Benchmark results for the AKTIVE reader

Parent ↗

Summary

• Replacing the 3-nested blitter reading each value separately with a 2-nested loop reading entire (partial) rows boosts the performance of both file and string readers.

• The boost is most significant for the file reader, and less so for the string reader.

• The new file reader is about 20x faster than the old.

• The new string reader is about 1.11x faster than the old

• Overall the new file reader is roughly on par with the new string reader. Whereas the old file reader was about 21x slower than the old string reader.

Takeaways

• There is no need to look into caching for the file reader at this time.

• Even so, memory mapping might still be able to provide another boost.

• Ditto for a vectorized endian conversion.

• See if the same kind of thing can be applied to the NETPBM binary formats.

Plots