From 5104a5f1878abb15f5824f8d867900c3752cf77a Mon Sep 17 00:00:00 2001 From: Kristian Larsson Date: Tue, 22 Sep 2026 22:37:34 +0200 Subject: [PATCH 1/6] Qualify benchmark container image names --- .github/languages.json | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-) diff --git a/.github/languages.json b/.github/languages.json index 40bd1689..5d8f3bf3 100644 --- a/.github/languages.json +++ b/.github/languages.json @@ -1,24 +1,24 @@ { "acton": {"compilers": ["acton"]}, "c": {"compilers": ["clang", "gcc"]}, - "chapel": {"image": "chapel/chapel:2.9.0", "compilers": ["chpl"]}, + "chapel": {"image": "docker.io/chapel/chapel:2.9.0", "compilers": ["chpl"]}, "codon": {"compilers": ["codon"]}, "cpp": {"compilers": ["clang++", "g++"]}, "crystal": {"compilers": ["crystal"]}, "csharp": {"compilers": ["dotnet:9"]}, "d": {"compilers": ["ldc2"]}, - "dart": {"image": "dart:stable", "compilers": ["dart/exe"]}, - "elixir": {"image": "elixir:1.18-otp-27", "compilers": ["elixir"]}, + "dart": {"image": "docker.io/dart:stable", "compilers": ["dart/exe"]}, + "elixir": {"image": "docker.io/elixir:1.18-otp-27", "compilers": ["elixir"]}, "fortran": {"compilers": ["gfortran"]}, - "go": {"image": "golang:1.26.8", "compilers": ["go"]}, - "hacklang": {"image": "hhvm/hhvm:latest", "compilers": ["hhvm"]}, + "go": {"image": "docker.io/golang:1.26.8", "compilers": ["go"]}, + "hacklang": {"image": "docker.io/hhvm/hhvm:latest", "compilers": ["hhvm"]}, "hare": {"compilers": ["hare"]}, "haskell": {"compilers": ["ghc"]}, "haxe": {"compilers": ["haxe/cpp"]}, - "java": {"image": "eclipse-temurin:21-jdk-noble", "compilers": ["openjdk:21", "openjdk/zgc:21"]}, - "javascript": {"image": "node:24-bookworm", "compilers": ["node:current", "bun"]}, - "julia": {"image": "julia:1.11-bookworm", "compilers": ["julia"]}, - "kotlin": {"image": "eclipse-temurin:21-jdk-noble", "compilers": ["kotlin/jvm"]}, + "java": {"image": "docker.io/eclipse-temurin:21-jdk-noble", "compilers": ["openjdk:21", "openjdk/zgc:21"]}, + "javascript": {"image": "docker.io/node:24-bookworm", "compilers": ["node:current", "bun"]}, + "julia": {"image": "docker.io/julia:1.11-bookworm", "compilers": ["julia"]}, + "kotlin": {"image": "docker.io/eclipse-temurin:21-jdk-noble", "compilers": ["kotlin/jvm"]}, "lisp": {"compilers": ["sbcl"]}, "lua": {"compilers": ["lua", "luajit"]}, "nelua": {"compilers": ["nelua/clang"]}, @@ -32,7 +32,7 @@ "racket": {"compilers": ["racket"]}, "ruby": {"compilers": ["ruby", "ruby/yjit"]}, "rust": {"compilers": ["rustc:stable"]}, - "swift": {"image": "swift:6.1-noble", "compilers": ["swift"]}, + "swift": {"image": "docker.io/swift:6.1-noble", "compilers": ["swift"]}, "typescript": {"compilers": ["deno"]}, "v": {"compilers": ["v/clang+gc"]}, "wasm": {"compilers": ["wasmtime"]}, From 159aff82a23635537425324b536464502374995a Mon Sep 17 00:00:00 2001 From: Kristian Larsson Date: Tue, 22 Sep 2026 22:37:34 +0200 Subject: [PATCH 2/6] Build C benchmarks for the runner CPU --- bench/include/c/app_ffi.rsp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/bench/include/c/app_ffi.rsp b/bench/include/c/app_ffi.rsp index 5a85682a..33931687 100644 --- a/bench/include/c/app_ffi.rsp +++ b/bench/include/c/app_ffi.rsp @@ -1 +1 @@ - -pipe -O3 -fomit-frame-pointer -march=broadwell -fopenmp -pthread -Wno-deprecated-declarations -mno-fma -o app app.c -lm -lcrypto + -pipe -O3 -fomit-frame-pointer -march=native -fopenmp -pthread -Wno-deprecated-declarations -mno-fma -o app app.c -lm -lcrypto From da556474559b903f35d272d1ff76cb909b188014 Mon Sep 17 00:00:00 2001 From: Kristian Larsson Date: Tue, 22 Sep 2026 22:37:34 +0200 Subject: [PATCH 3/6] Keep four lanes in C# benchmark vectors The native vector width can be two doubles, which drops the z coordinate in nbody and breaks Mandelbrot lane access. Use fixed 256-bit vectors so both programs retain all four lanes. --- bench/algorithm/mandelbrot/2.cs | 34 ++++++++++++++++----------------- bench/algorithm/nbody/9.cs | 18 ++++++++--------- 2 files changed, 26 insertions(+), 26 deletions(-) diff --git a/bench/algorithm/mandelbrot/2.cs b/bench/algorithm/mandelbrot/2.cs index c3336356..2728595c 100644 --- a/bench/algorithm/mandelbrot/2.cs +++ b/bench/algorithm/mandelbrot/2.cs @@ -1,12 +1,12 @@ using System; using System.Linq; -using System.Numerics; +using System.Runtime.Intrinsics; using System.Security.Cryptography; using System.Text; public class MandelBrot { - private static readonly Vector _threshold = new Vector(4); + private static readonly Vector256 _threshold = Vector256.Create(4.0); public static void Main(string[] args) { var size = args.Length == 0 ? 200 : int.Parse(args[0]); @@ -15,7 +15,7 @@ public static void Main(string[] args) var inv = 2.0 / size; Console.WriteLine($"P4\n{size} {size}"); - var xloc = new (Vector, Vector)[chunkSize]; + var xloc = new (Vector256, Vector256)[chunkSize]; Span array = stackalloc double[8]; for (var i = 0; i < chunkSize; i++) { @@ -24,7 +24,7 @@ public static void Main(string[] args) { array[j] = (offset + j) * inv - 1.5; } - xloc[i] = (new Vector(array.Slice(0, 4)), new Vector(array.Slice(4, 4))); + xloc[i] = (Vector256.Create((ReadOnlySpan)array.Slice(0, 4)), Vector256.Create((ReadOnlySpan)array.Slice(4, 4))); } var data = new byte[size * chunkSize]; @@ -47,19 +47,19 @@ public static void Main(string[] args) Console.WriteLine(ToHexString(hash)); } - static byte mbrot8((Vector, Vector) cr, double civ) + static byte mbrot8((Vector256, Vector256) cr, double civ) { - var ci = new Vector(new[] { civ, civ, civ, civ }); - var zr0 = new Vector(0); - var zr1 = new Vector(0); - var zi0 = new Vector(0); - var zi1 = new Vector(0); - var tr0 = new Vector(0); - var tr1 = new Vector(0); - var ti0 = new Vector(0); - var ti1 = new Vector(0); - var absz0 = new Vector(0); - var absz1 = new Vector(0); + var ci = Vector256.Create(civ); + var zr0 = Vector256.Zero; + var zr1 = Vector256.Zero; + var zi0 = Vector256.Zero; + var zi1 = Vector256.Zero; + var tr0 = Vector256.Zero; + var tr1 = Vector256.Zero; + var ti0 = Vector256.Zero; + var ti1 = Vector256.Zero; + var absz0 = Vector256.Zero; + var absz1 = Vector256.Zero; for (var _i = 0; _i < 10; _i++) { for (var _j = 0; _j < 5; _j++) @@ -80,7 +80,7 @@ static byte mbrot8((Vector, Vector) cr, double civ) } absz0 = tr0 + ti0; absz1 = tr1 + ti1; - if (Vector.GreaterThanAll(absz0, _threshold) && Vector.GreaterThanAll(absz1, _threshold)) + if (Vector256.GreaterThanAll(absz0, _threshold) && Vector256.GreaterThanAll(absz1, _threshold)) { return 0; } diff --git a/bench/algorithm/nbody/9.cs b/bench/algorithm/nbody/9.cs index 4b5c89af..e2af4067 100644 --- a/bench/algorithm/nbody/9.cs +++ b/bench/algorithm/nbody/9.cs @@ -10,7 +10,7 @@ modified by hanabi1224 to use simd-powered Vector */ using System; - using System.Numerics; + using System.Runtime.Intrinsics; public class NBody { @@ -30,14 +30,14 @@ public static void Main(String[] args) public class Body { - public Vector Pos { get; set; } - public Vector Velocity { get; set; } + public Vector256 Pos { get; set; } + public Vector256 Velocity { get; set; } public double Mass { get; } public Body(double x, double y, double z, double vx, double vy, double vz, double mass) { - Pos = new Vector(new[] { x, y, z, 0 }); - Velocity = new Vector(new[] { vx, vy, vz, 0 }); + Pos = Vector256.Create(x, y, z, 0); + Velocity = Vector256.Create(vx, vy, vz, 0); Mass = mass; } } @@ -113,7 +113,7 @@ public NBodySystem() public void OffsetMomentum() { - var p = new Vector(new[] { 0.0, 0.0, 0.0, 0.0 }); + var p = Vector256.Zero; foreach (var b in _bodies) { p -= b.Velocity * b.Mass; @@ -133,7 +133,7 @@ public void Advance(double dt) { var bj = _bodies[j]; var dpos = pos - bj.Pos; - double d2 = Vector.Dot(dpos, dpos); + double d2 = Vector256.Dot(dpos, dpos); double mag = dt / (d2 * Math.Sqrt(d2)); dpos *= mag; v -= dpos * bj.Mass; @@ -150,12 +150,12 @@ public double Energy() for (int i = 0; i < bodyCount; i++) { var bi = _bodies[i]; - e += 0.5 * bi.Mass * Vector.Dot(bi.Velocity, bi.Velocity); + e += 0.5 * bi.Mass * Vector256.Dot(bi.Velocity, bi.Velocity); for (int j = i + 1; j < bodyCount; j++) { var bj = _bodies[j]; var dpos = bi.Pos - bj.Pos; - e -= (bi.Mass * bj.Mass) / Math.Sqrt(Vector.Dot(dpos, dpos)); + e -= (bi.Mass * bj.Mass) / Math.Sqrt(Vector256.Dot(dpos, dpos)); } } return e; From b36a374c060d3fdaff405fae9ad2a128cc20035a Mon Sep 17 00:00:00 2001 From: Kristian Larsson Date: Tue, 22 Sep 2026 22:37:34 +0200 Subject: [PATCH 4/6] Yield between TypeScript sieve stages Calling upstream generators exhausts the stack at large inputs. Yield before each request so the coroutine chain resumes without recursing through every stage on the same call stack. --- bench/algorithm/coro-prime-sieve/1.ts | 2 ++ 1 file changed, 2 insertions(+) diff --git a/bench/algorithm/coro-prime-sieve/1.ts b/bench/algorithm/coro-prime-sieve/1.ts index 3e3fc1c3..3714dc16 100644 --- a/bench/algorithm/coro-prime-sieve/1.ts +++ b/bench/algorithm/coro-prime-sieve/1.ts @@ -12,6 +12,8 @@ async function* generate() { async function* filter(ch: AsyncGenerator, prime: number) { while (true) { + // Yield before calling upstream so long chains do not exhaust the stack. + await Promise.resolve(); var i = (await ch.next()).value; if (i % prime != 0) { yield i; From 4bbfc1c2644a84416e187bcbcd6ee603ce70722d Mon Sep 17 00:00:00 2001 From: Kristian Larsson Date: Tue, 22 Sep 2026 23:05:15 +0200 Subject: [PATCH 5/6] Publish results when some languages fail One failed language blocks every result from reaching Pages. Publish verified languages and show which ones are missing. Keep failed jobs visible and reject empty or mixed-run results. --- .github/suite.py | 33 +++++++++++++++++++++++++++++- .github/workflows/bench.yml | 8 +++++--- README.md | 26 ++++++++++++++---------- bench/tests/test_suite.py | 40 ++++++++++++++++++++++++++++++++++++- website/custom.d.ts | 11 ++++++++++ website/layouts/default.vue | 26 +++++++++++++++++++++++- website/nuxt.config.ts | 8 ++++++++ website/pages/index.vue | 12 ++++++----- 8 files changed, 143 insertions(+), 21 deletions(-) diff --git a/.github/suite.py b/.github/suite.py index dc7b5f78..d3229fb4 100644 --- a/.github/suite.py +++ b/.github/suite.py @@ -1,5 +1,5 @@ #!/usr/bin/env python3 -"""Plan the suite and reject incomplete benchmark artifacts.""" +"""Plan the suite and publish verified results from successful languages.""" import json import math import os @@ -130,6 +130,35 @@ def plan(): print(f'{key}={value}', file=out) +def collect_results(): + results = BENCH / 'build/_results' + available = sorted(p.name for p in results.iterdir() if p.is_dir()) if results.exists() else [] + if set(available) - LANGUAGES.keys(): + raise ValueError(f'Unexpected result languages: {set(available) - LANGUAGES.keys()}') + if not available: + raise ValueError('No successful languages to publish') + machines = set() + for language in available: + machines.update(verify(language, 'results')) + if len(machines) != 1: + raise ValueError('Results from different machines cannot be published together') + cpu, runner = next(iter(machines)) + missing = sorted(LANGUAGES.keys() - set(available)) + summary = dict(expectedLanguages=sorted(LANGUAGES), publishedLanguages=available, + missingLanguages=missing, cpuInfo=cpu, runnerName=runner, + githubSha=os.environ['GITHUB_SHA'], githubRunId=os.environ['GITHUB_RUN_ID'], + githubRepository=os.environ['GITHUB_REPOSITORY']) + (BENCH / 'build/run-summary.json').write_text(json.dumps(summary, indent=2) + '\n') + message = f'Publishing results for {len(available)} of {len(LANGUAGES)} languages.' + if missing: + message += f' Unavailable in this run: {", ".join(missing)}.' + print(message) + if os.environ.get('GITHUB_STEP_SUMMARY'): + with open(os.environ['GITHUB_STEP_SUMMARY'], 'a') as out: + print(message, file=out) + return summary + + if __name__ == '__main__': configured = {config['lang'] for _, config in CONFIGS} if