gpu::xetla::group::global_reduce_t< reduce_kind, tile_shape_acc, tile_shape_cnt, mem_desc_acc_t, mem_desc_cnt_t, num_group_reduction, counter_size, arch_tag, enable > Class Template Reference

Cross group global reduction. More...

#include <global_reduction.hpp>

Detailed Description

template<reduce_op reduce_kind, typename tile_shape_acc, typename tile_shape_cnt, typename mem_desc_acc_t, typename mem_desc_cnt_t, uint32_t num_group_reduction, uint32_t counter_size, gpu_arch arch_tag, class enable = void>
class gpu::xetla::group::global_reduce_t< reduce_kind, tile_shape_acc, tile_shape_cnt, mem_desc_acc_t, mem_desc_cnt_t, num_group_reduction, counter_size, arch_tag, enable >

Cross group global reduction.

Template Parameters

reduce_kind	Is the reduction type.
tile_shape_acc	Is the group-level tile shape for accumulation tile.
tile_shape_cnt	Is the group-level tile shape for counter tile.
mem_desc_acc_t	Is the memory descriptor of accumulation buffer.
mem_desc_cnt_t	Is the memory descriptor of counter buffer.
num_group_reduction	Is the number of workgroups to do the reduction.
counter_size	Is the stride to the next counter.
arch_tag	Is the HW architecture.