if I have a situation in which different thread would access he same memory location to read and to write (possible race condition I mean), if I use the texture memory and so each thread read data from the texture and write in global memory, race conditions will be avoided?
No. The only effect of using the texture in this case is that you might sometimes get stale data from the cache.
Hi Tera, sorry but what do you mean?
Moreover, I also think that I can speed up my operation using texture in this way, isn’t it?
Textures may only be used for data that is strictly read-only within a kernel.
Yes, textures would likely be faster, but they do not eliminate race conditions.