fs: dispatch ASCII and Latin1 via simdutf in ReadFileUtf8

[mertcanaltin]

In fs.readFileSync(path, 'utf8'), I dispatch the V8 string creation through simdutf. ASCII and Latin1-fits utf-8 use one-byte V8 strings, multibyte goes through simdutf to UTF-16, invalid utf-8 falls back to V8.

@nodejs/fs, @anonrig@addaleax, @lemire@mcollina

Bench results (gist): https://gist.github.com/mertcanaltin/a0096c3fad387d0bace821938754af44

fs/readfile-utf8-fastpath.js
latin1 path 4MB +803%
latin1 path 256KB +725%
utf8_mixed path 4MB +453%
utf8_mixed path 256KB +358%
latin1 path 16KB +309%
latin1 fd 4MB +238%

[nodejs-github-bot]

Review requested:

• @nodejs/performance

[codecov]

Codecov Report

❌ Patch coverage is 57.14286% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.06%. Comparing base (9de9b9f) to head (deaa5e1).
⚠️ Report is 134 commits behind head on main.

Files with missing lines	Patch %	Lines
src/util.cc	57.14%	2 Missing and 1 partial ⚠️

Additional details and impacted files

@@ Coverage Diff @@## main #63370 +/- ##
==========================================
- Coverage 90.07% 90.06% -0.02% 
==========================================
Files 714 714 Lines 225564 225740 +176 Branches 42656 42718 +62 ==========================================
+ Hits 203177 203306 +129 - Misses 14189 14221 +32 - Partials 8198 8213 +15 

Files with missing lines	Coverage Δ
src/util-inl.h	83.38% <ø> (+0.44%)	⬆️
src/util.h	90.98% <ø> (ø)
src/util.cc	86.95% <57.14%> (-0.49%)	⬇️

... and 65 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[addaleax]

Why would we make this change specifically for fs and not as part of the general ToV8Value() conversion or StringBytes?

[mertcanaltin]

Why would we make this change specifically for fs and not as part of the general ToV8Value() conversion or StringBytes?

Sure, I applied now,

new benchmark results:

have a one regresion, I solving today,
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='utf8_mixed' size=16384 *** -9.65 % ±2.96% ±3.94% ±5.13%

➜ node git:(mert/readfileutf8-zero-copy-fast-path) ✗ node-benchmark-compare ./result.csv
confidence improvement accuracy (*) (**) (***)
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='ascii' size=1024 0.11 % ±4.46% ±5.99% ±7.89%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='ascii' size=16384 2.08 % ±5.10% ±6.80% ±8.88%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='ascii' size=262144 -3.75 % ±3.75% ±5.00% ±6.50%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='ascii' size=4194304 *** 4.98 % ±2.36% ±3.14% ±4.08%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='ascii' size=64 0.11 % ±3.45% ±4.60% ±6.00%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='latin1' size=1024 1.61 % ±4.25% ±5.68% ±7.45%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='latin1' size=16384 ** 6.43 % ±3.72% ±4.96% ±6.46%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='latin1' size=262144 *** 27.97 % ±4.38% ±5.84% ±7.63%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='latin1' size=4194304 *** 221.38 % ±4.45% ±5.97% ±7.87%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='latin1' size=64 1.45 % ±2.80% ±3.73% ±4.85%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='utf8_mixed' size=1024 3.00 % ±5.51% ±7.37% ±9.67%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='utf8_mixed' size=16384 *** -9.65 % ±2.96% ±3.94% ±5.13%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='utf8_mixed' size=262144 *** 17.65 % ±3.20% ±4.26% ±5.55%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='utf8_mixed' size=4194304 *** 126.65 % ±3.25% ±4.34% ±5.68%
fs/readfile-utf8-fastpath.js n=3000 source='fd' content='utf8_mixed' size=64 0.33 % ±4.22% ±5.64% ±7.38%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='ascii' size=1024 2.40 % ±2.93% ±3.95% ±5.23%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='ascii' size=16384 0.24 % ±4.38% ±5.83% ±7.60%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='ascii' size=262144 *** 10.85 % ±1.06% ±1.42% ±1.86%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='ascii' size=4194304 *** 27.92 % ±3.49% ±4.68% ±6.18%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='ascii' size=64 0.79 % ±2.69% ±3.62% ±4.79%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='latin1' size=1024 *** 29.48 % ±3.18% ±4.27% ±5.65%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='latin1' size=16384 *** 264.74 % ±5.63% ±7.57% ±10.00%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='latin1' size=262144 *** 575.46 % ±7.93% ±10.65% ±14.06%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='latin1' size=4194304 *** 823.55 % ±4.83% ±6.46% ±8.47%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='latin1' size=64 3.02 % ±3.61% ±4.86% ±6.44%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='utf8_mixed' size=1024 *** 18.36 % ±3.05% ±4.11% ±5.43%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='utf8_mixed' size=16384 *** 136.45 % ±2.53% ±3.37% ±4.38%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='utf8_mixed' size=262144 *** 260.71 % ±4.50% ±5.99% ±7.80%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='utf8_mixed' size=4194304 *** 330.51 % ±2.51% ±3.38% ±4.48%
fs/readfile-utf8-fastpath.js n=3000 source='path' content='utf8_mixed' size=64 3.01 % ±3.61% ±4.86% ±6.43%
Be aware that when doing many comparisons the risk of a false-positive result increases.
In this case, there are 30 comparisons, you can thus expect the following amount of false-positive results:
1.50 false positives, when considering a 5% risk acceptance (*, **, ***),
0.30 false positives, when considering a 1% risk acceptance (**, ***),
0.03 false positives, when considering a 0.1% risk acceptance (***)
➜ node git:(mert/readfileutf8-zero-copy-fast-path) ✗

[mertcanaltin]

-9.65 % on fd utf8_mixed 16 KiB is addressed in #63385 it collapses the StringBytes UTF-16 path to a single simdutf pass for buffers ≤ 1 MiB. Both PRs together remove the regression @addaleax fyi

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/73552/

[gurgunday]

Lgtm

[nodejs-github-bot]

CI: https://ci.nodejs.org/job/node-test-pull-request/73736/

[addaleax]

Landed in 770385a