Remove remaining cuda-only tags and move failing subtests to test scripts

Author: mmakevic-amd

tensorflow/compiler/mlir/tensorflow/tests/tf_saved_model/BUILD

@@ -294,7 +294,6 @@ glob_lit_tests(
     default_tags = [
         "no_mac",  # TODO(b/191167848)
         "no_oss",  # TODO(b/190855110)
-        "cuda-only",
     ],
     driver = "@llvm-project//mlir:run_lit.sh",
     exclude = [

tensorflow/compiler/tests/BUILD

@@ -1788,7 +1788,6 @@ tf_xla_py_strict_test(
     srcs = ["unary_ops_test.py"],
     shard_count = 20,
     tags = [
-        "cuda-only",
         "no_aarch64",  # TODO(b/348125886)
         "no_cuda_asan",  # times out
         "no_pip",  # TODO(b/149738646): fix pip install so these tests run on kokoro pip
@@ -2861,7 +2860,6 @@ tf_cuda_cc_test(
     tags = [
         "config-cuda-only",
         "no_pip",  # TODO(b/149738646): fix pip install so these tests run on kokoro pip
-        "cuda-only",  # ROCmSoftwarePlatform #958
         "noasan",  # TODO(b/201651800)
         "requires-gpu-nvidia",
     ] + tf_cuda_tests_tags(),
@@ -2882,7 +2880,6 @@ tf_cuda_cc_test(
     tags = [
         "config-cuda-only",
         "no_pip",  # TODO(b/149738646): fix pip install so these tests run on kokoro pip
-        "cuda-only",  # ROCmSoftwarePlatform #958
         "noasan",  # TODO(b/201651800)
         "requires-gpu-nvidia",
     ] + tf_cuda_tests_tags(),

tensorflow/core/kernels/mkl/BUILD

@@ -427,7 +427,6 @@ tf_cc_test_mkl(
     size = "small",
     srcs = ["mkl_fused_batch_norm_op_test.cc"],
     linkstatic = 1,
-    tags = ["cuda-only"], # fails on AMD Rome CPUs as of 2021-03-29
     deps = [
         ":mkl_conv_op",
         ":mkl_fused_batch_norm_op",

tensorflow/core/kernels/mlir_generated/BUILD

@@ -557,7 +557,6 @@ tf_cuda_cc_test(
     tags = tf_cuda_tests_tags() + [
         "no_cuda",  # TODO(b/196608406): re-enable
         "no_cuda_asan",  # TODO(b/171341759): re-enable.
-        "cuda-only",
     ],
     deps = [
         ":base_binary_ops_test",

tensorflow/core/nccl/BUILD

@@ -61,7 +61,6 @@ tf_cuda_cc_test(
         "multi_gpu",
         "no_oss",
         "notap",
-        "cuda-only", # flaky on CI as of 2022-05-30
     ],
     deps = [
         "//tensorflow/core:test",

tensorflow/core/util/autotune_maps/BUILD

@@ -193,7 +193,6 @@ tf_cuda_only_cc_test(
     size = "small",
     srcs = ["autotune_serialize_test.cc"],
     features = ["-layering_check"],
-    tags = ["cuda-only"],
     deps = [
         ":autotune_serialize",
         ":conv_autotune_maps",

tensorflow/dtensor/cc/BUILD

@@ -225,7 +225,6 @@ tf_kernel_library(
         "dtensor_tpu_kernels.cc",
     ],
     tags = [
-        "cuda-only",
         "tpu",
     ],  # Disable building of TPU kernels on non-TPU platforms.
     deps = [

tensorflow/dtensor/python/tests/BUILD

@@ -746,9 +746,6 @@ dtensor_test(
         "tpu": 10,
         TPU_V3_DONUT_BACKEND: 32,
     },
-    tags = [
-        "cuda-only",
-    ],
     deps = [
         ":test_util",
         "//tensorflow/dtensor/python:api",
@@ -802,7 +799,6 @@ dtensor_test(
     },
     tags = [
         "no_oss_py38",  # TODO(b/267017937)
-        "cuda-only",
     ],
     deps = [
         ":test_util",

tensorflow/python/compiler/tensorrt/test/BUILD

@@ -74,7 +74,6 @@ filegroup(
 
 base_tags = [
     "no_cuda_on_cpu_tap",
-    "cuda-only",
     "no_windows",
     "nomac",
     # TODO(b/303453873): Re-enable tests once TensorRT has been updated

tensorflow/python/debug/lib/BUILD

@@ -331,7 +331,6 @@ cuda_py_strict_test(
     shard_count = 4,
     tags = [
         "no_windows",  # TODO(b/142475891): Enable this test on Windows.
-        "cuda-only", #TODO(ROCm) Re-enable after issue is fixed.
     ],
     xla_enable_strict_auto_jit = False,  # Node names are different with autojit
     deps = [
@@ -363,7 +362,6 @@ cuda_py_strict_test(
     python_version = "PY3",
     tags = [
         "no_windows_gpu",
-        "cuda-only", #TODO(ROCm) Re-enable after issue is fixed.
     ],
     deps = [
         ":debug_events_reader",