From: Mark Reid <mindmark@gmail.com>
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Subject: Re: [FFmpeg-devel] [PATCH] avcodec/tiff: add support for decoding compressed rgb floating point formats
Date: Sat, 1 Oct 2022 15:23:27 -0700
Message-ID: <CA+anCR=_FoL9TcaodCqB_NnODHZaXYXxBnD04JPPkP5CroeHjA@mail.gmail.com> (raw)
In-Reply-To: <GV1P250MB07378E19FCDBAECAB6EB68AB8F599@GV1P250MB0737.EURP250.PROD.OUTLOOK.COM>
On Sat, Oct 1, 2022 at 7:22 AM Andreas Rheinhardt <
andreas.rheinhardt@outlook.com> wrote:
> mindmark@gmail.com:
> > From: Mark Reid <mindmark@gmail.com>
> >
> > floating point uses a slightly different predictor technique describe
> here
> > http://chriscox.org/TIFFTN3d1.pdf
> >
> > Here is a link the test files, if someone could add them to fate me
> > https://www.dropbox.com/s/fg59h2os4gb4wug/tiff_fate_samples.zip
> >
> >
> > ---
> > libavcodec/tiff.c | 66 +++++++++++++++++++++-
> > tests/fate/image.mak | 18 ++++++
> > tests/ref/fate/tiff-lzw-rgbaf32le | 6 ++
> > tests/ref/fate/tiff-lzw-rgbf32le | 6 ++
> > tests/ref/fate/tiff-uncompressed-rgbaf32le | 6 ++
> > tests/ref/fate/tiff-uncompressed-rgbf32le | 6 ++
> > tests/ref/fate/tiff-zip-rgbaf32le | 6 ++
> > tests/ref/fate/tiff-zip-rgbf32le | 6 ++
> > 8 files changed, 119 insertions(+), 1 deletion(-)
> > create mode 100644 tests/ref/fate/tiff-lzw-rgbaf32le
> > create mode 100644 tests/ref/fate/tiff-lzw-rgbf32le
> > create mode 100644 tests/ref/fate/tiff-uncompressed-rgbaf32le
> > create mode 100644 tests/ref/fate/tiff-uncompressed-rgbf32le
> > create mode 100644 tests/ref/fate/tiff-zip-rgbaf32le
> > create mode 100644 tests/ref/fate/tiff-zip-rgbf32le
> >
> > diff --git a/libavcodec/tiff.c b/libavcodec/tiff.c
> > index 3a610ada85..c1d07f8c3e 100644
> > --- a/libavcodec/tiff.c
> > +++ b/libavcodec/tiff.c
> > @@ -1913,7 +1913,7 @@ static int decode_frame(AVCodecContext *avctx,
> AVFrame *p,
> > TiffContext *const s = avctx->priv_data;
> > unsigned off, last_off = 0;
> > int le, ret, plane, planes;
> > - int i, j, entries, stride;
> > + int i, j, k, entries, stride;
>
> Don't add another loop variable. We support "for (int k = 0;" nowadays.
>
> > unsigned soff, ssize;
> > uint8_t *dst;
> > GetByteContext stripsizes;
> > @@ -2249,6 +2249,70 @@ again:
> > }
> > }
> >
> > + /* Floating point predictor
> > + TIFF Technical Note 3 http://chriscox.org/TIFFTN3d1.pdf */
> > + if (s->predictor == 3) {
> > + int channels = s->bppcount;
> > + int lane_offset;
> > + uint8_t *tmpbuf;
> > + int bpc;
> > +
> > + dst = five_planes ? five_planes : p->data[plane];
> > + soff = s->bpp >> 3;
> > + if (s->planar) {
> > + soff = FFMAX(soff / s->bppcount, 1);
> > + channels = 1;
> > + }
> > + ssize = s->width * soff;
> > + bpc = FFMAX(soff / s->bppcount, 1); /* bytes per component
> */
> > + lane_offset = s->width * channels;
> > +
> > + tmpbuf = (uint8_t*)av_malloc(ssize);
>
> unnecessary cast.
>
> > + if (!tmpbuf)
> > + return AVERROR(ENOMEM);
> > +
> > + if (s->avctx->pix_fmt == AV_PIX_FMT_RGBF32LE ||
> > + s->avctx->pix_fmt == AV_PIX_FMT_RGBAF32LE) {
> > + for (i = 0; i < decoded_height; i++) {
> > + /* copy first sample byte for each channel */
> > + for (j = 0; j < channels; j++)
> > + tmpbuf[j] = dst[j];
> > +
> > + /* decode horizontal differences */
> > + for (j = channels; j < ssize; j++)
> > + tmpbuf[j] = dst[j] + tmpbuf[j-channels];
> > +
> > + /* combine shuffled bytes from their sepearate
> lanes */
>
> ^
>
> Did you not like this comment? I can remove it, I felt the code needed
some explanation and I don't think the comment is the best.