LANGUAGES.keys() != configured: @@ -139,6 +168,8 @@ def plan(): command = sys.argv[1] if command == 'plan': plan() + elif command == 'collect-results': + collect_results() elif command == 'list': for language in LANGUAGES: print(f'{language}: {len(programs(language))} programs') diff --git a/.github/workflows/bench.yml b/.github/workflows/bench.yml index 88b07655..5ba49151 100644 --- a/.github/workflows/bench.yml +++ b/.github/workflows/bench.yml @@ -86,7 +86,7 @@ jobs: results: needs: [plan, language] - if: needs.plan.outputs.publish == 'true' + if: ${{ !cancelled() && needs.plan.outputs.publish == 'true' }} runs-on: ubuntu-24.04 steps: - uses: actions/checkout@v4 @@ -96,20 +96,22 @@ jobs: pattern: benchmark-results-* merge-multiple: true path: bench/build - - name: Require complete results from this revision and run - run: /usr/bin/python3 .github/suite.py verify-results all + - name: Verify successful languages and record missing results + run: /usr/bin/python3 .github/suite.py collect-results - uses: actions/upload-artifact@v4 with: name: benchmark-results path: | bench/build/_results/ bench/build/environment-*.txt + bench/build/run-summary.json if-no-files-found: error retention-days: 90 overwrite: true publish: needs: results + if: ${{ !cancelled() && needs.results.result == 'success' }} permissions: contents: read pages: write diff --git a/README.md b/README.md index e7243cfe..efd83aef 100644 --- a/README.md +++ b/README.md @@ -3,7 +3,7 @@ Acton-maintained fork of [hanabi1224/Programming-Language-Benchmarks](https://github.com/hanabi1224/Programming-Language-Benchmarks). [Published results](https://actonlang.github.io/Programming-Language-Benchmarks/) -compare every upstream language with Acton. Acton implements all 18 problems; +compare upstream languages with Acton. Acton implements all 18 problems; other languages use the implementations enabled in the upstream configurations. ## Measurements @@ -24,8 +24,8 @@ Each language uses its own container with the same harness and inputs. The container runs on the host CPU without CPU or process-count quotas. The wrapper holds `~/.local/state/acton-perf.lock` throughout setup, checking, and measurement. Other measurements on the host must use that lock. A manual language selection -supports troubleshooting; only a complete run publishes the website. A new full -run cancels an older full run, while normal pushes leave it running. +supports troubleshooting; full-suite runs publish the languages that pass. +A new full run cancels an older full run, while normal pushes leave it running. Container storage is isolated in `/var/lib/acton-bench-containers`. Unused benchmark containers are removed before each job, dangling images afterwards, @@ -48,10 +48,14 @@ absolute path in the runner service environment, then restart the idle service. `.github/languages.json` selects all 39 upstream language entries and their primary toolchains, including WebAssembly. It does not select every historical compiler release or experimental backend. `.github/suite.py` requires every -selected program to build, pass correctness checks, and produce every expected -result before publication. A program that passes correctness checks but exceeds -a measurement time limit is shown as a timeout, without timing or memory values. -Crashes and incomplete repetitions fail the job. Toolchain versions, the container image ID, source +selected program in a language to build, pass correctness checks, and produce +every expected result before publishing that language. A program that passes +correctness checks but exceeds a measurement time limit is shown as a timeout, +without timing or memory values. +Crashes and incomplete repetitions fail the language job. Other languages can +still publish. The website lists unavailable languages and links to the run; +the combined artifact includes this inventory in `run-summary.json`. +Toolchain versions, the container image ID, source revision, and Actions