wgblas 0.1.2 → 1.0.0

package/src/ssymv/ssymv.d.mts

@@ -0,0 +1,109 @@
+import { GpuVector } from "../classes/GpuVector.mjs";
+import { GpuMatrix } from "../classes/GpuMatrix.mjs";
+
+/**
+ * Performs the symmetric matrix-vector operation y = alpha * A * x + beta * y
+ *
+ * A is an n×n symmetric matrix stored in row-major order. Only the triangle
+ * specified by `uplo` is referenced; the other triangle is inferred by symmetry.
+ *
+ * {@includeCode ../../examples/ssymv/ssymv.js}
+ *
+ * **Browser (standalone HTML):**
+ * {@includeCode ../../examples/ssymv/web/ssymv.html}
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param n      - order of the matrix A (number of rows and columns)
+ * @param alpha  - scalar multiplier for A*x
+ * @param A      - Float32Array or GpuMatrix, row-major, at least (n-1)*lda+n elements
+ * @param lda    - leading dimension of A (>= n)
+ * @param x      - Float32Array input vector, length at least (n-1)*incx+1
+ * @param incx   - stride for x (must be a positive integer)
+ * @param beta   - scalar multiplier for y
+ * @param y      - Float32Array input/output vector, length at least (n-1)*incy+1
+ * @param incy   - stride for y (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/ssymv/ssymv.mjs#L15">Source code: ssymv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function ssymv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  n: number,
+  alpha: number,
+  A: Float32Array | GpuMatrix,
+  lda: number,
+  x: Float32Array,
+  incx: number,
+  beta: number,
+  y: Float32Array,
+  incy: number,
+): Promise<{ y: Float32Array; gpuTimeMs?: number }>;
+
+/**
+ * Performs the symmetric matrix-vector operation y = alpha * A * x + beta * y
+ *
+ * A is kept GPU-resident; x and y are CPU Float32Arrays.
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param n      - order of the matrix A
+ * @param alpha  - scalar multiplier for A*x
+ * @param A      - GpuMatrix, row-major, GPU-resident
+ * @param lda    - leading dimension of A (must equal A.lda)
+ * @param x      - Float32Array input vector
+ * @param incx   - stride for x (must be a positive integer)
+ * @param beta   - scalar multiplier for y
+ * @param y      - Float32Array input/output vector
+ * @param incy   - stride for y (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/ssymv/ssymv.mjs#L15">Source code: ssymv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function ssymv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  n: number,
+  alpha: number,
+  A: GpuMatrix,
+  lda: number,
+  x: Float32Array,
+  incx: number,
+  beta: number,
+  y: Float32Array,
+  incy: number,
+): Promise<{ y: Float32Array; gpuTimeMs?: number }>;
+
+/**
+ * Performs the symmetric matrix-vector operation y = alpha * A * x + beta * y
+ *
+ * x and y are kept resident on the GPU. A must be a GpuMatrix.
