ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-18 20:09:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add declarations for atomic add for fp16 and unsigned short. (#2483)

Browse Source 
* add template for fp16 atomic add

* add template for unsigned short atomic add

* use atomicCAS in atomic add for fp16 and unsigned short

* revrt back to atomic add using casting

[ROCm/composable_kernel commit: 1b66f3f4a3]
This commit is contained in:
Illia Silin
2025-07-10 07:18:56 -07:00
committed by GitHub
parent 84a7b00497
commit e61ceee502
2 changed files with 22 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
include/ck/utility/generic_memory_space_atomic.hpp
View File

@@ -32,6 +32,22 @@ __device__ float atomic_add<float>(float* p_dst, const float& x)
return atomicAdd(p_dst, x);
}
template <>
__device__ unsigned short atomic_add<unsigned short>(unsigned short* p_dst, const unsigned short& x)
{
// Use atomicAdd with unsigned int
return static_cast<unsigned short>(
atomicAdd(reinterpret_cast<unsigned int*>(p_dst), static_cast<unsigned int>(x)));
}
template <>
__device__ _Float16 atomic_add<_Float16>(_Float16* p_dst, const _Float16& x)
{
// Use atomicAdd with unsigned int
return static_cast<_Float16>(
atomicAdd(reinterpret_cast<unsigned int*>(p_dst), static_cast<unsigned int>(x)));
}
template <>
__device__ double atomic_add<double>(double* p_dst, const double& x)
{