oneapi-src/oneDNN v3.1.1

on GitHub

This is a patch release containing the following changes to v3.1:

• Fixed correctness issue in pooling primitive with post-ops on Intel GPUs (4b7bc1a)
• Fixed segfault in bfloat16 convolution on processors with Intel AMX support (461d55e)
• Fixed correctness issue in deconvolution primitive with post-ops on Intel GPUs based on Xe-LP architecture (c8943f5, ad3c62f)
• Fixed performance regression in int8 convolution primitive with scales (7fa3b6f, bb3ecc4)
• Fixed correctness issue in int8 convolution primitive with zero points on processors with Intel AVX2 and Intel DL Boost support (d721767, f6365b1)
• Fixed performance regression in int8 inner product on processors with Intel AVX-512 and Intel DL Boost or Intel AMX support (2ede31e)
• Fixed segfault in pooling primitive with post-ops on processors with Intel SSE4.1 support (d712173, e4085a7)
• Fixed integer overflow in eltwise primitive on Intel GPUs (1932b3d, be05c33, 148006b, 2e64369, b4423fb, 87fd48f, 9a66ac6, 6ce52eb, 36bf079, 161d2b6, a5ef078, d058bd8)
• Fixed primitive creation error in large 3D convolutions on Intel GPUs (7c23d9e)
• Fixed performance regression in fp32 convolution primitive weight gradient on Intel GPUs (ff209f9, 8710839)
• Fixed primitive creation error in int8 convolution with zero points on Intel GPUs (cb91693, 85e58af)
• Fixed correctness issue in fp32 convolution with Winograd algorithm on Intel GPUs (97ac885)
• Fixed primitive creation error in depthwise convolution on Intel GPUs based on Xe-LP architecture (51d608d)
• Fixed segfault during Graph partition compilation (a5d3568)
• Fixed crashes in inner product with unsupported weight formats on Intel64 CPUs (c0f4e93)
• Fixed an issue with compilation of Graph partitions containing matmul and using destination tensor layout any on Intel GPUs (ab2041d, f2c457d)
• Improved accuracy of eltwise primitive with gelu_erf algorithm on Intel64 CPUs (e67abef)
• Fixed correctness issue in int8 matmul and inner product primitives on Intel GPUs based on Xe-HPG and Xe-HPC architecture (36aa622)
• Fixed potential correctness issue in bfloat16 convolution weight gradient on processors with Intel AMX support (c93e673, 8da1083, f7acf98)
• Fixed memory corruption in inner product weight gradient on processors with Intel AMX support (b56a89e)
• Fixed integer overflow issue in convolution primitive on Intel GPUs (774deab, 663c2e4, 12d5743, 31ac0e0, e3cb07d)
• Fixed correctness issue in matmul primitive with broadcasted bias on Intel GPUs (3ba7e8b)
• Fixed correctness issue in inner product primitive with post-ops on processors with Intel AVX2 support (69260f6)
• Fixed out of bounds prefetching in matmul and inner product primitives on Intel GPUs (2b8f6b1)
• Fixed dispatching issues for fp32 inner product implementation on processors with Intel AVX2 and Intel DL Boost supprt (f27dedb, f8d7c2e)
• Fixed division by zero issue in eltwise and eltwise post-op on Intel GPUs (f5654f5, a18c19e, a7c8cbc, 44355a6)
• Fixed correctness issue for 3D convolution primitive with post-ops (e6b93af)