[avr-gcc-list] suggestion

[igor_karpov]

I'm a total newbie to avr-gcc and just compiled a first program. It merely
fits the program memory so I've started to optimize..

In my code there lots of places where 1-2-3 bits are flipped in I/O port:


#define set_line_in_lcd_control_port(pin) sbi(PORTC, (pin))
#define set_lines_in_lcd_control_port(mask) PORTC |= (mask)
..
set_lines_in_lcd_control_port (_BV(LCD_CS1) | _BV(LCD_CS2));
..
set_line_in_lcd_control_port(LCD_E);

so, when 1 bit is toggled set_line_in_lcd_control_port is used, if more
- bits are set using a mask. This second #define compiles into

set_lines_in_lcd_control_port (_BV(LCD_CS1) | _BV(LCD_CS2));
2d40: 85 b3       in r24, 0x15 ; 21
2d42: 88 61       ori r24, 0x18 ; 24
2d44: 85 bb       out 0x15, r24 ; 21


whereas first into

set_line_in_lcd_control_port(LCD_E);
2d36: af 9a       sbi 0x15, 7 ; 21


that is, just one word to toggle one bit and 3 words to toggle 2+ bits.

My suggestion is to optimize the case when 2 bits are set into 2 sbi
commands, and when 2 bits are cleared into 2 cbi commands. It's quite
easy to count number of 1s or 0s in the mask - and if there are exacly
2, apply this optimization.

I was able to save ~100 bytes and fit my program into 4K, just by optimizing
these often used bit-tweaking places in my program.

If someone would have time to implement this kind of optimization for
I/O ports, that would be really great! I've never dealt with compilers
and just don't know how to do that..