diff options
| author | Richard Biener <rguenther@suse.de> | 2024-11-11 09:40:20 +0100 |
|---|---|---|
| committer | Richard Biener <rguenth@gcc.gnu.org> | 2024-11-11 13:03:16 +0100 |
| commit | da64698159fe69b68f5264b54cebcb67c501b3cf (patch) | |
| tree | 90c97c81ecc5eefb0adf760946a9b84f3dd22d96 | |
| parent | 95960cd473297cd0d2c9e75a1a424b870cee32f5 (diff) | |
| download | gcc-da64698159fe69b68f5264b54cebcb67c501b3cf.zip gcc-da64698159fe69b68f5264b54cebcb67c501b3cf.tar.gz gcc-da64698159fe69b68f5264b54cebcb67c501b3cf.tar.bz2 | |
tree-optimization/117510 - fix guard hoisting validity check
For the loop in the testcase we currently fail to hoist the guard
check of the inner loop (m > 0) out of the outer loop because
find_loop_guard checks all blocks of the outer loop for side-effects,
including those that are skipped by the guard. This usually
is harmless as the guard does not skip any blocks in the outer loop
but in this case store-motion was applied to the inner loop and thus
there's now a skipped store in the outer loop.
The following properly skips blocks that are dominated by the
entry to the skipped region.
PR tree-optimization/117510
* tree-ssa-loop-unswitch.cc (find_loop_guard): Only check
not skipped blocks for side-effects.
* gcc.dg/vect/vect-outer-pr117510.c: New testcase.
| -rw-r--r-- | gcc/testsuite/gcc.dg/vect/vect-outer-pr117510.c | 13 | ||||
| -rw-r--r-- | gcc/tree-ssa-loop-unswitch.cc | 6 |
2 files changed, 18 insertions, 1 deletions
diff --git a/gcc/testsuite/gcc.dg/vect/vect-outer-pr117510.c b/gcc/testsuite/gcc.dg/vect/vect-outer-pr117510.c new file mode 100644 index 0000000..e50b67c --- /dev/null +++ b/gcc/testsuite/gcc.dg/vect/vect-outer-pr117510.c @@ -0,0 +1,13 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target vect_double } */ +/* { dg-additional-options "-O3" } */ + +void f(int n, int m, double *a) +{ + a = __builtin_assume_aligned (a, __BIGGEST_ALIGNMENT__); + for (int i = 0; i < n; i++) + for (int j = 0; j < m; j++) + a[i] += 2*a[i] + j; +} + +/* { dg-final { scan-tree-dump "OUTER LOOP VECTORIZED" "vect" } } */ diff --git a/gcc/tree-ssa-loop-unswitch.cc b/gcc/tree-ssa-loop-unswitch.cc index 847f7ac..88516fd 100644 --- a/gcc/tree-ssa-loop-unswitch.cc +++ b/gcc/tree-ssa-loop-unswitch.cc @@ -1256,7 +1256,11 @@ find_loop_guard (class loop *loop, vec<gimple *> &dbg_to_reset) guard_edge = NULL; goto end; } - if (!empty_bb_without_guard_p (loop, bb, dbg_to_reset)) + /* If any of the not skipped blocks has side-effects or defs with + uses outside of the loop we cannot hoist the guard. */ + if (!dominated_by_p (CDI_DOMINATORS, + bb, guard_edge == te ? fe->dest : te->dest) + && !empty_bb_without_guard_p (loop, bb, dbg_to_reset)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, loc, |