Skip to content

Minor optimization#46

Merged
lemire merged 3 commits intomainfrom
minor_opti
Jun 24, 2024
Merged

Minor optimization#46
lemire merged 3 commits intomainfrom
minor_opti

Conversation

@lemire
Copy link
Member

@lemire lemire commented Jun 23, 2024 •
edited
Loading

This follows some initial benchmarking by @EgorBo at dotnet/runtime#103860 who reported some negative results on ARM. We are now going to include his data files in our benchmarks.

Results on AWS Graviton 3 (Neoverse V1):

Method FileName Mean Error StdDev Speed (GB/s)
SIMDUtf8ValidationRealData data/Arabic-Lipsum.utf8.txt 24,158.65 ns 171.516 ns 9.401 ns 3.38
DotnetRuntimeUtf8ValidationRealData data/Arabic-Lipsum.utf8.txt 41,360.25 ns 385.349 ns 21.122 ns 1.97
SIMDUtf8ValidationRealData data/Bogatov1069.utf8.txt 351.33 ns 1.369 ns 0.075 ns 3.04
DotnetRuntimeUtf8ValidationRealData data/Bogatov1069.utf8.txt 520.90 ns 7.581 ns 0.416 ns 2.05
SIMDUtf8ValidationRealData data/Bogatov136.utf8.txt 63.62 ns 1.739 ns 0.095 ns 2.14
DotnetRuntimeUtf8ValidationRealData data/Bogatov136.utf8.txt 72.47 ns 0.679 ns 0.037 ns 1.88
SIMDUtf8ValidationRealData data/Bogatov286.utf8.txt 118.04 ns 1.719 ns 0.094 ns 2.42
DotnetRuntimeUtf8ValidationRealData data/Bogatov286.utf8.txt 139.18 ns 3.668 ns 0.201 ns 2.05
SIMDUtf8ValidationRealData data/Bogatov527.utf8.txt 191.22 ns 1.704 ns 0.093 ns 2.76
DotnetRuntimeUtf8ValidationRealData data/Bogatov527.utf8.txt 255.25 ns 3.556 ns 0.195 ns 2.06
SIMDUtf8ValidationRealData data/Chinese-Lipsum.utf8.txt 20,654.74 ns 58.997 ns 3.234 ns 3.38
DotnetRuntimeUtf8ValidationRealData data/Chinese-Lipsum.utf8.txt 27,390.48 ns 156.597 ns 8.584 ns 2.55
SIMDUtf8ValidationRealData data/Emoji-Lipsum.utf8.txt 19,388.81 ns 112.299 ns 6.156 ns 3.38
DotnetRuntimeUtf8ValidationRealData data/Emoji-Lipsum.utf8.txt 81,125.90 ns 1,094.791 ns 60.009 ns .81
SIMDUtf8ValidationRealData data/Hebrew-Lipsum.utf8.txt 19,681.41 ns 94.898 ns 5.202 ns 3.38
DotnetRuntimeUtf8ValidationRealData data/Hebrew-Lipsum.utf8.txt 33,771.58 ns 4,770.220 ns 261.472 ns 1.97
SIMDUtf8ValidationRealData data/Hindi-Lipsum.utf8.txt 26,303.99 ns 135.452 ns 7.425 ns 3.35
DotnetRuntimeUtf8ValidationRealData data/Hindi-Lipsum.utf8.txt 53,610.39 ns 4,890.714 ns 268.077 ns 1.64
SIMDUtf8ValidationRealData data/Japanese-Lipsum.utf8.txt 20,109.93 ns 100.564 ns 5.512 ns 3.37
DotnetRuntimeUtf8ValidationRealData data/Japanese-Lipsum.utf8.txt 28,609.20 ns 76.647 ns 4.201 ns 2.37
SIMDUtf8ValidationRealData data/Korean-Lipsum.utf8.txt 19,912.58 ns 150.700 ns 8.260 ns 3.34
DotnetRuntimeUtf8ValidationRealData data/Korean-Lipsum.utf8.txt 52,203.96 ns 335.365 ns 18.382 ns 1.28
SIMDUtf8ValidationRealData data/Latin-Lipsum.utf8.txt 2,055.81 ns 3.447 ns 0.189 ns 42.29
DotnetRuntimeUtf8ValidationRealData data/Latin-Lipsum.utf8.txt 5,029.18 ns 35.681 ns 1.956 ns 17.29
SIMDUtf8ValidationRealData data/Russian-Lipsum.utf8.txt 31,369.36 ns 21.119 ns 1.158 ns 3.34
DotnetRuntimeUtf8ValidationRealData data/Russian-Lipsum.utf8.txt 110,679.04 ns 1,797.161 ns 98.508 ns .95
SIMDUtf8ValidationRealData data/arabic.utf8.txt 136,276.58 ns 4,705.752 ns 257.938 ns 3.92
DotnetRuntimeUtf8ValidationRealData data/arabic.utf8.txt 335,772.47 ns 3,591.631 ns 196.869 ns 1.59
SIMDUtf8ValidationRealData data/chinese.utf8.txt 46,978.44 ns 8,208.924 ns 449.959 ns 3.86
