pjs/tools/reorder/grope.cpp

565 строки
17 KiB
C++
Исходник Обычный вид История

2001-12-01 04:07:39 +03:00
/* ***** BEGIN LICENSE BLOCK *****
* Version: MPL 1.1/GPL 2.0/LGPL 2.1
*
* The contents of this file are subject to the Mozilla Public License Version
* 1.1 (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
* http://www.mozilla.org/MPL/
*
* Software distributed under the License is distributed on an "AS IS" basis,
* WITHOUT WARRANTY OF ANY KIND, either express or implied. See the License
* for the specific language governing rights and limitations under the
* License.
*
* The Original Code is ``grope''
*
* The Initial Developer of the Original Code is Netscape
* Communications Corp. Portions created by the Initial Developer are
* Copyright (C) 2001 the Initial Developer. All Rights Reserved.
*
* Contributor(s):
* Chris Waterson <waterson@netscape.com>
*
* Alternatively, the contents of this file may be used under the terms of
* either the GNU General Public License Version 2 or later (the "GPL"), or
* the GNU Lesser General Public License Version 2.1 or later (the "LGPL"),
* in which case the provisions of the GPL or the LGPL are applicable instead
* of those above. If you wish to allow use of your version of this file only
* under the terms of either the GPL or the LGPL, and not to allow others to
* use your version of this file under the terms of the MPL, indicate your
* decision by deleting the provisions above and replace them with the notice
* and other provisions required by the GPL or the LGPL. If you do not delete
* the provisions above, a recipient may use your version of this file under
* the terms of any one of the MPL, the GPL or the LGPL.
*
* ***** END LICENSE BLOCK ***** */
/*
A program that computes a function ordering for an executable based
on runtime profile information.
This program is directly based on work done by Roger Chickering
<rogc@netscape.com> in
<http://bugzilla.mozilla.org/show_bug.cgi?id=65845>
to implement Nat Friedman's <nat@nat.org> `grope',
_GNU Rope - A Subroutine Position Optimizer_
<http://www.hungry.com/~shaver/grope/grope.ps>
Specifically, it implements the procedure re-ordering algorithm
described in:
K. Pettis and R. Hansen. ``Profile-Guided Core Position.'' In
_Proceedings of the Int. Conference on Programming Language Design
and Implementation_, pages 16-27, June 1990.
*/
#include <assert.h>
#include <fstream>
#include <hash_set>
#include <map>
#include <set>
#include <list>
#include <vector>
#include <limits.h>
#include <unistd.h>
#include <stdio.h>
#include <fcntl.h>
#include "elf_symbol_table.h"
#define _GNU_SOURCE
#include <getopt.h>
elf_symbol_table symtab;
// Straight outta plhash.c!
#define GOLDEN_RATIO 0x9E3779B9U
struct hash<const Elf32_Sym *>
{
size_t operator()(const Elf32_Sym *sym) const {
return (reinterpret_cast<size_t>(sym) >> 4) * GOLDEN_RATIO; }
};
typedef unsigned int call_count_t;
struct call_graph_arc
{
const Elf32_Sym *m_from;
const Elf32_Sym *m_to;
call_count_t m_count;
call_graph_arc(const Elf32_Sym *left, const Elf32_Sym *right, call_count_t count = 0)
: m_count(count)
{
assert(left != right);
if (left > right) {
m_from = left;
