16 GCC HEAD regressions, 1 new, with your patch on 2004-07-22T01:29:41Z.

[Steven Bosscher]

On Thursday 22 July 2004 20:36, Geoffrey Keating wrote:
> > native gcc.sum gcc.c-torture/execute/simd-2.c
>
> This happens only at -O3 (both -O3 -fomit-frame-pointer and -O3 -g), so
> I'm suspecting:
>
> +2004-07-21  Josef Zlomek  <zlomekj@suse.cz>
> +
> +	* var-tracking.c (vt_find_locations): Set the in_pending bitmap at
> +	once.
> +

If you look at that patch (which FYI has been in my local tree for
several weeks now), you will see it is absolutely impossible for this
patch to cause that regression you're seeing.

diff -u -r2.20 -r2.21
--- gcc/gcc/var-tracking.c	2004/07/09 03:29:35	2.20
+++ gcc/gcc/var-tracking.c	2004/07/21 17:03:47	2.21
@@ -1700,13 +1700,10 @@
   in_worklist = sbitmap_alloc (last_basic_block);
   in_pending = sbitmap_alloc (last_basic_block);
   sbitmap_zero (in_worklist);
-  sbitmap_zero (in_pending);

   FOR_EACH_BB (bb)
-    {
-      fibheap_insert (pending, bb_order[bb->index], bb);
-      SET_BIT (in_pending, bb->index);
-    }
+    fibheap_insert (pending, bb_order[bb->index], bb);
+  sbitmap_ones (in_pending);

   while (!fibheap_empty (pending))
     {

Gr.
Steven