@danielsimonjr/mathts-wasm

High-performance WebAssembly module for MathTS, built with AssemblyScript.

Features

• SIMD Acceleration: Utilizes WebAssembly SIMD for vectorized operations
• Zero-Copy Operations: Works directly on typed arrays for minimal overhead
• Comprehensive Math: Scalar, array, matrix, and complex number operations
• Cross-Platform: Works in browsers, Node.js, Deno, and edge runtimes

Building

# Install dependencies
npm install

# Build debug WASM (with debugging info)
npm run asbuild:debug

# Build release WASM (optimized)
npm run asbuild:release

# Build both
npm run asbuild

Usage

The published entry is AssemblyScript-generated glue in build/mathts.js (see package.json exports["."]). Import named functions directly:

import { add_f64, sqrt_f64, sin_f64 } from '@danielsimonjr/mathts-wasm';

console.log(add_f64(2, 3)); // 5
console.log(sqrt_f64(16)); // 4
console.log(sin_f64(Math.PI / 2)); // 1

matrix and functions load the same binary as mathts-as.wasm via their own loaders and SHA-384 verification (core/src/wasm-loader.ts).

Dense decompositions (AS exports)

From algebra/decomposition.ts (used by matrix WASMBackend tests, not public lu()/qr() in JS):

• matrix_lu_decompose, matrix_qr_decompose, matrix_cholesky, matrix_inverse, matrix_determinant

Call heap_reset after managed sequences when using bump-allocated scratch (heap.ts). The functions WASM bridges call it after each managed call; matrix does not — it pools buffers.

Module Structure

assembly/
├── src/
│   ├── index.ts              # AS export surface
│   ├── algebra/decomposition.ts
│   ├── elementwise.ts, heap.ts, poly.ts, signal.ts, sort.ts, special.ts, tridiag.ts
│   ├── types/complex.ts
│   └── ops/                  # scalar, array, matrix, complex-ops, fft, …
├── build/                    # mathts.wasm, mathts.js, mathts.d.ts
├── tests/run.js              # debug wasm smoke tests
└── package.json

Available Operations

Scalar Operations

• Basic: add_f64, sub_f64, mul_f64, div_f64, mod_f64, neg_f64
• Power: sqrt_f64, pow_f64, square_f64, cube_f64, cbrt_f64
• Exp/Log: exp_f64, log_f64, log10_f64, log2_f64
• Trig: sin_f64, cos_f64, tan_f64, asin_f64, acos_f64, atan_f64
• Hyperbolic: sinh_f64, cosh_f64, tanh_f64, asinh_f64, acosh_f64, atanh_f64
• Rounding: abs_f64, floor_f64, ceil_f64, round_f64, trunc_f64

Array Operations

• Reductions: array_sum, array_product, array_mean, array_variance, array_min, array_max
• Norms: array_norm, array_norm_l1, array_norm_linf
• Element-wise: array_add, array_sub, array_mul, array_div, array_scale
• BLAS-like: array_dot, array_axpby, array_distance

Matrix Operations

• Creation: matrix_zeros, matrix_ones, matrix_identity, matrix_diag
• Arithmetic: matrix_add, matrix_sub, matrix_scale, matrix_multiply
• Properties: matrix_trace, matrix_norm_frobenius, matrix_transpose
• BLAS: matrix_gemm, matrix_gemv, matrix_axpy

Complex Operations

• Arithmetic: complex_add, complex_sub, complex_mul, complex_div
• Functions: complex_exp, complex_log, complex_sqrt, complex_pow
• Trig: complex_sin, complex_cos, complex_tan, complex_asin, complex_acos
• Properties: complex_abs, complex_arg, complex_conj

Performance Notes

1. SIMD: Enabled by default for vector operations
2. Memory Layout: Matrices use row-major order
3. Complex Arrays: Use interleaved storage [re0, im0, re1, im1, ...]

Testing

npm run asbuild          # release + debug (npm test needs release for test:diff)
npm test                 # tests/run.js (debug wasm) + nine kernel tests + test:diff (release wasm)

License

MIT