[sage-devel] Re: pari slowness
On Jul 22, 11:48 am, David Harvey <[EMAIL PROTECTED]> wrote: > On Jul 22, 2008, at 12:35 PM, David Harvey wrote: > This seems to have been fixed already in 3.0.5. Sorry for the noise. > > david Hi David, we reverted to the old gmp 4.2.1 spkg in 3.0.5 since the only reason to upgrade was to better support Cygwin and OSX 64 bit. Since both of those platforms are not ready for prime time we decided it would be the more prudent way to go. Obviously we want eMPIRe in Sage ASAP and I need to do my share to make this happen, so please remind me often :) Cheers, Michael --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jul 22, 2008, at 12:35 PM, David Harvey wrote: >>> >>> Okay, I can confirm that with sage 3.0.1, sage -gp has the same >>> speed >>> as my standalone GP build. So mostly likely the change to GMP 4.2.2 >>> introduced a speed regression (probably the core 2 patches not being >>> applied properly). >> >> Ok, I will investigate and made this a blocker for 3.0.3: #3388 >> >> As is the gmp.spkg with the Core2 patches and all that fun stuff is a >> giant mess including the OSX fixes made by William :) >> >> Let's hope MPIR is here sooner than later > > I just noticed that the slowness happens on amd64 as well, so > probably Gaudry's patches are not being applied properly either. This seems to have been fixed already in 3.0.5. Sorry for the noise. david --~--~-~--~~~---~--~~ You received this message because you are subscribed to the Google Groups "sage-devel" group. To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel?hl=en -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 2008, at 10:36 PM, mabshoff wrote: >> Okay, I can confirm that with sage 3.0.1, sage -gp has the same speed >> as my standalone GP build. So mostly likely the change to GMP 4.2.2 >> introduced a speed regression (probably the core 2 patches not being >> applied properly). > > Ok, I will investigate and made this a blocker for 3.0.3: #3388 > > As is the gmp.spkg with the Core2 patches and all that fun stuff is a > giant mess including the OSX fixes made by William :) > > Let's hope MPIR is here sooner than later I just noticed that the slowness happens on amd64 as well, so probably Gaudry's patches are not being applied properly either. This is on alhambra (2.6GHz opteron): -- | SAGE Version 3.0.1, Release Date: 2008-05-05 | | Type notebook() for the GUI, and license() for information.| -- sage: x = ZZ.random_element(2^1000) sage: y = ZZ.random_element(2^1000) sage: time z = x * y CPU times: user 0.19 s, sys: 0.02 s, total: 0.21 s Wall time: 0.21 sage: time z = x * y CPU times: user 0.19 s, sys: 0.01 s, total: 0.20 s Wall time: 0.20 sage: time z = x * y CPU times: user 0.20 s, sys: 0.00 s, total: 0.20 s Wall time: 0.20 -- | SAGE Version 3.0.2, Release Date: 2008-05-24 | | Type notebook() for the GUI, and license() for information.| -- sage: x = ZZ.random_element(2^1000) sage: y = ZZ.random_element(2^1000) sage: time z = x * y CPU times: user 0.41 s, sys: 0.00 s, total: 0.41 s Wall time: 0.42 s sage: time z = x * y CPU times: user 0.41 s, sys: 0.00 s, total: 0.41 s Wall time: 0.41 s sage: time z = x * y CPU times: user 0.41 s, sys: 0.00 s, total: 0.41 s Wall time: 0.41 s david --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 7:01 pm, David Harvey <[EMAIL PROTECTED]> wrote: > On Jun 9, 2008, at 5:58 PM, William Stein wrote: > > The last version, so that we could build on cygwin, and also it was > > needed > > for OS X 10.5 64-bit. We will switch to mpir soon, as soon as > > there is a > > release :-) > > Okay, I can confirm that with sage 3.0.1, sage -gp has the same speed > as my standalone GP build. So mostly likely the change to GMP 4.2.2 > introduced a speed regression (probably the core 2 patches not being > applied properly). Ok, I will investigate and made this a