https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119099
Jeffrey A. Law <law at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |vineetg at rivosinc dot com
--- Comment #4 from Jeffrey A. Law <law at gcc dot gnu.org> ---
So since in came up in the patchwork meeting today I figured a bit of a summary
of my morning's findings.
The reason I can't find the assert is I think *I* took it out at some point.
Regardless we can't have the sets expanding and contracting as it'd be a
nightmare to prove convergence.
What I was looking at was something like this Alexey:
diff --git a/gcc/ext-dce.cc b/gcc/ext-dce.cc
index e257e3bc873..5866cc0a3b8 100644
--- a/gcc/ext-dce.cc
+++ b/gcc/ext-dce.cc
@@ -1092,13 +1092,7 @@ ext_dce_rd_transfer_n (int bb_index)
/* We may have narrowed the set of live objects at the start
of this block. If so, update the bitmaps and indicate to
the generic dataflow code that something changed. */
- if (!bitmap_equal_p (&livein[bb_index], livenow))
- {
- bitmap_copy (&livein[bb_index], livenow);
- return true;
- }
-
- return false;
+ return bitmap_ior_into (&livein[bb_index], livenow);
}
/* Dummy function for the df_simple_dataflow API. */
That should result in the livein sets never contracting. That API will return
a boolean indicating if that livein set actually changed. Now I just have to
see if there's any fallout.
--
There's a *ton* of cleanup that can/should be done in here. I've got a rewrite
of the process_sets code that cleans it up considerably. But what worries me
much more is the process_uses code which is beyond bad in its
design/implementation.
