procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

re PR c++/17972 (const/pure functions result in bad asm)

Browse source 
PR c++/17972
	* tree-inline.c (expand_call_inline): Set TREE_SIDE_EFFECTS
	on the STMT_EXPR wrapping up the inlined body.

From-SVN: r92210
This commit is contained in:
Eric Botcazou 2004-12-15 19:14:55 +00:00
parent 172b87995c
commit 774a9b00ca
4 changed files with 44 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
gcc/ChangeLog
View file

@ -1,3 +1,9 @@
2004-12-15 Eric Botcazou <ebotcazou@libertysurf.fr>
PR c++/17972
* tree-inline.c (expand_call_inline): Set TREE_SIDE_EFFECTS
on the STMT_EXPR wrapping up the inlined body.
2004-12-15 Vladimir Makarov <vmakarov@redhat.com>
Steven Bosscher <stevenb@suse.de>
PR middle end/17340

4
gcc/testsuite/ChangeLog
View file

@ -1,3 +1,7 @@
2004-12-15 Alan Modra <amodra@bigpond.net.au>
* g++.dg/opt/inline9.C: New test.
2004-12-15 Tobias Schlueter <tobias.schlueter@physik.uni-muenchen.de>
PR fortran/18993

29
gcc/testsuite/g++.dg/opt/inline9.C Normal file
View file

@ -0,0 +1,29 @@
// PR c++/17972
// Origin: Michal Ostrowski <mostrows@watson.ibm.com>
// Testcase by Alan Modra <amodra@bigpond.net.au>
// { dg-do run }
// { dg-options "-O" }
// { dg-options "-O -mtune=i686" { target i?86-*-* } }
struct thread_info
{
short preempt_count;
} x;
static inline struct thread_info *cti (void) __attribute__ ((const));
static inline struct thread_info *cti (void)
{
return &x;
}
void fn (void) __attribute__ ((noinline));
void fn (void)
{
++cti()->preempt_count;
}
int main (void)
{
fn ();
return 0;
}

7
gcc/tree-inline.c
View file

@ -1625,8 +1625,11 @@ expand_call_inline (tree *tp, int *walk_subtrees, void *data)
splay_tree_delete (id->decl_map);
id->decl_map = st;
/* The new expression has side-effects if the old one did. */
TREE_SIDE_EFFECTS (expr) = TREE_SIDE_EFFECTS (t);
/* Although, from the semantic viewpoint, the new expression has
side-effects only if the old one did, it is not possible, from
the technical viewpoint, to evaluate the body of a function
multiple times without serious havoc. */
TREE_SIDE_EFFECTS (expr) = 1;
tsi_link_before (&id->tsi, expr, TSI_SAME_STMT);