https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119719
duane ellis <duane at duaneellis dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |duane at duaneellis dot com
--- Comment #4 from duane ellis <duane at duaneellis dot com> ---
(In reply to Andrew Pinski from comment #1)
> So this sounds like what autofdo does. I suspect you could hook up a similar
> method for any profiling like program to autofdo. Currently the autofdo uses
> perf (the Linux kernel userland tool) output to do it. but it is just a
> bunch of scripts which creates the autofdo output to the PC.
yea autofdo is too fat. one chip has 64k total ram about 15-16k free the app
is using about 45k now for self and there is a minimal amount of stack space
for each tasks run time stack.(we count bytes of stack space and adjust each
tasks stack to +128 bytes of margin)
because there is so little space left there is no usable malloc so anything c++
is a big fat memory hog and needs new/delete
resource constrained is hard,
this is not what i am using but it is very similar
https://www.st.com/en/microcontrollers-microprocessors/stm32f107vc.html
depending on the chip variant it has more ram but ours does not and i cannot
change the chip on the board i have to use the designed in chip
resource constrained is non trivial
