Slide 1

Slide 1 text

No content

Slide 2

Slide 2 text

No content

Slide 3

Slide 3 text

slomo the movie www. .com Special thanks

Slide 4

Slide 4 text

Introducing Dr. John Kitchin The neurologist

Slide 5

Slide 5 text

Introducing Dr. John Kitchin The inspiration

Slide 6

Slide 6 text

Introducing Dr. John Kitchin The skater

Slide 7

Slide 7 text

What is your goal in life?

Slide 8

Slide 8 text

Grow up

Slide 9

Slide 9 text

Pick a career

Slide 10

Slide 10 text

Buy a sports car

Slide 11

Slide 11 text

Retire

Slide 12

Slide 12 text

The faster the better

Slide 13

Slide 13 text

Slomo

Slide 14

Slide 14 text

Said “no”

Slide 15

Slide 15 text

Introducing Schneems The narrator

Slide 16

Slide 16 text

Sounds like “schnapps”

Slide 17

Slide 17 text

The narrator said, explaining his own introduction

Slide 18

Slide 18 text

No content

Slide 19

Slide 19 text

No content

Slide 20

Slide 20 text

Maintain Sprockets

Slide 21

Slide 21 text

Those are all my hobbies

Slide 22

Slide 22 text

Full time job:

Slide 23

Slide 23 text

Resisting the Trump Administration

Slide 24

Slide 24 text

I consider myself a very chill person

Slide 25

Slide 25 text

Like to mix my yoga habits with resistance

Slide 26

Slide 26 text

“ohm”

Slide 27

Slide 27 text

BTW

Slide 28

Slide 28 text

In this talk we will not use “nil”

Slide 29

Slide 29 text

Naan

Slide 30

Slide 30 text

Sorry for ny naan jokes

Slide 31

Slide 31 text

Sorry for ny naan jokes

Slide 32

Slide 32 text

This

Slide 33

Slide 33 text

talk

Slide 34

Slide 34 text

is

Slide 35

Slide 35 text

about

Slide 36

Slide 36 text

speed

Slide 37

Slide 37 text

No content

Slide 38

Slide 38 text

The fastest way to get where you’re going

Slide 39

Slide 39 text

Depends on where you’re going

Slide 40

Slide 40 text

To get there faster

Slide 41

Slide 41 text

Let’s slow down

Slide 42

Slide 42 text

Why would we want slower code?

Slide 43

Slide 43 text

Measure it

Slide 44

Slide 44 text

task "assets:profile" do puts "==============" StackProf.run(mode: :wall, out: "tmp/stackprof.dump") do Rake::Task["assets:precompile"].invoke end puts "Running: $ stackprof tmp/stackprof.dump" puts `stackprof tmp/stackprof.dump` end

Slide 45

Slide 45 text

task "assets:profile" do puts "==============" StackProf.run(mode: :wall, out: "tmp/stackprof.dump") do Rake::Task["assets:precompile"].invoke end puts "Running: $ stackprof tmp/stackprof.dump" puts `stackprof tmp/stackprof.dump` end

Slide 46

Slide 46 text

task "assets:profile" do puts "==============" StackProf.run(mode: :wall, out: "tmp/stackprof.dump") do Rake::Task["assets:precompile"].invoke end puts "Running: $ stackprof tmp/stackprof.dump" puts `stackprof tmp/stackprof.dump` end

Slide 47

Slide 47 text

$ stackprof tmp/stackprof.dump

Slide 48

Slide 48 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set

Slide 49

Slide 49 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set Class and method

Slide 50

Slide 50 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set Total # of samples

Slide 51

Slide 51 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set Percentage

Slide 52

Slide 52 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set Total at TOP of stack

Slide 53

Slide 53 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set Percent at TOP of stack

Slide 54

Slide 54 text

Running: $ stackprof tmp/stackprof.dump ================================== Mode: wall(1000) Samples: 2083 (62.23% miss rate) GC: 282 (13.54%) ================================== TOTAL (pct) SAMPLES (pct) FRAME 313 (15.0%) 313 (15.0%) Set#include? 307 (14.7%) 243 (11.7%) Sprockets::DigestUtils#digest 743 (35.7%) 137 (6.6%) Kernel#require 476 (22.9%) 124 (6.0%) Kernel#require 157 (7.5%) 92 (4.4%) SassC::Rails::Importer#imports 243 (11.7%) 81 (3.9%) # 67 (3.2%) 67 (3.2%) NumericWithFormat#to_s 160 (7.7%) 53 (2.5%) PathUtils#atomic_write 44 (2.1%) 44 (2.1%) # 239 (11.5%) 37 (1.8%) Sprockets::Cache::FileStore#set

Slide 55

Slide 55 text

26% of total execution samples

Slide 56

Slide 56 text

What is calling so much? Set#include?

