https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124795

--- Comment #8 from Jan Hubicka <hubicka at ucw dot cz> ---
So there is 
typedef struct instr_time
{
 int64 ticks;
} instr_time;

which is done to prevent people from misusing int64.
Indeed inline heuristics does not anticipate SRA of return values (it does some
limited guessing of SRA for function parameters).  How do you propose to
account for it?

Problem with SRA prediction from inliner usually is that it only happens when
multiple functions are inlined...

ipa-sra could promote this to scalar which may be useful elsewhere too, since
aggregate passing conventions may
be worse than scalar.

Reply via email to