run are recorded with the results. V uses its garbage-collected backend; the experimental autofree backend corrupts @@ -116,13 +120,15 @@ NODE_OPTIONS=--openssl-legacy-provider SITE_BASE_PATH=/Programming-Language-Benc The generated site is in `website/dist`. `SITE_BASE_PATH` defaults to `/` for local development. All internal links and assets respect the configured prefix. The checked-in upstream data is only for website development and PR build checks; -publishing replaces it completely with results from the successful benchmark run. +publishing replaces it completely with verified results from the current run. The `bench` workflow measures weekly and on manual dispatch. Its `publish` job calls `site.yml`, which downloads that run's results, builds the website on a GitHub-hosted runner, and deploys it to GitHub Pages. Configure the -repository's Pages publishing source as GitHub Actions. A failed build or -measurement job leaves the previous published site in place. +repository's Pages publishing source as GitHub Actions. Failed language jobs +remain red in Actions while verified languages publish. If no language passes, +result verification fails, or the run is cancelled, the previous site stays in +place. ## Attribution diff --git a/bench/tests/test_suite.py b/bench/tests/test_suite.py index c1f0a1ef..cb897e36 100644 --- a/bench/tests/test_suite.py +++ b/bench/tests/test_suite.py @@ -26,9 +26,11 @@ def setUp(self): buildLog={'finished': 'today'}, testLog={'finished': 'today'}, githubSha='current', githubRunId='123', githubRepository='owner/repo') for mock in [patch.object(suite, 'BENCH', self.root), + patch.object(suite, 'LANGUAGES', {'fixture': {}, 'missing': {}}), patch.object(suite, 'programs', return_value={'fixture': ('sample', '1.sh', 'sh', '1')}), patch.object(suite, 'PROBLEMS', {'sample': {'tests': [{'input': 'small'}, {'input': 'large'}]}}), - patch.dict(os.environ, GITHUB_SHA='current', GITHUB_RUN_ID='123', GITHUB_REPOSITORY='owner/repo')]: + patch.dict(os.environ, GITHUB_SHA='current', GITHUB_RUN_ID='123', + GITHUB_REPOSITORY='owner/repo', GITHUB_STEP_SUMMARY='')]: mock.start() self.addCleanup(mock.stop) self.write_records() @@ -76,6 +78,42 @@ def test_ignored_build_failure_is_rejected(self): with self.assertRaisesRegex(ValueError, 'Missing build output'): suite.verify('fixture', 'build') + def test_failed_language_does_not_block_verified_results(self): + summary = suite.collect_results() + self.assertEqual(summary['publishedLanguages'], ['fixture']) + self.assertEqual(summary['missingLanguages'], ['missing']) + self.assertEqual(summary['githubRunId'], '123') + self.assertEqual(json.loads((self.root / 'build/run-summary.json').read_text()), summary) + + def test_partial_artifact_is_still_rejected(self): + (self.results / 'fixture_large.json').unlink() + with self.assertRaisesRegex(ValueError, 'missing='): + suite.collect_results() + self.assertFalse((self.root / 'build/run-summary.json').exists()) + + def test_partial_publication_rejects_stale_results(self): + self.record['githubRunId'] = 'old run' + self.write_records() + with self.assertRaisesRegex(ValueError, 'Wrong githubRunId'): + suite.collect_results() + + def test_no_results_cannot_replace_the_site(self): + for path in self.results.iterdir(): + path.unlink() + self.results.rmdir() + with self.assertRaisesRegex(ValueError, 'No successful languages'): + suite.collect_results() + + def test_publication_rejects_different_language_machines(self): + other = self.results.parent / 'missing' + other.mkdir() + for path in self.results.iterdir(): + record = json.loads(path.read_text()) + record.update(lang='missing', runnerName='another runner') + (other / path.name).write_text(json.dumps(record)) + with self.assertRaisesRegex(ValueError, 'different machines'): + suite.collect_results() + if __name__ == '__main__': unittest.main() diff --git a/website/custom.d.ts b/website/custom.d.ts index 89e97fda..50a48b0f 100644 --- a/website/custom.d.ts +++ b/website/custom.d.ts @@ -5,6 +5,17 @@ declare module '*.vue' { type osType = 'linux' | 'osx' | 'windows' +type BenchmarkRun = { + expectedLanguages: string[] + publishedLanguages: string[] + missingLanguages: string[] + runnerName: string + cpuInfo: string + githubRepository: string + githubRunId: string + githubSha: string +} + type BenchResult = { cpuInfo: string lang: string diff --git a/website/layouts/default.vue b/website/layouts/default.vue index 3b56527d..23d4981c 100644 --- a/website/layouts/default.vue +++ b/website/layouts/default.vue @@ -8,6 +8,19 @@ Benchmarks +
