P2 Taqoz V2.8: Fast Local Variables in Cog Memory

Christof Eb. · 2023-04-28 07:26

Hi,
perhaps this is inspiring or useful to somebody. It shows a way to get fast local variables and also it shows a way to have several fast tiny assembler routines in cog memory.

As stated often before, for me it is very nice for readability of code, if I have local variables. These days I read: " Too much DUP SWAP DROP will ROT your brain. " in the docu for JFORTH for Amiga.

Of course it is best, when you can have speaking variable names. A way to do that can be found here: https://forums.parallax.com/discussion/174970/p2-taqoz-v2-8-lutlongs-value-type-variables-locals-now-in-lut-and-faster#latest

In the past I had learned, that it is essential to reduce HUB access, if speed is important. Faster than LUT is COG RAM. Fastest code is COG code, which can be called very fast in Taqoz, because in this case the assembler start address is the wordcode itself. So the goal was to have only one HUB access, which is code and address together. To achieve this, there must be separate a word to store and a word to read for each local variable. In the past it turned out, that 5 local variables are enough. These are located as registers at the upper end of cog ram, reducing space for COGMOD. Additionally the code block must be loaded into cog ram, before using the code!
To allow for nested words, the local variables can be saved "cLoc" and retrieved "cEndloc" on auxiliary L-Stack as a frame of 5 items.

The result is not too bad: The fibonacci result is three times faster compared to using global variables or named value type locals. (At the same time it is three times slower than doing all on stack.)

It uses absolute addresses in cog, so they are fix here for the version! So this will only work with the version _BOOT_P2.BIX in Taqoz.zip in https://sourceforge.net/projects/tachyon-forth/files/TAQOZ/binaries/ .

Have fun Christof

{   
   CogModLocalsB.fth

   27.04.23 Christof Eberspaecher 

}

!polls \ clear multitasking

IFDEF *Tests*   
    oldorgT org
    FORGET *Tests*   }

pub *Tests*     PRINT" P2 Test" ;

0 bytes oldorgT

: CogModLocals_Library ; \ { ###########################################

\ new cog registers for locals 
489 := acm
488 := bcm
487 := ccm
486 := dcm
485 := ecm

code cLoc ' \ save the locals from calling word to L-Stack
      wrlut ecm,auxptr
      add auxptr,#1
      wrlut dcm,auxptr
      add auxptr,#1
      wrlut ccm,auxptr
      add auxptr,#1
      wrlut bcm,auxptr
      add auxptr,#1
      wrlut acm,auxptr
   _ret_ add auxptr,#1
end

code cEndloc ' \ restore the ones from calling from L-Stack
      sub auxptr,#1 
      rdlut acm,auxptr
      sub auxptr,#1
      rdlut bcm,auxptr      
      sub auxptr,#1
      rdlut ccm,auxptr
      sub auxptr,#1
      rdlut dcm,auxptr
      sub auxptr,#1
   _ret_ rdlut ecm,auxptr
end

$086 := PUSHX

code CogModLocals ( cogmod for fast access to locals )
   mov acm,a ' \ >a
   jmp #\@DROP
   mov xx,acm ' \ a>
   jmp #\PUSHX
   mov bcm,a ' \ >b
   jmp #\@DROP
   mov xx,bcm ' \ b>
   jmp #\PUSHX   
   mov ccm,a ' \ >c
   jmp #\@DROP
   mov xx,ccm ' \ c>
   jmp #\PUSHX 
   mov dcm,a ' \ >d
   jmp #\@DROP
   mov xx,dcm ' \ d>
   jmp #\PUSHX   
   mov ecm,a ' \ >e
   jmp #\@DROP
   mov xx,ecm ' \ e>
   jmp #\PUSHX   
end

create$ >a 0 ' COGMOD + @words CPA W!
create$ a> 2 ' COGMOD + @words CPA W!
create$ >b 4 ' COGMOD + @words CPA W!
create$ b> 6 ' COGMOD + @words CPA W!
create$ >c 8 ' COGMOD + @words CPA W!
create$ c> 10 ' COGMOD + @words CPA W!
create$ >d 12 ' COGMOD + @words CPA W!
create$ d> 14 ' COGMOD + @words CPA W!
create$ >e 16 ' COGMOD + @words CPA W!
create$ e> 18 ' COGMOD + @words CPA W!

AT CogModLocals 2+ 20 LOADMOD 2drop \ bug in LOADMOD ==> 2drop

: fiboCogModLocals cLoc ( n -- f ) \ forth words for locals as inline code in cog
   AT CogModLocals 2+ 20 LOADMOD 2drop \ bug in LOADMOD ==> 2drop
    0 >a   1 >b   0 >c
    ( n ) for
        b> >a
        c> >b
        a> c> + >c
    next
    c> 
cEndloc ;

: fiboTest cLoc
   12456 >a \ just a test
   46 lap 
   fiboCogModLocals lap . .lap
   crlf a> . \ just a test
cEndloc ;

\ 46 lap fiboCogModLocals lap . .lap
\ }