- Bug: Among the list of AVX512 SGEMMSUP RD kernels, the ones handling
m_fringe = 3 had incorrect usage of ZMM on a vector-load instruction
that strictly needed YMMs.
- Further updated the existing micro-kernel test cases to simulate
these issues and validate the fix.
AMD-Internal: [CPUPL-5353]
Change-Id: Id86e60ce36bb9f8433a1a203cfe0b8c6347df2c1