Slide 57

Slide 57 text

$ stackprof tmp/stackprof.dump --method Set#include? Set#include? (/Users/richardschneeman/.rubies/ruby-2.3.1/lib/ruby/2.3.0/set.rb:214) samples: 313 self (15.0%) / 313 total (15.0%) callers: 312 ( 99.7%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 1 ( 0.3%) Sprockets::Utils#dfs_paths code: | 214 | def include?(o) 313 (15.0%) / 313 (15.0%) | 215 | @hash[o] | 216 | end

Slide 58

Slide 58 text

$ stackprof tmp/stackprof.dump --method Set#include? Set#include? (/Users/richardschneeman/.rubies/ruby-2.3.1/lib/ruby/2.3.0/set.rb:214) samples: 313 self (15.0%) / 313 total (15.0%) callers: 312 ( 99.7%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 1 ( 0.3%) Sprockets::Utils#dfs_paths code: | 214 | def include?(o) 313 (15.0%) / 313 (15.0%) | 215 | @hash[o] | 216 | end

Slide 59

Slide 59 text

What is calling so much? valid_processor_metadata_value?

Slide 60

Slide 60 text

$ stackprof tmp/stackprof.dump --method Sprockets::ProcessorUtils#valid_processor_metadata_value? Sprockets::ProcessorUtils#valid_processor_metadata_value? (/Users/richardschneeman/.gem/ruby/2.3.1/bundler/ gems/sprockets-3b0d6732c13f/lib/sprockets/processor_utils.rb:170) samples: 24 self (1.2%) / 2129 total (102.2%) callers: 1793 ( 84.2%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 336 ( 15.8%) Sprockets::ProcessorUtils#validate_processor_result! callees (2105 total): 1793 ( 85.2%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 312 ( 14.8%) Set#include? code: | 170 | def valid_processor_metadata_value?(value) 261 (12.5%) / 2 (0.1%) | 171 | if VALID_METADATA_VALUE_TYPES.include?(value.class) | 172 | true 61 (2.9%) / 8 (0.4%) | 173 | elsif VALID_METADATA_COMPOUND_TYPES.include?(value.class) 1806 (86.7%) / 13 (0.6%) | 174 | value.all? { |v| valid_processor_metadata_value?(v) } | 175 | else 1 (0.0%) / 1 (0.0%) | 176 | false | 177 | end

Slide 61

Slide 61 text

$ stackprof tmp/stackprof.dump --method Sprockets::ProcessorUtils#valid_processor_metadata_value? Sprockets::ProcessorUtils#valid_processor_metadata_value? (/Users/richardschneeman/.gem/ruby/2.3.1/bundler/ gems/sprockets-3b0d6732c13f/lib/sprockets/processor_utils.rb:170) samples: 24 self (1.2%) / 2129 total (102.2%) callers: 1793 ( 84.2%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 336 ( 15.8%) Sprockets::ProcessorUtils#validate_processor_result! callees (2105 total): 1793 ( 85.2%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 312 ( 14.8%) Set#include? code: | 170 | def valid_processor_metadata_value?(value) 261 (12.5%) / 2 (0.1%) | 171 | if VALID_METADATA_VALUE_TYPES.include?(value.class) | 172 | true 61 (2.9%) / 8 (0.4%) | 173 | elsif VALID_METADATA_COMPOUND_TYPES.include?(value.class) 1806 (86.7%) / 13 (0.6%) | 174 | value.all? { |v| valid_processor_metadata_value?(v) } | 175 | else 1 (0.0%) / 1 (0.0%) | 176 | false | 177 | end

