, unzip the following zipfile[1] in the FATE_SAMPLES directory, placing the test images in jxl/. [1]: https://buzo.us/y.zip sha256sum: 43a2eeb0dfdf471b47a9fdfb1653974fa156ceceb776891cc137569a8ebf0e75 signature: https://buzo.us/R.asc Leo Izen (5): avcodec/libjxldec: fix errors when decoding grayscale after rgb avcodec/libjxldec: use internal AVFrame as buffered space avcodec/jpegxl_parser: add JPEG XL parser avformat/jpegxl: remove jpegxl_probe, instead call avcodec/jpegxl_parse fate/jpegxl_anim: add demuxer fate test for jpegxl_anim libavcodec/Makefile | 3 + libavcodec/jpegxl.h | 94 ++ libavcodec/jpegxl_parse.c | 520 ++++++ libavcodec/jpegxl_parse.h | 72 + libavcodec/jpegxl_parser.c | 1477 +++++++++++++++++ libavcodec/libjxldec.c | 41 +- libavcodec/parsers.c | 1 + libavcodec/version.h | 2 +- libavformat/Makefile | 6 +- libavformat/img2dec.c | 4 +- libavformat/jpegxl_anim_dec.c | 132 +- .../{jpegxl_probe.h => jpegxl_parse.c} | 21 +- libavformat/jpegxl_probe.c | 412 ----- libavformat/version.h | 2 +- tests/Makefile | 1 + tests/fate/jxl.mak | 16 + tests/ref/fate/jxl-anim-demux-belgium | 6 + tests/ref/fate/jxl-anim-demux-icos4d | 6 + tests/ref/fate/jxl-anim-demux-lenna256 | 6 + tests/ref/fate/jxl-anim-demux-newton | 6 + 20 files changed, 2273 insertions(+), 555 deletions(-) create mode 100644 libavcodec/jpegxl.h create mode 100644 libavcodec/jpegxl_parse.c create mode 100644 libavcodec/jpegxl_parse.h create mode 100644 libavcodec/jpegxl_parser.c rename libavformat/{jpegxl_probe.h => jpegxl_parse.c} (55%) delete mode 100644 libavformat/jpegxl_probe.c create mode 100644 tests/fate/jxl.mak create mode 100644 tests/ref/fate/jxl-anim-demux-belgium create mode 100644 tests/ref/fate/jxl-anim-demux-icos4d create mode 100644 tests/ref/fate/jxl-anim-demux-lenna256 create mode 100644 tests/ref/fate/jxl-anim-demux-newton