|
From: | matt |
Subject: | Re: octave vs. matlab vectorization |
Date: | Tue, 15 Sep 2009 10:28:35 +0100 |
Hi John,First let me take this opportunity to thank you for all the great work you have done on Octave over the years. I am using it constantly and couldn't be happier.
On vectorization issue, I was staggered for two reasons. First, I couldn't believe that Matlab had "given up" on software vectorization. Did they tell anyone?! I only really care because I sometimes send Octave code to Matlab users; it's nice to know what the differences are. The second reason is that I was surprised the JIT compiler worked as well as it did--to use Jaroslav's quote, the elephant can dance.
Having said that, James found the vectorized code ran 5x faster on his Matlab--so who knows what is going on? Perhaps the Matlab compiler is cheating?
Anyway, many thanks again for Octave. Best, Matt. On 14 Sep 2009, at 18:29, John W. Eaton wrote:
On 14-Sep-2009, matt wrote: | ah, no! code 1 was *faster* under matlab! the ratio is 0.19:1 | | i find this result quite staggering. Why? As someone else already explained, Matlab has a JIT compiler that can, in some cases at least, compile loops into native code so that it runs faster and Octave does not have a JIT compiler. jwe
[Prev in Thread] | Current Thread | [Next in Thread] |