FYI : the code estimating parallelism of ABt operator is incorrect and may
result in grossly skewed figure. The result of it is incredible slow down
in performance, esp. over a few generations of multiplications in a
pipeline.

Reply via email to