support basic searcher #351

ShawnShawnYou · 2025-01-20T09:08:27Z

LHT129 · 2025-01-20T09:16:38Z

src/data_cell/flatten_interface.h

 public:
    InnerIdType total_count_{0};
    InnerIdType max_capacity_{1000000};
    uint32_t code_size_{0};
+    uint32_t prefetch_neighbor_codes_num_{1};


why flatten datacell need neighbor

used for batch query

LHT129 · 2025-01-20T09:46:45Z

src/allocator_wrapper.h

+template <typename T, typename U>
+bool
+operator==(const AllocatorWrapper<T>&, const AllocatorWrapper<U>&) noexcept {
+    return true;


always true？

LHT129 · 2025-01-20T09:55:10Z

src/data_cell/adapter_graph_datacell.h

+private:
+    std::shared_ptr<hnswlib::HierarchicalNSW> alg_hnsw_;
+};
+}  // namespace vsag


only used on test?

LHT129 · 2025-01-20T09:56:27Z

src/data_cell/flatten_datacell.h

@@ -70,6 +70,11 @@ class FlattenDataCell : public FlattenInterface {
        io_->Prefetch(id * code_size_);
    };

+    bool


not used here

LHT129 · 2025-01-20T09:58:57Z

src/data_cell/flatten_interface.h

+    virtual void
+    SetRuntimeParameters(const UnorderedMap<std::string, ParamValue>& new_params) {
+        // TODO(ZXY): implement
+        return;


LHT129 · 2025-01-20T09:59:27Z

src/impl/basic_searcher.cpp

+
+template <typename GraphTmpl, typename VectorDataTmpl>
+BasicSearcher<GraphTmpl, VectorDataTmpl>::BasicSearcher(std::shared_ptr<GraphTmpl> graph,
+                                                        std::shared_ptr<VectorDataTmpl> vector,


use interface is better

LHT129 · 2025-01-20T10:00:02Z

src/impl/basic_searcher.cpp

+
+#ifdef USE_SSE
+    for (uint32_t i = 0; i < prefetch_neighbor_visit_num_; i++) {
+        _mm_prefetch(vl->mass + neighbors[i], _MM_HINT_T0);


use Prefetch

LHT129 · 2025-01-20T10:00:12Z

src/impl/basic_searcher.cpp

+    for (uint32_t i = 0; i < neighbors.size(); i++) {
+#ifdef USE_SSE
+        if (i + prefetch_neighbor_visit_num_ < neighbors.size()) {
+            _mm_prefetch(vl->mass + neighbors[i + prefetch_neighbor_visit_num_], _MM_HINT_T0);


LHT129 · 2025-01-20T10:00:54Z

src/impl/basic_searcher.cpp

+        if (vl->mass[neighbors[i]] != vl->curV) {
+            to_be_visited_rid[count_no_visited] = i;
+            to_be_visited_id[count_no_visited] = neighbors[i];
+            count_no_visited++;


sometimes we don't need to_be_visited

for graph searcher, it's required

inabao · 2025-01-20T14:18:56Z

src/impl/basic_searcher.cpp

+
+BasicSearcher::BasicSearcher(GraphInterfacePtr graph,
+                             FlattenInterfacePtr vector,
+                             const IndexCommonParam& common_param) {


const GraphInterfacePtr& graph
const FlattenInterfacePtr& vector

modified to graph & flat interface

inabao · 2025-01-20T14:23:19Z

src/impl/basic_searcher.h

+#include <shared_mutex>
+
+#include "../utils.h"
+#include "ThreadPool.h"


remove unused header files

inabao · 2025-01-20T14:24:30Z

src/impl/basic_searcher.h

+                  const IndexCommonParam& common_param);
+
+    virtual MaxHeap
+    Search(const float* query, InnerSearchParam& inner_search_param) const;


Can we search different graph using one Searcher

modified, use SetGraphAndVector to allow search with different graph and vector

inabao · 2025-01-20T14:27:13Z

src/impl/runtime_parameter.h

+#include <string>
+#include <thread>
+#include <variant>
+#include <vector>


inabao · 2025-01-20T14:27:30Z

src/impl/basic_searcher_test.cpp

+#include "fixtures.h"
+#include "io/memory_io.h"
+#include "quantization/fp32_quantizer.h"
+#include "safe_allocator.h"


inabao · 2025-01-20T14:28:00Z

src/impl/basic_searcher_test.cpp

+        REQUIRE(result.top().second == valid_result.top().second);
+        REQUIRE(result.top().second == ids[i]);
+    }
+}


add new line

inabao · 2025-01-20T14:28:07Z

src/impl/runtime_parameter.h

+    int cur_{0};
+    bool is_end_{false};
+};
+}  // namespace vsag


add new line

inabao · 2025-01-20T14:28:22Z

src/impl/basic_searcher.h

+    uint32_t prefetch_neighbor_visit_num_{1};
+};
+
+}  // namespace vsag


add new line

inabao · 2025-01-20T14:29:04Z

src/impl/basic_searcher.cpp

+}
+
+MaxHeap
+BasicSearcher::Search(const float* query, InnerSearchParam& inner_search_param) const {


