Re: new-regalloc performance

[Daniel Berlin <dberlin at redhat dot com>]

On Thu, 1 Feb 2001, Richard Henderson wrote:

> You'll want to examine gcc.c-torture/compile/20001226-1.c.
>
> I'd fixed the previous big memory consumers for this test,
> but on the branch we're back to eating 470M of memory.

No surprise here, i missed a few memory leaks.
I'm on it.

>
> Moreover, (alpha) profile data shows
>
>  global alloc          : 192.79 (69%) usr   1.19 (34%) sys 194.04 (68%) wall
>  TOTAL                 : 280.18             3.46           283.75
>
>   %   cumulative   self              self     total
>  time   seconds   seconds    calls  ms/call  ms/call  name
>  67.49    161.15   161.15  2421348     0.07     0.07  varray_contains
>   6.60    176.91    15.76        3  5254.88  5281.60  compute_hash_table
>   4.68    188.09    11.18       13   859.83   874.58  verify_flow_info
>
> It looks like things could be fixed by using a regset
> instead of, or in addition to, a varray.

I added an sbitmap that mirrors the nodes on the selectStack, so we don't
have to look at it at all, just test a single bit (we would walk the
entire stack before until we found it).

I still have a large amount of tuning to do in the new allocator.



> >
> r~
>