Fix semi mask on scan node table

src/function/gds/shortest_paths.cpp

@@ -83,7 +83,7 @@
     explicit PathLengths(
         std::vector<std::tuple<common::table_id_t, uint64_t>> nodeTableIDAndNumNodes) {
         for (const auto& [tableID, numNodes] : nodeTableIDAndNumNodes) {
-            masks.insert({tableID, std::make_unique<processor::MaskData>(numNodes, UNVISITED)});
+            masks.insert({tableID, std::make_unique<MaskData>(numNodes, UNVISITED)});
         }
     }
 
@@ -131,10 +131,10 @@
 
 private:
     uint8_t curIter = 255;
-    common::table_id_map_t<std::unique_ptr<processor::MaskData>> masks;
+    common::table_id_map_t<std::unique_ptr<MaskData>> masks;
     common::table_id_t curFrontierFixedTableID;
-    processor::MaskData* curFrontierFixedMask;
-    processor::MaskData* nextFrontierFixedMask;
+    MaskData* curFrontierFixedMask;
+    MaskData* nextFrontierFixedMask;
 };
 
 class PathLengthsFrontiers : public Frontiers {
@@ -419,7 +419,7 @@
             }
             auto mask = sharedState->inputNodeOffsetMasks.at(tableID).get();
             for (auto offset = 0u; offset < sharedState->graph->getNumNodes(tableID); ++offset) {
-                if (!mask->isMasked(offset)) {
+                if (!mask->isMasked(offset, offset)) {
                     continue;
                 }
                 auto sourceNodeID = nodeID_t{offset, tableID};

src/include/processor/operator/mask.h → src/include/common/mask.h

@@ -7,7 +7,7 @@
 #include "common/types/internal_id_t.h"
 
 namespace kuzu {
-namespace processor {
+namespace common {
 
 // Note: Classes in this file are NOT thread-safe.
 struct MaskUtil {
@@ -52,7 +52,15 @@
         maskData = std::make_unique<MaskData>(maxOffset + 1);
     }
 
-    bool isMasked(common::offset_t offset) { return maskData->isMasked(offset, numMasks); }
+    // Return true if any offset between [startOffset, endOffset] is masked. Otherwise return false.
+    bool isMasked(common::offset_t startOffset, common::offset_t endOffset) const {
+        auto offset = startOffset;
+        auto numMasked = 0u;
+        while (offset <= endOffset) {
+            numMasked += maskData->isMasked(offset++, numMasks);
+        }
+        return numMasked > 0;
+    }
     // Increment mask value for the given nodeOffset if its current mask value is equal to
     // the specified `currentMaskValue`.
     // Note: blindly update mask does not parallelize well, so we minimize write by first checking
@@ -87,7 +95,7 @@
     virtual void init() = 0;
 
     virtual void incrementMaskValue(common::offset_t nodeOffset, uint8_t currentMaskValue) = 0;
-    virtual bool isMasked(common::offset_t nodeOffset) = 0;
+    virtual bool isMasked(common::offset_t startNodeOffset, common::offset_t endNodeOffset) = 0;
 
     bool isEnabled() const { return getNumMasks() > 0; }
     uint8_t getNumMasks() const { return maskCollection.getNumMasks(); }
@@ -115,8 +123,8 @@
         maskCollection.incrementMaskValue(nodeOffset, currentMaskValue);
     }
 
-    bool isMasked(common::offset_t nodeOffset) override {
-        return maskCollection.isMasked(nodeOffset);
+    bool isMasked(common::offset_t startNodeOffset, common::offset_t endNodeOffset) override {
+        return maskCollection.isMasked(startNodeOffset, endNodeOffset);
     }
 };
 
@@ -136,10 +144,11 @@
         maskCollection.incrementMaskValue(MaskUtil::getVectorIdx(nodeOffset), currentMaskValue);
     }
 
-    bool isMasked(common::offset_t nodeOffset) override {
-        return maskCollection.isMasked(MaskUtil::getVectorIdx(nodeOffset));
+    bool isMasked(common::offset_t startNodeOffset, common::offset_t endNodeOffset) override {
+        return maskCollection.isMasked(MaskUtil::getVectorIdx(startNodeOffset),
+            MaskUtil::getVectorIdx(endNodeOffset));
     }
 };
 
