[prev in list] [next in list] [prev in thread] [next in thread] 

List:       linux-arm-kernel
Subject:    Re: FastFPE, Floating point emulation issues
From:       Nicholas Clark <nick () unfortu ! net>
Date:       2002-06-30 17:40:53
[Download RAW message or body]

On Thu, Feb 07, 2002 at 02:54:58PM +0100, Teichmann Peter wrote:
> So one could assume that emulation does not cost much. But using FastFPE
> I get approximately 1.2 MFlops, and know that it is spending more time in
> undefined instruction trap handler and with parsing the FP instructions than
> with the actual fp calculations, even as it looks ahead if the next
> instructions are fp instructions too. This saves the undefined instruction
> trap overhead, but not the parsing overhead.

I can see some scope for minor optimisations in here.
(re-ordering to move LDRs earlier and gain a cycle, and one point where RSB
can be used to save an instruction)
Not that this would give much (if even measurable) speed gain.
Where should patches be sent? Is there any test suite to verify that no bugs
are introduced by doing this sort of thing?

Nicholas Clark
-- 
Even better than the real thing:	http://nms-cgi.sourceforge.net/

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ/Etiquette:       http://www.arm.linux.org.uk/armlinux/mailinglists.php

[prev in list] [next in list] [prev in thread] [next in thread]