Conversation
umar456 previously approved these changes Dec 6, 2018
| } | ||
|
|
||
| TEST(Scan, ExclusiveSum1D) { | ||
| const int block_work_size = 6912; |
| Binary<To, op> binop; | ||
| To accum = tptr[blockIdx_x - 1]; | ||
|
|
||
|
|
Fix output indices on broadcast step of 1D exclusive scan in CUDA
Fix output indices on broadcast step of exclusive scan in CUDA
Fix output indices on broadcast step of exclusive scan in CUDA
Fix output indices on broadcast step of exclusive scan
| } | ||
|
|
||
| template<typename To, af_op_t op, int dim> | ||
| template<typename To, af_op_t op, int dim, bool inclusive_scan> |
| const int wid = (blockIdx.y + blockIdx.z * gridDim.y) / blocks_y; | ||
| const int blockIdx_x = blockIdx.x - (blocks_x) * zid; | ||
| const int blockIdx_y = (blockIdx.y + blockIdx.z * gridDim.y) - (blocks_y) * wid; | ||
| const int blockIdx_x = blockIdx.x - (blocks_x)*zid; |
| int offset = !inclusive_scan; | ||
| for (int k = 0, id = xid; | ||
| k < lim && id < oInfo.dims[0]; | ||
| k < lim && id + offset < oInfo.dims[0]; |
| int offset = !inclusive_scan; | ||
| for (int k = 0, id = xid; | ||
| k < lim && id < out.dims[0]; | ||
| k < lim && id + offset < out.dims[0]; |
Mark Poscablo added 7 commits
December 20, 2018 07:11
… to loop initialization.
9prady9 pushed a commit to 9prady9/arrayfire that referenced this pull request
Mar 27, 2019
Fix output indices on broadcast step of 1D exclusive scan in CUDA (cherry picked from commit 70be3d4 )
umar456 pushed a commit that referenced this pull request
Apr 17, 2019
Fix output indices on broadcast step of 1D exclusive scan in CUDA (cherry picked from commit 70be3d4 )
Labels
None yet