gh-121562: optimized hex_from_char

[LimaBD]

Optimized hex_from_char performance

issue #121562

Timing with old hex_from_char:

$ ./python -m pyperf timeit --duplicate 100 "float.fromhex('0x123456.ffffp10')"
.....................
Mean +- std dev: 86.0 ns +- 0.9 ns

With new hex_from_char:

$ ./python -m pyperf timeit --duplicate 100 "float.fromhex('0x123456.ffffp10')"
.....................
Mean +- std dev: 81.8 ns +- 1.0 ns

If we measure hex_from_char directly, probably we will get a much better % improvement because there is additional calls that float.fromhex makes apart from this one.

• Issue: hex_from_char run faster replacing switch-case with lookup table #121562

[mdickinson]

LGTM in principle. The old code is agnostic to character encoding (works equally well with ASCII / EBCDIC / whatever), while the new code is based on the bytes of the UTF8-encoded string. That's fine, since the only use-case is applying this to a UTF-8-encoded value. But please could we add a comment that makes it clear that this table should only be used with encodings whose lower half is ASCII (like UTF-8).

[LimaBD]

Hello @mdickinson, thanks for your time, I just added the comment clarifying it but my English is average, anything let me know!

[mdickinson]

Thanks for the update; the comment looks good.

[mdickinson]

@LimaBD Merged. Thank you for the contribution!