unit test failure kernels:sparse_matmul_op_test on AARCH64

[elfringham]

Please make sure that this is a bug. As per our
GitHub Policy,
we only address code/doc bugs, performance issues, feature requests and
build/installation issues on GitHub. tag:bug_template

System information

• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 18.04
• Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: N/a
• TensorFlow installed from (source or binary): source
• TensorFlow version (use command below): git HEAD
• Python version: 3.6.9
• Bazel version (if compiling from source): 3.7.2
• GCC/Compiler version (if compiling from source): 10.3.0
• CUDA/cuDNN version: n/a
• GPU model and memory: n/a

You can collect some of this information using our environment capture
script
You can also obtain the TensorFlow version with:

1. TF 1.0: python -c "import tensorflow as tf; print(tf.GIT_VERSION, tf.VERSION)"
2. TF 2.0: python -c "import tensorflow as tf; print(tf.version.GIT_VERSION, tf.version.VERSION)"

Describe the current behavior

Test fails

Describe the expected behavior

Test passes

Contributing

• Do you want to contribute a PR? (yes/no): no
• Briefly describe your candidate solution(if contributing):

Standalone code to reproduce the issue
Provide a reproducible test case that is the bare minimum necessary to generate
the problem. If possible, please share a link to Colab/Jupyter/any notebook.

bazel test //tensorflow/core/kernels:sparse_matmul_op_test

Other info / logs Include any logs or source code that would be helpful to
diagnose the problem. If including tracebacks, please include the full
traceback. Large logs and files should be attached.

[ RUN ] SparseMatmulOpTest.BroadcastPacketTest
[0.170094 0.170094 0.170094 0.170094] != [ 0.170094 0.14922 -0.0823886 0.026985], differences: [ 0 -0.0208738 -0.252482 -0.143109]
tensorflow/core/kernels/sparse_matmul_op_test.cc:329: Failure
Value of: areApprox(ref, data2, PacketSize)
Actual: false
Expected: true
[ FAILED ] SparseMatmulOpTest.BroadcastPacketTest (0 ms)

[elfringham]

@cfRod @nSircombe

[elfringham]

It looks like AARCH64 is using the definitions of pbroadcast_first et al from

tensorflow/tensorflow/core/kernels/sparse_matmul_op.h

Line 92 in 9a9b33a

template <typename Packet>

which are only meant for scalars, so do not pick up the correct value.

[mohantym]

Hi @jvishnuvardhan! Could you please look at this issue.

[elfringham]

Fixed by commit 6d8220b