-} // namespace processor
+} // namespace common
 } // namespace kuzu

src/include/planner/operator/sip/side_way_info_passing.h

@@ -9,6 +9,7 @@ enum class SemiMaskPosition : uint8_t {
     NONE = 0,
     ON_BUILD = 1,
     ON_PROBE = 2,
+    // TODO(Xiyang): we should differentiate build to probe and probe to build.
     PROHIBIT = 3,
 };
 

src/include/processor/operator/gds_call.h

@@ -1,9 +1,9 @@
 #pragma once
 
+#include "common/mask.h"
 #include "function/gds/gds.h"
 #include "function/gds/gds_utils.h"
 #include "graph/graph.h"
-#include "processor/operator/mask.h"
 #include "processor/operator/sink.h"
 
 namespace kuzu {
@@ -13,10 +13,11 @@ struct GDSCallSharedState {
     std::mutex mtx;
     std::shared_ptr<FactorizedTable> fTable;
     std::unique_ptr<graph::Graph> graph;
-    common::table_id_map_t<std::unique_ptr<NodeOffsetLevelSemiMask>> inputNodeOffsetMasks;
+    common::table_id_map_t<std::unique_ptr<common::NodeOffsetLevelSemiMask>> inputNodeOffsetMasks;
 
     GDSCallSharedState(std::shared_ptr<FactorizedTable> fTable, std::unique_ptr<graph::Graph> graph,
-        common::table_id_map_t<std::unique_ptr<NodeOffsetLevelSemiMask>> inputNodeOffsetMasks)
+        common::table_id_map_t<std::unique_ptr<common::NodeOffsetLevelSemiMask>>
+            inputNodeOffsetMasks)
         : fTable{std::move(fTable)}, graph{std::move(graph)},
           inputNodeOffsetMasks{std::move(inputNodeOffsetMasks)} {}
     DELETE_COPY_AND_MOVE(GDSCallSharedState);
@@ -59,7 +60,7 @@ class GDSCall : public Sink {
           info{std::move(info)}, sharedState{std::move(sharedState)} {}
 
     bool hasSemiMask() const { return !sharedState->inputNodeOffsetMasks.empty(); }
-    std::vector<NodeSemiMask*> getSemiMasks() const;
+    std::vector<common::NodeSemiMask*> getSemiMasks() const;
 
     bool isSource() const override { return true; }
 

src/include/processor/operator/recursive_extend/recursive_join.h

@@ -3,9 +3,9 @@
 #include "bfs_state.h"
 #include "common/enums/extend_direction.h"
 #include "common/enums/query_rel_type.h"
+#include "common/mask.h"
 #include "frontier_scanner.h"
 #include "planner/operator/extend/recursive_join_type.h"
-#include "processor/operator/mask.h"
 #include "processor/operator/physical_operator.h"
 
 namespace kuzu {
@@ -14,10 +14,10 @@ namespace processor {
 class OffsetScanNodeTable;
 
 struct RecursiveJoinSharedState {
-    std::vector<std::unique_ptr<NodeOffsetLevelSemiMask>> semiMasks;
+    std::vector<std::unique_ptr<common::NodeOffsetLevelSemiMask>> semiMasks;
 
     explicit RecursiveJoinSharedState(
-        std::vector<std::unique_ptr<NodeOffsetLevelSemiMask>> semiMasks)
+        std::vector<std::unique_ptr<common::NodeOffsetLevelSemiMask>> semiMasks)
         : semiMasks{std::move(semiMasks)} {}
 };
 
@@ -116,7 +116,7 @@ class RecursiveJoin : public PhysicalOperator {
           info{std::move(info)}, sharedState{std::move(sharedState)},
           recursiveRoot{std::move(recursiveRoot)} {}
 
-    std::vector<NodeSemiMask*> getSemiMask() const;
+    std::vector<common::NodeSemiMask*> getSemiMask() const;
 
     void initLocalStateInternal(ResultSet* resultSet_, ExecutionContext* context) final;
 

src/include/processor/operator/scan/scan_node_table.h

@@ -16,7 +16,7 @@ struct ScanNodeTableProgressSharedState {
 
 class ScanNodeTableSharedState {
 public:
-    explicit ScanNodeTableSharedState(std::unique_ptr<NodeVectorLevelSemiMask> semiMask)
+    explicit ScanNodeTableSharedState(std::unique_ptr<common::NodeVectorLevelSemiMask> semiMask)
         : table{nullptr}, currentCommittedGroupIdx{common::INVALID_NODE_GROUP_IDX},
           currentUnCommittedGroupIdx{common::INVALID_NODE_GROUP_IDX}, numCommittedNodeGroups{0},
           numUnCommittedNodeGroups{0}, semiMask{std::move(semiMask)} {};
@@ -27,7 +27,7 @@ class ScanNodeTableSharedState {
     void nextMorsel(storage::NodeTableScanState& scanState,
         std::shared_ptr<ScanNodeTableProgressSharedState> progressSharedState);
 
