webgpu-bench 0.1.0

package/dist/stats.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"stats.d.ts","sourceRoot":"","sources":["../src/stats.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,KAAK,EAAE,MAAM,YAAY,CAAC;AAYxC,iGAAiG;AACjG,wBAAgB,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,MAAM,CAG/C;AAED,wBAAgB,YAAY,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,KAAK,CAkB7D;AAED,MAAM,WAAW,iBAAiB;IAChC,kEAAkE;IAClE,OAAO,EAAE,MAAM,CAAC;IAChB,8FAA8F;IAC9F,UAAU,EAAE,MAAM,CAAC;IACnB,uGAAuG;IACvG,SAAS,EAAE,MAAM,CAAC;CACnB;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,YAAY,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,EAAE,IAAI,EAAE,iBAAiB,GAAG,OAAO,CAQxF;AAED,wHAAwH;AACxH,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,GAAG,OAAO,CAGxF"}

package/dist/stats.js

@@ -0,0 +1,67 @@
+/**
+ * Two-sided 97.5% quantiles of Student's t distribution (i.e. the multiplier
+ * for a 95% confidence interval) indexed by degrees of freedom 1..30. Beyond
+ * 30 the value is within ~4% of the normal 1.96, so the df=30 entry is reused.
+ */
+const T_975 = [
+    12.706, 4.303, 3.182, 2.776, 2.571, 2.447, 2.365, 2.306, 2.262, 2.228, 2.201, 2.179, 2.16, 2.145, 2.131, 2.12, 2.11,
+    2.101, 2.093, 2.086, 2.08, 2.074, 2.069, 2.064, 2.06, 2.056, 2.052, 2.048, 2.045, 2.042,
+];
+/** Student-t multiplier for a two-sided 95% confidence interval with `df` degrees of freedom. */
+export function tQuantile975(df) {
+    if (df < 1)
+        return Number.POSITIVE_INFINITY;
+    return T_975[Math.min(df, T_975.length) - 1];
+}
+export function computeStats(values) {
+    if (values.length === 0) {
+        throw new Error('computeStats: values must be non-empty');
+    }
+    const n = values.length;
+    const sorted = values.toSorted((a, b) => a - b);
+    const sum = values.reduce((a, b) => a + b, 0);
+    const mean = sum / n;
+    const min = sorted[0];
+    const max = sorted[n - 1];
+    const mid = Math.floor(n / 2);
+    const median = n % 2 === 0 ? (sorted[mid - 1] + sorted[mid]) / 2 : sorted[mid];
+    const variance = n > 1 ? values.reduce((acc, v) => acc + (v - mean) ** 2, 0) / (n - 1) : 0;
+    const stddev = Math.sqrt(variance);
+    // Half-width of the 95% CI on the mean. With a single sample the
+    // t-multiplier is infinite (df = 0), so the interval is unbounded.
+    const ci95 = n > 1 ? (tQuantile975(n - 1) * stddev) / Math.sqrt(n) : Number.POSITIVE_INFINITY;
+    return { mean, min, max, median, stddev, ci95 };
+}
+/**
+ * Stopping rule for best-of-N sampling.
+ *
+ * The reported statistic is the *minimum* time, since every source of noise a
+ * benchmark meets (thermal throttling, clock ramp, compositor frames, other
+ * apps) only ever makes a run slower — the fastest run is the closest thing
+ * to the device's true capability. So sampling is done once the minimum has
+ * settled: the last `stableRuns` samples failed to beat the best seen before
+ * them by more than `tolerance`. Three tight samples converge immediately
+ * (minRuns 3, stableRuns 2: sample 1 sets the best, samples 2 and 3 confirm
+ * it); a still-ramping GPU that keeps producing faster runs keeps sampling.
+ *
+ * Degenerate data (non-positive or non-finite best) never converges, so the
+ * caller runs to its cap and can then diagnose the bad timings itself.
+ */
+export function isBestStable(values, opts) {
+    const n = values.length;
+    if (n < Math.max(opts.stableRuns + 1, opts.minRuns, 2))
+        return false;
+    const best = Math.min(...values);
+    if (!Number.isFinite(best) || best <= 0)
+        return false;
+    const priorBest = Math.min(...values.slice(0, n - opts.stableRuns));
+    // Stable iff none of the trailing samples improved on the prior best by more than the tolerance.
+    return best >= priorBest * (1 - opts.tolerance);
+}
+/** True when `sampleMs` is more than `threshold` (fraction) slower than `bestMs` — the signature of a throttled run. */
+export function isThrottled(sampleMs, bestMs, threshold) {
+    if (!Number.isFinite(bestMs) || bestMs <= 0)
+        return false;
+    return sampleMs > bestMs * (1 + threshold);
+}
+//# sourceMappingURL=stats.js.map

