GeezerVison

The World Wide Web, restored to 800 pixels of glory.

GeezerVison is showing github.com — Specialize BINARY_SUBTRACT · Issue #100 · faster-cpython/ideas

Specialize BINARY_SUBTRACT #100

New issue

Copy link

New issue

Copy link

Closed

Specialize BINARY_SUBTRACT#100

Copy link

Description

corona10

opened

on Oct 17, 2021

Issue body actions

I would like to test that specializing BINARY_SUBTRACT is worth implementing.
I will leave an experiment result by referencing an already done task.

Experiment commit:
corona10/cpython@fb360a2
https://github.com/corona10/cpython/tree/BINARY_SUBTRACT_specialize

cc Thanks to Ken Jin's tutorial and Dennis's work, It was easy to follow up on how to implement this task.
Thank you again.

Reacted by Ken Jin

Activity

corona10
changed the title ~~[-]Specialization BINARY_SUBTRACT[/-]~~ [+]Specialize BINARY_SUBTRACT[/+] on Oct 17, 2021

corona10 commented on Oct 17, 2021

corona10

on Oct 17, 2021

Author

More actions

Environment: MS Azure D8s v3(8 vcpu, 32GiB Memory) -> FYI, non-isolated environment.
I am waiting for @pablogsal 's isolated tests :) (Thanks Pablo!)

fannkuch / nbody / richards become faster.

Benchmark	baseline	specialize
chaos	114 ms	112 ms: 1.02x faster
crypto_pyaes	131 ms	128 ms: 1.02x faster
deltablue	7.57 ms	7.72 ms: 1.02x slower
django_template	64.0 ms	63.5 ms: 1.01x faster
dulwich_log	106 ms	105 ms: 1.01x faster
fannkuch	617 ms	593 ms: 1.04x faster
float	131 ms	132 ms: 1.01x slower
go	252 ms	251 ms: 1.00x faster
json_dumps	18.1 ms	18.3 ms: 1.01x slower
json_loads	39.2 us	39.8 us: 1.01x slower
logging_format	12.3 us	12.4 us: 1.01x slower
logging_silent	191 ns	195 ns: 1.02x slower
mako	19.6 ms	20.0 ms: 1.02x slower
meteor_contest	144 ms	145 ms: 1.01x slower
nbody	173 ms	164 ms: 1.05x faster
nqueens	134 ms	132 ms: 1.01x faster
pathlib	31.4 ms	31.6 ms: 1.00x slower
pickle_dict	37.2 us	37.8 us: 1.02x slower
pickle_list	5.85 us	5.77 us: 1.01x faster
pidigits	254 ms	254 ms: 1.00x faster
pyflate	802 ms	792 ms: 1.01x faster
python_startup	12.1 ms	12.1 ms: 1.01x slower
raytrace	524 ms	521 ms: 1.01x faster
regex_dna	268 ms	266 ms: 1.01x faster
regex_effbot	4.16 ms	4.14 ms: 1.00x faster
regex_v8	31.9 ms	31.8 ms: 1.00x faster
richards	87.6 ms	91.6 ms: 1.05x slower
scimark_fft	516 ms	495 ms: 1.04x faster
scimark_lu	224 ms	219 ms: 1.02x faster
scimark_monte_carlo	110 ms	109 ms: 1.01x faster
scimark_sor	225 ms	235 ms: 1.04x slower
spectral_norm	177 ms	175 ms: 1.01x faster
sympy_integrate	32.1 ms	32.5 ms: 1.01x slower
sympy_sum	257 ms	258 ms: 1.00x slower
sympy_str	456 ms	458 ms: 1.00x slower
unpack_sequence	56.2 ns	54.9 ns: 1.02x faster
unpickle	21.3 us	21.7 us: 1.02x slower
unpickle_list	6.56 us	6.65 us: 1.01x slower
unpickle_pure_python	417 us	413 us: 1.01x faster
xml_etree_process	96.3 ms	96.8 ms: 1.01x slower
Geometric mean	(ref)	1.00x faster