const InnerSearchParam& inner_search_param

inabao · 2025-01-20T14:37:51Z

src/data_cell/flatten_datacell.h

@@ -226,7 +226,18 @@ FlattenDataCell<QuantTmpl, IOTmpl>::query(float* result_dists,
                                          const std::shared_ptr<Computer<QuantTmpl>>& computer,
                                          const InnerIdType* idx,
                                          InnerIdType id_count) {
+    for (uint32_t i = 0; i < this->prefetch_neighbor_codes_num_ and i < id_count; i++) {


prefetch_neighbor_codes_num_ is always 1?

it can be further optimized by optimizer

codecov · 2025-01-20T16:35:32Z

Codecov Report

Attention: Patch coverage is 98.87640% with 1 line in your changes missing coverage. Please review.

@@            Coverage Diff             @@
##             main     #351      +/-   ##
==========================================
- Coverage   90.28%   88.02%   -2.27%     
==========================================
  Files         118      135      +17     
  Lines        7372     8693    +1321     
==========================================
+ Hits         6656     7652     +996     
- Misses        716     1041     +325

Flag	Coverage Δ
cpp	`88.02% <98.87%> (-2.27%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
common	`88.77% <ø> (+2.12%)`	⬆️
datacell	`91.31% <100.00%> (+1.13%)`	⬆️
index	`91.14% <100.00%> (+0.42%)`	⬆️
simd	`69.28% <ø> (-30.72%)`	⬇️

Continue to review full report in Codecov by Sentry.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 04283a8...cb903ef. Read the comment docs.

LHT129 · 2025-01-21T03:08:12Z

src/impl/basic_searcher.cpp

+
+void
+BasicSearcher::SetGraphAndVector(GraphInterfacePtr graph_data_cell,
+                                 FlattenInterfacePtr vector_data_cell) {


split two function

LHT129 · 2025-01-21T03:10:56Z

src/impl/basic_searcher.cpp

+                     std::pair<float, uint64_t>& current_node_pair,
+                     Vector<InnerIdType>& to_be_visited_rid,
+                     Vector<InnerIdType>& to_be_visited_id) const {
+    // to_be_visited_rid is used in redundant storage


redundant not used now

added comment for distinguish to_be_visited_rid and to_be_visited_id in basic_search.h

LHT129 · 2025-01-21T03:11:19Z

src/impl/basic_searcher.cpp

+}
+
+void
+BasicSearcher::SetGraphAndVector(GraphInterfacePtr graph_data_cell,


use interface name

LHT129 · 2025-01-21T03:12:18Z

src/impl/basic_searcher.cpp

+    uint32_t hops = 0;
+    uint32_t dist_cmp = 0;
+    uint32_t count_no_visited = 0;
+    Vector<InnerIdType> to_be_visited_rid(graph_data_cell_->MaximumDegree(), allocator_);


Signed-off-by: zhongxiaoyao.zxy <[email protected]>

inabao

LGTM

ShawnShawnYou requested review from inabao, wxyucs, LHT129 and jiaweizone as code owners January 20, 2025 09:08

pull-request-size bot added the size/XL label Jan 20, 2025

LHT129 reviewed Jan 20, 2025

View reviewed changes

ShawnShawnYou force-pushed the support-searcher branch from 0d51c38 to b267091 Compare January 20, 2025 11:47

inabao reviewed Jan 20, 2025

View reviewed changes

ShawnShawnYou force-pushed the support-searcher branch 2 times, most recently from 8af23a1 to 6450d83 Compare January 20, 2025 15:53

pull-request-size bot added size/L and removed size/XL labels Jan 20, 2025

ShawnShawnYou force-pushed the support-searcher branch from 6450d83 to a8a6a01 Compare January 20, 2025 16:14

LHT129 reviewed Jan 21, 2025

View reviewed changes

support basic searcher

cb903ef

Signed-off-by: zhongxiaoyao.zxy <[email protected]>

ShawnShawnYou force-pushed the support-searcher branch from a8a6a01 to cb903ef Compare January 21, 2025 03:48

ShawnShawnYou added kind/feature New feature or request version/0.13 labels Jan 21, 2025

ShawnShawnYou self-assigned this Jan 21, 2025

inabao approved these changes Jan 21, 2025

View reviewed changes

LHT129 added version/0.14 and removed version/0.13 labels Jan 22, 2025

support basic searcher #351

Are you sure you want to change the base?

support basic searcher #351

Conversation

ShawnShawnYou commented Jan 20, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

LHT129 Jan 20, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Jan 20, 2025 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

inabao left a comment

Choose a reason for hiding this comment

LHT129 Jan 20, 2025 •

edited

Loading

codecov bot commented Jan 20, 2025 •

edited

Loading