package/dist/stats.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"stats.js","sourceRoot":"","sources":["../src/stats.ts"],"names":[],"mappings":"AAEA;;;;GAIG;AACH,MAAM,KAAK,GAAG;IACZ,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI;IACnH,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK,EAAE,KAAK;CACxF,CAAC;AAEF,iGAAiG;AACjG,MAAM,UAAU,YAAY,CAAC,EAAU;IACrC,IAAI,EAAE,GAAG,CAAC;QAAE,OAAO,MAAM,CAAC,iBAAiB,CAAC;IAC5C,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,CAAC,CAAE,CAAC;AAChD,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,MAAyB;IACpD,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;IAC5D,CAAC;IACD,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,MAAM,MAAM,GAAG,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAChD,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IAC9C,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC;IACrB,MAAM,GAAG,GAAG,MAAM,CAAC,CAAC,CAAE,CAAC;IACvB,MAAM,GAAG,GAAG,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC;IAC3B,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IAC9B,MAAM,MAAM,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAE,GAAG,MAAM,CAAC,GAAG,CAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAE,CAAC;IAClF,MAAM,QAAQ,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3F,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACnC,iEAAiE;IACjE,mEAAmE;IACnE,MAAM,IAAI,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,MAAM,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,iBAAiB,CAAC;IAC9F,OAAO,EAAE,IAAI,EAAE,GAAG,EAAE,GAAG,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;AAClD,CAAC;AAWD;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,YAAY,CAAC,MAAyB,EAAE,IAAuB;IAC7E,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC;IACrE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC;IACjC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IACtD,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC;IACpE,iGAAiG;IACjG,OAAO,IAAI,IAAI,SAAS,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC;AAClD,CAAC;AAED,wHAAwH;AACxH,MAAM,UAAU,WAAW,CAAC,QAAgB,EAAE,MAAc,EAAE,SAAiB;IAC7E,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,MAAM,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IAC1D,OAAO,QAAQ,GAAG,MAAM,GAAG,CAAC,CAAC,GAAG,SAAS,CAAC,CAAC;AAC7C,CAAC"}

package/dist/suite.d.ts

@@ -0,0 +1,25 @@
+import { type GpuContext } from './gpu/context.ts';
+import type { BenchmarkResult, SuiteOptions } from './types.ts';
+export type { BenchmarkResult, SuiteOptions, DeviceInfo, Stats, BenchmarkCategory, BenchmarkStatus, SamplingStopReason, SuiteProgressEvent, MetricDef, } from './types.ts';
+export { computeStats } from './stats.ts';
+/**
+ * Runs the full benchmark suite, yielding a BenchmarkResult every time a
+ * row changes so a UI can render the results table incrementally: once
+ * when each benchmark is set up (`running`), after every measurement (still
+ * `running`, with the best-so-far), and once when it finishes (ok / skipped
+ * / error).
+ *
+ * Every benchmark isolates a single resource — memory read, memory write,
+ * or ALU throughput — rather than modeling a specific real-world op, so
+ * results are directly comparable to the device's published bandwidth /
+ * FLOPS specs.
+ *
+ * Measurements are scheduled round-robin across all benchmarks with idle
+ * gaps and thermal cooldowns (see `runSampling`), and the reported number
+ * for each benchmark is its *best* run — so a phone that throttles partway
+ * through the suite still reports what it can do rather than what it was
+ * doing while hot.
+ */
+export declare function runSuite(options?: SuiteOptions): AsyncGenerator<BenchmarkResult>;
+export type { GpuContext };
+//# sourceMappingURL=suite.d.ts.map

package/dist/suite.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"suite.d.ts","sourceRoot":"","sources":["../src/suite.ts"],"names":[],"mappings":"AAAA,OAAO,EAAqB,KAAK,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAkDtE,OAAO,KAAK,EAAqB,eAAe,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAEnF,YAAY,EACV,eAAe,EACf,YAAY,EACZ,UAAU,EACV,KAAK,EACL,iBAAiB,EACjB,eAAe,EACf,kBAAkB,EAClB,kBAAkB,EAClB,SAAS,GACV,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAgB1C;;;;;;;;;;;;;;;;;GAiBG;AACH,wBAAuB,QAAQ,CAAC,OAAO,GAAE,YAAiB,GAAG,cAAc,CAAC,eAAe,CAAC,CAiI3F;AAyCD,YAAY,EAAE,UAAU,EAAE,CAAC"}

