Rewrite LOAD_GLOBAL _Unbound to LOAD_IMMEDIATE in builtins module

[tekknolagi]

This should save some caches and be faster. Ideally we would do this in
all Skybison-controlled modules but I don't think we have a good signal
for that yet.

Partially addresses #395

[tekknolagi]

Hm. Looks like this is a big regression in the rewrite process since nqueens makes so many functions on the fly.

Potential ways around this:

• Inline generator comprehensions
• Hoist generator comprehensions to top-level functions
• Move rewriting to code objects (not great; they're supposed to be immutable)
• Do this rewrite in the (Python) compiler

[tekknolagi]

Bleh. Nqueens still has a 0.4% regression because rewriting takes longer.

[github-actions]

{
  "django_minimal_requests": {
    "benchmark": "django_minimal_requests",
    "cg_instructions before": 720237,
    "cg_instructions now": 720349,
    "cg_instructions ∆": "0.0%",
    "interpreter_args": [],
    "interpreter_name": "pyro",
    "version before": "5215bf66fdf658b5a1758e9e18955ad42310e570",
    "version now": "f10292b021f4e5ff072b466d771f0f0a378cbc15"
  }
}

[github-actions]

Summary

Metric	Average	Best	Worst	Notes
cg_instructions	0.1%	bench_base64 -0.1%	nqueens 0.6%	typically < 0.2% noise

Benchmark details

Base vs. New

benchmark	cg_instructions
2to3	0.1%
bench_base64	-0.1%
bench_compile	0.1%
bench_pickle	0.0%
deltablue	0.1%
fannkuch	-0.0%
go	0.1%
loadproperty	0.3%
nbody	0.0%
nqueens	0.6%
pyflate	-0.0%
pystone	0.1%
richards	0.1%

CPython vs New

benchmark	cg_instructions
2to3	-8.6%
bench_base64	-38.8%
bench_compile	1117.1%
bench_pickle	-19.4%
deltablue	-64.5%
fannkuch	-1.3%
go	-64.0%
loadproperty	-75.1%
nbody	20.7%
nqueens	42.8%
pyflate	-30.6%
pystone	-71.8%
richards	-80.0%

Base

benchmark	cg_instructions
2to3	2,342,001,511
bench_base64	2,954,425,156
bench_compile	3,006,175,526
bench_pickle	3,341,162,384
deltablue	1,459,393,975
fannkuch	5,521,747,288
go	1,893,969,866
loadproperty	434,780,405
nbody	9,406,459,048
nqueens	3,141,766,729
pyflate	10,038,792,535
pystone	1,122,683,016
richards	964,856,353

New

benchmark	cg_instructions
2to3	2,343,890,737
bench_base64	2,951,823,816
bench_compile	3,009,519,754
bench_pickle	3,341,455,395
deltablue	1,460,414,307
fannkuch	5,519,137,067
go	1,895,104,993
loadproperty	435,916,734
nbody	9,407,598,312
nqueens	3,160,980,700
pyflate	10,035,975,823
pystone	1,123,820,711
richards	965,996,283

CPython

benchmark	cg_instructions
2to3	2,563,610,701
bench_base64	4,820,248,126
bench_compile	247,270,261
bench_pickle	4,143,805,223
deltablue	4,109,440,109
fannkuch	5,594,149,369
go	5,264,709,090
loadproperty	1,747,774,500
nbody	7,793,864,282
nqueens	2,213,176,123
pyflate	14,468,046,983
pystone	3,981,585,959
richards	4,821,559,159