+ *
+ * {@includeCode ../../examples/ssymv/gpuvec.ssymv.js}
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param n      - order of the matrix A
+ * @param alpha  - scalar multiplier for A*x
+ * @param A      - GpuMatrix, row-major, GPU-resident
+ * @param lda    - leading dimension of A (must equal A.lda)
+ * @param x      - GpuVector input vector (not mutated)
+ * @param incx   - stride for x (must be a positive integer)
+ * @param beta   - scalar multiplier for y
+ * @param y      - GpuVector input/output vector (mutated in place)
+ * @param incy   - stride for y (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/ssymv/ssymv.mjs#L15">Source code: ssymv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function ssymv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  n: number,
+  alpha: number,
+  A: GpuMatrix,
+  lda: number,
+  x: GpuVector,
+  incx: number,
+  beta: number,
+  y: GpuVector,
+  incy: number,
+): Promise<{ gpuTimeMs?: number }>;

package/src/ssymv/ssymv.mjs

@@ -0,0 +1,130 @@
+import {
+  uploadBuffer,
+  createParamsBuffer,
+  stageReadback,
+  destroyBuffers,
+} from "../util/buffer.mjs";
+import { createBindGroup } from "../util/bindgroup.mjs";
+import { runComputePass, submit } from "../util/compute.mjs";
+import { extractResult } from "../util/result.mjs";
+import { extractTimestamp } from "../util/benchmark.mjs";
+import { getPipeline } from "../util/pipeline.mjs";
+import { GpuVector } from "../classes/GpuVector.mjs";
+import { GpuMatrix } from "../classes/GpuMatrix.mjs";
+
+export async function ssymv(device, uplo, n, alpha, A, lda, x, incx, beta, y, incy) {
+  const xIsGpu = x instanceof GpuVector;
+  const yIsGpu = y instanceof GpuVector;
+  const AIsGpu = A instanceof GpuMatrix;
+  const isLower = uplo === "lower";
+
+  if (!(device instanceof GPUDevice))
+    throw new Error("device must be a GPUDevice.");
+  if (!isLower && uplo !== "upper")
+    throw new Error("uplo must be 'lower' or 'upper'.");
+  if (
+    !Number.isInteger(n) ||
+    !Number.isInteger(incx) ||
+    !Number.isInteger(incy) ||
+    !Number.isInteger(lda)
+  )
+    throw new Error("n, incx, incy, and lda must be integers.");
+  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 (typeof beta !== "number")
+    throw new Error("beta must be a number.");
+  if (Number.isNaN(beta)) throw new Error("beta must not be NaN.");
+  if (!Number.isFinite(beta)) throw new Error("beta must be finite.");
+  if (incx <= 0 || incy <= 0)
+    throw new Error("incx and incy must be positive.");
+  if (lda < n) throw new Error("lda must be >= n.");
+  if (!AIsGpu && !(A instanceof Float32Array))
+    throw new Error("A must be a Float32Array or GpuMatrix.");
+  if (!xIsGpu && !(x instanceof Float32Array))
+    throw new Error("x must be a Float32Array or GpuVector.");
+  if (!yIsGpu && !(y instanceof Float32Array))
+    throw new Error("y must be a Float32Array or GpuVector.");
+  if (xIsGpu !== yIsGpu)
+    throw new Error(
+      "x and y must be the same type (both Float32Array or both GpuVector).",
+    );
+  if (xIsGpu && !AIsGpu)
+    throw new Error("A must be a GpuMatrix when x and y are GpuVectors.");
+  if (xIsGpu && x._buf === y._buf)
+    throw new Error("x and y must not reference the same GPU buffer when both are GpuVectors.");
+  if (AIsGpu && lda !== A.lda)
+    throw new Error("lda must match A.lda when A is a GpuMatrix.");
+  if (AIsGpu && (A.rows < n || A.cols < n))
+    throw new Error("A is too small for the given n.");
+  if (n < 0) throw new Error("n must be non-negative.");
+  if (n === 0) return yIsGpu ? {} : { y };
+
+  if (!AIsGpu && A.length < (n - 1) * lda + n)
+    throw new Error(
+      "A does not have enough elements for the given n and lda.",
+    );
+  if (x.length < (n - 1) * incx + 1)
+    throw new Error(