DotnetRuntimeUtf8ValidationRealData data/chinese.utf8.txt 87,147.05 ns 4,756.719 ns 260.732 ns 2.08
SIMDUtf8ValidationRealData data/czech.utf8.txt 41,123.35 ns 10,980.869 ns 601.898 ns 3.71
DotnetRuntimeUtf8ValidationRealData data/czech.utf8.txt 66,565.08 ns 2,311.022 ns 126.675 ns 2.29
SIMDUtf8ValidationRealData data/english.utf8.txt 17,417.32 ns 871.235 ns 47.755 ns 22.41
DotnetRuntimeUtf8ValidationRealData data/english.utf8.txt 24,774.87 ns 871.626 ns 47.777 ns 15.76
SIMDUtf8ValidationRealData data/esperanto.utf8.txt 11,975.72 ns 1,266.203 ns 69.405 ns 7.26
DotnetRuntimeUtf8ValidationRealData data/esperanto.utf8.txt 13,782.70 ns 134.035 ns 7.347 ns 6.31
SIMDUtf8ValidationRealData data/french.utf8.txt 144,430.40 ns 7,466.682 ns 409.274 ns 3.09
DotnetRuntimeUtf8ValidationRealData data/french.utf8.txt 128,505.50 ns 7,915.180 ns 433.858 ns 3.48
SIMDUtf8ValidationRealData data/german.utf8.txt 33,524.77 ns 4,730.929 ns 259.318 ns 6.14
DotnetRuntimeUtf8ValidationRealData data/german.utf8.txt 30,536.81 ns 1,116.981 ns 61.226 ns 6.74
SIMDUtf8ValidationRealData data/greek.utf8.txt 39,768.83 ns 4,038.720 ns 221.376 ns 4.56
DotnetRuntimeUtf8ValidationRealData data/greek.utf8.txt 91,992.03 ns 4,648.658 ns 254.809 ns 1.97
SIMDUtf8ValidationRealData data/hebrew.utf8.txt 55,497.71 ns 5,478.636 ns 300.303 ns 3.43
DotnetRuntimeUtf8ValidationRealData data/hebrew.utf8.txt 140,016.73 ns 9,507.891 ns 521.160 ns 1.36
SIMDUtf8ValidationRealData data/hindi.utf8.txt 93,533.46 ns 6,254.685 ns 342.840 ns 4.24
DotnetRuntimeUtf8ValidationRealData data/hindi.utf8.txt 267,074.96 ns 9,497.874 ns 520.611 ns 1.48
SIMDUtf8ValidationRealData data/japanese.utf8.txt 40,347.17 ns 4,436.762 ns 243.194 ns 4.07
DotnetRuntimeUtf8ValidationRealData data/japanese.utf8.txt 72,362.09 ns 13,707.373 ns 751.347 ns 2.27
SIMDUtf8ValidationRealData data/korean.utf8.txt 26,894.06 ns 5,179.031 ns 283.880 ns 3.64
DotnetRuntimeUtf8ValidationRealData data/korean.utf8.txt 54,747.39 ns 1,222.467 ns 67.008 ns 1.79
SIMDUtf8ValidationRealData data/persan.utf8.txt 39,748.97 ns 7,411.004 ns 406.222 ns 3.93
DotnetRuntimeUtf8ValidationRealData data/persan.utf8.txt 104,239.72 ns 453.056 ns 24.834 ns 1.50
SIMDUtf8ValidationRealData data/portuguese.utf8.txt 75,658.45 ns 4,730.071 ns 259.271 ns 3.71
DotnetRuntimeUtf8ValidationRealData data/portuguese.utf8.txt 63,024.48 ns 20,872.259 ns 1,144.079 ns 4.45
SIMDUtf8ValidationRealData data/russian.utf8.txt 108,903.30 ns 6,141.446 ns 336.633 ns 3.74
DotnetRuntimeUtf8ValidationRealData data/russian.utf8.txt 272,664.60 ns 2,284.552 ns 125.224 ns 1.49
SIMDUtf8ValidationRealData data/thai.utf8.txt 136,082.33 ns 4,574.564 ns 250.747 ns 4.36
DotnetRuntimeUtf8ValidationRealData data/thai.utf8.txt 239,407.58 ns 9,726.036 ns 533.117 ns 2.48
SIMDUtf8ValidationRealData data/turkish.utf8.txt 52,878.26 ns 6,966.967 ns 381.883 ns 3.69
DotnetRuntimeUtf8ValidationRealData data/turkish.utf8.txt 76,486.28 ns 5,998.832 ns 328.816 ns 2.55
SIMDUtf8ValidationRealData data/twitter.json 52,742.38 ns 175.435 ns 9.616 ns 11.97
DotnetRuntimeUtf8ValidationRealData data/twitter.json 72,448.82 ns 989.729 ns 54.250 ns 8.72
SIMDUtf8ValidationRealData data/vietnamese.utf8.txt 88,754.78 ns 12,962.133 ns 710.498 ns 3.59
DotnetRuntimeUtf8ValidationRealData data/vietnamese.utf8.txt 296,099.91 ns 7,875.808 ns 431.700 ns 1.08

Apple M2 results.

Method FileName Mean Error StdDev Speed (GB/s)
SIMDUtf8ValidationRealData data/Arabic-Lipsum.utf8.txt 11,399.64 ns 2,039.425 ns 111.788 ns 7.17
DotnetRuntimeUtf8ValidationRealData data/Arabic-Lipsum.utf8.txt 23,476.49 ns 1,255.384 ns 68.812 ns 3.48
SIMDUtf8ValidationRealData data/Bogatov1069.utf8.txt 179.33 ns 17.708 ns 0.971 ns 5.96
DotnetRuntimeUtf8ValidationRealData data/Bogatov1069.utf8.txt 288.72 ns 51.203 ns 2.807 ns 3.70
SIMDUtf8ValidationRealData data/Bogatov136.utf8.txt 31.83 ns 2.044 ns 0.112 ns 4.27
DotnetRuntimeUtf8ValidationRealData data/Bogatov136.utf8.txt 33.78 ns 28.276 ns 1.550 ns 4.03
SIMDUtf8ValidationRealData data/Bogatov286.utf8.txt 63.19 ns 23.075 ns 1.265 ns 4.53
DotnetRuntimeUtf8ValidationRealData data/Bogatov286.utf8.txt 68.33 ns 28.641 ns 1.570 ns 4.19
SIMDUtf8ValidationRealData data/Bogatov527.utf8.txt 97.59 ns 2.865 ns 0.157 ns 5.40
DotnetRuntimeUtf8ValidationRealData data/Bogatov527.utf8.txt 136.98 ns 7.925 ns 0.434 ns 3.85
SIMDUtf8ValidationRealData data/Chinese-Lipsum.utf8.txt 9,711.20 ns 843.972 ns 46.261 ns 7.19
DotnetRuntimeUtf8ValidationRealData data/Chinese-Lipsum.utf8.txt 14,650.94 ns 1,101.397 ns 60.371 ns 4.77
SIMDUtf8ValidationRealData data/Emoji-Lipsum.utf8.txt 9,109.92 ns 522.064 ns 28.616 ns 7.19
DotnetRuntimeUtf8ValidationRealData data/Emoji-Lipsum.utf8.txt 26,536.00 ns 14,264.855 ns 781.905 ns 2.47
SIMDUtf8ValidationRealData data/Hebrew-Lipsum.utf8.txt 9,261.09 ns 738.231 ns 40.465 ns 7.18
DotnetRuntimeUtf8ValidationRealData data/Hebrew-Lipsum.utf8.txt 19,261.95 ns 1,497.396 ns 82.077 ns 3.45
SIMDUtf8ValidationRealData data/Hindi-Lipsum.utf8.txt 12,268.32 ns 1,660.140 ns 90.998 ns 7.17
DotnetRuntimeUtf8ValidationRealData data/Hindi-Lipsum.utf8.txt 29,700.12 ns 3,732.162 ns 204.572 ns 2.96
SIMDUtf8ValidationRealData data/Japanese-Lipsum.utf8.txt 9,368.63 ns 201.405 ns 11.040 ns 7.24
DotnetRuntimeUtf8ValidationRealData data/Japanese-Lipsum.utf8.txt 14,596.26 ns 1,320.289 ns 72.369 ns 4.65
SIMDUtf8ValidationRealData data/Korean-Lipsum.utf8.txt 9,218.95 ns 299.560 ns 16.420 ns 7.22
DotnetRuntimeUtf8ValidationRealData data/Korean-Lipsum.utf8.txt 36,247.10 ns 5,763.164 ns 315.898 ns 1.84
SIMDUtf8ValidationRealData data/Latin-Lipsum.utf8.txt 991.56 ns 148.287 ns 8.128 ns 87.68
DotnetRuntimeUtf8ValidationRealData data/Latin-Lipsum.utf8.txt 2,329.52 ns 436.663 ns 23.935 ns 37.32
SIMDUtf8ValidationRealData data/Russian-Lipsum.utf8.txt 16,209.71 ns 49,686.589 ns 2,723.490 ns 6.46
