#include <stdio.h>
#include <cuda_runtime.h>

int main() 
{
  // Number of available devices (usually 1)
  int deviceCount;
  cudaError_t error = cudaGetDeviceCount(&deviceCount);

  if(error != cudaSuccess) 
  {
    printf("Error getting device count: %s\n",cudaGetErrorString(error));
    return 1;
  }

  if (deviceCount == 0) 
  {
    printf("There are no available CUDA devices.\n");
    return 0;
  }

  // Loop through all available devices
  for (int i=0; i<deviceCount; ++i) 
  {
    // Structure to hold the device properties
    cudaDeviceProp deviceProp;
    int device;


    // Get the properties for the current device
    error = cudaGetDeviceProperties(&deviceProp, i);
    if (error != cudaSuccess) 
    {
      printf("Error getting device properties for device %d: %s\n",i,cudaGetErrorString(error));
      continue;
    }

    cudaGetDevice(&device);
    int maxBlocksPerSM, maxWarpsPerSM;
    cudaDeviceGetAttribute(&maxBlocksPerSM,cudaDevAttrMaxBlocksPerMultiprocessor,device);
    maxWarpsPerSM = deviceProp.maxThreadsPerMultiProcessor / deviceProp.warpSize;

    // Print the device name and the number of SMs.
    printf("Device %d: %s\n",i,deviceProp.name);
    printf("  Compute capability (major.minor): %d.%d\n",deviceProp.major,deviceProp.minor);
    printf("Processors\n");
    printf("  Number of streaming multiprocessors (SMs): %d\n",deviceProp.multiProcessorCount);
    printf("  Clock rate: %.0fMHz\n",(double)deviceProp.clockRate/1000.0);
    printf("  Single to double precision ratio performance: %d\n",deviceProp.singleToDoublePrecisionPerfRatio);
    printf("  Max. number of blocks per SM: %d\n",maxBlocksPerSM);
    printf("  Max. number of warps per SM: %d\n",maxWarpsPerSM);
    printf("Memory\n");
    printf("  Global memory on device: %.0fGB\n",(double)deviceProp.totalGlobalMem / (double)(1<<30));
    printf("  Bus width between off chip global memory and on chip L2 cache: %d Bits\n",deviceProp.memoryBusWidth);
    printf("  Memory clock rate %.0fMHz:\n",(double)deviceProp.memoryClockRate / 1000.0);
    printf("  L2 cache size (shared by all SMs): %.0fkB\n",(double)deviceProp.l2CacheSize / (double)(1<<10) );
    printf("  Shared memory per SM (L1 cache): %.0fkB\n",(double)deviceProp.sharedMemPerMultiprocessor / (double)(1<<10));
    printf("  Maximum amount of shared memory available to each thread block: %.0fkB\n",(double)deviceProp.sharedMemPerBlock / (double)(1<<10));
    printf("  Constant memory on device %.0fkB\n",(double)deviceProp.totalConstMem / (double)(1<<10));
    printf("  Managed memory supported: %s\n",deviceProp.managedMemory==1?"yes":"no");
    printf("  Managed memory can be accessed concurrently with CPU: %s\n",deviceProp.concurrentManagedAccess==1?"yes":"no");
    printf("  Direct managed memory acces from host: %s\n",deviceProp.directManagedMemAccessFromHost==1?"yes":"no");
    printf("Registers\n");
    printf("  Number of 32bit registers available to each a SM: %d\n",deviceProp.regsPerMultiprocessor);
    printf("  Maximum number of 32bit registers available to each thread block: %d\n",deviceProp.regsPerBlock);
    printf("Threads\n");
    printf("  Number of threads per warp: %d\n",deviceProp.warpSize);
    printf("  Maximum number of active threads per SM: %d\n",deviceProp.maxThreadsPerMultiProcessor);
    printf("  Maximum number of active threads per block: %d\n",deviceProp.maxThreadsPerBlock);
    printf("Details\n");
    printf("  Device allows execution of concurrent kernels in the same context: %s\n",deviceProp.concurrentKernels==1?"yes":"no");
  }

  return 0;
}