cuda: demo no arg constructor in extended struct

demo_no_arg_constructor.cu

/// nvcc -std=c++20 -arch=sm_86 -o demo-no-arg-constructor demo_no_arg_constructor.cu
/// ❯ /opt/cuda/extras/compute-sanitizer/compute-sanitizer \
//     --leak-check=full  --report-api-errors=all              \
//     --check-cache-control                                                \
//     --check-api-memory-access=yes  --check-device-heap=yes               \
//     --tool memcheck                                                      \
//     ./demo-no-arg-constructor



#include <cuda_runtime.h>
#include <cstdint>
#include <iostream>


#define CUDA_CHECK(call) do {					\
    cudaError_t err = call;                                             \
    if (err != cudaSuccess) {                                           \
      std::cerr << "CUDA call failed: " << #call << "\n"                \
                << "Error code: " << err << "\n"                        \
                << "Error string: " << cudaGetErrorString(err)          \
                << " (at " << __FILE__ << ":" << __LINE__ << ")"        \
                << std::endl;                                           \
      throw std::runtime_error(cudaGetErrorString(err));                \
    }                                                                   \
  } while (0)



struct uint256 {
    uint32_t words[8];
};

struct evm_word_t : uint256 {
  // This default constructor will cause compute-sanitizer error:
  // ========= Invalid __global__ write of size 8 bytes
  // =========     at test_new_array(unsigned long)+0x110
  // =========     by thread (0,0,0) in block (0,0,0)
  // =========     Address 0x205dff918 is out of bounds
  __host__ __device__ evm_word_t() {}
};

__global__ void test_new_array(size_t N) {
  evm_word_t* arr = new evm_word_t[N]; // for large N this can fail with nullptr

  if (arr == nullptr) {
    printf("Thread %d: new failed\n", threadIdx.x);
    return;
  }

  const int x = blockIdx.x * blockDim.x + threadIdx.x;
  arr[0].words[0] = x;
  printf("Thread %d: arr[0].words[0] = %u\n", threadIdx.x, arr[0].words[0]);
  delete[] arr;
}


int main( int argc, char** argv ) {

  test_new_array<<<1, 32>>>(10);
  CUDA_CHECK(cudaDeviceSynchronize());


  return 0;
}