+

+ Results for {{ benchmarkRun.publishedLanguages.length }} of + {{ benchmarkRun.expectedLanguages.length }} languages from + this run + on {{ benchmarkRun.runnerName }}. +

+

+ Unavailable in this run: + {{ benchmarkRun.missingLanguages.join(', ') }}. Follow the run link for + failure details. +

+
@@ -17,5 +30,16 @@ import { Component, Vue } from 'nuxt-property-decorator' @Component({ components: {}, }) -export default class DefaultLayout extends Vue {} +export default class DefaultLayout extends Vue { + get benchmarkRun(): BenchmarkRun | null { + return this.$config.benchmarkRun || null + } + + get runUrl(): string { + const run = this.benchmarkRun + return run + ? `https://github.com/${run.githubRepository}/actions/runs/${run.githubRunId}` + : '' + } +} diff --git a/website/nuxt.config.ts b/website/nuxt.config.ts index f3d9a81b..90128411 100644 --- a/website/nuxt.config.ts +++ b/website/nuxt.config.ts @@ -1,14 +1,22 @@ +import { existsSync, readFileSync } from 'fs' +import { resolve } from 'path' import { NuxtConfig } from '@nuxt/types' import { $content } from '@nuxt/content' import _ from 'lodash' import { getLangBenchResults } from './contentUtils' const basePath = process.env.SITE_BASE_PATH || '/' +const runSummaryPath = resolve(__dirname, '../bench/build/run-summary.json') const config: NuxtConfig = { // Target: https://go.nuxtjs.dev/config-target target: 'static', ssr: true, + publicRuntimeConfig: { + benchmarkRun: existsSync(runSummaryPath) + ? JSON.parse(readFileSync(runSummaryPath, 'utf8')) + : null, + }, loading: { color: 'cyan', }, diff --git a/website/pages/index.vue b/website/pages/index.vue index 6132c77c..bc4afe23 100644 --- a/website/pages/index.vue +++ b/website/pages/index.vue @@ -27,13 +27,15 @@ languages and their different compilers or runtime

- All benchmark programs are measured sequentially in one CI job. Each - run records its machine and compiler versions. Compare numbers - within the same run, since GitHub-hosted hardware can change. + Benchmark programs are measured sequentially on a dedicated runner. + Each run records its machine and compiler versions. Compare numbers + within the same run.

- Successful benchmark runs publish this website on GitHub Pages. Runs - happen on changes to main and weekly. + Measurements run weekly or on request. Languages that pass publish + their results on GitHub Pages, even when another language fails. + Missing languages are listed above. Code changes also run + correctness checks on GitHub-hosted runners.

Main goals:

    From 473264a5e8740f8523c8836984126de3c7d52d9b Mon Sep 17 00:00:00 2001 From: Kristian Larsson Date: Tue, 22 Sep 2026 23:13:22 +0200 Subject: [PATCH 6/6] Preserve C Mandelbrot rounding GCC can use AVX-512 fused arithmetic even with -mno-fma. Disable expression contraction to keep the expected Mandelbrot output on these CPUs. --- bench/include/c/app_ffi.rsp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/bench/include/c/app_ffi.rsp b/bench/include/c/app_ffi.rsp index 33931687..36f645eb 100644 --- a/bench/include/c/app_ffi.rsp +++ b/bench/include/c/app_ffi.rsp @@ -1 +1 @@ - -pipe -O3 -fomit-frame-pointer -march=native -fopenmp -pthread -Wno-deprecated-declarations -mno-fma -o app app.c -lm -lcrypto + -pipe -O3 -fomit-frame-pointer -march=native -fopenmp -pthread -Wno-deprecated-declarations -ffp-contract=off -o app app.c -lm -lcrypto