Dear Elias,

I saw some posts about this in the past, but my issues regarding basic
functionality overrode my performance concerns at that time.  Now that I
believe basic functionality is getting good, my attention shifts to your
issue.  Several minutes to a fraction of a second to do anything bears
investigation.  I'd bet that, along with your test code, observations, and
input, Juergen can improve this situation significantly.

This is certainly a very significant issue.

Thanks.

Blake



On Sat, Jun 28, 2014 at 8:55 AM, Elias Mårtenson <[email protected]> wrote:

> Well, there is a very specific case that is exposed when trying to build
> up large arrays. That specific case has been discussed here a few times.
> It's underlying cause is that GNU APL proactively copies the array it's
> working on in case the user (i.e. the function that performs work on the
> data) of that array modifies it.
>
> I have had a few stabs at solving this one, but none of my solutions have
> been good enough yet. My experiments have, however, shown the potential for
> improvement. For example, in the program below, the time taken to load a
> 10k-line file was reduced from several munites to a fraction of a second:
>
> https://github.com/lokedhs/apl-tools/blob/master/io.apl#L8
>
> Regards,
> Elias
>
>
> On 28 June 2014 21:44, Blake McBride <[email protected]> wrote:
>
>> Interestingly, I think performance could be the real ticket to a large
>> resurgence of APL.  The world has moved from faster processors to more
>> processors (and threads).  APL is unique in its ability for the programmer
>> to express operations that can be parallelized.  Ultimately, APL's ability
>> to translate array operations into parallel operations will determine its
>> success.  As a first step, taking advantage of multiple cores is critical.
>>  Ultimately adding support for things like Nvidia Cuda would propel GNU APL
>> into a leading language!
>>
>> I know Juergen has worked on getting GNU APL to effectively use multiple
>> cores.  I really didn't follow that much because, at the time, I was just
>> looking for it to work.  Currently, there are no open items I am aware of
>> in terms of getting GNU APL to "work".  I'm sure I and others will find
>> more things as we use the system, but I am also confident that this is a
>> significantly decreasing number.  I am excited about all of this.
>>
>> In terms of your million row table, now that we've gotten as far as we
>> have, I strongly encourage you to create a simple example of the speed
>> issue you encountered for Juergen's evaluation.
>>
>> Thanks.
>>
>> Blake
>>
>>
>>
>> On Sat, Jun 28, 2014 at 8:18 AM, Elias Mårtenson <[email protected]>
>> wrote:
>>
>>> I agree with the suggestion to implement more extensions. There are a
>>> few really interesting ones such as the power operator.
>>>
>>> However, performance is something that needs to be worked on as well.
>>> Right now there are cases where performance deteriorates very quickly, such
>>> as when constructing and working with large arrays. Right now GNU APL is
>>> pretty much impossible to work with for 2D arrays of ten thousand rows or
>>> more.
>>>
>>> I had to work with a million row table once, and I very quickly gave up
>>> the idea of using APL for it.
>>>
>>> Regards,
>>> Elias
>>> On 28 Jun 2014 21:10, "Blake McBride" <[email protected]> wrote:
>>>
>>>> Dear Juergen,
>>>>
>>>> Thanks.  If you are using the Dyalog standard then you should match my
>>>> (WASD custom) keyboard, so I'll be interested in the layout once you
>>>> correct ]KEYBOARD.
>>>>
>>>> I am hoping you can included my WASD keyboard design and setup file
>>>> with GNU APL.  This would make a high quality GNU APL keyboard available to
>>>> everyone.
>>>>
>>>> Lastly, as a tangent, I hoped that down the road, once GNU APL is very
>>>> stable with respect to the IBM standard, you may start to add operators
>>>> provided by other APL vendors - so long as they don't conflict with the IBM
>>>> standard.  If you did this, you would slowly start making use of more of
>>>> the currently unused APL characters.
>>>>
>>>> Thanks for everything.  Your system is very usable now.  I really like
>>>> it.  GNU APL contained the critical mass needed to get me back into APL
>>>> after a 30 year hiatus.  I have spent a bunch of time getting my old tools
>>>> up and running on your system from old printouts I had.  I pretty much have
>>>> that going now.  I am looking forward to making real use of GNU APL and my
>>>> old framework again.  Thanks a lot!!
>>>>
>>>> Blake
>>>>
>>>>
>>>>
>>>> On Sat, Jun 28, 2014 at 7:33 AM, Juergen Sauermann <
>>>> [email protected]> wrote:
>>>>
>>>>>  Hi Blake,
>>>>>
>>>>> these days GNU APL is following Dyalog APL because they are selling
>>>>> their keyboards (and
>>>>> I bought one). I just had forgotten to update the ]KEYBOARD command;
>>>>> the config files shipped
>>>>> with GNU APL should already reflect the Dyalog layout.
>>>>>
>>>>> IBM seems to offer only APL keycaps, and Unicomp did not even bother
>>>>> to answer an inquiry
>>>>> that I sent them.
>>>>>
>>>>> /// Jürgen
>>>>>
>>>>>
>>>>>
>>>>> On 06/27/2014 07:41 PM, Blake McBride wrote:
>>>>>
>>>>> I noticed that for the ]KEYBOARD command, the comment symbol ⍝ appears
>>>>> in two places; on the C key, and on the \ key.  I think, perhaps, that
>>>>> keeping it on the \ key and removing it from the C key makes sense for the
>>>>> following reason.  The Z, X, C, and V keys are often mapped to the very
>>>>> convenient four characters each with a quad and one of the four arrow keys
>>>>> (up, down, right, left) which are very useful for keyed or component file
>>>>> IO.
>>>>>
>>>>>  I also, regretfully, noticed that I created my custom keyboard using
>>>>> the Dyalog standard rather than the standard used by GNU APL.  Apparently,
>>>>> both the Dyalog standard, and the GNU APL standard differ from the IBM
>>>>> standard.  Perhaps Standard is a poor word...
>>>>>
>>>>>
>>>>>  Thanks.
>>>>>
>>>>>  Blake
>>>>>
>>>>>
>>>>>
>>>>> On Fri, Jun 27, 2014 at 10:54 AM, Juergen Sauermann <
>>>>> [email protected]> wrote:
>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> Fred is correct. I have updated the ]KEYBOARD command to show
>>>>>> the new layout (as for dyalog keyboards). ⍙ is Alt-shift-. (two keys
>>>>>> right from M). SVN 345.
>>>>>>
>>>>>> /// Jürgen
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 06/27/2014 05:15 PM, Frederick H. Pitts wrote:
>>>>>>
>>>>>>> Elias,
>>>>>>>
>>>>>>>         My keyboard is configured to have ⍶, ⍹, and ⍷ above ⍺, ⍵ and
>>>>>>> ∊,
>>>>>>> respectively, so that their locations are easy to remember.  I would
>>>>>>> have
>>>>>>> liked to placed ⍙ above ∆ but that location was already occupied by
>>>>>>> ⍋,
>>>>>>> so I placed ⍙ above ⌊ on the 'd' key.
>>>>>>>
>>>>>>> Regards
>>>>>>>
>>>>>>> Fred
>>>>>>>
>>>>>>> On Fri, 2014-06-27 at 22:44 +0800, Elias Mårtenson wrote:
>>>>>>>
>>>>>>>> Any suggestions where I should put the ⍶ and ⍹ symbols on the Emacs
>>>>>>>> keymap? The ]KEYB command doesn't display a mapping for them.
>>>>>>>>
>>>>>>>>
>>>>>>>> Regards,
>>>>>>>> Elias
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>
>>
>

Reply via email to