-    NodeSemiMask* getSemiMask() const { return semiMask.get(); }
+    common::NodeSemiMask* getSemiMask() const { return semiMask.get(); }
 
 private:
     std::mutex mtx;
@@ -36,7 +36,7 @@ class ScanNodeTableSharedState {
     common::node_group_idx_t currentUnCommittedGroupIdx;
     common::node_group_idx_t numCommittedNodeGroups;
     common::node_group_idx_t numUnCommittedNodeGroups;
-    std::unique_ptr<NodeVectorLevelSemiMask> semiMask;
+    std::unique_ptr<common::NodeVectorLevelSemiMask> semiMask;
 };
 
 struct ScanNodeTableInfo {
@@ -52,7 +52,7 @@ struct ScanNodeTableInfo {
           columnPredicates{std::move(columnPredicates)} {}
     EXPLICIT_COPY_DEFAULT_MOVE(ScanNodeTableInfo);
 
-    void initScanState(NodeSemiMask* semiMask);
+    void initScanState(common::NodeSemiMask* semiMask);
 
 private:
     ScanNodeTableInfo(const ScanNodeTableInfo& other)
@@ -93,7 +93,7 @@ class ScanNodeTable final : public ScanTable {
         KU_ASSERT(this->nodeInfos.size() == this->sharedStates.size());
     }
 
-    std::vector<NodeSemiMask*> getSemiMasks() const;
+    std::vector<common::NodeSemiMask*> getSemiMasks() const;
 
     bool isSource() const override { return true; }
 

src/include/processor/operator/semi_masker.h

@@ -1,6 +1,6 @@
 #pragma once
 
-#include "processor/operator/mask.h"
+#include "common/mask.h"
 #include "processor/operator/physical_operator.h"
 
 namespace kuzu {
@@ -12,7 +12,7 @@ class BaseSemiMasker;
 // to indicate the execution sequence of its pipeline. Also, the maskerIdx is used as a flag to
 // indicate if a value in the mask is masked or not, as each masker will increment the selected
 // value in the mask by 1. More details are described in NodeTableSemiMask.
-using mask_with_idx = std::pair<NodeSemiMask*, uint8_t>;
+using mask_with_idx = std::pair<common::NodeSemiMask*, uint8_t>;
 
 class SemiMaskerInfo {
     friend class BaseSemiMasker;

src/include/storage/store/node_table.h

@@ -2,12 +2,11 @@
 
 #include <cstdint>
 
+#include "common/exception/not_implemented.h"
 #include "common/types/types.h"
-#include "processor/operator/mask.h"
 #include "storage/index/hash_index.h"
 #include "storage/store/node_group_collection.h"
 #include "storage/store/table.h"
-#include <common/exception/not_implemented.h>
 
 namespace kuzu {
 namespace evaluator {
@@ -26,25 +25,21 @@ class Transaction;
 namespace storage {
 
 struct NodeTableScanState final : TableScanState {
-    processor::NodeSemiMask* semiMask;
-
     // Scan state for un-committed data.
     // Ideally we shouldn't need columns to scan un-checkpointed but committed data.
     explicit NodeTableScanState(std::vector<common::column_id_t> columnIDs)
-        : TableScanState{std::move(columnIDs), {}}, semiMask{nullptr} {
+        : TableScanState{std::move(columnIDs), {}} {
         nodeGroupScanState = std::make_unique<NodeGroupScanState>(this->columnIDs.size());
     }
 
     NodeTableScanState(std::vector<common::column_id_t> columnIDs, std::vector<Column*> columns)
         : TableScanState{std::move(columnIDs), std::move(columns),
-              std::vector<ColumnPredicateSet>{}},
-          semiMask{nullptr} {
+              std::vector<ColumnPredicateSet>{}} {
         nodeGroupScanState = std::make_unique<NodeGroupScanState>(this->columnIDs.size());
     }
     NodeTableScanState(std::vector<common::column_id_t> columnIDs, std::vector<Column*> columns,
         std::vector<ColumnPredicateSet> columnPredicateSets)
-        : TableScanState{std::move(columnIDs), std::move(columns), std::move(columnPredicateSets)},
-          semiMask{nullptr} {
+        : TableScanState{std::move(columnIDs), std::move(columns), std::move(columnPredicateSets)} {
         nodeGroupScanState = std::make_unique<NodeGroupScanState>(this->columnIDs.size());
     }
 };

src/include/storage/store/table.h

@@ -2,6 +2,7 @@
 
 #include "catalog/catalog_entry/table_catalog_entry.h"
 #include "common/enums/zone_map_check_result.h"
+#include "common/mask.h"
 #include "storage/predicate/column_predicate.h"
 #include "storage/store/column.h"
 #include "storage/store/node_group.h"
@@ -20,8 +21,9 @@ struct TableScanState {
     common::ValueVector* IDVector;
     std::vector<common::ValueVector*> outputVectors;
     std::vector<common::column_id_t> columnIDs;
+    common::NodeSemiMask* semiMask;
 
