perf(ast/estree): speed up building UTF8-UTF16 translation table with SIMD #9359
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
github-actions
bot
added
A-ast
This was referenced Feb 25, 2025
How to use the Graphite Merge QueueAdd either label to this PR to merge it via the merge queue:
You must have a Graphite account in order to use the merge queue. Sign up using this link. An organization admin has enabled the Graphite Merge Queue in this repository. Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue. This stack of pull requests is managed by Graphite. Learn more about stacking. |
CodSpeed Performance ReportMerging #9359 will improve performances by 6.16%Comparing Summary
Benchmarks breakdown
|
overlookmotel
force-pushed
the
02-25-perf_ast_estree_speed_up_building_utf8-utf16_translation_table_with_simd
branch
from
ba4be9f to
6c2dd44
Compare
Merge activity
|
graphite-app
bot
force-pushed
the
02-25-perf_ast_estree_faster_utf-8_to_utf-16_span_conversion
branch
from
898ab78 to
61939ca
Compare
graphite-app bot
pushed a commit
that referenced
this pull request
Feb 26, 2025
… SIMD (#9359) Use SIMD operations to speed up building the table of UTF8-UTF16 offset translations. Previously, we looped through the source byte-by-byte, checking if each byte is ASCII, one-by-one. Instead, use SIMD to check blocks of 32 bytes at a time, and only drop down to byte-by-byte processing if the quick check does find any non-ASCII bytes in that block. Building the table is about 7x faster in our benchmark after this PR. Note: There are no explicit SIMD operations involved in the code. Just the code is written so that the compiler can auto-vectorize it. We could get even better performance if we did use SIMD proper - on x86_64 with AVX-512 or Apple M4, could process blocks of 64 bytes in one go, and with less instructions. But that involves a lot of complications, so I think this is a decent compromise for now.
graphite-app
bot
force-pushed
the
02-25-perf_ast_estree_speed_up_building_utf8-utf16_translation_table_with_simd
branch
from
6c2dd44 to
28a79f2
Compare
Base automatically changed from
02-25-perf_ast_estree_faster_utf-8_to_utf-16_span_conversion
to
main
February 26, 2025 02:18
… SIMD (#9359) Use SIMD operations to speed up building the table of UTF8-UTF16 offset translations. Previously, we looped through the source byte-by-byte, checking if each byte is ASCII, one-by-one. Instead, use SIMD to check blocks of 32 bytes at a time, and only drop down to byte-by-byte processing if the quick check does find any non-ASCII bytes in that block. Building the table is about 7x faster in our benchmark after this PR. Note: There are no explicit SIMD operations involved in the code. Just the code is written so that the compiler can auto-vectorize it. We could get even better performance if we did use SIMD proper - on x86_64 with AVX-512 or Apple M4, could process blocks of 64 bytes in one go, and with less instructions. But that involves a lot of complications, so I think this is a decent compromise for now.
overlookmotel
force-pushed
the
02-25-perf_ast_estree_speed_up_building_utf8-utf16_translation_table_with_simd
branch
from
28a79f2 to
82adab9
Compare
|
Miri passes (see #9370). I think this is good to go. |
graphite-app
bot
deleted the
02-25-perf_ast_estree_speed_up_building_utf8-utf16_translation_table_with_simd
branch
overlookmotel
added a commit
that referenced
this pull request
Feb 26, 2025
Boshen
added a commit
that referenced
this pull request
Feb 26, 2025