blocker for 3.0.3: #3388 As is the gmp.spkg with the Core2 patches and all that fun stuff is a giant mess including the OSX fixes made by William :) Let's hope MPIR is here sooner than later > david Cheers, Michael --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 2008, at 5:58 PM, William Stein wrote: > On Mon, Jun 9, 2008 at 2:43 PM, David Harvey > <[EMAIL PROTECTED]> wrote: >> >> On Jun 9, 2008, at 5:35 PM, Michael Abshoff wrote: >>> >>> I wonder if we are just building GMP incorrectly. That bernfrac() >>> >>> routine should depend mainly on the speed of long integer >>> multiplication and division. I am not a GP expert --- how does one >>> generate large random integers in GP? I could try multiplying >>> them to >>> see how long that takes. >> >> Well, I believe that out spkg-install for GMP is potentially >> broken since we >> updated to gmp 4.2.2. There are also some potential issues with >> pre-3.0.3 >> gmp.spkgs, so I am adding some explicite messages that lets you >> know that >> Jason's GMP patch was actually applied. >> >> In which version of sage did we switch to gmp 4.2.2? I will try >> building the >> previous version on this machine and compare results. > > The last version, so that we could build on cygwin, and also it was > needed > for OS X 10.5 64-bit. We will switch to mpir soon, as soon as > there is a > release :-) Okay, I can confirm that with sage 3.0.1, sage -gp has the same speed as my standalone GP build. So mostly likely the change to GMP 4.2.2 introduced a speed regression (probably the core 2 patches not being applied properly). david --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Mon, Jun 9, 2008 at 3:31 PM, mabshoff <[EMAIL PROTECTED]> wrote: > > > > On Jun 9, 3:20 pm, David Harvey <[EMAIL PROTECTED]> wrote: >> On Jun 9, 2008, at 5:17 PM, Jonathan Bober wrote: >> >> >> >> >> >> > On Mon, 2008-06-09 at 10:19 -0700, mabshoff wrote: >> >> >> [...] >> >> No clue. Can you actually compare the gp binary from Sage directly >> >> with the timings from your self builid binary to eliminate the >> >> problem >> >> that libPari is involved here? If the gp binary in Sage is slower >> >> by a >> >> factor of three compared to the one you build this sounds like a bug >> >> to me. But it could also be conversation overhead for example. >> >> > Definitely could be conversion overhead. On my machine (warning: I'm >> > still running the "ridiculously old" Sage 2.10.2) I get >> >> > sage: time y = pari(4).bernfrac() >> > CPU times: user 4.14 s, sys: 0.00 s, total: 4.14 s >> > Wall time: 4.15 >> > sage: type(y) >> > >> > sage: time x = Rational(y) >> > CPU times: user 1.50 s, sys: 0.00 s, total: 1.50 s >> > Wall time: 1.50 >> >> > It looks like the conversion from sage.lib.pari.gen.gen to >> > sage.rings.rational.Rational just converts y to a string and then >> > parses >> > the resulting string, which is why this takes so long. >> >> This is now >> >> http://trac.sagemath.org/sage_trac/ticket/3387 Patch up. Please review. William --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 3:20 pm, David Harvey <[EMAIL PROTECTED]> wrote: > On Jun 9, 2008, at 5:17 PM, Jonathan Bober wrote: > > > > > > > On Mon, 2008-06-09 at 10:19 -0700, mabshoff wrote: > > >> [...] > >> No clue. Can you actually compare the gp binary from Sage directly > >> with the timings from your self builid binary to eliminate the > >> problem > >> that libPari is involved here? If the gp binary in Sage is slower > >> by a > >> factor of three compared to the one you build this sounds like a bug > >> to me. But it could also be conversation overhead for example. > > > Definitely could be conversion overhead. On my machine (warning: I'm > > still running the "ridiculously old" Sage 2.10.2) I get > > > sage: time y = pari(4).bernfrac() > > CPU times: user 4.14 s, sys: 0.00 s, total: 4.14 s > > Wall