Specialize binary ops for floats

[tekknolagi]

• Add BINARY_{ADD,SUB,MUL,POWER}_FLOAT specializations in interpreter
• Add asm interpreter implementation of DUP_TOP_TWO
• Add asm interpreter implementation of ROT_THREE
• Add INPLACE_{ADD,SUB}_FLOAT specializations in interpreter
• Add asm interpreter implementation of BINARY_MUL_FLOAT
• Add the other ops; they are very similar

[github-actions]

{
  "django_minimal_requests": {
    "benchmark": "django_minimal_requests",
    "cg_instructions before": 703648,
    "cg_instructions now": 703440,
    "cg_instructions ∆": "-0.0%",
    "interpreter_args": [],
    "interpreter_name": "pyro",
    "version before": "42d68454344d08f625a7eb579a8f47a536da81ee",
    "version now": "1df60d7c3b77ef93012bae8ddb265bdee34d8e92"
  }
}

[github-actions]

Summary

Metric	Average	Best	Worst	Notes
cg_instructions	-5.6%	nbody -70.8%	bench_pickle 0.0%	typically < 0.2% noise

Benchmark details

Base vs. New

benchmark	cg_instructions
2to3	-0.0%
bench_base64	0.0%
bench_compile	-0.0%
bench_pickle	0.0%
deltablue	0.0%
fannkuch	-0.6%
go	-0.3%
loadproperty	0.0%
nbody	-70.8%
nqueens	-0.4%
pyflate	-0.6%
pystone	0.0%
richards	0.0%

CPython vs New

benchmark	cg_instructions
2to3	-10.1%
bench_base64	-38.7%
bench_compile	1105.6%
bench_pickle	-27.9%
deltablue	-64.6%
fannkuch	-4.0%
go	-64.2%
loadproperty	-75.2%
nbody	-64.8%
nqueens	32.8%
pyflate	-34.1%
pystone	-71.8%
richards	-80.0%

Base

benchmark	cg_instructions
2to3	2,304,831,233
bench_base64	2,952,625,190
bench_compile	2,981,208,631
bench_pickle	2,988,351,069
deltablue	1,452,746,645
fannkuch	5,403,085,367
go	1,890,208,748
loadproperty	433,033,184
nbody	9,404,712,556
nqueens	2,950,974,047
pyflate	9,589,659,608
pystone	1,120,926,632
richards	963,107,922

New

benchmark	cg_instructions
2to3	2,304,333,472
bench_base64	2,952,704,309
bench_compile	2,980,701,440
bench_pickle	2,989,027,574
deltablue	1,452,811,206
fannkuch	5,372,599,906
go	1,883,720,191
loadproperty	433,095,411
nbody	2,745,758,571
nqueens	2,939,113,226
pyflate	9,536,150,856
pystone	1,120,981,849
richards	963,160,438

CPython

benchmark	cg_instructions
2to3	2,563,594,680
bench_base64	4,820,238,937
bench_compile	247,234,552
bench_pickle	4,143,854,730
deltablue	4,109,339,679
fannkuch	5,594,144,075
go	5,264,681,155
loadproperty	1,747,769,383
nbody	7,793,857,481
nqueens	2,213,171,998
pyflate	14,468,019,192
pystone	3,981,578,475
richards	4,821,555,544