+      "x does not have enough elements for the given n and incx.",
+    );
+  if (y.length < (n - 1) * incy + 1)
+    throw new Error(
+      "y does not have enough elements for the given n and incy.",
+    );
+
+  const pipeline = await getPipeline(device, "ssymv");
+
+  let ABuffer = null;
+  let xBuffer = null;
+  let yBuffer = null;
+  let paramsBuffer = null;
+
+  try {
+    ABuffer = AIsGpu ? A._buf : uploadBuffer(A, "ssymv-A", false);
+    xBuffer = xIsGpu ? x._buf : uploadBuffer(x, "ssymv-x", false);
+    yBuffer = yIsGpu ? y._buf : uploadBuffer(y, "ssymv-y", true);
+    paramsBuffer = createParamsBuffer(
+      [
+        { value: n,               type: "u32" },
+        { value: alpha,           type: "f32" },
+        { value: beta,            type: "f32" },
+        { value: incx,            type: "u32" },
+        { value: incy,            type: "u32" },
+        { value: lda,             type: "u32" },
+        { value: isLower ? 0 : 1, type: "u32" },
+      ],
+      "ssymv-params",
+    );
+
+    const bindGroup = createBindGroup(pipeline.getBindGroupLayout(0), [
+      ABuffer,
+      xBuffer,
+      yBuffer,
+      paramsBuffer,
+    ]);
+
+    const wgCount = Math.min(n, device.limits.maxComputeWorkgroupsPerDimension);
+    const { commandEncoder, ts } = runComputePass(pipeline, bindGroup, wgCount);
+    const readBuffer = yIsGpu ? null : stageReadback(commandEncoder, yBuffer);
+
+    submit(commandEncoder);
+
+    const gpuTimeMs = await extractTimestamp(ts);
+
+    if (yIsGpu) {
+      if (gpuTimeMs !== undefined) return { gpuTimeMs };
+      return {};
+    }
+
+    const result = await extractResult(readBuffer, Float32Array);
+    if (gpuTimeMs !== undefined) return { y: result, gpuTimeMs };
+    return { y: result };
+  } finally {
+    if (!AIsGpu && ABuffer) destroyBuffers(ABuffer);
+    if (!xIsGpu && xBuffer) destroyBuffers(xBuffer);
+    if (!yIsGpu && yBuffer) destroyBuffers(yBuffer);
+    if (paramsBuffer) destroyBuffers(paramsBuffer);
+  }
+}

package/src/strmv/strmv.d.mts

@@ -0,0 +1,109 @@
+import { GpuVector } from "../classes/GpuVector.mjs";
+import { GpuMatrix } from "../classes/GpuMatrix.mjs";
+
+/**
+ * Performs the triangular matrix-vector operation y = op(A) * x
+ *
+ * A is an n×n triangular matrix stored in row-major order. Only the triangle
+ * specified by `uplo` is referenced; the other triangle is not accessed.
+ *
+ * {@includeCode ../../examples/strmv/strmv.js}
+ *
+ * **Browser (standalone HTML):**
+ * {@includeCode ../../examples/strmv/web/strmv.html}
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param trans  - `'no-transpose'` for A, `'transpose'` for A^T
+ * @param diag   - `'unit'` to treat the diagonal as all-ones (A's diagonal is not read), `'non-unit'` to read it
+ * @param n      - order of the matrix A (number of rows and columns)
+ * @param A      - Float32Array or GpuMatrix, row-major, at least (n-1)*lda+n elements
+ * @param lda    - leading dimension of A (>= n)
+ * @param x      - Float32Array input vector, length at least (n-1)*incx+1
+ * @param incx   - stride for x (must be a positive integer)
+ * @param y      - Float32Array output vector, length at least (n-1)*incy+1
+ * @param incy   - stride for y (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/strmv/strmv.mjs#L15">Source code: strmv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function strmv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  trans: 'no-transpose' | 'transpose',
+  diag: 'unit' | 'non-unit',
+  n: number,
+  A: Float32Array | GpuMatrix,
+  lda: number,
+  x: Float32Array,
+  incx: number,
+  y: Float32Array,
+  incy: number,
+): Promise<{ y: Float32Array; gpuTimeMs?: number }>;
+
+/**
+ * Performs the triangular matrix-vector operation y = op(A) * x
+ *
+ * A is kept GPU-resident; x and y are CPU Float32Arrays.
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param trans  - `'no-transpose'` for A, `'transpose'` for A^T
+ * @param diag   - `'unit'` to treat the diagonal as all-ones (A's diagonal is not read), `'non-unit'` to read it
+ * @param n      - order of the matrix A
+ * @param A      - GpuMatrix, row-major, GPU-resident
+ * @param lda    - leading dimension of A (must equal A.lda)
+ * @param x      - Float32Array input vector
+ * @param incx   - stride for x (must be a positive integer)
+ * @param y      - Float32Array output vector
+ * @param incy   - stride for y (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/strmv/strmv.mjs#L15">Source code: strmv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function strmv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  trans: 'no-transpose' | 'transpose',
+  diag: 'unit' | 'non-unit',
+  n: number,
+  A: GpuMatrix,
+  lda: number,
+  x: Float32Array,
+  incx: number,
+  y: Float32Array,
+  incy: number,
+): Promise<{ y: Float32Array; gpuTimeMs?: number }>;
+
+/**
+ * Performs the triangular matrix-vector operation y = op(A) * x
+ *
+ * x and y are kept resident on the GPU. A must be a GpuMatrix.
+ *
+ * {@includeCode ../../examples/strmv/gpuvec.strmv.js}
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param trans  - `'no-transpose'` for A, `'transpose'` for A^T
+ * @param diag   - `'unit'` to treat the diagonal as all-ones (A's diagonal is not read), `'non-unit'` to read it
+ * @param n      - order of the matrix A
+ * @param A      - GpuMatrix, row-major, GPU-resident
+ * @param lda    - leading dimension of A (must equal A.lda)
+ * @param x      - GpuVector input vector (not mutated)
+ * @param incx   - stride for x (must be a positive integer)
+ * @param y      - GpuVector output vector (mutated in place)
+ * @param incy   - stride for y (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/strmv/strmv.mjs#L15">Source code: strmv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function strmv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  trans: 'no-transpose' | 'transpose',
+  diag: 'unit' | 'non-unit',
+  n: number,
+  A: GpuMatrix,
+  lda: number,
+  x: GpuVector,
+  incx: number,
+  y: GpuVector,
+  incy: number,
+): Promise<{ gpuTimeMs?: number }>;

package/src/strmv/strmv.mjs

@@ -0,0 +1,132 @@
+import {
+  uploadBuffer,
+  createParamsBuffer,
+  stageReadback,
+  destroyBuffers,
+} from "../util/buffer.mjs";
+import { createBindGroup } from "../util/bindgroup.mjs";
+import { runComputePass, submit } from "../util/compute.mjs";
+import { extractResult } from "../util/result.mjs";
+import { extractTimestamp } from "../util/benchmark.mjs";
+import { getPipeline } from "../util/pipeline.mjs";
+import { GpuVector } from "../classes/GpuVector.mjs";
+import { GpuMatrix } from "../classes/GpuMatrix.mjs";
+
+export async function strmv(device, uplo, trans, diag, n, A, lda, x, incx, y, incy) {
+  const xIsGpu = x instanceof GpuVector;
+  const yIsGpu = y instanceof GpuVector;
+  const AIsGpu = A instanceof GpuMatrix;
+  const isLower = uplo === "lower";
+  const isNoTrans = trans === "no-transpose";
+  const isUnit = diag === "unit";
+
+  if (!(device instanceof GPUDevice))
+    throw new Error("device must be a GPUDevice.");
+  if (!isLower && uplo !== "upper")
+    throw new Error("uplo must be 'lower' or 'upper'.");
+  if (!isNoTrans && trans !== "transpose")
+    throw new Error("trans must be 'no-transpose' or 'transpose'.");
+  if (!isUnit && diag !== "non-unit")
+    throw new Error("diag must be 'unit' or 'non-unit'.");
+  if (
+    !Number.isInteger(n) ||
+    !Number.isInteger(incx) ||
+    !Number.isInteger(incy) ||
+    !Number.isInteger(lda)
+  )
+    throw new Error("n, incx, incy, and lda must be integers.");
