Re: Visit

[Camm Maguire-2]

Greetings!  I've got a local GCL version here which is almost ready
for upload.  Massive inlining work -- in short, just about every lisp
function can be automatically inlined and accordingly simplified under
one single fast-link function calling structure.  No need for
automatic recompilation at load time.  Full interface to all gmp
integer functions in their own package ....

Inlining everything in sight is obviously excessive.  I'm writing a
control function with user-settable parameters which will decide
whether or not to opt out of an inline.  Counting branch eliminations,
detecting whether fixnums and or explicit functions are provided as
arguments appear to me to be the key deciding factors.  Perhaps also
knowing that certain sequence arguments are lists.  What sort of
interface do you feel would be most useful in managing the tradeoff
between size and speed?  

Take care,

Robert Boyer <boyer@...> writes:

>> Separately, does Bob Boyer have a few extra cycles for some email
>> questions/consultation?  
>
> Most happy to share what little I know!
>
> Bob
>
>
>
>
>
>

--
Camm Maguire        camm@...
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah


_______________________________________________
Gcl-devel mailing list
Gcl-devel@...
http://lists.gnu.org/mailman/listinfo/gcl-devel

[Camm Maguire-2]

Greetings!  One other goal I'd like to achieve this go around is
profiling without having to have a separate slower gprof image handy.
This is now doable as all function calls can optionally be redirected
via the fast link mechanism to a statistics collecting procedure.  (Of
course, this would leave out low lying system profiling, such as GBC,
etc.)  I'm planning on leaving the gprof option, but providing this
new facility as a perhaps more convenient alternative.

Schelter had a direct interface to PROFIL, which produces a flat list
of functions with the time spend in each.  We have the potential to
also keep track of a call tree as gprof does, without gprof.  What do
you think of the gprof output?  What sort of profiling information is
most useful?  Are items like compiled size, fixnum conversion,
consing, data side effect calls (e.g. rplaca) useful reportables to
the end user?

Take care,

Robert Boyer <boyer@...> writes:

>> Separately, does Bob Boyer have a few extra cycles for some email
>> questions/consultation?  
>
> Most happy to share what little I know!
>
> Bob
>
>
>
>
>
>

--
Camm Maguire        camm@...
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah


_______________________________________________
Gcl-devel mailing list
Gcl-devel@...
http://lists.gnu.org/mailman/listinfo/gcl-devel

[Matt Kaufmann]

Hi --

The most useful profiling info to me is just a call tree, showing how
long we're inside each function and, for each function, how long we're
inside of each call within that function.  I use the Allegro CL
profiler and that's what it gives me; it's been a long time since I
used GCL's, so I don't recall what I get from gprof.

Even though I don't use the other things you mentioned, perhaps I
would if I knew how.  But even then, probably just time is what I'll
care about.

Of course, those who write a lot of Common Lisp (as opposed to ACL2)
might have different needs/wants than I!  Anyhow, having a simple way
to profile in GCL as one gets with Allegro CL would be cool.

Thanks --
-- Matt
   Cc: hunt@..., kaufmann@..., gcl-devel@...,
           boyer@...
   From: Camm Maguire <camm@...>
   Date: Thu, 06 Aug 2009 12:17:14 -0400
   X-MagicMail-UUID: 9a7fb33a-82a4-11de-8a32-00188bf6df8c
   X-SpamAssassin-Status: No, hits=-2.6 required=5.0
   X-UTCS-Spam-Status: No, hits=-282 required=165

   Greetings!  One other goal I'd like to achieve this go around is
   profiling without having to have a separate slower gprof image handy.
   This is now doable as all function calls can optionally be redirected
   via the fast link mechanism to a statistics collecting procedure.  (Of
   course, this would leave out low lying system profiling, such as GBC,
   etc.)  I'm planning on leaving the gprof option, but providing this
   new facility as a perhaps more convenient alternative.

   Schelter had a direct interface to PROFIL, which produces a flat list
   of functions with the time spend in each.  We have the potential to
   also keep track of a call tree as gprof does, without gprof.  What do
   you think of the gprof output?  What sort of profiling information is
   most useful?  Are items like compiled size, fixnum conversion,
   consing, data side effect calls (e.g. rplaca) useful reportables to
   the end user?

   Take care,

   Robert Boyer <boyer@...> writes:

   >> Separately, does Bob Boyer have a few extra cycles for some email
   >> questions/consultation?  
   >
   > Most happy to share what little I know!
   >
   > Bob
   >
   >
   >
   >
   >
   >

   --
   Camm Maguire        camm@...
   ==========================================================================
   "The earth is but one country, and mankind its citizens."  --  Baha'u'llah


_______________________________________________
Gcl-devel mailing list
Gcl-devel@...
http://lists.gnu.org/mailman/listinfo/gcl-devel