Skip to content

SSE4.2 Lookup Table Kernel#2

Closed
samyron wants to merge 3 commits into
arm-neon-simdfrom
simd-use-lookup-table
Closed

SSE4.2 Lookup Table Kernel#2
samyron wants to merge 3 commits into
arm-neon-simdfrom
simd-use-lookup-table

Conversation

@samyron

@samyron samyron commented Jan 24, 2025

Copy link
Copy Markdown
Owner

Benchmarks

SSE4.2 Rules Based Kernel

This is the existing implementation. It directly checks multiple conditions against a 16 byte chunk of data.

== Encoding mixed utf8 (5003001 bytes)
ruby 3.4.1 (2024-12-25 revision 48d4efcb85) +PRISM [x86_64-linux]
Warming up --------------------------------------
                json   110.000 i/100ms
                  oj    17.000 i/100ms
Calculating -------------------------------------
                json      1.112k (± 7.4%) i/s  (899.02 μs/i) -      5.610k in   5.073406s
                  oj    193.508 (± 4.7%) i/s    (5.17 ms/i) -    969.000 in   5.017873s

Comparison:
                json:     1112.3 i/s
                  oj:      193.5 i/s - 5.75x  slower


== Encoding mostly utf8 (5001001 bytes)
ruby 3.4.1 (2024-12-25 revision 48d4efcb85) +PRISM [x86_64-linux]
Warming up --------------------------------------
                json   101.000 i/100ms
                  oj    18.000 i/100ms
Calculating -------------------------------------
                json      1.140k (± 3.7%) i/s  (877.37 μs/i) -      5.757k in   5.058461s
                  oj    185.093 (± 6.5%) i/s    (5.40 ms/i) -    936.000 in   5.080750s

Comparison:
                json:     1139.8 i/s
                  oj:      185.1 i/s - 6.16x  slower

SSE4.2 Lookup Table Based Kernel

This uses the escape_table to look up the correct action for each byte in the 16 byte chunk of data. If no actions are needed it skips to the next 16 byte chunk.

== Encoding mixed utf8 (5003001 bytes)
ruby 3.4.1 (2024-12-25 revision 48d4efcb85) +PRISM [x86_64-linux]
Warming up --------------------------------------
                json    73.000 i/100ms
                  oj    18.000 i/100ms
Calculating -------------------------------------
                json    725.505 (± 4.1%) i/s    (1.38 ms/i) -      3.650k in   5.042347s
                  oj    202.460 (± 3.5%) i/s    (4.94 ms/i) -      1.026k in   5.073307s

Comparison:
                json:      725.5 i/s
                  oj:      202.5 i/s - 3.58x  slower


== Encoding mostly utf8 (5001001 bytes)
ruby 3.4.1 (2024-12-25 revision 48d4efcb85) +PRISM [x86_64-linux]
Warming up --------------------------------------
                json    70.000 i/100ms
                  oj    19.000 i/100ms
Calculating -------------------------------------
                json    708.133 (± 2.8%) i/s    (1.41 ms/i) -      3.570k in   5.045727s
                  oj    200.301 (± 2.5%) i/s    (4.99 ms/i) -      1.007k in   5.030429s

Comparison:
                json:      708.1 i/s
                  oj:      200.3 i/s - 3.54x  slower

@samyron samyron mentioned this pull request Jan 24, 2025
Closed
@samyron samyron closed this Apr 28, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@samyron