[tf-quant] Add TF compositions for the floating-point and integer tfl.conv2d
These ops are added as examples to show how to use the tf composite api. The
composition contains the some tfr built-ins ops which haven't been supported.
Simple unit tests, by comparing the floating-point and integer compositions,
are added to verify the correctness of the implementation.
There are some issues to be fixed:
- support per-channel quantization scales
- the tfr translator support for the built-in ops.
PiperOrigin-RevId: 371212467
Change-Id: Ifc04d21f0bb2a63ef97c8c594845f912c6f78435
diff --git a/tensorflow/compiler/mlir/tfr/BUILD b/tensorflow/compiler/mlir/tfr/BUILD
index b29bfdc..22c0af1 100644
--- a/tensorflow/compiler/mlir/tfr/BUILD
+++ b/tensorflow/compiler/mlir/tfr/BUILD
@@ -24,6 +24,7 @@
name = "friends",
includes = ["//third_party/mlir:subpackages"],
packages = [
+ "//learning/brain/experimental/mlir/quantization/...",
"//tensorflow/c/...",
"//tensorflow/compiler/...",
],
