[llvm-commits] [PATCH] Avoid overflow in scheduling

Reid Kleckner rnk at mit.edu
Sat Sep 26 15:25:38 CDT 2009


There is an assertion that checks that they are always > 0.  The
assertion could cast to signed before comparing, but you might as well
just declare them signed.

Reid

On Sat, Sep 26, 2009 at 4:22 PM, Nick Lewycky <nicholas at mxc.ca> wrote:
> Reid Kleckner wrote:
>>
>> Hey all,
>>
>> One of our nasty regression test cases generates a ginormous (4 MB of
>> bitcode) function that overflows a short integer in some instruction
>> scheduling code.  Is the attached fix OK?  If so, I'll commit it.
>>
>> This fixes PR4401.
>
> Any reason these aren't unsigned?
>
> Nick
>




More information about the llvm-commits mailing list