First benchmark result on the fast_init compared to the default branch. Warning: the patch is still a WIP, there are maybe performance issues (in this case, this benchmark should help to identify them).
---
haypo@speed-python$ python3 -m perf compare_to 2017-01-23_14-02-default-cebc9c7ad195.json fast_init-0_ref_cebc9c7ad195.json -G --min-speed=2
Slower (6):
- telco: 13.0 ms +- 0.2 ms -> 13.8 ms +- 0.3 ms: 1.06x slower (+6%)
- scimark_sor: 391 ms +- 7 ms -> 413 ms +- 11 ms: 1.06x slower (+6%)
- fannkuch: 938 ms +- 7 ms -> 973 ms +- 13 ms: 1.04x slower (+4%)
- pickle_dict: 53.7 us +- 3.4 us -> 55.5 us +- 5.0 us: 1.03x slower (+3%)
- sqlite_synth: 8.00 us +- 0.12 us -> 8.25 us +- 0.14 us: 1.03x slower (+3%)
- scimark_monte_carlo: 202 ms +- 5 ms -> 207 ms +- 5 ms: 1.03x slower (+3%)

Faster (19):
- scimark_lu: 362 ms +- 13 ms -> 335 ms +- 17 ms: 1.08x faster (-8%)
- chameleon: 23.5 ms +- 0.3 ms -> 21.7 ms +- 0.3 ms: 1.08x faster (-7%)
- xml_etree_process: 179 ms +- 3 ms -> 168 ms +- 3 ms: 1.07x faster (-6%)
- pickle_pure_python: 1.03 ms +- 0.02 ms -> 982 us +- 11 us: 1.05x faster (-5%)
- regex_dna: 283 ms +- 1 ms -> 271 ms +- 1 ms: 1.05x faster (-4%)
- chaos: 239 ms +- 2 ms -> 230 ms +- 3 ms: 1.04x faster (-4%)
- django_template: 356 ms +- 5 ms -> 343 ms +- 5 ms: 1.04x faster (-4%)
- call_simple: 10.7 ms +- 0.4 ms -> 10.3 ms +- 0.3 ms: 1.04x faster (-3%)
- hexiom: 18.0 ms +- 0.1 ms -> 17.4 ms +- 0.1 ms: 1.04x faster (-3%)
- pathlib: 42.3 ms +- 0.5 ms -> 40.9 ms +- 0.4 ms: 1.03x faster (-3%)
- regex_compile: 379 ms +- 2 ms -> 367 ms +- 4 ms: 1.03x faster (-3%)
- go: 501 ms +- 7 ms -> 485 ms +- 4 ms: 1.03x faster (-3%)
- xml_etree_generate: 208 ms +- 4 ms -> 202 ms +- 3 ms: 1.03x faster (-3%)
- deltablue: 14.5 ms +- 0.3 ms -> 14.1 ms +- 0.2 ms: 1.03x faster (-3%)
- unpickle_pure_python: 652 us +- 8 us -> 634 us +- 9 us: 1.03x faster (-3%)
- richards: 147 ms +- 2 ms -> 143 ms +- 4 ms: 1.03x faster (-3%)
- nqueens: 213 ms +- 2 ms -> 208 ms +- 2 ms: 1.03x faster (-2%)
- raytrace: 1.08 sec +- 0.01 sec -> 1.05 sec +- 0.02 sec: 1.02x faster (-2%)
- unpickle: 31.0 us +- 0.4 us -> 30.3 us +- 1.0 us: 1.02x faster (-2%)

Benchmark hidden because not significant (34): (...)
Ignored benchmarks (5) of 2017-01-23_14-02-default-cebc9c7ad195.json: mako, sympy_expand, sympy_integrate, sympy_str, sympy_sum
---

Reminder: The fast_init branch is based on the issue #29259 (tp_fastcall), so the result combines tp_fastcall and tp_fastnew/fastinit.