## [0.53.0] - 2025-02-26 - #9289 - 4a5a7cf napi/parser: [**BREAKING**] Remove magic string; enable utf16 span converter by default (#9291) (Boshen) ### Features - 5c775ea ast/estree: Enable serialization without TS fields (#9285) (overlookmotel) - f21740e data_structures: Add `CodeBuffer::print_bytes_iter_unchecked` method (#9337) (overlookmotel) - e10fb97 ecmascript: Improve may_have_side_effects for `.length` (#9366) (sapphi-red) - 35e5ca9 ecmascript: Improve may_have_side_effects for `instanceof` (#9365) (sapphi-red) - 11012c6 ecmascript: Improve ValueType for coalesce operator (#9354) (sapphi-red) - b7998fd ecmascript: To_number for object without toString (#9353) (sapphi-red) - e51d563 minifier: Concatenate strings with template literals on right side (#9356) (sapphi-red) - 9d7db54 minifier: Concatenate strings with template literals (#9355) (sapphi-red) - 835ee95 wasm: Return estree with utf16 span offsets (#9376) (Boshen) ### Bug Fixes - 6a8f53f ast/estree: Visit `JSXOpeningFragment` and `JSXClosingFragment` (#9342) (overlookmotel) - e303767 ast/estree: Fix ESTree AST for imports and exports (#9282) (overlookmotel) - 54d59f1 data_structures: Stack types correctly report allocation size if allocation failure during grow (#9317) (overlookmotel) - f5c8698 ecmascript: Correct may_have_side_effects for classes (#9367) (sapphi-red) - d3ed128 minifier: Do not remove `=== 0` if the lhs can be NaN (#9352) (sapphi-red) ### Performance - 82adab9 ast/estree: Speed up building UTF8-UTF16 translation table with SIMD (#9359) (overlookmotel) - 61939ca ast/estree: Faster UTF-8 to UTF-16 span conversion (#9349) (overlookmotel) - 1bfc459 ast/estree: Pre-allocate `CodeBuffer` for JSON output (#9340) (overlookmotel) - 018c523 ast/estree: `ESTree` serializer use `CodeBuffer` (#9331) (overlookmotel) - 35ee399 codegen: Use `iter::repeat_n` in `CodeBuffer` (#9325) (overlookmotel) ### Documentation - 8bd3e39 data_structures: Uppercase SAFETY comments (#9330) (overlookmotel) ### Refactor - d94fc15 allocator: Reduce scope of `unsafe` blocks (#9319) (overlookmotel) - 7427900 ast: Re-order `ExportDefaultDeclaration` fields (#9348) (overlookmotel) - b09249c ast/estree: Rename serializers and serialization methods (#9284) (overlookmotel) - 55ed1df ast/estree: Shorten `ESTree` impls for enums (#9275) (overlookmotel) - 9d98444 codegen, data_structures: Move `CodeBuffer` into `oxc_data_structures` crate (#9326) (overlookmotel) - 6a4e892 data_structures: Add debug assertion to `CodeBuffer::peek_nth_char_back` and improve safety docs (#9336) (overlookmotel) - fc46218 data_structures: `CodeBuffer::print_str` use `Vec::extend_from_slice` (#9332) (overlookmotel) - 690bae5 data_structures: Stack types const assert `T` is not zero-size type (#9318) (overlookmotel) - 10ba2ea data_structures: Reduce scope of `unsafe` blocks (#9316) (overlookmotel) - beb8382 data_structures: `CodeBuffer::print_bytes_unchecked` take a byte slice (#9327) (overlookmotel) - faf966f ecmascript: Don't check side effects in constant_evaluation (#9122) (sapphi-red) - 2faabe1 estree: Make `itoa` dependency optional (#9338) (overlookmotel) - 4e9e8cf lexer: Reduce scope of `unsafe` blocks (#9320) (overlookmotel) - c31b53f mangler: Reduce scope of `unsafe` blocks (#9321) (overlookmotel) - f10a6da mangler: Move base54 into seperate mod (#9278) (Cameron) - 12e89e0 syntax: Reduce scope of `unsafe` blocks (#9322) (overlookmotel) - f39be5f traverse: Reduce scope of `unsafe` blocks (#9323) (overlookmotel) Co-authored-by: Boshen <1430279+Boshen@users.noreply.github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Use SIMD operations to speed up building the table of UTF8-UTF16 offset translations.
Previously, we looped through the source byte-by-byte, checking if each byte is ASCII, one-by-one. Instead, use SIMD to check blocks of 32 bytes at a time, and only drop down to byte-by-byte processing if the quick check does find any non-ASCII bytes in that block.
Building the table is about 7x faster in our benchmark after this PR.
Note: There are no explicit SIMD operations involved in the code. Just the code is written so that the compiler can auto-vectorize it.
We could get even better performance if we did use SIMD proper - on x86_64 with AVX-512 or Apple M4, could process blocks of 64 bytes in one go, and with less instructions. But that involves a lot of complications, so I think this is a decent compromise for now.