Slide 62

Slide 62 text

$ stackprof tmp/stackprof.dump --method Sprockets::ProcessorUtils#valid_processor_metadata_value? Sprockets::ProcessorUtils#valid_processor_metadata_value? (/Users/richardschneeman/.gem/ruby/2.3.1/bundler/ gems/sprockets-3b0d6732c13f/lib/sprockets/processor_utils.rb:170) samples: 24 self (1.2%) / 2129 total (102.2%) callers: 1793 ( 84.2%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 336 ( 15.8%) Sprockets::ProcessorUtils#validate_processor_result! callees (2105 total): 1793 ( 85.2%) Sprockets::ProcessorUtils#valid_processor_metadata_value? 312 ( 14.8%) Set#include? code: | 170 | def valid_processor_metadata_value?(value) 261 (12.5%) / 2 (0.1%) | 171 | if VALID_METADATA_VALUE_TYPES.include?(value.class) | 172 | true 61 (2.9%) / 8 (0.4%) | 173 | elsif VALID_METADATA_COMPOUND_TYPES.include?(value.class) 1806 (86.7%) / 13 (0.6%) | 174 | value.all? { |v| valid_processor_metadata_value?(v) } | 175 | else 1 (0.0%) / 1 (0.0%) | 176 | false | 177 | end

Slide 63

Slide 63 text

def valid_processor_metadata_value?(value) if VALID_METADATA_VALUE_TYPES.include?(value.class) true elsif VALID_METADATA_COMPOUND_TYPES.include?(value.class) value.all? { |v| valid_processor_metadata_value?(v) } else false end end

Slide 64

Slide 64 text

Set#include?

Slide 65

Slide 65 text

# File set.rb, line 214 def include?(o) @hash[o] end

Slide 66

Slide 66 text

Set is powered by a Hash under the hood

Slide 67

Slide 67 text

Skip the Set, just use a Hash

Slide 68

Slide 68 text

But why?

Slide 69

Slide 69 text

Ruby has optimized instructions for hash calls

Slide 70

Slide 70 text

code = " foo = Hash.new foo[:bar] " puts RubyVM::InstructionSequence.compile(code).disasm # 0000 trace 1 ( 2) # 0002 getinlinecache 9, # 0005 getconstant :Hash # 0007 setinlinecache # 0009 opt_send_without_block , # 0012 setlocal_OP__WC__0 2 # 0014 trace 1 ( 3) # 0016 getlocal_OP__WC__0 2 # 0018 putobject :bar # 0020 opt_aref , # 0023 leave

Slide 71

Slide 71 text

Compare to Set

Slide 72

Slide 72 text

code = " s = Set.new s.include?(:bar) " puts RubyVM::InstructionSequence.compile(code).disasm # 0000 trace 1 ( 2) # 0002 getinlinecache 9, # 0005 getconstant :Set # 0007 setinlinecache # 0009 opt_send_without_block , # 0012 setlocal_OP__WC__0 2 # 0014 trace 1 ( 3) # 0016 getlocal_OP__WC__0 2 # 0018 putobject :bar # 0020 opt_send_without_block , # 0023 leave

Slide 73

Slide 73 text

No content

Slide 74

Slide 74 text

insns.def

Slide 75

Slide 75 text

/** @c optimize @e [] @j 最適化された recv[obj]。 */ DEFINE_INSN opt_aref (CALL_INFO ci, CALL_CACHE cc) (VALUE recv, VALUE obj) (VALUE val) { if (!SPECIAL_CONST_P(recv)) { if (RBASIC_CLASS(recv) == rb_cArray && BASIC_OP_UNREDEFINED_P(BOP_AREF, ARRAY_REDEFINED_OP_FLAG) && FIXNUM_P(obj)) { val = rb_ary_entry(recv, FIX2LONG(obj)); } else if (RBASIC_CLASS(recv) == rb_cHash && BASIC_OP_UNREDEFINED_P(BOP_AREF, HASH_REDEFINED_OP_FLAG)) { val = rb_hash_aref(recv, obj); } else { goto INSN_LABEL(normal_dispatch); } } else { INSN_LABEL(normal_dispatch): PUSH(recv); PUSH(obj); CALL_SIMPLE_METHOD(recv); } }

Slide 76

Slide 76 text

/** @c optimize @e [] @j 最適化された recv[obj]。 */ DEFINE_INSN opt_aref (CALL_INFO ci, CALL_CACHE cc) (VALUE recv, VALUE obj) (VALUE val) { if (!SPECIAL_CONST_P(recv)) { if (RBASIC_CLASS(recv) == rb_cArray && BASIC_OP_UNREDEFINED_P(BOP_AREF, ARRAY_REDEFINED_OP_FLAG) && FIXNUM_P(obj)) { val = rb_ary_entry(recv, FIX2LONG(obj)); } else if (RBASIC_CLASS(recv) == rb_cHash && BASIC_OP_UNREDEFINED_P(BOP_AREF, HASH_REDEFINED_OP_FLAG)) { val = rb_hash_aref(recv, obj); } else { goto INSN_LABEL(normal_dispatch); } } else { INSN_LABEL(normal_dispatch): PUSH(recv); PUSH(obj); CALL_SIMPLE_METHOD(recv); } }

