Tacuda 0.7.1

TaCuda — .NET/C# Bindings

Managed bindings for the native tacuda CUDA library. Full TA-Lib coverage (161 functions, 395 native entrypoints incl. the batch API): generated P/Invoke signatures for host API, device API, MACDEXT, 19 vector math kernels, and typed _ex entrypoints. The DllImport name is platform-neutral (tacuda), so the same assembly loads tacuda.dll, libtacuda.so, or libtacuda.dylib automatically.

Prerequisites

• .NET 8.0+ or .NET 9.0+
• 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

• NativeMethods.ct_* methods generated from include/tacuda.h / include/tacuda_device.h
• Precision-aware core entrypoints (*_ex) with runtime dtype pair selection
• OHLCV helper model (OhlcvSeries) for convenient column-major inputs

Installation

NuGet

dotnet add package Tacuda

From Source

# Build the native library first
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build

# Reference the binding project
dotnet add reference bindings/csharp/Tacuda.Bindings.csproj

Note: The tacuda native library must be discoverable at runtime. Place it next to the published binary or add its location to PATH / LD_LIBRARY_PATH.

Quick Start

using Tacuda.Bindings;

// Prepare data
var prices = new float[100000];
var sma = new float[100000];
var rsi = new float[100000];

// Fill prices with your data...

// Calculate indicators (GPU-accelerated)
NativeMethods.ct_sma(prices, sma, prices.Length, 20);
NativeMethods.ct_rsi(prices, rsi, prices.Length, 14);

// MACD (multiple outputs)
var macd = new float[100000];
var signal = new float[100000];
var hist = new float[100000];
NativeMethods.ct_macd(prices, macd, signal, hist, prices.Length, 12, 26, 9, 0);

// OHLCV container
var candles = new OhlcvSeries(open, high, low, close, volume);
var output = new float[candles.Length];
NativeMethods.ct_atr(candles.High, candles.Low, candles.Close, output, candles.Length, 14);

// MACDEXT with independent MA types (1=EMA, 0=SMA, 2=WMA)
NativeMethods.ct_macdext(prices, macd, signal, hist, prices.Length,
                         12, 1, 26, 0, 9, 2);

// Vector math (element-wise GPU kernels)
var logs = new float[100000];
NativeMethods.ct_ln(prices, logs, prices.Length);

// Zero-copy device pipeline
IntPtr dIn = NativeMethods.ct_device_alloc(prices.Length);
IntPtr dOut = NativeMethods.ct_device_alloc(prices.Length);
NativeMethods.ct_device_upload(dIn, prices, prices.Length);
NativeMethods.ct_sma_d(dIn, dOut, prices.Length, 20);   // compute on GPU
NativeMethods.ct_sin_d(dOut, dOut, prices.Length);      // chain kernels
NativeMethods.ct_device_sync();
var result = new float[prices.Length];
NativeMethods.ct_device_download(result, dOut, prices.Length);
NativeMethods.ct_device_free(dIn);
NativeMethods.ct_device_free(dOut);

Available Indicators (full TA-Lib coverage)

All indicators are available as static NativeMethods.ct_* methods:

Device API variants (ct_*_d and ct_*_device) plus ct_device_alloc/upload/download/sync/free and buffer-pool controls are available for zero-copy GPU pipelines.

NuGet Packaging

To build the NuGet package with native libraries:

# 1. Build native libraries for target platforms
# 2. Place them in runtimes/ folder:
#    runtimes/win-x64/native/tacuda.dll
#    runtimes/linux-x64/native/libtacuda.so

# 3. Pack
cd bindings/csharp
dotnet pack -c Release

To publish:

dotnet nuget push bin/Release/*.nupkg --source https://api.nuget.org/v3/index.json --api-key <NUGET_API_KEY>

Regenerating Bindings

After editing include/tacuda.h:

python bindings/generate_bindings.py

CTest verifies that the generated files are up to date.

Running Tests

dotnet run --project bindings/csharp/ConsoleExample -- --test
# Runs 100 tests covering all indicator categories, vector math, MACDEXT and device API

Changelog

Canonical release history is maintained in CHANGELOG.md.

License

Apache License 2.0 — see LICENSE.