ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-13 17:55:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor

Browse Source
This commit is contained in:
Chao Liu
2018-11-15 23:53:23 -06:00
parent adf4b173b3
commit 73480fee36
4 changed files with 115 additions and 209 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
src/include/direct_convolution_2.cuh
View File

@@ -92,10 +92,10 @@ __device__ void blockwise_convolution(InDesc,
auto f_copy = [](const TFloat& src, TFloat& dst) { dst = src; };
// copy input tensor into register
threadwise_4d_tensor_op_in<TFloat,
decltype(in_thread_src_desc),
decltype(in_thread_dst_desc),
decltype(f_copy)>(
threadwise_4d_tensor_op_binary<TFloat,
decltype(in_thread_src_desc),
decltype(in_thread_dst_desc),
decltype(f_copy)>(
in_thread_src_desc,
p_in_lds + in_desc.Get1dIndex(
n_thread_work_begin, 0, hi_thread_work_begin, wi_thread_work_begin),
@@ -107,10 +107,10 @@ __device__ void blockwise_convolution(InDesc,
++k_thread_work_begin)
{
// copy weight tensor into register
threadwise_4d_tensor_op_wei<TFloat,
decltype(wei_thread_src_desc),
decltype(wei_thread_dst_desc),
decltype(f_copy)>(
threadwise_4d_tensor_op_binary<TFloat,
decltype(wei_thread_src_desc),
decltype(wei_thread_dst_desc),
decltype(f_copy)>(
wei_thread_src_desc,
p_wei_lds + wei_desc.Get1dIndex(k_thread_work_begin, 0, 0, 0),
wei_thread_dst_desc,
@@ -118,10 +118,10 @@ __device__ void blockwise_convolution(InDesc,
f_copy);
// copy output tensor into register
threadwise_4d_tensor_op_out<TFloat,
decltype(out_thread_src_desc),
decltype(out_thread_dst_desc),
decltype(f_copy)>(
threadwise_4d_tensor_op_binary<TFloat,
decltype(out_thread_src_desc),
decltype(out_thread_dst_desc),
decltype(f_copy)>(
out_thread_src_desc,
p_out_lds + out_desc.Get1dIndex(n_thread_work_begin,
k_thread_work_begin,
@@ -143,10 +143,10 @@ __device__ void blockwise_convolution(InDesc,
p_out_thread);
// accumulate output tensor into LDS
threadwise_4d_tensor_op_out<TFloat,
decltype(out_thread_dst_desc),
decltype(out_thread_src_desc),
decltype(f_copy)>(
threadwise_4d_tensor_op_binary<TFloat,
decltype(out_thread_dst_desc),
decltype(out_thread_src_desc),
decltype(f_copy)>(
out_thread_dst_desc,
p_out_thread,
out_thread_src_desc,