[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-apl] Make infinity and pi symbols into constants?

From: Juergen Sauermann
Subject: Re: [Bug-apl] Make infinity and pi symbols into constants?
Date: Thu, 16 Jan 2014 17:03:05 +0100
User-agent: Mozilla/5.0 (X11; Linux i686; rv:17.0) Gecko/20130330 Thunderbird/17.0.5


not sure. Maybe the University of Saarbrücken can provide copies.
It helps if you speak German (because the thesis is not in English).

/// Jürgen

On 01/16/2014 04:25 PM, Elias Mårtenson wrote:
That sounds interesting. Where can I find a copy of the thesis to read?


On 16 January 2014 22:23, Juergen Sauermann <address@hidden> wrote:

actually, cells have no reference count. Values have one and in the example below
there would be rather few reference count updates.

But obviously +/25000000⍴1 can be parallelized (see my 1990 Ph.D thesis on that  :-) ).

The type checking is not taking too much time, so a specialised type may cost more than
it saves. Internal type conversion costs a little more but can't be avoided by specialised types).
All of that can be parallelized.

//// Jürgen

On 01/16/2014 02:42 PM, Elias Mårtenson wrote:
On 16 January 2014 21:19, Juergen Sauermann <address@hidden> wrote:
GNU APL 1.4: Parallel / multicore implementation of primitive functions and operators

Regarding this part, it does indeed look very interesting.

This reminded me of something I was thinking about one night before going to sleep (to take it with whatever amount of NaCl you want) after having run some simple matrix operations in a profiler.

I realised that for a simple operation such as  +/25000000⍴1, most of the time seems to be spent updating the reference counts for the individual cells. Is this correct?

Is there any obvious way this could be optimised? Say, for example, by creating a specialised array type that does not hold cells but rather raw numbers?


reply via email to

[Prev in Thread] Current Thread [Next in Thread]