[PicForth] Re: -! optimisation problem

[David McNab]

Samuel Tardieu wrote:
> Great! When you upload your new version on the wiki, I'll be happy to
> add it either in PicForth or as a test case, to make sure that there
> are no regressions.

new lib3, pure-picforth version, now uploaded to wiki

> When I wrote that the two operations were
> equivalent, I was not talking about the status bits whose behaviour
> are not covered by Forth :-)

This exposes a philosophical question.

The fact that status bits like z, c etc are readily settable/testable 
from PicForth might tempt some people to debate that assertion.

IMHO, a good position might be to document the impact of the various 
PicForth words on the status bits (wherever the impact is clear and 
deterministic).

As you can see from lib3, access to the status bits is an absolute 
necessity when dealing in number sizes greater than 8 bits.

Also, in general, access to Z and C from pure-picforth code can produce 
much more optimal code. For instance, incrementing a counter variable, 
and doing something whenever the count reaches zero.

The construct:

     1 somevar +!
     z bit-set? if
        do-something
     then

produces:

     0x0005  0AA2    incf    0x22,f
     0x0006  1903    btfsc   0x03,2
     0x0007  2804    goto    0x004   ; do-something

while:

     1 somevar +!
     somevar @ 0= if
        do-something
     then

produces:

     0x0009  0AA2    incf    0x22,f
     0x000A  0822    movf    0x22,w
     0x000B  1903    btfsc   0x03,2
     0x000C  2804    goto    0x004   ; do-something

which takes an extra mword.

(Hey, there might be scope to add an optimiser rule which uses incfsz 
etc where possible).

-- 
Cheers
David