[GHC] #9740: D380 caused fft2 regressions

newer
Re: [GHC] #2496: Invalid Eq/Ord...

older
Re: [GHC] #2496: Invalid Eq/Ord...

GHC

28 Oct 2014 28 Oct '14

9:24 p.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.10.1 Component: Compiler | Version: 7.8.3 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Easy (less than 1 | Type of failure: Runtime hour) | performance bug Blocked By: | Test Case: Related Tickets: | Blocking: | Differential Revisions: Phab:D390 -------------------------------------+------------------------------------- When I got rid of the hand-unboxing in GHC.List, I missed the fact that a `take` helper function took an unboxed value and was therefore strict. When I changed that to a regular function, I didn't make it strict, and this caused a regression in fft2 and (I believe) also a test suite perf test. The function is properly strictified in D390. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

Show replies by date

GHC

28 Oct 28 Oct

9:25 p.m.

GHC

9:26 p.m.

GHC

29 Oct 29 Oct

3:20 a.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: patch Priority: normal | Milestone: 7.10.1 Component: Core | Version: 7.8.3 Libraries | Keywords: Resolution: | Architecture: Unknown/Multiple Operating System: | Difficulty: Easy (less than 1 Unknown/Multiple | hour) Type of failure: Runtime | Blocked By: performance bug | Related Tickets: Test Case: | Blocking: | Differential Revisions: Phab:D390 | -------------------------------------+------------------------------------- Comment (by Joachim Breitner ): In [changeset:"5f69c8efd94862261bc6730f8dd80c2b67b430ad/ghc"]: {{{ #!CommitTicketReference repository="ghc" revision="5f69c8efd94862261bc6730f8dd80c2b67b430ad" Reorder GHC.List; fix performance regressions Rearrange some oddly placed code. Modify `take` to make the fold unconditionally strict in the passed `Int`. This clears up the `fft2` regression. This fixes #9740. Differential Revision: https://phabricator.haskell.org/D390 }}} -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

GHC

3:51 a.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: closed Priority: normal | Milestone: 7.10.1 Component: Core | Version: 7.8.3 Libraries | Keywords: Resolution: fixed | Architecture: Unknown/Multiple Operating System: | Difficulty: Easy (less than 1 Unknown/Multiple | hour) Type of failure: Runtime | Blocked By: performance bug | Related Tickets: Test Case: | Blocking: | Differential Revisions: Phab:D390 | -------------------------------------+------------------------------------- Changes (by nomeata): * status: patch => closed * resolution: => fixed Comment: Thanks! -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:4 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

GHC

4:13 a.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.10.1 Component: Core | Version: 7.8.3 Libraries | Keywords: Resolution: | Architecture: Unknown/Multiple Operating System: | Difficulty: Easy (less than 1 Unknown/Multiple | hour) Type of failure: Runtime | Blocked By: performance bug | Related Tickets: Test Case: | Blocking: | Differential Revisions: Phab:D390 | -------------------------------------+------------------------------------- Changes (by nomeata): * status: closed => new * resolution: fixed => Comment: It fixed the regression in T7257: http://ghcspeed- nomeata.rhcloud.com/timeline/?exe=2&base=2%2B68&ben=tests%2Falloc%2FT7257&env=1&revs=50&equid=on but not the one in fft2: http://ghcspeed- nomeata.rhcloud.com/timeline/?exe=2&base=2%2B68&ben=nofib%2Fallocs%2Ffft2&env=1&revs=50&equid=on Do your measurements disagree, or were you just expecting this to fix it. Could you have a second look? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:5 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

GHC

8:24 a.m.

...

It fixed the regression in T7257: http://ghcspeed- nomeata.rhcloud.com/timeline/?exe=2&base=2%2B68&ben=tests%2Falloc%2FT7257&env=1&revs=50&equid=on but not the one in fft2: http://ghcspeed- nomeata.rhcloud.com/timeline/?exe=2&base=2%2B68&ben=nofib%2Fallocs%2Ffft2&env=1&revs=50&equid=on

Do your measurements disagree, or were you just expecting this to fix it. Could you have a second look?

That's most peculiar. I had not measured T7257, but I ''definitely'' saw an improvement in fft2, unless I read something all wrong. I'm running nofib now and will review the numbers again. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:6 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

GHC

9:21 a.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.10.1 Component: Core | Version: 7.8.3 Libraries | Keywords: Resolution: | Architecture: Unknown/Multiple Operating System: | Difficulty: Easy (less than 1 Unknown/Multiple | hour) Type of failure: Runtime | Blocked By: performance bug | Related Tickets: Test Case: | Blocking: | Differential Revisions: Phab:D390 | -------------------------------------+------------------------------------- Comment (by dfeuer): I think I found the problem. I accidentally had an unrestricted `INLINE` on `take`. I'm trying again now. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:7 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

GHC

10:04 a.m.

GHC

10:48 a.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: patch Priority: normal | Milestone: 7.10.1 Component: Core | Version: 7.8.3 Libraries | Keywords: Resolution: | Architecture: Unknown/Multiple Operating System: | Difficulty: Easy (less than 1 Unknown/Multiple | hour) Type of failure: Runtime | Blocked By: performance bug | Related Tickets: Test Case: | Blocking: | Differential Revisions: Phab:D394 | -------------------------------------+------------------------------------- Comment (by Joachim Breitner ): In [changeset:"64d0a198be05c7baff36e43ab96928a402f00a19/ghc"]: {{{ #!CommitTicketReference repository="ghc" revision="64d0a198be05c7baff36e43ab96928a402f00a19" Really fix fft2 regression. #9740 Rewrite `take` more aggressively for fusion. Add some more explicit strictness to `unsafeTake` and `unsafeDrop` that seems to help code size and allocation just a drop in some nofib tests. They were not previously strict in their numerical arguments, but always called in contexts where those had been forced; it didn't make a difference in simple test cases, but made a small difference for nofib. See #9740. Differential Revision: https://phabricator.haskell.org/D394 }}} -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:9 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

GHC

11:33 a.m.

#9740: D380 caused fft2 regressions -------------------------------------+------------------------------------- Reporter: dfeuer | Owner: Type: bug | Status: closed Priority: normal | Milestone: 7.10.1 Component: Core | Version: 7.8.3 Libraries | Keywords: Resolution: fixed | Architecture: Unknown/Multiple Operating System: | Difficulty: Easy (less than 1 Unknown/Multiple | hour) Type of failure: Runtime | Blocked By: performance bug | Related Tickets: Test Case: | Blocking: | Differential Revisions: Phab:D394 | -------------------------------------+------------------------------------- Changes (by nomeata): * status: patch => closed * resolution: => fixed Comment: Fix confirmed. The regression had also affected mandel, where performance is recovered as well. Thanks for taking the time to hunt down the issue! -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/9740#comment:10 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

3859

Age (days ago)

3859

Last active (days ago)

List overview

Download

10 comments

1 participants

[GHC] #9740: D380 caused fft2 regressions

GHC

GHC

GHC

GHC

GHC

GHC

GHC

GHC

GHC

GHC

GHC

tags

participants (1)