[FEATURE] Track occupancy in IBF

include/hibf/interleaved_bloom_filter.hpp

@@ -23,6 +23,7 @@
 #include <cereal/cereal.hpp>           // for make_nvp
 #include <cereal/macros.hpp>           // for CEREAL_SERIALIZE_FUNCTION_NAME
 #include <cereal/types/base_class.hpp> // for base_class
+#include <cereal/types/vector.hpp>     // for vector
 
 #include <hibf/cereal/concepts.hpp>           // for cereal_archive
 #include <hibf/contrib/aligned_allocator.hpp> // for aligned_allocator
@@ -192,10 +193,6 @@ class interleaved_bloom_filter : private seqan::hibf::bit_vector
         return h;
     }
 
-    //!\brief Helper function to reduce code-duplication between emplace and emplace_exists.
-    template <bool check_exists>
-    inline auto emplace_impl(size_t const value, bin_index const bin) noexcept;
-
 public:
     class membership_agent_type; // documented upon definition below
     template <std::integral value_t>
@@ -215,16 +212,18 @@ class interleaved_bloom_filter : private seqan::hibf::bit_vector
      * \param bins_ The number of bins.
      * \param size The bitvector size.
      * \param funs The number of hash functions. Default 2. At least 1, at most 5.
+     * \param track_occupancy_ Whether to track the occupancy of the bins.
      *
      * \details
      *
      * ### Example
      *
      * \include test/snippet/ibf/interleaved_bloom_filter_constructor.cpp
      */
-    interleaved_bloom_filter(seqan::hibf::bin_count bins_,
-                             seqan::hibf::bin_size size,
-                             seqan::hibf::hash_function_count funs = seqan::hibf::hash_function_count{2u});
+    interleaved_bloom_filter(seqan::hibf::bin_count const bins_,
+                             seqan::hibf::bin_size const size,
+                             seqan::hibf::hash_function_count const funs = seqan::hibf::hash_function_count{2u},
+                             bool const track_occupancy_ = false);
 
     /*!\brief Construct an Interleaved Bloom Filter.
      * \param configuration The seqan::hibf::config.
@@ -249,20 +248,14 @@ class interleaved_bloom_filter : private seqan::hibf::bit_vector
      *
      * \details
      *
+     * If `track_occupancy` is set to `true`, the occupancy of the bin is tracked.
+     *
      * ### Example
      *
      * \include test/snippet/ibf/interleaved_bloom_filter_emplace.cpp
      */
     void emplace(size_t const value, bin_index const bin) noexcept;
 
-    /*!\brief Inserts a value into a specific bin and returns whether the value already existed.
-     * \param[in] value The raw numeric value to process.
-     * \param[in] bin The bin index to insert into.
-     * \returns `true` if the value already existed, `false` otherwise.
-     * \sa seqan::hibf::interleaved_bloom_filter::emplace
-    */
-    [[nodiscard]] bool emplace_exists(size_t const value, bin_index const bin) noexcept;
-
     /*!\brief Clears a specific bin.
      * \param[in] bin The bin index to clear.
      *
@@ -293,7 +286,7 @@ class interleaved_bloom_filter : private seqan::hibf::bit_vector
                       "The reference type of the range to clear must be seqan::hibf::bin_index.");
 #ifndef NDEBUG
         for (auto && bin : bin_range)
-            assert(bin.value < bins);
+            assert(bin.value < technical_bins);
 #endif // NDEBUG
 
         for (size_t offset = 0, i = 0; i < bin_size_; offset += technical_bins, ++i)
@@ -438,6 +431,17 @@ class interleaved_bloom_filter : private seqan::hibf::bit_vector
     using base_t::data;
     //!\}
 
