On Mon, 26 Mar 2012, Razya Ladelsky wrote:
> This is (hopefully) a permanent fix to pr46886.c
> I removed the condition preventing parallelization of do_while loops, as
> was blocking parallelizing important loops in spec-2006.
> The patch fixes the number of iterations for cases where the body could
> appear in the latch, as in pr46886.c.
> 2012-03-26 Razya Ladelsky <ra...@il.ibm.com>
> PR tree-optimization/46886
> * tree-parloops.c (transform_to_exit_first_loop):Set
> number of iterations correctly when the body may appear at the latch.
> (pallelize_loops): Remove the condition preventing
> do-while loops.
> Bootstrap and testsuite psss successfully on power7 linux machine.
> Ok to commit?
+ /* if the latch contains more than the one statemnt of control variable
+ increment then it contains the body. */
+ if (exit_1->dest == loop->latch && last_and_only_stmt (loop->latch))
new_rhs = gimple_cond_rhs (cond_stmt);
please check what the comment suggests, thus,
&& last_and_only_stmt (loop->latch) == cond_stmt
tree-parloops.c is quite fragile in what it expects the IL to look like
and tests do not cover what later code expects. Please do not add to
@@ -2146,7 +2149,6 @@ parallelize_loops (void)
spurious whitespace change.
@@ -2213,6 +2212,7 @@ parallelize_loops (void)
changed = true;
if (dump_file && (dump_flags & TDF_DETAILS))
Ok with the above change.