Vectorization speeds up your Python code

pythonspeed.com

14 points·itamarst··3 comments

Great article. I especially liked the comparison of float32 to float64. I had no idea it was as large as a 10x improvement in a simple case like the one demonstrated.

* Elapsed (secs)

* CPU instructions (G)

* Cache misses (M)

* float64 : 4.7 1.6 30.4

* float32 : 0.4 1.2 16.7

This looks quite odd to me. I don't have perf installed on my machine but when I do

DATA = np.random.rand(300_000_000).astype(np.float32)

I get ~0.24s as a wall time for the normalization calculation on my machine, and

DATA = np.random.rand(300_000_000).astype(np.float64)

Is giving me ~0.33s

Note that line needs to be omitted from the time measurement, since it's the same for both scripts and is just overhead, I was just measuring the actual mean + substraction code.

I reran a few more times, got a bunch of variability, but did get some runs where f64 is just twice as slow, not 10×. Probably my initial run hit swapping. So I will update the article.

That being said, worth noting that this is also quite hardware dependent. Like, I just have 16MB L3 cache, I imagine higher L3 would change ratios a lot.

UPDATE: Oops. Original runs were with NumPy 1.18, this run was with NumPy 1.22. So will rerun again with that.

UPDATE2: NumPy version doesn't matter. So yeah was probably swapping.

UPDATE3: Fixed article should be up in a minute.

Reply on news.ycombinator.com