Re: re motion 50 (Just to you...)
On 18 Sep 2013, at 12:49, Vincent Lefevre wrote:
> On 2013-09-18 11:08:11 +0100, David Lester wrote:
>> Richard, Ulrich,
>>
>> Here's a sketch of an EDP algorithm for a modern-style supercomputer chip.
>>
>> Assumptions:
>>
>> (*) EDP is the main thing you are interested in.
>>
>> (*) we have 1024 cpus.
>>
>> Then the algorithm is:
>>
>> stage 1: read-in A[i] and B[i] into core [i]. If there are more than 1024 elements in A/B, then
>> read-in A[i % 1024] and B[i % 1024] to core i.
> [...]
>
> If EDP is to be implemented by a P1788 system library and non-parallel
> programming language like C or Fortran, such an algorithm could not
> be used, because such libraries or programming languages work in a
> sequential model, parallelism being controlled by the developer at
> a higher level. That's also why more complex features like EDP should
> be reserved for the developer or specific libraries, and possibly be
> grouped in another standard.
Vincent,
I'm not disagreeing with you for a moment. I was pointing out that one can obtain higher throughput at lower energy consumption on stock hardware, rather than going for a highly specialised piece of rarely-used hardware.
Regards,
Dave
>
> --
> Vincent Lefèvre <vincent@xxxxxxxxxx> - Web: <http://www.vinc17.net/>
> 100% accessible validated (X)HTML - Blog: <http://www.vinc17.net/blog/>
> Work: CR INRIA - computer arithmetic / AriC project (LIP, ENS-Lyon)