trunk/640f90e15bc0e9af5caaa054939015d14a086131: flex_gemm: compose paired fragment reductions (#192894)
- PyTorch: 2301 events in the last 90 days
- PyTorch: 2271th Release in the last 90 days
- Previous: earlier the same day · trunk/10dc63912abee5b1c6b942e670872e515cbf9210: [CPP FakeTensor] Don't look up the op in register_fake (#200798)
What happened
Human Note At a high level we are now allowing chunking + reductions along the n axis. So this is two forms of reductions; 1 is the chunked or interleave pairing and the other is the reduction along the N axis follwing this up. Below we have 2 for the first and then 32 for the second or a total of 64 elements. We needed so mechanize to enforce we have 64 acc values in regsisters and we do the reductions only in regi…
Summary assembled by rule from the sources below