GitLab

- gen_onnx.py changes for onnx output of empty const input branches (seen in resnext50)
- updated onnx_test.cpp to validate parsing of input.
- new onnx files generated from onnx tests

- Handle checks for each IF output
- add const to inputs of all_but_last_dims_equal
- add std::equal instead of using equal
- Use .back() for vectors in getting last value
- Use input().front() instead of prev(prev()) when replacing the last value.

Allows for rank 2 tensors into batchnorm.  Specifically when spatial dimensions are all 1 and removed

Removes use_dynamic_same_auto_pad
Change padding_mode to be used for dynamic padding
Move compute_padded_shape to pad_calc.cpp as it will be used in other dynamic padding cases
Fix same_lower compute_padded_shape bug and add a test.

Rewrites the TF batch norm like operators to other MIGX operators
Removes the code related to batch_norm_inference

Hand verified results before running test using random data generated from
the new protobuf files for this.

Everything is correctly functionally within tolerances.

Had to readjust input sizes to that of the .onnx file as one input is deliberately
made smaller (without the trailing 1) for theses tests to test the case correctly

Fixes test which verifies how we interpret vectors with trailing ones, between input branches to
an IF module block.

Update test for new generated onnx file, as well as using unsqueeze output
for when we parse in a network with mismatched shapes with a trailing 1

Inverted logic when I wrote generated statements for these. Regenerated files

- Gets past to the split section of the resnext model
- adding outline seems to solve if issues but verify calls broken
- Referencing wrong element now instead of output of correct if block?
- Need to determine proper output through verify tests.
- Modified protobuf to handle case of extra 1 to "vectorize" scalar
- Modified verify/tests to get things to "work", may need to be revised further.

Simplified algebraic operations (x*1), x*(-1), x/1, 0+x & x+0,  x-0, 0-x, 0*x, x*0, and 0/x operations

- Update if_then/else_empty test protobuff and cases
- Need to update rand() vector used
- Make y empty instead of x for if_else_empty_test.onnx
- Regenerate protobufs with updates
- Add changes to handle empty/scalar input branch size to if operator.
- Add case where if both branches empty throw an error.
- Update verify tests with gold vectors and new shapes for empty input vec
  which we handle like a scalar before broadcasting

- Verification tests that test each then/else branches for parsed IF operator
- Testing empty shape tensors for one branch -> output must be the other branch's shape
- Testing trailing 1 shape for one branch -> output must be union of both inputs

Current issue with IF operator is that we can't handle training vectors that match
in size correclty while also running into issues with empty inputs for one of the
branches for size/type checks.

This seemed to be missing, just leveraging the existing protobuf made
to test parsing of if_then_test.onnx for this and using the tensor of all
ones to default to an ADD operation to ensure cond =1 is being handled and parsed
in correctly.

Stream sync changes and associated API level changes

optimize the softmax operator

Adds two methods for the custom_ops virtual class.

bool runs_on_offload_target(), if the custom op runs directly on the gpu then it should be set to true. in this case, custom op expects its parameters to reside in GPU memory and writes output to the GPU memory. If it is set to false then, custom op expects it's parameter to reside on the host and puts back the result into the host memory.

output_alias, if output of the custom op is aliasing the input buffer. i.e. interpreting the same input buffer with differnet shape and strides.

Update as_vector() in C++ API to handle non-standard shapes. It required exposing element_index to space_index conversion method for the shape class.

Improvements/Additions to be made:

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

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 

* Remove device functions
* Update tests

This PR allows for other values of epsilon to be matched when finding layernorm. Similarly, the calculation now uses the variable for epsilon.

Change find_mul_conv to work with multibroadcast also. Checks the strides instead of the broadcast axis.

Compute mean and variance in same reduction
Set block size to numbers divisible by 32 instead powers of 2
Global is also set exactly instead of being divisible by block size
More exact matching of global/local can help get rid of branching/loops
Reduce vectors first before doing dpp_reduce
Explicitly vectorize array operators since the compiler doesnt always vectorize them
Still uses old for loop when its computing at compile-time since the reinterpret_cast nor the all the vector types is supported

* fix typo for add_sigmoid

* Replaced `find_library` with `find_package` to locate MLIR static library
* Unified the include dir for headers and remove backward compatibility
* Embedded the external/include dir into the exported library

The verify tests from pr #1354 were still causing some codecov timeouts after merge. This PR further reduces the problem sizes to avoid these failures.

* fix slice_dim1 for case

* Implement concat using jit compilation

Improves performance for 4/6 GEMMs used by huggingface BERT models with batch_size>1 by using a non-batched rocBLAS call for GEMMs where the B input has a broadcasted batch dimension.
The four verify tests added reflect the actual configurations used by bert-base-cased, with varied batch sizes.

Also adds a matcher to simplify_reshapes to move multibroadcasts after concats.

* Remove unused headers

Fixes TF literal parsing for relu6.  previously always made a float type literal, breaks for float16 as an example

* Fix accuracy bug when vectorizing slices

Using not and or improves readability. The cppcheck rule will help ensure we are doing it consistently.

Rewrite_gelu pass replaces the gelu formula of x * (1/2) * (1 + erf(x/sqrt(2))) with the sigmoid approximation of x * Sigmoid(x * 1.702)

reshape op requires standard shape. During simplify_algebra, it inserts reshapes without checking for this requirement.

This will rewrite dot operators like X(Y + b) to XY + Xb when b is constant as we can fold the add away.
This improves handling pointwise with broadcasted operators, this helps improves const propagation.
Improve gemm fusion with a mul_add
Improve support for broadcast shapes in gemm