add asm into lds_double_buffer version

This commit is contained in:
Chao Liu
2019-04-04 10:38:49 -05:00
parent 155d78594b
commit fbc7817bbb
4 changed files with 109 additions and 54 deletions

View File

@@ -271,7 +271,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
for(index_t i = 0; i < nrepeat; ++i)
{
constexpr auto gridwise_conv =
#if 1
#if 0
gridwise_implicit_gemm_convolution_2_chwn_cyxk_khwn
#else
gridwise_implicit_gemm_convolution_2_chwn_cyxk_khwn_lds_double_buffer
@@ -306,7 +306,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
float time = launch_kernel(gridwise_conv.Run,
dim3(GridSize),
dim3(BlockSize),
gridwise_conv.GetSharedMemoryUsage(),
gridwise_conv.GetDynamicSharedMemoryUsage(),
static_cast<T*>(in_chwn_device_buf.GetDeviceBuffer()),
static_cast<T*>(wei_cyxk_device_buf.GetDeviceBuffer()),
static_cast<T*>(out_khwn_device_buf.GetDeviceBuffer()));