nomeata (Joachim Breitner)
User

Today

  • Clear sailing ahead.

Tomorrow

  • Clear sailing ahead.

Tuesday

  • Clear sailing ahead.

Badges

User Since
Jun 6 2014, 3:01 AM (146 w, 2 d)
Availability
Available

Recent Activity

Fri, Mar 24

nomeata resigned from rGHC1dd60ac9a949: Typos in comments (notes too) [ci skip].
Fri, Mar 24, 9:34 AM

Sun, Mar 19

nomeata resigned from rGHCcbe569a56e2a: Upgrade UniqSet to a newtype.

Sure.

Sun, Mar 19, 9:49 PM

Fri, Mar 17

nomeata raised a concern with rGHC1cbc7c315fb6: Bump nofib submodule.

Sorry for not checking earlier, but no, this is not sufficient: https://raw.githubusercontent.com/nomeata/ghc-speed-logs/f37051fd6649e468621ee197179a514587b78e27/e0c433c81182c934ee4c4cc5c6cf25a1b6fb8d83.log.broken

Fri, Mar 17, 5:14 PM

Thu, Mar 16

nomeata added a comment to D3055: WIP Add HeapView functionality.
In D3055#96358, @erikd wrote:

There's a lot of work to be done on this and I can't see myself finding the time.

Thu, Mar 16, 5:26 PM

Wed, Mar 15

nomeata accepted rNOFIB313812d319e0: spectral: fix secretary.
Wed, Mar 15, 3:38 PM
nomeata added a comment to rNOFIB313812d319e0: spectral: fix secretary.

I was actually thinking about this problem (but in the context of benchmarks in parallel) - maybe we could check-in (or add as submodules?) the packages that are needed for the benchmarks. This would make nofib more hermetic (and we could measure the compile-time required for those packages as another signal for measuring the performance of GHC itself).

Wed, Mar 15, 3:38 PM
nomeata accepted D3347: spectral: temporarily disable secretary.

LGTM.

Wed, Mar 15, 3:37 PM
nomeata raised a concern with rNOFIB313812d319e0: spectral: fix secretary.

This broke testing on perf.haskell.org:

+== make boot --no-print-directory;
+ in /home/nomeata/logs/ghc-tmp-REV/nofib/spectral/secretary
+------------------------------------------------------------------------
+/home/nomeata/logs/ghc-tmp-REV/inplace/bin/ghc-stage2  -M -dep-suffix "" -dep-makefile .depend -osuf o     -O2 -Rghc-timing -H32m -hisuf hi -package random -rtsopts Main.lhs
+<command line>: cannot satisfy -package random
+    (use -v for more information)
+<<ghc: 10186952 bytes, 7 GCs, 530340/819800 avg/max bytes residency (2 samples), 23M in use, 0.001 INIT (0.000 elapsed), 0.007 MUT (0.006 elapsed), 0.012 GC (0.012 elapsed) :ghc>>
Wed, Mar 15, 3:05 PM

Thu, Mar 9

nomeata committed rGHC712c45d03579: integerConstantFolding: no longer broken with -DDEBUG (#1106) (authored by nomeata).
integerConstantFolding: no longer broken with -DDEBUG (#1106)
Thu, Mar 9, 11:16 AM
nomeata committed rNOFIB9a1b1a0cd06b: Simon’s notes: Refer mailing list about binary-tree (authored by nomeata).
Simon’s notes: Refer mailing list about binary-tree
Thu, Mar 9, 10:20 AM
nomeata committed rGHC665cefe80d11: Add a comment to the mapFB rules (authored by nomeata).
Add a comment to the mapFB rules
Thu, Mar 9, 9:14 AM

Tue, Mar 7

nomeata updated the diff for D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).

