mirror of
https://git.savannah.gnu.org/git/guile.git
synced 2025-05-20 11:40:18 +02:00
Remove code to run GC more frequently as process image size increased
* libguile/gc.c: Remove code that would try to run GC more frequently as the process image size was increasing. Before, it was often the case that the heap was the main component of image size, but with expandable stacks and statically allocated data that is no longer true. Also, once scm_gc_register_allocation was incorporated, we don't need to be so conservative any more. It seems this code was simply causing too many gc's to run. Removing it improves some micro-benchmarks; time will tell.
This commit is contained in:
Andy Wingo
parent
aef1fcf94e
commit
0320b1fc3f
1 changed files with 1 additions and 147 deletions
148
libguile/gc.c
148
libguile/gc.c
|
|
@ -1,5 +1,5 @@
|
|||
/* Copyright (C) 1995,1996,1997,1998,1999,2000,2001, 2002, 2003, 2006,
|
||||
* 2008, 2009, 2010, 2011, 2012, 2013 Free Software Foundation, Inc.
|
||||
* 2008, 2009, 2010, 2011, 2012, 2013, 2014 Free Software Foundation, Inc.
|
||||
*
|
||||
* This library is free software; you can redistribute it and/or
|
||||
* modify it under the terms of the GNU Lesser General Public License
|
||||
|
|
@ -684,151 +684,9 @@ accumulate_gc_timer (void * hook_data SCM_UNUSED,
|
|||
return NULL;
|
||||
}
|
||||
|
||||
/* Return some idea of the memory footprint of a process, in bytes.
|
||||
Currently only works on Linux systems. */
|
||||
static size_t
|
||||
get_image_size (void)
|
||||
{
|
||||
unsigned long size, resident, share;
|
||||
size_t ret = 0;
|
||||
|
||||
FILE *fp = fopen ("/proc/self/statm", "r");
|
||||
|
||||
if (fp && fscanf (fp, "%lu %lu %lu", &size, &resident, &share) == 3)
|
||||
ret = resident * 4096;
|
||||
|
||||
if (fp)
|
||||
fclose (fp);
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
/* These are discussed later. */
|
||||
static size_t bytes_until_gc = DEFAULT_INITIAL_HEAP_SIZE;
|
||||
static scm_i_pthread_mutex_t bytes_until_gc_lock = SCM_I_PTHREAD_MUTEX_INITIALIZER;
|
||||
|
||||
/* Make GC run more frequently when the process image size is growing,
|
||||
measured against the number of bytes allocated through the GC.
|
||||
|
||||
If Guile is allocating at a GC-managed heap size H, libgc will tend
|
||||
to limit the process image size to H*N. But if at the same time the
|
||||
user program is mallocating at a rate M bytes per GC-allocated byte,
|
||||
then the process stabilizes at H*N*M -- assuming that collecting data
|
||||
will result in malloc'd data being freed. It doesn't take a very
|
||||
large M for this to be a bad situation. To limit the image size,
|
||||
Guile should GC more often -- the bigger the M, the more often.
|
||||
|
||||
Numeric functions that produce bigger and bigger integers are
|
||||
pessimal, because M is an increasing function of time. Here is an
|
||||
example of such a function:
|
||||
|
||||
(define (factorial n)
|
||||
(define (fac n acc)
|
||||
(if (<= n 1)
|
||||
acc
|
||||
(fac (1- n) (* n acc))))
|
||||
(fac n 1))
|
||||
|
||||
It is possible for a process to grow for reasons that will not be
|
||||
solved by faster GC. In that case M will be estimated as
|
||||
artificially high for a while, and so GC will happen more often on
|
||||
the Guile side. But when it stabilizes, Guile can ease back the GC
|
||||
frequency.
|
||||
|
||||
The key is to measure process image growth, not mallocation rate.
|
||||
For maximum effectiveness, Guile reacts quickly to process growth,
|
||||
and exponentially backs down when the process stops growing.
|
||||
|
||||
See http://thread.gmane.org/gmane.lisp.guile.devel/12552/focus=12936
|
||||
for further discussion.
