Djordje Novakovic

Merging forward and backward kernels into one

1

Replaced two kernels with one because they are similar. Merged two dropout functions that are calling kernels into one with parameter is_backward to determine which kernel should be executed.

This PR is replacing `std::vector` with `InlineVector` in `TensorDescriptor` for lens/strides. `InlineVector` https://github.com/ROCm/MIOpen/pull/3419 I did some performance testing to compare `InlineVector` with `std::vector`(initialization, elements access etc). The results are that...

This PR is for new, upgraded, Op4dTensorGeneric kernel, this is part of porting kernels from OCL to HIP Below is performance (speed-up and drops in performance) comparison between new Op4dTensorGeneric...