1515521 - Add memory reporters for interning machinery, and fix up GPU cache debug display

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, enhancement)

Description

[Bobby Holley (:bholley)]

We need the memory reporters, the GPU cache stuff is a ridealong.

Comment 1

[Bobby Holley (:bholley)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=5590a319b07ba49a7c035bbb36274c026c95e68a

Comment 2

[Bobby Holley (:bholley)]

Attachment: Bug 1515521 - Add memory reporters for interning data.

Comment 3

[Bobby Holley (:bholley)]

Attachment: Bug 1515521 - Support the gpu-cache debugger in Firefox.

Depends on D15044

Comment 4

[Bobby Holley (:bholley)]

Attachment: Bug 1515521 - Route all debug flag changes through the SetFlags command.

This avoids having separate paths for Firefox and wrench/vue to
enable/disable flags.

Depends on D15046

Comment 5

[Bobby Holley (:bholley)]

Attachment: Bug 1515521 - Remove buggy fresh calculation.

The |block_index| member in this case is an index into pending_blocks,
not blocks, and thus this can trigger out-of-bounds access. Given that
there's no fast way to go from a GpuCacheAddress a Block, and given that
this member is unused, we just drop it.

Depends on D15047

Comment 6

[Bobby Holley (:bholley)]

https://github.com/servo/webrender/pull/3440

Comment 7

[Kartikaya Gupta (email:[e-mail])]

Try push at https://treeherder.mozilla.org/#/jobs?repo=try&revision=1cf9bc743152bfa7e3bd555f71796ba23a907d5b is showing awsy failures. Do you know what's up with that? The sane failures happen with just the Rust patches I think (see first try push on the wr-future-update bug)

Comment 8

[Bobby Holley (:bholley)]

Hm, that's weird. The try push in comment 1 was green.

I'm away from my computer now and possibly tomorrow morning, so may not get to look at this inmediately. Might be worth diffing with the try push and seeing what changed. I made a few cosmetic review fixups to the Rust code but nothing that should have triggered such failures.

Comment 9

[Kartikaya Gupta (email:[e-mail])]

I rebased the patches from comment 1 to the same base rev as my try push and diffed them. The only changes were cosmetic (a comment and the MsgSender use statement changes) plus changes to examples/ which shouldn't matter. I kicked off a new try push with your original patches rebased at https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&revision=670244511f2bedaf250dc0b337bc99f2c05d7baf - if it's not green then likely something changed in m-c that is tickled by your patches.

Comment 10

[Kartikaya Gupta (email:[e-mail])]

So yeah something must have changed in m-c since the failures show up on your original patches rebased as well.

It looks like the code at [1] is throwing a TypeError somehow.

[1] https://searchfox.org/mozilla-central/source/testing/awsy/awsy/awsy_test_case.py#166-167

Comment 11

[Kartikaya Gupta (email:[e-mail])]

The breakage in the try push is from 1510929, which has since been backed out. The most recent try pushes are showing different failures but that might be because they don't have the C++ side changes. I'll do a fresh try push to see what the current state is.

Comment 12

[Kartikaya Gupta (email:[e-mail])]

Looking green so far.

Comment 13

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/ee1524718628
Add memory reporters for interning data. r=kvark
https://hg.mozilla.org/integration/autoland/rev/ce0f2042c1e2
Support the gpu-cache debugger in Firefox. r=kvark

Comment 14

[Cristina Coroiu [:ccoroiu]]

https://hg.mozilla.org/mozilla-central/rev/ee1524718628
https://hg.mozilla.org/mozilla-central/rev/ce0f2042c1e2