This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[Java] PATCH for optimization/12547
- From: Andrew Haley <aph at redhat dot com>
- To: Jason Merrill <jason at redhat dot com>
- Cc: jsturm at redhat dot com, gcc-patches at gcc dot gnu dot org
- Date: Thu, 13 Nov 2003 08:33:50 +0000
- Subject: [Java] PATCH for optimization/12547
- References: <wvlislpt65u.fsf@prospero.boston.redhat.com>
Jason Merrill writes:
> This bug on the tree-ssa branch turned out to be an unsharing problem.
> jsturm had removed java_tree_inlining_walk_subtrees as part of his work on
> Java gimplification; as a result, the unsharing pass wasn't walking into
> BLOCKs, so it wasn't really doing any unsharing of Java code. The affected
> testcases have shared COND_EXPRs.
>
> I restored the hook, but things were still broken. The problem turned out
> to be that it was walking the TREE_CHAIN chain of a BLOCK_EXPR_BODY.
> TREE_CHAIN has no defined meaning for an arbitrary expression, and the Java
> frontend uses it for chaining the loop stack and such. So if the thing
> directly inside a BLOCK is a loop, we walk from there to the next outermost
> loop. Boom.
>
> Fixed thus. Andrew, was there something else you were trying to do with
> this code?
>
> Tested x86_64-pc-linux-gnu, applied to trunk. tree-ssa version follows.
>
> 2003-11-12 Jason Merrill <jason@redhat.com>
>
> PR optimization/12547
> * lang.c (java_tree_inlining_walk_subtrees): Just walk
> BLOCK_EXPR_BODY directly.
>
> *** java/lang.c.~1~ 2003-11-03 16:39:20.000000000 -0500
> --- java/lang.c 2003-11-12 18:34:43.000000000 -0500
> *************** java_tree_inlining_walk_subtrees (tree *
> *** 840,860 ****
> switch (code)
> {
> case BLOCK:
> ! if (BLOCK_EXPR_BODY (t))
> ! {
> ! tree *prev = &BLOCK_EXPR_BODY (*tp);
> ! while (*prev)
> ! {
> ! WALK_SUBTREE (*prev);
> ! prev = &TREE_CHAIN (*prev);
> ! }
> ! }
> return NULL_TREE;
> - break;
>
> default:
> return NULL_TREE;
> }
> }
>
> /* Called from unsafe_for_reeval. */
> --- 840,853 ----
> switch (code)
> {
> case BLOCK:
> ! WALK_SUBTREE (BLOCK_EXPR_BODY (t));
> return NULL_TREE;
>
> default:
> return NULL_TREE;
> }
> +
> + #undef WALK_SUBTREE
> }
>
> /* Called from unsafe_for_reeval. */
I don't know for certain that a BLOCK must have a BLOCK_EXPR_BODY.
Maybe you shouldn't remove if (BLOCK_EXPR_BODY (t)).
Andrew.