Layout talos regressions in Visual Studio 2015

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Layout talos regressions in Visual Studio 2015

Gregory Szorc-3
According to https://bugzilla.mozilla.org/show_bug.cgi?id=1254767#c2 a
number of apparently layout related Talos figures regressed on PGO builds
performed with Visual Studio 2015. This is concerning and represents a
potential showstopper for switching automation to VS2015 this release cycle.

I'm sending this email to ask for help from layout technical experts to
help identify reasons for the regressions. Links to builds you can
test/profile with are in the bug.

I'm reasonably confident that the PGO profiling and optimization process
completed successfully. But don't rule out a failure there that's causing
layout code to not get PGO benefits.

Can someone please help us understand where layout code is slower with
these builds so we can switch to VS2015 in the next week or two?

gps
_______________________________________________
dev-tech-layout mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-tech-layout
Reply | Threaded
Open this post in threaded view
|

Re: Layout talos regressions in Visual Studio 2015

Jet Villegas-2
Are there other config changes that occurred between the last run and the
run cited here?
https://bugzilla.mozilla.org/show_bug.cgi?id=1186060#c13

--Jet

On Tue, Mar 15, 2016 at 10:29 AM, Gregory Szorc <[hidden email]> wrote:

> According to https://bugzilla.mozilla.org/show_bug.cgi?id=1254767#c2 a
> number of apparently layout related Talos figures regressed on PGO builds
> performed with Visual Studio 2015. This is concerning and represents a
> potential showstopper for switching automation to VS2015 this release cycle.
>
> I'm sending this email to ask for help from layout technical experts to
> help identify reasons for the regressions. Links to builds you can
> test/profile with are in the bug.
>
> I'm reasonably confident that the PGO profiling and optimization process
> completed successfully. But don't rule out a failure there that's causing
> layout code to not get PGO benefits.
>
> Can someone please help us understand where layout code is slower with
> these builds so we can switch to VS2015 in the next week or two?
>
> gps
>
_______________________________________________
dev-tech-layout mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-tech-layout
Reply | Threaded
Open this post in threaded view
|

Re: Layout talos regressions in Visual Studio 2015

Gregory Szorc-3
I have been constantly tweaking the automation configs since that comment
was posted. However, I don't believe I've changed anything that would cause
a difference in PGO behavior from the time I first got PGO working with
VS2015 in automation.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=46e456744f3a has
the most recent PGO builds I've performed in automation. They used the
automation config that we plan on landing.

On Mon, Mar 14, 2016 at 7:50 PM, Jet Villegas <[hidden email]> wrote:

> Are there other config changes that occurred between the last run and the
> run cited here?
> https://bugzilla.mozilla.org/show_bug.cgi?id=1186060#c13
>
> --Jet
>
> On Tue, Mar 15, 2016 at 10:29 AM, Gregory Szorc <[hidden email]> wrote:
>
>> According to https://bugzilla.mozilla.org/show_bug.cgi?id=1254767#c2 a
>> number of apparently layout related Talos figures regressed on PGO builds
>> performed with Visual Studio 2015. This is concerning and represents a
>> potential showstopper for switching automation to VS2015 this release cycle.
>>
>> I'm sending this email to ask for help from layout technical experts to
>> help identify reasons for the regressions. Links to builds you can
>> test/profile with are in the bug.
>>
>> I'm reasonably confident that the PGO profiling and optimization process
>> completed successfully. But don't rule out a failure there that's causing
>> layout code to not get PGO benefits.
>>
>> Can someone please help us understand where layout code is slower with
>> these builds so we can switch to VS2015 in the next week or two?
>>
>> gps
>>
>
>
_______________________________________________
dev-tech-layout mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-tech-layout
Reply | Threaded
Open this post in threaded view
|

Re: Layout talos regressions in Visual Studio 2015

Gregory Szorc-3
In reply to this post by Gregory Szorc-3
On Mon, Mar 14, 2016 at 7:29 PM, Gregory Szorc <[hidden email]> wrote:

> According to https://bugzilla.mozilla.org/show_bug.cgi?id=1254767#c2 a
> number of apparently layout related Talos figures regressed on PGO builds
> performed with Visual Studio 2015. This is concerning and represents a
> potential showstopper for switching automation to VS2015 this release cycle.
>
> I'm sending this email to ask for help from layout technical experts to
> help identify reasons for the regressions. Links to builds you can
> test/profile with are in the bug.
>
> I'm reasonably confident that the PGO profiling and optimization process
> completed successfully. But don't rule out a failure there that's causing
> layout code to not get PGO benefits.
>
> Can someone please help us understand where layout code is slower with
> these builds so we can switch to VS2015 in the next week or two?
>

In case you aren't following the bug, for reasons we can't explain
automation decided to run Talos jobs against a non-PGO build despite -
despite Treeherder clearly showing only a PGO build for the Try push that
triggered the Talos jobs. We're scratching our heads too.

https://bugzilla.mozilla.org/show_bug.cgi?id=1254767#c9 and after contain
actual PGO results. The bad news is there are some regressions. The good
news is there aren't as many regressions and people seem to be helping with
the investigation.

I'm sorry for the wild goose chase. And thank you to those helping us chase
down why things changed.
_______________________________________________
dev-tech-layout mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-tech-layout