[Gcl-devel] Re: [Axiom-developer] Re: counting allocated bytes

[Gregory Vanuxem]

Hello Stephen and Camm,

I'm particularly interested by this addition. Can it be considered as
comparable to get-byte-consed in SBCL or CMUCL ?

For information I resurrected code of Axiom relative to Lisp memory
management (g-timer.boot) and I use get- byte-consed on CMUCL and SBCL.
I wonder if this new function can be used. As far as I know (correct me
if I'm wrong) get-byte-consed (SBCL, CMUCL) returns the number of bytes
allocated, freed and "purified" so total-bytes-allocated and
get-bytes-consed seem to have different purposes. Stephen, can you tell
me what is the counterpart of TOTAL-BYTES-ALLOCATED on SBCL (or CMUCL or
Clisp) ?

Greg


Le mercredi 11 juillet 2007 à 16:14 -0400, Camm Maguire a écrit :
> Greetings, and thank you so much!
> 
> This is now in.  Your assumptions are correct, but alas the n_used
> counter is not reliably maintained.  Here is the version now checked
> in commensurate with (room):
> 
> DEFUN_NEW("TOTAL-BYTES-ALLOCATED",object,fStotal_bytes_allocated,SI,0,0,NONE,OO,OO,OO,OO,(void),"")
>  {
>   int i;
>   unsigned long bytes;
>   for(bytes=0,i=t_start; i < t_end; i++) {
>     struct typemanager *tm=tm_of(i);
>     if (tm->tm_type!=i)
>       continue;
>     bytes += TM_NUSED(*tm) * tm->tm_size;
>   }
>   bytes+=ncbpage*PAGESIZE;
>   bytes+=rb_pointer-rb_start;
>   RETURN1(make_fixnum(bytes));
> 
> }
> 
> Quite a big difference.  I need to fix tm_nused someday.  Notes:
> 
> 1) This is a slight overestimate -- we really should subtract the free
>    block space from the contiguous allocation usage
> 2) The algorithm needs correcting when sgc is on -- is one interested
>    in the active read-write set, of the static read-only set here?
> 3) This might motivate me to provide a function to print the
>    contiguous block structure for Matt Kaufmann ...
> 
> Also think I fixed the install of clcs/packages.lisp and bin/gcl.
> Please let me know if problems persist.
> 
> Thanks again so much!
> 
> 
> Stephen Wilson <address@hidden> writes:
> 
> > Hello,
> > 
> > I have been working on evaluating/integrating a Lisp based testing
> > framework into Axiom (based on Lift).  One of the features is the
> > ability to benchmark the running of tests.  
> > 
> > To accommodate the benchmark code, I needed a function which returns
> > the total number of bytes allocated since the image was started.  
> > 
> > I could not find such a function.  Perhaps I missed it? There is
> > SI:ALLOCATED, but it reports information for a specific type only.
> > 
> > I considered writing a function using SI:ALLOCATED and SI:SIZE-OF for
> > personal use.  However, when I considered if such a function would be
> > of use to GCL, I realized it may be better to attempt it in C.
> > 
> > Of course, if the facility already exists, then this is just a
> > learning experience for me.  Attached is a patch against alloc.c.
> > 
> > I have two issues with this code:
> > 
> >   * Is it correct to assume that all entries in tm_table with the
> >     tm_size field equal share the same allocation space?  I assumed
> >     this from the documentation for SI:ALLOCATED.
> > 
> >   * Im slightly concerned about overflow of the `bytes' variable.  Is
> >     this likely?  Should bignum arith be used?
> > 
> > Hopefully this might be of some use to GCL, however I completely
> > understand if it is deemed unnecessary, badly implemented, etc.
> > 
> > Sincerely,
> > Steve
> > 
> > 
>