m_to = right;
}
else {
m_from = right;
m_to = left;
}
}
friend bool
operator==(const call_graph_arc &lhs, const call_graph_arc &rhs)
{
return (lhs.m_from == rhs.m_from) && (lhs.m_to == rhs.m_to);
}
friend ostream &
operator<<(ostream &out, const call_graph_arc &arc)
{
out << &arc << ": ";
out.form("<(%s %s) %d>",
symtab.get_symbol_name(arc.m_from),
symtab.get_symbol_name(arc.m_to),
arc.m_count);
return out;
}
};
struct hash<call_graph_arc *>
{
size_t
operator()(const call_graph_arc* arc) const
{
size_t result;
result = reinterpret_cast<unsigned long>(arc->m_from);
result ^= reinterpret_cast<unsigned long>(arc->m_to) >> 16;
result ^= reinterpret_cast<unsigned long>(arc->m_to) << 16;
result *= GOLDEN_RATIO;
return result;
}
};
struct equal_to<call_graph_arc *>
{
bool
operator()(const call_graph_arc* lhs, const call_graph_arc *rhs) const
{
return *lhs == *rhs;
}
};
typedef hash_set<call_graph_arc *> arc_container_t;
arc_container_t arcs;
typedef multimap<const Elf32_Sym *, call_graph_arc *> arc_map_t;
arc_map_t from_map;
arc_map_t to_map;
struct less_call_graph_arc_count
{
bool
operator()(const call_graph_arc *lhs, const call_graph_arc *rhs) const
{
if (lhs->m_count == rhs->m_count) {
if (lhs->m_from == rhs->m_from)
return lhs->m_to > rhs->m_to;
return lhs->m_from > rhs->m_from;
}
return lhs->m_count > rhs->m_count;
}
};
typedef set<call_graph_arc *, less_call_graph_arc_count> arc_count_index_t;
bool opt_debug = false;
const char *opt_out = "order.out";
int opt_tick = 0;
bool opt_verbose = false;
int opt_window = 16;
static struct option long_options[] = {
{ "debug", no_argument, 0, 'd' },
{ "exe", required_argument, 0, 'e' },
{ "help", no_argument, 0, '?' },
{ "out", required_argument, 0, 'o' },
{ "tick", optional_argument, 0, 't' },
{ "verbose", no_argument, 0, 'v' },
{ "window", required_argument, 0, 'w' },
{ 0, 0, 0, 0 }
};
//----------------------------------------------------------------------
static void
usage(const char *name)
{
cerr << "usage: " << name << " [options] [<file> ...]" << endl;
cerr << " Options:" << endl;
cerr << " --debug, -d" << endl;
cerr << " Print lots of verbose debugging cruft." << endl;
cerr << " --exe=<image>, -e <image> (required)" << endl;
cerr << " Specify the executable image from which to read symbol information." << endl;
cerr << " --help, -?" << endl;
cerr << " Print this message and exit." << endl;
cerr << " --out=<file>, -o <file>" << endl;
cerr << " Specify the output file to which to dump the symbol ordering of the" << endl;
cerr << " best individual (default is `order.out')." << endl;
cerr << " --tick[=<num>], -t [<num>]" << endl;
cerr << " When reading address data, print a dot to stderr every <num>th" << endl;
cerr << " address processed from the call trace. If specified with no argument," << endl;
cerr << " a dot will be printed for every million addresses processed." << endl;
cerr << " --verbose, -v" << endl;
cerr << " Issue progress messages to stderr." << endl;
cerr << " --window=<num>, -w <num>" << endl;
cerr << " Use a sliding window instead of pagination to score orderings." << endl;
}
/**
* Using the symbol table, map a stream of address references into a
* callgraph.
