ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-25 07:14:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

adding implicit gemm

Browse Source 
[ROCm/composable_kernel commit: 84d9802d30]
This commit is contained in:
Chao Liu
2019-01-15 00:11:30 -06:00
parent 50256bbcfe
commit aa885b185d
15 changed files with 268 additions and 169 deletions
Download Patch File Download Diff File Expand all files Collapse all files

46
src/include/ConstantMatrixDescriptor.cuh Normal file
View File

@@ -0,0 +1,46 @@
#pragma once
#include "common.cuh"
template <unsigned NRow, unsigned NCol, unsigned RowStride>
struct ConstantMatrixDescriptor
{
__host__ __device__ ConstantMatrixDescriptor()
{
static_assert(NCol <= RowStride, "wrong! NCol > RowStride!");
}
__host__ __device__ constexpr unsigned GetNumberOfRow() const { return NRow; }
__host__ __device__ constexpr unsigned GetNumberOfColumn() const { return NCol; }
__host__ __device__ constexpr unsigned GetRowStride() const { return RowStride; }
__host__ __device__ constexpr unsigned GetElementSize() const { return NRow * NCol; }
__host__ __device__ constexpr unsigned GetElementSpace() const { return NRow * RowStride; }
__host__ __device__ unsigned Get1dIndex(unsigned irow, unsigned icol) const
{
return irow * RowStride + icol;
}
template <unsigned SubNRow, unsigned SubNCol>
__host__ __device__ constexpr auto MakeSubMatrixDescriptor(Number<SubNRow>,
Number<SubNCol>) const
{
return ConstantMatrixDescriptor<SubNRow, SubNCol, RowStride>{};
}
};
template <unsigned NRow, unsigned NCol>
__host__ __device__ constexpr auto make_ConstantMatrixDescriptor(Number<NRow>, Number<NCol>)
{
return ConstantMatrixDescriptor<NRow, NCol, NCol>{};
}
template <unsigned NRow, unsigned NCol, unsigned RowStride>
__host__ __device__ constexpr auto
make_ConstantMatrixDescriptor(Number<NRow>, Number<NCol>, Number<RowStride>)
{
return ConstantMatrixDescriptor<NRow, NCol, RowStride>{};
}