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] function inlining when optimize for size (PR11830)


I have examined if it is possible to use the function inlining when
optimizing for size. I've found we can use it without any risk because
when the max-inline-insns-* parameters were set with a small value
the function inlining doesn't inline large functions.

I measuerd the size effect on CSiBE and this patch made about 1% code
size save (arm:1.212%, i386:1.024%, i686:1.029%, m68k:0.795%,
mips:1.096%, ppc:1.106%).

This looks very nice. Do you have any numbers for single/auto limits set to 10? That would allow inlining the functions that have exactly one function call in them possibly elliminating the need for wrapper, but not sure if it would pay back.

Yes, I have some numbers (code size save):


Single/Auto/RTL, arm-elf, i686-linux, ppc-elf
 0/ 0/ 0, 1.194%, 1.027%, 1.092%
 5/ 5/10, 1.212%, 1.029%, 1.106%
10/10/20, 1.054%, 0.921%, 0.964%

There is a significant difference between 5/5/10 and 10/10/20
combinations. I think when we optimizing for size the best combination
is the 5/5/10 for single/auto/rtl limits, am I correct?

Regards,
  Gábor Lóki




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