misc changes

tensorflow/python/framework/test_util.py

@@ -1368,6 +1368,21 @@ def is_gpu_available(cuda_only=False, min_cuda_compute_capability=None):
     min_cuda_compute_capability: a (major,minor) pair that indicates the minimum
       CUDA compute capability required, or None if no requirement.
 
+  Note that the keyword arg name "cuda_only" is misleading (since routine will
+  return true when a GPU device is available irrespective of whether TF was
+  built with CUDA support or ROCm support. However no changes here because
+
+  ++ Changing the name "cuda_only" to something more generic would break
+     backward compatibility
+
+  ++ Adding an equivalent "rocm_only" would require the implementation check
+     the build type. This in turn would require doing the same for CUDA and thus
+     potentially break backward compatibility
+
+  ++ Adding a new "cuda_or_rocm_only" would not break backward compatibility,
+     but would require most (if not all) callers to update the call to use
+     "cuda_or_rocm_only" instead of "cuda_only"
+
   Returns:
     True if a GPU device of the requested kind is available.
   """

tensorflow/tools/ci_build/linux/rocm/run_cc_core.sh

@@ -19,9 +19,10 @@ set -e
 set -x
 
 N_JOBS=$(grep -c ^processor /proc/cpuinfo)
+N_GPUS=$(lspci|grep 'VGA'|grep 'AMD/ATI'|wc -l)
 
 echo ""
-echo "Bazel will use ${N_JOBS} concurrent job(s)."
+echo "Bazel will use ${N_JOBS} concurrent build job(s) and ${N_GPUS} concurrent test job(s)."
 echo ""
 
 # Run configure.
@@ -29,11 +30,15 @@ export PYTHON_BIN_PATH=`which python3`
 export CC_OPT_FLAGS='-mavx'
 
 export TF_NEED_ROCM=1
+export TF_GPU_COUNT=${N_GPUS}
 
 yes "" | $PYTHON_BIN_PATH configure.py
 
 # Run bazel test command. Double test timeouts to avoid flakes.
-bazel test --config=rocm --test_tag_filters=-no_oss,-oss_serial,-no_gpu,-benchmark-test -k \
+bazel test --config=rocm --test_tag_filters=-no_oss,-oss_serial,-no_gpu,-no_rocm,-benchmark-test -k \
     --test_lang_filters=cc --jobs=${N_JOBS} --test_timeout 300,450,1200,3600 \
-    --build_tests_only --test_output=errors --local_test_jobs=1 --config=opt \
+    --build_tests_only --test_output=errors --local_test_jobs=${TF_GPU_COUNT} --config=opt \
+    --test_sharding_strategy=disabled \
+    --test_size_filters=small,medium \
+    --run_under=//tensorflow/tools/ci_build/gpu_build:parallel_gpu_execute -- \
     //tensorflow/... -//tensorflow/compiler/... -//tensorflow/contrib/...