Closed Bug 1478704 Opened 6 years ago Closed 6 years ago

Client Storage isn't work with Intel GPUs

Categories

(Core :: Graphics: Layers, defect)

Product:
Core
Component:
Graphics: Layers
Version:
63 Branch
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED FIXED
Milestone:
mozilla63
Tracking Flags:
Tracking Status
firefox63 --- fixed

People

(Reporter: jrmuizel, Assigned: dthayer)

References

(Blocks 1 open bug)

Details

(Whiteboard: [fxperf:p1])

Attachments

(3 files)

Bug 1478704 - Fully generalize shader generation to rect textures
(deleted), text/x-review-board-request
jrmuizel
: review+
Details
Bug 1478704 - Use GL_TEXTURE_RECTANGLE_ARB for client storage
(deleted), text/x-review-board-request
jrmuizel
: review+
Details
Bug 1478704 - Compute shader adjustments for mask separately
(deleted), text/x-review-board-request
jrmuizel
: review+
Details
I'm still seeing large chunks of time in texture upload even with client storage enabled: https://perfht.ml/2LnH5jb Perhaps the Intel driver is more strict about when it can go fast.
Assignee: nobody → dothayer
Status: NEW → ASSIGNED
Whiteboard: [fxperf:p1]
Looks like this is indeed related to using GL_TEXTURE_2D instead of GL_TEXTURE_RECTANGLE_ARB as recommended. I went to GL_TEXTURE_2D to side step an issue with some videos not showing up with GL_TEXTURE_RECTANGLE_ARB since I couldn't observe any performance difference. So I'll just be looking into the video issue now :)
Blocks: 1400787
Comment on attachment 8995374 [details] Bug 1478704 - Fully generalize shader generation to rect textures https://reviewboard.mozilla.org/r/259826/#review266852
Attachment #8995374 - Flags: review?(jmuizelaar) → review+
Comment on attachment 8995375 [details] Bug 1478704 - Use GL_TEXTURE_RECTANGLE_ARB for client storage https://reviewboard.mozilla.org/r/259828/#review266854
Attachment #8995375 - Flags: review?(jmuizelaar) → review+
Comment on attachment 8995560 [details] Bug 1478704 - Compute shader adjustments for mask separately https://reviewboard.mozilla.org/r/259956/#review267046
Attachment #8995560 - Flags: review+
Comment on attachment 8995560 [details] Bug 1478704 - Compute shader adjustments for mask separately https://reviewboard.mozilla.org/r/259956/#review267048
Attachment #8995560 - Flags: review?(jmuizelaar) → review+
Pushed by dothayer@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/3329c8f5582a Fully generalize shader generation to rect textures r=jrmuizel https://hg.mozilla.org/integration/autoland/rev/c77acf366e1c Use GL_TEXTURE_RECTANGLE_ARB for client storage r=jrmuizel https://hg.mozilla.org/integration/autoland/rev/c57be31d3c06 Compute shader adjustments for mask separately r=jrmuizel,mstange
With this build https://queue.taskcluster.net/v1/task/ZHV4IsxlSJa1ucqoPj9RCQ/runs/0/artifacts/public/build/target.dmg I still see memmove's showing up in texture upload in profiles of gsmarena. https://perfht.ml/2v8D5Is
Here's a new profile without the BasicTextureImage stuff caused by layers.acceleration.draw-fps
And the actual profile: https://perfht.ml/2vcmp2V
Pushed by dothayer@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/bdeb62b3ec57 Fully generalize shader generation to rect textures r=jrmuizel https://hg.mozilla.org/integration/autoland/rev/393d903674b9 Use GL_TEXTURE_RECTANGLE_ARB for client storage r=jrmuizel https://hg.mozilla.org/integration/autoland/rev/45250a02b2ed Compute shader adjustments for mask separately r=jrmuizel,mstange
I've filed a follow up to see if we can improve the performance on Intel further: bug 1479145.
Perf wins on OS X: == Change summary for alert #14640 (as of Fri, 27 Jul 2018 20:05:07 GMT) == Improvements: 4% tsvgx osx-10-10 opt e10s stylo 306.42 -> 292.66 3% tpaint osx-10-10 opt e10s stylo 268.45 -> 261.64 2% tresize osx-10-10 opt e10s stylo 28.19 -> 27.52 For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=14640
Flags: needinfo?(dothayer)
AWSY wins on OS X: == Change summary for alert #14641 (as of Fri, 27 Jul 2018 20:05:07 GMT) == Improvements: 14% Resident Memory osx-10-10 opt stylo 654,781,186.19 -> 560,147,258.66 For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=14641
Regressions: 1549311
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: