[patch] Fix PR tree-optimization/35252

[Richard Guenther]

On Thu, Jul 24, 2008 at 4:50 PM, Victor Kaplansky <VICTORK@il.ibm.com> wrote:
>
> Hi,
>
> The function in PR35252 which performs summation of two complex float
> arrays
> can be vectorized using loop SLP.
>
> SLP analysis checks that all operations across vector have the same
> TREE_CODE.
> This patch fixes SLP to consider the following two loads:
>
>  CR.6_21 = REALPART_EXPR <ad[i_17]>;
>  CI.7_22 = IMAGPART_EXPR <ad[i_17]>;
>
> as having the same operation in this SLP analysis.
>
> Also added three additional test cases to test vectorization of complex
> data
> type.
>
> Bootstrapped and tested on x86_linx and PowerPC. O.K. for mainline?

Ok.

Thanks,
Richard.

> Thanks,
> Victor
>
> ChangeLog:
>
>      PR tree-optimization/35252
>      * tree-vect-analyze.c (vect_build_slp_tree): Make IMAGPART_EXPR and
>      REALPART_EXPR to be considered as same load operation.
>
> Testsuite:
>
>      PR tree-optimization/35252
>      * gcc.dg/vect/vect-complex-1.c, gcc.dg/vect/vect-complex-1.c,:
>      gcc.dg/vect/fast-math-vect-complex-3.c: New tests.
>
> (See attached file: patch.complex-slp2.diff.txt)