Import benchmarks from prior work

[bennn]

Two sources:

1. Static Python https://github.com/facebookincubator/cinder/tree/0baa6051949aef10db8f02fca37005d48849aca9/Tools/benchmarks
2. Reticulated performance https://github.com/nuprl/retic_performance/tree/master/benchmarks

TODO:

• [x] import the SP benchmarks that already have types
  • [x] find way to organize between untyped / shallow typed / concrete typed / primitive typed
  • [x] generate all combinations (what is all?)
  • [x] import the SP benchmarks for all the files in the repo:
  • [x] deltablue
  • [x] fannkuch
  • [x] nbody
  • [x] nqueens
  • [x] pystone
  • [x] richards
• [ ] import the Retic benchmarks, check that the types work in SP
  • [x] start with pidigits, should be easy
  • [x] 2nd, try spectralnorm
  • [x] go
  • [x] http2
  • [x] slowsha
  • [x] Espionage
  • [x] Evolution
  • [x] PythonFlow
  • [ ] aespython
  • [x] call_method
  • [x] call_method_slots
  • [x] call_simple
  • [x] chaos
  • [x] fannkuch
  • [x] float
  • [x] futen
  • [x] meteor
  • [x] nbody
  • [x] nqueens
  • [x] pystone
  • [x] sample_fsm
  • [x] sieve
  • [x] stats
  • [x] take5
• [ ] try adding types to other SP benchmarks
• [ ] look for new benchmarks

[bennn]

@vivaan2006 for next time (05/10):

• [x] make a new file Benchmarks/spectralnorm/untyped/main.py, copy in spectralnorm and remove all types, make sure it runs with normal python
• [x] make file Benchmarks/spectralnorm/shallow/main.py, copy in spectral norm, type updating the types for Static Python
  • [x] List(T) chages to List[T]
  • [x] all files need standard imports from __static__
  • [x] every variable declaration needs a type
• [x] try the same for float: make untyped & shallow versions, check than untyped runs

[bennn]

Great job with spectralnorm and float.

Try Espionage for next time.

[bennn]

For next time, try the c benchmarks: call_x and chaos

[mrigankpawagi]

Though there is a lot of overlap, there are some more benchmarks with PyPy and with pyperformance.

[mrigankpawagi]

I was actually able to run pyperformance on Python 3.10.5+cinder, although I had to use release 1.0.8 (so pip install pyperformance==1.0.8) since the next version introduced changes to a few benchmarks that work only with Python 3.11+.

There was one error related to nargs, in the msgpack module. This was "fixed" by changing Packer(**kwargs).pack(o) to Packer(**kwargs).pack.__call__(o) in L36 of the module's __init__.py (the version being that installed by pyperformance). I think we can include these benchmarks as well to this repository. The pyperformance module comes with a driver like the one we need.

PS one of the benchmarks in this is django_template which is a very simple benchmark but it runs (reconfirming that Django indeed can work with static-python).

[bennn]

Ok, both these leads sound great (pyperformance + django)!