PHP GPU Tensors

PHP GPU Tensors

PHP GPU Tensors is a native PHP extension designed specifically for GPU computing. This extension provides efficient handling of tensor operations and machine-learning tasks leveraging NVIDIA GPU acceleration. It enables developers to create tensor operations and data pipelines directly in PHP, without the need for a Python runtime. The core features include:

Key Features

• GPU Tensors: Utilize the CudaArray class to perform a range of operations including arithmetic, broadcasting, comparisons, and matrix multiplications. The library supports batch processing and reductions, as well as slicing similar to Python’s style.
• Explicit Data Movement: Control data transfers between the host and the GPU explicitly, with options for packed buffers and .npy file imports to optimize execution.
• Custom CUDA Kernels: Compile and execute custom CUDA C++ kernels at runtime with NVRTC, facilitating high-performance tasks directly within PHP.
• Kernel Fusion: Improve performance by fusing tensor operations, enabling quicker execution and reduced overhead during repetitive processes. Fused execution can significantly enhance training workloads by minimizing computation time.
• Low-Level Framework: Unlike comprehensive machine-learning frameworks, PHP GPU Tensors focuses on providing a low-level, efficient library for tensor computation without built-in support for features like automatic differentiation or CPU fallbacks.

Example Usage

Here is a quick example demonstrating basic tensor operations:

use Cuda\CudaArray;

$a = CudaArray::ones([4]);
$b = CudaArray::full([4], 2.0);
$c = CudaArray::full([4], 3.0);

// Eager execution (default): one GPU operation per call.
$eager = $a + $b * $c;

// Fusion (opt-in): capture once, replay as fused GPU kernels.
$plan = Fusion::compile(fn($a, $b, $c) => $a + $b * $c, inputs: [$a, $b, $c]);
$fused = $plan->run($a, $b, $c);

print_r($fused->toArray()); // [7, 7, 7, 7]

This code snippet highlights how to perform tensor arithmetic operations seamlessly using PHP.

Performance Results

The efficiency of Fusion compared to eager execution has shown notable improvements:

This performance data illustrates the potential benefits of adopting kernel fusion in training routines.

Installation

The extension is available via Packagist and can be easily installed by running:

pie install lcmialichi/php-gpu-tensors:0.1.0-beta.4

Conclusion

PHP GPU Tensors opens new possibilities for conducting high-performance tensor operations directly within PHP environments. By utilizing CUDA for acceleration, it empowers developers to harness the speed and efficiency of GPU computing for machine learning and numerical workloads.