IPA merge part 2: operand-cache privatization

[Andrew MacLeod]

On Thu, 2006-11-30 at 10:06 +0100, Jan Hubicka wrote:
> Hi,

>   
> + /* Per-function operand caches.  */
> + struct ssa_operands GTY(()) {
> +    /* These arrays are the cached operand vectors for call clobbered calls.  */

^^^ this comment doesn't seem relevant here.  I think its a left over
from before dberlin changed the call clobbered stuff. The 2 vectors
disappeared, but the comment didn't :-)

> --- 243,260 ----
>   void
>   init_ssa_operands (void)
>   {
> !   if (!n_initialized++)
> !     {
> !       build_defs = VEC_alloc (tree, heap, 5);
> !       build_uses = VEC_alloc (tree, heap, 10);
> !       build_vuses = VEC_alloc (tree, heap, 25);
> !       build_v_may_defs = VEC_alloc (tree, heap, 25);
> !       build_v_must_defs = VEC_alloc (tree, heap, 25);
> !     }

OK. If we ever need to make the ssa optimizer pass
reentrant/pipelined/whatever, we will deal with these build-vector
static variables then. 

Looks good to me.

Andrew