time: 4.15 > > sage: type(y) > > > > sage: time x = Rational(y) > > CPU times: user 1.50 s, sys: 0.00 s, total: 1.50 s > > Wall time: 1.50 > > > It looks like the conversion from sage.lib.pari.gen.gen to > > sage.rings.rational.Rational just converts y to a string and then > > parses > > the resulting string, which is why this takes so long. > > This is now > > http://trac.sagemath.org/sage_trac/ticket/3387 Good. > (this ticket does not include the discrepancy in GP timings) > > david Craig Citro did *a lot* of work on pari<->Sage conversation a while back and there are still some cases open. So once Craig sees this he might have some explanation what needs to get done since he knows the code well and might already have something that was never submitted since it needs cleanup. Cheers, Michael --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 2008, at 5:17 PM, Jonathan Bober wrote: > > On Mon, 2008-06-09 at 10:19 -0700, mabshoff wrote: >> >> [...] >> No clue. Can you actually compare the gp binary from Sage directly >> with the timings from your self builid binary to eliminate the >> problem >> that libPari is involved here? If the gp binary in Sage is slower >> by a >> factor of three compared to the one you build this sounds like a bug >> to me. But it could also be conversation overhead for example. > > Definitely could be conversion overhead. On my machine (warning: I'm > still running the "ridiculously old" Sage 2.10.2) I get > > sage: time y = pari(4).bernfrac() > CPU times: user 4.14 s, sys: 0.00 s, total: 4.14 s > Wall time: 4.15 > sage: type(y) > > sage: time x = Rational(y) > CPU times: user 1.50 s, sys: 0.00 s, total: 1.50 s > Wall time: 1.50 > > It looks like the conversion from sage.lib.pari.gen.gen to > sage.rings.rational.Rational just converts y to a string and then > parses > the resulting string, which is why this takes so long. This is now http://trac.sagemath.org/sage_trac/ticket/3387 (this ticket does not include the discrepancy in GP timings) david --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
Hi Bill, On Jun 9, 3:07 pm, Bill Hart <[EMAIL PROTECTED]> wrote: > Doesn't gmp 4.2.1 build on cygwin? Do you happen to recall what the > issues were, as it'll be a problem for mpir too (though I did manage > one build on cygwin already). > > Bill. At some point after a recent Cygwin update gmp 4.2.1 ceased to build with some sort of assembly error. Cygwin itself updated to gmp 4.2.2, so I do not know if it is a bug they just missed. I also noticed that even with vanilla gmp 4.2.2 on OSX "ABI=64" is not supported which really confused me since I got gmp to compile to compile in 64 bit mode on OSX before. But it seems like config.guess misidentifies some recent Core2 Quad CPUs, i.e. http://groups.google.com/group/sage-support/t/3371db245e33a79c from today and also some mention about that with that recent Japanese Core2 patch for gmp discussed on mpir-devel this week. Cheers, Michael --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
Doesn't gmp 4.2.1 build on cygwin? Do you happen to recall what the issues were, as it'll be a problem for mpir too (though I did manage one build on cygwin already). Bill. On 9 Jun, 22:58, "William Stein" <[EMAIL PROTECTED]> wrote: > On Mon, Jun 9, 2008 at 2:43 PM, David Harvey <[EMAIL PROTECTED]> wrote: > > > On Jun 9, 2008, at 5:35 PM, Michael Abshoff wrote: > > >> I wonder if we are just building GMP incorrectly. That bernfrac() > > >> routine should depend mainly on the speed of long integer > >> multiplication and division. I am not a GP expert --- how does one > >> generate large random integers in GP? I could try multiplying them to > >> see how long that takes. > > > Well, I believe that out spkg-install for GMP is potentially broken since we > > updated to gmp 4.2.2. There are also some potential issues with pre-3.0.3 > > gmp.spkgs, so I am adding some explicite messages that lets you know that > > Jason's GMP patch was actually applied. > > > In which version of sage did we switch to gmp 4.2.2? I will try building the > > previous version on this machine and compare results. > > The last version, so that we could build on cygwin, and also it was needed > for OS X 10.5 64-bit. We will switch to mpir soon, as soon as there is a > release :-) > > -- William --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Mon, Jun 9, 2008 at 2:43 PM, David Harvey <[EMAIL PROTECTED]> wrote: > > On Jun 9, 2008, at 5:35 PM, Michael Abshoff wrote: >> >> I wonder if we are just building GMP incorrectly. That bernfrac() >> >> routine should depend mainly on the speed of long integer >> multiplication and division. I am not a GP expert --- how does one >> generate large random integers in GP? I could try multiplying them to >> see how long that takes. > > Well, I believe that out spkg-install for GMP is potentially broken since we > updated to gmp 4.2.2. There are also some potential issues with pre-3.0.3 > gmp.spkgs, so I am adding some explicite messages that lets you know that > Jason's GMP patch was actually applied. > > In which version of sage did we switch to gmp 4.2.2? I will try building the > previous version on this machine and compare results. The last version, so that we could build on cygwin, and also it was needed for OS X 10.5 64-bit. We will switch to mpir soon, as soon as there is a release :-) -- William --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 2008, at 5:35 PM, Michael Abshoff wrote: > I wonder if we are just building GMP incorrectly. That bernfrac() > routine should depend mainly on the speed of long integer > multiplication and division. I am not a GP expert --- how does one > generate large random integers in GP? I could try multiplying them to > see how long that takes. > > Well, I believe that out spkg-install for GMP is potentially broken > since we updated to gmp 4.2.2. There are also some potential issues > with pre-3.0.3 gmp.spkgs, so I am adding some explicite messages > that lets you know that Jason's GMP patch was actually applied. In which version of sage did we switch to gmp 4.2.2? I will try building the previous version on this machine and compare results. david --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Mon, Jun 9, 2008 at 2:27 PM, David Harvey <[EMAIL PROTECTED]> wrote: > > > On Jun 9, 2008, at 5:17 PM, Jonathan Bober wrote: > > > > > On Mon, 2008-06-09 at 10:19 -0700, mabshoff wrote: > >> > >> [...] > >> No clue. Can you actually compare the gp binary from Sage directly > >> with the timings from your self builid binary to eliminate the > >> problem > >> that libPari is involved here? If the gp binary in Sage is slower > >> by a > >> factor of three compared to the one you build this sounds like a bug > >> to me. But it could also be conversation overhead for example. > > > > Definitely could be conversion overhead. On my machine (warning: I'm > > still running the "ridiculously old" Sage 2.10.2) I get > > > > sage: time y = pari(4).bernfrac() > > CPU times: user 4.14 s, sys: 0.00 s, total: 4.14 s > > Wall time: 4.15 > > sage: type(y) > > > > sage: time x = Rational(y) > > CPU times: user 1.50 s, sys: 0.00 s, total: 1.50 s > > Wall time: 1.50 > > > > It looks like the conversion from sage.lib.pari.gen.gen to > > sage.rings.rational.Rational just converts y to a string and then > > parses > > the resulting string, which is why this takes so long. > > That's true, that definitely accounts for part of it. Yep. But there's still quite a big gap between the times from sage -gp and > from my standalone GP build (see my followup email earlier on this > thread). > Yes. > > I wonder if we are just building GMP incorrectly. That bernfrac() > routine should depend mainly