Closed Bug 857031 Opened 12 years ago Closed 3 years ago

pdf.js performance on an ice map is not acceptable

Categories

(Firefox :: PDF Viewer, enhancement, P1)

Product:
Firefox
Component:
PDF Viewer
Platform:
x86_64
Linux
Type:
enhancement

Tracking

()

Status:
RESOLVED FIXED
Milestone:
101 Branch
Project Flags:
Performance Impact high
Tracking Flags:
Tracking Status
firefox101 --- affected

People

(Reporter: smaug, Assigned: calixte)

References

Details

(Keywords: perf:pageload, Whiteboard: [pdfjs-performance])

Attachments

(6 files)

test
(deleted), application/pdf
Details
Link to GitHub pull-request: https://github.com/mozilla/pdf.js/pull/14754
(deleted), text/x-github-pull-request
Details
Link to GitHub master push: https://github.com/mozilla/pdf.js/commit/08e1abe68e5df09c753665739e09c4529e1f6e61
(deleted), text/x-github-pull-request
Details
Link to GitHub pull-request: https://github.com/mozilla/pdf.js/pull/14777
(deleted), text/x-github-pull-request
Details
Bug 857031 - Add a performance test for the rendering of a pdf with a lot of image masks r=#pdfjs-reviewers
(deleted), text/x-phabricator-request
Details
Link to GitHub master push: https://github.com/mozilla/pdf.js/commit/379125c37ff58cfb86e3bbd2eca5760fb2fcb061
(deleted), text/x-github-pull-request
Details
Attached file test (deleted) —
Looks like there are many layers in the map. Pdf.js really can't handle this too well. Other pdf viewers seem to behave significantly faster.
Priority: -- → P3
Whiteboard: [pdfjs-c-performance]
How slow is slow? It takes about 4--5 seconds to render for me, which isn't so bad for a detailed vector map like this one.
Flags: needinfo?(bugs)
Compare to native pdf viewers. Especially zoom level changes are painful.
Flags: needinfo?(bugs)
Redrawing after zoom takes about 1.5s for me. evince takes about 0.5s.
Redraw after zoom takes ~3s here. Evince doesn't seem to really redraw. Is just scales + does some antialiasing.

I think this regressed, it's way slower than a few seconds.

Still pretty quick on MacOS: 2159ms for page 1.

It is pretty ok on Linux. And zoom is around the same level what gnome's native pdf viewer has.

Weird, on my Linux machine it takes around 20 seconds.

Worth to profile?

Whiteboard: [pdfjs-c-performance] → [pdfjs-performance]

hmm, that profile misses quite a bit details. Was it recorded using WebDev settings or what?
But anyhow, canvas is taking almost all the time.

The profile I got is quite different https://share.firefox.dev/3HkKDvK.
Canvas code is taking way less time (vs other stuff happening)

On Win10 it takes about 10sec+ to load and render.
https://share.firefox.dev/3g9jlMJ

Let me put this to performance triage, that way also the performance flag will be set.

This is better these days, but still not great.

Performance Impact: --- → ?
Keywords: perf:pageload

Can reproduce on mac, linux, windows. Significant difference from Chrome on speed. P1 performance page load.

Priority: P3 → P1
Whiteboard: [pdfjs-performance] → [pdfjs-performance, performance-pageload]
Performance Impact: ? → P1

(In reply to Benjamin De Kosnik [:bdekoz] from comment #15)

Can reproduce on mac, linux, windows. Significant difference from Chrome on speed. P1 performance page load.

I imagine you meant to set the Performance field to P1 and not change the Priority field.

Severity: normal → --
Type: defect → enhancement
Priority: P1 → P3
Assignee: nobody → cdenizet
Keywords: perf:pageload
Whiteboard: [pdfjs-performance, performance-pageload] → [pdfjs-performance]

Can we add this PDF to the set of PDFs we use for performance/memory tests?

Flags: needinfo?(cdenizet)
Commit merged into master by GitHub Authored by calixteman (calixteman)
Priority: P3 → P1
Pushed by cdenizet@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/666e6f622fa6 Add a performance test for the rendering of a pdf with a lot of image masks r=pdfjs-reviewers,perftest-reviewers,marco,sparky DONTBUILD

https://hg.mozilla.org/mozilla-central/rev/666e6f622fa6

Status: NEW → RESOLVED
Closed: 3 years ago
status-firefox101: --- → fixed
Resolution: --- → FIXED
Target Milestone: --- → 101 Branch
Status: RESOLVED → REOPENED
Flags: needinfo?(cdenizet)
Resolution: FIXED → ---
Commit merged into master by GitHub Authored by calixteman (calixteman)
Regressions: 1765558

This is not instantaneous like in Evince, but I think it's fast enough now.

Status: REOPENED → RESOLVED
Closed: 3 years ago → 3 years ago
Resolution: --- → FIXED
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: