Support system-wide and block-wide atomics - Githubissues

eyalroz / cuda-kat

CUDA kernel author's tools

BSD 3-Clause "New" or "Revised" License

104 stars 8 forks source link

Support system-wide and block-wide atomics #55

Open eyalroz opened 4 years ago

eyalroz commented 4 years ago

Beginning with CUDA 10 (or maybe 9?) we have three kinds of atomics:

atomicFoo() - atomic w.r.t. other memory access from within the same GPU.
atomicFoo_system() - atomic w.r.t. memory access from any GPU and from the host, on the same system.
atomicFoo_block() - atomic w.r.t. memory accesses from threads in the same thread block only.

We currently support only the first kind, but should support the other two.