+  if (incx <= 0 || incy <= 0)
+    throw new Error("incx and incy must be positive.");
+  if (lda < n) throw new Error("lda must be >= n.");
+  if (!AIsGpu && !(A instanceof Float32Array))
+    throw new Error("A must be a Float32Array or GpuMatrix.");
+  if (!xIsGpu && !(x instanceof Float32Array))
+    throw new Error("x must be a Float32Array or GpuVector.");
+  if (!yIsGpu && !(y instanceof Float32Array))
+    throw new Error("y must be a Float32Array or GpuVector.");
+  if (xIsGpu !== yIsGpu)
+    throw new Error(
+      "x and y must be the same type (both Float32Array or both GpuVector).",
+    );
+  if (xIsGpu && x._buf === y._buf)
+    throw new Error(
+      "x and y must not reference the same GPU buffer when both are GpuVectors.",
+    );
+  if (xIsGpu && !AIsGpu)
+    throw new Error("A must be a GpuMatrix when x and y are GpuVectors.");
+  if (AIsGpu && yIsGpu && A._buf === y._buf)
+    throw new Error("A and y must not reference the same GPU buffer.");
+  if (AIsGpu && lda !== A.lda)
+    throw new Error("lda must match A.lda when A is a GpuMatrix.");
+  if (AIsGpu && (A.rows < n || A.cols < n))
+    throw new Error("A is too small for the given n.");
+  if (n < 0) throw new Error("n must be non-negative.");
+  if (n === 0) return yIsGpu ? {} : { y };
+
+  if (!AIsGpu && A.length < (n - 1) * lda + n)
+    throw new Error(
+      "A does not have enough elements for the given n and lda.",
+    );
+  if (x.length < (n - 1) * incx + 1)
+    throw new Error(
+      "x does not have enough elements for the given n and incx.",
+    );
+  if (y.length < (n - 1) * incy + 1)
+    throw new Error(
+      "y does not have enough elements for the given n and incy.",
+    );
+
+  const pipeline = await getPipeline(device, "strmv");
+
+  let ABuffer = null;
+  let xBuffer = null;
+  let yBuffer = null;
+  let paramsBuffer = null;
+
+  try {
+    ABuffer = AIsGpu ? A._buf : uploadBuffer(A, "strmv-A", false);
+    xBuffer = xIsGpu ? x._buf : uploadBuffer(x, "strmv-x", false);
+    yBuffer = yIsGpu ? y._buf : uploadBuffer(y, "strmv-y", true);
+    paramsBuffer = createParamsBuffer(
+      [
+        { value: n,             type: "u32" },
+        { value: incx,          type: "u32" },
+        { value: incy,          type: "u32" },
+        { value: lda,           type: "u32" },
+        { value: isNoTrans ? 0 : 1, type: "u32" },
+        { value: isLower ? 0 : 1,   type: "u32" },
+        { value: isUnit ? 1 : 0,    type: "u32" },
+      ],
+      "strmv-params",
+    );
+
+    const bindGroup = createBindGroup(pipeline.getBindGroupLayout(0), [
+      ABuffer,
+      xBuffer,
+      yBuffer,
+      paramsBuffer,
+    ]);
+
+    const wgCount = Math.min(n, device.limits.maxComputeWorkgroupsPerDimension);
+    const { commandEncoder, ts } = runComputePass(pipeline, bindGroup, wgCount);
+    const readBuffer = yIsGpu ? null : stageReadback(commandEncoder, yBuffer);
+
+    submit(commandEncoder);
+
+    const gpuTimeMs = await extractTimestamp(ts);
+
+    if (yIsGpu) {
+      if (gpuTimeMs !== undefined) return { gpuTimeMs };
+      return {};
+    }
+
+    const result = await extractResult(readBuffer, Float32Array);
+    if (gpuTimeMs !== undefined) return { y: result, gpuTimeMs };
+    return { y: result };
+  } finally {
+    if (!AIsGpu && ABuffer) destroyBuffers(ABuffer);
+    if (!xIsGpu && xBuffer) destroyBuffers(xBuffer);
+    if (!yIsGpu && yBuffer) destroyBuffers(yBuffer);
+    if (paramsBuffer) destroyBuffers(paramsBuffer);
+  }
+}

package/src/strsv/strsv.d.mts

@@ -0,0 +1,98 @@
+import { GpuVector } from "../classes/GpuVector.mjs";
+import { GpuMatrix } from "../classes/GpuMatrix.mjs";
+
+/**
+ * Solves the triangular system op(A) * x = b for x, in place (x holds b on
+ * input, the solution on output).
