mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-11 17:00:18 +00:00
experimenting
This commit is contained in:
@@ -160,7 +160,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
|
||||
constexpr unsigned WeiBlockCopyDataPerRead = 4;
|
||||
|
||||
constexpr unsigned BlockSize = 128;
|
||||
#elif 1
|
||||
#elif 0
|
||||
// 1x1, 28x28, 256 thread
|
||||
constexpr unsigned BPerBlock = 128;
|
||||
constexpr unsigned KPerBlock = 128;
|
||||
@@ -211,7 +211,7 @@ void device_implicit_gemm_convolution_2_chwn_cyxk_khwn(InDesc,
|
||||
for(unsigned i = 0; i < nrepeat; ++i)
|
||||
{
|
||||
float time = launch_kernel(
|
||||
#if 0
|
||||
#if 1
|
||||
gridwise_implicit_gemm_convolution_2_chwn_cyxk_khwn
|
||||
#else
|
||||
gridwise_implicit_gemm_convolution_2_chwn_cyxk_khwn_lds_double_buffer
|
||||
|
||||
@@ -661,9 +661,9 @@ int main(int argc, char* argv[])
|
||||
device_direct_convolution_2_nchw_kcyx_nkhw
|
||||
#elif 0
|
||||
device_direct_convolution_2_vectorized_nchw_kcyx_nkhw
|
||||
#elif 1
|
||||
device_implicit_gemm_convolution_1_chwn_cyxk_khwn
|
||||
#elif 0
|
||||
device_implicit_gemm_convolution_1_chwn_cyxk_khwn
|
||||
#elif 1
|
||||
device_implicit_gemm_convolution_2_chwn_cyxk_khwn
|
||||
#endif
|
||||
(in_nchw_desc, in_nchw, wei_kcyx_desc, wei_kcyx, out_nkhw_desc, out_nkhw_device, nrepeat);
|
||||
|
||||
Reference in New Issue
Block a user