SparseConvCUDAKernels.h

Go to the documentation of this file.

// ----------------------------------------------------------------------------
// -                        CloudViewer: www.cloudViewer.org                  -
// ----------------------------------------------------------------------------
// Copyright (c) 2018-2024 www.cloudViewer.org
// SPDX-License-Identifier: MIT
// ----------------------------------------------------------------------------
 
#pragma once
#include <vector>
 
namespace cloudViewer {
namespace ml {
namespace impl {
 
template <class TReal, class TIndex, class TKernelIndex>
void FillColumn(const cudaStream_t& stream,
                TReal* columns,
                int in_channels,
                TIndex begin_idx,
                TIndex end_idx,
                TIndex num_out,
                TIndex num_inp,
                const TReal* const __restrict__ inp_features,
                const TReal* const __restrict__ inp_importance,
                size_t neighbors_index_size,
                const TIndex* const __restrict__ neighbors_index,
                const TKernelIndex* const __restrict__ neighbors_kernel_index,
                const TReal* const __restrict__ neighbors_importance,
                const int64_t* const __restrict__ neighbors_row_splits,
                const int num_kernel_elements,
                bool normalize);
 
template <class TReal, class TIndex, class TKernelIndex>
void FillColumnTranspose(
        const cudaStream_t& stream,
        TReal* columns,
        int in_channels,
        TIndex begin_idx,
        TIndex end_idx,
        TIndex num_out,
        TIndex num_inp,
        const TReal* const __restrict__ inp_features,
        const TReal* const __restrict__ inp_neighbors_importance_sum,
        const int64_t* const __restrict__ inp_neighbors_prefix_sum,
        size_t neighbors_index_size,
        const TIndex* const __restrict__ neighbors_index,
        const TKernelIndex* const __restrict__ neighbors_kernel_index,
        const TReal* const __restrict__ neighbors_importance,
        const int64_t* const __restrict__ neighbors_row_splits,
        const int num_kernel_elements,
        bool normalize);
 
}  // namespace impl
}  // namespace ml
}  // namespace cloudViewer