dict: Use DK_LOG_SIZE in hot loop.

[methane]

DK_LOG_SIZE(key) < 8 is faster than DK_SIZE(key) <= 0xff, at least on
GCC.

[methane]

• Ubuntu 20.04 LTS
• gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
• without --enable-optimization --with-lto

$ cpython/python -m pyperf compare_to main.json main2.json -G --min-speed=1
Slower (3):
- scimark_sparse_mat_mult: 6.70 ms +- 0.10 ms -> 6.95 ms +- 0.10 ms: 1.04x slower
- nqueens: 114 ms +- 1 ms -> 117 ms +- 1 ms: 1.03x slower
- float: 95.9 ms +- 1.6 ms -> 97.4 ms +- 1.3 ms: 1.02x slower

Faster (25):
- pickle_dict: 39.2 us +- 4.6 us -> 36.3 us +- 0.2 us: 1.08x faster
- regex_effbot: 3.98 ms +- 0.04 ms -> 3.72 ms +- 0.02 ms: 1.07x faster
- regex_v8: 29.9 ms +- 0.4 ms -> 28.4 ms +- 0.3 ms: 1.05x faster
- deltablue: 5.79 ms +- 0.42 ms -> 5.58 ms +- 0.08 ms: 1.04x faster
- sympy_expand: 694 ms +- 11 ms -> 675 ms +- 4 ms: 1.03x faster
- mako: 13.7 ms +- 0.1 ms -> 13.3 ms +- 0.1 ms: 1.03x faster
- sympy_str: 416 ms +- 7 ms -> 405 ms +- 3 ms: 1.03x faster
- scimark_fft: 475 ms +- 24 ms -> 464 ms +- 7 ms: 1.02x faster
- json_dumps: 16.6 ms +- 0.3 ms -> 16.2 ms +- 0.1 ms: 1.02x faster
- scimark_lu: 153 ms +- 7 ms -> 149 ms +- 2 ms: 1.02x faster
- unpack_sequence: 56.1 ns +- 1.4 ns -> 54.9 ns +- 0.9 ns: 1.02x faster
- regex_dna: 238 ms +- 2 ms -> 233 ms +- 1 ms: 1.02x faster
- unpickle_list: 6.06 us +- 0.21 us -> 5.94 us +- 0.08 us: 1.02x faster
- regex_compile: 185 ms +- 3 ms -> 182 ms +- 1 ms: 1.02x faster
- spectral_norm: 147 ms +- 3 ms -> 145 ms +- 1 ms: 1.02x faster
- sympy_integrate: 27.6 ms +- 0.4 ms -> 27.2 ms +- 0.3 ms: 1.02x faster
- pickle_list: 5.45 us +- 0.07 us -> 5.36 us +- 0.05 us: 1.02x faster
- django_template: 51.0 ms +- 1.2 ms -> 50.3 ms +- 0.5 ms: 1.02x faster
- dulwich_log: 100 ms +- 5 ms -> 99.1 ms +- 0.7 ms: 1.01x faster
- crypto_pyaes: 111 ms +- 2 ms -> 110 ms +- 1 ms: 1.01x faster
- chameleon: 9.18 ms +- 0.08 ms -> 9.07 ms +- 0.10 ms: 1.01x faster
- hexiom: 8.40 ms +- 0.08 ms -> 8.30 ms +- 0.14 ms: 1.01x faster
- xml_etree_parse: 183 ms +- 4 ms -> 181 ms +- 1 ms: 1.01x faster
- tornado_http: 153 ms +- 4 ms -> 152 ms +- 2 ms: 1.01x faster
- unpickle_pure_python: 349 us +- 8 us -> 345 us +- 2 us: 1.01x faster

Benchmark hidden because not significant (31): 2to3, chaos, fannkuch, go, html5lib, json_loads, logging_format, logging_silent, logging_simple, meteor_contest, nbody, pathlib, pickle, pickle_pure_python, pidigits, pyflate, python_startup, python_startup_no_site, raytrace, richards, scimark_monte_carlo, scimark_sor
, sqlalchemy_declarative, sqlalchemy_imperative, sqlite_synth, sympy_sum, telco, unpickle, xml_etree_iterparse, xml_etree_generate, xml_etree_process

Geometric mean: 1.01x faster

[markshannon]

Apart from one variable name, looks good.