Paolo Bonzini writes: > On 24/08/2016 22:44, Pranith Kumar wrote: >> Use the __atomic_*_n() primitives which take the value as argument. It >> is not necessary to store the value locally before calling the >> primitive, hence saving us a stack store and load. > > If you do this, you might as well do it for __atomic_load and > __atomic_compare_exchange. However, you'd still need typeof_strip_qual > for __atomic_compare_exchange's "_old" local variable.
OK, I will update the patch doing the same for load and compare exchange. > > The question is, does this actually cause a change in generated code? I > would be very surprised if it did, but then it's perfectly possible to > have a bug in GCC. It looks like it does. I tested atomic_load() with previous and the atomic_load_n() version and it generates the following code: current: mov 0x200c4a(%rip),%eax # 601030 <value> mov %eax,-0x4(%rsp) mov -0x4(%rsp),%eax atomic_load_n(): mov 0x200c4a(%rip),%eax # 601030 <value> So looks like it's worth it. Thanks, -- Pranith
