[Python-Dev] Timings [Was: Python-checkins] python/dist/src/Python ceval.c, 2.383, 2.384

[Raymond Hettinger]

> A 2% speed improvement with gcc on low-endian machines.  My guess is
that
> this
> new pattern for NEXTARG() is detected and optimized as a single
(*short)
> loading.

One other thought:  When testing these kind of optimizations, do not use
PyStone!

Unfortunately, part of its design is to time a long for-loop and the net
that timing out of the total.  At best, that means that improvements to
for-loops don't showup on PyStone.

At worst, PyStone gives results *opposite* of reality.  Because of cache
effects, empty for-loops run a tad faster than for-loops that do
something.


Raymond Hettinger