Just getting the first telemetry data in: https://telemetry.mozilla.org/new-pipeline/dist.html#!cumulative=0&end_date=2018-11-01&include_spill=0&keys=__none__!__none__!__none__&max_channel_version=nightly%252F65&measure=IMAGE_DECODE_SPEED_WEBP&min_channel_version=null&processType=*&product=Firefox&sanitize=1&sort_keys=submissions&start_date=2018-11-01&table=0&trim=1&use_submission_date=0 It is less than half the speed of JPEG and PNG decoding. Might be an integration problem on my side (either the building of the library itself, or something in the decoder).

Webp is lot more complex that jpeg and png, so it's normal to be slower with the same CPU. Webp it's less complex that HEIC/HEIF and AV1F, so it's normal to be faster with the same CPU.

But maybe half the speed is to much...

> But maybe half the speed is to much... I don't see why. On any somewhat-modern hardware (as in: doesn't belong in a museum) this is hardly something any user would notice.

I don't notice a difference on giphy.com or soeren-hentzschel.at, but on http://littlesvr.ca/apng/gif_apng_webp3.html I still have a huge delay to load the images. So I wonder if it may depend on an issue on the server?

(In reply to sjw from comment #4) > I don't notice a difference on giphy.com or soeren-hentzschel.at, but on > http://littlesvr.ca/apng/gif_apng_webp3.html I still have a huge delay to > load the images. > So I wonder if it may depend on an issue on the server? For me in that page FF2018-11-05 not allways load webp images. Chrome and Edge load fast and always fine.

(In reply to Marco Sousa from comment #5) > (In reply to sjw from comment #4) > > I don't notice a difference on giphy.com or soeren-hentzschel.at, but on > > http://littlesvr.ca/apng/gif_apng_webp3.html I still have a huge delay to > > load the images. > > So I wonder if it may depend on an issue on the server? > > For me in that page FF2018-11-05 not allways load webp images. > Chrome and Edge load fast and always fine. Bug 1504448 should resolve that. I think. There are some buffering related problems. The telemetry will probably shift somewhat after that patch lands.

If some concrete examples come up versus Chrome, please comment. Local testing of decoding on the command line vs profiling firefox did not seem to show a significant difference. The buffering problem before was causing some weirdness that could explain the apparent performance (at least in part).