Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.forth > #135504

Re: First result in riscv optimiser

Show all headers | View raw

albert@spenarnc.xs4all.nl writes:
>The ciforth works in reverse of the usual optimisers. Peephole is the
>last step.

That's exactly like usual optimizers.  Peephole optimization tends to
be used to catch some things that other optimizations have missed,
typically across the boundaries of the other optimizations.

>Now I have completed the second step: if code works on constant data,
>execute it at compile time, if possible.
>An example:
>albert@sinas2:~/PROJECT/optim$ optimiser
>
>: nonsense DUP 'DROP EXECUTE ;
> OK
> : test 2 4 * nonsense 13 + ;
>  OK
>  'test inline&fold
>  2
>  LIT
>  *
>  DUP
>  LIT
>  EXECUTE
>  nonsense
>  LIT
>  +
>  test
>   OK
>   SEE test
>
>: test
>0000,0000,0000,0015
>;

Gforth does not do inlining by itself yet, but it does have a literal
stack for constant folding and more.  See
<2019Aug5.121829@mips.complang.tuwien.ac.at>.  Let's take your
example, but using explicit inlining for NONSENSE:

inline: nonsense dup `drop execute ;inline
: test 2 4 * nonsense 13 + ;
see test
\ output follows:
: test  
  #21 ; ok

Gforth will do that on RISC-V as well as on other architectures,
because these things happen at the threaded-code level.

I don't think that these contrived examples prove much, and in general
I don't think that full constant folding will trigger often, but
partial constant folding (e.g., optimizing "5 -" into lit+ with the
immediate argument -5) is probably relatively frequent, and
occasionally, constant folding works in cases where a much
heavier-weight optimization would otherwise be needed.  E.g.,

5e fvalue x
synonym y x
: foo to y ;

FOO is compiled to

: foo  
  <x> f! ;

where <X> is the body address of X.  In the process of this
compilation, several levels of partial constant folding are involved.
There may be other ways to achieve this kind of compilation from this
code, but I think they would be substantially more complicated.  IIRC
our EuroForth 2019 paper on the new Gforth header gives an example of
how that works.

- anton
-- 
M. Anton Ertl  http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
     New standard: https://forth-standard.org/
EuroForth 2026 CFP: http://www.euroforth.org/ef26/cfp.html

Back to comp.lang.forth | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

First result in riscv optimiser albert@spenarnc.xs4all.nl - 2026-09-01 10:49 +0200
  Re: First result in riscv optimiser anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2026-09-01 15:33 +0000
    Re: First result in riscv optimiser albert@spenarnc.xs4all.nl - 2026-09-01 18:54 +0200
    Re: First result in riscv optimiser albert@spenarnc.xs4all.nl - 2026-09-02 11:18 +0200
    Re: First result in riscv optimiser Kragen Javier Sitaker <kragen@canonical.org> - 2026-09-04 13:14 -0300
      Re: First result in riscv optimiser albert@spenarnc.xs4all.nl - 2026-09-05 13:06 +0200
        Re: First result in riscv optimiser Kragen Javier Sitaker <kragen@canonical.org> - 2026-09-08 02:21 -0300
          Re: First result in riscv optimiser albert@spenarnc.xs4all.nl - 2026-09-08 12:38 +0200
            Re: First result in riscv optimiser Kragen Javier Sitaker <kragen@canonical.org> - 2026-09-11 07:50 -0300
              Re: First result in riscv optimiser Paul Rubin <no.email@nospam.invalid> - 2026-09-11 17:09 -0700
              Re: First result in riscv optimiser albert@spenarnc.xs4all.nl - 2026-09-13 14:03 +0200
            Re: First result in riscv optimiser antispam@fricas.org (Waldek Hebisch) - 2026-09-13 02:53 +0000

csiph-web