Re: gains from vectorization

Hi,

vectorization certainly won't help any kind of code, so it is important
to understand what part of the code is consuming most CPU time.
(compile the code with -pg, run it, and examine the output of gprof
executable_name gmon.out).

If this indicates some vectorizable loops are consuming most time,
-ftree-vectorize might do something (but in most of the cases less than
you might hope for).

All this assumes that you're using some other kind of optimisation
already (-O3 -march=XXX -ffast-math ...), and that your code is already
'well' written.

Notice that some commercial compiler can be quite a bit faster than g95
(also quite a bit more expensive ;-).

Joost

.

Relevant Pages

  • Re: Compiling with Modules
    ... I compile these module separate using the scheme: ... thor1.f90:: undefined reference to ... Regarding the "LOOP WAS VECTORIZED": ... A caveat with vectorization ...
    (comp.lang.fortran)
  • Reenable plug & play service
    ... I've had a problem on my laptop: "svchost.exe" was consuming in the ... last 4 days together with "system" nearly 100% of the cpu time, ... disabled all services not being able to shutdown and rebooted. ... play service, or any service again. ...
    (microsoft.public.win2000.general)
  • Reenable plug & play service
    ... I've had a problem on my laptop: "svchost.exe" was consuming in the ... last 4 days together with "system" nearly 100% of the cpu time, ... disabled all services not being able to shutdown and rebooted. ... play service, or any service again. ...
    (microsoft.public.win2000.setup)
  • Unable to open IE but can see IE process running in task manager
    ... These multiple processes of iexplore is consuming who of my CPU time. ... And this is affecting other applications like mcafee, ... I am using Mcafee as my security system. ...
    (microsoft.public.windows.inetexplorer.ie6.browser)
  • Re: real*4 array, wrong output compile with pgf77 -r8
    ... I use pgf77 to compile my fortran code. ... > The reason why I do this is trying to save cpu time by specify real*4. ... diligent you are in your variable declarations. ... using a -r4 or -r8 type of switch. ...
    (comp.lang.fortran)