[llvm-commits] [llvm] r153165 - /llvm/trunk/test/CodeGen/Generic/misched.ll

Andrew Trick atrick at apple.com
Wed Mar 21 10:04:22 PDT 2012


On Mar 21, 2012, at 6:26 AM, Hal Finkel <hfinkel at anl.gov> wrote:

> Andrew,
> 
> Out of curiosity, why would we want to have a test that just runs
> 'true'?


Short answer: I didn't feel the need to file a bug for an unsupported feature, but wanted others to see what was broken. I expected RUN: true to go away after my next checkin.

Long answer: I've been relying on stress testing without reducing them to nice unit tests. But in the long run misched=shuffle won't be tested outside of individual unit tests, so I need to start collecting those interesting cases that I find. I'd like to easily share the set of passing and failing tests during bringup, but hadn't decided how to do that. From now on, I'll probably put an XFAIL set of tests in /X86 with a target triple and move them to /Generic once they're stable.

-Andy



More information about the llvm-commits mailing list