cory-master/BabelStream
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Enqueue CUDA kernels

Browse Source
This commit is contained in:
Tom Deakin 2015-07-16 18:06:35 +01:00
parent 2c2dbf2c3f
commit f2536f8764
1 changed files with 22 additions and 29 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
cuda-stream.cu
View File

@ -168,12 +168,19 @@ int main(int argc, char *argv[])
h_c[i] = 0.0; h_c[i] = 0.0;
} }
// Create device buffers
DATATYPE * d_a, * d_b, *d_c;
cudaMalloc(&d_a, ARRAY_SIZE*sizeof(DATATYPE));
cudaMalloc(&d_b, ARRAY_SIZE*sizeof(DATATYPE));
cudaMalloc(&d_c, ARRAY_SIZE*sizeof(DATATYPE));
// Copy host memory to device // Copy host memory to device
cudaMemcpy(d_a, h_a, ARRAY_SIZE*sizeof(DATATYPE), cudaMemcpyHostToDevice);
cudaMemcpy(d_b, h_b, ARRAY_SIZE*sizeof(DATATYPE), cudaMemcpyHostToDevice);
cudaMemcpy(d_c, h_c, ARRAY_SIZE*sizeof(DATATYPE), cudaMemcpyHostToDevice);
// Make sure the copies are finished // Make sure the copies are finished
cudaDeviceSynchronize();
// List of times // List of times
std::vector< std::vector<double> > timings; std::vector< std::vector<double> > timings;
@ -184,56 +191,42 @@ int main(int argc, char *argv[])
// Main loop // Main loop
for (unsigned int k = 0; k < NTIMES; k++) for (unsigned int k = 0; k < NTIMES; k++)
{ {
/*std::vector<double> times; std::vector<double> times;
t1 = std::chrono::high_resolution_clock::now(); t1 = std::chrono::high_resolution_clock::now();
copy( copy<<<ARRAY_SIZE/1024, 1024>>>(d_a, d_c);
cl::EnqueueArgs( cudaDeviceSynchronize();
queue,
cl::NDRange(ARRAY_SIZE)),
d_a, d_c);
queue.finish();
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count()); times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
t1 = std::chrono::high_resolution_clock::now(); t1 = std::chrono::high_resolution_clock::now();
mul( mul<<<ARRAY_SIZE/1024, 1024>>>(d_b, d_c);
cl::EnqueueArgs( cudaDeviceSynchronize();
queue,
cl::NDRange(ARRAY_SIZE)),
d_b, d_c);
queue.finish();
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count()); times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
t1 = std::chrono::high_resolution_clock::now(); t1 = std::chrono::high_resolution_clock::now();
add( add<<<ARRAY_SIZE/1024, 1024>>>(d_a, d_b, d_c);
cl::EnqueueArgs( cudaDeviceSynchronize();
queue,
cl::NDRange(ARRAY_SIZE)),
d_a, d_b, d_c);
queue.finish();
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count()); times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
t1 = std::chrono::high_resolution_clock::now(); t1 = std::chrono::high_resolution_clock::now();
triad( triad<<<ARRAY_SIZE/1024, 1024>>>(d_a, d_b, d_c);
cl::EnqueueArgs( cudaDeviceSynchronize();
queue,
cl::NDRange(ARRAY_SIZE)),
d_a, d_b, d_c);
queue.finish();
t2 = std::chrono::high_resolution_clock::now(); t2 = std::chrono::high_resolution_clock::now();
times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count()); times.push_back(std::chrono::duration_cast<std::chrono::duration<double> >(t2 - t1).count());
timings.push_back(times);*/ timings.push_back(times);
} }
// Check solutions // Check solutions
cudaMemcpy(h_a, d_a, ARRAY_SIZE*sizeof(DATATYPE), cudaMemcpyDeviceToHost);
cudaMemcpy(h_b, d_b, ARRAY_SIZE*sizeof(DATATYPE), cudaMemcpyDeviceToHost);
cudaMemcpy(h_c, d_c, ARRAY_SIZE*sizeof(DATATYPE), cudaMemcpyDeviceToHost);
check_solution(h_a, h_b, h_c); check_solution(h_a, h_b, h_c);
// Crunch results // Crunch results