🐛 Describe the bug

Number of functional/performance related regressions associated with oneDNN update seems to have increased recently, which points to the lack of test plan/coverage when such updates happen.

Creating this issue to track recent regressions and how they were fixed.

• [inductor][cpu]rexnet100 dynamic FP32 multiple threads performance regression #115346
• Segmentation Fault in PyTorch 2.2 with F.conv3d function #120406
• permute + conv2d crash/memory corruption on CPU in torch 2.2 #120211
• F.conv1d fails with RuntimeError: cannot create std::vector larger than max_size() after transpose of input tensor #120547
• Pytorch 2.1.X SIGILL on CPUs that do not support SSE4.1 #116623
• python -c "import torch;print(torch.nn.GELU()(torch.rand(2)))" crashes on aarch64 #115482
• [aarch64] nn.Linear(20, 1) inference fails #114750
• Runnings SentenceTransformer encoding step causes Docker containers on Mac (Silicon) to crash with code 139 #111695
• Silent incorrect result for addmm for noncontiguous input #104439
• conv_transpose1d is 1000x times slow in torch 2.2.1+cpu vs torch 1.13.1+cpu #120982 (not verified yet)
• Operator performance plummets on PyTorch 2.2.1 Windows platform #120788 (not verified yet)

Versions

2.2/nightly

cc @ezyang @gchanan @zou3519 @kadeng @gujinghui @PenghuiCheng @XiaobingSuper @jianyuh @jgong5 @mingfeima @sanchitintel @ashokei @jingxu10 @min-jean-cho @yanbing-j @Guobing-Chen @Xia-Weiwen