wgblas 2.0.0 → 2.2.0

package/src/isamax/isamax.mjs

@@ -12,14 +12,14 @@ import { extractTimestamp } from "../util/benchmark.mjs";
 import { extractResult } from "../util/result.mjs";
 import { getPipeline } from "../util/pipeline.mjs";
 import { GpuVector } from "../classes/GpuVector.mjs";
-
-const WGS = 64;
+import { WGS } from "../util/constants.mjs";
+import { requireGpuDevice, requireSameDevice } from "../util/device.mjs";
 
 export async function isamax(device, n, x, incx) {
   const xIsGpu = x instanceof GpuVector;
 
-  if (!(device instanceof GPUDevice))
-    throw new Error("device must be a GPUDevice.");
+  requireGpuDevice(device);
+  requireSameDevice(device, "isamax", { x });
   if (!Number.isInteger(n) || !Number.isInteger(incx))
     throw new Error("n and incx must be integers.");
   if (incx <= 0) throw new Error("incx must be positive.");
@@ -42,17 +42,20 @@ export async function isamax(device, n, x, incx) {
   let readBuffer = null;
 
   try {
-    xBuffer = xIsGpu ? x._buf : uploadBuffer(x, "isamax-x", false);
+    xBuffer = xIsGpu ? x._buf : uploadBuffer(device, x, "isamax-x", false);
     partialsValBuffer = createStorageBuffer(
+      device,
       2 * WGS * 4,
       "isamax-partials-val",
     ); //to hold 2*WGS partial max values of f32
     partialsIdxBuffer = createStorageBuffer(
+      device,
       2 * WGS * 4,
       "isamax-partials-idx",
     ); //to hold 2*WGS partial max indices of u32
-    resultBuffer = createResultBuffer(4, "isamax-result"); // u32 index
+    resultBuffer = createResultBuffer(device, 4, "isamax-result"); // u32 index
     paramsBuffer = createParamsBuffer(
+      device,
       [
         { value: n, type: "u32" },
         { value: incx, type: "u32" },
@@ -60,33 +63,35 @@ export async function isamax(device, n, x, incx) {
       "isamax-params",
     );
 
-    const bgMain = createBindGroup(pipelineMain.getBindGroupLayout(0), [
+    const bgMain = createBindGroup(device, pipelineMain.getBindGroupLayout(0), [
       xBuffer,
       partialsValBuffer,
       partialsIdxBuffer,
       paramsBuffer,
     ]);
     const { commandEncoder: enc1, ts: ts1 } = runComputePass(
+      device,
       pipelineMain,
       bgMain,
       2 * WGS,
     ); //dispatch 2*WGS workgroups
 
-    submit(enc1);
+    submit(device, enc1);
 
-    const bgReduce = createBindGroup(pipelineReduce.getBindGroupLayout(0), [
-      partialsValBuffer,
-      partialsIdxBuffer,
-      resultBuffer,
-    ]);
+    const bgReduce = createBindGroup(
+      device,
+      pipelineReduce.getBindGroupLayout(0),
+      [partialsValBuffer, partialsIdxBuffer, resultBuffer],
+    );
     const { commandEncoder: enc2, ts: ts2 } = runComputePass(
+      device,
       pipelineReduce,
       bgReduce,
       1,
     ); // dispatch 1 workgroup to reduce the partial sums to a single result
-    readBuffer = stageReadback(enc2, resultBuffer);
+    readBuffer = stageReadback(device, enc2, resultBuffer);
 
-    submit(enc2);
+    submit(device, enc2);
 
     const resultPromise = extractResult(readBuffer, Uint32Array);
     readBuffer = null; // ownership transferred — extractResult's own finally destroys it
@@ -108,7 +113,7 @@ export async function isamax(device, n, x, incx) {
     if (partialsIdxBuffer) destroyBuffers(partialsIdxBuffer);
     if (resultBuffer) destroyBuffers(resultBuffer);
     if (paramsBuffer) destroyBuffers(paramsBuffer);
-    // Only reached if submit(enc2) threw before ownership was transferred above.
+    // Only reached if submit(device, enc2) threw before ownership was transferred above.
     if (readBuffer) destroyBuffers(readBuffer);
   }
 }

package/src/random/random.d.mts

@@ -4,22 +4,19 @@
  * @param n - number of elements
  * @param low - lower bound (default: -1)
  * @param high - upper bound (default: 1)
+ * @param seed - omit for genuine randomness (`Math.random`, the default);
+ *   pass any number for a deterministic, reproducible sequence (mulberry32) —
+ *   useful for regression tests that need "random-looking" data without
+ *   flaking between runs
  *
- * @example Default range [-1, 1)
- * ```js
- * import { randomFloat32Array } from "wgblas";
+ * **Default range [-1, 1):**
+ * {@includeCode ../../examples/randomfloat32array/randomfloat32array.js}
  *
- * const x = randomFloat32Array(4);
- * console.log(x); // Float32Array [ -0.42, 0.81, -0.07, 0.55 ]
- * ```
+ * **Custom range [0, 10):**
+ * {@includeCode ../../examples/randomfloat32array-custom/randomfloat32array-custom.js}
  *
- * @example Custom range [0, 10)
- * ```js
- * import { randomFloat32Array } from "wgblas";
- *
- * const x = randomFloat32Array(4, 0, 10);
- * console.log(x);
- * ```
+ * **Seeded (deterministic):**
+ * {@includeCode ../../examples/randomfloat32array-seeded/randomfloat32array-seeded.js}
  * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/random/random.mjs#L1">Source code: random.mjs (L1)</a>
  * @category Utilities
  */
@@ -27,6 +24,7 @@ export declare function randomFloat32Array(
   n: number,
   low?: number,
   high?: number,
+  seed?: number,
 ): Float32Array;
 
 /**
@@ -35,22 +33,19 @@ export declare function randomFloat32Array(
  * @param n - number of elements
  * @param low - lower bound (default: -1)
  * @param high - upper bound (default: 1)
+ * @param seed - omit for genuine randomness (`Math.random`, the default);
+ *   pass any number for a deterministic, reproducible sequence (mulberry32) —
+ *   useful for regression tests that need "random-looking" data without
+ *   flaking between runs
  *
- * @example Default range [-1, 1)
- * ```js
- * import { randomFloat64Array } from "wgblas";
- *
- * const x = randomFloat64Array(4);
- * console.log(x); // Float64Array [ -0.42, 0.81, -0.07, 0.55 ]
- * ```
+ * **Default range [-1, 1):**
+ * {@includeCode ../../examples/randomfloat64array/randomfloat64array.js}
  *
- * @example Custom range [0, 10)
- * ```js
- * import { randomFloat64Array } from "wgblas";
+ * **Custom range [0, 10):**
+ * {@includeCode ../../examples/randomfloat64array-custom/randomfloat64array-custom.js}
  *
- * const x = randomFloat64Array(4, 0, 10);
- * console.log(x);
- * ```
+ * **Seeded (deterministic):**
+ * {@includeCode ../../examples/randomfloat64array-seeded/randomfloat64array-seeded.js}
  * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/random/random.mjs#L7">Source code: random.mjs (L7)</a>
  * @category Utilities
  */
@@ -58,16 +53,20 @@ export declare function randomFloat64Array(
   n: number,
   low?: number,
   high?: number,
+  seed?: number,
 ): Float64Array;
 
 /**
- * Returns a Float32Array of n*lda elements, row-major with leading dimension
- * `lda`, representing an actual lower- or upper-triangular matrix for a
- * triangular routine (`strmv`/`strsv`) example: entries in the `uplo`
- * triangle are uniform in `[low, high)`, the n diagonal entries
- * (`A[i*lda+i]`) are uniform in `[diagLow, diagHigh)` — kept well away from 0
- * so a triangular solve doesn't divide by a near-zero pivot — and every entry
- * in the other triangle is 0.
+ * Returns a Float32Array of n*lda elements, with leading dimension `lda`
+ * under the given `layout`, representing an actual lower- or
+ * upper-triangular matrix for a triangular routine (`strmv`/`strsv`)
+ * example: entries in the `uplo` triangle are uniform in `[low, high)`, the
+ * n diagonal entries are uniform in `[diagLow, diagHigh)` — kept well away
+ * from 0 so a triangular solve doesn't divide by a near-zero pivot — and
+ * every entry in the other triangle is 0. `uplo` always describes the
+ * logical triangle, regardless of storage order: only the flat index each
+ * (row, col) maps to changes between layouts (`A[row*lda+col]` for
+ * row-major, `A[col*lda+row]` for column-major).
  *
  * @param n - matrix order (rows/cols read by the triangular routine)
  * @param lda - leading dimension; throws if `lda < n`
@@ -76,14 +75,12 @@ export declare function randomFloat64Array(
  * @param high - upper bound for off-diagonal entries (default: 1)
  * @param diagLow - lower bound for diagonal entries (default: 5)
  * @param diagHigh - upper bound for diagonal entries (default: 15)
+ * @param layout - `'row-major'` or `'column-major'` storage order (default: `'row-major'`)
  *
- * @example
- * ```js
- * import { randomTriangularFloat32Array } from "wgblas";
+ * {@includeCode ../../examples/randomtriangularfloat32array/randomtriangularfloat32array.js}
  *
- * const n = 4, lda = n;
- * const A = randomTriangularFloat32Array(n, lda, "lower");
- * ```
+ * **Column-major storage:**
+ * {@includeCode ../../examples/randomtriangularfloat32array-columnmajor/randomtriangularfloat32array-columnmajor.js}
  * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/random/random.mjs#L13">Source code: random.mjs (L13)</a>
  * @category Utilities
  */
@@ -95,4 +92,5 @@ export declare function randomTriangularFloat32Array(
   high?: number,
   diagLow?: number,
   diagHigh?: number,
+  layout?: 'row-major' | 'column-major',
 ): Float32Array;

