commit 712fd5cfe4a5de420a4226c874664423eba4cb1a
author Adrian Kuegel <akuegel@google.com> Tue Sep 22 03:27:52 2020 -0700
committer TensorFlower Gardener <gardener@tensorflow.org> Tue Sep 22 03:32:29 2020 -0700
tree 5fe849d0fca46778e49b8b40ff2ee65a34476ae0
parent a0a4b95f501f55e9c49bbc628e9b8cb3a5b788f6

Add back the "no_rocm" tag to cwise_unary_op target.

It seems in the Kokoro build environment, hexdump is not available.

PiperOrigin-RevId: 333042998
Change-Id: I2a5fe58ffed008f37ddb807378cc6e4f0d03e33f

tensorflow/core/kernels/mlir_generated/BUILD

diff --git a/tensorflow/core/kernels/mlir_generated/BUILD b/tensorflow/core/kernels/mlir_generated/BUILD
index 427886b..29e7341 100644
--- a/tensorflow/core/kernels/mlir_generated/BUILD
+++ b/tensorflow/core/kernels/mlir_generated/BUILD
@@ -41,6 +41,9 @@
         "cwise_op_gpu_base.h",
         "cwise_op_gpu_tanh.cc",
     ]),
+    # Building abs_kernels and tanh_kernels requires hexdump on the ROCM
+    # platform, which might not be available.
+    tags = ["no_rocm"],
     deps = if_cuda_or_rocm([
         ":abs_kernels",
         ":tanh_kernels",