RE: Ann: HAllInOne bug fix release
Something very odd is going on. GHC does not generate programs than run 3x faster between GHC versions. If, perhaps, you compiled both without -O, no cross-module inlining takes place. That would account for a big slow down when using separate compilation. I see you use -O2 this time. I don't know why the all in one version might go slower though S | -Original Message- | From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Hal Daume | Sent: 31 July 2003 00:23 | To: Wolfgang Jeltsch; The Haskell Mailing List | Subject: RE: Ann: HAllInOne bug fix release | | Ah, apparently it is not (at least not the Cygwin version). | | I recompiled NHC with GHC -O2, both the separate compilation version and | the all-in-over version. Averaged over five runs, we see that the | separate compilation version is actually *faster* than the ai1 version: | | standard nhc compiled with ghc: | | real0m27.167s | user0m9.991s | sys 0m1.304s | | nhc all-in-one: | | real0m31.411s | user0m10.007s | sys 0m1.299s | | | | i am completely unable to explain this. someone want to hazard a guess? | | - hal, who is a bit disappointed now :( | | -- | Hal Daume III | [EMAIL PROTECTED] | Arrest this man, he talks in maths. | www.isi.edu/~hdaume | | | -Original Message- | From: [EMAIL PROTECTED] | [mailto:[EMAIL PROTECTED] On Behalf Of Wolfgang Jeltsch | Sent: Wednesday, July 30, 2003 2:59 PM | To: The Haskell Mailing List | Subject: Re: Ann: HAllInOne bug fix release | | | On Wednesday, 2003-07-30, 23:36, CEST, Hal Daume III wrote: | [...] | | A few people have asked me for speed-up results from | All-In-One-ifying code, | so here's a good one. We take two versions of NHC. One is | the original | binary distribution and the other is the All-In-One-ified | version, compiled | by GHC. | | Hi, | | is the original binary distribution compiled with GHC? If | not, the speed-up | may also be the result of using a different compiler (i.e., GHC). | | [...] | | Wolfgang | | ___ | Haskell mailing list | [EMAIL PROTECTED] | http://www.haskell.org/mailman/listinfo/haskell | | ___ | Haskell mailing list | [EMAIL PROTECTED] | http://www.haskell.org/mailman/listinfo/haskell ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell
RE: Ann: HAllInOne bug fix release
At 08:44 31/07/03 +0100, Simon Peyton-Jones wrote: I don't know why the all in one version might go slower though Virtual memory thrashing? (Hal did say something about needing lots of RAM.) #g --- Graham Klyne [EMAIL PROTECTED] PGP: 0FAA 69FF C083 000B A2E9 A131 01B9 1C7A DBCA CB5E ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell
RE: Ann: HAllInOne bug fix release
Simon Peyton-Jones wrote: | I don't know why the all in one version might go | slower though. How about some (artificial) cut-offs during optimization phases? Optimizing a hugs module could lead to some kind of combinatorial explosion (which gets cut off by the optimizer) which does not happen when you have separate modules? Just my 2 öre! /Koen ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell
RE: Ann: HAllInOne bug fix release
(moved to haskell cafe -- hopefully the other threads will follow) I don't know why the all in one version might go slower though Virtual memory thrashing? (Hal did say something about needing lots of RAM.) only to compile. i haven't measured, but the resulting executable shouldn't require any more memory. it only hits about 400mbs max, though, which won't swap on my machine. - hal ___ Haskell-Cafe mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell-cafe
RE: Ann: HAllInOne bug fix release
(haskell-cafe now) Interesting. I've stripped them and they are exactly the same size. diff says they differ though. pre-stripping, the normal nhc (non ai1) is marginally larger (6.8m versus 7.0m). I'll try to measure memory usage if someone tells me how :) - Hal -- Hal Daume III | [EMAIL PROTECTED] Arrest this man, he talks in maths. | www.isi.edu/~hdaume -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of John Meacham Sent: Wednesday, July 30, 2003 5:22 PM To: The Haskell Mailing List Subject: Re: Ann: HAllInOne bug fix release Another useful thing would be the striped binary sizes of the two versions. I am curious how much AllInOneing affects the resultant binary. Memory usage would also be interesting. John On Wed, Jul 30, 2003 at 04:22:58PM -0700, Hal Daume wrote: Ah, apparently it is not (at least not the Cygwin version). I recompiled NHC with GHC -O2, both the separate compilation version and the all-in-over version. Averaged over five runs, we see that the separate compilation version is actually *faster* than the ai1 version: standard nhc compiled with ghc: real0m27.167s user0m9.991s sys 0m1.304s nhc all-in-one: real0m31.411s user0m10.007s sys 0m1.299s i am completely unable to explain this. someone want to hazard a guess? - hal, who is a bit disappointed now :( -- Hal Daume III | [EMAIL PROTECTED] Arrest this man, he talks in maths. | www.isi.edu/~hdaume -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Wolfgang Jeltsch Sent: Wednesday, July 30, 2003 2:59 PM To: The Haskell Mailing List Subject: Re: Ann: HAllInOne bug fix release On Wednesday, 2003-07-30, 23:36, CEST, Hal Daume III wrote: [...] A few people have asked me for speed-up results from All-In-One-ifying code, so here's a good one. We take two versions of NHC. One is the original binary distribution and the other is the All-In-One-ified version, compiled by GHC. Hi, is the original binary distribution compiled with GHC? If not, the speed-up may also be the result of using a different compiler (i.e., GHC). [...] Wolfgang ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell -- Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell -- -- - John Meacham - California Institute of Technology, Alum. - [EMAIL PROTECTED] -- - ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell ___ Haskell-Cafe mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell-cafe