+    /*!\brief Contains the number of unique values inserted into each bin.
+     * \details
+     * Only contains non-zero values if `track_occupancy` is true.
+     *
+     * A value is unique if inserting it into the IBF would set at least one previously unset bit.
+     */
+    std::vector<size_t> occupancy{};
+
+    //!\brief Whether to track the occupancy of the bins.
+    bool track_occupancy{false};
+
     /*!\cond DEV
      * \brief The version of the HIBF.
      */
@@ -461,6 +465,8 @@ class interleaved_bloom_filter : private seqan::hibf::bit_vector
         archive(bin_words);
         archive(hash_funs);
         archive(cereal::base_class<base_t>(this));
+        archive(occupancy);
+        archive(track_occupancy);
     }
     //!\endcond
 };

src/build/construct_ibf.cpp

@@ -51,7 +51,9 @@ seqan::hibf::interleaved_bloom_filter construct_ibf(robin_hood::unordered_flat_s
     local_index_allocation_timer.start();
     seqan::hibf::interleaved_bloom_filter ibf{bin_count,
                                               bin_size,
-                                              seqan::hibf::hash_function_count{data.config.number_of_hash_functions}};
+                                              seqan::hibf::hash_function_count{data.config.number_of_hash_functions},
+                                              data.config.empty_bin_fraction > 0.0};
+
     local_index_allocation_timer.stop();
     data.index_allocation_timer += local_index_allocation_timer;
 

src/build/insert_into_ibf.cpp

@@ -34,14 +34,34 @@ void insert_into_ibf(robin_hood::unordered_flat_set<uint64_t> const & kmers,
 
     serial_timer local_fill_ibf_timer{};
     local_fill_ibf_timer.start();
-    for (auto chunk : kmers | seqan::stl::views::chunk(chunk_size))
+    auto chunk_view = seqan::stl::views::chunk(kmers, chunk_size);
+    for (auto && chunk : chunk_view)
     {
         assert(chunk_number < number_of_bins);
         seqan::hibf::bin_index const bin_idx{bin_index + chunk_number};
         ++chunk_number;
-        for (size_t const value : chunk)
+        for (auto && value : chunk)
             ibf.emplace(value, bin_idx);
     }
+
+    assert(chunk_view.size() <= number_of_bins);
+    // Edge case: If there are not enough k-mers to emplace at least one value into each bin, set the occupancy of
+    // the left over bins to 1.
+    // GCOVR_EXCL_START
+    if (ibf.track_occupancy && chunk_view.size() < number_of_bins)
+    {
+        size_t const diff = number_of_bins - chunk_view.size();
+        auto it = ibf.occupancy.begin() + bin_index + chunk_view.size();
+        assert(std::ranges::all_of(it,
+                                   it + diff,
+                                   [](size_t value)
+                                   {
+                                       return value == 0u;
+                                   }));
+        std::ranges::fill_n(it, diff, 1u);
+    }
+    // GCOVR_EXCL_STOP
+
     local_fill_ibf_timer.stop();
     fill_ibf_timer += local_fill_ibf_timer;
 }

src/interleaved_bloom_filter.cpp

@@ -28,14 +28,15 @@ namespace seqan::hibf
 #    pragma GCC diagnostic ignored "-Wattributes"
 #endif // HIBF_COMPILER_IS_GCC
 
-interleaved_bloom_filter::interleaved_bloom_filter(seqan::hibf::bin_count bins_,
-                                                   seqan::hibf::bin_size size,
-                                                   seqan::hibf::hash_function_count funs)
+interleaved_bloom_filter::interleaved_bloom_filter(seqan::hibf::bin_count const bins_,
+                                                   seqan::hibf::bin_size const size,
+                                                   seqan::hibf::hash_function_count const funs,
+                                                   bool const track_occupancy_) :
+    bins{bins_.value},
+    bin_size_{size.value},
+    hash_funs{funs.value},
+    track_occupancy{track_occupancy_}
 {
-    bins = bins_.value;
-    bin_size_ = size.value;
-    hash_funs = funs.value;
-
     if (bins == 0)
         throw std::logic_error{"The number of bins must be > 0."};
     if (hash_funs == 0 || hash_funs > 5)
@@ -47,6 +48,7 @@ interleaved_bloom_filter::interleaved_bloom_filter(seqan::hibf::bin_count bins_,
     bin_words = divide_and_ceil(bins, 64u);
     technical_bins = bin_words * 64u;
     resize(technical_bins * bin_size_);
+    occupancy.resize(technical_bins, 0u);
 }
 
