[Liboil] Fw: ORC performance for NEON

Prateek Mathur hiprateek007 at yahoo.co.in
Wed Jan 5 21:57:13 PST 2011

Hi David,

Thanks for your reply. I have tried running the code many times over but the range of the time taken remains the same. For measuring time I have also checked with clock_gettime which also suggests the same result.
I am not sure if replacing for loop by orcc generated C function is the best way to get optimization and this is where I need some help. If that is not the case should I then go to intrinsics?

Prateek Mathur

--- On Wed, 1/5/11, David Schleef <ds at entropywave.com> wrote:

From: David Schleef <ds at entropywave.com>
Subject: Re: [Liboil] Fw: ORC performance for NEON
To: "Prateek Mathur" <hiprateek007 at yahoo.co.in>
Cc: liboil at lists.freedesktop.org
Date: Wednesday, January 5, 2011, 3:05 AM

On Tue, Jan 04, 2011 at 02:28:37PM +0530, Prateek Mathur wrote:
> Now when I run the ORC binary with ORC_DEBUG=3 I get the statement "compiling for target "neon" " which makes me beleive that ORC is working for correct platform.
> But when I run both the versions the normal addition is working much better (more than 100 times better) than the ORC code.

It looks like you are running the code once.  First of all, any
performance measurements should be done many times with an accurate
time counter (gettimeofday() is not).  Also, the first time you
run an orc function, it will compile the code, which generally
slower than running the resulting code.


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/liboil/attachments/20110106/875fc4e6/attachment.htm>

More information about the Liboil mailing list