2 files changed, 70 insertions, 0 deletions

diff --git a/hello/CMakeLists.txt b/hello/CMakeLists.txt
new file mode 100644
index 0000000..e4b4acc
--- /dev/null
+++ b/hello/CMakeLists.txt
@@ -0,0 +1,11 @@
+cmake_minimum_required(VERSION 3.28)
+
+project(cuda_hello LANGUAGES CUDA CXX)
+
+add_executable(cuda_hello
+  hello.cu)
+
+# -Wpedantic causes warnings due to nvcc emitting non-standard (gcc-specific)
+# host code.
+# https://stackoverflow.com/questions/31000996/warning-when-compiling-cu-with-wpedantic-style-of-line-directive-is-a-gcc-ex
+target_compile_options(cuda_hello PRIVATE -Wall -Wextra -Wno-pedantic)
diff --git a/hello/hello.cu b/hello/hello.cu
new file mode 100644
index 0000000..691b18c
--- /dev/null
+++ b/hello/hello.cu
@@ -0,0 +1,59 @@
+#include <cstdio>
+
+void logDevices() {
+  int count;
+  if (cudaGetDeviceCount(&count) != cudaSuccess) {
+    printf("No CUDA devices found\n");
+    return;
+  }
+
+  printf("CUDA devices found: %d\n", count);
+  for (int i = 0; i < count; ++i) {
+    cudaDeviceProp properties;
+    if (cudaGetDeviceProperties(&properties, i) == cudaSuccess) {
+      printf("Device [%d]: %s\n", i, properties.name);
+    }
+  }
+}
+
+__global__ void kernel(int* array, int N) {
+  for (int i = 0; i < N; ++i) {
+    array[i] = i;
+  }
+}
+
+int main() {
+  logDevices();
+
+  constexpr int N = 100;
+
+  int* host_array   = new int[N];
+  int* device_array = nullptr;
+  bool success      = false;
+
+  if (cudaMalloc(&device_array, N * sizeof(int)) != cudaSuccess) {
+    goto cleanup;
+  }
+
+  kernel<<<1, 1>>>(device_array, N);
+
+  if (cudaMemcpy(
+          host_array, device_array, N * sizeof(int), cudaMemcpyDeviceToHost) !=
+      cudaSuccess) {
+    goto cleanup;
+  }
+
+  for (int i = 0; i < N; ++i) {
+    printf("%d ", host_array[i]);
+  }
+  printf("\n");
+
+  success = true;
+
+cleanup:
+  delete[] host_array;
+  if (device_array != nullptr) {
+    cudaFree(device_array);
+  }
+  return success ? 0 : 1;
+}