Richard Henderson wrote:
> On 08/03/2011 07:07 AM, Georg-Johann Lay wrote:
>> #include <stdio.h>
>>
>> void foo ()
>> {
>> printf ("%d %d %d", 1, 2, 3);
>> printf ("%d %d %d", 3, 4, 5);
>> printf ("%d %d %d", 1, 4, 5);
>> }
>>
>> Attached the output: The compiler happily pushes onto the stack
>> but pops only at the end of the function. So in a function with
>> many such calls that would eat up great deal of RAM. It that
>> what we want?
>
> Add more printfs and find out. We'll consume 32 bytes and then
> pop it all off in the middle of the function, then start again.
>
> See the use of pending_stack_adjust in expand_call.
>
>
> r~
>
Yes, the following blocks are sprinkled all over the function:
in r24,__SP_L__ ; 222 *movhi_sp/2 [length = 2]
in r25,__SP_H__
adiw r24,32 ; 134 *addhi3/2 [length = 1]
in __tmp_reg__,__SREG__ ; 223 *movhi_sp/1 [length = 5]
cli
out __SP_H__,r25
out __SREG__,__tmp_reg__
out __SP_L__,r24
With ACCUMULATE_OUTGOING_ARGS it looks much better: there is just
one such block in the prologue/epilogue.
I think ACCUMULATE_OUTGOING_ARGS would be a win definitely.
Pushing more than 63 bytes for one function is no real world
code for AVR and I think we can focus on functions receiving
<= 63 bytes on stack.
Johann
