Re: [Sdcc-user] missing possible optimizations?

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

> <test.asm snip>
> 188:	mov	r2,#0x00
> 191:	mov	r3,_main_response_1_1
> 193:	mov	a,r3
> 195:	orl	a,r2
> 196:	mov	_P_Data,a
>
> This part is a huge simplification of my macro WRITE_P_DATA.  With the
> given values of On_Mask and Inv_Mask, almost the entire expression can be
> reduced. I could do this by hand, but I would rather leave it, as I may
> need to change either the On_Mask or Inv_Mask later.  When the masks are
> any other values, the expression actually does a fair amount of work.  Yes,
> me being lazy now means more work for the compiler, but that's what
> computers are for, right?  So anyway, it looks like the compiler was
> (correctly) able to reduce the macro down to:
> 	P_Data = 0x00 | response[0];
> That it got that far is actually pretty impressive.  But, ORing with zero
> changes nothing.  So this could be reduced further to:
> 	P_Data = response[0];
> And in asm, this could be written with a single line:
> 	mov _P_Data,_main_response_1_1
I've commited a better optimization for these cases:
a & 0
a & 0xff
a | 0
a | 0xff
a ^ 0

Please don't complain, if your macro still isn't optimized to
> 	mov _P_Data,_main_response_1_1

There's a write access on P_Data, and P_Data is a SFR, and SFRs are volatile. 
Therefore the read access must not be optimized away. Finally your code 
exposes a bug in gen.c. I will have a look on this next time.

Bernhard