RE: Ann: HAllInOne bug fix release
(cafe) This seems like a reasonable hypothesis. Is there a way to get GHC to keep trying, despite the enormity of the input file? Specifically, I expect it to do a better job of (a) inlining and (b) specialization. Is there a way to hint to it to try a bit harder? :) - Hal -- Hal Daume III | [EMAIL PROTECTED] Arrest this man, he talks in maths. | www.isi.edu/~hdaume -Original Message- From: Koen Claessen [mailto:[EMAIL PROTECTED] Sent: Thursday, July 31, 2003 1:49 AM To: Simon Peyton-Jones Cc: Hal Daume; Wolfgang Jeltsch; The Haskell Mailing List Subject: RE: Ann: HAllInOne bug fix release Simon Peyton-Jones wrote: | I don't know why the all in one version might go | slower though. How about some (artificial) cut-offs during optimization phases? Optimizing a hugs module could lead to some kind of combinatorial explosion (which gets cut off by the optimizer) which does not happen when you have separate modules? Just my 2 öre! /Koen ___ Haskell-Cafe mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell-cafe
Re: Ann: HAllInOne bug fix release
On Wednesday, 2003-07-30, 23:36, CEST, Hal Daume III wrote: [...] A few people have asked me for speed-up results from All-In-One-ifying code, so here's a good one. We take two versions of NHC. One is the original binary distribution and the other is the All-In-One-ified version, compiled by GHC. Hi, is the original binary distribution compiled with GHC? If not, the speed-up may also be the result of using a different compiler (i.e., GHC). [...] Wolfgang ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell
RE: Ann: HAllInOne bug fix release
Ah, apparently it is not (at least not the Cygwin version). I recompiled NHC with GHC -O2, both the separate compilation version and the all-in-over version. Averaged over five runs, we see that the separate compilation version is actually *faster* than the ai1 version: standard nhc compiled with ghc: real0m27.167s user0m9.991s sys 0m1.304s nhc all-in-one: real0m31.411s user0m10.007s sys 0m1.299s i am completely unable to explain this. someone want to hazard a guess? - hal, who is a bit disappointed now :( -- Hal Daume III | [EMAIL PROTECTED] Arrest this man, he talks in maths. | www.isi.edu/~hdaume -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Wolfgang Jeltsch Sent: Wednesday, July 30, 2003 2:59 PM To: The Haskell Mailing List Subject: Re: Ann: HAllInOne bug fix release On Wednesday, 2003-07-30, 23:36, CEST, Hal Daume III wrote: [...] A few people have asked me for speed-up results from All-In-One-ifying code, so here's a good one. We take two versions of NHC. One is the original binary distribution and the other is the All-In-One-ified version, compiled by GHC. Hi, is the original binary distribution compiled with GHC? If not, the speed-up may also be the result of using a different compiler (i.e., GHC). [...] Wolfgang ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell
Re: Ann: HAllInOne bug fix release
Another useful thing would be the striped binary sizes of the two versions. I am curious how much AllInOneing affects the resultant binary. Memory usage would also be interesting. John On Wed, Jul 30, 2003 at 04:22:58PM -0700, Hal Daume wrote: Ah, apparently it is not (at least not the Cygwin version). I recompiled NHC with GHC -O2, both the separate compilation version and the all-in-over version. Averaged over five runs, we see that the separate compilation version is actually *faster* than the ai1 version: standard nhc compiled with ghc: real0m27.167s user0m9.991s sys 0m1.304s nhc all-in-one: real0m31.411s user0m10.007s sys 0m1.299s i am completely unable to explain this. someone want to hazard a guess? - hal, who is a bit disappointed now :( -- Hal Daume III | [EMAIL PROTECTED] Arrest this man, he talks in maths. | www.isi.edu/~hdaume -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Wolfgang Jeltsch Sent: Wednesday, July 30, 2003 2:59 PM To: The Haskell Mailing List Subject: Re: Ann: HAllInOne bug fix release On Wednesday, 2003-07-30, 23:36, CEST, Hal Daume III wrote: [...] A few people have asked me for speed-up results from All-In-One-ifying code, so here's a good one. We take two versions of NHC. One is the original binary distribution and the other is the All-In-One-ified version, compiled by GHC. Hi, is the original binary distribution compiled with GHC? If not, the speed-up may also be the result of using a different compiler (i.e., GHC). [...] Wolfgang ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell -- Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell -- --- John Meacham - California Institute of Technology, Alum. - [EMAIL PROTECTED] --- ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell