PyCL-GPU: A GPGPU Framework for Python

PyCL-GPU is a streamlined, Python-based framework for General-Purpose computing on Graphics Processing Units (GPGPU). It abstracts away the boilerplate of PyOpenCL, providing a clean, modular API for numerical and image processing tasks.

Key Features

• Clean API: Manage GPU contexts, buffers, and kernels with minimal code.
• AMP Integrated: Includes the Adaptive Multiprocessing Pool (AMP) to orchestrate parallel GPU tasks across all available CPU cores.
• Speedmaxxing Methodology: High-velocity optimizations including FP16 vectorization, Memory Grip alignment, and Radix Reordering.
• Intelligent Routing: A* PCIe Data Routing to avoid hardware bus congestion.
• Cross-Platform: Works with any OpenCL-compliant hardware (NVIDIA, AMD, Intel).

Research Paper

The methodology and performance results of this framework are documented in the paper: "Speedmaxxing: Direct Hardware Access and Optimization for GPGPU using PyCL-GPU" (Included as paper.pdf).

Project Structure

• framework/: The core library (Context, Buffer, Program, Task management).
• amp/: Adaptive Multiprocessing Pool implementation.
• kernels/: OpenCL C kernel source files.
• main_dpj.py: Dot-Product Join + Radix Reordering demo.
• main_astar.py: A* PCIe Routing visualization demo.
• main_neural.py: Latency Predictor neural-driven scaling demo.
• main.py: General ensemble example using AMP.

Installation

1. Install core dependencies:

   pip install -r requirements.txt

2. The amp library is now bundled directly in the PyCL-GPU-main folder. No separate installation is required.

Quick Start

Simple Task

from framework.task import GeneralTask
# See main_simple.py for full context

Multiprocessing Ensemble (AMP)

from framework.task import MultiParallelTask
# Uses all CPU cores to feed the GPU

License

Polyform Non-Commercial License 1.0.0. "Personal use, teaching, and research are allowed. For-profit business use requires a separate license."