[Bug c/59448] Code generation doesn't respect C11 address-dependency

[joseph at codesourcery dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59448

--- Comment #3 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
The code generated appears fully in accordance with the semantics of C11.  
You refer to 5.1.2.4#14, the definition of "carries a dependency".  The 
term "carries a dependency" is used outside its own definition only in the 
definition of "dependency-ordered before".  This testcase contains no 
instances of "dependency-ordered before", because any such instance must 
directly or indirectly involve a case of "A performs a release operation 
on an atomic object M, and, in another thread, B performs a consume 
operation on M and reads a value written by any side effect in the release 
sequence headed by A", and the test involves no atomic objects.

"dependency-ordered before", in turn, is only used in the definition of 
"inter-thread happens before", which is only used in the definition of 
"happens before".

Please provide a complete testcase, using _Atomic or <stdatomic.h> (and so 
using GCC 4.9, of course) that demonstrates any bug: that is, that does 
not contain a data race according to the C11 definition, but where GCC has 
reordered code in a way that introduces one (and so one can be 
demonstrated by enough iterations of the threads in the testcase) - or 
where the code generated fails some other semantics associated with 
"happens before", such as those for "visible side effect" and "visible 
sequence of side effects".  I expect any such bug, and fix, would probably 
not be a front-end issue but an issue with the atomic built-in functions 
failing to ensure appropriate ordering in the presence of dependencies 
involving atomic operations (dependencies not involving such operations 
having no effects on C11 semantics).