package/dist/suite.js

@@ -0,0 +1,196 @@
+import { acquireGpuContext } from "./gpu/context.js";
+import { generateMatVecData, padToMultipleOf4 } from "./data/generate.js";
+import { KernelSampler } from "./gpu/benchmarkRunner.js";
+import { runSampling } from "./sampling.js";
+import { prepareReadBandwidth, prepareWriteBandwidth } from "./benchmarks/streamBandwidth.js";
+import { prepareFlopsF32Scalar, prepareFlopsF32Vec4, prepareFlopsF32Mat4, prepareFlopsF32Matvec, } from "./benchmarks/flopsF32.js";
+import { prepareFlopsF16Scalar, prepareFlopsF16Vec4, prepareFlopsF16Mat4, prepareFlopsF16Matvec, } from "./benchmarks/flopsF16.js";
+import { prepareFlopsI8Scalar, prepareFlopsI8Vec4, prepareFlopsI8Mat4, prepareFlopsI8Matvec, prepareFlopsI8MatvecDp4a, prepareFlopsI8Dp4a, } from "./benchmarks/flopsI8.js";
+import { prepareFlopsF32Div, prepareFlopsI32Div, prepareFlopsF32Sqrt, prepareFlopsF32Rsqrt, prepareFlopsF32Pow, prepareFlopsF32Sincos, prepareFlopsF32Log, } from "./benchmarks/flopsMath.js";
+import { prepareFlopsU32PackUnpack, prepareFlopsI32F32Convert, prepareFlopsF32F16Convert, prepareFlopsI32F16Convert, } from "./benchmarks/flopsConvert.js";
+import { errorResult, metricPerSecond, rowFromMeta, FLOPS_METRIC, BYTES_METRIC, } from "./benchmarks/common.js";
+export { computeStats } from "./stats.js";
+const DEFAULT_ROWS = 4096;
+const DEFAULT_COLS = 4096;
+const bandwidth = (id, prepare) => [id, 'bandwidth', prepare];
+const compute = (id, prepare) => [id, 'compute', prepare];
+/**
+ * Runs the full benchmark suite, yielding a BenchmarkResult every time a
+ * row changes so a UI can render the results table incrementally: once
+ * when each benchmark is set up (`running`), after every measurement (still
+ * `running`, with the best-so-far), and once when it finishes (ok / skipped
+ * / error).
+ *
+ * Every benchmark isolates a single resource — memory read, memory write,
+ * or ALU throughput — rather than modeling a specific real-world op, so
+ * results are directly comparable to the device's published bandwidth /
+ * FLOPS specs.
+ *
+ * Measurements are scheduled round-robin across all benchmarks with idle
+ * gaps and thermal cooldowns (see `runSampling`), and the reported number
+ * for each benchmark is its *best* run — so a phone that throttles partway
+ * through the suite still reports what it can do rather than what it was
+ * doing while hot.
+ */
+export async function* runSuite(options = {}) {
+    const rows = padToMultipleOf4(options.rows ?? DEFAULT_ROWS);
+    const cols = padToMultipleOf4(options.cols ?? DEFAULT_COLS);
+    const ctx = await acquireGpuContext();
+    options.onDeviceInfo?.(ctx.info);
+    const data = generateMatVecData(rows, cols);
+    const harness = {
+        targetMs: options.targetMs,
+        targetDispatchMs: options.targetDispatchMs,
+        warmups: options.warmups,
+    };
+    const flopsHarness = { ...harness, threads: options.computeThreads, iterations: options.computeIterations };
+    const benchmarks = [
+        bandwidth('read-bandwidth', () => prepareReadBandwidth(ctx, data, harness)),
+        bandwidth('write-bandwidth', () => prepareWriteBandwidth(ctx, data, harness)),
+        compute('flops-f32-scalar', () => prepareFlopsF32Scalar(ctx, flopsHarness)),
+        compute('flops-f32-vec4', () => prepareFlopsF32Vec4(ctx, flopsHarness)),
+        compute('flops-f32-mat4', () => prepareFlopsF32Mat4(ctx, flopsHarness)),
+        compute('flops-f32-matvec', () => prepareFlopsF32Matvec(ctx, flopsHarness)),
+        compute('flops-f16-scalar', () => prepareFlopsF16Scalar(ctx, flopsHarness)),
+        compute('flops-f16-vec4', () => prepareFlopsF16Vec4(ctx, flopsHarness)),
+        compute('flops-f16-mat4', () => prepareFlopsF16Mat4(ctx, flopsHarness)),
+        compute('flops-f16-matvec', () => prepareFlopsF16Matvec(ctx, flopsHarness)),
+        compute('flops-i8-scalar', () => prepareFlopsI8Scalar(ctx, flopsHarness)),
+        compute('flops-i8-vec4', () => prepareFlopsI8Vec4(ctx, flopsHarness)),
+        compute('flops-i8-mat4', () => prepareFlopsI8Mat4(ctx, flopsHarness)),
+        compute('flops-i8-matvec', () => prepareFlopsI8Matvec(ctx, flopsHarness)),
+        compute('flops-i8-matvec-dp4a', () => prepareFlopsI8MatvecDp4a(ctx, flopsHarness)),
+        compute('flops-i8-dp4a', () => prepareFlopsI8Dp4a(ctx, flopsHarness)),
+        compute('flops-f32-div', () => prepareFlopsF32Div(ctx, flopsHarness)),
+        compute('flops-i32-div', () => prepareFlopsI32Div(ctx, flopsHarness)),
+        compute('flops-f32-sqrt', () => prepareFlopsF32Sqrt(ctx, flopsHarness)),
+        compute('flops-f32-rsqrt', () => prepareFlopsF32Rsqrt(ctx, flopsHarness)),
+        compute('flops-f32-pow', () => prepareFlopsF32Pow(ctx, flopsHarness)),