Slide 77

Slide 77 text

Versus

Slide 78

Slide 78 text

/** @c optimize @e Invoke method without block @j Invoke method without block */ DEFINE_INSN opt_send_without_block (CALL_INFO ci, CALL_CACHE cc) (...) (VALUE val) // inc += -ci->orig_argc; { struct rb_calling_info calling; calling.blockptr = NULL; vm_search_method(ci, cc, calling.recv = TOPN(calling.argc = ci->orig_argc)); CALL_METHOD(&calling, ci, cc); }

Slide 79

Slide 79 text

/** @c optimize @e Invoke method without block @j Invoke method without block */ DEFINE_INSN opt_send_without_block (CALL_INFO ci, CALL_CACHE cc) (...) (VALUE val) // inc += -ci->orig_argc; { struct rb_calling_info calling; calling.blockptr = NULL; vm_search_method(ci, cc, calling.recv = TOPN(calling.argc = ci->orig_argc)); CALL_METHOD(&calling, ci, cc); }

Slide 80

Slide 80 text

Ruby optimizes Hash calls by Skipping Method Lookup

Slide 81

Slide 81 text

BTW

Slide 82

Slide 82 text

Don’t subclass Hash

Slide 83

Slide 83 text

/** @c optimize @e [] @j 最適化された recv[obj]。 */ DEFINE_INSN opt_aref (CALL_INFO ci, CALL_CACHE cc) (VALUE recv, VALUE obj) (VALUE val) { if (!SPECIAL_CONST_P(recv)) { if (RBASIC_CLASS(recv) == rb_cArray && BASIC_OP_UNREDEFINED_P(BOP_AREF, ARRAY_REDEFINED_OP_FLAG) && FIXNUM_P(obj)) { val = rb_ary_entry(recv, FIX2LONG(obj)); } else if (RBASIC_CLASS(recv) == rb_cHash && BASIC_OP_UNREDEFINED_P(BOP_AREF, HASH_REDEFINED_OP_FLAG)) { val = rb_hash_aref(recv, obj); } else { goto INSN_LABEL(normal_dispatch); } } else { INSN_LABEL(normal_dispatch): PUSH(recv); PUSH(obj); CALL_SIMPLE_METHOD(recv); } } You lose speed

Slide 84

Slide 84 text


 “I don’t subclass hash”

Slide 85

Slide 85 text

Hash With Indifferent Access

Slide 86

Slide 86 text

Hashie

Slide 87

Slide 87 text

Rack:: Utils::
 HeaderHash

Slide 88

Slide 88 text

No content

Slide 89

Slide 89 text

Rack is 23% faster without HeaderHash

Slide 90

Slide 90 text

Don’t subclass Hash

Slide 91

Slide 91 text

No content

Slide 92

Slide 92 text

Switch Set to a hash

Slide 93

Slide 93 text

VALID_METADATA_VALUE_TYPES_HASH = VALID_METADATA_VALUE_TYPES. each_with_object({}) do |type, hash| hash[type] = true end.freeze def valid_processor_metadata_value?(value) if VALID_METADATA_VALUE_TYPES_HASH[value.class] true elsif VALID_METADATA_COMPOUND_TYPES_HASH[value.class] value.all? { |v| valid_processor_metadata_value?(v) } else false end end

Slide 94

Slide 94 text

Skips Method Lookup

Slide 95

Slide 95 text

Did it help?

Slide 96

Slide 96 text

18.325s 17.981s

Slide 97

Slide 97 text

1.8% Faster!

