Quantcast

Better perf

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Better perf

GHC - devs mailing list

I’ve just committed this patch sequence

fb9ae288088a3eabc4e1bb4e86fa473a3881d2e2 Make FloatOut/SetLevels idemoptent on bottoming functions

995ab74b3c55fe3a0299bd94b49e948c942e76d6 Comments only

1163f4f2fe9aabd722c963497c67c5f8c71ef71b Tiny refactor

9b2c73ea8082199245bfa6a28390b70b38f87fd1 Make TH_Roles2 less fragile

9304df5230a7a29d3e992916d133e462b854e55f Fix CSE (again) on literal strings

In my final validate run (after updating to HEAD) I saw

Unexpected stat failures:

   perf/compiler/T13035.run    T13035 [stat too good] (normal)

-6.4% alloc

 

   perf/compiler/T12425.run    T12425 [stat too good] (optasm)

-6.6% alloc

 

   perf/compiler/T9675.run     T9675 [stat too good] (optasm)

-10.4% alloc

 

   perf/compiler/T1969.run     T1969 [stat too good] (normal)

-21% peak megabytes

 

   perf/space_leaks/T4029.run  T4029 [stat too good] (ghci)

-14% peak megabytes

This is good.  I did not see these in earlier validations (perhaps I did not rebuild the libraries sufficiently), so I have left them.  If Harbormaster agrees that perf has improved, could someone re-centre the numbers?  Ideally say which patch is responsible.

Simon


_______________________________________________
ghc-devs mailing list
[hidden email]
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Better perf

Ben Gamari-2
Simon Peyton Jones via ghc-devs <[hidden email]> writes:

> I've just committed this patch sequence
>
> fb9ae288088a3eabc4e1bb4e86fa473a3881d2e2 Make FloatOut/SetLevels idemoptent on bottoming functions
>
> 995ab74b3c55fe3a0299bd94b49e948c942e76d6 Comments only
>
> 1163f4f2fe9aabd722c963497c67c5f8c71ef71b Tiny refactor
>
> 9b2c73ea8082199245bfa6a28390b70b38f87fd1 Make TH_Roles2 less fragile
>
> 9304df5230a7a29d3e992916d133e462b854e55f Fix CSE (again) on literal strings
> In my final validate run (after updating to HEAD) I saw
>
> Unexpected stat failures:
>
>    perf/space_leaks/T4029.run  T4029 [stat too good] (ghci)
>
This one is rather interesting since I have also been seeing this for
quite some time locally, but Harbormaster consistently disagrees.
Moreover, it passes locally if my machine isn't under load. I
haven't yet investigated why allocations are so inconsistent.

Cheers,

- Ben


_______________________________________________
ghc-devs mailing list
[hidden email]
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

signature.asc (497 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Better perf

Joachim Breitner-2
In reply to this post by GHC - devs mailing list
Hi,

perf.haskell.org has something to say about these:

Am Montag, den 06.03.2017, 15:45 +0000 schrieb Simon Peyton Jones via ghc-devs:
> I’ve just committed this patch sequence
> fb9ae288088a3eabc4e1bb4e86fa473a3881d2e2 Make FloatOut/SetLevels idemoptent on bottoming functions

increases lambda runtime by 3%. Maybe an environment-dependent
performance cliff, given that you did not report this regression in
your nofib listing.

> 995ab74b3c55fe3a0299bd94b49e948c942e76d6 Comments only

No change reported. Good :-)

> 1163f4f2fe9aabd722c963497c67c5f8c71ef71b Tiny refactor

No change reported.

> 9b2c73ea8082199245bfa6a28390b70b38f87fd1 Make TH_Roles2 less fragile

No change reported.

> 9304df5230a7a29d3e992916d133e462b854e55f Fix CSE (again) on literal strings

This is where most of the changes are:
https://perf.haskell.org/ghc/#revision/9304df5230a7a29d3e992916d133e462b854e55f

There are some nice runtime improvements in cryptarithm1 (-12%), fasta,
integer and scs (each -3%).

But: binary-trees runtime increases by 5%.


This was your daily performance weather report.

Joachim

--
Joachim “nomeata” Breitner
  [hidden email]https://www.joachim-breitner.de/
  XMPP: [hidden email] • OpenPGP-Key: 0xF0FBF51F
  Debian Developer: [hidden email]
_______________________________________________
ghc-devs mailing list
[hidden email]
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

RE: Better perf

GHC - devs mailing list
| But: binary-trees runtime increases by 5%.

David: might you look to see if there is any obvious reason for this regression?  We could just accept it, but it's always good to know why, and to document it.

