Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Le mercredi 06 avril 2016 à 07:25 -0700, Johannes Wagner a écrit : > and last update, I disabled Hyperthreading on the i7, and now it > performes as expected. > > i7 no HT: > > 100 loops, best of 3: 879.57 ns per loop > 10 loops, best of 3: 9.88 µs per loop > 100 loops, best of 3: 4.46

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

and last update, I disabled Hyperthreading on the i7, and now it performes as expected. i7 no HT: 100 loops, best of 3: 879.57 ns per loop 10 loops, best of 3: 9.88 µs per loop 100 loops, best of 3: 4.46 ms per loop 1 loops, best of 3: 69.89 µs per loop 1 loops, best of 3: 26.67

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

On Tuesday, April 5, 2016 at 7:54:16 PM UTC+2, Milan Bouchet-Valat wrote: > > Le mardi 05 avril 2016 à 10:18 -0700, Johannes Wagner a écrit : > > hey Milan, > > so consider following code: > > > > Pkg.clone("git://github.com/kbarbary/TimeIt.jl.git") > > using TimeIt > > > > v = rand(3) >

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Le mardi 05 avril 2016 à 10:18 -0700, Johannes Wagner a écrit : > hey Milan, > so consider following code: > > Pkg.clone("git://github.com/kbarbary/TimeIt.jl.git") > using TimeIt > > v = rand(3) > r = rand(6000,3) > x = linspace(0.0, 10.0, 500) * (v./sqrt(sumabs2(v)))' > > dotprods = r * x[2,:]

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

hey Milan, so consider following code: Pkg.clone("git://github.com/kbarbary/TimeIt.jl.git") using TimeIt v = rand(3) r = rand(6000,3) x = linspace(0.0, 10.0, 500) * (v./sqrt(sumabs2(v)))' dotprods = r * x[2,:] imexp= cis(dotprods) sumprod = sum(imexp) * sum(conj(imexp)) f(r, x) = r *

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Le lundi 04 avril 2016 à 10:36 -0700, Johannes Wagner a écrit : > hey guys, > so attached you find text files with @code_native output for the > instructions  > - r * x[1,:] > - cis(imexp) > - sum(imexp) * sum(conj(imexp)) > > for julia 0.5.  > > Hardware I run on is a Haswell i5 machine, a

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

hey guys, so attached you find text files with @code_native output for the instructions - r * x[1,:] - cis(imexp) - sum(imexp) * sum(conj(imexp)) for julia 0.5. Hardware I run on is a Haswell i5 machine, a Haswell i7 machine, and a IvyBridge i5 machine. Turned out on an Haswell i5 machine

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Le mercredi 30 mars 2016 à 15:16 -0700, Johannes Wagner a écrit : > > > > Le mercredi 30 mars 2016 à 04:43 -0700, Johannes Wagner a écrit :  > > > Sorry for not having expressed myself clearly, I meant the latest  > > > version of fedora to work fine (24 development). I always used the  > > >

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

On Wednesday, March 30, 2016 at 1:58:23 PM UTC+2, Milan Bouchet-Valat wrote: > > Le mercredi 30 mars 2016 à 04:43 -0700, Johannes Wagner a écrit : > > Sorry for not having expressed myself clearly, I meant the latest > > version of fedora to work fine (24 development). I always used the > >

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Le mercredi 30 mars 2016 à 04:43 -0700, Johannes Wagner a écrit : > Sorry for not having expressed myself clearly, I meant the latest > version of fedora to work fine (24 development). I always used the > latest julia nightly available on the copr nalimilan repo. Right now > that is:

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Sorry for not having expressed myself clearly, I meant the latest version of fedora to work fine (24 development). I always used the latest julia nightly available on the copr nalimilan repo. Right now that is: 0.5.0-dev+3292, Commit 9d527c5*, all use LLVM: libLLVM-3.7.1 (ORCJIT, haswell)

[julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

just a little update. Tested some other fedoras: Fedora 22 with llvm 3.8 is also slow with julia 0.5, whereas a fedora 24 branch with llvm 3.7 is faster on julia 0.5 compared to julia 0.4, as it should be (speedup from inner loop parts translated into speedup to whole function). don't know if

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

Le mercredi 16 mars 2016 à 09:25 -0700, Johannes Wagner a écrit : > just a little update. Tested some other fedoras: Fedora 22 with llvm > 3.8 is also slow with julia 0.5, whereas a fedora 24 branch with llvm > 3.7 is faster on julia 0.5 compared to julia 0.4, as it should be > (speedup from inner

[julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

as an example, the data looks like this: using(TimeIt) v = rand(3) r = rand(6000,3) x = linspace(1.0, 2.0, 300) * (v./sqrt(sumabs2(v)))' *# Julia 0.4 function* function s04(xl, rl) result = zeros(size(xl,1)) for i = 1:size(xl,1) dotprods = rl * xl[i,:]'

[julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

as an example, the data looks like this: v = rand(3) r = rand(6000,3) x = linspace(1.0, 2.0, 300) * (v./sqrt(sumabs2(v)))' my function in 0.4 looks like this: function s04(xl, rl) result = zeros(size(xl,1)) for i = 1:size(xl,1) dotprods = rl * xl[i,:]'

[julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

spending time mostly on exp() calls on an array, other than that a dot product and a sum... openblas I only mentioned because it uninstalled openblas-threads while uninstalling julia 0.5 (by dnf dependency in fedora) and while installing again, openblas-threads did not get installed as

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

I did not compile them, just installed them from the nalimilan/julia and nalimilan/julia-nightlies repos with dnf on fedora23. On Friday, February 26, 2016 at 4:39:28 PM UTC+1, Yichao Yu wrote: > > On Fri, Feb 26, 2016 at 10:28 AM, Kristoffer Carlsson > wrote: > > What

Re: [julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

On Fri, Feb 26, 2016 at 10:28 AM, Kristoffer Carlsson wrote: > What code and where is it spending time? You talk about openblas, does it > mean that blas got slower for you? How about peakflops() on the different > versions? > > > On Friday, February 26, 2016 at 4:08:06 PM

[julia-users] Re: regression from 0.43 to 0.5dev, and back to 0.43 on fedora23

What code and where is it spending time? You talk about openblas, does it mean that blas got slower for you? How about peakflops() on the different versions? On Friday, February 26, 2016 at 4:08:06 PM UTC+1, Johannes Wagner wrote: > > hey guys, > I just experienced something weird. I have some