ARM-software/ComputeLibrary
A collection of low-level machine learning functions optimized for ARM Cortex and Mali GPU architectures.

Velocity · 7d
+0.9
★ / day
Trend
→steady
star history
The Compute Library provides optimized implementations of neural network and computer vision operations including multiple convolution algorithms (GeMM, Winograd, FFT, Direct). It supports acceleration via SIMD technologies (NEON, SVE) on CPUs and OpenCL on GPUs. The library targets multiple data types (FP32, FP16, INT8, UINT8, BFLOAT16) and includes advanced optimization techniques such as kernel fusion and micro-architecture-specific tuning.