I have started recently to learn a bit about Lisp and would like to ask for
one question:
As a compiler I just like a good compiler, but as an evaluator I would like
to know whether there is any significative improvement in performance
(let us say more than a 10%) when recompiling the compiler from its
sources (I use currently both an AMD Athlon and a 486) instead of using
the provided binary image from www.cons.org.
The reason is just I am very happy when I only use float-point operations
(it approaches C performance and the cycle type-compile-debug would be
clearly better for me after three or four months of seriously programming
in Lisp) but with integer ones (unsigned-byte 32), I am not so happy.
Anyway, I use much more floating-point arithmetics than integer one.
That's all.
······@no.al.Xspan.es (Juan Pablo Hierro Álvarez) writes:
Note that CMU CL specific questions are likely to be read and answered
by more folks on the CMU CL mailing lists (see www.cons.org for the
addresses).
> As a compiler I just like a good compiler, but as an evaluator I would like
> to know whether there is any significative improvement in performance
> (let us say more than a 10%) when recompiling the compiler from its
> sources (I use currently both an AMD Athlon and a 486) instead of using
> the provided binary image from www.cons.org.
While there are special backend feature flags for :pentium, :ppro and
even :athlon, only a handful of floating point VOPs dependend on those
features, so that recompiling your implementation will not really gain
you much.
> The reason is just I am very happy when I only use float-point operations
> (it approaches C performance and the cycle type-compile-debug would be
> clearly better for me after three or four months of seriously programming
> in Lisp) but with integer ones (unsigned-byte 32), I am not so happy.
If there are specific instances where you deem integer performance to
be suboptimal, please take it up on the mailing list, and we might be
able to improve on this performance level.
Note that you can write new compiler transformations and VOPs for the
compiler without recompiling the compiler.
Regs, Pierre.
--
Pierre R. Mai <····@acm.org> http://www.pmsf.de/pmai/
The most likely way for the world to be destroyed, most experts agree,
is by accident. That's where we come in; we're computer professionals.
We cause accidents. -- Nathaniel Borenstein