Front page | perl.perl5.porters |
Postings from December 2000
The following simple patch decreases the overhead of
creation/destruction of objects 3-4 times (estimated for optimized
build: I get 2.2x times speedup for -g build). Tested with: create
an array reference with 100 objects with make_many(100) in a loop:
$x = make_many(100) while --$c;
(XSUB code at the end). My estimates: if the time to create one
object is 1 unit, then the time to DESTROY it is around 1 unit, and
the time to (re)store signals is around 9 units. [The rest is the
dispatch overhead.]
Enjoy,
Ilya
P.S. Well, I applied another patch to my Perl so it has a sped-up
DESTROY. But the digits without this additional patch should be
the same.
--- ./config.h~ Fri Nov 17 02:57:04 2000
+++ ./config.h Fri Dec 15 03:20:13 2000
@@ -1905,7 +1905,7 @@
* traditional longjmp() if siglongjmp isn't available.
* See HAS_SIGSETJMP.
*/
-#define HAS_SIGSETJMP /**/
+/*#define HAS_SIGSETJMP*/ /**/
#ifdef HAS_SIGSETJMP
#define Sigjmp_buf sigjmp_buf
#define Sigsetjmp(buf,save_mask) sigsetjmp((buf),(save_mask))
==================================================================
int no_object;
#define noobject() (no_object = 1)
#define DESTROY(sv) 1
HV *mystash;
SV* make_many(int c)
{
SV *rv;
AV *av = newAV();
av_extend(av,c-1);
while (c--) {
SV *ss = NEWSV(833, 0);
SV *sv = newRV_noinc(ss);
sv_bless(sv, mystash);
if (no_object) {
SvOBJECT_off(ss);
SvREFCNT_dec(mystash);
}
av_store(av, c, sv);
}
return newRV_noinc((SV*)av);
}
MODULE = time_DESTROY PACKAGE = time_DESTROY_d
void
DESTROY(sv)
SV *sv
MODULE = time_DESTROY PACKAGE = time_DESTROY
void
noobject()
SV*
make_many(c)
int c
BOOT:
mystash = gv_stashpv("time_DESTROY", 1);
Thread Previous
|
Thread Next