|
||||
*/
|
||||
static void *
|
||||
adjust_gc_frequency (void * hook_data SCM_UNUSED,
|
||||
void *fn_data SCM_UNUSED,
|
||||
void *data SCM_UNUSED)
|
||||
{
|
||||
static size_t prev_image_size = 0;
|
||||
static size_t prev_bytes_alloced = 0;
|
||||
size_t image_size;
|
||||
size_t bytes_alloced;
|
||||
|
||||
scm_i_pthread_mutex_lock (&bytes_until_gc_lock);
|
||||
bytes_until_gc = GC_get_heap_size ();
|
||||
scm_i_pthread_mutex_unlock (&bytes_until_gc_lock);
|
||||
|
||||
image_size = get_image_size ();
|
||||
bytes_alloced = GC_get_total_bytes ();
|
||||
|
||||
#define HEURISTICS_DEBUG 0
|
||||
|
||||
#if HEURISTICS_DEBUG
|
||||
fprintf (stderr, "prev image / alloced: %lu / %lu\n", prev_image_size, prev_bytes_alloced);
|
||||
fprintf (stderr, " image / alloced: %lu / %lu\n", image_size, bytes_alloced);
|
||||
fprintf (stderr, "divisor %lu / %f\n", free_space_divisor, target_free_space_divisor);
|
||||
#endif
|
||||
|
||||
if (prev_image_size && bytes_alloced != prev_bytes_alloced)
|
||||
{
|
||||
double growth_rate, new_target_free_space_divisor;
|
||||
double decay_factor = 0.5;
|
||||
double hysteresis = 0.1;
|
||||
|
||||
growth_rate = ((double) image_size - prev_image_size)
|
||||
/ ((double)bytes_alloced - prev_bytes_alloced);
|
||||
|
||||
#if HEURISTICS_DEBUG
|
||||
fprintf (stderr, "growth rate %f\n", growth_rate);
|
||||
#endif
|
||||
|
||||
new_target_free_space_divisor = minimum_free_space_divisor;
|
||||
|
||||
if (growth_rate > 0)
|
||||
new_target_free_space_divisor *= 1.0 + growth_rate;
|
||||
|
||||
#if HEURISTICS_DEBUG
|
||||
fprintf (stderr, "new divisor %f\n", new_target_free_space_divisor);
|
||||
#endif
|
||||
|
||||
if (new_target_free_space_divisor < target_free_space_divisor)
|
||||
/* Decay down. */
|
||||
target_free_space_divisor =
|
||||
(decay_factor * target_free_space_divisor
|
||||
+ (1.0 - decay_factor) * new_target_free_space_divisor);
|
||||
else
|
||||
/* Jump up. */
|
||||
target_free_space_divisor = new_target_free_space_divisor;
|
||||
|
||||
#if HEURISTICS_DEBUG
|
||||
fprintf (stderr, "new target divisor %f\n", target_free_space_divisor);
|
||||
#endif
|
||||
|
||||
if (free_space_divisor + 0.5 + hysteresis < target_free_space_divisor
|
||||
|| free_space_divisor - 0.5 - hysteresis > target_free_space_divisor)
|
||||
{
|
||||
free_space_divisor = lround (target_free_space_divisor);
|
||||
#if HEURISTICS_DEBUG
|
||||
fprintf (stderr, "new divisor %lu\n", free_space_divisor);
|
||||
#endif
|
||||
GC_set_free_space_divisor (free_space_divisor);
|
||||
}
|
||||
}
|
||||
|
||||
prev_image_size = image_size;
|
||||
prev_bytes_alloced = bytes_alloced;
|
||||
|
||||
return NULL;
|
||||
}
|
||||
|
||||
/* The adjust_gc_frequency routine handles transients in the process
|
||||
image size. It can't handle instense non-GC-managed steady-state
|
||||
allocation though, as it decays the FSD at steady-state down to its
|
||||
minimum value.
|
||||
|
||||
The only real way to handle continuous, high non-GC allocation is to
|
||||
let the GC know about it. This routine can handle non-GC allocation
|
||||
rates that are similar in size to the GC-managed heap size.
|
||||
*/
|
||||
|
||||
void
|
||||
scm_gc_register_allocation (size_t size)
|
||||
{
|
||||
|
|
@ -866,10 +724,6 @@ scm_init_gc ()
|
|||
scm_c_hook_add (&scm_before_gc_c_hook, start_gc_timer, NULL, 0);
|
||||
scm_c_hook_add (&scm_after_gc_c_hook, accumulate_gc_timer, NULL, 0);
|
||||
|
||||
/* GC_get_heap_usage does not take a lock, and so can run in the GC
|
||||
start hook. */
|
||||
scm_c_hook_add (&scm_before_gc_c_hook, adjust_gc_frequency, NULL, 0);
|
||||
|
||||
GC_set_start_callback (run_before_gc_c_hook);
|
||||
|
||||
#include "libguile/gc.x"
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue