[llvm] r177730 - R600: Use legacy (0 * anything = 0) MUL instructions for pow intrinsics

[Christian König]

Am 26.03.2013 18:58, schrieb Justin Holewinski:
> On Tue, Mar 26, 2013 at 1:50 PM, Christian König 
> <christian.koenig at amd.com <mailto:christian.koenig at amd.com>> wrote:
>
>     Am 26.03.2013 18 <tel:26.03.2013%2018>:18, schrieb Rafael Espíndola:
>
>             As Michel already wrote in the commit message, it fixes
>             two piglit tests and
>             so doesn't need to be covered by any llvm tests.
>
>             While it might make sense for some low level instructions
>             to be covered by
>             llvm tests, it probably doesn't make any sense for complex
>             behavior changes
>             like this one, since they are covered pretty well by piglit.
>
>         That is not the case at all! Please check the developer policy:
>
>         http://llvm.org/docs/DeveloperPolicy.html#test-cases
>
>
>     I know the policy, but it still doesn't make much sense for us
>     covering the same code by two different unit testing frameworks.
>
>     Christian.
>
>
> The issue here is more about non-R600 developers.  If I modify shared 
> codegen code somewhere, I want some way of verifying that I haven't 
> broken your change.  That's the purpose of running the unit tests 
> pre-commit.  These tests have to be in-tree, and I certainly don't 
> want to have to run mesa tests as a pre-commit check for LLVM!

Completely agree to that, and I think adding test cases for new 
patterns, instructions important backend behaviour etc.. does indeed 
make allot of sense.

But for example for the last change of our ELSE pseudo operation I don't 
even know how to cover that with llvm IR (except for adding compiled IR 
output of piglit and that's something around 25kb in size and covering 
allot of different things and not only that bug in specific).

Well I can easily make an IR dump of all falling/fixed piglit tests and 
that to the repository, but that probably wouldn't be welcome either.

Christian.

>
>
>
>             Christian.
>
>         Cheers,
>         Rafael
>
>
>
>
>     _______________________________________________
>     llvm-commits mailing list
>     llvm-commits at cs.uiuc.edu <mailto:llvm-commits at cs.uiuc.edu>
>     http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
>
> -- 
>
> Thanks,
>
> Justin Holewinski

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130326/e9483ea6/attachment.html>