C++ API. More...


Go to the source code of this file.
Namespaces | |
| namespace | gpu |
| namespace | gpu::xetla |
| namespace | gpu::xetla::subgroup |
Functions | |
| template<reduce_op reduce_kind, typename dtype_out , typename dtype_acc , int dim, typename mat_t > | |
| __XETLA_API std::enable_if_t<(dim==1), xetla_vector< dtype_out, mat_t::tile_size_y > > | gpu::xetla::subgroup::tile_reduce (mat_t &src) |
| template<reduce_op reduce_kind, typename dtype_out , typename dtype_acc , int dim, typename mat_t > | |
| __XETLA_API std::enable_if_t<(dim==0), xetla_vector< dtype_out, mat_t::tile_size_x > > | gpu::xetla::subgroup::tile_reduce (mat_t &src) |
| template<typename T_dst , typename T_src , bool accumulate = true, typename dtype_acc = float, uint32_t num_acc = 4> | |
| gpu::xetla::subgroup::XETLA_MARKER ("This is only for reduce add, and will be deprecated in future. " "Please use tile_reduce instead.") __XETLA_API typename std | |
| Reduce 2d src tile to the 1d tile, and output to 1d dst. | |
C++ API.