Re: Simple cost model for ipcp

[Diego Novillo <dnovillo at google dot com>]

On 8/24/08 5:56 AM, Jan Hubicka wrote:
> Hi
> this patch adds simple cost model into ipcp.  It makes ipcp to realize
> what constant propagation is for free and what cause code size growth
> and code size growth is limited to 10%.

Nice.


> + 
> + /* Estimate cost of clonning NODE.  */
> + static long

Blank line after comment.

> + ipcp_estimate_clonning_cost (struct cgraph_node *node)
> + {
> +   int freq_sum = 1;
> +   gcov_type count_sum = 1;
> +   struct cgraph_edge *e;
> +   int cost;
> + 
> +   /* When we don't need original clone; we should always propagate.  */
> +   if (!ipcp_need_original_clone_p (node))
> +     return 0;
> + 
> +   for (e = node->callers; e; e = e->next_caller)
> +     if (!bitmap_bit_p (dead_nodes, e->caller->uid)
> +         && !ipcp_need_redirect_p (e))
> +       {
> + 	count_sum += e->count;
> + 	freq_sum += e->frequency + 1;
> +       }
> + 
> +   cost = node->local.inline_summary.self_insns * 1000;
> +   if (max_count)
> +     cost /= count_sum * 1000 / max_count + 1;
> +   else
> +     cost /= freq_sum * 1000 / REG_BR_PROB_BASE + 1;
> +   if (dump_file)
> +     fprintf (dump_file, "Cost of versioning %s is %i, (size: %i, freq: %i)\n",
> +              cgraph_node_name (node), cost, node->local.inline_summary.self_insns,
> + 	     freq_sum);
> +   return false;
> + }

This function returns 'long' but it seems to be returning a boolean
value.  It also computes a cost, but does nothing significant with it.

> + 
> + /* Return number of live constant parameters.  */
> + static int

Blank line after comment.  Seems to happen all over the file, so at
least it's consistent.

>   	 "inline-unit-growth",
>   	 "how much can given compilation unit grow because of the inlining (in percent)",
>   	 30, 0, 0)
> + DEFPARAM(PARAM_IPCP_UNIT_GROWTH,
> + 	 "ipcp-unit-growth",
> + 	 "how much can given compilation unit grow because of the interprocedural constant propagation (in percent)",

How about: "Maximum percentage that a given compilation unit may grow
due to interprocedural constant propagation."


Diego.