+        compute('flops-f32-sincos', () => prepareFlopsF32Sincos(ctx, flopsHarness)),
+        compute('flops-f32-log', () => prepareFlopsF32Log(ctx, flopsHarness)),
+        compute('flops-u32-packunpack', () => prepareFlopsU32PackUnpack(ctx, flopsHarness)),
+        compute('flops-i32-f32-convert', () => prepareFlopsI32F32Convert(ctx, flopsHarness)),
+        compute('flops-f32-f16-convert', () => prepareFlopsF32F16Convert(ctx, flopsHarness)),
+        compute('flops-i32-f16-convert', () => prepareFlopsI32F16Convert(ctx, flopsHarness)),
+    ];
+    // Phase 1: build every benchmark's GPU resources up front. Rows that can't
+    // run at all (missing feature, failed shader compile) resolve right here;
+    // the rest show up as `running` so the table has its final shape before
+    // the first measurement lands.
+    const scheduled = [];
+    for (const [id, category, prepare] of benchmarks) {
+        const fallbackMeta = {
+            id,
+            label: id,
+            description: '',
+            source: '',
+            category,
+            rows,
+            cols,
+            metric: category === 'bandwidth' ? BYTES_METRIC : FLOPS_METRIC,
+            amountPerOp: 0,
+        };
+        let prepared;
+        try {
+            prepared = await prepare();
+        }
+        catch (error) {
+            yield errorResult(fallbackMeta, error);
+            continue;
+        }
+        if (prepared.kind === 'skipped') {
+            yield prepared.result;
+            continue;
+        }
+        scheduled.push({
+            meta: prepared.meta,
+            sampler: new KernelSampler(prepared.harness),
+            metaAtWork: prepared.metaAtWork,
+        });
+        yield rowFromMeta(prepared.meta);
+    }
+    // Phase 2: sample everything round-robin. `runSampling` pushes updates
+    // through a callback; bridge them into this generator via a queue.
+    const byId = new Map(scheduled.map((k) => [k.meta.id, k]));
+    const queue = [];
+    let wake = null;
+    const push = (r) => {
+        queue.push(r);
+        wake?.();
+        wake = null;
+    };
+    const sampleables = scheduled.map(({ meta, sampler }) => ({
+        id: meta.id,
+        calibrate: () => sampler.calibrate(),
+        sample: () => sampler.sample(),
+    }));
+    const run = runSampling(sampleables, {
+        ...pickSamplingOptions(options),
+        onProgress: options.onProgress,
+        onUpdate: (state) => push(resultFromState(byId.get(state.id), state)),
+    }).finally(() => {
+        for (const k of scheduled)
+            k.sampler.destroy();
+    });
+    let done = false;
+    void run.then(() => {
+        done = true;
+        wake?.();
+    }, () => {
+        done = true;
+        wake?.();
+    });
+    try {
+        while (true) {
+            while (queue.length > 0)
+                yield queue.shift();
+            if (done)
+                break;
+            await new Promise((resolve) => {
+                wake = resolve;
+            });
+        }
+        await run;
+    }
+    finally {
+        ctx.device.destroy();
+    }
+}
+function pickSamplingOptions(o) {
+    return {
+        minRounds: o.minRounds,
+        maxRounds: o.maxRounds,
+        stableRounds: o.stableRounds,
+        improvementTolerance: o.improvementTolerance,
+        throttleThreshold: o.throttleThreshold,
+        idleMs: o.idleMs,
+        cooldownMs: o.cooldownMs,
+        maxCooldowns: o.maxCooldowns,
+        throttledFraction: o.throttledFraction,
+    };
+}
+function resultFromState(kernel, state) {
+    const { sampler } = kernel;
+    // Once calibrated, the problem size (and so FLOPs per op) is whatever the sampler settled on.
+    const meta = kernel.metaAtWork && sampler.work !== undefined ? kernel.metaAtWork(sampler.work) : kernel.meta;
+    if (state.error !== undefined)
+        return errorResult(meta, state.error);
+    const row = rowFromMeta(meta);
+    row.status = state.stopReason ? 'ok' : 'running';
+    row.innerIterations = sampler.innerIterations;
+    row.timingMethod = sampler.timingMethod;
+    row.timesMs = state.timesMs;
+    row.throttledMs = state.throttledMs;
+    row.stats = state.stats;
+    row.stopReason = state.stopReason;
+    if (state.stats) {
+        // Headline number comes from the best run: every noise source only ever
+        // slows a run down, so the minimum is the least-contaminated estimate.
+        row.metricValue = metricPerSecond(meta.amountPerOp, state.stats.min);
+    }
+    if (state.stopReason === 'throttled') {
+        row.message = 'Device stayed thermally throttled through every cooldown; best run may understate it.';
+    }
+    return row;
+}
+//# sourceMappingURL=suite.js.map

