On 11/8/2010 1:07 AM, Francis Moreau wrote:
Ian Lance Taylor<iant@xxxxxxxxxx> writes:
I think I would say this slightly differently. The code is still
undefined according to the C/C++ language standards. When you use
-fno-strict-aliasing, gcc is promising to avoid using a class of
optimizations based on those portions of the language standards.
Another way to say this is that gcc is promising to not optimize based
on the fact that memory locations are accessed using different types.
That is, the code is still incorrect; gcc is just avoiding a class of
optimizations which are known to cause trouble with real programs.
I also am having trouble understanding what the OP is asking for here.
Clarification on GCC's strict aliasing rules.
But I think I got it now, although I still don't know which exactly
undefined behaviours are defined by GCC's -fno-strict-aliasing.
Ian just explained it well. -fno-strict-aliasing avoids code movement
optimizations which depend on compliance with -fstrict-aliasing. If you
mean you want incorrect code to give the same result as it does with
other compilers which accept it, on the same architecture, that's not
entirely the same thing, no one can recommend it, but I can't stop you
from trying.
--
Tim Prince