The docs example code refers to these lanes as rowIncrements, which I
thought was a confusing name too.
Each byte of every floating point value in a row of pixels is split and
combined into separate groups.
A group of all the sign/exponents bytes in the row and groups for each of
the upper, mid, and lower mantissa bytes in the row.
Maybe group_size is a better name than lane_offset now that I typed that
out.
> > + for (j = 0; j < lane_offset; j++) {
> > + for (k = 0; k < bpc; k++) {
> > + dst[bpc * j + k] = tmpbuf[(bpc - k - 1) *
> lane_offset + j];
> > + }
> > + }
> > + dst += stride;
> > + }
> > + } else if (s->avctx->pix_fmt == AV_PIX_FMT_RGBF32BE ||
> > + s->avctx->pix_fmt == AV_PIX_FMT_RGBAF32BE) {
> > + /* same as LE only the shuffle at the end is reversed */
> > + for (i = 0; i < decoded_height; i++) {
> > + for (j = 0; j < channels; j++)
> > + tmpbuf[j] = dst[j];
> > +
> > + for (j = channels; j < ssize; j++)
> > + tmpbuf[j] = dst[j] + tmpbuf[j-channels];
> > +
> > + for (j = 0; j < lane_offset; j++) {
> > + for (k = 0; k < bpc; k++) {
> > + dst[bpc * j + k] = tmpbuf[k * lane_offset +
> j];
> > + }
> > + }
> > + dst += stride;
> > + }
> > + } else {
> > + av_log(s->avctx, AV_LOG_ERROR, "unsupported floating
> point pixel format\n");
> > + }
> > + av_free(tmpbuf);
> > + }
> > +
> > if (s->photometric == TIFF_PHOTOMETRIC_WHITE_IS_ZERO) {
> > int c = (s->avctx->pix_fmt == AV_PIX_FMT_PAL8 ? (1<<s->bpp)
> - 1 : 255);
> > dst = p->data[plane];
> > diff --git a/tests/fate/image.mak b/tests/fate/image.mak
> > index 03e794dc48..971531520d 100644
> > --- a/tests/fate/image.mak
> > +++ b/tests/fate/image.mak
> > @@ -501,6 +501,24 @@ fate-tiff-fax-g3: CMD = framecrc -i
> $(TARGET_SAMPLES)/CCITT_fax/G31D.TIF
> > FATE_TIFF += fate-tiff-fax-g3s
> > fate-tiff-fax-g3s: CMD = framecrc -i
> $(TARGET_SAMPLES)/CCITT_fax/G31DS.TIF
> >
> > +FATE_TIFF += fate-tiff-uncompressed-rgbf32le
> > +fate-tiff-uncompressed-rgbf32le: CMD = framecrc -i
> $(TARGET_SAMPLES)/tiff/uncompressed_rgbf32le.tif
> > +
> > +FATE_TIFF += fate-tiff-uncompressed-rgbaf32le
> > +fate-tiff-uncompressed-rgbaf32le: CMD = framecrc -i
> $(TARGET_SAMPLES)/tiff/uncompressed_rgbaf32le.tif
> > +
> > +FATE_TIFF += fate-tiff-lzw-rgbf32le
> > +fate-tiff-lzw-rgbf32le: CMD = framecrc -i
> $(TARGET_SAMPLES)/tiff/lzw_rgbf32le.tif
> > +
> > +FATE_TIFF += fate-tiff-lzw-rgbaf32le
> > +fate-tiff-lzw-rgbaf32le: CMD = framecrc -i
> $(TARGET_SAMPLES)/tiff/lzw_rgbaf32le.tif
> > +
> > +FATE_TIFF += fate-tiff-zip-rgbf32le
> > +fate-tiff-zip-rgbf32le: CMD = framecrc -i
> $(TARGET_SAMPLES)/tiff/zip_rgbf32le.tif
> > +
> > +FATE_TIFF += fate-tiff-zip-rgbaf32le
> > +fate-tiff-zip-rgbaf32le: CMD = framecrc -i
> $(TARGET_SAMPLES)/tiff/zip_rgbaf32le.tif
> > +
> > FATE_TIFF-$(call DEMDEC, IMAGE2, TIFF) += $(FATE_TIFF)
>
> I already wanted to write that framecrc tests don't really work with
> floating point formats, but then I saw that there is no real floating
> point code in this patch. Very weird for a floating point format.
> (Apart from that: You want to use FRAMECRC instead of DEMDEC.)