DotnetRuntimeUtf8ValidationRealData data/Russian-Lipsum.utf8.txt 38,774.13 ns 15,205.809 ns 833.482 ns 2.70
SIMDUtf8ValidationRealData data/arabic.utf8.txt 82,135.62 ns 75,320.317 ns 4,128.561 ns 6.50
DotnetRuntimeUtf8ValidationRealData data/arabic.utf8.txt 213,231.93 ns 94,725.565 ns 5,192.228 ns 2.50
SIMDUtf8ValidationRealData data/chinese.utf8.txt 22,018.20 ns 14,750.411 ns 808.520 ns 8.24
DotnetRuntimeUtf8ValidationRealData data/chinese.utf8.txt 48,770.69 ns 46,322.436 ns 2,539.089 ns 3.72
SIMDUtf8ValidationRealData data/czech.utf8.txt 17,871.31 ns 23,368.427 ns 1,280.902 ns 8.55
DotnetRuntimeUtf8ValidationRealData data/czech.utf8.txt 35,975.96 ns 9,030.069 ns 494.969 ns 4.25
SIMDUtf8ValidationRealData data/english.utf8.txt 8,887.52 ns 385.530 ns 21.132 ns 43.92
DotnetRuntimeUtf8ValidationRealData data/english.utf8.txt 16,732.80 ns 1,301.154 ns 71.321 ns 23.33
SIMDUtf8ValidationRealData data/esperanto.utf8.txt 5,540.34 ns 214.626 ns 11.764 ns 15.70
DotnetRuntimeUtf8ValidationRealData data/esperanto.utf8.txt 9,494.63 ns 112.688 ns 6.177 ns 9.16
SIMDUtf8ValidationRealData data/french.utf8.txt 83,480.49 ns 29,775.365 ns 1,632.088 ns 5.35
DotnetRuntimeUtf8ValidationRealData data/french.utf8.txt 88,632.88 ns 10,779.051 ns 590.836 ns 5.04
SIMDUtf8ValidationRealData data/german.utf8.txt 16,116.36 ns 25,195.670 ns 1,381.060 ns 12.77
DotnetRuntimeUtf8ValidationRealData data/german.utf8.txt 22,503.68 ns 1,330.839 ns 72.948 ns 9.14
SIMDUtf8ValidationRealData data/greek.utf8.txt 18,711.08 ns 5,684.170 ns 311.569 ns 9.69
DotnetRuntimeUtf8ValidationRealData data/greek.utf8.txt 57,779.24 ns 12,729.790 ns 697.763 ns 3.14
SIMDUtf8ValidationRealData data/hebrew.utf8.txt 26,752.19 ns 21,639.367 ns 1,186.127 ns 7.11
DotnetRuntimeUtf8ValidationRealData data/hebrew.utf8.txt 83,893.26 ns 62,240.251 ns 3,411.598 ns 2.27
SIMDUtf8ValidationRealData data/hindi.utf8.txt 44,453.68 ns 26,985.647 ns 1,479.174 ns 8.92
DotnetRuntimeUtf8ValidationRealData data/hindi.utf8.txt 175,719.39 ns 107,454.435 ns 5,889.940 ns 2.26
SIMDUtf8ValidationRealData data/japanese.utf8.txt 19,261.51 ns 4,880.817 ns 267.534 ns 8.53
DotnetRuntimeUtf8ValidationRealData data/japanese.utf8.txt 50,006.96 ns 79,439.404 ns 4,354.342 ns 3.29
SIMDUtf8ValidationRealData data/korean.utf8.txt 10,816.74 ns 820.990 ns 45.001 ns 9.05
DotnetRuntimeUtf8ValidationRealData data/korean.utf8.txt 31,789.57 ns 8,716.355 ns 477.773 ns 3.08
