/* Interprocedural constant propagation
Copyright (C) 2005, 2006, 2007, 2008, 2009 Free Software Foundation, Inc.
Contributed by Razya Ladelsky <RAZYA@il.ibm.com>
-
+
This file is part of GCC.
-
+
GCC is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free
Software Foundation; either version 3, or (at your option) any later
version.
-
+
GCC is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or
FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
for more details.
-
+
You should have received a copy of the GNU General Public License
along with GCC; see the file COPYING3. If not see
<http://www.gnu.org/licenses/>. */
{
printf ("value is %d",y);
}
-
+
int f (int x)
{
g (x);
f (3);
h (3);
}
-
-
+
+
The IPCP algorithm will find that g's formal argument y is always called
with the value 3.
The algorithm used is based on "Interprocedural Constant Propagation", by
Challahan David, Keith D Cooper, Ken Kennedy, Linda Torczon, Comp86, pg
152-161
-
+
The optimization is divided into three stages:
First stage - intraprocedural analysis
=======================================
This phase computes jump_function and modification flags.
-
+
A jump function for a callsite represents the values passed as an actual
arguments of a given callsite. There are three types of values:
Pass through - the caller's formal parameter is passed as an actual argument.
Constant - a constant is passed as an actual argument.
Unknown - neither of the above.
-
+
The jump function info, ipa_jump_func, is stored in ipa_edge_args
structure (defined in ipa_prop.h and pointed to by cgraph_node->aux)
modified_flags are defined in ipa_node_params structure
(defined in ipa_prop.h and pointed to by cgraph_edge->aux).
-
+
-ipcp_init_stage() is the first stage driver.
Second stage - interprocedural analysis
TOP - unknown.
BOTTOM - non constant.
CONSTANT - constant value.
-
+
Lattice describing a formal parameter p will have a constant value if all
callsites invoking this function have the same constant value passed to p.
-
+
The lattices are stored in ipcp_lattice which is itself in ipa_node_params
structure (defined in ipa_prop.h and pointed to by cgraph_edge->aux).
and many calls redirected back to fit the description above.
-ipcp_insert_stage() is the third phase driver.
-
+
*/
#include "config.h"
if (cgraph_maybe_hot_edge_p (e))
n_hot_calls ++;
}
-
+
if (!n_calls)
{
if (dump_file)
fprintf (dump_file, "Considering %s for cloning; code would shrink.\n",
cgraph_node_name (node));
return true;
- }
+ }
if (!flag_ipa_cp_clone)
{
/* Compute the proper scale for NODE. It is the ratio between the number of
direct calls (represented on the incoming cgraph_edges) and sum of all
- invocations of NODE (represented as count in cgraph_node). */
+ invocations of NODE (represented as count in cgraph_node).
+
+ FIXME: This code is wrong. Since the callers can be also clones and
+ the clones are not scaled yet, the sums gets unrealistically high.
+ To properly compute the counts, we would need to do propagation across
+ callgraph (as external call to A might imply call to non-clonned B
+ if A's clone calls clonned B). */
static void
ipcp_compute_node_scale (struct cgraph_node *node)
{
/* Compute sum of all counts of callers. */
for (cs = node->callers; cs != NULL; cs = cs->next_caller)
sum += cs->count;
+ /* Work around the unrealistically high sum problem. We just don't want
+ the non-cloned body to have negative or very low frequency. Since
+ majority of execution time will be spent in clones anyway, this should
+ give good enough profile. */
+ if (sum > node->count * 9 / 10)
+ sum = node->count * 9 / 10;
if (node->count == 0)
ipcp_set_node_scale (node, 0);
else
ipa_count_arguments (cs);
if (ipa_get_cs_argument_count (IPA_EDGE_REF (cs))
!= ipa_get_param_count (IPA_NODE_REF (cs->callee)))
- {
- /* Handle cases of functions with
- a variable number of parameters. */
- ipa_set_called_with_variable_arg (IPA_NODE_REF (cs->callee));
- if (flag_indirect_inlining)
- ipa_compute_jump_functions (cs);
- }
- else
- ipa_compute_jump_functions (cs);
+ ipa_set_called_with_variable_arg (IPA_NODE_REF (cs->callee));
+ ipa_compute_jump_functions (cs);
}
}
}
ipa_check_create_node_params ();
ipa_check_create_edge_args ();
ipa_register_cgraph_hooks ();
- /* 1. Call the init stage to initialize
+ /* 1. Call the init stage to initialize
the ipa_node_params and ipa_edge_args structures. */
ipcp_init_stage ();
}
static bool
cgraph_gate_cp (void)
{
- /* FIXME lto. IPA-CP does not tolerate running when the inlining decisions
- have not been applied. This happens when WPA modifies the callgraph.
- Since those decisions are not applied until after all the IPA passes
- have been run in LTRANS, this means that IPA passes may see partially
- modified callgraphs. The solution to this is to apply WPA decisions
- early during LTRANS. */
- return flag_ipa_cp && !flag_ltrans;
+ return flag_ipa_cp;
}
struct ipa_opt_pass_d pass_ipa_cp =
ipcp_write_summary, /* write_summary */
ipcp_read_summary, /* read_summary */
NULL, /* function_read_summary */
+ lto_ipa_fixup_call_notes, /* stmt_fixup */
0, /* TODOs */
NULL, /* function_transform */
NULL, /* variable_transform */
OSDN Git Service
