Author: Nolan Denson
Date: 15:15:29 09/27/01
Ok I did some researching about the P4 code and I got this response from some key programmers from Intel. All Below is the response: The only comment I would add is that for the benchmark crafty, we are handicapped by not being allowed to change source by the rules of the benchmark whereas for an external application crafty, it may be possible to do some rewrite and see more benefits. - Milind -----Original Message----- From: Sent: Thursday, September 27, 2001 1:19 PM To: 'Nolan Denson'; Subject: RE: P4 code Optimization Hi Nolan, A few quick responses to your question, my manager may have some more input. I have been working on Intel's vectorizer for quite some years now, and the responses that we have got vary from very negative to great enthusiasm. It is true that vectorization does not help every application, but it can help to improve performance substantially. For example, in the context of FP intensive kernels, great speedups can be obtained (see e.g. http://www.aceshardware.com/Spades/read.php?article_id=40000193 <http://www.aceshardware.com/Spades/read.php?article_id=40000193> ). In the context of whole application, speedup is obviously limited by the fraction of the execution time spent in vector loops (Amdahl's law). Besides, vectorization alone is not always the way to befenit from P4. This processor has other rich features that our compiler can exploit to boost performance. Another note. SSE2 *does* support integer operations, in fact it is a full extension of 64-bit MMX(tm) technology into 128-bits. So tell your friends! SSE/SSE2 is not just FP alone, it is also integers. Crafty, indeed is very hard to optimize, and vectorization has not given us any benefit at all so far. But maybe you are happy to hear that our team has two chess hobbyist as well, and that our near future focus will be on improving integer performance. I keep you posted if we have something interesting to report. Aart
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.