Hi, I wonder how waves interact with caching.
Detailed description of the my question:
A CU scheduler can process 40 * 64 = 2560 threads at a time. Each CU has a L1 cache.
Question 1: Are the 2560 threads sharing the L1 cache in the CU?
Question 2: does a wave access L2 cache only if L1 misses?
Question 3: Thinking from wavefront point of view, when will a cache line(L1 and L2) be invalidated ?