This is a patch release containing the following changes to v3.14:
- Fixed correctness issues in matmul with binary post-ops on x64 CPUs (719dca4, 80d6f31, b07f2ff)
- Fixed correctness issues in grouped matmul with sparse groups on Intel GPUs (e6789e7, 7d0d404)
- Fixed issues in grouped matmul with post-ops on Intel GPUs (14a8dd2, 87e2ac2, 8925469)
- Improved primitive creation performance for grouped matmul cases on Intel GPUs (a58c5ca, 45d6dc7, e2d6906, 693776d, a8f70c8, cca6ab1, a032339, 0b0b929)
- Updated ZenDNN minimal supported version to v6.1.2 (6259781)
- Fixed correctness issue in batched
f32matmul on x64 CPUs with Intel AVX-512 instruction set support (bcad6d0) - Fixed excessive memory allocation in benchdnn cold-cache mode for small tensors (f2c81be)
- Improved performance of
f32matmul for matrix-vector multiplication cases on x64 CPUs (d6c0515, f6027e6)