Clarification on MMA0 Results Handling in the Latest Code

[ziyuhuang123]

I noticed that the latest code includes Mma1_is_RS. Does this imply that the results of MMA0 might be stored in shared memory (smem) and then loaded back into registers for mma1? I remember that in previous versions, the results were directly reused in registers.

[tridao]

If Mma1_is_RS==false then the output of MMA0 is stored to smem and used directly by Mma1 without loading to registers. It's just an option to reduce register usage.