PERF: Optimize point transformations

[snowman2]

Related: #1202

In [1]: from pyproj import Transformer

In [2]: transformer = Transformer.from_crs(4326, 3857, always_xy=True)

In [3]: %timeit transformer.transform(1, 2)
4.69 µs ± 37.9 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

In [4]: %timeit transformer.transform_point(1, 2)
2.08 µs ± 20.2 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

Version in #1203:

In [5]: %timeit transformer.transform_point(1, 2)
2.99 µs ± 132 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

[codecov]

Codecov Report

Merging #1204 (c9126d3) into main (c4cb8f0) will increase coverage by 0.00%.
The diff coverage is 100.00%.

@@           Coverage Diff           @@
##             main    #1204   +/-   ##
=======================================
  Coverage   96.28%   96.29%           
=======================================
  Files          20       20           
  Lines        1804     1808    +4     
=======================================
+ Hits         1737     1741    +4     
  Misses         67       67           

Impacted Files	Coverage Δ
pyproj/transformer.py	95.04% <100.00%> (+0.09%)	⬆️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[greglucas]

Nice! This works great for me.

[snowman2]

With #1208

In [3]: transformer = Transformer.from_crs(4326, 3857, always_xy=True)

In [4]: %timeit transformer.transform(1, 2)
1.44 µs ± 38.6 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [5]: %timeit transformer.transform(1, 2)
1.42 µs ± 14.1 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

In [6]: %timeit transformer.transform(1, 2)
1.44 µs ± 19.1 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)

[greglucas]

Looks good to me! I get similar timings to you, and not a major slowdown for trying this before going down the other path. So, I do like this better than adding a new method, and theoretically downstream users should be able to get immediate benefits without needing to change the call sites.

[snowman2]

I do like this better than adding a new method, and theoretically downstream users should be able to get immediate benefits without needing to change the call sites.

I agree