ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-24 06:44:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

Dynamic tensor descriptor (#24)

Browse Source 
* support dynamic tensor descriptor

* use buffer load OOB feature for padding case

* add navi support

* add int8x4 inference kernel

Co-authored-by: Chao Liu <chao@ixt-rack-81.local.lan>
Co-authored-by: Jing Zhang <jizhan@amd.com>
This commit is contained in:
Chao Liu
2021-03-25 13:51:11 -05:00
committed by GitHub
parent bbcb67d0aa
commit fcbb978828
85 changed files with 14129 additions and 2532 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
driver/include/conv_common.hpp
View File

@@ -51,26 +51,24 @@ constexpr auto get_convolution_output_default_4d_tensor_descriptor(
}
template <class InDesc, class WeiDesc, class OutDesc>
constexpr std::size_t calculate_convolution_flops(InDesc, WeiDesc, OutDesc)
constexpr std::size_t
calculate_convolution_flops(const InDesc& in_desc, const WeiDesc& wei_desc, const OutDesc& out_desc)
{
using namespace ck;
constexpr auto wei_desc = WeiDesc{};
constexpr auto out_desc = OutDesc{};
constexpr auto I0 = Number<0>{};
constexpr auto I1 = Number<1>{};
constexpr auto I2 = Number<2>{};
constexpr auto I3 = Number<3>{};
constexpr index_t N = out_desc.GetLength(I0);
constexpr index_t K = out_desc.GetLength(I1);
constexpr index_t Ho = out_desc.GetLength(I2);
constexpr index_t Wo = out_desc.GetLength(I3);
const index_t N = out_desc.GetLength(I0);
const index_t K = out_desc.GetLength(I1);
const index_t Ho = out_desc.GetLength(I2);
const index_t Wo = out_desc.GetLength(I3);
constexpr index_t C = wei_desc.GetLength(I1);
constexpr index_t Y = wei_desc.GetLength(I2);
constexpr index_t X = wei_desc.GetLength(I3);
const index_t C = wei_desc.GetLength(I1);
const index_t Y = wei_desc.GetLength(I2);
const index_t X = wei_desc.GetLength(I3);
return std::size_t(2) * N * K * Ho * Wo * C * Y * X;
}