[NFC] polish colossalai/kernel/cuda_native/csrc/kernels/transform_kernels.cu code stype (#629)

pull/673/head
puck_WCR 2022-04-01 20:20:54 +08:00 committed by binmakeswell
parent c1bed0d998
commit 01cc941e1d
1 changed files with 6 additions and 8 deletions

View File

@ -135,10 +135,9 @@ __global__ void bias_add_transform_20314(T *output, const T *input,
const T *bias, int dim_3, int dim_4); const T *bias, int dim_3, int dim_4);
template <> template <>
__global__ void bias_add_transform_20314<float>(float *output, __global__ void
const float *input, bias_add_transform_20314<float>(float *output, const float *input,
const float *bias, int dim_3, const float *bias, int dim_3, int dim_4) {
int dim_4) {
int id0 = blockIdx.x; int id0 = blockIdx.x;
int id1 = blockIdx.y; int id1 = blockIdx.y;
int id2 = blockIdx.z; int id2 = blockIdx.z;
@ -174,10 +173,9 @@ __global__ void bias_add_transform_20314<float>(float *output,
} }
template <> template <>
__global__ void bias_add_transform_20314<__half>(__half *output, __global__ void
const __half *input, bias_add_transform_20314<__half>(__half *output, const __half *input,
const __half *bias, int dim_3, const __half *bias, int dim_3, int dim_4) {
int dim_4) {
int id0 = blockIdx.x; int id0 = blockIdx.x;
int id1 = blockIdx.y; int id1 = blockIdx.y;
int id2 = blockIdx.z; int id2 = blockIdx.z;