mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-12 01:10:17 +00:00
b8b2d0a6d1
* add threadwise copy the copy a tensor in one copy, added kpack to DL GEMM * add kpack into fwd v4r5 nchw fp32
46 lines
1014 B
C++
46 lines
1014 B
C++
#ifndef CK_COMMON_HEADER_HPP
|
|
#define CK_COMMON_HEADER_HPP
|
|
|
|
#include "config.hpp"
|
|
#include "array.hpp"
|
|
#include "container_helper.hpp"
|
|
#include "statically_indexed_array.hpp"
|
|
#include "container_element_picker.hpp"
|
|
#include "multi_index.hpp"
|
|
#include "data_type.hpp"
|
|
#include "float_type.hpp"
|
|
#include "functional.hpp"
|
|
#include "functional2.hpp"
|
|
#include "functional3.hpp"
|
|
#include "functional4.hpp"
|
|
#include "in_memory_operation.hpp"
|
|
#include "integral_constant.hpp"
|
|
#include "math.hpp"
|
|
#include "number.hpp"
|
|
#include "sequence.hpp"
|
|
#include "sequence_helper.hpp"
|
|
#include "synchronization.hpp"
|
|
#include "tuple.hpp"
|
|
#include "tuple_helper.hpp"
|
|
#include "type.hpp"
|
|
#include "utility.hpp"
|
|
#include "magic_division.hpp"
|
|
#include "static_buffer.hpp"
|
|
#include "dynamic_buffer.hpp"
|
|
|
|
// TODO: remove this
|
|
#if CK_USE_AMD_INLINE_ASM
|
|
#include "amd_inline_asm.hpp"
|
|
#endif
|
|
|
|
#if CK_USE_AMD_DLOP
|
|
#include "amd_dlop.hpp"
|
|
#endif
|
|
|
|
#if CK_USE_AMD_XDLOPS
|
|
#include "amd_xdlops.hpp"
|
|
#include "amd_xdlops_inline_asm.hpp"
|
|
#endif
|
|
|
|
#endif
|