[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: help with vectorization of update rule
From: |
John W. Eaton |
Subject: |
Re: help with vectorization of update rule |
Date: |
Thu, 21 Oct 2010 16:00:17 -0400 |
On 21-Oct-2010, James Sherman Jr. wrote:
| On Thu, Oct 21, 2010 at 11:48 AM, grg <address@hidden> wrote:
|
| I believe that it is because the vectorization in this case takes many
| more multiplications and additions than the non-vectorized version. Your
| initial implementation takes one matrix multiplication (s by s times s by 20)
| per "it" (T of them in this case).
|
| The vectorized version essentially calculates the matrix exponential at every
| time instant independently of each other, so for say it = 3, it calculates A^3
| then multiplies it by x0 (2 multiplications of s by s, then one of s by s
| times s by 20). Then for it = 4, it calculates A^4 then multiplies it by x0.
| (3 multiplications of s by s, then one of s by s times s by 20) So you have
| many more calculations to perform.
|
| Thats my semi-informed guess of why the vectorization in this case takes much
| longer.
Also, the loop is only from 1:20. Try setting T to something like
5000 and then compare.
jwe
- help with vectorization of update rule, grg, 2010/10/21
- help with vectorization of update rule, grg, 2010/10/21
- Re: help with vectorization of update rule, Søren Hauberg, 2010/10/21
- Re: help with vectorization of update rule, Søren Hauberg, 2010/10/22
- Re: help with vectorization of update rule, grg, 2010/10/22
- Re: help with vectorization of update rule, Søren Hauberg, 2010/10/22
- Re: help with vectorization of update rule, Moo, 2010/10/23
- Re: help with vectorization of update rule, Jaroslav Hajek, 2010/10/25