moebiusband/MD-Bench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Measure memory transfer of CPU to GPU, add explanation how to distribute calculation among multiple GPUs

Browse Source
This commit is contained in:
Maximilian Gaul
2021-12-01 17:16:32 +01:00
parent da90466f98
commit 7691b23d67
1 changed files with 6 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/force.cu
View File

@@ -144,10 +144,15 @@ double computeForce(
for(int i = 0; i < nDevices; ++i) { for(int i = 0; i < nDevices; ++i) {
cudaDeviceProp prop; cudaDeviceProp prop;
cudaGetDeviceProperties(&prop, i); cudaGetDeviceProperties(&prop, i);
printf("DEVICE NAME: %s\r\n", prop.name); printf("DEVICE %d/%d NAME: %s\r\n", i + 1, nDevices, prop.name);
} }
// Choose GPU where you want to execute code on
// It is possible to execute the same kernel on multiple GPUs but you have to copy the data multiple times
// Executing `cudaSetDevice(N)` before cudaMalloc / cudaMemcpy / calc_force <<< >>> will set the GPU accordingly
*/ */
// HINT: Run with cuda-memcheck ./MDBench-NVCC in case of error // HINT: Run with cuda-memcheck ./MDBench-NVCC in case of error
// HINT: Only works for data layout = AOS!!! // HINT: Only works for data layout = AOS!!!