[Bug target/55377] ipa-pure-cont produces wrong code on AVR

["pinskia at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55377

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2012-11-19 04:36:36 UTC ---
This looks correct as the only thing as:
  while (!serial.canWrite()) {}

cannot change its state once in that loop as far as I can tell.  There are no
volatile variables read either.

canWrite is inlined which was:
 bool canWrite() {
  return !(_txPointers.full(sizeof(TXBuffer)));
 }

And we decided mhvlib::RingBuffer::full only reads from its arguments which are
not volatile.

full does:
 bool full(uint8_t blockLength) {
  return length() > (_size - 1 - blockLength);
 }

Where length was:

 uint8_t length() {
  int16_t length = _head - _tail;
  if (length < 0) {

   length = (_size - _tail) + _head + 1;
  }

  return (uint8_t) length;
 }

And both _head and _tail are not marked as volatile so we only need to read
them once (including through the loop).

So the code that ipa-pure-const is producing is correct.
The reason why you can't produce a simpler testcase is because it depends on
other optimization chooses that the compiler has decided to take.  

So in conclusion, I think you are missing some variables marked as volatile in
the RingBuffer implementation and/or Device_TXImplementation implementation.