portelli/Grid
1
0
Fork 0
mirror of https://github.com/paboyle/Grid.git synced 2025-04-16 08:45:55 +01:00
Code Releases Activity

No compile fix

Browse Source
This commit is contained in:
Peter Boyle 2025-04-04 16:44:12 -04:00
parent adc90d3a86
commit d66a9af6a3
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Grid/lattice/Lattice_slicesum_core.h
View File

@ -55,7 +55,7 @@ inline void sliceSumReduction_cub_small(const vobj *Data,
d_offsets = static_cast<int*>(acceleratorAllocDevice((rd+1)*sizeof(int)));
//copy offsets to device
acceleratorCopyToDeviceAsync(&offsets[0],d_offsets,sizeof(int)*(rd+1),computeStream);
acceleratorCopyToDeviceAsynch(&offsets[0],d_offsets,sizeof(int)*(rd+1),computeStream);
gpuError_t gpuErr = gpucub::DeviceSegmentedReduce::Reduce(temp_storage_array, temp_storage_bytes, rb_p,d_out, rd, d_offsets, d_offsets+1, ::gpucub::Sum(), zero_init, computeStream);
@ -88,7 +88,7 @@ inline void sliceSumReduction_cub_small(const vobj *Data,
exit(EXIT_FAILURE);
}
acceleratorCopyFromDeviceAsync(d_out,&lvSum[0],rd*sizeof(vobj),computeStream);
acceleratorCopyFromDeviceAsynch(d_out,&lvSum[0],rd*sizeof(vobj),computeStream);
//sync after copy
accelerator_barrier();