Benchmark hidden because not significant (16): 2to3, chameleon, hexiom, logging_simple, pickle, pickle_pure_python, python_startup_no_site, regex_compile, scimark_sparse_mat_mult, sqlite_synth, sympy_expand, telco, tornado_http, xml_etree_parse, xml_etree_iterparse, xml_etree_generate

corona10 commented on Oct 17, 2021

corona10

on Oct 17, 2021

Author

More actions

fannkuch

    binary_add.specialization_success : 44
    binary_add.specialization_failure : 5
    binary_add.hit : 6930513
    binary_add.deferred : 318
    binary_add.miss : 28
    binary_add.deopt : 6
    binary_add.unquickened : 191

    binary_multiply.specialization_success : 12
    binary_multiply.specialization_failure : 8
    binary_multiply.hit : 288
    binary_multiply.deferred : 198
    binary_multiply.miss : 0
    binary_multiply.deopt : 0
    binary_multiply.unquickened : 83

    binary_subtract.specialization_success : 27
    binary_subtract.specialization_failure : 5
    binary_subtract.hit : 1043817
    binary_subtract.deferred : 244
    binary_subtract.miss : 0
    binary_subtract.deopt : 0
    binary_subtract.unquickened : 75

nbody

    binary_add.specialization_success : 49
    binary_add.specialization_failure : 5
    binary_add.hit : 1605746
    binary_add.deferred : 318
    binary_add.miss : 28
    binary_add.deopt : 6
    binary_add.unquickened : 220

    binary_multiply.specialization_success : 38
    binary_multiply.specialization_failure : 8
    binary_multiply.hit : 10800700
    binary_multiply.deferred : 200
    binary_multiply.miss : 0
    binary_multiply.deopt : 0
    binary_multiply.unquickened : 231

    binary_subtract.specialization_success : 32
    binary_subtract.specialization_failure : 5
    binary_subtract.hit : 2401427
    binary_subtract.deferred : 244
    binary_subtract.miss : 0
    binary_subtract.deopt : 0
    binary_subtract.unquickened : 104

raytrace

    binary_add.specialization_success : 346
    binary_add.specialization_failure : 2219
    binary_add.hit : 4462044
    binary_add.deferred : 159961
    binary_add.miss : 394452
    binary_add.deopt : 289
    binary_add.unquickened : 363

    binary_multiply.specialization_success : 15739
    binary_multiply.specialization_failure : 10779
    binary_multiply.hit : 7814370
    binary_multiply.deferred : 1694199
    binary_multiply.miss : 376198
    binary_multiply.deopt : 15703
    binary_multiply.unquickened : 274


    binary_subtract.specialization_success : 16807
    binary_subtract.specialization_failure : 42058
    binary_subtract.hit : 2128006
    binary_subtract.deferred : 3763883
    binary_subtract.miss : 325179
    binary_subtract.deopt : 16760
    binary_subtract.unquickened : 239

pablogsal commented on Oct 18, 2021

pablogsal

on Oct 18, 2021

Collaborator

More actions

Results on my benchmark machines:

❯ pyperf compare_to json/* --table --table-format=md -G

Benchmark	2021-10-17_12-06-main-b3f0ceae919c	2021-10-17_17-39-binary-subscript
regex_effbot	4.05 ms	3.75 ms: 1.08x faster
unpack_sequence	63.4 ns	59.9 ns: 1.06x faster
regex_dna	270 ms	258 ms: 1.05x faster
regex_v8	30.8 ms	29.5 ms: 1.05x faster
float	107 ms	103 ms: 1.04x faster
chaos	95.0 ms	91.9 ms: 1.03x faster
mako	17.4 ms	16.9 ms: 1.03x faster
fannkuch	511 ms	500 ms: 1.02x faster
pickle	13.4 us	13.2 us: 1.02x faster
go	217 ms	213 ms: 1.02x faster
hexiom	9.58 ms	9.41 ms: 1.02x faster
unpickle_pure_python	343 us	337 us: 1.02x faster
scimark_sor	189 ms	186 ms: 1.02x faster
pickle_dict	36.9 us	36.3 us: 1.02x faster
telco	7.68 ms	7.56 ms: 1.02x faster
scimark_monte_carlo	96.4 ms	94.9 ms: 1.02x faster
pyflate	663 ms	652 ms: 1.02x faster
sqlite_synth	3.22 us	3.17 us: 1.01x faster
xml_etree_process	79.8 ms	78.8 ms: 1.01x faster
json_dumps	16.8 ms	16.6 ms: 1.01x faster
dulwich_log	83.9 ms	83.2 ms: 1.01x faster
crypto_pyaes	107 ms	106 ms: 1.01x faster
sympy_integrate	27.3 ms	27.1 ms: 1.01x faster
xml_etree_parse	188 ms	186 ms: 1.01x faster
python_startup	11.5 ms	11.4 ms: 1.01x faster
sympy_sum	212 ms	211 ms: 1.01x faster
python_startup_no_site	8.48 ms	8.43 ms: 1.01x faster
2to3	319 ms	318 ms: 1.00x faster
pickle_pure_python	463 us	461 us: 1.00x faster
pidigits	235 ms	235 ms: 1.00x slower
meteor_contest	132 ms	133 ms: 1.00x slower
deltablue	6.21 ms	6.24 ms: 1.00x slower
sqlalchemy_imperative	22.6 ms	22.8 ms: 1.01x slower
raytrace	413 ms	416 ms: 1.01x slower
chameleon	9.84 ms	9.94 ms: 1.01x slower
pathlib	20.8 ms	21.0 ms: 1.01x slower
django_template	48.6 ms	49.2 ms: 1.01x slower
pickle_list	5.30 us	5.37 us: 1.01x slower
scimark_sparse_mat_mult	5.73 ms	5.83 ms: 1.02x slower
nbody	136 ms	139 ms: 1.02x slower
spectral_norm	138 ms	141 ms: 1.02x slower
unpickle	16.3 us	16.8 us: 1.03x slower
Geometric mean	(ref)	1.01x faster