SIMDUtf8ValidationRealData data/persan.utf8.txt 17,584.40 ns 8,038.834 ns 440.636 ns 8.88
DotnetRuntimeUtf8ValidationRealData data/persan.utf8.txt 49,597.62 ns 33,180.391 ns 1,818.729 ns 3.15
SIMDUtf8ValidationRealData data/portuguese.utf8.txt 32,645.46 ns 54,609.499 ns 2,993.331 ns 8.60
DotnetRuntimeUtf8ValidationRealData data/portuguese.utf8.txt 46,334.64 ns 25,046.242 ns 1,372.869 ns 6.06
SIMDUtf8ValidationRealData data/russian.utf8.txt 54,265.85 ns 42,360.008 ns 2,321.895 ns 7.50
DotnetRuntimeUtf8ValidationRealData data/russian.utf8.txt 168,228.80 ns 46,455.269 ns 2,546.370 ns 2.42
SIMDUtf8ValidationRealData data/thai.utf8.txt 68,946.68 ns 35,710.480 ns 1,957.412 ns 8.61
DotnetRuntimeUtf8ValidationRealData data/thai.utf8.txt 140,746.17 ns 97,987.595 ns 5,371.031 ns 4.22
SIMDUtf8ValidationRealData data/turkish.utf8.txt 18,900.31 ns 31,626.706 ns 1,733.566 ns 10.32
DotnetRuntimeUtf8ValidationRealData data/turkish.utf8.txt 42,712.48 ns 79,564.867 ns 4,361.219 ns 4.57
SIMDUtf8ValidationRealData data/twitter.json 25,863.13 ns 8,125.716 ns 445.398 ns 24.42
DotnetRuntimeUtf8ValidationRealData data/twitter.json 43,760.80 ns 6,249.167 ns 342.538 ns 14.43
SIMDUtf8ValidationRealData data/vietnamese.utf8.txt 44,278.79 ns 57,707.891 ns 3,163.164 ns 7.21
DotnetRuntimeUtf8ValidationRealData data/vietnamese.utf8.txt 210,026.00 ns 49,423.902 ns 2,709.091 ns 1.52

@lemire lemire requested a review from Nick-Nuon June 23, 2024 21:35
@lemire lemire mentioned this pull request Jun 24, 2024
Closed
@lemire lemire merged commit b919f83 into main Jun 24, 2024
Nick-Nuon
Nick-Nuon reviewed Jun 24, 2024
View reviewed changes
Copy link
Collaborator

@Nick-Nuon Nick-Nuon left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seen the PR and I take note of the merge!
Other than the typo , it looks good to me!

src/UTF8.cs

if (AdvSimd.Arm64.MaxAcross(currentBlock).ToScalar() <= 127)
if (AdvSimd.Arm64.MaxAcross(Vector128.AsUInt32(AdvSimd.And(currentBlock, v80))).ToScalar() == 0)
// We could it with (AdvSimd.Arm64.MaxAcross(currentBlock).ToScalar() <= 127) but it is slower on some
Copy link
Collaborator

@Nick-Nuon Nick-Nuon Jun 24, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There seems to be a very minor typo here in the comments

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@Nick-Nuon Nick-Nuon Nick-Nuon left review comments

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

2 participants

@lemire @Nick-Nuon