This fixes another fallout of the PR65962 fix, gcc.dg/vect/vect-62.c
failing for targets that use constant pool entries for the array
initializer.
On x86_64 we fail to vectorize the 2nd loop in the testcase because
PRE makes the latch block non-empty (and adds a loop carried dependency).
The patch fixes the code that is supposed to inhibit PRE from doing this.
Bootstrapped on x86_64-unknown-linux-gnu, testing in progress.
Richard.
2015-10-27 Richard Biener <rguent...@suse.de>
PR tree-optimization/65962
* tree-ssa-pre.c (eliminate_dom_walker::before_dom_children):
Avoid creating loop carried dependences also for outer loops
of the loop a use to replace is in.
* gcc.dg/vect/vect-62.c: Adjust.
Index: gcc/tree-ssa-pre.c
===================================================================
*** gcc/tree-ssa-pre.c (revision 229404)
--- gcc/tree-ssa-pre.c (working copy)
*************** eliminate_dom_walker::before_dom_childre
*** 4082,4089 ****
gimple *def_stmt = SSA_NAME_DEF_STMT (sprime);
basic_block def_bb = gimple_bb (def_stmt);
if (gimple_code (def_stmt) == GIMPLE_PHI
! && b->loop_father->header == def_bb)
{
ssa_op_iter iter;
tree op;
bool found = false;
--- 4090,4098 ----
gimple *def_stmt = SSA_NAME_DEF_STMT (sprime);
basic_block def_bb = gimple_bb (def_stmt);
if (gimple_code (def_stmt) == GIMPLE_PHI
! && def_bb->loop_father->header == def_bb)
{
+ loop_p loop = def_bb->loop_father;
ssa_op_iter iter;
tree op;
bool found = false;
*************** eliminate_dom_walker::before_dom_childre
*** 4092,4100 ****
affine_iv iv;
def_bb = gimple_bb (SSA_NAME_DEF_STMT (op));
if (def_bb
! && flow_bb_inside_loop_p (b->loop_father, def_bb)
! && simple_iv (b->loop_father,
! b->loop_father, op, &iv, true))
{
found = true;
break;
--- 4101,4108 ----
affine_iv iv;
def_bb = gimple_bb (SSA_NAME_DEF_STMT (op));
if (def_bb
! && flow_bb_inside_loop_p (loop, def_bb)
! && simple_iv (loop, loop, op, &iv, true))
{
found = true;
break;
*************** eliminate_dom_walker::before_dom_childre
*** 4110,4116 ****
print_generic_expr (dump_file, sprime, 0);
fprintf (dump_file, " which would add a loop"
" carried dependence to loop %d\n",
! b->loop_father->num);
}
/* Don't keep sprime available. */
sprime = NULL_TREE;
--- 4118,4124 ----
print_generic_expr (dump_file, sprime, 0);
fprintf (dump_file, " which would add a loop"
" carried dependence to loop %d\n",
! loop->num);
}
/* Don't keep sprime available. */
sprime = NULL_TREE;
Index: gcc/testsuite/gcc.dg/vect/vect-62.c
===================================================================
*** gcc/testsuite/gcc.dg/vect/vect-62.c (revision 229404)
--- gcc/testsuite/gcc.dg/vect/vect-62.c (working copy)
*************** int main1 ()
*** 33,41 ****
}
/* Multidimensional array. Aligned. The "inner" dimensions
! are invariant in the inner loop. Vectorizable, but the
! vectorizer detects that everything is invariant and that
! the loop is better left untouched. (it should be optimized away). */
for (i = 0; i < N; i++)
{
for (j = 0; j < N; j++)
--- 33,40 ----
}
/* Multidimensional array. Aligned. The "inner" dimensions
! are invariant in the inner loop. The outer loop is
! vectorizable after invariant/store motion. */
for (i = 0; i < N; i++)
{
for (j = 0; j < N; j++)
*************** int main (void)
*** 65,69 ****
return main1 ();
}
! /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 0
"vect" } } */
--- 64,68 ----
return main1 ();
}
! /* { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" } } */
/* { dg-final { scan-tree-dump-times "Vectorizing an unaligned access" 0
"vect" } } */
