XeTLA GEMM_UNIVERSAL

This is a gemm_universal API built on top of xetla group level API to provide a more convenient way to compose a GEMM_UNIVERSAL kernel. More...

Classes
class	gpu::xetla::kernel::gemm_universal_t< dispatch_policy, gemm_t, epilogue_t, enable >
	GEMM_UNIVERSAL functor. More...
struct	gpu::xetla::kernel::group_swizzle_default< arch_tag_ >
	Default GROUP_SWIZZLE implementation. More...
struct	gpu::xetla::kernel::group_swizzle_snake< wg_num_n_, arch_tag_ >
	GROUP_SWIZZLE implementation of snake curve. More...
struct	gpu::xetla::kernel::dispatch_policy_default< group_swizzle_policy_ >
	Default GEMM_UNIVERSAL implementation. More...
struct	gpu::xetla::kernel::dispatch_policy_kslicing< group_swizzle_policy_, global_ratio_, local_ratio_ >
	Kslicing GEMM_UNIVERSAL implementation. More...
struct	gpu::xetla::kernel::dispatch_policy_stream_k< arch_tag_ >
	StreamK GEMM implementation. More...
class	gpu::xetla::kernel::gemm_universal_t< dispatch_policy_default< group_swizzle_ >, gemm_t_, epilogue_t_, std::enable_if_t<(group_swizzle_::arch_tag==gpu_arch::Xe)> >
	Default GEMM_UNIVERSAL functor, specialized for Xe architecture. More...
class	gpu::xetla::kernel::gemm_universal_t< dispatch_policy_kslicing< group_swizzle_, num_global_kslicing_, num_local_kslicing_ >, gemm_t_, epilogue_t_, std::enable_if_t<(group_swizzle_::arch_tag==gpu_arch::Xe)> >
	Is the gemm_universal functor, specialized in kslicing dispatch policy and Xe architecture. More...
class	gpu::xetla::kernel::gemm_universal_t< dispatch_policy_stream_k< gpu_arch::Xe >, gemm_t_, epilogue_t_ >
	Default GEMM_UNIVERSAL functor, specialized for Xe architecture. More...

Detailed Description

This is a gemm_universal API built on top of xetla group level API to provide a more convenient way to compose a GEMM_UNIVERSAL kernel.