There is no need to make the fp16 full reduction kernel a static function.

This commit is contained in:
Benoit Steiner 2016-05-24 23:11:56 -07:00
parent b5d6b52a4d
commit 0835667329

View File

@ -159,7 +159,7 @@ __global__ void ReductionInitKernelHalfFloat(Reducer reducer, const Self input,
template <int BlockSize, int NumPerThread, typename Self,
typename Reducer, typename Index>
static __global__ void FullReductionKernelHalfFloat(Reducer reducer, const Self input, Index num_coeffs,
__global__ void FullReductionKernelHalfFloat(Reducer reducer, const Self input, Index num_coeffs,
half* output, half2* scratch) {
eigen_assert(NumPerThread % 2 == 0);