[llvm-commits] Review request for test-suite test "unix-smail" disabling

Chris Lattner clattner at apple.com
Fri Mar 30 11:49:04 PDT 2012


On Mar 30, 2012, at 11:19 AM, Joel Jones wrote:

> The test in test-suite in the directory MultiSource/Benchmarks/Prolangs-C,
> when measured, appears to spend almost no time in user-space, but in the
> system executing getpwent().  Over time, as more entries in the password file
> (or entries loaded from a network server) are added, the time spent in getpwent
> grows.  This leads to the appearance of a performance regression.  This patch
> disables unix-smail from being run.  However, the test will still be visible in the
> output web pages in sections labeled: "Removed Tests - Compile Time" and 
> "Removed Tests - Execution Time".

Hi Joel,

Why not just hack the benchmark to not call getpwent?

-Chris

> 
> Joel Jones
> 
> Index: MultiSource/Benchmarks/Prolangs-C/Makefile
> ===================================================================
> --- MultiSource/Benchmarks/Prolangs-C/Makefile	(revision 153756)
> +++ MultiSource/Benchmarks/Prolangs-C/Makefile	(working copy)
> @@ -6,6 +6,10 @@
> # Get the $(ARCH) setting
> include $(LEVEL)/Makefile.config
> 
> +# disable unix-smail completely, as it really only tests getpwent, which isn't
> +# much of a compiler test
> +PARALLEL_DIRS := $(filter-out unix-smail, $(PARALLEL_DIRS))
> +
> # Disable bison due to insane running times of bison.native
> ifeq ($(ARCH), Sparc)
> PARALLEL_DIRS := $(filter-out bison, $(PARALLEL_DIRS))
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits




More information about the llvm-commits mailing list