# corsix/amx

Apple AMX Instruction Set

Repository: https://github.com/corsix/amx
Canonical: https://ross.abutalabs.com/products/amx
Language: C
License: MIT
License Family: permissive
Last push: 2024-12-26T15:29:28+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 1464, "days_push": 615, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1261, forks 59 (observed 2026-08-28T04:04:10.158064+00:00)

## What it is
A C library and documentation of Apple's undocumented AMX (Apple Matrix eXtension) instructions found on M1-M4 Apple Silicon chips, enabling CPU-issued matrix and vector multiply-accumulate operations on a special accelerator unit. It includes reverse-engineered instruction encodings, header files, and research notes across chip generations.

## Use cases
- use Apple AMX matrix accelerator on M1/M2/M3/M4
- write fast matrix multiply code on Apple Silicon
- learn undocumented Apple AMX instruction encodings
- accelerate linear algebra without GPU or Neural Engine
- compare AMX vs NEON performance on Apple chips
- access bf16 matrix operations on Apple Silicon

## When to choose
- You need maximum CPU-side matrix/vector compute throughput on Apple M-series hardware
- You are doing low-level performance research or reverse engineering of Apple Silicon
- You want to avoid GPU/Neural Engine overhead for small matrix workloads

## When to avoid
- You need documented, officially supported APIs (use Accelerate, Metal, or Core ML instead)
- You target non-Apple or pre-M1 hardware
- You need long-term stability, since AMX is undocumented and may change between chip generations

## Facets
- artifact type: library
- maturity: active
- function: gpu-computing, math, developer-tools
- domain: hardware, performance, developer-tools
- platform: cpp, c
- tags: apple-silicon, amx, matrix-multiply, undocumented-instructions, arm64, simd, reverse-engineering, algorithms, macos

## Member repositories
- corsix/amx (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:10.158064+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T05:04:18.279994+00:00, confidence not recorded.
  - readme: https://github.com/corsix/amx (fetched 2026-08-28T04:04:10.158064+00:00, sha 7ec2b16386fa)
- Data as of 2026-08-30T08:39:29.467469+00:00.