[llvm-commits] [test-suite] r165698 - /test-suite/trunk/MultiSource/Benchmarks/Makefile

Hal Finkel hfinkel at anl.gov
Fri Oct 12 17:11:07 PDT 2012


Daniel,

Sure. The number of outer iterations is specified as a command-line parameter in the Makefile for each test. For example, if you look in:
MultiSource/Benchmarks/TSVC/Equivalencing-flt/Makefile
you'll see:
RUN_OPTIONS = 3125 5
where the 3135 is the number of outer loop iterations. I tuned the numbers so each test takes ~5 seconds on my build machine. Would dividing the number by 10 be enough to bring the tests in line with the rest of the test suite on ARM? How does the SMALL_PROBLEM_SIZE variable interact with the reference outputs?

Thanks again,
Hal

----- Original Message -----
> From: "Daniel Dunbar" <daniel at zuster.org>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: llvm-commits at cs.uiuc.edu
> Sent: Friday, October 12, 2012 7:00:48 PM
> Subject: Re: [llvm-commits] [test-suite] r165698 - /test-suite/trunk/MultiSource/Benchmarks/Makefile
> 
> Hi Hal,
> 
> Do you think that you could implement support for SMALL_PROBLEM_SIZE
> in these tests?
> 
> They significantly bloat the total testing time of the entire suite,
> especially at -O0 (or on ARM). The amount of time that they take to
> run is pretty disproportionate to the amount of time it takes the
> rest
> of the suite to run, particularly with SMALL_PROBLEM_SIZE=1 enabled.
> 
> For example, on x86_64 for -O0, these tests take about 1/3 of the
> total testing time. On ARM they take over half the time.
> 
> I'm not familiar with these tests but it looks like this may be as
> simple as changing LEN and LEN2 to be substantial smaller when that
> Makefile variable is set.
> 
>  - Daniel
> 
> On Wed, Oct 10, 2012 at 9:05 PM, Hal Finkel <hfinkel at anl.gov> wrote:
> > Author: hfinkel
> > Date: Wed Oct 10 23:05:05 2012
> > New Revision: 165698
> >
> > URL: http://llvm.org/viewvc/llvm-project?rev=165698&view=rev
> > Log:
> > Reenable TSVC on Darwin.
> >
> > All TSVC tests now pass on both my Linux and Mac machines,
> > and I hope that the buildbots will agreee.
> >
> >
> > Modified:
> >     test-suite/trunk/MultiSource/Benchmarks/Makefile
> >
> > Modified: test-suite/trunk/MultiSource/Benchmarks/Makefile
> > URL:
> > http://llvm.org/viewvc/llvm-project/test-suite/trunk/MultiSource/Benchmarks/Makefile?rev=165698&r1=165697&r2=165698&view=diff
> > ==============================================================================
> > --- test-suite/trunk/MultiSource/Benchmarks/Makefile (original)
> > +++ test-suite/trunk/MultiSource/Benchmarks/Makefile Wed Oct 10
> > 23:05:05 2012
> > @@ -9,12 +9,7 @@
> >                   McCat Olden Ptrdist llubenchmark \
> >                   sim FreeBench MallocBench Prolangs-C SciMark2-C
> >                   mediabench\
> >                   nbench ASCI_Purple MiBench Trimaran VersaBench
> >                   NPB-serial\
> > -                 BitBench ASC_Sequoia
> > -
> > -# Disable TSVC on Darwin until PR-14025 is resolved.
> > -ifneq ($(TARGET_OS),Darwin)
> > -PARALLEL_DIRS += TSVC
> > -endif
> > +                 BitBench ASC_Sequoia TSVC
> >
> >  ifndef DISABLE_CXX
> >  PARALLEL_DIRS += Prolangs-C++ PAQ8p tramp3d-v4 Bullet
> >
> >
> > _______________________________________________
> > llvm-commits mailing list
> > llvm-commits at cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 

-- 
Hal Finkel
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory



More information about the llvm-commits mailing list