Slide 98

Slide 98 text

Bugger

Slide 99

Slide 99 text

No content

Slide 100

Slide 100 text

Keep gliding

Slide 101

Slide 101 text

Round 2

Slide 102

Slide 102 text

TOTAL (pct) SAMPLES (pct) FRAME 2328 (109.6%) 362 (17.0%) Sprockets::ProcessorUtils# valid_processor_metadata_value? 348 (16.4%) 256 (12.1%) Sprockets::DigestUtils#digest 486 (22.9%) 106 (5.0%) Kernel#require 97 (4.6%) 97 (4.6%) ActiveSupport:: NumericWithFormat#to_s 123 (5.8%) 94 (4.4%) Sprockets::PathUtils#atomic_write 581 (27.4%) 76 (3.6%) Kernel#require 61 (2.9%) 61 (2.9%) # .mechanism 193 (9.1%) 52 (2.4%) Sprockets::Cache::FileStore#set 95 (4.5%) 48 (2.3%) SassC::Rails::Importer# imports 36 (1.7%) 36 (1.7%) ExecJS::ExternalRuntime# exec_runtime 59 (2.8%) 25 (1.2%) Kernel#require 75 (3.5%) 25 (1.2%) Module#delegate

Slide 103

Slide 103 text

$ stackprof tmp/stackprof.dump --method Sprockets::DigestUtils#digest # . . . Sprockets::DigestUtils#digest (lib/sprockets/digest_utils.rb:46) samples: 4 self (0.2%) / 7 total (0.3%) callers: 5 ( 71.4%) Sprockets::Cache#expand_key 2 ( 28.6%) Sprockets::Loader#load_from_unloaded callees (3 total): 3 ( 100.0%) Sprockets::DigestUtils#digest_class code: | 46 | def digest(obj) 4 (0.2%) / 1 (0.0%) | 47 | digest = digest_class.new | 48 | queue = [obj] | 49 | | 50 | while queue.length > 0 | 51 | obj = queue.shift | 52 | klass = obj.class | 53 | 2 (0.1%) / 2 (0.1%) | 54 | if klass == String | 55 | digest << obj | 56 | elsif klass == Symbol | 57 | digest << 'Symbol' | 58 | digest << obj.to_s | 59 | elsif klass == Fixnum

Slide 104

Slide 104 text

$ stackprof tmp/stackprof.dump --method Sprockets::DigestUtils#digest # . . . Sprockets::DigestUtils#digest (lib/sprockets/digest_utils.rb:46) samples: 4 self (0.2%) / 7 total (0.3%) callers: 5 ( 71.4%) Sprockets::Cache#expand_key 2 ( 28.6%) Sprockets::Loader#load_from_unloaded callees (3 total): 3 ( 100.0%) Sprockets::DigestUtils#digest_class code: | 46 | def digest(obj) 4 (0.2%) / 1 (0.0%) | 47 | digest = digest_class.new | 48 | queue = [obj] | 49 | | 50 | while queue.length > 0 | 51 | obj = queue.shift | 52 | klass = obj.class | 53 | 2 (0.1%) / 2 (0.1%) | 54 | if klass == String | 55 | digest << obj | 56 | elsif klass == Symbol | 57 | digest << 'Symbol' | 58 | digest << obj.to_s | 59 | elsif klass == Fixnum

Slide 105

Slide 105 text

def digest(obj) digest = digest_class.new queue = [obj] while queue.length > 0 obj = queue.shift klass = obj.class if klass == String digest << obj elsif klass == Symbol digest << 'Symbol' digest << obj.to_s elsif klass == Fixnum digest << 'Fixnum' digest << obj.to_s elsif klass == Bignum digest << 'Bignum' digest << obj.to_s elsif klass == TrueClass digest << 'TrueClass' elsif klass == FalseClass digest << 'FalseClass'

Slide 106

Slide 106 text

Let me paraphrase

Slide 107

Slide 107 text

if String

Slide 108

Slide 108 text

if String elsif Symbol

Slide 109

Slide 109 text

if String elsif Symbol elsif Fixnum

Slide 110

Slide 110 text

if String elsif Symbol elsif Fixnum elsif Bignum

Slide 111

Slide 111 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass

Slide 112

Slide 112 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass

Slide 113

Slide 113 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass elsif NilClass

Slide 114

Slide 114 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass elsif NilClass elsif Array

Slide 115

Slide 115 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass elsif NilClass elsif Array elsif Hash

Slide 116

Slide 116 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass elsif NilClass elsif Array elsif Hash elsif Set

Slide 117

Slide 117 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass elsif NilClass elsif Array elsif Hash elsif Set elsif Encoding

Slide 118

Slide 118 text

If we pass in a Set object, we must make 10 comparisons

Slide 119

Slide 119 text

if String elsif Symbol elsif Fixnum elsif Bignum elsif TrueClass elsif FalseClass elsif NilClass elsif Array elsif Hash elsif Set elsif Encoding Expand and Iterate

Slide 120

Slide 120 text

if/elsif is hidden iteration

Slide 121

Slide 121 text

How do we go faster?

