Slawomir Siwek

@jiangjiajun please review

Hi @yeliang2258, currently fuses with `elementwise_add` require both elementwise_addends to have same dimensions. That's because `append_sum` combined with `ShareDataWith` accumulates both of these tensors in one memory buffer. So if...

Merged - [x] `abs`, `abs_grad`, `relu6` https://github.com/PaddlePaddle/Paddle/pull/45397 - [x] `gaussian_random` https://github.com/PaddlePaddle/Paddle/pull/45481 - [x] Moving `fluid/mkldnn_reuse.h` to `phi/onednn_reuse.h` https://github.com/PaddlePaddle/Paddle/pull/45626 - [x] `scale` https://github.com/PaddlePaddle/Paddle/pull/45537 - [x] `reduce_sum`, `reduce_sum_grad`, `reduce_mean`, `reduce_mean_grad`, `reduce_min`, `reduce_max`...

**Part II** Merged - [x] `conv2d`, `conv2d_grad` https://github.com/PaddlePaddle/Paddle/pull/46342 - [x] `softmax` https://github.com/PaddlePaddle/Paddle/pull/47339 - [x] `softplus` https://github.com/PaddlePaddle/Paddle/pull/47406 - [x] `pool2d`, `pool2d_grad` https://github.com/PaddlePaddle/Paddle/pull/47423 - [x] `depthwise_conv2d`, `conv3d` https://github.com/PaddlePaddle/Paddle/pull/47658 - [x] `depthwise_conv2d_grad`, `conv3d_grad`...

I checked locally via `test_activation_bf16_mkldnn_op.py` and I get same error message as previously in other OneDNN ops https://github.com/PaddlePaddle/Paddle/pull/45770 _UnavailableError: There are no kernels which are registered in the gelu_grad operator._...

@YuanRisheng I checked your suggestion and after mentioned fix `gelu_grad` kernel is called. Could you please mark CI-Coverage as successful?

> The HardSwishGradKernel in this PR is not called. May be you can modify this unittest in next PR, thanks. @YuanRisheng I enabled HardSwishGrad unit test, as you suggested. It...

> Could you help to review this PR and evaluate these changes on your internal system? After this PR is merged, we can start the migration of remaining kernels @jczaja...

We've run first performance and functional tests and this PR passed everything with expected results. We will run one more test on other platform, to make sure that everything is...

LGTM There are small differences in accuracy (some positive, some negative). We will run more test after this PR is merged, as they are more stable.