Rebase

Tue, Mar 7, 4:55 AM
nomeata added a comment to D3275: Add rule mapFB c (λx.x) = c.
In D3275#95081, @dfeuer wrote:

We use one mechanism to deal with map/coerce and now a different one to deal with map/id. My impression is that these basically do the same thing in general. The map/id rule (as I understand it) simplifies mapFB forms so they'll eventually trip the foldr/cons rule. Is that right? Is it more effective than the simpler approach we take for map/coerce?

Tue, Mar 7, 4:51 AM

Mon, Mar 6

nomeata added a comment to D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).

Ok, all reported performance changes are due to the changes to the list functions, and not due to fusing short lists.

Mon, Mar 6, 6:52 AM

Sun, Mar 5

nomeata added a comment to D3275: Add rule mapFB c (λx.x) = c.
In D3275#94927, @dfeuer wrote:

Can you add a test to verify directly that the rule fires?

Sun, Mar 5, 5:42 PM
nomeata raised a concern with rGHCa694cee77b64: TcTypeable: Try to reuse KindReps.

There are two small, but stable runtime changes in nofib. Expected, or a sign that something went wrong?
https://perf.haskell.org/ghc/#revision/a694cee77b64235b42029fea248453ddf6b17d17

Sun, Mar 5, 5:32 AM

Sat, Mar 4

nomeata created D3275: Add rule mapFB c (λx.x) = c.
Sat, Mar 4, 2:20 PM
nomeata added a comment to D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).

Bumping out of the review queue for now.

Sat, Mar 4, 2:03 PM
nomeata committed rGHC2e58c3b21b40: Update dangling Note reference (authored by nomeata).
Update dangling Note reference
Sat, Mar 4, 1:51 PM

Fri, Mar 3

nomeata raised a concern with rGHC2effe18ab51d: The Early Inline Patch.

This might be interesting for @lukemaurer: According to https://perf.haskell.org/ghc/#revision/2effe18ab51d66474724d38b20e49cc1b8738f60 this patch undid the effect of join points on fannkuch-redux (i..e allocation goes up by 1318762.9%, but runtime goes down by 5%). Was this ever cleared up?

Fri, Mar 3, 5:31 AM

Thu, Mar 2

nomeata added a comment to D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).

Gipeda results are in: https://perf.haskell.org/ghc/#revision/b112f4604811faa6f5aeb399d8b5ee575f413bb7.

It definitely looks like there is more work to do here.

Thu, Mar 2, 4:23 PM

Wed, Mar 1

nomeata updated the diff for D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).

Test suite update

Wed, Mar 1, 8:32 PM
nomeata accepted rGHCd0508ef001e9: When floating, don't box an expression that's okay for speculation (#13338).

Nevermind me then :-)

Wed, Mar 1, 1:13 PM
nomeata raised a concern with rGHCd0508ef001e9: When floating, don't box an expression that's okay for speculation (#13338).

Do you have a grip on the performance regressions?

Benchmark name 	                    previous 	change 	       now 	
nofib/time/fannkuch-redux 	4.706 	+ 11.62% 	5.253 	seconds
nofib/time/k-nucleotide 	        5.323 	+ 4.3% 	        5.552 	seconds
Wed, Mar 1, 12:59 PM
nomeata updated the diff for D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).

Try to make T2110 more robust

Wed, Mar 1, 12:15 PM

Tue, Feb 28

nomeata created D3246: Revise list fusion for and, or, all, any, elem, notElem (#13351).
Tue, Feb 28, 3:42 PM

Feb 22 2017

nomeata resigned from rGHC8ccbc2e5252a: Bump Cabal and containers submodules.
Feb 22 2017, 9:51 AM
nomeata raised a concern with rGHC8ccbc2e5252a: Bump Cabal and containers submodules.

The last-piece nofib program has regressed considerably (7%), likely due to changes in Data.Map. How should we handle that? Ignore it? Report to the cabal project? I guess I’ll do the latter: https://github.com/haskell/containers/issues/415

Feb 22 2017, 9:50 AM

Feb 21 2017

nomeata added a comment to rGHC27a2854124cc: A number of Typeable wibbles from review.

