[MISC] C++23 md subscript

Author: eseiler

include/hibf/hierarchical_interleaved_bloom_filter.hpp

@@ -21,8 +21,9 @@
 #include <hibf/interleaved_bloom_filter.hpp> // for interleaved_bloom_filter
 #include <hibf/layout/layout.hpp>            // for layout
 #include <hibf/misc/counting_vector.hpp>     // for counting_vector
-#include <hibf/misc/timer.hpp>               // for concurrent_timer
-#include <hibf/platform.hpp>                 // for HIBF_CONSTEXPR_VECTOR
+#include <hibf/misc/md_vector.hpp>
+#include <hibf/misc/timer.hpp> // for concurrent_timer
+#include <hibf/platform.hpp>   // for HIBF_CONSTEXPR_VECTOR
 
 namespace seqan::hibf
 {
@@ -202,7 +203,7 @@ class hierarchical_interleaved_bloom_filter
      * If `j != i` is returned, there is a lower level IBF, bin `b` is a merged bin, and `j` is the ID of the lower
      * level IBF in ibf_vector.
      */
-    std::vector<std::vector<int64_t>> next_ibf_id;
+    md_vector<int64_t> next_ibf_id;
 
     /*!\brief Stores for each bin in each IBF of the HIBF the user bin ID.
     * \details
@@ -211,7 +212,7 @@ class hierarchical_interleaved_bloom_filter
     * lower level IBF.
     * Otherwise, the returned value `j` is the corresponding user bin ID.
     */
-    std::vector<std::vector<int64_t>> ibf_bin_to_user_bin_id{};
+    md_vector<int64_t> ibf_bin_to_user_bin_id{};
 
     //!\brief Returns a membership_agent to be used for counting.
     membership_agent_type membership_agent() const;
@@ -280,16 +281,16 @@ class hierarchical_interleaved_bloom_filter::membership_agent_type
         {
             sum += result[bin];
 
-            auto const current_filename_index = hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx][bin];
+            auto const current_filename_index = hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx, bin];
 
             if (current_filename_index < 0) // merged bin
             {
                 if (sum >= threshold)
-                    membership_for_impl(values, hibf_ptr->next_ibf_id[ibf_idx][bin], threshold);
+                    membership_for_impl(values, hibf_ptr->next_ibf_id[ibf_idx, bin], threshold);
                 sum = 0u;
             }
             else if (bin + 1u == result.size() ||                                                  // last bin
-                     current_filename_index != hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx][bin + 1]) // end of split bin
+                     current_filename_index != hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx, bin + 1]) // end of split bin
             {
                 if (sum >= threshold)
                     result_buffer.emplace_back(current_filename_index);
@@ -415,16 +416,16 @@ class hierarchical_interleaved_bloom_filter::counting_agent_type
         for (size_t bin{}; bin < result.size(); ++bin)
         {
             sum += result[bin];
-            auto const current_filename_index = hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx][bin];
+            auto const current_filename_index = hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx, bin];
 
             if (current_filename_index < 0) // merged bin
             {
                 if (sum >= threshold)
-                    bulk_count_impl(values, hibf_ptr->next_ibf_id[ibf_idx][bin], threshold);
+                    bulk_count_impl(values, hibf_ptr->next_ibf_id[ibf_idx, bin], threshold);
                 sum = 0u;
             }
             else if (bin + 1u == result.size() ||                                                  // last bin
-                     current_filename_index != hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx][bin + 1]) // end of split bin
+                     current_filename_index != hibf_ptr->ibf_bin_to_user_bin_id[ibf_idx, bin + 1]) // end of split bin
             {
                 if (sum >= threshold)
                     result_buffer[current_filename_index] = sum;

include/hibf/layout/hierarchical_binning.hpp

@@ -13,7 +13,8 @@
 #include <hibf/build/bin_size_in_bits.hpp> // for bin_size_in_bits
 #include <hibf/config.hpp>                 // for config
 #include <hibf/layout/data_store.hpp>      // for data_store
-#include <hibf/platform.hpp>               // for HIBF_WORKAROUND_GCC_BOGUS_MEMCPY
+#include <hibf/misc/md_vector.hpp>
+#include <hibf/platform.hpp> // for HIBF_WORKAROUND_GCC_BOGUS_MEMCPY
 
 namespace seqan::hibf::layout
 {
@@ -140,9 +141,9 @@ class hierarchical_binning
      *
      * \image html hierarchical_dp_init.png
      */
-    void initialization(std::vector<std::vector<size_t>> & matrix,
-                        std::vector<std::vector<size_t>> & ll_matrix,
-                        std::vector<std::vector<std::pair<size_t, size_t>>> & trace);
+    void initialization(md_vector<size_t> & matrix,
+                        md_vector<size_t> & ll_matrix,
+                        md_vector<std::pair<size_t, size_t>> & trace);
 
     /*!\brief Performs the recursion.
      *
@@ -182,9 +183,8 @@ class hierarchical_binning
      * this algorithm. It would be too computational intensive to compute the splitting for every possibility.
      *
      */
-    void recursion(std::vector<std::vector<size_t>> & matrix,
-                   std::vector<std::vector<size_t>> & ll_matrix,
-                   std::vector<std::vector<std::pair<size_t, size_t>>> & trace);
+    void
+    recursion(md_vector<size_t> & matrix, md_vector<size_t> & ll_matrix, md_vector<std::pair<size_t, size_t>> & trace);
 
     void backtrack_merged_bin(size_t trace_j,
                               size_t const next_j,
@@ -198,7 +198,7 @@ class hierarchical_binning
                              maximum_bin_tracker & max_tracker);
 
     //!\brief Backtracks the trace matrix and writes the resulting binning into the output file.
-    size_t backtracking(std::vector<std::vector<std::pair<size_t, size_t>>> const & trace);
+    size_t backtracking(md_vector<std::pair<size_t, size_t>> const & trace);
 
     data_store initialise_libf_data(size_t const trace_j) const;
 

include/hibf/misc/md_vector.hpp

@@ -0,0 +1,42 @@
+// SPDX-FileCopyrightText: 2006-2024, Knut Reinert & Freie Universität Berlin
+// SPDX-FileCopyrightText: 2016-2024, Knut Reinert & MPI für molekulare Genetik
+// SPDX-License-Identifier: BSD-3-Clause
+
+/*!\file
+ * \brief Provides seqan::hibf::md_vector.
+ * \author Enrico Seiler <enrico.seiler AT fu-berlin.de>
+ */
+
+#pragma once
+
+#include <vector> // for vector
+
+#include <hibf/platform.hpp>
+
+namespace seqan::hibf
+{
+
+template <typename value_t>
+struct md_vector : public std::vector<std::vector<value_t>>
+{
+    using base_t = std::vector<std::vector<value_t>>;
+    using base_t::base_t;
+    using base_t::operator[];
+#if defined(__cpp_explicit_this_parameter) && __cpp_explicit_this_parameter >= 202110L
+    decltype(auto) operator[](this auto & self, size_t const x, size_t const y)
+    {
+        return self[x][y];
+    }
+#else
+    value_t & operator[](size_t const x, size_t const y)
+    {
+        return (*this)[x][y];
+    }
+    value_t const & operator[](size_t const x, size_t const y) const
+    {
+        return (*this)[x][y];
+    }
+#endif
+};
+
+} // namespace seqan::hibf

include/hibf/sketch/minhashes.hpp

@@ -17,6 +17,7 @@
 #include <cereal/access.hpp> // for access
 #include <cereal/cereal.hpp> // for make_nvp, CEREAL_NVP
 
+#include <hibf/misc/md_vector.hpp>
 #include <hibf/platform.hpp>
 
 namespace seqan::hibf::sketch
@@ -39,7 +40,7 @@ struct minhashes
     static constexpr size_t sketch_size{40};
 
     //!\brief A table of sketches. For LSH we need multiple sketches, stored in a table.
-    std::vector<std::vector<uint64_t>> table{}; // Each element (vector<uint64_t>) is a minhash.
+    md_vector<uint64_t> table{}; // Each element (vector<uint64_t>) is a minhash.
 
     /*!\name Constructors, destructor and assignment
      * \{

src/layout/hierarchical_binning.cpp

@@ -54,13 +54,13 @@ size_t hierarchical_binning::execute()
     }
 
     // technical bins (outer) = rows; user bins (inner) = columns
-    std::vector<std::vector<size_t>> matrix(num_technical_bins, std::vector<size_t>(num_user_bins, max_size_t));
+    md_vector<size_t> matrix(num_technical_bins, std::vector<size_t>(num_user_bins, max_size_t));
 
     // technical bins (outer) = rows; user bins (inner) = columns
-    std::vector<std::vector<size_t>> ll_matrix(num_technical_bins, std::vector<size_t>(num_user_bins, 0u));
+    md_vector<size_t> ll_matrix(num_technical_bins, std::vector<size_t>(num_user_bins, 0u));
 
     // technical bins (outer) = rows; user bins (inner) = columns
-    std::vector<std::vector<std::pair<size_t, size_t>>> trace(
+    md_vector<std::pair<size_t, size_t>> trace(
         num_technical_bins,
         std::vector<std::pair<size_t, size_t>>(num_user_bins, {max_size_t, max_size_t}));
 
@@ -87,9 +87,9 @@ size_t hierarchical_binning::execute()
     return static_cast<size_t>(std::ceil(levels));
 }
 
-void hierarchical_binning::initialization(std::vector<std::vector<size_t>> & matrix,
-                                          std::vector<std::vector<size_t>> & ll_matrix,
-                                          std::vector<std::vector<std::pair<size_t, size_t>>> & trace)
+void hierarchical_binning::initialization(md_vector<size_t> & matrix,
+                                          md_vector<size_t> & ll_matrix,
+                                          md_vector<std::pair<size_t, size_t>> & trace)
 {
     assert(data != nullptr);
 
@@ -99,8 +99,8 @@ void hierarchical_binning::initialization(std::vector<std::vector<size_t>> & mat
     for (size_t i = 0; i < num_technical_bins; ++i)
     {
         size_t const corrected_ub_cardinality = static_cast<size_t>(ub_cardinality * data->fpr_correction[i + 1]);
-        matrix[i][0] = divide_and_ceil(corrected_ub_cardinality, i + 1u);
-        trace[i][0] = {0u, 0u}; // unnecessary?
+        matrix[i, 0] = divide_and_ceil(corrected_ub_cardinality, i + 1u);
+        trace[i, 0] = {0u, 0u}; // unnecessary?
     }
 
     // initialize first row
@@ -118,9 +118,9 @@ void hierarchical_binning::initialization(std::vector<std::vector<size_t>> & mat
         for (size_t j = 1; j < num_user_bins; ++j)
         {
             sum += (*data->kmer_counts)[data->positions[j]];
-            matrix[0][j] = data->union_estimates[j];
-            ll_matrix[0][j] = max_merge_levels(j + 1) * sum;
-            trace[0][j] = {0u, j - 1}; // unnecessary?
+            matrix[0, j] = data->union_estimates[j];
+            ll_matrix[0, j] = max_merge_levels(j + 1) * sum;
+            trace[0, j] = {0u, j - 1}; // unnecessary?
         }
     }
     else
@@ -130,16 +130,16 @@ void hierarchical_binning::initialization(std::vector<std::vector<size_t>> & mat
             assert(j < data->positions.size());
             assert(data->positions[j] < data->kmer_counts->size());
             sum += (*data->kmer_counts)[data->positions[j]];
-            matrix[0][j] = sum;
-            ll_matrix[0][j] = max_merge_levels(j + 1) * sum;
-            trace[0][j] = {0u, j - 1}; // unnecessary?
+            matrix[0, j] = sum;
+            ll_matrix[0, j] = max_merge_levels(j + 1) * sum;
+            trace[0, j] = {0u, j - 1}; // unnecessary?
         }
     }
 }
 
-void hierarchical_binning::recursion(std::vector<std::vector<size_t>> & matrix,
-                                     std::vector<std::vector<size_t>> & ll_matrix,
-                                     std::vector<std::vector<std::pair<size_t, size_t>>> & trace)
+void hierarchical_binning::recursion(md_vector<size_t> & matrix,
+                                     md_vector<size_t> & ll_matrix,
+                                     md_vector<std::pair<size_t, size_t>> & trace)
 {
     assert(data != nullptr);
 
@@ -182,23 +182,23 @@ void hierarchical_binning::recursion(std::vector<std::vector<size_t>> & matrix,
                 size_t const corrected_ub_cardinality =
                     static_cast<size_t>(ub_cardinality * data->fpr_correction[(i - i_prime)]);
                 size_t score =
-                    std::max<size_t>(divide_and_ceil(corrected_ub_cardinality, i - i_prime), matrix[i_prime][j - 1]);
-                size_t full_score = score * (i + 1) /*#TBs*/ + config.alpha * ll_matrix[i_prime][j - 1];
+                    std::max<size_t>(divide_and_ceil(corrected_ub_cardinality, i - i_prime), matrix[i_prime, j - 1]);
+                size_t full_score = score * (i + 1) /*#TBs*/ + config.alpha * ll_matrix[i_prime, j - 1];
 
                 // std::cout << " ++ j:" << j << " i:" << i << " i':" << i_prime << " score:" << score << std::endl;
 
                 if (full_score < full_minimum)
                 {
                     minimum = score;
                     full_minimum = full_score;
-                    trace[i][j] = {i_prime, j - 1};
-                    ll_matrix[i][j] = ll_matrix[i_prime][j - 1];
+                    trace[i, j] = {i_prime, j - 1};
+                    ll_matrix[i, j] = ll_matrix[i_prime, j - 1];
                 }
             }
 
             // seqan3::debug_stream << "current vertical minimum of " << "j:" << j << " i:" << i
             //                      << " -> score:" << full_minimum << " (M_ij=" << minimum << ")"
-            //                      << " trace:" << trace[i][j]
+            //                      << " trace:" << trace[i, j]
             //                      << std::endl;
 
             // check horizontal cells
@@ -216,16 +216,16 @@ void hierarchical_binning::recursion(std::vector<std::vector<size_t>> & matrix,
 
             // if the user bin j-1 was not split into multiple technical bins!
             // I may merge the current user bin j into the former
-            while (j_prime != 0 && ((i - trace[i][j_prime].first) < 2) && get_weight() < minimum)
+            while (j_prime != 0 && ((i - trace[i, j_prime].first) < 2) && get_weight() < minimum)
             {
                 weight += (*data->kmer_counts)[data->positions[j_prime]];
                 --j_prime;
 
                 // score: The current maximum technical bin size for the high-level IBF (score for the matrix M)
                 // ll_kmers: estimate for the number of k-mers that have to be resolved on lower levels
                 // full_score: The score to minimize -> score * #TB-high_level + low_level_memory footprint
-                size_t const score = std::max<size_t>(matrix[i - 1][j_prime], get_weight());
-                size_t const ll_kmers = ll_matrix[i - 1][j_prime] + max_merge_levels(j - j_prime) * weight;
+                size_t const score = std::max<size_t>(matrix[i - 1, j_prime], get_weight());
+                size_t const ll_kmers = ll_matrix[i - 1, j_prime] + max_merge_levels(j - j_prime) * weight;
                 size_t const full_score = score * (i + 1) /*#TBs*/ + config.alpha * ll_kmers;
 
                 // seqan3::debug_stream << " -- " << "j_prime:" << j_prime
@@ -236,12 +236,12 @@ void hierarchical_binning::recursion(std::vector<std::vector<size_t>> & matrix,
                 {
                     minimum = score;
                     full_minimum = full_score;
-                    trace[i][j] = {i - 1, j_prime};
-                    ll_matrix[i][j] = ll_kmers;
+                    trace[i, j] = {i - 1, j_prime};
+                    ll_matrix[i, j] = ll_kmers;
                 }
             }
 
-            matrix[i][j] = minimum;
+            matrix[i, j] = minimum;
         }
     }
 }
@@ -307,7 +307,7 @@ void hierarchical_binning::backtrack_split_bin(size_t trace_j,
     // std::cout << "split " << trace_j << " into " << number_of_bins << ": " << cardinality_per_bin << std::endl;
 }
 
-size_t hierarchical_binning::backtracking(std::vector<std::vector<std::pair<size_t, size_t>>> const & trace)
+size_t hierarchical_binning::backtracking(md_vector<std::pair<size_t, size_t>> const & trace)
 {
     assert(data != nullptr);
 
@@ -323,8 +323,8 @@ size_t hierarchical_binning::backtracking(std::vector<std::vector<std::pair<size
     while (trace_j > 0u && trace_i > 0u)
     {
         // std::cout << "\t I am now at " << trace_i << "," << trace_j << std::endl;
-        size_t next_i = trace[trace_i][trace_j].first;
-        size_t next_j = trace[trace_i][trace_j].second;
+        size_t next_i = trace[trace_i, trace_j].first;
+        size_t next_j = trace[trace_i, trace_j].second;
 
         size_t number_of_bins = (trace_i - next_i);
 
@@ -339,7 +339,7 @@ size_t hierarchical_binning::backtracking(std::vector<std::vector<std::pair<size
         {
             backtrack_split_bin(trace_j, number_of_bins, bin_id, max_tracker);
 
-            trace_i = trace[trace_i][trace_j].first;
+            trace_i = trace[trace_i, trace_j].first;
             --trace_j;
         }