mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-17 19:40:04 +00:00
b827099a27
* start fixing 16bit data packing
* adding StaticTensor
* adding StaticTensor
* adding StaticTensor
* add missing constexpr
* adding static tensor
* adding static tensor
* adding transpose
* add inline asm for transpose 2x2 of half_t
* add general transpose_vectors(), but have unnecessary register initialization using v_mov
* fix unnecessary register initialization in transpose_vector by using more pass-by-reference
* add hardcoded logic for NHWC wrw
* improve asm for v_pack
* make ThreadwiseTensorSliceTransfer_v3r2 support any tensor
* tweak
* reorganize file
[ROCm/composable_kernel commit: b491ebf384]
51 lines
1.2 KiB
C++
51 lines
1.2 KiB
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 "data_type_enum.hpp"
|
|
#include "data_type_enum_helper.hpp"
|
|
#include "functional.hpp"
|
|
#include "functional2.hpp"
|
|
#include "functional3.hpp"
|
|
#include "functional4.hpp"
|
|
#include "enable_if.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 "magic_division.hpp"
|
|
#include "utility.hpp"
|
|
#include "c_style_pointer_cast.hpp"
|
|
#include "amd_address_space.hpp"
|
|
#include "amd_buffer_addressing.hpp"
|
|
#include "static_buffer.hpp"
|
|
// TODO remove this
|
|
#include "static_buffer_of_vector_type_v2.hpp"
|
|
#include "dynamic_buffer.hpp"
|
|
#include "is_known_at_compile_time.hpp"
|
|
#include "transpose_vectors.hpp"
|
|
|
|
#include "inner_product.hpp"
|
|
|
|
// TODO: remove this
|
|
#if CK_USE_AMD_INLINE_ASM
|
|
#include "amd_inline_asm.hpp"
|
|
#endif
|
|
|
|
#if CK_USE_AMD_XDLOPS
|
|
#include "amd_xdlops.hpp"
|
|
#endif
|
|
|
|
#endif
|