r198422 - [OpenCL] Add test case for previous commit.

David Tweed david.tweed at gmail.com
Fri Jan 3 08:07:39 PST 2014


Hi all,

> Added: cfe/trunk/test/SemaOpenCL/invalid-constant.cl
> URL: http://llvm.org/viewvc/llvm-project/cfe/trunk/test/SemaOpenCL/invalid-constant.cl?rev=198422&view=auto
> ==============================================================================
> --- cfe/trunk/test/SemaOpenCL/invalid-constant.cl (added)
> +++ cfe/trunk/test/SemaOpenCL/invalid-constant.cl Fri Jan  3 09:11:57 2014
> @@ -0,0 +1,2 @@
> +// RUN: %clang_cc1 -verify %s
> +constant int no_init; // expected-error {{variable in constant address space must be initialized}}

So I absolutely think this is a fine commit in isolation and tests are
needed for everything, but I wonder
if -- given the presence of native build testing on
low power -- and hence less performant machines it might be something
to consider looking to see analyse if even a
little bit of aggregating similar-ish tests in (llvm and clang
projects) together into files would make noticeably reduce the
testing time. This is motivated by noticing that, eg, in

http://lab.llvm.org:8011/builders/clang-native-arm-cortex-a9/builds/14538

virtually 25% of the "build+test" time was taken on executing tests. I
have no idea what proportion of that is due
to start-up costs before the "actual test bit" happens; it wouldn't
surprise me if this was a large proportion of
the time. Obviously it's a non-issue for those testing on highly
performant machines,
and it would be very foolish to merge together tests that are testing
significantly different things even if they
would technically work in the same file, but maybe stuff that's
testing very close concepts could be merged.

Actually, now I think about it that can't be done as it'll break
historical databases of test results. But if anyone
has any other bright ideas on the subject...

-- 
cheers, dave tweed__________________________
high-performance computing and machine vision expert: david.tweed at gmail.com
"while having code so boring anyone can maintain it, use Python." --
attempted insult seen on slashdot



More information about the cfe-commits mailing list