 size_t find_biggest_bin(config const & configuration)
@@ -101,7 +103,8 @@ size_t max_bin_size(config & configuration, size_t const max_bin_elements)
 interleaved_bloom_filter::interleaved_bloom_filter(config & configuration, size_t const max_bin_elements) :
     interleaved_bloom_filter{seqan::hibf::bin_count{configuration.number_of_user_bins},
                              seqan::hibf::bin_size{max_bin_size(configuration, max_bin_elements)},
-                             seqan::hibf::hash_function_count{configuration.number_of_hash_functions}}
+                             seqan::hibf::hash_function_count{configuration.number_of_hash_functions},
+                             configuration.empty_bin_fraction > 0.0}
 {
     size_t const chunk_size = std::clamp<size_t>(std::bit_ceil(bin_count() / configuration.threads), 8u, 64u);
 
@@ -112,39 +115,27 @@ interleaved_bloom_filter::interleaved_bloom_filter(config & configuration, size_
     }
 }
 
-template <bool check_exists>
-inline auto interleaved_bloom_filter::emplace_impl(size_t const value, bin_index const bin) noexcept
+[[gnu::always_inline]] void interleaved_bloom_filter::emplace(size_t const value, bin_index const bin) noexcept
 {
     assert(bin.value < bins);
 
-    [[maybe_unused]] bool exists{true};
+    bool exists{track_occupancy};
 
     for (size_t i = 0; i < hash_funs; ++i)
     {
-        size_t idx = hash_and_fit(value, hash_seeds[i]);
-        idx += bin.value;
+        size_t const idx = hash_and_fit(value, hash_seeds[i]) + bin.value;
         assert(idx < size());
 
-        // Constructing the reference twice for emplace_exists would impact performance.
+        // Constructing the reference twice for tracking occupancy would impact performance.
         // No difference for emplace.
         seqan::hibf::bit_vector::reference bit_reference{(*this)[idx]};
-        if constexpr (check_exists)
+        if (track_occupancy)
             exists &= bit_reference;
-        bit_reference = 1;
+        bit_reference = true;
     };
 
-    if constexpr (check_exists)
-        return exists;
-};
-
-[[gnu::always_inline]] void interleaved_bloom_filter::emplace(size_t const value, bin_index const bin) noexcept
-{
-    return emplace_impl<false>(value, bin);
-}
-
-[[gnu::always_inline]] bool interleaved_bloom_filter::emplace_exists(size_t const value, bin_index const bin) noexcept
-{
-    return emplace_impl<true>(value, bin);
+    if (track_occupancy && !exists)
+        ++occupancy[bin.value];
 }
 
 void interleaved_bloom_filter::clear(bin_index const bin) noexcept
@@ -205,6 +196,7 @@ void interleaved_bloom_filter::increase_bin_number_to(seqan::hibf::bin_count con
     bins = new_bins;
     bin_words = new_bin_words;
     technical_bins = new_technical_bins;
+    occupancy.resize(technical_bins, 0u);
 }
 
 [[gnu::always_inline]] bit_vector const &

test/performance/ibf/interleaved_bloom_filter_benchmark.cpp

@@ -92,7 +92,7 @@ inline benchmark::Counter elements_per_second(size_t const count)
     return benchmark::Counter(count, benchmark::Counter::kIsIterationInvariantRate, benchmark::Counter::OneK::kIs1000);
 }
 
-template <bool check_exists>
+template <bool track_occupancy>
 inline void emplace_benchmark_impl(::benchmark::State & state)
 {
     auto const & [values, original_ibf] = set_up(state);
@@ -102,23 +102,18 @@ inline void emplace_benchmark_impl(::benchmark::State & state)
 
     seqan::hibf::interleaved_bloom_filter ibf{seqan::hibf::bin_count{original_ibf.bin_count()},
                                               seqan::hibf::bin_size{original_ibf.bin_size()},
-                                              seqan::hibf::hash_function_count{original_ibf.hash_function_count()}};
+                                              seqan::hibf::hash_function_count{original_ibf.hash_function_count()},
+                                              track_occupancy};
 