+ *
+ * A is an n×n triangular matrix stored in row-major order. Only the triangle
+ * specified by `uplo` is referenced; the other triangle is not accessed.
+ *
+ * {@includeCode ../../examples/strsv/strsv.js}
+ *
+ * **Browser (standalone HTML):**
+ * {@includeCode ../../examples/strsv/web/strsv.html}
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param trans  - `'no-transpose'` to solve A*x=b, `'transpose'` to solve A^T*x=b
+ * @param diag   - `'unit'` to treat the diagonal as all-ones (A's diagonal is not read), `'non-unit'` to read it
+ * @param n      - order of the matrix A (number of rows and columns)
+ * @param A      - Float32Array or GpuMatrix, row-major, at least (n-1)*lda+n elements
+ * @param lda    - leading dimension of A (>= n)
+ * @param x      - Float32Array holding b on input, the solution on output; length at least (n-1)*incx+1
+ * @param incx   - stride for x (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/strsv/strsv.mjs#L15">Source code: strsv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function strsv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  trans: 'no-transpose' | 'transpose',
+  diag: 'unit' | 'non-unit',
+  n: number,
+  A: Float32Array | GpuMatrix,
+  lda: number,
+  x: Float32Array,
+  incx: number,
+): Promise<{ x: Float32Array; gpuTimeMs?: number }>;
+
+/**
+ * Solves the triangular system op(A) * x = b for x, in place.
+ *
+ * A is kept GPU-resident; x is a CPU Float32Array.
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param trans  - `'no-transpose'` to solve A*x=b, `'transpose'` to solve A^T*x=b
+ * @param diag   - `'unit'` to treat the diagonal as all-ones (A's diagonal is not read), `'non-unit'` to read it
+ * @param n      - order of the matrix A
+ * @param A      - GpuMatrix, row-major, GPU-resident
+ * @param lda    - leading dimension of A (must equal A.lda)
+ * @param x      - Float32Array holding b on input, the solution on output
+ * @param incx   - stride for x (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/strsv/strsv.mjs#L15">Source code: strsv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function strsv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  trans: 'no-transpose' | 'transpose',
+  diag: 'unit' | 'non-unit',
+  n: number,
+  A: GpuMatrix,
+  lda: number,
+  x: Float32Array,
+  incx: number,
+): Promise<{ x: Float32Array; gpuTimeMs?: number }>;
+
+/**
+ * Solves the triangular system op(A) * x = b for x, in place.
+ *
+ * x is kept resident on the GPU (mutated in place). A must be a GpuMatrix.
+ *
+ * {@includeCode ../../examples/strsv/gpuvec.strsv.js}
+ *
+ * @param device - GPUDevice from `init()`
+ * @param uplo   - `'lower'` to use the lower triangle, `'upper'` to use the upper triangle
+ * @param trans  - `'no-transpose'` to solve A*x=b, `'transpose'` to solve A^T*x=b
+ * @param diag   - `'unit'` to treat the diagonal as all-ones (A's diagonal is not read), `'non-unit'` to read it
+ * @param n      - order of the matrix A
+ * @param A      - GpuMatrix, row-major, GPU-resident
+ * @param lda    - leading dimension of A (must equal A.lda)
+ * @param x      - GpuVector holding b on input, the solution on output (mutated in place)
+ * @param incx   - stride for x (must be a positive integer)
+ * @see <a href="https://github.com/manit2004/wgblas/blob/main/src/strsv/strsv.mjs#L15">Source code: strsv.mjs (L15)</a>
+ * @category BLAS Level 2
+ */
+export declare function strsv(
+  device: GPUDevice,
+  uplo: 'lower' | 'upper',
+  trans: 'no-transpose' | 'transpose',
+  diag: 'unit' | 'non-unit',
+  n: number,
+  A: GpuMatrix,
+  lda: number,
+  x: GpuVector,
+  incx: number,
+): Promise<{ gpuTimeMs?: number }>;