Runtime performance improvements in fasta and lambda:
https://perf.haskell.org/ghc/#revision/27a2854124cc1c101570104501beea234a4ee921
A bit weird, isn’t it?

Feb 21 2017, 8:56 AM

Feb 16 2017

nomeata added a comment to D3141: testsuite: Move echoing commands in make invocations to VERBOSE=5.
In D3141#92010, @Phyx wrote:

I rather not have another environment variable to do another thing, as it's already hard to remember what to use for the various things.

Feb 16 2017, 10:17 AM

Feb 15 2017

nomeata requested changes to D3141: testsuite: Move echoing commands in make invocations to VERBOSE=5.

I have doubts about the VERBOSE setting. If increasing the verbosity can make tests fail, then that is a harsh violation of the principle of least surprise!

Feb 15 2017, 10:40 PM
nomeata raised a concern with rGHCa3704409acc3: Fix various issues with testsuite code on Windows.

Sorry to jump on this after it has been commited, but I have doubts about the VERBOSE setting. If increasing the verbosity can make tests fail, then that is a harsh violation of the principle of least surprise!

Feb 15 2017, 10:40 PM

Feb 11 2017

nomeata updated the diff for D3089: Improve the Occurrence Analyzer’s handling of one-shot functions.

Also recenter haddock.compiler

Feb 11 2017, 12:46 PM
nomeata updated the diff for D3089: Improve the Occurrence Analyzer’s handling of one-shot functions.

Update perf data.

Feb 11 2017, 12:44 PM
nomeata updated the diff for D3089: Improve the Occurrence Analyzer’s handling of one-shot functions.

Rebase to master

Feb 11 2017, 10:49 AM
nomeata added inline comments to D3055: WIP Add HeapView functionality.
Feb 11 2017, 9:23 AM

Feb 9 2017

nomeata added a comment to rGHCb9bebd8cedcc: Implement addCStub in template-haskell..

https://travis-ci.org/ghc/ghc/builds/200216540 looks good, thanks!

Feb 9 2017, 10:39 PM
nomeata added inline comments to D810: Ensure that Literals are in range.
Feb 9 2017, 5:39 PM
nomeata added a comment to rGHCb9bebd8cedcc: Implement addCStub in template-haskell..

Hmm, baking in the output of a tool that is not under our control, has several implementations and may of course change its error message from versin to version does not strike me as very useful in the long run. Maybe just grep the output for TH_addCStub2.hs:13:13, assuming that at least that part stays stable?

Feb 9 2017, 2:30 PM
nomeata raised a concern with rGHCb9bebd8cedcc: Implement addCStub in template-haskell..

On Travis, this causes test cases to fail: https://api.travis-ci.org/jobs/200040619/log.txt?deansi=true

Feb 9 2017, 12:19 PM
nomeata committed rGHCd266aac9586d: Library docs: Document the order for sort and sortOn. (authored by Alexander Kjeldaas <ak@formalprivacy.com>).
Library docs: Document the order for sort and sortOn.
Feb 9 2017, 12:17 PM

Feb 8 2017

nomeata committed rGHC8c4c084f858c: Travis experiment: Try “quickest” build flavor (authored by nomeata).
Travis experiment: Try “quickest” build flavor
Feb 8 2017, 1:38 PM

Feb 7 2017

nomeata committed rGHC2f31711f9291: Travis experiment: bump boot GHC (authored by nomeata).
Travis experiment: bump boot GHC
Feb 7 2017, 3:39 PM
nomeata committed rGHC8882aa770588: Improve the Occurrence Analyzer’s handling of one-shot functions (authored by nomeata).
Improve the Occurrence Analyzer’s handling of one-shot functions
Feb 7 2017, 3:35 PM
nomeata added a comment to rGHCb8f58d79ee3e: Another improvement to SetLevels.

