// CUDA programming

// Exercise n. 01

#include <errno.h>

#include <cuda.h>

#include <stdio.h>

// Prototype

__host__ void print_dev_prop(cudaDeviceProp dev_prop);

int main(void)

{

// Number of CUDA-capable devices attached to this system

int dev_cnt;

cudaGetDeviceCount(&dev_cnt);

// Calculate the theoretical peak bandwidth for each device

for(int i = 0; i < dev_cnt; i++)

{

cudaDeviceProp dev_prop;

cudaGetDeviceProperties(&dev_prop, i);

printf("Device Number: %d\n", i);

print_dev_prop(dev_prop);

}

}

// Print device properties

__host__ void print_dev_prop(cudaDeviceProp dev_prop)

{

printf(" Major revision number: %d\n", dev_prop.major);

printf(" Minor revision number: %d\n", dev_prop.minor);

printf(" Name: %s\n", dev_prop.name);

printf(" Total global memory: %zu\n", dev_prop.totalGlobalMem);

printf(" Total shared memory per block: %zu\n", dev_prop.sharedMemPerBlock);

printf(" Total registers per block: %d\n", dev_prop.regsPerBlock);

printf(" Warp size: %d\n", dev_prop.warpSize);

printf(" Maximum memory pitch: %zu\n", dev_prop.memPitch);

printf(" Maximum threads per block: %d\n", dev_prop.maxThreadsPerBlock);

for(int i = 0; i < 3; ++i)

printf(" Maximum block dimension #%02d: %d\n", i, dev_prop.maxThreadsDim[i]);

for(int i = 0; i < 3; ++i)

printf(" Maximum grid dimension #%02d: %d\n", i, dev_prop.maxGridSize[i]);

printf(" Clock rate: %d\n", dev_prop.clockRate);

printf(" Memory Bus Width (bits): %d\n", dev_prop.memoryBusWidth);

printf(" Peak Memory Bandwidth (GB/s): %f\n\n", 2.0 * dev_prop.memoryClockRate * (dev_prop.memoryBusWidth / 8) / 1.0e6);

printf(" Total constant memory: %zu\n", dev_prop.totalConstMem);

printf(" Texture alignment: %zu\n", dev_prop.textureAlignment);

printf(" Concurrent copy and execution: %s\n", (dev_prop.deviceOverlap ? "Yes" : "No"));

printf(" Number of multiprocessors: %d\n", dev_prop.multiProcessorCount);

printf(" Kernel execution timeout: %s\n", (dev_prop.kernelExecTimeoutEnabled ? "Yes" : "No"));

return;

}