David Brown <address@hidden> wrote:
You don't need to convince me - I've seen plenty of horror stories
of this type, both myself and with my customers (although not with
IAR - I have never bought a compiler from them).
Even though we got IAR for free at work, well, free of charge at least
-- I never use it. It's a pain in the bottom to convince that
compiler that there's a valid license, and it's virtually impossible
to convince it there's a valid network license when the compiler run
within a VMware machine. After spending many hours with IAR support
about that, sending them detailed reports about all the history track
record back to at least year 1900, they eventually offered me that
running it within a virtual machine is not supported, and won't be.
As I, in turn, don't want to sit in front of a Windows machine, and
IAR doesn't want to support anything else but Windows -- there's now
zero motivation for me to even use it at all. That's even though I'm
convinced they are better in some details than GCC, at least for the
AVR target.
Some c.a.e. regulars seem to believe that commercial compilers
regularly produce smaller and tighter code than gcc because they are
x years ahead in terms of optimisations.
It was kinda funny. Occasionally, some people complain about certain
GCC optimizations being too aggressive, in particular in the
German-language mikrocontroller.net forums. A couple of times, I've
got curious enough to find an instance of IAR I could really get
access to, and compare the results from both compilers for the same
code. Guess what? In all analyzed cases, IAR performed almost the
same kind of optimizations as those people complained about on GCC. I
thought this to be really impressive: both compilers have a completely
different code base, grew out of a completely different history, yet
produce similar optimization techniques, almost down to the code
sequences involved. (IAR and GCC use different ABIs on AVR, so of
course, different actual CPU registers are used.)