Effects of Single and Double Precision

While I have been in a rush so far (and without time to write here), I thought I could share the below graph regarding single and double precision multiplication.

While the actual results aren’t mine (I just ran the code provided in the BLIS package and generated the graph), it’s interesting to see how the performance on single precision and double precision compares. Maybe it’s obvious that the performance would be doubled as the number of bits for single precision is half of the ones at the double, but it’s interesting to see this nevertheless.