|
From: <sv...@va...> - 2006-05-09 01:10:40
|
Author: sewardj
Date: 2006-05-07 15:37:03 +0100 (Sun, 07 May 2006)
New Revision: 5882
Log:
Increase scheduling quantum to 100k basic blocks. Nowadays even
modest cpus can run 5-10M memcheck'd bbs per second and the previous
limit of 50k gives a 100Hz switch rate, which causes cache pollution
(a known performance problem) and other context-switch overheads.
Modified:
trunk/coregrind/m_scheduler/scheduler.c
Modified: trunk/coregrind/m_scheduler/scheduler.c
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
--- trunk/coregrind/m_scheduler/scheduler.c 2006-05-04 16:43:34 UTC (rev =
5881)
+++ trunk/coregrind/m_scheduler/scheduler.c 2006-05-07 14:37:03 UTC (rev =
5882)
@@ -96,7 +96,7 @@
/* Defines the thread-scheduling timeslice, in terms of the number of
basic blocks we attempt to run each thread for. Smaller values
give finer interleaving but much increased scheduling overheads. */
-#define SCHEDULING_QUANTUM 50000
+#define SCHEDULING_QUANTUM 100000
=20
/* If False, a fault is Valgrind-internal (ie, a bug) */
Bool VG_(in_generated_code) =3D False;
|