[avr-gcc-list] Question about code size

[Dave Hylands]

Hi,

I've tried building a bootloader I have using 4.1 and 4.2 and the code
size is about 50% larger than what I get from 3.4.4 (or 3.4.5). Now, I
got the sources for these toolchains from different sources, so its
conceivable that one has a patch that other one doesn't, but the
difference seems rather extreme.

I've put all of my file over here:
http://www.davehylands.com/Misc/bootloader/

The b34.sh and b42.sh are the scripts I used to build the two

An excerpt of  a typical function is:

void nothing_response(void)
{
   if (getch() == ' ')
   {
       putch(0x14);
       putch(0x10);
   }
}

Under 3.4.4, using -Os -mshort-calls (on boot.c)

00003c72 <nothing_response>:
   3c72:        f7 df           rcall   .-18            ; 0x3c62 <getch>
   3c74:        80 32           cpi     r24, 0x20       ; 32
   3c76:        21 f4           brne    .+8             ; 0x3c80 
<nothing_response+0xe>
   3c78:        84 e1           ldi     r24, 0x14       ; 20
   3c7a:        da df           rcall   .-76            ; 0x3c30 <putch>
   3c7c:        80 e1           ldi     r24, 0x10       ; 16
   3c7e:        d8 df           rcall   .-80            ; 0x3c30 <putch>
   3c80:        08 95           ret

Under 4.2.0, using the same options, I get:

00003ca2 <nothing_response>:
   3ca2:        80 91 c0 00     lds     r24, 0x00C0
   3ca6:        87 ff           sbrs    r24, 7
   3ca8:        fc cf           rjmp    .-8             ; 0x3ca2 
<nothing_response>
   3caa:        80 91 c6 00     lds     r24, 0x00C6
   3cae:        80 32           cpi     r24, 0x20       ; 32
   3cb0:        71 f4           brne    .+28            ; 0x3cce 
<nothing_response+0x2c>
   3cb2:        80 91 c0 00     lds     r24, 0x00C0
   3cb6:        85 ff           sbrs    r24, 5
   3cb8:        fc cf           rjmp    .-8             ; 0x3cb2 
<nothing_response+0x10>
   3cba:        84 e1           ldi     r24, 0x14       ; 20
   3cbc:        80 93 c6 00     sts     0x00C6, r24
   3cc0:        80 91 c0 00     lds     r24, 0x00C0
   3cc4:        85 ff           sbrs    r24, 5
   3cc6:        fc cf           rjmp    .-8             ; 0x3cc0 
<nothing_response+0x1e>
   3cc8:        80 e1           ldi     r24, 0x10       ; 16
   3cca:        80 93 c6 00     sts     0x00C6, r24
   3cce:        08 95           ret

Is there a different set of options I can use with 4.x to get smaller code size?

It seems to be inlining the putch function which is called many times,
and the inlined version is quite a bit bigger than calling the
function.

--
Dave Hylands
Vancouver, BC, Canada
http://www.DaveHylands.com/