Re: [Sdcc-user] missing possible optimizations?

[Bernhard H. <ber...@be...>]

> So I understand that the mov command given above would not work, since it
> never performs the read on the SFR.  In my case I don't need the read, so
> to avoid it I'll have to change the way my macro works.  For the general
> case though, wouldn't the following be correct?
>
> mov a,_P_Data                   ; Read the SFR into A
>                                 ; (or any register, if we can't trash A)
> mov _P_Data,_main_response_1_1  ; Directly write the optimized result
SDCC emits now exactly this code :-)

There was a new iCode opcode DUMMY_READ_VOLATILE necessary. Code is generated 
with the mcs51 port and (may be buggy) with the ds390/400 port. For all other 
ports I need the help from the port maintainers (I speak neither z80 nor 
pic*).

volatile-ness of operands are quite often not considered in SDCC :-(
But with the help of the new op it should be easy to fix bug #663539.

I've written a regression-test too. But it triggers one or two further bugs 
(could be connected with other reported problems), which I'm going to hunt 
now.

Bernhard

P.S.: Authentication is broken again on the compile farm. But they have added 
a 4 CPU opteron server with SuSe linux :-)) I started a build, but it didn't 
finish --- authentication was broken. I couldn't login again. Weired :-(