uhcr 4.0.0

Universal Hardware-Aware Compute Runtime — a modular, capability-driven execution stack

UHCR

A Python framework for hardware-optimized computation with JIT compilation across x86_64, AArch64, RISC-V, and CUDA.

Full documentation

Install

pip install uhcr

Quick Example

import uhcr

@uhcr.jit(eager=True)
def compute(a, b):
    return (a + b) * 2

Features

• JIT Compilation - Traces Python functions and compiles to native machine code
• Multi-ISA Code Generation - Targets x86_64 (AVX2), AArch64 (NEON), and RISC-V (RVV)
• CUDA Backend - GPU acceleration via PTX JIT for NVIDIA hardware
• Optimization Pipeline - Constant folding, dead code elimination, strength reduction, CSE
• Hardware Detection - Automatic CPUID, GPU probe, and NUMA topology discovery
• Distributed Computing - Protocol server (REST & gRPC) with distributed worker coordination
• Storage Optimization - High-performance memory pooling and hierarchical caching
• Plugin System - Extend with custom backends, kernels, and passes via TOML manifests
• Tensor API - High-level tensor operations dispatched to the optimal backend
• Built-in Benchmarks - Performance measurement suite for comparing execution paths

Documentation

• Architecture
• API Reference
• JIT Guide
• Plugin Guide
• Optimization Passes
• Multi-ISA
• Benchmarks
• Contributing

Experimental Status

UHCR v3.5.0 includes the CLI tool (uhcr) with Docker and Kubernetes containerization support (--docker, --kubernetes), expanded hardware detection (RAM speed/type, CPU cache topology), and multi-ISA code generation. These are now stable and production-ready as of v3.5.0.

License

Apache-2.0