I see different runtime results from nofib:
https://perf.haskell.org/ghc/#revision/b8f58d79ee3e34840beeea2fab846a9f47bff21a

Feb 7 2017, 2:06 PM
nomeata committed rGHCa0440d8d12b0: Travis experiment: bump boot GHC (authored by nomeata).
Travis experiment: bump boot GHC
Feb 7 2017, 1:54 PM
nomeata committed rGHCc729798f5a07: Travis experiment: try trusty (authored by nomeata).
Travis experiment: try trusty
Feb 7 2017, 1:54 PM
nomeata committed rGHCf9a0639fe00c: Travis experiment: language: generic (authored by nomeata).
Travis experiment: language: generic
Feb 7 2017, 1:51 PM
nomeata committed rGHC6fce907bf573: Travis experiment: Build in non-container environment (authored by nomeata).
Travis experiment: Build in non-container environment
Feb 7 2017, 1:51 PM
nomeata committed rGHC5ce39f6395ef: Add Wredundant-constraints to list of flags excluded from -Wall (authored by siddhanathan).
Add Wredundant-constraints to list of flags excluded from -Wall
Feb 7 2017, 10:20 AM
nomeata committed rGHC078beea9e970: Docs: Fix typo in glasgow_exts.rst (authored by DanielG).
Docs: Fix typo in glasgow_exts.rst
Feb 7 2017, 10:18 AM
nomeata updated the diff for D3089: Improve the Occurrence Analyzer’s handling of one-shot functions.

s/info/information/

Feb 7 2017, 9:59 AM
nomeata updated the diff for D3089: Improve the Occurrence Analyzer’s handling of one-shot functions.

Improve notes

Feb 7 2017, 9:36 AM

Feb 6 2017

nomeata committed rGHC52fc50301279: Improve the Occurrence Analyzer’s handling of one-shot functions (authored by nomeata).
Improve the Occurrence Analyzer’s handling of one-shot functions
Feb 6 2017, 8:39 PM
nomeata added a comment to D3073: Do not drop dead code in the desugarer.

@nomeata looks like there's a perf regression though in the testsuite (T4029).

Feb 6 2017, 7:41 PM

Feb 5 2017

nomeata raised a concern with rGHCfbcef83a3aa1: Use proper primitives in Utils.Binary.

https://perf.haskell.org/ghc/#revision/fbcef83a3aa130d976a201f2a21c5afc5a43d000 reports some speed-ups, but also that some stats in the test suite might need to be adjusted

Feb 5 2017, 10:13 PM
nomeata accepted rGHCa2f39da0461b: Add liftA2 to Applicative class.

https://perf.haskell.org/ghc/#revision/a2f39da0461b5da62a9020b0d98a1ce2765dd700 reports some speed-ups!

Feb 5 2017, 10:12 PM
nomeata committed rGHC4f92864d0404: Improve the Occurrence Analyzer’s handling of one-shot functions (authored by nomeata).
Improve the Occurrence Analyzer’s handling of one-shot functions
Feb 5 2017, 3:47 PM
nomeata retitled D3089: Improve the Occurrence Analyzer’s handling of one-shot functions from to Improve the Occurrence Analyzer’s handling of one-shot functions.
Feb 5 2017, 3:46 PM
nomeata committed rGHC88dcbc63ae34: Bind educated guess: count OneShotInfos in occ_one_shots (authored by nomeata).
Bind educated guess: count OneShotInfos in occ_one_shots
Feb 5 2017, 11:46 AM
nomeata committed rGHC6564330e7be1: Bind educated guess: count OneShotInfos in occ_one_shots (authored by nomeata).
Bind educated guess: count OneShotInfos in occ_one_shots
Feb 5 2017, 11:35 AM
nomeata committed rGHC4a78993377b2: Improve the Occurrence Analyzer’s handling of one-shot functions (authored by nomeata).
Improve the Occurrence Analyzer’s handling of one-shot functions
Feb 5 2017, 11:28 AM
nomeata committed rGHCa9dc47a1b995: Improve the Occurrence Analyzer’s handling of one-shot functions (authored by nomeata).
Improve the Occurrence Analyzer’s handling of one-shot functions
Feb 5 2017, 9:54 AM

Feb 4 2017

nomeata added a comment to D3073: Do not drop dead code in the desugarer.