package/dist/suite.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"suite.js","sourceRoot":"","sources":["../src/suite.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAmB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,kBAAkB,EAAE,gBAAgB,EAAE,MAAM,oBAAoB,CAAC;AAC1E,OAAO,EAAE,aAAa,EAAE,MAAM,0BAA0B,CAAC;AACzD,OAAO,EAAE,WAAW,EAA6C,MAAM,eAAe,CAAC;AACvF,OAAO,EAAE,oBAAoB,EAAE,qBAAqB,EAAE,MAAM,iCAAiC,CAAC;AAC9F,OAAO,EACL,qBAAqB,EACrB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,GACtB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EACrB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,GACtB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,oBAAoB,EACpB,kBAAkB,EAClB,kBAAkB,EAClB,oBAAoB,EACpB,wBAAwB,EACxB,kBAAkB,GACnB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EACL,kBAAkB,EAClB,kBAAkB,EAClB,mBAAmB,EACnB,oBAAoB,EACpB,kBAAkB,EAClB,qBAAqB,EACrB,kBAAkB,GACnB,MAAM,2BAA2B,CAAC;AACnC,OAAO,EACL,yBAAyB,EACzB,yBAAyB,EACzB,yBAAyB,EACzB,yBAAyB,GAC1B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,WAAW,EACX,eAAe,EACf,WAAW,EACX,YAAY,EACZ,YAAY,GAIb,MAAM,wBAAwB,CAAC;AAchC,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE1C,MAAM,YAAY,GAAG,IAAI,CAAC;AAC1B,MAAM,YAAY,GAAG,IAAI,CAAC;AAG1B,MAAM,SAAS,GAAG,CAAC,EAAU,EAAE,OAAyC,EAAkB,EAAE,CAAC,CAAC,EAAE,EAAE,WAAW,EAAE,OAAO,CAAC,CAAC;AACxH,MAAM,OAAO,GAAG,CAAC,EAAU,EAAE,OAAyC,EAAkB,EAAE,CAAC,CAAC,EAAE,EAAE,SAAS,EAAE,OAAO,CAAC,CAAC;AASpH;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,CAAC,KAAK,SAAS,CAAC,CAAC,QAAQ,CAAC,UAAwB,EAAE;IACxD,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,IAAI,YAAY,CAAC,CAAC;IAC5D,MAAM,IAAI,GAAG,gBAAgB,CAAC,OAAO,CAAC,IAAI,IAAI,YAAY,CAAC,CAAC;IAE5D,MAAM,GAAG,GAAG,MAAM,iBAAiB,EAAE,CAAC;IACtC,OAAO,CAAC,YAAY,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IAEjC,MAAM,IAAI,GAAG,kBAAkB,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IAC5C,MAAM,OAAO,GAAkB;QAC7B,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,gBAAgB,EAAE,OAAO,CAAC,gBAAgB;QAC1C,OAAO,EAAE,OAAO,CAAC,OAAO;KACzB,CAAC;IACF,MAAM,YAAY,GAAG,EAAE,GAAG,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,cAAc,EAAE,UAAU,EAAE,OAAO,CAAC,iBAAiB,EAAE,CAAC;IAE5G,MAAM,UAAU,GAA8B;QAC5C,SAAS,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAC3E,SAAS,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC;QAC7E,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAClF,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,gBAAgB,EAAE,GAAG,EAAE,CAAC,mBAAmB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACvE,OAAO,CAAC,iBAAiB,EAAE,GAAG,EAAE,CAAC,oBAAoB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACzE,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,kBAAkB,EAAE,GAAG,EAAE,CAAC,qBAAqB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QAC3E,OAAO,CAAC,eAAe,EAAE,GAAG,EAAE,CAAC,kBAAkB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACrE,OAAO,CAAC,sBAAsB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACnF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACpF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;QACpF,OAAO,CAAC,uBAAuB,EAAE,GAAG,EAAE,CAAC,yBAAyB,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;KACrF,CAAC;IAEF,2EAA2E;IAC3E,0EAA0E;IAC1E,wEAAwE;IACxE,+BAA+B;IAC/B,MAAM,SAAS,GAAsB,EAAE,CAAC;IACxC,KAAK,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,CAAC,IAAI,UAAU,EAAE,CAAC;QACjD,MAAM,YAAY,GAAkB;YAClC,EAAE;YACF,KAAK,EAAE,EAAE;YACT,WAAW,EAAE,EAAE;YACf,MAAM,EAAE,EAAE;YACV,QAAQ;YACR,IAAI;YACJ,IAAI;YACJ,MAAM,EAAE,QAAQ,KAAK,WAAW,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,YAAY;YAC9D,WAAW,EAAE,CAAC;SACf,CAAC;QACF,IAAI,QAA2B,CAAC;QAChC,IAAI,CAAC;YACH,QAAQ,GAAG,MAAM,OAAO,EAAE,CAAC;QAC7B,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,WAAW,CAAC,YAAY,EAAE,KAAK,CAAC,CAAC;YACvC,SAAS;QACX,CAAC;QACD,IAAI,QAAQ,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YAChC,MAAM,QAAQ,CAAC,MAAM,CAAC;YACtB,SAAS;QACX,CAAC;QACD,SAAS,CAAC,IAAI,CAAC;YACb,IAAI,EAAE,QAAQ,CAAC,IAAI;YACnB,OAAO,EAAE,IAAI,aAAa,CAAC,QAAQ,CAAC,OAAO,CAAC;YAC5C,UAAU,EAAE,QAAQ,CAAC,UAAU;SAChC,CAAC,CAAC;QACH,MAAM,WAAW,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;IACnC,CAAC;IAED,uEAAuE;IACvE,mEAAmE;IACnE,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3D,MAAM,KAAK,GAAsB,EAAE,CAAC;IACpC,IAAI,IAAI,GAAwB,IAAI,CAAC;IACrC,MAAM,IAAI,GAAG,CAAC,CAAkB,EAAE,EAAE;QAClC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACd,IAAI,EAAE,EAAE,CAAC;QACT,IAAI,GAAG,IAAI,CAAC;IACd,CAAC,CAAC;IAEF,MAAM,WAAW,GAAiB,SAAS,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC,CAAC;QACtE,EAAE,EAAE,IAAI,CAAC,EAAE;QACX,SAAS,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,SAAS,EAAE;QACpC,MAAM,EAAE,GAAG,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE;KAC/B,CAAC,CAAC,CAAC;IAEJ,MAAM,GAAG,GAAG,WAAW,CAAC,WAAW,EAAE;QACnC,GAAG,mBAAmB,CAAC,OAAO,CAAC;QAC/B,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,QAAQ,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAE,EAAE,KAAK,CAAC,CAAC;KACvE,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE;QACd,KAAK,MAAM,CAAC,IAAI,SAAS;YAAE,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;IACjD,CAAC,CAAC,CAAC;IACH,IAAI,IAAI,GAAG,KAAK,CAAC;IACjB,KAAK,GAAG,CAAC,IAAI,CACX,GAAG,EAAE;QACH,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,EAAE,EAAE,CAAC;IACX,CAAC,EACD,GAAG,EAAE;QACH,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,EAAE,EAAE,CAAC;IACX,CAAC,CACF,CAAC;IAEF,IAAI,CAAC;QACH,OAAO,IAAI,EAAE,CAAC;YACZ,OAAO,KAAK,CAAC,MAAM,GAAG,CAAC;gBAAE,MAAM,KAAK,CAAC,KAAK,EAAG,CAAC;YAC9C,IAAI,IAAI;gBAAE,MAAM;YAChB,MAAM,IAAI,OAAO,CAAO,CAAC,OAAO,EAAE,EAAE;gBAClC,IAAI,GAAG,OAAO,CAAC;YACjB,CAAC,CAAC,CAAC;QACL,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;YAAS,CAAC;QACT,GAAG,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;IACvB,CAAC;AACH,CAAC;AAED,SAAS,mBAAmB,CAAC,CAAe;IAC1C,OAAO;QACL,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,SAAS,EAAE,CAAC,CAAC,SAAS;QACtB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,oBAAoB,EAAE,CAAC,CAAC,oBAAoB;QAC5C,iBAAiB,EAAE,CAAC,CAAC,iBAAiB;QACtC,MAAM,EAAE,CAAC,CAAC,MAAM;QAChB,UAAU,EAAE,CAAC,CAAC,UAAU;QACxB,YAAY,EAAE,CAAC,CAAC,YAAY;QAC5B,iBAAiB,EAAE,CAAC,CAAC,iBAAiB;KACvC,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,MAAuB,EAAE,KAA0B;IAC1E,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,CAAC;IAC3B,8FAA8F;IAC9F,MAAM,IAAI,GAAG,MAAM,CAAC,UAAU,IAAI,OAAO,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC;IAC7G,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS;QAAE,OAAO,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,KAAK,CAAC,CAAC;IAErE,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC;IAC9B,GAAG,CAAC,MAAM,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IACjD,GAAG,CAAC,eAAe,GAAG,OAAO,CAAC,eAAe,CAAC;IAC9C,GAAG,CAAC,YAAY,GAAG,OAAO,CAAC,YAAY,CAAC;IACxC,GAAG,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;IAC5B,GAAG,CAAC,WAAW,GAAG,KAAK,CAAC,WAAW,CAAC;IACpC,GAAG,CAAC,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC;IACxB,GAAG,CAAC,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC;IAClC,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;QAChB,wEAAwE;QACxE,uEAAuE;QACvE,GAAG,CAAC,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACvE,CAAC;IACD,IAAI,KAAK,CAAC,UAAU,KAAK,WAAW,EAAE,CAAC;QACrC,GAAG,CAAC,OAAO,GAAG,uFAAuF,CAAC;IACxG,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/types.d.ts

@@ -0,0 +1,171 @@
+/** Summary statistics computed over the kept (non-throttled) timed measurements. */
+export interface Stats {
+    mean: number;
+    /** The best (fastest) kept measurement — the headline number, see `BenchmarkResult.metricValue`. */
+    min: number;
+    max: number;
+    median: number;
+    stddev: number;
+    /** Half-width of the 95% confidence interval on the mean (Student-t); Infinity with a single sample. */
+    ci95: number;
+}
+/**
+ * Why sampling stopped for a benchmark:
+ * - `converged`: the best run stopped improving (see `SuiteOptions.stableRounds`).
+ * - `max-rounds`: the per-benchmark cap was hit — the best was still improving, or the benchmark
+ *   kept producing throttled runs on its own while the rest of the suite was fine.
+ * - `throttled`: the device kept coming back throttled even after the suite's cooldown pauses
+ *   were exhausted, so the best run may understate the device (treat the row with suspicion).
+ */
+export type SamplingStopReason = 'converged' | 'max-rounds' | 'throttled';
+export type TimingMethod = 'gpu-timestamp' | 'cpu-wallclock';
+/** `bandwidth`: streams memory with ~no compute. `compute`: pure ALU with ~no memory traffic. */
+export type BenchmarkCategory = 'bandwidth' | 'compute';
+export type BenchmarkStatus = 'running' | 'ok' | 'skipped' | 'error';
+/**
+ * Identifies what a benchmark's throughput number counts. `key` is a
+ * JSON-safe id shared by every benchmark that measures the same kind of
+ * thing (so it's stable to key off, not unique per-benchmark); `unit` is
+ * the short symbol shown as `<unit>/s` (FLOP, OP, B); `name` is a short
+ * friendly label for display (e.g. in a tooltip).
+ */
+export interface MetricDef {
+    key: string;
+    unit: string;
+    name: string;
+}
+/**
+ * One row of the results table. `runSuite` yields a fresh object for a given
+ * `id` every time the row changes — when the benchmark starts (`running`),
+ * after every sample it takes (still `running`, with the partial stats so
+ * far), and when it finishes (ok/skipped/error) — so the UI can render
+ * progress incrementally.
+ */
+export interface BenchmarkResult {
+    id: string;
+    label: string;
+    description: string;
+    /** WGSL source of the kernel that was run, for display alongside the result. */
+    source: string;
+    category: BenchmarkCategory;
+    status: BenchmarkStatus;
+    message?: string;
+    /**
+     * Logical problem size actually benchmarked. For `bandwidth` kernels:
+     * the streamed buffer's rows/cols (after padding to a multiple of 4).
+     * For `compute` kernels: thread count / in-shader loop trip count (as
+     * calibrated for this GPU, see `SuiteOptions.targetDispatchMs`).
+     */
+    rows: number;
+    cols: number;
+    /** Number of GPU dispatches batched into each timed measurement. */
+    innerIterations: number;
+    /** Per-op time (ms) for each kept timed measurement, in the order taken. Length varies: sampling is adaptive. */
+    timesMs: number[];
+    /**
+     * Per-op time (ms) for measurements that were discarded as throttled —
+     * more than `throttleThreshold` slower than the best run seen so far. Kept for diagnostics; never feeds `stats`.
+     */
+    throttledMs: number[];
+    /** Summary over `timesMs`. `stats.min` is the best run. */
+    stats?: Stats;
+    /** Set for `ok` rows: see `SamplingStopReason`. */
+    stopReason?: SamplingStopReason;
+    /** What `metricValue` counts. Set from the start, before any measurement lands. */
+    metric: MetricDef;
+    /** Peak `<metric.unit>/s`, computed from the best run (`stats.min`). */
+    metricValue?: number;
+    timingMethod: TimingMethod;
+}
+/** Suite-level progress events (round boundaries and thermal cooldown pauses). */
+export type SuiteProgressEvent = {
+    type: 'round';
+    round: number;
+    active: number;
+} | {
+    type: 'cooldown';
+    attempt: number;
+    maxAttempts: number;
+    ms: number;
+    throttledIds: string[];
+} | {
+    type: 'throttle-abort';
+    throttledIds: string[];
+};
+export interface SuiteOptions {
+    /** Streamed buffer rows (for the bandwidth benchmarks). Padded up to a multiple of 4. Default 4096. */
+    rows?: number;
+    /** Streamed buffer cols (for the bandwidth benchmarks). Padded up to a multiple of 4. Default 4096. */
+    cols?: number;
+    /**
+     * Target duration of a single timed measurement, in ms. Sets how many
+     * dispatches get batched per measurement. Kept short (default 100) so the
+     * suite's total GPU-busy time — and therefore heat — stays low.
+     */
+    targetMs?: number;
+    /**
+     * Target duration of a single GPU dispatch, in ms. A dispatch can't be
+     * preempted, so a long one freezes the display for its whole duration;
+     * the raw-FLOPS kernels have their loop trip count calibrated (from a tiny
+     * first probe upward, never more than 4x per step) so one dispatch lands
+     * near this on whatever GPU is running. Default 10.
+     */
+    targetDispatchMs?: number;
+    /** Number of discarded warmup measurements per benchmark before timing starts (the calibration pass is a further untimed run). Default 1. */
+    warmups?: number;
+    /** Fewest kept measurements per benchmark before it can be declared converged. Default 3. */
+    minRounds?: number;
+    /** Most kept measurements per benchmark, reached only if the best run keeps improving. Default 10. */
+    maxRounds?: number;
+    /**
+     * A benchmark is converged once its best run has not improved by more than
+     * `improvementTolerance` over this many consecutive kept measurements.
+     * Default 2.
+     */
+    stableRounds?: number;
+    /** Relative improvement of the best run that counts as "still improving" for `stableRounds`. Default 0.01 (1%). */
+    improvementTolerance?: number;
+    /**
+     * A measurement more than this fraction slower than the benchmark's best
+     * run is classified as thermally throttled and discarded. Default 0.20
+     * (20%): real throttling is 20–50%, while ordinary wall-clock jitter on
+     * Safari runs to ~15%.
+     */
+    throttleThreshold?: number;
+    /** Idle gap between consecutive measurements, in ms, so the GPU duty-cycles instead of running flat out. Default 100. */
+    idleMs?: number;
+    /** How long the whole suite pauses when it detects throttling, in ms. Default 3000. */
+    cooldownMs?: number;
+    /** How many cooldown pauses to attempt before giving up and reporting the still-unconverged benchmarks as `throttled`. Default 3. */
+    maxCooldowns?: number;
+    /**
+     * Suite-wide throttle trigger: a cooldown starts when at least this
+     * fraction (and at least two) of the benchmarks measured in a round come
+     * back throttled. Default 0.5.
+     */
+    throttledFraction?: number;
+    /** Number of GPU threads (invocations) launched by the raw-FLOPS compute benchmarks. Default 1,048,576. */
+    computeThreads?: number;
+    /**
+     * Pins the trip count of the in-shader loop in the raw-FLOPS compute
+     * benchmarks, disabling per-dispatch calibration. Left unset, each kernel's
+     * default (256–1024) is the *ceiling* and the sampler picks the count that
+     * makes one dispatch take ~`targetDispatchMs`.
+     */
+    computeIterations?: number;
+    /** Called once with the resolved GPU device/adapter info before benchmarks start. */
+    onDeviceInfo?: (info: DeviceInfo) => void;
+    /** Called at round boundaries and when the suite pauses for a thermal cooldown. */
+    onProgress?: (event: SuiteProgressEvent) => void;
+}
+export interface DeviceInfo {
+    vendor?: string;
+    architecture?: string;
+    description?: string;
+    features: string[];
+    limits: Record<string, number>;
+    supportsF16: boolean;
+    supportsI8Dot: boolean;
+    supportsTimestampQuery: boolean;
+}
+//# sourceMappingURL=types.d.ts.map