package/src/random/random.mjs

@@ -1,28 +1,60 @@
-export function randomFloat32Array(n, low = -1, high = 1) {
+// mulberry32: a small, fast, deterministic PRNG — not cryptographic, but
+// good enough distribution for test/benchmark data. Returns a () => number
+// generator producing values in [0, 1), same contract as Math.random.
+function mulberry32(seed) {
+  let a = seed >>> 0;
+  return function () {
+    a = (a + 0x6d2b79f5) | 0;
+    let t = Math.imul(a ^ (a >>> 15), 1 | a);
+    t = (t + Math.imul(t ^ (t >>> 7), 61 | t)) ^ t;
+    return ((t ^ (t >>> 14)) >>> 0) / 4294967296;
+  };
+}
+
+export function randomFloat32Array(n, low = -1, high = 1, seed) {
   const x = new Float32Array(n);
-  for (let i = 0; i < n; i++) x[i] = low + Math.random() * (high - low);
+  const next = seed === undefined ? Math.random : mulberry32(seed);
+  for (let i = 0; i < n; i++) x[i] = low + next() * (high - low);
   return x;
 }
 
-export function randomFloat64Array(n, low = -1, high = 1) {
+export function randomFloat64Array(n, low = -1, high = 1, seed) {
   const x = new Float64Array(n);
-  for (let i = 0; i < n; i++) x[i] = low + Math.random() * (high - low);
+  const next = seed === undefined ? Math.random : mulberry32(seed);
+  for (let i = 0; i < n; i++) x[i] = low + next() * (high - low);
   return x;
 }
 
-export function randomTriangularFloat32Array(n, lda, uplo = "lower", low = -1, high = 1, diagLow = 5, diagHigh = 15) {
+export function randomTriangularFloat32Array(
+  n,
+  lda,
+  uplo = "lower",
+  low = -1,
+  high = 1,
+  diagLow = 5,
+  diagHigh = 15,
+  layout = "row-major",
+) {
   if (uplo !== "lower" && uplo !== "upper")
     throw new Error("uplo must be 'lower' or 'upper'.");
+  if (layout !== "row-major" && layout !== "column-major")
+    throw new Error("layout must be 'row-major' or 'column-major'.");
   if (lda < n) throw new Error("lda must be >= n.");
 
+  // Row-major stores row i at A[i*lda+j]; column-major stores column j at
+  // A[j*lda+i] instead. `uplo` describes the logical triangle (unaffected by
+  // storage order) — only which flat index each (i, j) maps to changes.
+  const isColMajor = layout === "column-major";
+  const idx = (i, j) => (isColMajor ? j * lda + i : i * lda + j);
+
   const A = new Float32Array(n * lda);
   for (let i = 0; i < n; i++) {
     for (let j = 0; j < n; j++) {
       if (i === j) continue;
       const inTriangle = uplo === "lower" ? j < i : j > i;
-      if (inTriangle) A[i * lda + j] = low + Math.random() * (high - low);
+      if (inTriangle) A[idx(i, j)] = low + Math.random() * (high - low);
     }
-    A[i * lda + i] = diagLow + Math.random() * (diagHigh - diagLow);
+    A[idx(i, i)] = diagLow + Math.random() * (diagHigh - diagLow);
   }
   return A;
 }

package/src/sasum/sasum.d.mts

@@ -1,7 +1,7 @@
 import { GpuVector } from "../classes/GpuVector.mjs";
 
 /**
- * Computes the sum of absolute values of a vector: result = sum(|x[i]|)
+ * Computes the sum of absolute values of a vector: $$\text{result} = \sum_{i} |x_i|$$
  *
  * {@includeCode ../../examples/sasum/sasum.js}
  *
@@ -24,7 +24,7 @@ export declare function sasum(
 ): Promise<{ asum: number } | { asum: number; gpuTimeMs: number }>;
 
 /**
- * Computes the sum of absolute values of a vector: result = sum(|x[i]|)
+ * Computes the sum of absolute values of a vector: $$\text{result} = \sum_{i} |x_i|$$
  *
  * {@includeCode ../../examples/sasum/gpu.sasum.js}
  *

package/src/sasum/sasum.mjs

@@ -12,14 +12,14 @@ import { extractTimestamp } from "../util/benchmark.mjs";
 import { extractResult } from "../util/result.mjs";
 import { getPipeline } from "../util/pipeline.mjs";
 import { GpuVector } from "../classes/GpuVector.mjs";
-
-const WGS = 64; // workgroup size
+import { WGS } from "../util/constants.mjs";
+import { requireGpuDevice, requireSameDevice } from "../util/device.mjs";
 
 export async function sasum(device, n, x, incx) {
   const xIsGpu = x instanceof GpuVector;
 
-  if (!(device instanceof GPUDevice))
-    throw new Error("device must be a GPUDevice.");
+  requireGpuDevice(device);
+  requireSameDevice(device, "sasum", { x });
   if (!Number.isInteger(n) || !Number.isInteger(incx))
     throw new Error("n and incx must be integers.");
   if (incx <= 0) throw new Error("incx must be positive.");
@@ -41,10 +41,11 @@ export async function sasum(device, n, x, incx) {
   let readBuffer = null;
 
   try {
-    xBuffer = xIsGpu ? x._buf : uploadBuffer(x, "sasum-x", false);
-    partialsBuffer = createStorageBuffer(2 * WGS * 4, "sasum-partials"); // 2*WGS partial sums of f32
-    resultBuffer = createResultBuffer(4, "sasum-result"); // final f32 scalar
+    xBuffer = xIsGpu ? x._buf : uploadBuffer(device, x, "sasum-x", false);
+    partialsBuffer = createStorageBuffer(device, 2 * WGS * 4, "sasum-partials"); // 2*WGS partial sums of f32
+    resultBuffer = createResultBuffer(device, 4, "sasum-result"); // final f32 scalar
     paramsBuffer = createParamsBuffer(
+      device,
       [
         { value: n, type: "u32" },
         { value: incx, type: "u32" },
@@ -52,31 +53,34 @@ export async function sasum(device, n, x, incx) {
       "sasum-params",
     );
 
-    const bgMain = createBindGroup(pipelineMain.getBindGroupLayout(0), [
+    const bgMain = createBindGroup(device, pipelineMain.getBindGroupLayout(0), [
       xBuffer,
       partialsBuffer,
       paramsBuffer,
     ]);
     const { commandEncoder: enc1, ts: ts1 } = runComputePass(
+      device,
       pipelineMain,
       bgMain,
       2 * WGS,
     ); // dispatch 2*WGS workgroups
 
-    submit(enc1);
+    submit(device, enc1);
 
-    const bgReduce = createBindGroup(pipelineReduce.getBindGroupLayout(0), [
-      partialsBuffer,
-      resultBuffer,
-    ]);
+    const bgReduce = createBindGroup(
+      device,
+      pipelineReduce.getBindGroupLayout(0),
+      [partialsBuffer, resultBuffer],
+    );
     const { commandEncoder: enc2, ts: ts2 } = runComputePass(
+      device,
       pipelineReduce,
       bgReduce,
       1,
     ); // dispatch 1 workgroup to reduce the partial sums to a single result
-    readBuffer = stageReadback(enc2, resultBuffer);
+    readBuffer = stageReadback(device, enc2, resultBuffer);
 
