This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[tree-ssa] PATCH to mostly_copy_tree_r
- From: Jason Merrill <jason at redhat dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Tue, 20 May 2003 16:07:39 -0400
- Subject: [tree-ssa] PATCH to mostly_copy_tree_r
I have been noticing a lot of
if (foo)
{
...
}
else
{
(void)0;
}
in tree dumps recently; the pretty-printer is supposed to recognize this
case and avoid printing an 'else' clause. The cause turned out to be
unnecessary unsharing of the size_zero_node in empty statements, which
causes IS_EMPTY_STMT to return false.
The fix is to just not unshare constants; I don't know why I used to think
they should be unshared.
Hmm...now it occurs to me that we'll probably still end up with copies when
inlining, so IS_EMPTY_STMT should use integer_zerop. And in most places we
should be using !TREE_SIDE_EFFECTS instead, anyway.
Booted and tested i686-pc-linux-gnu, applied to tree-ssa.
2003-05-20 Jason Merrill <jason@redhat.com>
* gimplify.c (mostly_copy_tree_r): Don't unshare constants.
*** gimplify.c.~1~ Tue May 13 15:44:24 2003
--- gimplify.c Tue May 20 15:45:58 2003
*************** mostly_copy_tree_r (tp, walk_subtrees, d
*** 2406,2413 ****
void *data;
{
enum tree_code code = TREE_CODE (*tp);
! /* Don't unshare types and SAVE_EXPR nodes. */
if (TREE_CODE_CLASS (code) == 't'
|| code == SAVE_EXPR)
*walk_subtrees = 0;
else if (code == BIND_EXPR)
--- 2492,2500 ----
void *data;
{
enum tree_code code = TREE_CODE (*tp);
! /* Don't unshare types, constants and SAVE_EXPR nodes. */
if (TREE_CODE_CLASS (code) == 't'
+ || TREE_CODE_CLASS (code) == 'c'
|| code == SAVE_EXPR)
*walk_subtrees = 0;
else if (code == BIND_EXPR)