use static linked cupti internally.

PiperOrigin-RevId: 313822331
Change-Id: Ie2c1e4a2d92ed9f92e2dcd8d3bc354546ab4663f

tensorflow/core/profiler/internal/gpu/BUILD

@@ -15,6 +15,10 @@ load(
     "//tensorflow/core/platform:build_config_root.bzl",
     "tf_cuda_tests_tags",
 )
+load(
+    "//tensorflow/stream_executor:build_defs.bzl",
+    "tf_additional_cupti_deps",
+)
 
 package(
     default_visibility = ["//tensorflow:internal"],
@@ -91,9 +95,7 @@ tf_cuda_library(
     deps = [
         "//tensorflow/core:lib",
         "//tensorflow/core:platform_base",
-    ] + if_cuda_is_configured_compat([
-        "//tensorflow/stream_executor/cuda:cupti_stub",
-    ]),
+    ] + tf_additional_cupti_deps(),
 )
 
 tf_cuda_library(
@@ -104,9 +106,7 @@ tf_cuda_library(
     visibility = ["//visibility:public"],
     deps = [
         ":cupti_interface",
-    ] + if_cuda_is_configured_compat([
-        "//tensorflow/stream_executor/cuda:cupti_stub",
-    ]),
+    ] + tf_additional_cupti_deps(),
 )
 
 tf_cuda_library(

tensorflow/stream_executor/build_defs.bzl

@@ -10,6 +10,9 @@ def tf_additional_cuda_platform_deps():
 def tf_additional_cuda_driver_deps():
     return [":cuda_stub"]
 
+def tf_additional_cupti_deps():
+    return ["//tensorflow/stream_executor/cuda:cupti_stub"]
+
 def tf_additional_cudnn_plugin_deps():
     return []
 

tensorflow/stream_executor/cuda/cupti_stub.cc

@@ -23,12 +23,16 @@ limitations under the License.
 namespace {
 // Returns DSO handle or null if loading the DSO fails.
 void* GetDsoHandle() {
+#if defined(PLATFORM_GOOGLE) && (CUDA_VERSION > 10000)
+  return nullptr;
+#else
   static auto handle = []() -> void* {
     auto handle_or = stream_executor::internal::DsoLoader::GetCuptiDsoHandle();
     if (!handle_or.ok()) return nullptr;
     return handle_or.ValueOrDie();
   }();
   return handle;
+#endif
 }
 
 template <typename T>