This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Fix PR 18746, fab causing more problems



On Dec 1, 2004, at 2:02 PM, Andrew MacLeod wrote:


On Wed, 2004-12-01 at 13:22, Diego Novillo wrote:
On Wed, Dec 01, 2004 at 09:37:55AM -0800, Richard Henderson wrote:

So, the call to mark_new_vars_to_rename actually prevents the
alias analyzer from tripping itself over incomplete alias info.

This is not the ideal solution, but is the least intrusive I can
think of. Ideally, we could probably have a flag for
get_stmt_operands to only do real operands.  Or, the alias
analyzer could do a prior pass of refreshing statement operands
before it clears all the alias sets.

Do you mean call get_stmt_operands() on all stmts before clearing the alias sets? Doesnt help for 4.0, but in 4.1 modify_stmt() should no longer be updating this stuff lazily, so the operands will always be up to date. Meaning you get the equivalent of a prior pass of refreshing for free. It just might change how you view the solution for 4.0....

That is a better way of fixing it ... In fact sine there are other passes which call modify_stmt which causes the same problem, I propose the following patch which just goes through before clearing the aliasing sets and gets the statement operands.

OK? Bootstrapping right now with disable-checking, it is passed stage2
(I turned off ENABLE_CHECKING with stage1).

-- Andrew Pinski

ChangeLog:
	* tree-ssa-alias.c (compute_may_aliases): Get all statement operands
	before doing anything so we don't get operands which are not renamed
	already.

Index: tree-ssa-alias.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/tree-ssa-alias.c,v
retrieving revision 2.60
diff -u -p -r2.60 tree-ssa-alias.c
--- tree-ssa-alias.c 1 Dec 2004 00:33:05 -0000 2.60
+++ tree-ssa-alias.c 1 Dec 2004 19:17:42 -0000
@@ -291,6 +291,18 @@ static void
compute_may_aliases (void)
{
struct alias_info *ai;
+ basic_block bb;
+
+ /* We cannot be lazy getting the statement operands as if we are then we
+ would get non SSA_NAMEs in the VOPs which would we balk at below. */
+ FOR_EACH_BB (bb)
+ {
+ block_stmt_iterator si;
+ for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si))
+ {
+ get_stmt_operands (bsi_stmt (si));
+ }
+ }


memset (&alias_stats, 0, sizeof (alias_stats));



Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]