2 files changed, 83 insertions, 0 deletions
diff --git a/Tests/CudaOnly/Fatbin/CMakeLists.txt b/Tests/CudaOnly/Fatbin/CMakeLists.txt
new file mode 100644
index 0000000..db0dc22
--- /dev/null
+++ b/Tests/CudaOnly/Fatbin/CMakeLists.txt
@@ -0,0 +1,25 @@
+cmake_minimum_required(VERSION 3.18)
+project(CudaFATBIN LANGUAGES CUDA)
+
+
+set(CMAKE_CUDA_ARCHITECTURES all-major)
+
+add_library(CudaFATBIN OBJECT
+${CMAKE_CURRENT_SOURCE_DIR}/../CUBIN/kernelA.cu
+${CMAKE_CURRENT_SOURCE_DIR}/../CUBIN/kernelB.cu
+${CMAKE_CURRENT_SOURCE_DIR}/../CUBIN/kernelC.cu)
+
+set_property(TARGET CudaFATBIN PROPERTY CUDA_FATBIN_COMPILATION ON)
+
+# Will use `cuModuleLoadFatBinary` to load the fatbinaries
+add_executable(CudaOnlyFatbin main.cu)
+target_compile_features(CudaOnlyFatbin PRIVATE cuda_std_11)
+target_compile_definitions(CudaOnlyFatbin PRIVATE "FATBIN_FILE_PATHS=\"$<JOIN:$<TARGET_OBJECTS:CudaFATBIN>,~_~>\"")
+
+find_package(CUDAToolkit REQUIRED)
+target_link_libraries(CudaOnlyFatbin PRIVATE CUDA::cuda_driver)
+
+if(APPLE)
+  # Help the static cuda runtime find the driver (libcuda.dyllib) at runtime.
+  set_property(TARGET CudaOnlyFatbin PROPERTY BUILD_RPATH ${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES})
+endif()
diff --git a/Tests/CudaOnly/Fatbin/main.cu b/Tests/CudaOnly/Fatbin/main.cu
new file mode 100644
index 0000000..89af0e3
--- /dev/null
+++ b/Tests/CudaOnly/Fatbin/main.cu
@@ -0,0 +1,58 @@
+#include <iostream>
+#include <string>
+#include <vector>
+
+#include <cuda.h>
+
+#define GENERATED_HEADER(x) GENERATED_HEADER1(x)
+#define GENERATED_HEADER1(x) <x>
+
+static std::string input_paths = { FATBIN_FILE_PATHS };
+
+int main()
+{
+  const std::string delimiter = "~_~";
+  input_paths += delimiter;
+
+  size_t end = 0;
+  size_t previous_end = 0;
+  std::vector<std::string> actual_paths;
+  while ((end = input_paths.find(delimiter, previous_end)) !=
+         std::string::npos) {
+    actual_paths.emplace_back(
+      input_paths.substr(previous_end, end - previous_end));
+    previous_end = end + 3;
+  }
+
+  cuInit(0);
+  int count = 0;
+  cuDeviceGetCount(&count);
+  if (count == 0) {
+    std::cerr << "No CUDA devices found\n";
+    return 1;
+  }
+
+  CUdevice device;
+  cuDeviceGet(&device, 0);
+
+  CUcontext context;
+  cuCtxCreate(&context, 0, device);
+
+  CUmodule module;
+  for (auto p : actual_paths) {
+    if (p.find(".fatbin") == std::string::npos) {
+      std::cout << p << " Doesn't have the .fatbin suffix" << p << std::endl;
+      return 1;
+    }
+    std::cout << "trying to load fatbin: " << p << std::endl;
+    CUresult result = cuModuleLoad(&module, p.c_str());
+    std::cout << "module pointer: " << module << '\n';
+    if (result != CUDA_SUCCESS || module == nullptr) {
+      std::cerr << "Failed to load the embedded fatbin with error: "
+                << static_cast<unsigned int>(result) << '\n';
+      return 1;
+    }
+  }
+
+  return 0;
+}