g++ V3.3.1 not faster with __restrict

g++ V3.3.1 not faster with __restrict

[Kaspar Thommen]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="us-ascii", Size: 895 bytes --]

Hi,

I ran some benchmarks using g++ V3.3.1. Two functions are called from a main
program. While both doing exactly the same array manipulation (dot product
on doubles), one of the functions uses the __restrict keyword to tell the
compiler that pointer aliasing is "turned off" in that function.

To my surprise, the __restrict version is not faster than the conventional
one. In fact, they show the same speed. Do I have to specify any compilation
option to tell g++ to really take advantage from the __restrict information?
Didn't find anything in the docs however. I tried using -o, -o2, -o3, giving
always equal speed for both functions.

Regards
Kaspar

-- 
GMX Weihnachts-Special: Seychellen-Traumreise zu gewinnen!

Rentier entlaufen. Finden Sie Rudolph! Als Belohnung winken
tolle Preise. http://www.gmx.net/de/cgi/specialmail/

+++ GMX - die erste Adresse für Mail, Message, More! +++

Re: g++ V3.3.1 not faster with __restrict

[Falk Hueffner]

"Kaspar Thommen" <kaspar@gmx.ch> writes:

> To my surprise, the __restrict version is not faster than the conventional
> one. In fact, they show the same speed. Do I have to specify any compilation
> option to tell g++ to really take advantage from the __restrict information?

No, although probably -O2 is required to show any difference.

> Didn't find anything in the docs however. I tried using -o, -o2,
> -o3, giving always equal speed for both functions.

Try -funroll-loops or -funroll-all-loops, this produces more
possibilities to take advantage of the non-aliasing.

-- 
	Falk