Date: Thu, 5 Sep 2002 10:35:10 -0400 (EDT) From: Andrew Gallatin <gallatin@cs.duke.edu> To: ticso@cicely.de Cc: John Baldwin <jhb@FreeBSD.ORG>, freebsd-alpha@FreeBSD.ORG Subject: Re: alpha performance on -current Message-ID: <15735.27550.588643.958037@grasshopper.cs.duke.edu> In-Reply-To: <20020905084824.GJ87724@cicely5.cicely.de> References: <XFMail.20020904090455.jhb@FreeBSD.org> <15734.29725.515274.183629@grasshopper.cs.duke.edu> <20020904223255.GI87724@cicely5.cicely.de> <15734.39976.326598.176742@grasshopper.cs.duke.edu> <20020905084824.GJ87724@cicely5.cicely.de>
next in thread | previous in thread | raw e-mail | index | archive | help
Bernd Walter writes:
> > with MB:
> > 8699.25 real 6985.64 user 1379.72 sys
> >
> > without MB:
> > 8298.44 real 7010.03 user 969.02 sys
>
> Would be insteresting what happens if you remove only the obsolete
> barriers.
The lmbench perf is just the same as with the wrongheaded way I was
doing it earlier -- 1.6us with kern.giant.proc=1, and 1.4us with
kern.giant.proc=0. To refresh, it is 2.0/1.6 with the stock code.
I'm starting a buildworld now.
> What is required is the following:
> ...
> /* acq */
> atomic (locked) memory access
> mb
> ...
> /* rel */
> mb
> atomic (locked) memory access
> ...
>
> But what we currently have is
> ...
> /* acq */
> atomic (locked) memory access
> mb
> mb
> ...
> /* rel */
> mb
> atomic (locked) memory access
> mb
I'm using the appended diff, which I think is what you're suggesting:
Cheers,
Drew
Index: atomic.h
===================================================================
RCS file: /home/ncvs/src/sys/alpha/include/atomic.h,v
retrieving revision 1.14
diff -u -r1.14 atomic.h
--- atomic.h 17 May 2002 05:45:39 -0000 1.14
+++ atomic.h 5 Sep 2002 14:08:50 -0000
@@ -416,35 +416,61 @@
static __inline u_int32_t
atomic_cmpset_acq_32(volatile u_int32_t *p, u_int32_t cmpval, u_int32_t newval)
{
- int retval;
-
- retval = atomic_cmpset_32(p, cmpval, newval);
- alpha_mb();
- return (retval);
+ return (atomic_cmpset_32(p, cmpval, newval));
}
static __inline u_int32_t
atomic_cmpset_rel_32(volatile u_int32_t *p, u_int32_t cmpval, u_int32_t newval)
{
+ u_int32_t ret;
+
alpha_mb();
- return (atomic_cmpset_32(p, cmpval, newval));
+ __asm __volatile (
+ "1:\tldl_l %0, %4\n\t" /* load old value */
+ "cmpeq %0, %2, %0\n\t" /* compare */
+ "beq %0, 2f\n\t" /* exit if not equal */
+ "mov %3, %0\n\t" /* value to store */
+ "stl_c %0, %1\n\t" /* attempt to store */
+ "beq %0, 3f\n\t" /* if it failed, spin */
+ "2:\n" /* done */
+ ".section .text3,\"ax\"\n" /* improve branch prediction */
+ "3:\tbr 1b\n" /* try again */
+ ".previous\n"
+ : "=&r" (ret), "=m" (*p)
+ : "r" ((long)(int)cmpval), "r" (newval), "m" (*p)
+ : "memory");
+
+ return ret;
}
static __inline u_int64_t
atomic_cmpset_acq_64(volatile u_int64_t *p, u_int64_t cmpval, u_int64_t newval)
{
- int retval;
-
- retval = atomic_cmpset_64(p, cmpval, newval);
- alpha_mb();
- return (retval);
+ return (atomic_cmpset_64(p, cmpval, newval));
}
static __inline u_int64_t
atomic_cmpset_rel_64(volatile u_int64_t *p, u_int64_t cmpval, u_int64_t newval)
{
+ u_int64_t ret;
+
alpha_mb();
- return (atomic_cmpset_64(p, cmpval, newval));
+ __asm __volatile (
+ "1:\tldq_l %0, %4\n\t" /* load old value */
+ "cmpeq %0, %2, %0\n\t" /* compare */
+ "beq %0, 2f\n\t" /* exit if not equal */
+ "mov %3, %0\n\t" /* value to store */
+ "stq_c %0, %1\n\t" /* attempt to store */
+ "beq %0, 3f\n\t" /* if it failed, spin */
+ "2:\n" /* done */
+ ".section .text3,\"ax\"\n" /* improve branch prediction */
+ "3:\tbr 1b\n" /* try again */
+ ".previous\n"
+ : "=&r" (ret), "=m" (*p)
+ : "r" (cmpval), "r" (newval), "m" (*p)
+ : "memory");
+
+ return ret;
}
#define atomic_cmpset_acq_int atomic_cmpset_acq_32
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-alpha" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?15735.27550.588643.958037>