This is a patch release containing the following changes to v3.13.4:
- Fixed correctness issue in matmul with multiple post-ops on Intel GPUs based on Xe architecture (8afe311)
- Fixed correctness issue in batched
f32matmul on x64 CPUs with Intel AVX-512 instruction set support (067bb12) - Fixed correctness issue in matmul with per-N destination scales (767c027, 31def54)
- Fixed excessive memory allocation in benchdnn cold-cache mode for small tensors (6476bf1)
- Improved performance of
f32matmul for matrix-vector multiplication cases on x64 CPUs (8c86975, f5f6012)