ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-21 13:29:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

experimenting

Browse Source
This commit is contained in:
Chao Liu
2019-03-24 12:09:57 -05:00
parent f35c64eb78
commit 766b0a9eaf
33 changed files with 1886 additions and 1822 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/include/ConstantMatrixDescriptor.hip.hpp
View File

@@ -1,7 +1,7 @@
#pragma once
#include "common.hip.hpp"
template <unsigned NRow_, unsigned NCol_, unsigned RowStride_>
template <index_t NRow_, index_t NCol_, index_t RowStride_>
struct ConstantMatrixDescriptor
{
__host__ __device__ constexpr ConstantMatrixDescriptor()
@@ -9,24 +9,28 @@ struct ConstantMatrixDescriptor
static_assert(NCol_ <= RowStride_, "wrong! NCol > RowStride!");
}
__host__ __device__ constexpr unsigned NRow() const { return NRow_; }
__host__ __device__ constexpr index_t NRow() const { return NRow_; }
__host__ __device__ constexpr unsigned NCol() const { return NCol_; }
__host__ __device__ constexpr index_t NCol() const { return NCol_; }
__host__ __device__ constexpr unsigned RowStride() const { return RowStride_; }
__host__ __device__ constexpr index_t RowStride() const { return RowStride_; }
__host__ __device__ constexpr auto GetLengths() const { return Sequence<NRow_, NCol_>{}; }
__host__ __device__ constexpr unsigned GetElementSize() const { return NRow_ * NCol_; }
__host__ __device__ constexpr index_t GetElementSize() const { return NRow_ * NCol_; }
__host__ __device__ constexpr unsigned GetElementSpace() const { return NRow_ * RowStride_; }
__host__ __device__ constexpr index_t GetElementSpace() const { return NRow_ * RowStride_; }
__host__ __device__ unsigned Get1dIndex(unsigned irow, unsigned icol) const
__host__ __device__ index_t Get1dIndex(index_t irow, index_t icol) const
{
#if DEVICE_BACKEND_HIP
return __mul24(irow, RowStride_) + icol;
#else
return irow * RowStride_ + icol;
#endif
}
template <unsigned SubNRow, unsigned SubNCol>
template <index_t SubNRow, index_t SubNCol>
__host__ __device__ constexpr auto MakeSubMatrixDescriptor(Number<SubNRow>,
Number<SubNCol>) const
{
@@ -34,13 +38,13 @@ struct ConstantMatrixDescriptor
}
};
template <unsigned NRow, unsigned NCol>
template <index_t NRow, index_t NCol>
__host__ __device__ constexpr auto make_ConstantMatrixDescriptor(Number<NRow>, Number<NCol>)
{
return ConstantMatrixDescriptor<NRow, NCol, NCol>{};
}
template <unsigned NRow, unsigned NCol, unsigned RowStride>
template <index_t NRow, index_t NCol, index_t RowStride>
__host__ __device__ constexpr auto
make_ConstantMatrixDescriptor(Number<NRow>, Number<NCol>, Number<RowStride>)
{