Thanks

Simon

| -----Original Message-----
| From: ghc-devs [mailto:[hidden email]] On Behalf Of Joachim
| Breitner
| Sent: 07 March 2017 10:43
| To: [hidden email]
| Subject: Re: Better perf
|
| Hi,
|
| perf.haskell.org has something to say about these:
|
| Am Montag, den 06.03.2017, 15:45 +0000 schrieb Simon Peyton Jones via
| ghc-devs:
| > I’ve just committed this patch sequence
| > fb9ae288088a3eabc4e1bb4e86fa473a3881d2e2 Make FloatOut/SetLevels
| > idemoptent on bottoming functions
|
| increases lambda runtime by 3%. Maybe an environment-dependent
| performance cliff, given that you did not report this regression in your
| nofib listing.
|
| > 995ab74b3c55fe3a0299bd94b49e948c942e76d6 Comments only
|
| No change reported. Good :-)
|
| > 1163f4f2fe9aabd722c963497c67c5f8c71ef71b Tiny refactor
|
| No change reported.
|
| > 9b2c73ea8082199245bfa6a28390b70b38f87fd1 Make TH_Roles2 less fragile
|
| No change reported.
|
| > 9304df5230a7a29d3e992916d133e462b854e55f Fix CSE (again) on literal
| > strings
|
| This is where most of the changes are:
| https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fperf.has
| kell.org%2Fghc%2F%23revision%2F9304df5230a7a29d3e992916d133e462b854e55f&d
| ata=02%7C01%7Csimonpj%40microsoft.com%7C68ad9d20aa794e70cdb708d46546c5ff%
| 7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636244802039385263&sdata=pvj
| YNwtA3NHHxF7z7bwagYX5Cjun8%2FIztPvXEO1AtdY%3D&reserved=0
|
| There are some nice runtime improvements in cryptarithm1 (-12%), fasta,
| integer and scs (each -3%).
|
| But: binary-trees runtime increases by 5%.
|
|
| This was your daily performance weather report.
|
| Joachim
|
| --
| Joachim “nomeata” Breitner
|   [hidden email]
| https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.joac
| him-
| breitner.de%2F&data=02%7C01%7Csimonpj%40microsoft.com%7C68ad9d20aa794e70c
| db708d46546c5ff%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636244802039
| 385263&sdata=hbp839x4Boa3J6gY4sOchfHGsLZeHiEsNatxsSv08iQ%3D&reserved=0
|   XMPP: [hidden email] • OpenPGP-Key: 0xF0FBF51F
|   Debian Developer: [hidden email]
_______________________________________________
ghc-devs mailing list
[hidden email]
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Better perf

Joachim Breitner-2
Hi,

Am Dienstag, den 07.03.2017, 22:55 +0000 schrieb Simon Peyton Jones via
ghc-devs:
> > But: binary-trees runtime increases by 5%.
>
> David: might you look to see if there is any obvious reason for this
> regression?  We could just accept it, but it's always good to know
> why, and to document it.

Turns out that my commit
Add rule mapFB c (λx.x) = c
fixed that regression:
https://perf.haskell.org/ghc/#revision/2fa44217c1d9722227297eefb0d6c6aed7e128ca

Maybe there is just a performance cliff there, and these jumps don’t
really mean anything.

Greetings,
Joachim

--
Joachim “nomeata” Breitner
  [hidden email]https://www.joachim-breitner.de/
  XMPP: [hidden email] • OpenPGP-Key: 0xF0FBF51F
  Debian Developer: [hidden email]
_______________________________________________
ghc-devs mailing list
[hidden email]
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

RE: Better perf

GHC - devs mailing list
Interesting!

I keep nofib/Simon-nofib-notes for per-benchmark notes on perf.

Would you like to add a note for 'binary-trees' pointing to these emails?  So if someone later is looking for perf changes in binary-trees, they have some back-pointers to chase up.

Thanks

Simon

|  -----Original Message-----
|  From: ghc-devs [mailto:[hidden email]] On Behalf Of
|  Joachim Breitner
|  Sent: 09 March 2017 15:10
|  To: [hidden email]
|  Subject: Re: Better perf
|  
|  Hi,
|  
|  Am Dienstag, den 07.03.2017, 22:55 +0000 schrieb Simon Peyton Jones
|  via
|  ghc-devs:
|  > > But: binary-trees runtime increases by 5%.
|  >
|  > David: might you look to see if there is any obvious reason for this
|  > regression?  We could just accept it, but it's always good to know
|  > why, and to document it.
|  
|  Turns out that my commit
|  Add rule mapFB c (λx.x) = c
|  fixed that regression:
|  https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fperf.
|  haskell.org%2Fghc%2F%23revision%2F2fa44217c1d9722227297eefb0d6c6aed7e1
|  28ca&data=02%7C01%7Csimonpj%40microsoft.com%7C874797bc62de4b6aecf208d4
|  66fe7720%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C6362466904875493
|  08&sdata=ysGORaGgIlhCc88uQHB3v0sh5LNWhFY06iSgnTm6Pgo%3D&reserved=0
|  
|  Maybe there is just a performance cliff there, and these jumps don’t
|  really mean anything.
|  
|  Greetings,
|  Joachim
|  
|  --
|  Joachim “nomeata” Breitner
|    [hidden email]
|  https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.j
|  oachim-
|  breitner.de%2F&data=02%7C01%7Csimonpj%40microsoft.com%7C874797bc62de4b
|  6aecf208d466fe7720%7C72f988bf86f141af91ab2d7cd011db47%7C1%7C0%7C636246
|  690487549308&sdata=HxOTFdoh4hIqXVbPCgZOwMjob%2B572b1ymbAdBEXTF6A%3D&re
|  served=0
|    XMPP: [hidden email] • OpenPGP-Key: 0xF0FBF51F
|    Debian Developer: [hidden email]
_______________________________________________
ghc-devs mailing list
[hidden email]
http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs
Loading...