XeTLA v0.3.6
IntelĀ® Xe Templates for Linear Algebra - API Definition Document
 
Loading...
Searching...
No Matches
store_xe.hpp File Reference

C++ API. More...

Include dependency graph for store_xe.hpp:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes

struct  gpu::xetla::subgroup::detail::check_store_type< tile_t, payload_t >
 

Namespaces

namespace  gpu
 
namespace  gpu::xetla
 
namespace  gpu::xetla::subgroup
 
namespace  gpu::xetla::subgroup::detail
 

Functions

template<cache_hint L1 = cache_hint::write_back, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t >
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_global_2d_xe > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload)
 Is the func storing data from register file to global memory.
 
template<cache_hint L1 = cache_hint::write_back, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t >
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_global_block_1d_xe > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload)
 Is the func storing data from register file to global memory.
 
template<cache_hint L1 = cache_hint::write_back, cache_hint L3 = cache_hint::write_back, typename tile_t , typename payload_t , typename oob_check_tag = global_atomic_oob_check_on_tag>
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_global_unaligned_2d_xe > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload, oob_check_tag tag={})
 Is the func storing data from register file to unaligned global memory surface.
 
template<cache_hint L1 = cache_hint::uncached, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t , typename oob_check_tag = global_atomic_oob_check_on_tag>
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_global_atomic_xe > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload, oob_check_tag tag={})
 Is the func storing data from register file to global memory enable atomic adding data into the same buffer, but support float32, float64, uint32_t, uint64_t and int type.
 
template<cache_hint L1 = cache_hint::write_back, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t >
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_local_scatter_xe > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload)
 Is the func storing data from register file to shared local memory, which supports the memory surface 2d scenario.
 
template<cache_hint L1 = cache_hint::write_back, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t >
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_local_scatter_vnni_col_xe > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload)
 Is the data store func from register file to local shared memory, where the data in register is vnni packed and col major.
 
template<cache_hint L1 = cache_hint::write_back, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t >
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_local_block_1d_xe &&tile_t::block_size_y !=1 > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload)
 Is the data store func from register file to shared local memory, where supports memory surface 1d or 2d scenario, and we always assume dst memory layout is row major.
 
template<cache_hint L1 = cache_hint::write_back, cache_hint L2 = cache_hint::write_back, typename tile_t , typename payload_t >
__XETLA_API std::enable_if_t< detail::check_store_type< tile_t, payload_t >::is_local_block_1d_xe &&tile_t::tile_size_y==1 &&tile_t::block_size_y==1 > gpu::xetla::subgroup::tile_store (tile_t &tile, payload_t &payload)
 Is the func storing data from register file to shared local memory, the data in registers will be stored to SLM in 1d mode, and we always assume dst memory layout is row major.
 

Detailed Description

C++ API.