| Message ID | 20170719190049.17909-1-pegro@friiks.de |
|---|---|
| State | Superseded |
| Headers | show |
On Wed, 19 Jul 2017 21:00:49 +0200 Peter Große <pegro@friiks.de> wrote: > I was able to reproduce your checksums only after removing > > --disable-optimizations --disable-mmx --disable-stripping > > from my configure command. I thought optimizations should have no > impact on the result?! After some further tests, it seems swscale (used as the auto_scaler in the test) produces slightly different output depending on whether MMX is enabled or not. When I re-add --disable-optimizations --enable-mmx --disable-stripping to my configure command line, the test succeeds. Any further comments to the patch and test case? Regards Peter
Le duodi 2 thermidor, an CCXXV, Peter Große a écrit : > After some further tests, it seems swscale (used as the auto_scaler in the > test) produces slightly different output depending on whether MMX is enabled or > not. > > When I re-add --disable-optimizations --enable-mmx --disable-stripping to my > configure command line, the test succeeds. > > Any further comments to the patch and test case? If you use testsrc2 instead of file inputs, it produces streams with the requested colorspace directly. testsrc2 does not handle alpha, though, so most of the video is transparent. I will have a look to see if it can be improved. Regards,
On 20.07.2017 17:06, Peter Große wrote: > On Wed, 19 Jul 2017 21:00:49 +0200 > Peter Große <pegro@friiks.de> wrote: > >> I was able to reproduce your checksums only after removing >> >> --disable-optimizations --disable-mmx --disable-stripping >> >> from my configure command. I thought optimizations should have no >> impact on the result?! > > After some further tests, it seems swscale (used as the auto_scaler in the > test) produces slightly different output depending on whether MMX is enabled or > not. > > When I re-add --disable-optimizations --enable-mmx --disable-stripping to my > configure command line, the test succeeds. > > Any further comments to the patch and test case? Maybe try adding -sws_flags +accurate_rnd+bitexact to CMD? Regards, Tobias
On Thu, 20 Jul 2017 17:19:36 +0200 Tobias Rapp <t.rapp@noa-archive.com> wrote: > > When I re-add --disable-optimizations --enable-mmx --disable-stripping to my > > configure command line, the test succeeds. > > > > Any further comments to the patch and test case? > > Maybe try adding -sws_flags +accurate_rnd+bitexact to CMD? I tried that, but that didn't help. Regards Peter
Le duodi 2 thermidor, an CCXXV, Nicolas George a écrit : > testsrc2 does not handle alpha, though, so most of the video is > transparent. I will have a look to see if it can be improved. Have a look at the attached patch. (Not complete, but working.) testsrc2=s=1024x768:alpha=255[a];testsrc2=alpha=128[b];[a][b]overlay=x=100:y=50 gives the expected result. Regards,
On Thu, Jul 20, 2017 at 05:38:18PM +0200, Peter Große wrote: > On Thu, 20 Jul 2017 17:19:36 +0200 > Tobias Rapp <t.rapp@noa-archive.com> wrote: > > > > When I re-add --disable-optimizations --enable-mmx --disable-stripping to my > > > configure command line, the test succeeds. > > > > > > Any further comments to the patch and test case? > > > > Maybe try adding -sws_flags +accurate_rnd+bitexact to CMD? > > I tried that, but that didn't help. make sure these flags actually get passed into the used swscale if they do, please open a ticket, i probably wont find the time to look into this in the near future but when i or someone else works on sws the next time being able to quickly find all issues affecting it shuld help "time per bug" thx [...]
diff --git a/tests/fate/filter-video.mak b/tests/fate/filter-video.mak index 53fc7a6528..b30d5b7b7c 100644 --- a/tests/fate/filter-video.mak +++ b/tests/fate/filter-video.mak @@ -351,6 +351,11 @@ FATE_FILTER_SAMPLES-$(call ALLYES, MATROSKA_DEMUXER OVERLAY_FILTER H264_DECODER fate-filter-overlay-dvdsub-2397: tests/data/filtergraphs/overlay-dvdsub-2397 fate-filter-overlay-dvdsub-2397: CMD = framecrc -flags bitexact -i $(TARGET_SAMPLES)/filter/242_4.mkv -filter_complex_script $(TARGET_PATH)/tests/data/filtergraphs/overlay-dvdsub-2397 -c:a copy +FATE_FILTER_SAMPLES-$(call ALLYES, MXF_DEMUXER OVERLAY_FILTER DNXHD_DECODER PNG_DECODER) += fate-filter-overlay-main-alpha +fate-filter-overlay-main-alpha: SRC = $(TARGET_SAMPLES)/mxf/track_01_v02.mxf +fate-filter-overlay-main-alpha: OVERLAY = $(TARGET_SAMPLES)/png1/lena-rgba.png +fate-filter-overlay-main-alpha: CMD = framecrc -flags bitexact -i $(SRC) -filter_complex "movie='$(OVERLAY)'[logo];[v:0][logo]overlay" -an + FATE_FILTER_HQX-$(call ALLYES, IMAGE2_DEMUXER PNG_DECODER HQX_FILTER) = fate-filter-hq2x fate-filter-hq3x fate-filter-hq4x FATE_FILTER_SAMPLES-yes += $(FATE_FILTER_HQX-yes) fate-filter-hq2x: CMD = framecrc -i $(TARGET_SAMPLES)/filter/pixelart%d.png -vf hqx=2 -pix_fmt bgra diff --git a/tests/ref/fate/filter-overlay-main-alpha b/tests/ref/fate/filter-overlay-main-alpha new file mode 100644 index 0000000000..95bb59ea13 --- /dev/null +++ b/tests/ref/fate/filter-overlay-main-alpha @@ -0,0 +1,15 @@ +#tb 0: 1001/24000 +#media_type 0: video +#codec_id 0: rawvideo +#dimensions 0: 1280x720 +#sar 0: 1/1 +0, 0, 0, 1, 2304000, 0x49ef8134 +0, 1, 1, 1, 2304000, 0x49ef8134 +0, 2, 2, 1, 2304000, 0x49ef8134 +0, 3, 3, 1, 2304000, 0x49ef8134 +0, 4, 4, 1, 2304000, 0x49ef8134 +0, 5, 5, 1, 2304000, 0x49ef8134 +0, 6, 6, 1, 2304000, 0x49ef8134 +0, 7, 7, 1, 2304000, 0x49ef8134 +0, 8, 8, 1, 2304000, 0x49ef8134 +0, 9, 9, 1, 2304000, 0x49ef8134
Signed-off-by: Peter Große <pegro@friiks.de> --- Updated version of the test. I was able to reproduce your checksums only after removing --disable-optimizations --disable-mmx --disable-stripping from my configure command. I thought optimizations should have no impact on the result?! tests/fate/filter-video.mak | 5 +++++ tests/ref/fate/filter-overlay-main-alpha | 15 +++++++++++++++ 2 files changed, 20 insertions(+) create mode 100644 tests/ref/fate/filter-overlay-main-alpha