[PD] tabread / write broken on 64bit AMD Linux

zack zack at sat.qc.ca
Fri Mar 24 13:56:42 CET 2006


Thanks much for the quick reply ---



I am short on time too,  especially since I would have spend a lot  
of  time just understanding how things work, before making a good  
change.


For the moment,  I can get by with the ugly  fix I made to the  
accessor methods  (read/write) for tabread / tabwrite   and  
tab4read   ---   I did not touch the "tilde" objs, since I haven't  
heard anything "off"  (yet ?) ---   if you know off hand what the  
other affected objects are, or could be ,  could you tell me so I  
could do the same to them --


thanks


Zack










On 23-Mar-06, at 5:13 PM, zmoelnig at iem.at wrote:

> Zitat von zack <zack at sat.qc.ca>:
>
>>
>> On my AMD dual-core 64bit (linux) running pd, recently checked out  
>> of  the CVS:
>>
>>
>> It looks like tabread misses (returns 0 for) every other (odd)   
>> array  value when reading the contents of an array --- UNLESS the  
>> contents  was set using tabwrite, in which case, the data is  
>> there-- as if the  reader was indexing  with   a 32 bit-size  
>> increment...
>
> this is a long known issue (quite as long as there are amd64  
> computers out there)
>
> however, nobody has found time yet to fix it (the fix is not trivial).
> if you don't want to fix it yourself but need this feature, i see  
> only 2 possibilities:
> a) make a amd64 available to mr.puckette (lend him one, buy him  
> one,...)
> b) fund some developer who already has an amd64 but never had the  
> time to fix the tables.
>
> the 2 other solutions are to run pd in 32bit mode (i never found  
> any real speed gains in any application when running in 64bit - but  
> of course it is cooler) OR rethink whether you really cannot do  
> without tables for now (i have been living with pd on x86_64 for >6  
> month now and it seems like i never really needed the tables. but  
> then it is just my laptop which i hardly ever use) all the best.
>
> mfg.asdr.
> IOhannes
> <mime-attachment>





More information about the Pd-list mailing list