gpu::xetla::group::gemm_selector_t< dtype_a, dtype_b, mem_layout_a, mem_layout_b, mem_space_a, mem_space_b, alignment_a, alignment_b, dtype_acc, tile_shape, k_stride, engine, arch_tag, stages, sync_freq, enable > Class Template Reference

Gemm selection functor. More...

#include <api.hpp>

Detailed Description

template<typename dtype_a, typename dtype_b, mem_layout mem_layout_a, mem_layout mem_layout_b, mem_space mem_space_a, mem_space mem_space_b, int alignment_a, int alignment_b, typename dtype_acc, typename tile_shape, int k_stride, mma_engine engine, gpu_arch arch_tag, int stages = 3, int sync_freq = 0, class enable = void>
class gpu::xetla::group::gemm_selector_t< dtype_a, dtype_b, mem_layout_a, mem_layout_b, mem_space_a, mem_space_b, alignment_a, alignment_b, dtype_acc, tile_shape, k_stride, engine, arch_tag, stages, sync_freq, enable >

Gemm selection functor.

Template Parameters

dtype_a	Is the memory data type of matA.
dtype_b	Is the memory data type of matB.
mem_layout_a	Is the memory layout of matA.
mem_layout_b	Is the memory layout of matB.
mem_space_a	Is the memory space of matA.
mem_space_b	Is the memory space of matB.
alignment_a	Is the memory alignment of matA.
alignment_b	Is the memory alignment of matB.
dtype_acc	Is the compute data type.
tile_shape	Is the workgroup-level tile shape.
k_stride	Is the accumulate stride along k-dim.
engine	Is the compute engine type.
arch_tag	Is the HW architecture.
stages	Is the prefetch distance.
sync_freq	Is the group sync frequency.