github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-12 20:54:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

[Inference] Delete duplicated copy_vector (#5716)

Browse Source
This commit is contained in:
傅剑寒
2024-05-14 14:35:33 +08:00
committed by GitHub
parent 7806842f2d
commit 121d7ad629
6 changed files with 28 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
extensions/csrc/kernel/cuda/utils/vec_copy.h
View File

@@ -8,25 +8,8 @@ namespace colossalAI {
namespace cuda {
namespace utils {
// Note(LiuYang): Depreciated
template <typename T, int VecSize>
__device__ __inline__ void copy_vector(T *dst, const T *src) {
using VT = typename common::VecTypeTrait<T, VecSize>::Type;
*(reinterpret_cast<VT *>(dst)) = *(reinterpret_cast<const VT *>(src));
}
template <>
__device__ __inline__ void copy_vector<float, 8>(float *dst, const float *src) {
// Since the maximum memory alignment length is 128 bits, we choose float4
// here.
*(reinterpret_cast<float4 *>(dst)) = *(reinterpret_cast<const float4 *>(src));
*(reinterpret_cast<float4 *>(dst + 4)) =
*(reinterpret_cast<const float4 *>(src + 4));
}
// Note(LiuYang): Depreciated
template <typename T, int VecSize>
__device__ __inline__ void copy_zero_vector(T *dst) {
__device__ __inline__ void copy_zero(T *dst) {
using VT = typename common::VecTypeTrait<T, VecSize>::Type;
*(reinterpret_cast<VT *>(dst)) = funcs::CastFunctor<float, VT>()(0.0f);
}