------- Comment #12 from jakub at gcc dot gnu dot org 2008-04-09 14:32 -------
Created an attachment (id=15455)
--> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=15455&action=view)
gcc43-pr35634.patch
Here is the updated FE only patch. One change is that it avoids
P{RE,OST}{IN,DE}CREMENT_EXPR only for the promoting types, and has some
(admittedly very ugly) OpenMP parsing changes to counter that. Unfortunately
unlike #pragma omp for increment, #pragma omp atomic can have some_lvalue++
, not necessarily a variable_name++, so I have no idea how to handle that.
On x86_64 with this patch I get 3 omp failures (2 in libgomp atomic-10.c, one
in gcc/testsuite atomic-1.c) and:
FAIL: gcc.dg/vect/pr18536.c scan-tree-dump-times vect "vectorized 1 loops" 1
FAIL: gcc.dg/vect/pr30771.c scan-tree-dump-times vect "vectorized 1 loops" 1
FAIL: gcc.dg/vect/vect-iv-8a.c scan-tree-dump-times vect "vectorized 1 loops" 1
FAIL: gcc.dg/vect/vect-multitypes-11.c scan-tree-dump-times vect "vectorized 1
loops" 2
so (depending on where the other patch was tested) doing this in the FE doesn't
help much or at all.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35634
