This is a patch release containing the following changes to v2.7.3:
- Fixed potential
NaNissue in convolution weight gradient on Intel CPUs (6d80bb4, 4c34f89, 017950a, 796a600) - Improved bfloat16 convolution weight gradient performance for processors with Intel AMX support (21bdc21, 82cb7d3, b2e948f, 0a33f70, ff05d0e)
- Fixed out of bounds writes in bfloat16 inner product weight gradient for processors with Intel AMX support (caead72)
- Fixed illegal instruction in matmul for processors with Intel AMX support (be942a2, 28ddb5b, d264ba4)
- Fixed segfault in convolution with depthwise post-op for processors with Intel SSE4.1 support (f708100)
- Worked around segfaults for builds with Intel C/C++ Compiler 2021 for macOS (1382605)
- Fixed segfault in bfloat16 convolution with strides for processors with Intel AMX support (c3b1dcd)
- Fixed correctness issue in int8 convolution with zero points for processors with Intel AMX support (5e76d8b)
- Fixed assertion fail in int8 convolution for processors with Intel AMX support (05629a5)
- Fixed incorrect results in vanilla GRU for Intel CPUs (2089770)
- Improved bfloat16 convolution performance for cases with large number of channels and spatial dimensions (c67f46b, c9cb51d, 4e2c573, 474527f, 87e8ea9)
- Fixed an issue with incorrect header files location when using oneDNN as subproject (be6abca)