AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

EXSWHTEC-267 - Implement tests for multi_grid_group APIs (#156)

Browse Source 
Change-Id: I2820ec19831f0dad3b8ce60172c86791f274107e
This commit is contained in:
Nives Vukovic
2023-11-20 18:23:26 +01:00
committed by Rakesh Roy
parent 10594ecb3a
commit 9b96b707ee
5 changed files with 701 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
catch/include/cpu_grid.h
+26
View File
@@ -78,6 +78,32 @@ struct CPUGrid {
unsigned int thread_count_;
};
struct CPUMultiGrid {
CPUMultiGrid(const unsigned int num_grids, const dim3 grid_dims[], const dim3 block_dims[]) {
thread_count_ = 0;
grid_count_ = num_grids;
grids_.reserve(grid_count_);
for (int i = 0; i < grid_count_; i++) {
grids_.emplace_back(grid_dims[i], block_dims[i]);
thread_count_ += grids_[i].thread_count_;
}
}
inline unsigned int thread0_rank_in_multi_grid(const unsigned int grid_rank) const {
unsigned int multi_grid_thread_rank_0 = 0;
unsigned int multi_grid_thread_count = 0;
for (int i = 0; i <= grid_rank; i++) {
multi_grid_thread_rank_0 = multi_grid_thread_count;
multi_grid_thread_count += grids_[i].thread_count_;
}
return multi_grid_thread_rank_0;
}
std::vector<CPUGrid> grids_;
unsigned int grid_count_;
unsigned int thread_count_;
};
/* Generate dimensions for 1D, 2D and 3D blocks of threads */
inline dim3 GenerateThreadDimensions() {
hipDeviceProp_t props;