Support static builds of nvbench with nvml enabled.

To do this we need to ensure that the nvml init handler is
both contained in the library/executable that uses nvbench.

The original implementation fails since the singleton can be dropped
since it has no usages. So instead we move to a function static
which we ensure will always be used.

cmake/NVBenchExports.cmake

@@ -21,6 +21,18 @@ macro(nvbench_generate_exports)
       )
     endif()
 
+    if (TARGET nvbench_json)
+      set(nvbench_json_code_block
+        [=[
+        add_library(nvbench_json INTERFACE IMPORTED)
+        if (TARGET nlohmann_json::nlohmann_json)
+          target_link_libraries(nvbench_json INTERFACE nlohmann_json::nlohmann_json)
+        endif()
+        ]=])
+      string(APPEND nvbench_build_export_code_block ${nvbench_json_code_block})
+      string(APPEND nvbench_install_export_code_block ${nvbench_json_code_block})
+    endif()
+
     rapids_export(BUILD NVBench
       EXPORT_SET nvbench-targets
       NAMESPACE "nvbench::"

nvbench/CMakeLists.txt

@@ -24,16 +24,14 @@ set(srcs
   detail/measure_cold.cu
   detail/measure_hot.cu
   detail/state_generator.cxx
+
+  internal/nvml.cxx
 )
 
 if (NVBench_ENABLE_CUPTI)
   list(APPEND srcs detail/measure_cupti.cu cupti_profiler.cxx)
 endif()
 
-if (NVBench_ENABLE_NVML)
-  list(APPEND srcs internal/nvml.cxx)
-endif()
-
 # CUDA 11.0 can't compile json_printer without crashing
 # So for that version fall back to C++ with degraded
 # output ( no PTX version info )

nvbench/device_info.cu

@@ -45,6 +45,9 @@ device_info::device_info(int id)
     , m_nvml_device(nullptr)
 {
   NVBENCH_CUDA_CALL(cudaGetDeviceProperties(&m_prop, m_id));
+  // NVML's lifetime should extend for the entirety of the process, so store in a
+  // global.
+  [[maybe_unused]] static auto nvml_lifetime = nvbench::nvml::NVMLLifetimeManager();
 
 #ifdef NVBENCH_HAS_NVML
   // Retrieve the current device's pci_id as a null-terminated string.

nvbench/internal/nvml.cuh

@@ -32,6 +32,16 @@
 namespace nvbench::nvml
 {
 
+// RAII struct that initializes and shuts down NVML
+// Beeds to be constructed and kept alive while using nvml
+struct NVMLLifetimeManager
+{
+  NVMLLifetimeManager();
+  ~NVMLLifetimeManager();
+private:
+  bool m_inited{false};
+};
+
 /// Base class for NVML-specific exceptions
 struct error : std::runtime_error
 {

nvbench/internal/nvml.cxx

@@ -16,56 +16,46 @@
  *  limitations under the License.
  */
 
-#include <nvbench/internal/nvml.cuh>
-
 #include <nvbench/config.cuh>
-
-#include <fmt/format.h>
-
-#include <nvml.h>
+#include <nvbench/internal/nvml.cuh>
 
 #include <stdexcept>
 
-namespace
-{
+#include <fmt/format.h>
+#include <nvml.h>
 
-// RAII struct that initializes and shuts down NVML
-struct NVMLLifetimeManager
+namespace nvbench::nvml
 {
-  NVMLLifetimeManager()
+NVMLLifetimeManager::NVMLLifetimeManager()
+{
+#ifdef NVBENCH_HAS_NVML
+  try
+  {
+    NVBENCH_NVML_CALL_NO_API(nvmlInit());
+    m_inited = true;
+  }
+  catch (std::exception &e)
+  {
+    fmt::print("NVML initialization failed:\n {}", e.what());
+  }
+#endif
+}
+
+NVMLLifetimeManager::~NVMLLifetimeManager()
+{
+#ifdef NVBENCH_HAS_NVML
+  if (m_inited)
   {
     try
     {
-      NVBENCH_NVML_CALL_NO_API(nvmlInit());
-      m_inited = true;
+      NVBENCH_NVML_CALL_NO_API(nvmlShutdown());
     }
     catch (std::exception &e)
     {
-      fmt::print("NVML initialization failed:\n {}", e.what());
+      fmt::print("NVML shutdown failed:\n {}", e.what());
     }
   }
+#endif
+}
 
-  ~NVMLLifetimeManager()
-  {
-    if (m_inited)
-    {
-      try
-      {
-        NVBENCH_NVML_CALL_NO_API(nvmlShutdown());
-      }
-      catch (std::exception &e)
-      {
-        fmt::print("NVML shutdown failed:\n {}", e.what());
-      }
-    }
-  }
-
-private:
-  bool m_inited{false};
-};
-
-// NVML's lifetime should extend for the entirety of the process, so store in a
-// global.
-auto nvml_lifetime = NVMLLifetimeManager{};
-
-} // namespace
+} // namespace nvbench::nvml