     for (auto _ : state)
     {
         size_t bin_index = 0u;
-        [[maybe_unused]] size_t result{};
         for (auto && chunk : seqan::stl::views::chunk(values, chunk_size))
         {
             for (auto value : chunk)
-                if constexpr (check_exists)
-                    result += ibf.emplace_exists(value, seqan::hibf::bin_index{bin_index});
-                else
-                    ibf.emplace(value, seqan::hibf::bin_index{bin_index});
+                ibf.emplace(value, seqan::hibf::bin_index{bin_index});
             ++bin_index;
         }
-        if constexpr (check_exists)
-            benchmark::DoNotOptimize(result);
     }
 
     state.counters["elements"] = elements_per_second(number_of_elements);
@@ -129,7 +124,7 @@ void emplace_benchmark(::benchmark::State & state)
     emplace_benchmark_impl<false>(state);
 }
 
-void emplace_exists_benchmark(::benchmark::State & state)
+void emplace_with_occupancy_benchmark(::benchmark::State & state)
 {
     emplace_benchmark_impl<true>(state);
 }
@@ -209,7 +204,7 @@ void bulk_count_benchmark(::benchmark::State & state)
 }
 
 BENCHMARK(emplace_benchmark)->RangeMultiplier(2)->Range(64, 1024);
-BENCHMARK(emplace_exists_benchmark)->RangeMultiplier(2)->Range(64, 1024);
+BENCHMARK(emplace_with_occupancy_benchmark)->RangeMultiplier(2)->Range(64, 1024);
 BENCHMARK(clear_benchmark)->RangeMultiplier(2)->Range(64, 1024);
 BENCHMARK(clear_range_benchmark)->RangeMultiplier(2)->Range(64, 1024);
 BENCHMARK(bulk_contains_benchmark)->RangeMultiplier(2)->Range(64, 1024);

test/unit/hibf/interleaved_bloom_filter_test.cpp

@@ -196,26 +196,45 @@ TEST(ibf_test, emplace)
         auto & res = agent.bulk_contains(hash);
         EXPECT_RANGE_EQ(res, expected);
     }
+
+    EXPECT_EQ(ibf.occupancy.size(), 64u);
+    EXPECT_TRUE(std::ranges::all_of(ibf.occupancy,
+                                    [](size_t const occ)
+                                    {
+                                        return occ == 0u;
+                                    }));
 }
 
-TEST(ibf_test, emplace_exists)
+TEST(ibf_test, emplace_with_occupancy)
 {
     // 1. Construct and emplace
     seqan::hibf::interleaved_bloom_filter ibf{seqan::hibf::bin_count{128u},
                                               seqan::hibf::bin_size{512},
-                                              seqan::hibf::hash_function_count{2u}};
+                                              seqan::hibf::hash_function_count{2u},
+                                              true};
 
     for (size_t bin_idx : std::views::iota(0, 64))
         for (size_t hash : std::views::iota(0, 64))
             ibf.emplace(hash, seqan::hibf::bin_index{bin_idx});
 
     // 2. Test for correctness
+
+    auto agent = ibf.membership_agent();
+    std::vector<bool> expected(128);
+    std::fill(expected.begin(), expected.begin() + 64u, true);
+    for (size_t hash : std::views::iota(0, 64))
+    {
+        auto & res = agent.bulk_contains(hash);
+        EXPECT_RANGE_EQ(res, expected);
+    }
+
+    ASSERT_EQ(ibf.occupancy.size(), 128u);
+
     for (size_t bin_idx : std::views::iota(0, 64))
-        for (size_t hash : std::views::iota(0, 64))
-            ASSERT_TRUE(ibf.emplace_exists(hash, seqan::hibf::bin_index{bin_idx}));
+        EXPECT_NE(ibf.occupancy[bin_idx], 0u);
 
     for (size_t bin_idx : std::views::iota(64, 128))
-        ASSERT_FALSE(ibf.emplace_exists(0u, seqan::hibf::bin_index{bin_idx}));
+        EXPECT_EQ(ibf.occupancy[bin_idx], 0u);
 }
 
 TEST(ibf_test, clear)