Slide 122

Slide 122 text

Get rid of iteration

Slide 123

Slide 123 text

Case statements

Slide 124

Slide 124 text

or

Slide 125

Slide 125 text

Hash loOkups

Slide 126

Slide 126 text

Before

Slide 127

Slide 127 text

def digest(obj) digest = digest_class.new queue = [obj] while queue.length > 0 obj = queue.shift klass = obj.class if klass == String digest << obj elsif klass == Symbol digest << 'Symbol' digest << obj.to_s elsif klass == Fixnum digest << 'Fixnum' digest << obj.to_s elsif klass == Bignum digest << 'Bignum' digest << obj.to_s elsif klass == TrueClass digest << 'TrueClass' elsif klass == FalseClass digest << 'FalseClass' elsif klass == NilClass digest << 'NilClass'.freeze elsif klass == Array digest << 'Array' queue.concat(obj) elsif klass == Hash digest << 'Hash' queue.concat(obj.sort) elsif klass == Set digest << 'Set' queue.concat(obj.to_a) elsif klass == Encoding digest << 'Encoding' digest << obj.name else raise TypeError, "couldn't digest #{klass}" end end digest.digest end

Slide 128

Slide 128 text

After

Slide 129

Slide 129 text

def digest(obj) digest = digest_class.new ADD_VALUE_TO_DIGEST[obj.class].call(obj, digest) digest.digest end

Slide 130

Slide 130 text

Store logic in a Hash

Slide 131

Slide 131 text

def digest(obj) digest = digest_class.new ADD_VALUE_TO_DIGEST[obj.class].call(obj, digest) digest.digest end Constant time lookup

Slide 132

Slide 132 text

Logic Lives in Lambdas

Slide 133

Slide 133 text

ADD_VALUE_TO_DIGEST = { String => ->(val, digest) { digest << val }, FalseClass => ->(val, digest) { digest << 'FalseClass'.freeze }, TrueClass => ->(val, digest) { digest << 'TrueClass'.freeze }, NilClass => ->(val, digest) { digest << 'NilClass'.freeze }, Symbol => ->(val, digest) { digest << 'Symbol'.freeze digest << val.to_s }, Integer => ->(val, digest) { digest << 'Integer'.freeze digest << val.to_s }, Array => ->(val, digest) { digest << 'Array'.freeze val.each do |element| ADD_VALUE_TO_DIGEST[element.class].call(element, digest) end },

Slide 134

Slide 134 text

ADD_VALUE_TO_DIGEST = { String => ->(val, digest) { digest << val }, FalseClass => ->(val, digest) { digest << 'FalseClass'.freeze }, TrueClass => ->(val, digest) { digest << 'TrueClass'.freeze }, NilClass => ->(val, digest) { digest << 'NilClass'.freeze }, Symbol => ->(val, digest) { digest << 'Symbol'.freeze digest << val.to_s }, Integer => ->(val, digest) { digest << 'Integer'.freeze digest << val.to_s }, Array => ->(val, digest) { digest << 'Array'.freeze val.each do |element| ADD_VALUE_TO_DIGEST[element.class].call(element, digest) end },

Slide 135

Slide 135 text

ADD_VALUE_TO_DIGEST = { String => ->(val, digest) { digest << val }, FalseClass => ->(val, digest) { digest << 'FalseClass'.freeze }, TrueClass => ->(val, digest) { digest << 'TrueClass'.freeze }, NilClass => ->(val, digest) { digest << 'NilClass'.freeze }, Symbol => ->(val, digest) { digest << 'Symbol'.freeze digest << val.to_s }, Integer => ->(val, digest) { digest << 'Integer'.freeze digest << val.to_s }, Array => ->(val, digest) { digest << 'Array'.freeze val.each do |element| ADD_VALUE_TO_DIGEST[element.class].call(element, digest) end },

Slide 136

Slide 136 text

ADD_VALUE_TO_DIGEST = { String => ->(val, digest) { digest << val }, FalseClass => ->(val, digest) { digest << 'FalseClass'.freeze }, TrueClass => ->(val, digest) { digest << 'TrueClass'.freeze }, NilClass => ->(val, digest) { digest << 'NilClass'.freeze }, Symbol => ->(val, digest) { digest << 'Symbol'.freeze digest << val.to_s }, Integer => ->(val, digest) { digest << 'Integer'.freeze digest << val.to_s }, Array => ->(val, digest) { digest << 'Array'.freeze val.each do |element| ADD_VALUE_TO_DIGEST[element.class].call(element, digest) end },

Slide 137

Slide 137 text

zOMG WAT

Slide 138

Slide 138 text

Recursive hash is Recursive

Slide 139

Slide 139 text

Faster?

