Message-Id:	<199906291400.KAA09526@delorie.com>
Sender:	root AT delorie DOT com
Date:	Tue, 29 Jun 1999 10:56:07 -0300
From:	salvador <salvador AT inti DOT gov DOT ar>
Organization:	INTI
X-Accept-Language:	es-AR, en, es
To:	djgpp-announce AT delorie DOT com, allegro AT canvaslink DOT com, balug-lst AT balug DOT org DOT ar
Subject:	Benchmarks: Beating MSVC in string sort.
Reply-To:	djgpp AT delorie DOT com

Hi All:

  I was playing a little more with the benchmarks here are some details:

1) I emulated the MSVC aligment using -malign-jumps=0, -malign-loops=0,
-malign-functions=4, -malign-double and doing a trick to prevent ld from
breaking my aligment.
  The result was very small, some times faster, some times slower. Nothing
that really makes the difference. I only tested with the K6-2.

2) I taked a look to the code generated by gcc 2.95 and is quite close to
the code from MSVC, things that still slower:

a) Slower epilog and not generated for super-scalar processors.
b) gcc 2.95 is emiting 2 stack adjusts instead of just one.

After trying regparm(3) for the numeric sort I got about 4% of speeup and
the code is now only 5% slower than MSVC code.

3) I was wondering why MSVC was so much faster for string sorts, so I
taked a quick look to this benchmark and saw it uses memmove, so I
replaced djgpp v2.02 memmove to the one I submited for v2.03 (properly
aligning the routine) and now gcc is 13% faster than MSVC ;-)
The string sort jumped from around 0.58 to 1.13! (1=MSVC 5.0).

SET

--
Salvador Eduardo Tropea (SET). (Electronics Engineer)
Visit my home page: http://welcome.to/SetSoft or
http://www.geocities.com/SiliconValley/Vista/6552/
Alternative e-mail: set-soft AT usa DOT net set AT computer DOT org
                    set AT ieee DOT org set-soft AT bigfoot DOT com
Address: Curapaligue 2124, Caseros, 3 de Febrero
Buenos Aires, (1678), ARGENTINA Phone: +(5411) 4759 0013

- Raw text -