Michael Schmitt wrote:
>What do you expect to get when the compiler optimizer inlines the call?
A fair question. If it's a compile-time thing, I'd say it would be an option as
to whether such non-calls "count" or not. Or maybe you compile without inlining
if it's enabled. I'd expect there to at least be a way to make it pretend it
called the function. This does make me lean slightly toward compile-time, since
otherwise it's inserting code that it will be skipping around in normal
operation. Though I suppose it could be something as simple as the following,
which is sort of assembler pseudo-code:
TM someflag,set Tracing calls?
BNO around
BAS R14,traceit Yes, I'm playing fast and loose with R14 here
DC A(C'THINGBEINGTRACED') And yes, I realize this line doesn't
work
around DS 0H
That bogus DC is a pointer to a constant with the function name. So the traceit
function knows that that such a constant will be at 4 past R14, and to return
to the instruction after that. This way you're not generating a DC for every
call to the same function in a module. This would add a TM/BNO for every call
in normal operation; on modern hardware, that seems like it would be
insignificant. But it's probably still a compile-time option whether even THAT
gets put in, since otherwise a new compiler that adds this capability will
suddenly generate larger modules for the same code (16 bytes per function call,
if I'm counting it right, which could add up).
The point, of course, is for debugging, so you (might) want to know even about
those inlined calls. If it's a trivial function (strlen or something), maybe
not. Also, if it's a compile-time thing, you'd want to be able to select "all"
or "not stdlib stuff". These are things the compiler knows, yes? (Well, I guess
if you replaced strlen with your own function, the compiler might not know, but
that's your problem!)
This is fun...
Sorta related, I was just reading a novel in which a "hacker" has decompiled
some code and can tell the programmer was highly skilled. Clearly the author
knows nothing of compiler optimization, since even a HELLO WORLD program looks
like something complex when it's optimized!
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN
