gpu::xetla::subgroup::elemwise_reduce_op_stream_k_t< reduce_kind, dtype_in, arch_tag > Struct Template Reference

Is the element-wise reduce op functor, specialized for stream_k dispatch Load partial sum from scratchspace Reduce in GRF Store zero to scratchspace Do these steps with smaller tiles to minimize GRF pressure. More...

#include <tile_op_functor.hpp>

Detailed Description

template<reduce_op reduce_kind, typename dtype_in, gpu_arch arch_tag = gpu_arch::Xe>
struct gpu::xetla::subgroup::elemwise_reduce_op_stream_k_t< reduce_kind, dtype_in, arch_tag >

Is the element-wise reduce op functor, specialized for stream_k dispatch Load partial sum from scratchspace Reduce in GRF Store zero to scratchspace Do these steps with smaller tiles to minimize GRF pressure.

Template Parameters

reduce_kind	Is the reduce type, can be sum, prod, min and max.
dtype_in	Is the memory side buffer data type.
arch_tag	Is the hardware architecture tag.