Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
  • Mar 04 11:26
    timholy commented #160
  • Mar 04 07:53
    tjaeuth commented #160
  • Feb 14 2020 13:06
    ViralBShah removed as member
  • Feb 14 2020 13:06
    ViralBShah removed as member
  • Feb 14 2020 13:06
    ViralBShah removed as member
  • Oct 09 2019 08:33
    KristofferC closed #161
  • Oct 09 2019 08:28
    KristofferC closed #2
  • Oct 09 2019 08:28
    KristofferC closed #2
  • Sep 04 2019 18:14
    KristofferC opened #161
  • Sep 04 2019 15:28
    KristofferC opened #2
  • Sep 04 2019 15:28
    KristofferC opened #2
  • Jun 01 2019 12:21
    timholy commented #160
  • Jun 01 2019 05:05
    javadba opened #160
  • Sep 12 2018 21:30
    NaelsonDouglas opened #159
  • Sep 12 2018 21:29
    NaelsonDouglas commented #158
  • Aug 12 2018 19:06
    GravityAssisted opened #158
  • Jul 09 2018 16:40
    RoyiAvital commented #142
  • Dec 18 2017 11:19
    alirezamecheng commented #157
  • Dec 12 2017 12:05
    proflage commented #157
  • Dec 12 2017 12:05
    proflage commented #157
Christopher Rackauckas
@ChrisRackauckas
did you run it twice?
No
f
@f_kazemian_twitter
No
Christopher Rackauckas
@ChrisRackauckas
So your timings don't mean anything.
f
@f_kazemian_twitter
of course if I run the both code
I can get this
@time f([1,2,3,4,5]) is 0.000004
and @time f2([1,2,3,4,5]) is 0.000109
why f2 is worse than f?
I mean It's their example
I think parallelaccelerator is not good
Christopher Rackauckas
@ChrisRackauckas
because you're probably are multithreading a computation which only has length 5?
f
@f_kazemian_twitter
because I couldn't get enough speed up
no
I think Julia is good for only sequential I know It's a bad result but
I tried every thing every where
I never get a good result in each package that told me it's parallel it's good in speed up it's bla bla
Christopher Rackauckas
@ChrisRackauckas
Yes, and I've been telling you why for weeks
f
@f_kazemian_twitter
I expect to get a good result here!
but
Christopher Rackauckas
@ChrisRackauckas
no, not in this 5 element case
f
@f_kazemian_twitter
you're right
in main program that I'm working with
Christopher Rackauckas
@ChrisRackauckas
that's just not how parallelism works. 5 elements added and multiplied is not a good idea to parallelize
f
@f_kazemian_twitter
I tried to work with 50 elements and I got only 2 seconds better than sequential
Christopher Rackauckas
@ChrisRackauckas
What chip are you using and are you sure that SIMD isn't just using all of your FPUs or something like that?
f
@f_kazemian_twitter
I run it in several pc
I just tried this example for 50 elements too
I got a bad result too
Christopher Rackauckas
@ChrisRackauckas
yes...
f
@f_kazemian_twitter
what number is big ? 50 is big
Christopher Rackauckas
@ChrisRackauckas
50 is very very small
f
@f_kazemian_twitter
but my main program has a lot of work and I think 50 is too big for that
I think yes for this function it's small..... but 50 times for a loop
Christopher Rackauckas
@ChrisRackauckas
if what you're multithreading is costly then 50 is very big
f
@f_kazemian_twitter
and parallelization a loop with that size is not small
right?
Christopher Rackauckas
@ChrisRackauckas
depends
f
@f_kazemian_twitter
oh I used multitreading too and I got a really bad result
Christopher Rackauckas
@ChrisRackauckas
I told you to check for inferrability issues though
f
@f_kazemian_twitter
I can push all of my work on git and shows to you If you want
Christopher Rackauckas
@ChrisRackauckas
and you never did
(this is off topic for ParallelAccelerator and should probably go to the main Julia channel)
f
@f_kazemian_twitter
yep sorry
ArGxento
@ArGxento
Hello, I have a question. How can I get the C++ code generated by the current version of ParallelAccelerator? I think old version of ParallelAccelerator can save generated C++ codes using PROSPECT_DEV_MODE. Or at least can I check and modify the compile options when the generated codes are compiled?
Todd A. Anderson
@DrTodd13
@ArGxento yes, you need PROSPECT_DEV_MODE=1 but you also need PROSPECT_MODE=cgen and you also probably want ParallelAccelerator.CGen.setCreateMain(true). The default is now native-threading so you need the PROSPECT_MODE=cgen to generate C code. The setCreateMain call will generate a C file with a main function that calls the accelerated function with the same arguments you passed to it. It also generates a .sh file you can run to compile everything.
ArGxento
@ArGxento
@DrTodd13 Thank you very much!
Jayakrishnan Nair
@jayvn
Hi
Just wanted to ask if ParallelAccelerator works on Julia 0.6 ? I looked up the github conversations but, nothing definite came of it
And the @acc compiled function is slow on my system and @code_warntype on it shows too many reds so I figured it wasn't ready yet for 0.6