[llvm-dev] MCJIT on cross-toolchain?

[Sergej Jaskiewicz via llvm-dev]

I apologize in advance if this is a somewhat silly question (I’m making my first steps in the infrastructure), but still.

Is it true that if an LLVM toolchain is compiled as a cross-toolchain, MCJIT is not expected to work?
I’m still working on a toolchain that cross-compiles from Windows to ARM Linux, and I’ve managed to get clang and lld tests to pass, but some (not all) MCJIT tests are failing.

If the answer is yes, what would be the appropriate way to skip those tests?

Sergej.
_______________________________________________
LLVM Developers mailing list
llvm...@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

[Sergej Jaskiewicz via llvm-dev]

Correction: I’m talking about unit tests. They fail with access violation following the message "'skylake' is not a recognized processor for this target (ignoring processor)”.

[David Blaikie via llvm-dev]

I think the answer is: These tests are buggy & should be fixed. They should be "not run" if the native target isn't compiled in but is needed for a test.

[David Blaikie via llvm-dev]

On Mon, Sep 2, 2019 at 11:36 AM Sergej Jaskiewicz <jaski...@icloud.com> wrote:

On 2 Sep 2019, at 21:16, David Blaikie <dbla...@gmail.com> wrote:

I think the answer is: These tests are buggy & should be fixed. They should be "not run" if the native target isn't compiled in but is needed for a test.

Thank you!

By native target you mean the host? If yes, I think I could try to fix that.

Right - I think 'lit' tests have a way to say "REQUIRES: native-target" or something and perhaps the unit tests haven't had the equivalent thing done to them. (it'd probably be programmatic in the unit tests - calling some LLVM API to see if the native target is available, then bailing out/skipping the test functionality entirely if it isn't).