MIOpen
AMDMIOpen is AMD's library of high-performance machine learning primitives for ROCm, covering convolution, RNN, pooling, batch normalization and related deep learning kernels - the same role cuDNN plays for NVIDIA. AMD develops it as part of the ROCm stack; the project is being folded into the consolidated ROCm/rocm-libraries monorepo, and this repository is kept as a retired, read-only mirror during that migration.
Openness
5 high confidence- license
- MIT(OSI)
- source
- public
- core features withheld
- no
MIT license body confirmed. The repository is public and unarchived and builds the whole product, and the README describes no paid tier, enterprise edition or license-gated build beside it, so source is public and the core ungated.
- https://api.github.com/repos/ROCm/MIOpen recorded 2026-09-02
Repo metadata for miopen's canonical repository - private false, archived false, fork false.
- https://github.com/ROCm/MIOpen/blob/develop_deprecated/LICENSE.txt recorded 2026-10-07
LICENSE body opens with an AMD copyright line, declares the MIT text for the whole repository, and separately dual-licenses two files (kernel_cache.hpp/.cpp) under MIT plus an Apache-2.0 grant from their original Vratis, Ltd. author - a widening, not a restriction, since both are OSI tiers.
- https://github.com/ROCm/MIOpen/blob/develop_deprecated/README.md recorded 2026-10-07
README carries a retirement notice pointing to ROCm/rocm-libraries, describes MIOpen as AMD's library for high-performance machine learning primitives, and names no paid tier, enterprise edition or license-gated build beside the published source.
Adoption
2 low confidenceMIOpen ships inside ROCm's own distribution packages (apt, dnf, zypper) and as a source build, and AMD publishes no download count for either channel, leaving GitHub stars as the only adoption signal.
- https://api.github.com/repos/ROCm/MIOpen recorded 2026-09-02
Repo metadata - stargazers_count = 1,193 - for ROCm/MIOpen.
Capability
3 medium confidenceUnlike cutlass and composable-kernel, which PyTorch and FlashAttention vendor directly as submodules, no outside project documents a dependency on MIOpen; it ships bundled inside ROCm instead. That gap in evidence, not a smaller operator surface, is what keeps it well below Apache TVM's retargetable, multi-vendor compilation.
- https://github.com/ROCm/MIOpen/blob/develop_deprecated/README.md recorded 2026-10-07
README carries a retirement notice pointing to ROCm/rocm-libraries, describes MIOpen as AMD's library for high-performance machine learning primitives, and names no paid tier, enterprise edition or license-gated build beside the published source.
Verified 2026-09-27