TaCuda — Node.js Bindings

Node.js package for the native tacuda CUDA library. Full TA-Lib coverage (161 functions): indicators, candlestick patterns, MACDEXT, 19 vector math kernels, precision (_ex) variants, and the zero-copy device API.

Prerequisites

• Node.js >= 18.0.0
• NVIDIA GPU (Compute Capability >= 6.0)
• CUDA Toolkit 11.x or 12.x
• Built tacuda shared library (tacuda.dll / libtacuda.so / libtacuda.dylib)

Package Scope

• JavaScript wrappers for the complete 395-function native surface (host + device + batch API)
• Structured OHLCV inputs (Ohlcv class or plain {open, high, low, close, volume} objects)
• Batched multi-symbol execution: rsiBatch, smaBatch, macdBatch, ..., device.*Batch
• Full candlestick registry via cdlPattern() and CDL_NAMES
• Works with Float32Array inputs (and Float64Array for _ex precision variants)
• tacuda.device.* namespace for zero-copy GPU pipelines, tacuda.bufferPool controls

Installation

npm install tacuda

Note: The tacuda native library must be built separately and placed where the Node.js bindings can find it. See "Library Location" below.

Quick Start

const tacuda = require('tacuda');

const close = Float32Array.from(
    { length: 100000 }, (_, i) => 100 + Math.sin(i * 0.01) * 10
);

// Moving averages
const sma20 = tacuda.sma(close, 20);
const ema12 = tacuda.ema(close, 12);

// RSI
const rsi14 = tacuda.rsi(close, 14);

// MACD (returns { macd, signal, hist })
const macdResult = tacuda.macd(close, 12, 26, 9);

// Bollinger Bands (returns { upper, middle, lower })
const bbands = tacuda.bbands(close, 20, 2.0, 2.0);

// MACDEXT with independent MA types (returns { macd, signal, hist })
const ext = tacuda.macdext(close, 12, tacuda.MA_TYPE.EMA,
                           26, tacuda.MA_TYPE.SMA, 9, tacuda.MA_TYPE.WMA);

// Vector math (element-wise GPU kernels)
const logClose = tacuda.ln(close);
const spread = tacuda.sub(high, low);

// Candlestick patterns (63 available)
const doji = tacuda.cdlDoji(open, high, low, close);
const hammer = tacuda.cdlPattern('hammer', open, high, low, close);

console.log(`RSI[100]: ${rsi14[100]}`);
console.log(`MACD signal: ${macdResult.signal[100]}`);

Zero-copy device pipeline

const n = close.length;
const dIn = tacuda.device.alloc(n);      // GPU buffer (opaque pointer)
const dOut = tacuda.device.alloc(n);

tacuda.device.upload(dIn, close);        // 1 PCIe transfer in
tacuda.device.sma(dIn, dOut, n, 20);     // compute on GPU
tacuda.device.sin(dOut, dOut, n);        // chain more kernels, zero copies
tacuda.device.sync();
const result = tacuda.device.download(dOut, n);  // 1 PCIe transfer out

tacuda.device.free(dIn);
tacuda.device.free(dOut);

Double precision (_ex variants)

const close64 = Float64Array.from(close);
const sma64 = tacuda.smaEx(close64, 20);   // float64 storage + compute
const { macd } = tacuda.macdEx(close64, 12, 26, 9, tacuda.MA_TYPE.EMA);

Available Indicators (full TA-Lib coverage)

Moving Averages

sma, ema, wma, dema, tema, trima, kama, t3, mama, mavp, ma

Momentum & Oscillators

rsi, macd, macdext, macdfix, macdLine, stochastic, stochf, stochRsi, cci, cmo, roc, rocp, rocr, rocr100, momentum, change, willr, apo, ppo, pvo, trix, ultosc, bop

Volatility

atr, natr, trange, bbands, accbands, stddev, variance

Trend & Directional

adx, adxr, dx, plusDi, minusDi, plusDm, minusDm, sar, sarext, aroon, aroonOsc

Volume

ad, adosc, obv, mfi, imi, nvi, pvi

Statistical

linearReg, linearRegSlope, linearRegIntercept, linearRegAngle, tsf, correl, beta, avgdev, sum, max, min, maxIndex, minIndex, minMax, minMaxIndex, midpoint

Price Transform

avgPrice, medPrice, typPrice, wclPrice, midPrice

Hilbert Transform

htDcPeriod, htDcPhase, htPhasor, htSine, htTrendline, htTrendmode

Vector Math (19, element-wise)

sin, cos, tan, asin, acos, atan, sinh, cosh, tanh, exp, ln, log10, sqrt, ceil, floor, add, sub, mult, div

Precision variants (Float64Array in/out)

smaEx, emaEx, rsiEx, macdEx, atrEx, bbandsEx

Device API (tacuda.device.*, raw GPU pointers)

Memory: alloc, allocEx, free, freeEx, upload, uploadEx, download, downloadEx, sync. Indicators: sma, ema, wma, dema, tema, trima, kama, t3, rsi, momentum, roc, rocp, rocr, rocr100, cmo, stddev, var, sum, max, min, linearreg family, tsf, atr, natr, trange, adx, adxr, dx, plusDi, minusDi, willr, cci, macd, macdext, stochastic, stochasticFast, ultosc, bbands, accbands, ad, adosc, obv, mfi, price transforms, aroon, aroonosc, all 19 vector math kernels, _ex precision variants, plus change, trix, maxindex, minindex, ma, macdLine, apo, ppo, pvo. Pool: tacuda.bufferPool.cleanup(), allocationCount(), setMaxCacheBytes(bytes).

Candlestick Patterns (63)

cdlDoji, cdlHammer, cdlEngulfing, cdlMorningStar, cdlEveningStar, cdlThreeWhiteSoldiers, cdlThreeBlackCrows, and 56 more via cdlPattern(name, open, high, low, close).

Full list available via tacuda.CDL_NAMES.

Library Location

The bindings search for the native library in this order:

1. TACUDA_LIBRARY environment variable
2. ../../build/Release/tacuda.dll (relative to this directory)
3. ../../build/Debug/tacuda.dll
4. Common system locations

Running Tests

npm test
# Runs 173 tests covering all indicators, vector math, MACDEXT, _ex and device API

Build & Publish (npm)

cd bindings/nodejs
npm pack --dry-run
npm publish --access public

Changelog

Canonical release history is maintained in CHANGELOG.md.

License

Apache License 2.0 — see LICENSE.