Slide 140

Slide 140 text

14.9s 12.54s

Slide 141

Slide 141 text

16% faster asset compilation with no cache

Slide 142

Slide 142 text

Repeat for all supported types

Slide 143

Slide 143 text

Who likes audience participation?

Slide 144

Slide 144 text

Can we go faster?

Slide 145

Slide 145 text

ADD_VALUE_TO_DIGEST[String] ADD_VALUE_TO_DIGEST[Hash] ADD_VALUE_TO_DIGEST[true] ADD_VALUE_TO_DIGEST[false] ADD_VALUE_TO_DIGEST[nil] ADD_VALUE_TO_DIGEST[Array] ADD_VALUE_TO_DIGEST[Set] ADD_VALUE_TO_DIGEST[Integer]

Slide 146

Slide 146 text

String Hash true false nil Array Set Integer All Constants

Slide 147

Slide 147 text

Hash# compare_by_identity

Slide 148

Slide 148 text

Compare object, not value

Slide 149

Slide 149 text

String Hash true false nil Array Set Integer All Constants

Slide 150

Slide 150 text

No content

Slide 151

Slide 151 text

7% speed improvement

Slide 152

Slide 152 text

Let’s slow it down

Slide 153

Slide 153 text

What did we do?

Slide 154

Slide 154 text

Identify a goal

Slide 155

Slide 155 text

No content

Slide 156

Slide 156 text

No content

Slide 157

Slide 157 text

Gather potentially useful tools

Slide 158

Slide 158 text

No content

Slide 159

Slide 159 text

No content

Slide 160

Slide 160 text

No content

Slide 161

Slide 161 text

No content

Slide 162

Slide 162 text

Iterate on the problem

Slide 163

Slide 163 text

No content

Slide 164

Slide 164 text

No content

Slide 165

Slide 165 text

No content

Slide 166

Slide 166 text

No content

Slide 167

Slide 167 text

RubyVM::InstructionSequence.compile(code).disasm

Slide 168

Slide 168 text

Shoulders of giants

Slide 169

Slide 169 text

No content

Slide 170

Slide 170 text

The more you understand about _how_ your code works, the faster you can make it

Slide 171

Slide 171 text

Find a speed buddy

Slide 172

Slide 172 text

(or buddies)

Slide 173

Slide 173 text

“C” how fast your program will go

Slide 174

Slide 174 text

gem 'sassc-rails'

Slide 175

Slide 175 text

Sassc uses libsass

Slide 176

Slide 176 text

31% faster

Slide 177

Slide 177 text

Sprockets 4 will support sassc out of the box (but not yet)

Slide 178

Slide 178 text

Don’t abandon your language for speed

Slide 179

Slide 179 text

Find the slow parts and tag team them with other languages

Slide 180

Slide 180 text

C-extensions

Slide 181

Slide 181 text

Use Rust with helix

Slide 182

Slide 182 text

Use Go with gorb

Slide 183

Slide 183 text

p.s. have you tried JRuby?

Slide 184

Slide 184 text

What about cached compile times

Slide 185

Slide 185 text

No content

Slide 186

Slide 186 text

Bootscale caches require lookups

Slide 187

Slide 187 text

Without it, the more gems on your system the LONGER `require` takes

Slide 188

Slide 188 text

6s cached compile times 2s cached compile times

Slide 189

Slide 189 text

300% faster

Slide 190

Slide 190 text

No content

Slide 191

Slide 191 text

Don’t be an asshole

Slide 192

Slide 192 text

Don’t appologize for the programing language you love

Slide 193

Slide 193 text

Set your own goals

Slide 194

Slide 194 text

Go at your own speed

Slide 195

Slide 195 text

These are the good days of our programming life

Slide 196

Slide 196 text

Live them to the fullest

Slide 197

Slide 197 text

No content

Slide 198

Slide 198 text

No content

Slide 199

Slide 199 text

No content

Slide 200

Slide 200 text

No content