Benchmark hidden because not significant (16): tornado_http, logging_format, sqlalchemy_declarative, logging_simple, xml_etree_iterparse, unpickle_list, xml_etree_generate, scimark_lu, logging_silent, scimark_fft, sympy_expand, nqueens, json_loads, richards, regex_compile, sympy_st

❯ pyperf system
Show the system configuration
Unit irqbalance.service could not be found.

System state
============

CPU: use 2 logical CPUs: 17,35
Perf event: Maximum sample rate: 1 per second
ASLR: Full randomization
Linux scheduler: Isolated CPUs (2/36): 17,35
Linux scheduler: RCU disabled on CPUs (2/36): 17,35
CPU Frequency: 0-16,18-34=min=1200 MHz, max=3000 MHz; 17,35=min=max=3000 MHz
CPU scaling governor (intel_pstate): performance
Turbo Boost (intel_pstate): Turbo Boost disabled
IRQ affinity: Default IRQ affinity: CPU 0-16,18-34
IRQ affinity: IRQ affinity: IRQ 0,2=CPU 0-35; IRQ 1,3-16,21,25-31,56-59,68-85,87,89-90,92-93,95-104=CPU 0-16,18-34; IRQ 32=CPU 0,18; IRQ 33=CPU 1,19; IRQ 34=CPU 2,20; IRQ 35=CPU 3,21; IRQ 36=CPU 4,22; IRQ 37=CPU 5; IRQ 38=CPU 6; IRQ 39=CPU 7; IRQ 40=CPU 8; IRQ 41=CPU 9; IRQ 42=CPU 10; IRQ 43=CPU 11; IRQ 44=CPU 12; IRQ 45=CPU 13; IRQ 46=CPU 14; IRQ 47=CPU 15; IRQ 48=CPU 16; IRQ 49=CPU 17; IRQ 50=CPU 23; IRQ 51=CPU 24; IRQ 52=CPU 25; IRQ 53=CPU 26; IRQ 54=CPU 27; IRQ 55=CPU 28; IRQ 60=CPU 29; IRQ 61=CPU 30; IRQ 62=CPU 31; IRQ 63=CPU 32; IRQ 64=CPU 33; IRQ 65=CPU 34; IRQ 66=CPU 35

OK! System ready for benchmarking

Reacted by Donghee Na

markshannon commented on Oct 18, 2021
markshannon
on Oct 18, 2021
Member
More actions
See #101
corona10
closed this as completed on Nov 28, 2021
gramster
added this to Fancy CPython Boardon Jan 10, 2022
gramster
moved this to Todo in Fancy CPython Boardon Jan 10, 2022
gramster
moved this from Todo to Other in Fancy CPython Boardon Jan 10, 2022
gramster
removed this from Fancy CPython Boardon Jan 24, 2022

Metadata

Assignees

No one assigned

Labels

No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions

You are visitor 01402520 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.

GeezerVison

Specialize BINARY_SUBTRACT #100

Description

Activity

corona10 commented on Oct 17, 2021

corona10 commented on Oct 17, 2021

fannkuch

nbody

raytrace

pablogsal commented on Oct 18, 2021

markshannon commented on Oct 18, 2021

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions