Loop removal was processing blocks in the wrong order, reversing
debug stmt execution order between blocks (but having correct
order within a block).
Bootstrap & regtest running on x86_64-unknown-linux-gnu.
The testcase exposes PR90717 and thus currently FAILs with
-flto -fuse-linker-plugin. We seem to lack a dg-xfail-if
and dg-xfail-run-if isn't applicable here. Anyhow, hope
that Alex can fix this issue.
Richard.
2019-06-03 Richard Biener <rguent...@suse.de>
PR tree-optimization/90716
* tree-loop-distribution.c (destroy_loop): Process blocks in
correct order.
* gcc.dg/guality/pr90716.c: New testcase.
Index: gcc/tree-loop-distribution.c
===================================================================
--- gcc/tree-loop-distribution.c (revision 271803)
+++ gcc/tree-loop-distribution.c (working copy)
@@ -1104,15 +1104,13 @@ destroy_loop (struct loop *loop)
gimple_stmt_iterator dst_gsi = gsi_after_labels (exit->dest);
bool safe_p = single_pred_p (exit->dest);
- i = nbbs;
- do
+ for (unsigned i = 0; i < nbbs; ++i)
{
/* We have made sure to not leave any dangling uses of SSA
names defined in the loop. With the exception of virtuals.
Make sure we replace all uses of virtual defs that will remain
outside of the loop with the bare symbol as delete_basic_block
will release them. */
- --i;
for (gphi_iterator gsi = gsi_start_phis (bbs[i]); !gsi_end_p (gsi);
gsi_next (&gsi))
{
@@ -1147,7 +1145,6 @@ destroy_loop (struct loop *loop)
gsi_next (&gsi);
}
}
- while (i != 0);
redirect_edge_pred (exit, src);
exit->flags &= ~(EDGE_TRUE_VALUE|EDGE_FALSE_VALUE);
Index: gcc/testsuite/gcc.dg/guality/pr90716.c
===================================================================
--- gcc/testsuite/gcc.dg/guality/pr90716.c (nonexistent)
+++ gcc/testsuite/gcc.dg/guality/pr90716.c (working copy)
@@ -0,0 +1,25 @@
+/* { dg-do run } */
+/* { dg-options "-g" } */
+
+void __attribute__((noinline))
+optimize_me_not ()
+{
+ __asm__ volatile ("" : : : "memory");
+}
+int a[7][8];
+int main()
+{
+ int b, j;
+ b = 0;
+ for (; b < 7; b++) {
+ j = 0;
+ for (; j < 8; j++)
+ a[b][j] = 0;
+ }
+ /* j may very well be optimized out, so we cannot test for j == 8.
+ Instead test j + 1 which will make the test UNSUPPORTED if i
+ is optimized out. Since the test previously had wrong debug
+ with j == 0 this is acceptable. */
+ optimize_me_not(); /* { dg-final { gdb-test . "j + 1" "9" } } */
+ return 0;
+}
