WasmGPU.compute.kernels.luFactorC64Batched¶
Summary¶
luFactorC64Batched performs in-place batched LU factorization with partial pivoting over row-major complex64 matrices stored as interleaved real and imaginary f32 values.
Syntax¶
WasmGPU.compute.kernels.luFactorC64Batched(matrices: StorageBuffer, ipiv: StorageBuffer, batchCount: number, n: number, opts?: KernelDispatchOptions): void
Parameters¶
| Name | Type | Required | Description |
|---|---|---|---|
matrices |
StorageBuffer |
Yes | Row-major complex64 matrices; overwritten with compact L/U factors. |
ipiv |
StorageBuffer |
Yes | Output u32 pivot rows, one per matrix row. |
batchCount |
number |
Yes | Number of matrices; a non-negative integer. |
n |
number |
Yes | Square matrix order; a non-negative integer. |
opts |
KernelDispatchOptions |
No | Label and workgroup-limit validation. External encoders are not supported. |
Returns¶
void - Factorization and pivot data are written into matrices and ipiv.
Type Details¶
type KernelDispatchOptions = {
encoder?: GPUCommandEncoder;
label?: string;
validateLimits?: boolean;
};
matrices needs batchCount * n * n * 8 bytes and ipiv needs batchCount * n * 4 bytes. They must be distinct. Zero dimensions perform no work. This multi-stage kernel rejects opts.encoder and submits its own commands; its internal factorization strategy is an implementation detail.
Example¶
const wgpu = await WasmGPU.create(document.querySelector("canvas"));
const matrices = wgpu.compute.createStorageBuffer({ data: new Float32Array([4, 0, 3, 0, 6, 0, 3, 0]) });
const ipiv = wgpu.compute.createStorageBuffer({ byteLength: 8, copySrc: true });
wgpu.compute.kernels.luFactorC64Batched(matrices, ipiv, 1, 2);