This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch,avr] PR70677: Use -fno-caller-saves for avr


Denis Chertykov writes:

> 2016-08-01 15:17 GMT+03:00 Georg-Johann Lay <avr@gjlay.de>:
>> Problem with -fcaller-saves is that there are situations where it triggers
>> an expensive frame just to store a variable around a function call even
>> though there are plenty of call-saved registers.
>>
>> Example:
>>
>> typedef __UINT8_TYPE__ uint8_t;
>>
>> extern uint8_t uart0_getc (void);
>>
>> void foo (uint8_t *buffer, uint8_t cnt)
>> {
>>   while (--cnt)
>>     {
>>       *buffer++ = uart0_getc();
>>     }
>> }
>>
>> $ avr-gcc -Os -S -dp -mmcu=atmega8 loop-buf.c
>>
>> $ avr-gcc gcc -B$TV -Os -c -save-temps -dp -mmcu=atmega8 loop-buf.c &&
>> avr-size loop-buf.o
>>    text    data     bss     dec     hex filename
>>      50       0       0      50      32 loop-buf.o
>>
>> $ avr-gcc -Os -c -save-temps -dp -mmcu=atmega8 loop-buf.c -fno-caller-saves
>> && avr-size loop-buf.o
>>    text    data     bss     dec     hex filename
>>      32       0       0      32      20 loop-buf.o
>>
>> I actually came never across a situation where -fcaller-saves improved the
>> code performance, hence this patch proposes to switch off -fcaller-saved per
>> default.

Like you mentioned in the bug report, would fixing the costs be a better
way to fix this rather than a blanket disabling of the option?

Regards
Senthil


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]