package/dist/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":"AAAA,oFAAoF;AACpF,MAAM,WAAW,KAAK;IACpB,IAAI,EAAE,MAAM,CAAC;IACb,oGAAoG;IACpG,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,wGAAwG;IACxG,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,kBAAkB,GAAG,WAAW,GAAG,YAAY,GAAG,WAAW,CAAC;AAE1E,MAAM,MAAM,YAAY,GAAG,eAAe,GAAG,eAAe,CAAC;AAE7D,iGAAiG;AACjG,MAAM,MAAM,iBAAiB,GAAG,WAAW,GAAG,SAAS,CAAC;AAExD,MAAM,MAAM,eAAe,GAAG,SAAS,GAAG,IAAI,GAAG,SAAS,GAAG,OAAO,CAAC;AAErE;;;;;;GAMG;AACH,MAAM,WAAW,SAAS;IACxB,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,gFAAgF;IAChF,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,MAAM,EAAE,eAAe,CAAC;IACxB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;;;;OAKG;IACH,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,CAAC;IACb,oEAAoE;IACpE,eAAe,EAAE,MAAM,CAAC;IACxB,iHAAiH;IACjH,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB;;;OAGG;IACH,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,2DAA2D;IAC3D,KAAK,CAAC,EAAE,KAAK,CAAC;IACd,mDAAmD;IACnD,UAAU,CAAC,EAAE,kBAAkB,CAAC;IAChC,mFAAmF;IACnF,MAAM,EAAE,SAAS,CAAC;IAClB,wEAAwE;IACxE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,EAAE,YAAY,CAAC;CAC5B;AAED,kFAAkF;AAClF,MAAM,MAAM,kBAAkB,GAC1B;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,GAChD;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,EAAE,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,EAAE,CAAA;CAAE,GAC9F;IAAE,IAAI,EAAE,gBAAgB,CAAC;IAAC,YAAY,EAAE,MAAM,EAAE,CAAA;CAAE,CAAC;AAEvD,MAAM,WAAW,YAAY;IAC3B,uGAAuG;IACvG,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,uGAAuG;IACvG,IAAI,CAAC,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;;;OAMG;IACH,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,6IAA6I;IAC7I,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,6FAA6F;IAC7F,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,sGAAsG;IACtG,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,mHAAmH;IACnH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;OAKG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,yHAAyH;IACzH,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,uFAAuF;IACvF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,qIAAqI;IACrI,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,2GAA2G;IAC3G,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB;;;;;OAKG;IACH,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,qFAAqF;IACrF,YAAY,CAAC,EAAE,CAAC,IAAI,EAAE,UAAU,KAAK,IAAI,CAAC;IAC1C,mFAAmF;IACnF,UAAU,CAAC,EAAE,CAAC,KAAK,EAAE,kBAAkB,KAAK,IAAI,CAAC;CAClD;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC/B,WAAW,EAAE,OAAO,CAAC;IACrB,aAAa,EAAE,OAAO,CAAC;IACvB,sBAAsB,EAAE,OAAO,CAAC;CACjC"}

