[LLVMdev] Compiler-RT buildbot on ARM

Evgeniy Stepanov eugeni.stepanov at gmail.com
Wed Sep 10 04:12:56 PDT 2014


On Wed, Sep 10, 2014 at 2:37 PM, Renato Golin <renato.golin at linaro.org> wrote:
> On 10 September 2014 11:21, Evgeniy Stepanov <eugeni.stepanov at gmail.com> wrote:
>> As for the aarch64 failures, do you want to build the aarch64 bit
>> stuff, but not run the tests?
>
> Sort of. I'd like to test everything I can, but execution AArch64
> tests will obviously fail on ARM.
>
> Are there any tests that don't need execution? Like the lit tests we
> have for LLVM?

All (or almost all) sanitizer lit tests need execution, unlike llvm
lit tests. They are runtime library tests, after all, not compiler
tests.

>
> If not, I'd still like to build AArch64 code, since that stresses LLVM
> on AArch64.
>
>> In cmake we try to detect target
>> platforms supported by the compiler, but we don't check that binaries
>> actuall run.
>
> I was wondering if this could be fixed in the test selection process
> (CMake?) to not only choose tests of the built back-ends, but the ones
> that could actually run. I imagine if you just run "cmake $SRC", LLVM
> build all back-ends by default, wouldn't that try to *run* all tests
> from all platforms?

We check that compilation of a simple source succeeds (see
cmake/config-ix.cmake), that requires that target headers and
libraries are available. This normally limits us to multilib variants
of the build platform.

We probably should check that test binaries can be executed, too - but
there are exceptions like android where we run tests remotely (on an
attached device or emulator).

>
>
>> But why is there so little failures, if the hardware does
>> not support 64-bit (or does it)?
>
> It doesn't. Not sure why so little failures...
>
> cheers,
> --renato



More information about the llvm-dev mailing list