[116] in Pthreads mailing list archive
Re: Linux ELF
daemon@ATHENA.MIT.EDU (Christopher Provenzano)
Mon Sep 11 05:42:04 1995
To: wabbit@corner.iaf.nl
Cc: raeburn@cygnus.com (Ken Raeburn), hjl@nynexst.com, pthreads@MIT.EDU,
proven@MIT.EDU
In-Reply-To: Your message of "Fri, 18 Aug 1995 00:27:59 +0200."
<m0sjDPs-0000edC@gizmo.corner.iaf.nl>
Date: Mon, 11 Sep 1995 04:52:36 EDT
From: Christopher Provenzano <proven@MIT.EDU>
> A message from Ken Raeburn arrived and said:
> > Fails to *complete* (i.e., runs forever), or completes with a failure
> > indication?
>
> SigsegV. Need I say more? The debugger can't help me on this one, I must
> add.
Not sure why it was segfaulting but I've determined the original test
was bogus and I went ahead and rewrote it. Whatever was the problem was in
the test because now it completes successfully under Linux, so the
current floating point save/restore should be just fine.
CAP
For those speed freaks out there here are some bench marks for context
switches I just ran on my 486DX4-100.
With floating point save/restore (p_bench_yield -O 1) 13 usec/context switch
Without (p_bench_yield -O 2) 8 usec/context switch
and on a 486DX-25
With 35 usec/context switch
Without 22 usec/context switch
