ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Add templates for fp16 and unsigned short atomic add to fix FBGEMM bu…" (#2474)

Browse Source 
This reverts commit 112b47e885.
This commit is contained in:
Illia Silin
2025-07-08 19:01:26 -07:00
committed by GitHub
parent 112b47e885
commit 93420ecf89
2 changed files with 22 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
include/ck/utility/generic_memory_space_atomic.hpp
View File

@@ -32,33 +32,6 @@ __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)
{
unsigned short old_val, new_val;
do
{
old_val = *p_dst;
new_val = old_val + x;
} while(atomicCAS(p_dst, old_val, new_val) != old_val);
return old_val;
}
template <>
__device__ _Float16 atomic_add<_Float16>(_Float16* p_dst, const _Float16& x)
{
_Float16 old_val, new_val;
do
{
old_val = *p_dst;
new_val = old_val + x; // Proper FP16 addition
} while(atomicCAS(reinterpret_cast<unsigned short*>(p_dst),
*reinterpret_cast<unsigned short*>(&old_val),
*reinterpret_cast<unsigned short*>(&new_val)) !=
*reinterpret_cast<unsigned short*>(&old_val));
return old_val;
}
template <>
__device__ double atomic_add<double>(double* p_dst, const double& x)
{