-    // Only used when scan from checkpointed data.
+    // Only used when scan from persistent data.
     std::vector<Column*> columns;
 
     TableScanSource source = TableScanSource::NONE;
@@ -33,18 +35,19 @@ struct TableScanState {
     common::ZoneMapCheckResult zoneMapResult = common::ZoneMapCheckResult::ALWAYS_SCAN;
 
     explicit TableScanState(std::vector<common::column_id_t> columnIDs)
-        : IDVector(nullptr), columnIDs{std::move(columnIDs)} {
+        : IDVector(nullptr), columnIDs{std::move(columnIDs)}, semiMask{nullptr} {
         rowIdxVector = std::make_unique<common::ValueVector>(common::LogicalType::INT64());
     }
     TableScanState(std::vector<common::column_id_t> columnIDs, std::vector<Column*> columns,
         std::vector<ColumnPredicateSet> columnPredicateSets)
-        : IDVector(nullptr), columnIDs{std::move(columnIDs)}, columns{std::move(columns)},
-          columnPredicateSets{std::move(columnPredicateSets)} {
+        : IDVector(nullptr), columnIDs{std::move(columnIDs)}, semiMask{nullptr},
+          columns{std::move(columns)}, columnPredicateSets{std::move(columnPredicateSets)} {
         rowIdxVector = std::make_unique<common::ValueVector>(common::LogicalType::INT64());
     }
     explicit TableScanState(std::vector<common::column_id_t> columnIDs,
         std::vector<Column*> columns)
-        : IDVector(nullptr), columnIDs{std::move(columnIDs)}, columns{std::move(columns)} {
+        : IDVector(nullptr), columnIDs{std::move(columnIDs)}, semiMask{nullptr},
+          columns{std::move(columns)} {
         rowIdxVector = std::make_unique<common::ValueVector>(common::LogicalType::INT64());
     }
     virtual ~TableScanState() = default;

src/processor/map/map_recursive_extend.cpp

@@ -12,11 +12,11 @@ namespace processor {
 
 static std::shared_ptr<RecursiveJoinSharedState> createSharedState(
     const binder::NodeExpression& nbrNode, const main::ClientContext& context) {
-    std::vector<std::unique_ptr<NodeOffsetLevelSemiMask>> semiMasks;
+    std::vector<std::unique_ptr<common::NodeOffsetLevelSemiMask>> semiMasks;
     for (auto tableID : nbrNode.getTableIDs()) {
         auto table = context.getStorageManager()->getTable(tableID)->ptrCast<storage::NodeTable>();
         semiMasks.push_back(
-            std::make_unique<NodeOffsetLevelSemiMask>(tableID, table->getNumRows()));
+            std::make_unique<common::NodeOffsetLevelSemiMask>(tableID, table->getNumRows()));
     }
     return std::make_shared<RecursiveJoinSharedState>(std::move(semiMasks));
 }

src/processor/operator/gds_call.cpp

@@ -10,8 +10,8 @@
     return "Algorithm: " + funcName;
 }
 
-std::vector<NodeSemiMask*> GDSCall::getSemiMasks() const {
-    std::vector<NodeSemiMask*> masks;
+std::vector<common::NodeSemiMask*> GDSCall::getSemiMasks() const {
+    std::vector<common::NodeSemiMask*> masks;
     for (auto& [_, mask] : sharedState->inputNodeOffsetMasks) {
         masks.push_back(mask.get());
     }

src/processor/operator/recursive_extend/recursive_join.cpp

@@ -280,7 +280,7 @@ void RecursiveJoin::populateTargetDstNodes(ExecutionContext*) {
         auto numNodes = mask->getMaxOffset() + 1;
         if (mask->isEnabled()) {
             for (auto offset = 0u; offset < numNodes; ++offset) {
-                if (mask->isMasked(offset)) {
+                if (mask->isMasked(offset, offset)) {
                     targetNodeIDs.insert(nodeID_t{offset, mask->getTableID()});
                     numTargetNodes++;
                 }