google/gemmlowp

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1843 stars · 462 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

gemmlowp is a small, self-contained, headers-only C++ library for low-precision (quantized) general matrix multiplication (GEMM). It provides optimized code paths for ARM NEON and x86 SSE4.1 and is widely used as a quantized inference kernel in machine-learning runtimes like TensorFlow Lite.

Use cases

• perform quantized 8-bit matrix multiplication on mobile CPUs
• accelerate low-precision GEMM kernels for neural network inference
• embed a lightweight matrix multiply library in an embedded ARM project
• speed up integer matrix multiplication with NEON or SSE4.1 intrinsics
• implement quantized convolution layers for on-device ML
• profile and benchmark low-precision GEMM performance

When to choose

• you need fast quantized/low-precision GEMM on ARM or x86 with minimal dependencies
• you want a header-only library that is easy to embed in C++11 projects
• you are building or extending a mobile or embedded ML inference engine

When to avoid

• you need a full linear algebra library (BLAS, solvers, decompositions)
• you need floating-point GEMM with broad GPU acceleration
• you require frequent updates or official vendor support, as the project is in maintenance mode

Facets

library · maturity maintenance

machine-learning math machine-learning deep-learning performance cpp windows embedded cross-platform gemm low-precision quantization header-only matrix-multiplication neon simd inference algorithms linux macos android

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google/gemmlowp")