*/
static void
map_addrs(int fd)
{
long long total_calls = 0;
typedef list<const Elf32_Sym *> window_t;
window_t window;
int window_size = 0;
unsigned int buf[128];
ssize_t cb;
while ((cb = read(fd, buf, sizeof buf)) > 0) {
if (cb % sizeof buf[0])
fprintf(stderr, "unaligned read\n");
unsigned int *addr = buf;
unsigned int *limit = buf + (cb / 4);
for (; addr < limit; ++addr) {
const Elf32_Sym *sym = symtab.lookup(*addr);
if (! sym)
continue;
++total_calls;
window.push_front(sym);
if (window_size >= opt_window)
window.pop_back();
else
++window_size;
window_t::const_iterator i = window.begin();
window_t::const_iterator end = window.end();
for (; i != end; ++i) {
if (sym != *i) {
call_graph_arc *arc;
call_graph_arc key(sym, *i);
arc_container_t::iterator iter = arcs.find(&key);
if (iter == arcs.end()) {
arc = new call_graph_arc(sym, *i);
arcs.insert(arc);
from_map.insert(arc_map_t::value_type(arc->m_from, arc));
to_map.insert(arc_map_t::value_type(arc->m_to, arc));
}
else
arc = const_cast<call_graph_arc *>(*iter);
++(arc->m_count);
}
}
if (opt_verbose && opt_tick && (total_calls % opt_tick == 0)) {
cerr << ".";
flush(cerr);
}
}
}
if (opt_verbose) {
if (opt_tick)
cerr << endl;
cerr << "Total calls: " << total_calls << endl;
}
}
static void
remove_from(arc_map_t& map, const Elf32_Sym *sym, const call_graph_arc *arc)
{
pair<arc_map_t::iterator, arc_map_t::iterator> p
= map.equal_range(sym);
for (arc_map_t::iterator i = p.first; i != p.second; ++i) {
if (i->second == arc) {
map.erase(i);
break;
}
}
}
/**
* The main program
*/
int
main(int argc, char *argv[])
{
const char *opt_exe = 0;
int c;
while (1) {
int option_index = 0;
c = getopt_long(argc, argv, "?de:o:t:vw:", long_options, &option_index);
if (c < 0)
break;
switch (c) {
case '?':
usage(argv[0]);
return 0;
case 'd':
opt_debug = true;
break;
case 'e':
opt_exe = optarg;
break;
case 'o':
opt_out = optarg;
break;
case 't':
opt_tick = optarg ? atoi(optarg) : 1000000;
break;
case 'v':
opt_verbose = true;
break;
case 'w':
opt_window = atoi(optarg);
if (opt_window <= 0) {
cerr << "invalid window size: " << opt_window << endl;
return 1;
}
break;
default:
usage(argv[0]);
return 1;
}
}
// Make sure an image was specified
if (! opt_exe) {
usage(argv[0]);
return 1;
}
// Read the sym table.
symtab.init(opt_exe);
// Process addresses to construct the weighted call graph.
if (optind >= argc) {
map_addrs(STDIN_FILENO);
}
else {
do {
int fd = open(argv[optind], O_RDONLY);
if (fd < 0) {
perror(argv[optind]);
return 1;
}
map_addrs(fd);
close(fd);
} while (++optind < argc);
}
// Emit the ordering.
ofstream out(opt_out);
// Collect all of the arcs into a sorted container, with arcs
// having the largest weight at the front.
arc_count_index_t sorted_arcs(arcs.begin(), arcs.end());
while (sorted_arcs.size()) {
if (opt_debug) {
cerr << "==========================================" << endl << endl;
cerr << "Sorted Arcs:" << endl;
for (arc_count_index_t::const_iterator iter = sorted_arcs.begin();
iter != sorted_arcs.end();
++iter) {
cerr << **iter << endl;
}
cerr << endl << "Arc Container:" << endl;
for (arc_container_t::const_iterator iter = arcs.begin();
iter != arcs.end();
++iter) {
cerr << **iter << endl;
}
cerr << endl << "From Map:" << endl;
for (arc_map_t::const_iterator iter = from_map.begin();
iter != from_map.end();
++iter) {
cerr << symtab.get_symbol_name(iter->first) << ": " << *(iter->second) << endl;
}
cerr << endl << "To Map:" << endl;
for (arc_map_t::const_iterator iter = to_map.begin();
iter != to_map.end();
++iter) {
cerr << symtab.get_symbol_name(iter->first) << ": " << *(iter->second) << endl;
}
cerr << endl;
}
// The first arc in the sorted set will have the largest
// weight. Pull it out, and emit it's sink.
arc_count_index_t::iterator max = sorted_arcs.begin();
call_graph_arc *arc = const_cast<call_graph_arc *>(*max);
sorted_arcs.erase(max);
if (opt_debug)
cerr << "pulling " << *arc << endl;
arcs.erase(arc);
remove_from(from_map, arc->m_from, arc);
remove_from(to_map, arc->m_to, arc);
out << symtab.get_symbol_name(arc->m_to) << endl;
// Merge arc->m_to into arc->m_from. First, modify anything
// that used to point to arc->m_to to point to arc->m_from.
