[PATCH] D25155: [Polly] Build and run isl_test as part of check-polly

Tobias Grosser via llvm-commits llvm-commits at lists.llvm.org
Tue Oct 4 06:45:07 PDT 2016


grosser marked 3 inline comments as done.
grosser added a comment.

In https://reviews.llvm.org/D25155#560724, @Meinersbur wrote:

> Thanks for the lit implementation!
>
> I tried and got:
>
>  ********************
>
> Testing Time: 1.17s
>
>  ********************
>
> Failing Tests (1):
>
>     Polly - isl unit tests :: isl_test.sh
>   
>   Unexpected Failures: 1
>
> FAILED: cd /root/build/llvm/release/tools/polly/test && /usr/bin/python2.7 /root/src/llvm/utils/lit/lit.py -sv --param polly_site_config=/root/build/llvm/release/tools/polly/test/UnitIsl/lit.site.cfg /root/build/llvm/release/tools/polly/test/UnitIsl
>  ninja: build stopped: subcommand failed.
>
>  


Interesting. It works for me. In which directory is the isl_codegen binary generated for you, if at all. To which value is the PATH variable set? (Does setting PATH even work on windows?)

> It also runs with `ninja polly-check-tests`, presumbly because lit searches for `lit.site.cfg`'s in all subdirectories. But, if you allow the question, if you don't want to run this by the default `ninja check-polly`, why implementing this in the first place?

Ideally, I wanted this to run as part of polly-check and polly-check-isl, but not polly-check-tests. The reason is that I thought you introduced polly-check-test to have an option to only run .ll tests, but no
unit tests. However, if polly-check-tests anyhow searches recursively, this is probably difficult to achieve.

Best,
Tobias



> Meinersbur wrote in CMakeLists.txt:77
> Why is this line duplicated?

I dropped it.

The original intention was to include the UnitIsl/ folder explicitly, but I forgot to rename /Unit to /UnitIsl. However, it seems to work even without this line.

> Meinersbur wrote in lit.cfg:109-111
> Given that PollyISL doesn't need anything from LLVM (here: `opt`), can we remove a lot of such lines?

Yes.

https://reviews.llvm.org/D25155





More information about the llvm-commits mailing list