[PATCH] overzealous instcombine of fptrunc(OP(fpext(x), fpext(y)))

[Nadav Rotem]

Hi Steve!

You are the expert in this domain so I trust you that the code is correct :)

Can you please add a CHECK: line for the function name and for ‘ret’ in your tests? It helps in debugging failing tests.  Also, InstCombine/fpextend.ll probably needs a CHECK line. 

The rest LGTM. 

Thanks,
Nadav


On Nov 27, 2013, at 10:48 AM, Stephen Canon <scanon at apple.com> wrote:

> InstCombiner::visitFPTrunc is currently overzealous and does some "optimizations" that it shouldn't (because they change results).  The attached patch reins it in so that we still get the benefit of this optimization for common cases, but we get rid of the misbehavior, and adds some test cases for situations where the optimization *shouldn't* happen.
> 
> Thanks to Pascal Quoc for bringing the issue to my attention (http://blog.frama-c.com/index.php?post/2013/09/25/The-problem-with-differential-testing-is-that-at-least-one-of-the-compilers-must-get-it-right).
> 
> This is also <rdar://problem/15080092>.
> 
> Thanks,
> – Steve
> 
> <fpextend_patch.txt>_______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits