123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234 |
- #include "ceres/cuda_block_structure.h"
- #ifndef CERES_NO_CUDA
- namespace ceres::internal {
- namespace {
- inline int Dimension(const std::vector<Block>& blocks) {
- if (blocks.empty()) {
- return 0;
- }
- const auto& last = blocks.back();
- return last.size + last.position;
- }
- }
- CudaBlockSparseStructure::CudaBlockSparseStructure(
- const CompressedRowBlockStructure& block_structure, ContextImpl* context)
- : CudaBlockSparseStructure(block_structure, 0, context) {}
- CudaBlockSparseStructure::CudaBlockSparseStructure(
- const CompressedRowBlockStructure& block_structure,
- const int num_col_blocks_e,
- ContextImpl* context)
- : first_cell_in_row_block_(context),
- value_offset_row_block_f_(context),
- cells_(context),
- row_blocks_(context),
- col_blocks_(context) {
-
- std::vector<Block> row_blocks;
-
- const auto& col_blocks = block_structure.cols;
-
- std::vector<int> first_cell_in_row_block;
-
- std::vector<int> value_offset_row_block_f;
-
- std::vector<Cell> cells;
- int f_values_offset = -1;
- num_nonzeros_e_ = 0;
- is_crs_compatible_ = true;
- num_row_blocks_ = block_structure.rows.size();
- num_col_blocks_ = col_blocks.size();
- row_blocks.reserve(num_row_blocks_);
- first_cell_in_row_block.reserve(num_row_blocks_ + 1);
- value_offset_row_block_f.reserve(num_row_blocks_ + 1);
- num_nonzeros_ = 0;
-
-
-
- bool sequential_layout = true;
- int row_block_id = 0;
- num_row_blocks_e_ = 0;
- for (; row_block_id < num_row_blocks_; ++row_block_id) {
- const auto& r = block_structure.rows[row_block_id];
- const int row_block_size = r.block.size;
- const int num_cells = r.cells.size();
- if (num_col_blocks_e == 0 || r.cells.size() == 0 ||
- r.cells[0].block_id >= num_col_blocks_e) {
- break;
- }
- num_row_blocks_e_ = row_block_id + 1;
-
-
-
-
- if (num_cells > 2 && row_block_size > 1) {
- is_crs_compatible_ = false;
- }
- row_blocks.emplace_back(r.block);
- first_cell_in_row_block.push_back(cells.size());
- for (int cell_id = 0; cell_id < num_cells; ++cell_id) {
- const auto& c = r.cells[cell_id];
- const int col_block_size = col_blocks[c.block_id].size;
- const int cell_size = col_block_size * row_block_size;
- cells.push_back(c);
- if (cell_id == 0) {
- DCHECK(c.position == num_nonzeros_e_);
- num_nonzeros_e_ += cell_size;
- } else {
- if (f_values_offset == -1) {
- num_nonzeros_ = c.position;
- f_values_offset = c.position;
- }
- sequential_layout &= c.position == num_nonzeros_;
- num_nonzeros_ += cell_size;
- if (cell_id == 1) {
-
-
- for (auto it = value_offset_row_block_f.rbegin();
- it != value_offset_row_block_f.rend();
- ++it) {
- if (*it != -1) break;
- *it = c.position;
- }
- value_offset_row_block_f.push_back(c.position);
- }
- }
- }
- if (num_cells == 1) {
- value_offset_row_block_f.push_back(-1);
- }
- }
- for (; row_block_id < num_row_blocks_; ++row_block_id) {
- const auto& r = block_structure.rows[row_block_id];
- const int row_block_size = r.block.size;
- const int num_cells = r.cells.size();
-
-
-
- if (num_cells > 1 && row_block_size > 1) {
- is_crs_compatible_ = false;
- }
- row_blocks.emplace_back(r.block);
- first_cell_in_row_block.push_back(cells.size());
- if (r.cells.empty()) {
- value_offset_row_block_f.push_back(-1);
- } else {
- for (auto it = value_offset_row_block_f.rbegin();
- it != value_offset_row_block_f.rend();
- --it) {
- if (*it != -1) break;
- *it = cells[0].position;
- }
- value_offset_row_block_f.push_back(r.cells[0].position);
- }
- for (const auto& c : r.cells) {
- const int col_block_size = col_blocks[c.block_id].size;
- const int cell_size = col_block_size * row_block_size;
- cells.push_back(c);
- DCHECK(c.block_id >= num_col_blocks_e);
- if (f_values_offset == -1) {
- num_nonzeros_ = c.position;
- f_values_offset = c.position;
- }
- sequential_layout &= c.position == num_nonzeros_;
- num_nonzeros_ += cell_size;
- }
- }
- if (f_values_offset == -1) {
- f_values_offset = num_nonzeros_e_;
- num_nonzeros_ = num_nonzeros_e_;
- }
-
- for (auto it = value_offset_row_block_f.rbegin();
- it != value_offset_row_block_f.rend();
- ++it) {
- if (*it != -1) break;
- *it = num_nonzeros_;
- }
- value_offset_row_block_f.push_back(num_nonzeros_);
- CHECK_EQ(num_nonzeros_e_, f_values_offset);
- first_cell_in_row_block.push_back(cells.size());
- num_cells_ = cells.size();
- num_rows_ = Dimension(row_blocks);
- num_cols_ = Dimension(col_blocks);
- CHECK(sequential_layout);
- if (VLOG_IS_ON(3)) {
- const size_t first_cell_in_row_block_size =
- first_cell_in_row_block.size() * sizeof(int);
- const size_t cells_size = cells.size() * sizeof(Cell);
- const size_t row_blocks_size = row_blocks.size() * sizeof(Block);
- const size_t col_blocks_size = col_blocks.size() * sizeof(Block);
- const size_t total_size = first_cell_in_row_block_size + cells_size +
- col_blocks_size + row_blocks_size;
- const double ratio =
- (100. * total_size) / (num_nonzeros_ * (sizeof(int) + sizeof(double)) +
- num_rows_ * sizeof(int));
- VLOG(3) << "\nCudaBlockSparseStructure:\n"
- "\tRow block offsets: "
- << first_cell_in_row_block_size
- << " bytes\n"
- "\tColumn blocks: "
- << col_blocks_size
- << " bytes\n"
- "\tRow blocks: "
- << row_blocks_size
- << " bytes\n"
- "\tCells: "
- << cells_size << " bytes\n\tTotal: " << total_size
- << " bytes of GPU memory (" << ratio << "% of CRS matrix size)";
- }
- first_cell_in_row_block_.CopyFromCpuVector(first_cell_in_row_block);
- cells_.CopyFromCpuVector(cells);
- row_blocks_.CopyFromCpuVector(row_blocks);
- col_blocks_.CopyFromCpuVector(col_blocks);
- if (num_col_blocks_e || num_row_blocks_e_) {
- value_offset_row_block_f_.CopyFromCpuVector(value_offset_row_block_f);
- }
- }
- }
- #endif
|