NCCL fix.

Author: ysiraichi

.bazelrc

@@ -47,7 +47,7 @@ build:native_arch_posix --host_copt=-march=native
 build:mkl_open_source_only --define=tensorflow_mkldnn_contraction_kernel=1
 
 build:cuda --repo_env TF_NEED_CUDA=1
-build:cuda --repo_env TF_NCCL_USE_STUB=1
+build:cuda --repo_env TF_NCCL_USE_STUB=0
 # "sm" means we emit only cubin, which is forward compatible within a GPU generation.
 # "compute" means we emit both cubin and PTX, which is larger but also forward compatible to future GPU generations.
 build:cuda --crosstool_top=@local_config_cuda//crosstool:toolchain
@@ -92,7 +92,8 @@ build:tpu --define=with_tpu_support=true
 test:tpu --local_test_jobs=1
 test:cuda --local_test_jobs=1
 
-test:cuda --@local_config_cuda//cuda:include_cuda_libs=true
+test:cuda --repo_env TF_NCCL_USE_STUB=1
+test:cuda --@local_config_cuda//cuda:include_cuda_libs=false
 
 #########################################################################
 # RBE config options below.