Re: [Sdcc-user] any better way- how to implement?

[Johan K. <joh...@id...>]

----- Original Message -----
From: Royce & Sharal Pereira <be...@im...>
To: <sdc...@li...>
Sent: Tuesday, July 31, 2001 10:08 AM
Subject: Fw: [Sdcc-user] any better way- how to implement?


> Hi, all,
>
> I optimised the inefficient code by using a peephole optimiser:-
> Make a text file say, peep.opt, & insert the following text
> rule:= replace
>         {
>             clr a
>             mov c,%1
>             rlc a
>             cpl a
>             mov  r2,a
>             clr c
>             jz  %2
>             cpl c
> %2:
>             mov  %1,c
>         }
>       by
>         {
>             cpl  %1 ;optm. by Author
>         }
> Save the file in the same dir as your sourcefile.
>
> Then compile, adding the command line option --peep-file peep.opt
> This worked for my application. Has any one tried this? Wonder if there
> could be any side effects...

There certainly are. Depending on how where what and why, "r2" could be
anything like r2-r7 and even a DSEG _sloc. Further more, "r2" could be the
result of a subexpression that will be used later, like in:

bit0 = ~bit1;
bit1= ~bit1;

You will gain a lot by declaring bit vars as volatile. I wonder if they
shouldn't be by default, it would be much more efficient than keeping them
registered. Also, shouldn't bitvar _sloc's be in BSEG instead of DSEG?
Sandeep?

Johan

Johan