[autovect] [patch] some cleanups and additional testcases for recentpatterns

[Dorit Naishlos <DORIT at il dot ibm dot com>]

Some fixes to the recently added pattern recognition functions. Also added
a couple testcases.
Bootstrapped with vectorization enabled on powerpc-linux, tested on the
vectorizer testcases. Committed to autovcet-branch.

dorit

Changelog:
+       * tree-vect-analyze.c (vect_pattern_recog_funcs): Move
+       vect_recog_sad_pattern to later in the initialization so it could
use
+       previously detected widen-sum.
+       (vect_recog_sad_pattern): Push last_stmt instead of stmt. Check for
+       optab of SAD_EXPR (accidentally was MULT_HI_EXPR).
+       (vect_recog_mult_hi_pattern): Don't push the widen-mult
pattern-stmt
+       into the stmt-list.
+       (vect_recog_dot_prod_pattern): Rely on previously detected
widen-mult
+       and widen-sum.
+       * tree-vect-transform.c (vectorizable_reduction): Allow other kinds
+       of def-types for the uses of the reduction.

+       * gcc.dg/vect/vect-mult-hi-and-dot.c: New.
+       * gcc.dg/vect/vect-sad-s8.c: New.

Patch:
(See attached file: patch2.oct23.txt)

Attachment: patch2.oct23.txt
Description: Text document