https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117116
Hongtao Liu <liuhongt at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|liuhongt at gcc dot gnu.org |uros at gcc dot gnu.org
--- Comment #13 from Hongtao Liu <liuhongt at gcc dot gnu.org> ---
(In reply to Uroš Bizjak from comment #12)
>It's not a problem with the above pattern, which *does* support V2DI:
Oh, yes, I must have been blind not to see that v2di.
> Created attachment 59348 [details]
> Proposed patch
>
> Patch in testing.
>
> Just leave CONST_VECTOR RTXes that won't simplify to integer in memory. The
> follow-up code that handles (offsettable_memref_p (operand)) handles
> CONST_VECTOR RTXes in memory just fine.
>
> I tried to split V2SImode CONST_VECTOR to two SImode CONST elements, but
> there were some further ICEs generated, and it is getting a bit late here...?
The codegen looks much better, please go ahead with the patch.