>
>
yes there is no floating point calculations involved, the process should
always be bitexact.
you mean like this right?
FATE_TIFF-$(call FRAMECRC, IMAGE2, TIFF) += $(FATE_TIFF)
Thanks for the quick review!
>
> > FATE_IMAGE_FRAMECRC += $(FATE_TIFF-yes)
> > diff --git a/tests/ref/fate/tiff-lzw-rgbaf32le
> b/tests/ref/fate/tiff-lzw-rgbaf32le
> > new file mode 100644
> > index 0000000000..c99aa02ef0
> > --- /dev/null
> > +++ b/tests/ref/fate/tiff-lzw-rgbaf32le
> > @@ -0,0 +1,6 @@
> > +#tb 0: 1/25
> > +#media_type 0: video
> > +#codec_id 0: rawvideo
> > +#dimensions 0: 8x8
> > +#sar 0: 0/1
> > +0, 0, 0, 1, 1024, 0x877e1d5f
> > diff --git a/tests/ref/fate/tiff-lzw-rgbf32le
> b/tests/ref/fate/tiff-lzw-rgbf32le
> > new file mode 100644
> > index 0000000000..a6d3fabfda
> > --- /dev/null
> > +++ b/tests/ref/fate/tiff-lzw-rgbf32le
> > @@ -0,0 +1,6 @@
> > +#tb 0: 1/25
> > +#media_type 0: video
> > +#codec_id 0: rawvideo
> > +#dimensions 0: 8x8
> > +#sar 0: 0/1
> > +0, 0, 0, 1, 768, 0xad26ed90
> > diff --git a/tests/ref/fate/tiff-uncompressed-rgbaf32le
> b/tests/ref/fate/tiff-uncompressed-rgbaf32le
> > new file mode 100644
> > index 0000000000..c99aa02ef0
> > --- /dev/null
> > +++ b/tests/ref/fate/tiff-uncompressed-rgbaf32le
> > @@ -0,0 +1,6 @@
> > +#tb 0: 1/25
> > +#media_type 0: video
> > +#codec_id 0: rawvideo
> > +#dimensions 0: 8x8
> > +#sar 0: 0/1
> > +0, 0, 0, 1, 1024, 0x877e1d5f
> > diff --git a/tests/ref/fate/tiff-uncompressed-rgbf32le
> b/tests/ref/fate/tiff-uncompressed-rgbf32le
> > new file mode 100644
> > index 0000000000..a6d3fabfda
> > --- /dev/null
> > +++ b/tests/ref/fate/tiff-uncompressed-rgbf32le
> > @@ -0,0 +1,6 @@
> > +#tb 0: 1/25
> > +#media_type 0: video
> > +#codec_id 0: rawvideo
> > +#dimensions 0: 8x8
> > +#sar 0: 0/1
> > +0, 0, 0, 1, 768, 0xad26ed90
> > diff --git a/tests/ref/fate/tiff-zip-rgbaf32le
> b/tests/ref/fate/tiff-zip-rgbaf32le
> > new file mode 100644
> > index 0000000000..c99aa02ef0
> > --- /dev/null
> > +++ b/tests/ref/fate/tiff-zip-rgbaf32le
> > @@ -0,0 +1,6 @@
> > +#tb 0: 1/25
> > +#media_type 0: video
> > +#codec_id 0: rawvideo
> > +#dimensions 0: 8x8
> > +#sar 0: 0/1
> > +0, 0, 0, 1, 1024, 0x877e1d5f
> > diff --git a/tests/ref/fate/tiff-zip-rgbf32le
> b/tests/ref/fate/tiff-zip-rgbf32le
> > new file mode 100644
> > index 0000000000..a6d3fabfda
> > --- /dev/null
> > +++ b/tests/ref/fate/tiff-zip-rgbf32le
> > @@ -0,0 +1,6 @@
> > +#tb 0: 1/25
> > +#media_type 0: video
> > +#codec_id 0: rawvideo
> > +#dimensions 0: 8x8
> > +#sar 0: 0/1
> > +0, 0, 0, 1, 768, 0xad26ed90
> > --
> > 2.31.1.windows.1
> >
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
next prev parent reply other threads:[~2022-10-01 22:23 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-10-01 14:06 mindmark
2022-10-01 14:21 ` Andreas Rheinhardt
2022-10-01 22:23 ` Mark Reid [this message]
2022-10-01 22:28 ` Andreas Rheinhardt
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CA+anCR=_FoL9TcaodCqB_NnODHZaXYXxBnD04JPPkP5CroeHjA@mail.gmail.com' \
--to=mindmark@gmail.com \
--cc=ffmpeg-devel@ffmpeg.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git