From a very brief look at this, and assuming for a moment the plugin architecture allowed for hooks prior to the desugarer, would this change still be necessary

Feb 4 2017, 11:07 PM
nomeata added a comment to D3073: Do not drop dead code in the desugarer.

perf.haskell.org reports the absence of significant regressions: https://perf.haskell.org/ghc/#revision/b59c2de7abe3cd4e046f11c3536ba8e7137c4f84

Feb 4 2017, 11:06 PM
nomeata accepted rGHC283acec1d730: Make split sections by default work again.

Whoohoo! Build time down by 13%. And sizes down. Well spotted!

Feb 4 2017, 11:05 PM
nomeata committed rGHC88950a11b9df: Ensure that Literals in an Int# case are in range (authored by nomeata).
Ensure that Literals in an Int# case are in range
Feb 4 2017, 10:25 PM
nomeata updated the diff for D810: Ensure that Literals are in range.

Redone the patch, and make mkMachInt etc. do the wrapping.

Feb 4 2017, 10:24 PM
nomeata updated the diff for D810: Ensure that Literals are in range.

Rebase to master

Feb 4 2017, 9:30 PM
nomeata committed rGHC54b9b064fc79: Expose cseExpr from CSE (authored by nomeata).
Expose cseExpr from CSE
Feb 4 2017, 5:09 PM
nomeata committed rGHC82121357f5bf: New internal dynamic flag: Very aggressive inlining (authored by nomeata).
New internal dynamic flag: Very aggressive inlining
Feb 4 2017, 5:09 PM
nomeata closed D3069: Two new knobs for GHC API users by committing rGHC82121357f5bf: New internal dynamic flag: Very aggressive inlining.
Feb 4 2017, 5:09 PM
nomeata committed rGHCb59c2de7abe3: Do not drop dead code in the desugarer (authored by nomeata).
Do not drop dead code in the desugarer
Feb 4 2017, 8:14 AM
nomeata committed rGHC122c65592718: Expose cseExpr from CSE (authored by nomeata).
Expose cseExpr from CSE
Feb 4 2017, 8:14 AM
nomeata committed rGHC6d5efdb9ab64: New internal dynamic flag: Very aggressive inlining (authored by nomeata).
New internal dynamic flag: Very aggressive inlining
Feb 4 2017, 8:14 AM
nomeata added a comment to D3073: Do not drop dead code in the desugarer.

I disagree with that review.

Feb 4 2017, 8:13 AM

Feb 3 2017

nomeata added a comment to D3073: Do not drop dead code in the desugarer.

I have a better approach now. Do you like it?

Feb 3 2017, 9:13 PM
nomeata updated the diff for D3073: Do not drop dead code in the desugarer.

New approach to not dropping dead code in the desugarer

Feb 3 2017, 9:11 PM
nomeata retitled D3073: Do not drop dead code in the desugarer from to Do not drop dead code in the desugarer.
Feb 3 2017, 5:30 PM
nomeata added a comment to D3069: Two new knobs for GHC API users.

I'm curious: what is a hypothetical API user going to do with this flag? It seems likely that inlining everything possible would essentially never be a win in any realistic program due to code bloat. Is there something I'm missing here?

Feb 3 2017, 4:26 PM
nomeata added a comment to D3069: Two new knobs for GHC API users.

Obviously I’ll push these as separate commits; but I’d like to get a quick nod from someone else before I do so.

