thanks for the link, indeed
"
For the following intrinsics, <type> is either a 32-bit or 64-bit integer.
Atomic Fetch-and-op Operations
<type> __sync_fetch_and_add(<type> *ptr,<type> val)
...
"
clearly states that the types are signed and the same for the
input/output values. So GCC implementation is probably bogus but as you
say fixing it might imply legacy problems.
I logged the issue and the patch with
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41639
Thanks,
-c
Ian Lance Taylor wrote:
Christian BRUEL <christian.bruel@xxxxxx> writes:
Can I assume from the gcc sources that the __sync_fetch_ builtins take
only unsigned value as inputs, and thus also return unsigned values.
These builtins came from Intel. This page
http://software.intel.com/sites/products/documentation/hpc/compilerpro/en-us/cpp/lin/compiler_c/intref_cls/common/intref_itanium_synchro_prim.htm
suggests but does not clearly state that the return type should be the
same as the value type. We could probably implement that in gcc, but
I think it would be awkward to simply change the sign of the return
type now; it might break currently working programs.
This is probably worth a bug report in any case.
Ian