-    submit(enc2);
+    submit(device, enc2);
 
     const resultPromise = extractResult(readBuffer, Float32Array);
     readBuffer = null; // ownership transferred — extractResult's own finally destroys it
@@ -96,7 +100,7 @@ export async function sasum(device, n, x, incx) {
     if (partialsBuffer) destroyBuffers(partialsBuffer);
     if (resultBuffer) destroyBuffers(resultBuffer);
     if (paramsBuffer) destroyBuffers(paramsBuffer);
-    // Only reached if submit(enc2) threw before ownership was transferred above.
+    // Only reached if submit(device, enc2) threw before ownership was transferred above.
     if (readBuffer) destroyBuffers(readBuffer);
   }
 }

package/src/saxpy/saxpy.d.mts

@@ -1,7 +1,7 @@
 import { GpuVector } from "../classes/GpuVector.mjs";
 
 /**
- * Performs the operation y = alpha * x + y
+ * Performs the operation $$y \leftarrow \alpha x + y$$
  *
  * {@includeCode ../../examples/saxpy/saxpy.js}
  *
@@ -29,7 +29,7 @@ export declare function saxpy(
 ): Promise<{ y: Float32Array } | { y: Float32Array; gpuTimeMs: number }>;
 
 /**
- * Performs the operation y = alpha * x + y
+ * Performs the operation $$y \leftarrow \alpha x + y$$
  *
  * {@includeCode ../../examples/saxpy/gpu.saxpy.js}
  *

package/src/saxpy/saxpy.mjs

@@ -11,21 +11,21 @@ import { extractTimestamp } from "../util/benchmark.mjs";
 import { getPipeline } from "../util/pipeline.mjs";
 import { calcWorkgroups } from "../util/workgroup.mjs";
 import { GpuVector } from "../classes/GpuVector.mjs";
+import { requireGpuDevice, requireSameDevice } from "../util/device.mjs";
 
 export async function saxpy(device, n, alpha, x, incx, y, incy) {
   const xIsGpu = x instanceof GpuVector;
   const yIsGpu = y instanceof GpuVector;
 
-  if (!(device instanceof GPUDevice))
-    throw new Error("device must be a GPUDevice.");
+  requireGpuDevice(device);
+  requireSameDevice(device, "saxpy", { x, y });
   if (
     !Number.isInteger(n) ||
     !Number.isInteger(incx) ||
     !Number.isInteger(incy)
   )
     throw new Error("n, incx, and incy must be integers.");
-  if (typeof alpha !== "number")
-    throw new Error("alpha must be a number.");
+  if (typeof alpha !== "number") throw new Error("alpha must be a number.");
   if (Number.isNaN(alpha)) throw new Error("alpha must not be NaN.");
   if (!Number.isFinite(alpha)) throw new Error("alpha must be finite.");
   if (incx <= 0 || incy <= 0)
@@ -56,9 +56,10 @@ export async function saxpy(device, n, alpha, x, incx, y, incy) {
   let readBuffer = null;
 
   try {
-    xBuffer = xIsGpu ? x._buf : uploadBuffer(x, "saxpy-x", false);
-    yBuffer = yIsGpu ? y._buf : uploadBuffer(y, "saxpy-y", true);
+    xBuffer = xIsGpu ? x._buf : uploadBuffer(device, x, "saxpy-x", false);
+    yBuffer = yIsGpu ? y._buf : uploadBuffer(device, y, "saxpy-y", true);
     paramsBuffer = createParamsBuffer(
+      device,
       [
         { value: n, type: "u32" },
         { value: alpha, type: "f32" },
@@ -68,23 +69,25 @@ export async function saxpy(device, n, alpha, x, incx, y, incy) {
       "saxpy-params",
     );
 
-    const bindGroup = createBindGroup(pipeline.getBindGroupLayout(0), [
+    const bindGroup = createBindGroup(device, pipeline.getBindGroupLayout(0), [
       xBuffer,
       yBuffer,
       paramsBuffer,
     ]);
     const { commandEncoder, ts } = runComputePass(
+      device,
       pipeline,
       bindGroup,
-      calcWorkgroups(n),
+      calcWorkgroups(device, n),
     );
-    readBuffer = yIsGpu ? null : stageReadback(commandEncoder, yBuffer);
+    readBuffer = yIsGpu ? null : stageReadback(device, commandEncoder, yBuffer);
 
-    submit(commandEncoder);
+    submit(device, commandEncoder);
 
     const gpuTimeMs = await extractTimestamp(ts);
 
-    if (yIsGpu && xIsGpu) {
+    if (yIsGpu) {
+      // xIsGpu === yIsGpu, enforced above
       if (gpuTimeMs !== undefined) return { gpuTimeMs };
       return {};
     }

package/src/scopy/scopy.d.mts

@@ -1,7 +1,7 @@
 import { GpuVector } from "../classes/GpuVector.mjs";
 
 /**
- * Performs the operation y = x
+ * Performs the operation $$y \leftarrow x$$
  *
  * {@includeCode ../../examples/scopy/scopy.js}
  *
@@ -27,7 +27,7 @@ export declare function scopy(
 ): Promise<{ y: Float32Array } | { y: Float32Array; gpuTimeMs: number }>;
 
 /**
- * Performs the operation y = x
+ * Performs the operation $$y \leftarrow x$$
  *
  * {@includeCode ../../examples/scopy/gpu.scopy.js}
  *

package/src/scopy/scopy.mjs

@@ -11,13 +11,14 @@ import { extractTimestamp } from "../util/benchmark.mjs";
 import { getPipeline } from "../util/pipeline.mjs";
 import { calcWorkgroups } from "../util/workgroup.mjs";
 import { GpuVector } from "../classes/GpuVector.mjs";
+import { requireGpuDevice, requireSameDevice } from "../util/device.mjs";
 
 export async function scopy(device, n, x, incx, y, incy) {
   const xIsGpu = x instanceof GpuVector;
   const yIsGpu = y instanceof GpuVector;
 
-  if (!(device instanceof GPUDevice))
-    throw new Error("device must be a GPUDevice.");
+  requireGpuDevice(device);
+  requireSameDevice(device, "scopy", { x, y });
   if (
     !Number.isInteger(n) ||
     !Number.isInteger(incx) ||
@@ -52,9 +53,10 @@ export async function scopy(device, n, x, incx, y, incy) {
   let readBuffer = null;
 
   try {
-    xBuffer = xIsGpu ? x._buf : uploadBuffer(x, "scopy-x", false);
-    yBuffer = yIsGpu ? y._buf : uploadBuffer(y, "scopy-y", true);
+    xBuffer = xIsGpu ? x._buf : uploadBuffer(device, x, "scopy-x", false);
+    yBuffer = yIsGpu ? y._buf : uploadBuffer(device, y, "scopy-y", true);
     paramsBuffer = createParamsBuffer(
+      device,
       [
         { value: n, type: "u32" },
         { value: incx, type: "u32" },
@@ -63,23 +65,25 @@ export async function scopy(device, n, x, incx, y, incy) {
       "scopy-params",
     );
 
-    const bindGroup = createBindGroup(pipeline.getBindGroupLayout(0), [
+    const bindGroup = createBindGroup(device, pipeline.getBindGroupLayout(0), [
       xBuffer,
       yBuffer,
       paramsBuffer,
     ]);
     const { commandEncoder, ts } = runComputePass(
+      device,
       pipeline,
       bindGroup,
-      calcWorkgroups(n),
+      calcWorkgroups(device, n),
     );
-    readBuffer = yIsGpu ? null : stageReadback(commandEncoder, yBuffer);
+    readBuffer = yIsGpu ? null : stageReadback(device, commandEncoder, yBuffer);
 
-    submit(commandEncoder);
+    submit(device, commandEncoder);
 
     const gpuTimeMs = await extractTimestamp(ts);
 
-    if (yIsGpu && xIsGpu) {
+    if (yIsGpu) {
+      // xIsGpu === yIsGpu, enforced above
       if (gpuTimeMs !== undefined) return { gpuTimeMs };
       return {};
     }

package/src/sdot/sdot.d.mts

@@ -1,7 +1,7 @@
 import { GpuVector } from "../classes/GpuVector.mjs";
 
 /**
- * Computes the dot product of two vectors: result = sum(x[i] * y[i])
+ * Computes the dot product of two vectors: $$\text{result} = \sum_{i} x_i y_i$$
  *
  * {@includeCode ../../examples/sdot/sdot.js}
  *
@@ -28,7 +28,7 @@ export declare function sdot(
 ): Promise<{ dot: number } | { dot: number; gpuTimeMs: number }>;
 
 /**
- * Computes the dot product of two vectors: result = sum(x[i] * y[i])
+ * Computes the dot product of two vectors: $$\text{result} = \sum_{i} x_i y_i$$
  *
  * {@includeCode ../../examples/sdot/gpu.sdot.js}
  *