tree c3eee7b77aa7f5afb3f1707b852b96906f7273b6
parent fd1dee7bd7264287cbbc7de706e70e6fab0077e7
author Frank Barchard <fbarchard@google.com> 1614381225 -0800
committer XNNPACK Team <xnnpack-github-robot@google.com> 1614381271 -0800

QS8 LD64 C4 dot product GEMM microkernel reduced remainder handling

Main loop handles remainder of 5 to 7, so if there is a remainder it is 1 to 4.
The A pointers only advance by 8 in the mainloop.
The old remainder register is replaced with a rewind register (kc + 3) & ~7

PiperOrigin-RevId: 359848654