GitLab

* doesn't make much sense to make broadcast use two inputs or handle
dynamic shapes
* compute the common shape for dynamic multibroadcast in the
multibroadcast op
* multibroadcast all combinations of the dynamic inputs

Constructs a dynamic shape from three vectors of lengths,
the minimums, maximums, and optimals

multibroadcast and broadcast take two inputs

For dynamic shape handling

Merge branch 'refactor_dynamic_compute' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into dyn_ref_broadcast

Not needed, special case with dynamic padding

Merge branch 'refactor_auto_pad_conv' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into refactor_dynamic_compute

Improvements/Additions to be made:

changes for the quant_convolution,
changes for the deconvolution,
Macros for MIOpen status checks

* Fix invalid program from find_splits

test_gpu_pack_int8_args fails on gfx908 machine, because it doesn't set compute_fp32 flag correctly. This PR fixes the test such that it checks for the device-name, and rocblas-versions and sets this flag accordingly.

Implement operator for CPU and GPU implementations

Rewrites the BatchNormalization ONNX operator into other MIGX operators
- Added handling of 1D input tensor case (edge case in ONNX spec)
Removes the spatial and per_activation functionality (not in the ONNX spec)
- Did not remove the batch_norm_inference related code as the TensorFlow parser still uses it
- Can remove that code when the TF version is updated

Upgrade cppcheck to 2.9 

Workflow has concurrency reintroduced with different set of rules. New expected behavior is to check concurrency on PR level with one running and one pending performance tests. In case of multiple commits in same PR, always the latest commit is queued after initiated performance test execution is completed. Any other PRs/commits are in pending/queued state

Codecov announced deprecating the bash uploader. Using updated uploader

* Remove device functions
* Update tests