The fix was subsumed by that for PR82129.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied.
Richard.
2017-10-23 Richard Biener <rguent...@suse.de>
PR tree-optimization/82129
Revert
2017-08-01 Richard Biener <rguent...@suse.de>
PR tree-optimization/81181
* tree-ssa-pre.c (compute_antic_aux): Defer clean() to ...
(compute_antic): ... end of iteration here.
Index: gcc/tree-ssa-pre.c
===================================================================
--- gcc/tree-ssa-pre.c (revision 254004)
+++ gcc/tree-ssa-pre.c (working copy)
@@ -2082,8 +2082,7 @@ static sbitmap has_abnormal_preds;
ANTIC_OUT[BLOCK] = phi_translate (ANTIC_IN[succ(BLOCK)])
ANTIC_IN[BLOCK] = clean(ANTIC_OUT[BLOCK] U EXP_GEN[BLOCK] - TMP_GEN[BLOCK])
-
- Note that clean() is deferred until after the iteration. */
+*/
static bool
compute_antic_aux (basic_block block, bool block_has_abnormal_pred_edge)
@@ -2219,8 +2218,7 @@ compute_antic_aux (basic_block block, bo
bitmap_value_insert_into_set (ANTIC_IN (block),
expression_for_id (bii));
- /* clean (ANTIC_IN (block)) is defered to after the iteration converged
- because it can cause non-convergence, see for example PR81181. */
+ clean (ANTIC_IN (block));
if (!bitmap_set_equal (old, ANTIC_IN (block)))
changed = true;
@@ -2453,12 +2451,6 @@ compute_antic (void)
gcc_checking_assert (num_iterations < 500);
}
- /* We have to clean after the dataflow problem converged as cleaning
- can cause non-convergence because it is based on expressions
- rather than values. */
- FOR_EACH_BB_FN (block, cfun)
- clean (ANTIC_IN (block));
-
statistics_histogram_event (cfun, "compute_antic iterations",
num_iterations);
