building polly with isl + imath instead of gmp

[Sebastian Pop]

Tobias Grosser wrote:
> On 04/01/2013 07:41 PM, Sebastian Pop wrote:
> >Hi,
> >
> >I have just posted on isl-development list a set of patches with which one can
> >build ISL with imath instead of GMP:
> >
> >https://groups.google.com/forum/#!msg/isl-development/GuEKe0Sunp8/c7gTL7MFOWgJ
> >
> >Attached to this email are the patches needed to build polly without GMP.
> 
> Hi Sebastian,
> 
> the patch that removes explicit references to GMP is OK to commit.
> 
> I am unsure about the other patches. The problem I see is that if we
> compile against an unmodified isl (that uses gmp) the isl_int macros
> are replaced by gmp function calls. This requires us to detect and
> link against libgmp. If we just remove the 'required' we may get
> unexpected link-time errors.
> 
> I believe we need to work with Sven to figure out how to distinguish
> the different isl back-ends. Based on these discussions we may
> either add a flag to configure that chooses a certain isl backend or
> we may be able to auto-detect the isl-backend. The library
> requirements will then
> need to be set accordingly.

I posted all the patches in the form we are using them: that will allow other
people to build polly with isl + imath.  I would like to not commit these
patches to Polly until we settled down with the changes to isl and imath: the
polly patches depend on the changes to isl.

Sebastian
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation