Add int16 support to MINIMUM and MAXIMUM

[andresovela]

This PR adds support for int16 to the MINIMUM and MAXIMUM kernels. This is necessary to work with some audio models using int16x8 quantization.

This is my first contribution to TFLM, let me know if there's anything else that needs to be done :)

bug=#2679

[suleshahid]

Hi Andres, thanks for the PR.

Seems like TfLite already has this so it would be great to support in TFLM as well. Could you add a unit test in the maximum_minimum_test.cc for this? You can see the TfLite one for reference, but we will need to use the quantized versions as in our unit tests.

Thanks!

[andresovela]

Sure, will do. Can you point me to some documentation about how to run the unit tests in this project?

[ddavis-2015]

Sure, will do. Can you point me to some documentation about how to run the unit tests in this project?

To test with x86 use:
make -f tensorflow/lite/micro/tools/make/Makefile BUILD_TYPE=default test_kernel_maximum_minimum_test -j$(nproc)

[andresovela]

@ddavis-2015 is this documented somewhere? I took a look at CONTRIBUTING.md but I didn't see anything. I'd open a PR with testing instructions if there isn't anything yet.

[ddavis-2015]

@ddavis-2015 is this documented somewhere? I took a look at CONTRIBUTING.md but I didn't see anything. I'd open a PR with testing instructions if there isn't anything yet.

@andresovela

https://github.com/tensorflow/tflite-micro/blob/main/tensorflow/lite/micro/docs/porting_reference_ops.md#5-port-the-op-from-lite-to-micro-pr3

Step 6.

[andresovela]

make -f tensorflow/lite/micro/tools/make/Makefile  BUILD_TYPE=default test_kernel_maximum_minimum_test -j4
tensorflow/lite/micro/tools/make/downloads/flatbuffers already exists, skipping the download.
tensorflow/lite/micro/tools/make/downloads/kissfft already exists, skipping the download.
tensorflow/lite/micro/tools/make/downloads/pigweed already exists, skipping the download.
tensorflow/lite/micro/tools/make/test_latency_log.sh kernel_maximum_minimum_test  gen/linux_x86_64_default_gcc/bin/kernel_maximum_minimum_test '~~~ALL TESTS PASSED~~~' linux
Testing FloatTest
Testing Int8Test
Testing Int16Test
Testing FloatWithBroadcastTest
Testing Int32WithBroadcastTest
5/5 tests passed
~~~ALL TESTS PASSED~~~

Running kernel_maximum_minimum_test took 0.001 seconds

[ddavis-2015]

@andresovela ok looks good. Now for the procedural aspects of this process (we ALL have to do this): The code must pass Google code style checks, and there must be an issue # in your initial description. For the code style issues, find the code style issue test in the checks section and click on Details.

For the issue # please create an issue in this repo (it can just be a very short description of the problem). Then at the end of your initial description of this PR add:
bug=#xxxx
where xxxx is the issue number.

Finally, you can check that the Continuous Integration tests pass by going to labels on the right hand side of the PR page, and selecting ci:run. CI runs take approx. 20 minutes. If everything passes, you can add @suleshahid to the reviewers and congrats on your addition to the TFLM (or LiteRT for Microcontrollers) code base.

[andresovela]

@ddavis-2015 I fixed the code style issue but it looks like I don't have permissions to either add labels nor add reviewers to PRs

[ddavis-2015]

@ddavis-2015 I fixed the code style issue but it looks like I don't have permissions to either add labels nor add reviewers to PRs

@andresovela Sorry, didnt know about the permissions. All tests pass, reviewer added.

[suleshahid]

Thanks again Andres, and thank you David for the help!