package/dist/types.js

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=types.js.map

package/dist/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../src/types.ts"],"names":[],"mappings":""}

package/package.json

@@ -0,0 +1,45 @@
+{
+  "name": "webgpu-bench",
+  "version": "0.1.0",
+  "description": "Benchmark suite for WebGPU's raw ceilings: memory bandwidth and fp32/fp16/int8 FLOPS.",
+  "keywords": [
+    "webgpu",
+    "benchmark",
+    "gpu",
+    "flops",
+    "bandwidth",
+    "profiler",
+    "wgsl",
+    "vgpu"
+  ],
+  "homepage": "https://github.com/bhouston/webgpu-profiler",
+  "bugs": {
+    "url": "https://github.com/bhouston/webgpu-profiler/issues"
+  },
+  "license": "MIT",
+  "author": "Ben Houston (https://ben3d.ca)",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/bhouston/webgpu-profiler.git",
+    "directory": "packages/webgpu-bench"
+  },
+  "type": "module",
+  "sideEffects": false,
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "scripts": {
+    "build": "tsc",
+    "dev": "tsc -w --preserveWatchOutput --noEmitOnError",
+    "test": "vitest run",
+    "test:browser": "vitest run --project browser"
+  },
+  "dependencies": {
+    "vgpu": "^0.4.0"
+  },
+  "devDependencies": {
+    "@webgpu/types": "^0.1.69"
+  },
+  "engines": {
+    "node": ">=22.0.0"
+  }
+}