This commit is contained in:
Chao Liu
2019-09-26 14:59:19 -05:00
parent 51a9fa1dbd
commit b12bbceebc
16 changed files with 2377 additions and 2304 deletions

View File

@@ -3,7 +3,7 @@
#include "device.hpp"
#include "tensor.hpp"
#include "gridwise_convolution_kernel_wrapper.hpp"
#include "gridwise_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw.hpp"
//#include "gridwise_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw.hpp"
#include "gridwise_convolution_implicit_gemm_v4r1_nchw_kcyx_nkhw_lds_double_buffer.hpp"
template <class T,

View File

@@ -3,7 +3,7 @@
#include "device.hpp"
#include "tensor.hpp"
#include "gridwise_convolution_kernel_wrapper.hpp"
#include "gridwise_convolution_implicit_gemm_v4r4_nchw_kcyx_nkhw.hpp"
//#include "gridwise_convolution_implicit_gemm_v4r4_nchw_kcyx_nkhw.hpp"
#include "gridwise_convolution_implicit_gemm_v4r4_nchw_kcyx_nkhw_lds_double_buffer.hpp"
using namespace ck;
@@ -164,7 +164,7 @@ void device_convolution_implicit_gemm_v4r4_nchw_kcyx_nkhw(InDesc,
printf("%s: BlockSize %u, GridSize %u \n", __func__, BlockSize, GridSize);
constexpr auto gridwise_conv =
#if 1
#if 0
GridwiseConvolutionImplicitGemm_v4r4_nchw_kcyx_nkhw
#else
GridwiseConvolutionImplicitGemm_v4r4_nchw_kcyx_nkhw_lds_double_buffer

View File

@@ -51,7 +51,7 @@ void device_convolution_implicit_gemm_v4r4_nchw_kcyx_nkhw_padded(InDesc,
wei_kcyx_device_buf.ToDevice(wei_kcyx.mData.data());
out_nkhw_device_buf.ToDevice(out_nkhw.mData.data());
#if 0
#if 1
constexpr index_t BlockSize = 256;
constexpr index_t BPerBlock = 128;