Feb 3 2017, 3:29 PM
nomeata retitled D3069: Two new knobs for GHC API users from to Two new knobs for GHC API users.
Feb 3 2017, 3:04 PM
nomeata committed rGHCc2becee48aa7: Bump performance mark for T9020 (authored by nomeata).
Bump performance mark for T9020
Feb 3 2017, 10:39 AM
nomeata committed rGHC09b8332df924: Get rid of ProbOneShot (authored by nomeata).
Get rid of ProbOneShot
Feb 3 2017, 10:39 AM
nomeata closed D3067: Get rid of ProbOneShot by committing rGHC09b8332df924: Get rid of ProbOneShot.
Feb 3 2017, 10:39 AM

Feb 2 2017

nomeata updated the diff for D3067: Get rid of ProbOneShot.

Clean up some code is simpler without ProbOneShot

Feb 2 2017, 7:38 PM
nomeata retitled D3067: Get rid of ProbOneShot from to Get rid of ProbOneShot.
Feb 2 2017, 7:31 PM
nomeata committed rGHC4c5af213a9ff: Get rid of ProbOneShot (authored by nomeata).
Get rid of ProbOneShot
Feb 2 2017, 2:32 PM
nomeata committed rGHC99857796bdac: Get rid of ProbOneShot (authored by nomeata).
Get rid of ProbOneShot
Feb 2 2017, 11:54 AM
nomeata committed rGHCd8ac64e782b8: Add a testcase for #13227 (authored by nomeata).
Add a testcase for #13227
Feb 2 2017, 11:06 AM

Feb 1 2017

nomeata added a comment to D3055: WIP Add HeapView functionality.

Does this have to live in a separate library? Can it not be part of base (which already has a bunch of low-level GHC. modules) or ghc-prim?

Feb 1 2017, 2:54 PM
nomeata added a comment to D3055: WIP Add HeapView functionality.
In D3055#89677, @erikd wrote:

@nomeata What do you thing of moving the haskell definition of those types to a different file (say GHC.Exts.HeapView.Closures`) and then generating that file automatically from Closure.h? Doing it this way means no test is needed.

Feb 1 2017, 2:33 PM
nomeata added inline comments to D3055: WIP Add HeapView functionality.
Feb 1 2017, 9:13 AM

Jan 24 2017

nomeata raised a concern with rGHC53e2e70a4778: Ensure that scrutinee constant folding wraps numbers.

This had a reproducible effect of increasing runtime of fasta by +4.47%. Are runtime effects here expected?

Jan 24 2017, 8:46 PM

Jan 19 2017

nomeata accepted rGHCbf1e1f3206f8: Add explicit foldMap implementation for Maybe.

Good job, tests/alloc/T13056 improves by 4%!

Jan 19 2017, 11:14 PM

Jan 17 2017

nomeata accepted rGHC715be013d405: Typos in manual and comments [ci skip].
Jan 17 2017, 11:09 AM

Jan 6 2017

nomeata added a comment to D2871: Add a CSE pass to Stg (#9291).

Well, I expect programs that use Monad (Either s) or Functor (Either s) or similar constructs to benefit, but we might not just have them in nofib.

Jan 6 2017, 11:58 AM
nomeata added a comment to D2871: Add a CSE pass to Stg (#9291).

Yes, at https://perf.haskell.org/ghc/#revision/19d5c7312bf0ad9ae764168132aecf3696d5410b. Unfortunately, nothing significant shows up. When I tested it locally, I saw some relevant improvements with the runtime of binary-trees. I continue to conclude that our benchmark situation sucks when it comes to producing hard evidence.

Jan 6 2017, 11:07 AM

Jan 5 2017

nomeata committed rGHC19d5c7312bf0: Add a CSE pass to Stg (#9291) (authored by nomeata).
Add a CSE pass to Stg (#9291)
Jan 5 2017, 8:18 AM
nomeata closed D2871: Add a CSE pass to Stg (#9291) by committing rGHC19d5c7312bf0: Add a CSE pass to Stg (#9291).
Jan 5 2017, 8:18 AM