on the speed of long integer > multiplication and division. I am not a GP expert --- how does one > generate large random integers in GP? I could try multiplying them to > see how long that takes. > Well, I believe that out spkg-install for GMP is potentially broken since we updated to gmp 4.2.2. There are also some potential issues with pre-3.0.3 gmp.spkgs, so I am adding some explicite messages that lets you know that Jason's GMP patch was actually applied. > > david > Cheers, Michael > > > > > --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 2008, at 5:17 PM, Jonathan Bober wrote: > > On Mon, 2008-06-09 at 10:19 -0700, mabshoff wrote: >> >> [...] >> No clue. Can you actually compare the gp binary from Sage directly >> with the timings from your self builid binary to eliminate the >> problem >> that libPari is involved here? If the gp binary in Sage is slower >> by a >> factor of three compared to the one you build this sounds like a bug >> to me. But it could also be conversation overhead for example. > > Definitely could be conversion overhead. On my machine (warning: I'm > still running the "ridiculously old" Sage 2.10.2) I get > > sage: time y = pari(4).bernfrac() > CPU times: user 4.14 s, sys: 0.00 s, total: 4.14 s > Wall time: 4.15 > sage: type(y) > > sage: time x = Rational(y) > CPU times: user 1.50 s, sys: 0.00 s, total: 1.50 s > Wall time: 1.50 > > It looks like the conversion from sage.lib.pari.gen.gen to > sage.rings.rational.Rational just converts y to a string and then > parses > the resulting string, which is why this takes so long. That's true, that definitely accounts for part of it. But there's still quite a big gap between the times from sage -gp and from my standalone GP build (see my followup email earlier on this thread). I wonder if we are just building GMP incorrectly. That bernfrac() routine should depend mainly on the speed of long integer multiplication and division. I am not a GP expert --- how does one generate large random integers in GP? I could try multiplying them to see how long that takes. david --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Mon, 2008-06-09 at 10:19 -0700, mabshoff wrote: > > [...] > No clue. Can you actually compare the gp binary from Sage directly > with the timings from your self builid binary to eliminate the problem > that libPari is involved here? If the gp binary in Sage is slower by a > factor of three compared to the one you build this sounds like a bug > to me. But it could also be conversation overhead for example. Definitely could be conversion overhead. On my machine (warning: I'm still running the "ridiculously old" Sage 2.10.2) I get sage: time y = pari(4).bernfrac() CPU times: user 4.14 s, sys: 0.00 s, total: 4.14 s Wall time: 4.15 sage: type(y) sage: time x = Rational(y) CPU times: user 1.50 s, sys: 0.00 s, total: 1.50 s Wall time: 1.50 It looks like the conversion from sage.lib.pari.gen.gen to sage.rings.rational.Rational just converts y to a string and then parses the resulting string, which is why this takes so long. --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 2008, at 1:19 PM, mabshoff wrote: > > > > On Jun 9, 10:05 am, David Harvey <[EMAIL PROTECTED]> wrote: >> Hi, > > Hi David, > >> This is on an 8-core 2GHz xeon running debian. (Tom Boothby's >> machine.) >> >> In a clean build of sage-3.0.2: >> >> sage: time x = bernoulli(4) >> CPU times: user 4.19 s, sys: 0.01 s, total: 4.20 s >> Wall time: 4.20 s >> sage: time x = bernoulli(4) >> CPU times: user 3.18 s, sys: 0.01 s, total: 3.18 s >> Wall time: 3.19 s >> sage: time x = bernoulli(4) >> CPU times: user 3.18 s, sys: 0.00 s, total: 3.19 s >> Wall time: 3.19 s >> sage: time x = bernoulli(4) >> CPU times: user 3.18 s, sys: 0.00 s, total: 3.19 s >> Wall time: 3.19 s >> >> Then I tried building my own PARI/GP. I first built gmp 4.2.1 with >> jason martin's core 2 patches. Then I built pari/gp. I get: >> >> ? # >> timer = 1 (on) >> ? x = bernfrac(4); >> time = 1,972 ms. >> ? x = bernfrac(4); >> time = 1,317 ms. >> ? x = bernfrac(4); >> time = 1,316 ms. >> ? x = bernfrac(4); >> time = 1,316 ms. >> >> Why is the sage version three times slower than the gp version? > > No clue. Can you actually compare the gp binary from Sage directly > with the timings from your self builid binary to eliminate the problem > that libPari is involved here? If the gp binary in Sage is slower by a > factor of three compared to the one you build this sounds like a bug > to me. But it could also be conversation overhead for example. With sage -gp, I get something *in between*: ? x = bernfrac(4); time = 3,272 ms. ? x = bernfrac(4); time = 2,260 ms. ? x = bernfrac(4); time = 2,260 ms. The banner printout is exactly the same for both gp builds: GP/PARI CALCULATOR Version 2.3.3 (released) amd64 running linux (x86-64/GMP-4.2.1 kernel) 64- bit version compiled: Jun 9 2008, gcc-4.1.2 20061115 (prerelease) (Debian 4.1.1-21) (readline v5.2 enabled, extended help available) except that my standalone build says "readline not compiled in". david --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
FYI, when I computed bernoulli(10^7+4), I did so from sage -gp -- not from the sage interface to gp. On Mon, 9 Jun 2008, David Harvey wrote: > > Hi, > > This is on an 8-core 2GHz xeon running debian. (Tom Boothby's machine.) > > In a clean build of sage-3.0.2: > > sage: time x = bernoulli(4) > CPU times: user 4.19 s, sys: 0.01 s, total: 4.20 s > Wall time: 4.20 s > sage: time x = bernoulli(4) > CPU times: user 3.18 s, sys: 0.01 s, total: 3.18 s > Wall time: 3.19 s > sage: time x = bernoulli(4) > CPU times: user 3.18 s, sys: 0.00 s, total: 3.19 s > Wall time: 3.19 s > sage: time x = bernoulli(4) > CPU times: user 3.18 s, sys: 0.00 s, total: 3.19 s > Wall time: 3.19 s > > Then I tried building my own PARI/GP. I first built gmp 4.2.1 with > jason martin's core 2 patches. Then I built pari/gp. I get: > > ? # >timer = 1 (on) > ? x = bernfrac(4); > time = 1,972 ms. > ? x = bernfrac(4); > time = 1,317 ms. > ? x = bernfrac(4); > time = 1,316 ms. > ? x = bernfrac(4); > time = 1,316 ms. > > Why is the sage version three times slower than the gp version? > > david > > > > > --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---
[sage-devel] Re: pari slowness
On Jun 9, 10:05 am, David Harvey <[EMAIL PROTECTED]> wrote: > Hi, Hi David, > This is on an 8-core 2GHz xeon running debian. (Tom Boothby's machine.) > > In a clean build of sage-3.0.2: > > sage: time x = bernoulli(4) > CPU times: user 4.19 s, sys: 0.01 s, total: 4.20 s > Wall time: 4.20 s > sage: time x = bernoulli(4) > CPU times: user 3.18 s, sys: 0.01 s, total: 3.18 s > Wall time: 3.19 s > sage: time x = bernoulli(4) > CPU times: user 3.18 s, sys: 0.00 s, total: 3.19 s > Wall time: 3.19 s > sage: time x = bernoulli(4) > CPU times: user 3.18 s, sys: 0.00 s, total: 3.19 s > Wall time: 3.19 s > > Then I tried building my own PARI/GP. I first built gmp 4.2.1 with > jason martin's core 2 patches. Then I built pari/gp. I get: > > ? # > timer = 1 (on) > ? x = bernfrac(4); > time = 1,972 ms. > ? x = bernfrac(4); > time = 1,317 ms. > ? x = bernfrac(4); > time = 1,316 ms. > ? x = bernfrac(4); > time = 1,316 ms. > > Why is the sage version three times slower than the gp version? No clue. Can you actually compare the gp binary from Sage directly with the timings from your self builid binary to eliminate the problem that libPari is involved here? If the gp binary in Sage is slower by a factor of three compared to the one you build this sounds like a bug to me. But it could also be conversation overhead for example. > david Cheers, Michael --~--~-~--~~~---~--~~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~--~~~~--~~--~--~---