gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag > Struct Template Reference

xetla nbarrier definition API. More...

#include <raw_send_nbarrier.hpp>

Public Member Functions
__XETLA_API void	init_nbarrier (uint8_t nbarrier_id, nbarrier_role role=nbarrier_role::producer_consumer)
__XETLA_API void	arrive ()
	named barrier signal from subgroup.
__XETLA_API void	wait ()
	named barrier wait within subgroup.
__XETLA_API void	arrive_wait ()
	named barrier signal from subgroup.

Public Attributes
xetla_vector< uint32_t, 16 >	nbar
	Description of named barrier objection.
uint32_t	barrier_id

Detailed Description

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>
struct gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >

xetla nbarrier definition API.

This is the API to define a named barrier within subgroup.

Template Parameters

num_producers	is the number of subgroups participating the barrier as producer.
num_consumers	is the number of subgroups participating the barrier as consumer.

Member Function Documentation

arrive()

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>

__XETLA_API void gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >::arrive ( )

inline

named barrier signal from subgroup.

Parameters

bar	is the named barrier object.

arrive_wait()

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>

__XETLA_API void gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >::arrive_wait ( )

inline

named barrier signal from subgroup.

init_nbarrier()

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>

__XETLA_API void gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >::init_nbarrier ( uint8_t  nbarrier_id, nbarrier_role  role = nbarrier_role::producer_consumer  )

inline

Parameters

role	is the role of subgroup when participating the barrier.
nbarrier_id	[in] is the id of the barrier. note: all subgroups participating the barrier should have the same barrier_id. Here is the bspec link https://gfxspecs.intel.com/Predator/Home/Index/54006

wait()

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>

__XETLA_API void gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >::wait ( )

inline

named barrier wait within subgroup.

Member Data Documentation

barrier_id

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>

uint32_t gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >::barrier_id

nbar

template<uint8_t num_producers = 1, uint8_t num_consumers = 1, gpu_arch arch_tag = gpu_arch::Xe>

xetla_vector<uint32_t, 16> gpu::xetla::xetla_nbarrier_t< num_producers, num_consumers, arch_tag >::nbar

Description of named barrier objection.

Structure is defined in here.