Re: Teraspace memory allocation

["Simon Coulter" <shc@xxxxxxxxxxxxxxxxx>]

Hello Walden,

You wrote:
>I have a need to allocate more than 16Meg of memory. Teraspace allocations
>seem to be the ticket and it looks like I can simply call _C_TS_malloc,
>but I'm looking for someone that has used it from RPGLE. Anyone? Bueller?

I have been following this thread and don't really see the problem.  Given
that the ILE Concepts manual says that teraspace can be addressed from a
space pointer, I created the following program using my definitions of the
C memory allocation functions including definitions for the _C_TS_xxxxxx
teraspace functions.

H DFTACTGRP(*NO) BNDDIR('QC2LE')

 /define TERASPACE__
 /copy rpgleinc,stdlib

D realPointer     S               *
D based           DS                  BASED(realPointer)
D   fld1                        10
D   fld2                        10I 0
D   fld3                         5S 0

C                   EVAL      realPointer = malloc(100)
C                   EVAL      realPointer = realloc(realPointer:200)
C                   EVAL      based = *BLANKS
C                   EVAL      fld1 = 'TESTXYZ'
C                   EVAL      fld2 = 1234567890
C                   EVAL      fld3 = 98765
C                   CALLP     free(realPointer)

C                   SETON                                        LR
C                   RETURN

This program works on 440 (with latest Cume).  Debug shows me that the
'realPointer' is set to 9E580280000810C0 after the malloc (which actually
uses _C_TS_malloc).  The realloc (which uses _C_TS_realloc) doesn't change
the pointer address (normal behaviour in this sequence of events) but does
seem to increase the allocation and I can successfully address the
teraspace storage.

Removing the /define for TERASPACE__ causes the normal malloc/realloc/free
functions to be used.  The program functions identically but debug shows
that 'realPointer' is set to DE636812A30010B0 which is a "normal" space
pointer address.  Given the difference in addresses in 'realPointer' I
presume the teraspace version is actually using teraspace.

I tried tracing the function calls but that didn't show anything useful
because the IBM service program implementing the memory allocation
functions doesn't log trace data.

Given that my code seems to do what you require I suggest the problem is in
your prototypes.  Either that or you are doing something wierd with the
pointer addresses after the allocation.

In a further attempt to prove that teraspace really was being used (even
though the compiler listing indicated it was) I wrote the following code.

H DFTACTGRP(*NO) BNDDIR('QC2LE')

 /define TERASPACE__
 /copy rpgleinc,stdlib

D realPointer     S               *
D based           DS                  BASED(@based)
D   fld1                        10
D   fld2                        10I 0
D   fld3                         5S 0

D $MAX_ALLOC      C                   CONST(2147483408)

C                   EVAL      realPointer = malloc($MAX_ALLOC)
C                   EVAL      @based = realPointer +
C                                      ( $MAX_ALLOC - %SIZE(based) )
C                   EVAL      based = *BLANKS
C                   EVAL      fld1 = 'TESTXYZ'
C                   EVAL      fld2 = 1234567890
C                   EVAL      fld3 = 98765
C                   CALLP     free(realPointer)

C                   SETON                                        LR
C                   RETURN

It took a mucking huge amount of time to complete the malloc.  WRKSYSACT
also showed the storage management task SMP00001 doing a fair bit of work.
WRKSYSSTS showed disk usage going up.  So I reckon it worked.  A second run
of the same program completed the malloc instantly.  That is consistent
with the ILE Concepts information about there being a single teraspace per
process which lasts until the process is terminated thus the second run
used the space allocated by the first run.

Even though a 2GB allocation indicates teraspace is being used I modified
the same program to increase the allocation by the $MAX_ALLOC amount a few
times using realloc (just as well I have 12GB of free disk space) until I
exceeded the amount I could possibly allocate if I were not using
teraspace.  Each realloc took quite a while (as one would expect -- copying
GB of space must hurt) but they did work.

So the net is that teraspace is accessible and addressable from RPG IV.

Regards,
Simon Coulter.

--------------------------------------------------------------------
   FlyByNight Software         AS/400 Technical Specialists
   http://www.flybynight.com.au/

   Phone: +61 3 9419 0175   Mobile: +61 0411 091 400        /"\
   Fax:   +61 3 9419 0175   mailto: shc@flybynight.com.au   \ /
                                                             X
                 ASCII Ribbon campaign against HTML E-Mail  / \
--------------------------------------------------------------------