Can several CPU threads write to constant memory in CUDA 4 ?
If several CPU threads are writing to constant memory, will they override each other memory ?
depends if those threads are writing to the same context or not
Let me see if I understand correctly -
in CUDA 4.0 if you run several CPU threads using C-for-CUDA all running the same CPU code,
which writes some CPU-thread-dependent values to the same const variable in the host, there will be problems, right ?
If right, is there a workaround for that ? How can you make the device regard each such write to the constant memory as different and use a different portion
of the constant memory per CPU thread (as done for different constant variables) ?
Along the same lines - what is the behavior of a similar case, with texture memory ? same problems ?
(Lot’s of questions because the programmer’s guide provide enough documentation about this).
Thanks for your comment. Could you please elaborate ? There is not much documentation regarding these issue…