typedef list<call_graph_arc *> arc_list_t;
arc_list_t map_add_queue;
pair<arc_map_t::iterator, arc_map_t::iterator> p;
// Find all the arcs that point to arc->m_to.
p = to_map.equal_range(arc->m_to);
for (arc_map_t::iterator i = p.first; i != p.second; ++i) {
// Remove the arc that points to arc->m_to (`doomed') from
// all of our indicies.
call_graph_arc *doomed = i->second;
const Elf32_Sym *source = doomed->m_from;
sorted_arcs.erase(doomed);
arcs.erase(doomed);
remove_from(from_map, source, doomed);
// N.B. that `doomed' will be removed from the `to_map'
// after the loop completes.
// Create a new (or locate an existing) arc whose source
// was the doomed arc's source, and whose sink is
// arc->m_from (i.e., the node that subsumed arc->m_to).
call_graph_arc *merge;
call_graph_arc key = call_graph_arc(source, arc->m_from);
arc_container_t::iterator iter = arcs.find(&key);
if (iter == arcs.end()) {
merge = new call_graph_arc(source, arc->m_from);
arcs.insert(merge);
from_map.insert(arc_map_t::value_type(merge->m_from, merge));
map_add_queue.push_back(merge);
}
else {
// We found an existing arc: temporarily remove it
// from the sorted index.
merge = const_cast<call_graph_arc *>(*iter);
sorted_arcs.erase(merge);
}
// Include the doomed arc's weight in the merged arc, and
// (re-)insert it into the sorted index.
merge->m_count += doomed->m_count;
sorted_arcs.insert(merge);
delete doomed;
}
to_map.erase(p.first, p.second);
for (arc_list_t::iterator merge = map_add_queue.begin();
merge != map_add_queue.end();
map_add_queue.erase(merge++)) {
to_map.insert(arc_map_t::value_type((*merge)->m_to, *merge));
}
// Now, roll anything that arc->m_to used to point at into
// what arc->m_from now points at.
// Collect all of the arcs that originate at arc->m_to.
p = from_map.equal_range(arc->m_to);
for (arc_map_t::iterator i = p.first; i != p.second; ++i) {
// Remove the arc originating from arc->m_to (`doomed')
// from all of our indicies.
call_graph_arc *doomed = i->second;
const Elf32_Sym *sink = doomed->m_to;
// There shouldn't be any self-referential arcs.
assert(sink != arc->m_to);
sorted_arcs.erase(doomed);
arcs.erase(doomed);
remove_from(to_map, sink, doomed);
// N.B. that `doomed' will be removed from the `from_map'
// once the loop completes.
// Create a new (or locate an existing) arc whose source
// is arc->m_from and whose sink was the removed arc's
// sink.
call_graph_arc *merge;
call_graph_arc key = call_graph_arc(arc->m_from, sink);
arc_container_t::iterator iter = arcs.find(&key);
if (iter == arcs.end()) {
merge = new call_graph_arc(arc->m_from, sink);
arcs.insert(merge);
map_add_queue.push_back(merge);
to_map.insert(arc_map_t::value_type(merge->m_to, merge));
}
else {
// We found an existing arc: temporarily remove it
// from the sorted index.
merge = const_cast<call_graph_arc *>(*iter);
sorted_arcs.erase(merge);
}
// Include the doomed arc's weight in the merged arc, and
// (re-)insert it into the sorted index.
merge->m_count += doomed->m_count;
sorted_arcs.insert(merge);
delete doomed;
}
from_map.erase(p.first, p.second);
for (arc_list_t::iterator merge = map_add_queue.begin();
merge != map_add_queue.end();
map_add_queue.erase(merge++)) {
from_map.insert(arc_map_t::value_type((*merge)->m_from, *merge));
}
}
out.close();
return 0;
}