Just like PR 31995, this was found on the pointer plus branch after fixing forwprop (note fowrwprop could be improved even more but that is a different story and then we just get back to PR 31995 really), This testcase is really vect-103 but modified to confuse data-ref: /* { dg-require-effective-target vect_int } */ #include <stdlib.h> #include <stdarg.h> #include "tree-vect.h" #define N 9 struct extraction { int a[N]; int b[N]; }; static int a[N] = {1,2,3,4,5,6,7,8,9}; static int b[N] = {17,24,7,0,2,3,4,31,82}; static int c[N] = {9,17,24,7,0,2,3,4,31}; int main1 (int x, int y) { int i; struct extraction *p; p = (struct extraction *) malloc (sizeof (struct extraction)); for (i = 0; i < N; i++) { p->a[i] = a[i]; p->b[i] = b[i]; if (x == 135) abort (); /* to avoid vectorization */ } /* Vectorizable: distance > VF. */ for (i = 0; i < N; i++) { p->a[x + i] = *((int *)p + x + i + 8); } /* check results: */ for (i = 0; i < N; i++) { if (p->a[i] != c[i]) abort(); } return 0; } int main (void) { check_vect (); return main1 (0, N); } /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */ /* { dg-final { scan-tree-dump-times "dependence distance modulo vf == 0" 1 "vect" } } */ /* { dg-final { cleanup-tree-dump "vect" } } */
Confirmed, this is now on the trunk. http://gcc.gnu.org/ml/gcc-testresults/2007-07/msg00395.html FAIL: gcc.dg/vect/vect-103.c scan-tree-dump-times vectorized 1 loops 1 FAIL: gcc.dg/vect/vect-103.c scan-tree-dump-times dependence distance modulo vf == 0 1
The testcase was fixed after: 2007-07-09 Richard Guenther <rguenther@suse.de> PR middle-end/32698 * fold-const.c (fold_plusminus_mult_expr): Move constant arguments second to allow decomposing.