Date: Mon, 4 Jan 1999 01:47:22 +0200 (EET) From: Vladimir Kushnir <kushn@mail.kar.net> To: "Jordan K. Hubbard" <jkh@zippy.cdrom.com> Cc: FBSD_CURRENT <freebsd-current@FreeBSD.ORG> Subject: Re: pgcc-1.1.1 Message-ID: <Pine.BSF.4.05.9901040028530.34505-100000@kushnir.kiev.ua> In-Reply-To: <71148.915402183@zippy.cdrom.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Sun, 3 Jan 1999, Jordan K. Hubbard wrote:
> > same some of us need fast code). Works fine. Only question: does anybody
> > know why Linux cross-compiled (within emulation) code runs _much_ faster?
>
> 1. Please don't cross-post to both -current and -questions, whether you
> originated the cross-post or not.
>
Sorry for cross-posting - it wasn't intended. My fault.
> 2. Without any details, even speculating on this question is a meaningless
> exercise and a waste of time. What code? Compiled how? Etc.
Just an example: parts of the BYTE benchmark (nbench) compiled with
pgcc-1.1.1 from ports collection (first) and with pgcc-1.1-1.i586.rpm
for glibc (second):
Native FreeBSD pgcc:
/usr/local/bin/nbench: ELF 32-bit LSB executable, Intel 80386, version 1
(FreeBSD), statically linked, stripped
BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)
TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 46.961 : 1.20 : 0.40
STRING SORT : 2.4854 : 1.11 : 0.17
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
BITFIELD : 6.3344e+06 : 1.09 : 0.23
FP EMULATION : 2.721 : 1.31 : 0.30
FOURIER : 663.85 : 0.75 : 0.42
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
ASSIGNMENT : 0.40413 : 1.54 : 0.40
IDEA : 59.795 : 0.91 : 0.27
HUFFMAN : 37.112 : 1.03 : 0.33
NEURAL NET : 0.69246 : 1.11 : 0.47
Part of Linux pgcc:
file nbench
nbench: ELF 32-bit LSB executable, Intel 80386, version 1 (Linux),
statically linked, stripped
BYTEmark* Native Mode Benchmark ver. 2 (10/95)
Index-split by Andrew D. Balsa (11/97)
Linux/Unix* port by Uwe F. Mayer (12/96,11/97)
TEST : Iterations/sec. : Old Index : New Index
: : Pentium 90* : AMD K6/233*
--------------------:------------------:-------------:------------
NUMERIC SORT : 46.744 : 1.20 : 0.39
STRING SORT : 3.4938 : 1.56 : 0.24
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
BITFIELD : 6.3761e+06 : 1.09 : 0.23
FP EMULATION : 2.794 : 1.34 : 0.31
FOURIER : 1143.7 : 1.30 : 0.73
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
For some reason (can't say yet why) here Linux binary stopped, but pls
note an that while most of the benchmarks are of the same value, in some
cases (underlined values here) linux binary runs essentially faster.
Besides, I had more or less the same experience with g77 (can't provide
with precise numbers now, but sometimes difference was around 20-30%).
Could it be that there're some linux specific optimizations that we don't
use?
>
> - Jordan
>
>
Regards,
Vladimir
===========================|=======================
Vladimir Kushnir |
kushn@mail.kar.net, | Powered by FreeBSD
kushnir@ap3.bitp.kiev.ua |
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.05.9901040028530.34505-100000>