Following preliminary investigation and tuning with the auto-tuner, these are the new configurations for gemm that provide the best performance.
The selection of the configuration is now based on the arithmetic intensity and not only on _M and _N dimension.
Following preliminary investigation and tuning with the auto-tuner, these are the new configurations for
gemm
that provide the best performance. The selection of the configuration is now based on the arithmetic intensity and not only on_M
and_N
dimension.