Path: nntp.eternal-september.org!news.eternal-september.org!eternal-september.org!.POSTED!not-for-mail
From: Stephen Pelc <stephen@vfxforth.com>
Newsgroups: comp.lang.forth
Subject: Re: OOS approach revisited
Date: Sat, 28 Jun 2025 09:37:33 -0000 (UTC)
Organization: A noiseless patient Spider
Lines: 65
Message-ID: <103od4s$pis9$1@dont-email.me>
References: <fdbeb48de8ceb748d44c67dc3981a566@www.novabbs.com> <bc63996456fe967e5c66d17cbbeb21c2@www.novabbs.com> <nnd$39000884$25328e8c@2d18b2c67eb1f6e0> <mc8dkkFeh4uU1@mid.individual.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8; format=fixed
Content-Transfer-Encoding: 8bit
Injection-Date: Sat, 28 Jun 2025 11:37:33 +0200 (CEST)
Injection-Info: dont-email.me; posting-host="2ba889cbcec782843888680410bc9a37";
	logging-data="838537"; mail-complaints-to="abuse@eternal-september.org";	posting-account="U2FsdGVkX18u0YqRriFg+Q2X0tda+BXD"
User-Agent: Usenapp for MacOS
Cancel-Lock: sha1:284+Dn2vp/92KFlOCD/yOWXpTuM=
X-Usenapp: v1.27.4/l - Full License

On 27 Jun 2025 at 22:35:32 CEST, "minforth" <minforth@gmx.net> wrote:

> Am 27.06.2025 um 20:15 schrieb albert@spenarnc.xs4all.nl:
>> In article <bc63996456fe967e5c66d17cbbeb21c2@www.novabbs.com>,
>> LIT <zbigniew2011@gmail.com> wrote:
>>>> It really depends on how counted loops are implemented.
>>>> Most CPUs have operators for register-based count-down loops
>>>> that are blazingly fast.
>>>> 
>>>> If they can be used within Forth-based loop constructs
>>>> I would expect a greater speed increase than what you measured.
>>> 
>>> In that old fig-Forth it's rather short and simple:
>>> 
>>>         sqHeader '(LOOP)'
>>> XLOOP   dw   $ + 2
>>>         mov  BX,1
>>> XLOO1:  add  [BP],BX
>>>         mov  AX,[BP]
>>>         sub  AX,[BP+2]
>>>         xor  AX,BX
>>>         js   BRAN1
>>>         add  BP,4
>>>         inc  SI
>>>         inc  SI
>>>         jmp  NEXT
>>> 
>>> It doesn't look that bad. Can it be
>>> done even shorter?
>> 
>> My optimiser looks into the combination of DO and LOOP,
>> transfers the returns stack into registers after inlining
>> everything. It is near vfx performance.
>> All experimental, but yes there is much to be gained.
> 
> Must be tricky to do UNLOOP in a register-based loop.  ;-)

Here are the code generators for VFX x64 LOOP and UNLOOP.
All the complexity is in the DO and ?DO code.

: c_loop        \ mrk> drbid -- ; compile code for LOOP ; SFP094
  c_shuffle reset-opt            \ SFP097
  a[ INC    r14 ]a  use-a        \ update index
  a[ INC    r15 ]a  use-a        \ update limit-index-$8000.0000
  a[ JNO ]a  <ares  use-a        \ resolve backward branch
  c_unloop                              \ remove DO ... LOOP state
  >RES                                  \ resolve forward branch
;

: c_unloop      \ -- ; compile code for UNLOOP
  c_shuffle  reset-opt
  a[  pop    r14        \ restore old index
      pop    r15        \ restore old index-limit-xorbit63
      pop    rax        \ discarded
  ]a  use-a
;

Stephen
-- 
Stephen Pelc, stephen@vfxforth.com
Wodni & Pelc GmbH
Vienna, Austria
Tel: +44 (0)7803 903612, +34 649 662 974
http://www.vfxforth.com/downloads/VfxCommunity/
  free VFX Forth downloads