WIP: v1 recursive

homework_1/Makefile

@@ -23,11 +23,11 @@ PROJECT         := PDS_homework_1
 # Excecutable's name
 TARGET          := knnsearch
 # Source directories list(space seperated). Makefile-relative path, UNDER current directory.
-SRC_DIR_LIST    := src
+SRC_DIR_LIST    := src gtest
 # Include directories list(space seperated). Makefile-relative path.
 INC_DIR_LIST    := inc \
                    src \
-                   Libs/matrix/include/ \
+                   gtest \
                    /usr/include/hdf5/serial/
 #                   Libs/MATLAB/R2019b/include/ \
 
@@ -56,7 +56,6 @@ PRE_DEFS		:=
 LDFLAGS         := -pthread -lopenblas \
                    -L/usr/lib/x86_64-linux-gnu/hdf5/serial -lhdf5
 #                   -LLibs/MATLAB/R2019b/bin/ -lmat -lmx -Wl,-rpath,Libs/MATLAB/R2019b/bin/ \
-#                   -Wl,-rpath,Libs/unwind/bin/
 
 # Map output file
 MAP_FILE        := output.map
@@ -198,8 +197,8 @@ local_v0_opt: TARGET := local_v0_opt
 local_v0_opt: $(BUILD_DIR)/$(TARGET)
 	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
 	
-local_v1: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=4
-local_v1: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=4
+local_v1: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=1
+local_v1: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=1
 local_v1: TARGET := local_v1
 local_v1: $(BUILD_DIR)/$(TARGET)
 	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
@@ -235,7 +234,13 @@ v1: TARGET := knnsearch_v1
 v1: $(BUILD_DIR)/$(TARGET)
 	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
 
-
+tests: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=0 -DTESTING
+tests: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=0 -DTESTING
+tests: TARGET := tests
+tests: $(BUILD_DIR)/$(TARGET)
+	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
+	
+	
 #
 # ========= Inside CSAL Image build rules ===========
 #
@@ -280,6 +285,6 @@ csal_v1: $(BUILD_DIR)/$(TARGET)
 # make IMAGE=hpcimage EXEC=knnsearch_v1 run
 # make IMAGE=hpcimage EXEC=knnsearch_v1 run
 #
-run: DOCKER := $(DOCKER_CMD)
+
 run:
 	$(DOCKER) ./out/$(EXEC)

homework_1/inc/config.h

@@ -13,7 +13,7 @@
 #include <iostream>
 #include <string>
 
-#include <matrix.hpp>
+#include "matrix.hpp"
 
 // HDF5 supported types
 enum class HDF5_type {
@@ -32,21 +32,11 @@ enum class HDF5_type {
  #define CODE_VERSION   V1
 #endif
 
-// matrix alias template dispatcher based on pre-define flag from compiler (see Makefile)
-#if   CODE_VERSION == V0
-   #define NAMESPACE_VERSION using namespace v0
-   using MatrixDst = mtx::Matrix<double>;
-   using MatrixIdx = mtx::Matrix<uint32_t>;
-   static constexpr HDF5_type DstHDF5Type = HDF5_type::DOUBLE;
-   static constexpr HDF5_type IdxHDF5Type = HDF5_type::INT;
-#elif CODE_VERSION == V1
-   #define NAMESPACE_VERSION using namespace v1
-   using MatrixDst = mtx::Matrix<double>;
-   using MatrixIdx = mtx::Matrix<uint32_t>;
-   static constexpr HDF5_type DstHDF5Type = HDF5_type::DOUBLE;
-   static constexpr HDF5_type IdxHDF5Type = HDF5_type::INT;
-#endif
-
+// matrix alias template dispatcher
+using MatrixDst = mtx::Matrix<double>;
+using MatrixIdx = mtx::Matrix<uint32_t>;
+static constexpr HDF5_type DstHDF5Type = HDF5_type::DOUBLE;
+static constexpr HDF5_type IdxHDF5Type = HDF5_type::INT;
 
 
 //! enumerator for output handling

homework_1/Libs/matrix/include/matrix.hpp → homework_1/inc/matrix.hpp

@@ -83,25 +83,49 @@ struct Matrix {
    static constexpr MatrixOrder matrixOrder = Order; //!< meta:export of array order
    static constexpr MatrixType matrixType = Type;  //!< meta:export of array type
    static constexpr bool symmetric = Symmetric;    //!< meta:export symmetric flag
+
    /*!
     * \name Obj lifetime
     */
    //! @{
 
    //! Construct an empty matrix with dimensions rows x columns
-   Matrix(IndexType rows = IndexType{}, IndexType columns = IndexType{}) noexcept :
-      m_(capacity(rows, columns)), rows_(rows), cols_(columns) { }
+   Matrix(IndexType rows = IndexType{}, IndexType columns = IndexType{}) noexcept
+       : vector_storage_(capacity(rows, columns)),
+         raw_storage_(nullptr),
+         use_vector_(true),
+         rows_(rows),
+         cols_(columns) {
+       data_ = vector_storage_.data();
+    }
 
    //! Construct a matrix by copying existing data with dimensions rows x columns
-   Matrix(DataType* data, IndexType rows, IndexType columns) noexcept :
-      m_(data, data + capacity(rows, columns)), rows_(rows), cols_(columns) { }
+   Matrix(DataType* data, IndexType major_start, IndexType major_length, IndexType minor_length) noexcept
+      :  vector_storage_(),
+         raw_storage_ (data + major_start * minor_length),
+         use_vector_ (false) {
+      if constexpr (Order == MatrixOrder::ROWMAJOR) {
+         rows_ = major_length;
+         cols_ = minor_length;
+      }
+      else {
+         rows_ = minor_length;
+         cols_ = major_length;
+      }
+      data_ = raw_storage_;
+   }
 
    //! Construct a matrix using an initializer list
    Matrix(IndexType rows, IndexType columns, std::initializer_list<DataType> list)
-      : m_(list), rows_(rows), cols_(columns) {
+      : vector_storage_(list),
+        raw_storage_(nullptr),
+        use_vector_(true),
+        rows_(rows),
+        cols_(columns) {
       if (list.size() != capacity(rows, columns)) {
-         throw std::invalid_argument("Matrix initializer list size does not match matrix dimensions.");
+          throw std::invalid_argument("Matrix initializer list size does not match matrix dimensions.");
       }
+      data_ = vector_storage_.data();
    }
 
    //! move ctor
@@ -126,9 +150,12 @@ struct Matrix {
    }
    //! Set the interface size of the Matrix (what appears to be the size)
    IndexType resize(IndexType rows, IndexType columns) {
-      rows_ = rows;
-      cols_ = columns;
-      m_.reserve(capacity(rows_, cols_));
+      if (use_vector_) {
+         rows_ = rows;
+         cols_ = columns;
+         vector_storage_.resize(capacity(rows_, cols_));
+         data_ = vector_storage_.data();
+      }
       return capacity(rows_, cols_);
    }
 
@@ -148,42 +175,49 @@ struct Matrix {
          auto T = [](size_t i)->size_t { return i*(i+1)/2; };          // Triangular number of i
          if constexpr (Order == MatrixOrder::COLMAJOR) {
             // In column major we use the lower triangle of the matrix
-            if (i>=j)   return m_[j*rows_ - T(j) + i];  // Lower, use our notation
-            else        return m_[i*rows_ - T(i) + j];  // Upper, use opposite index
+            if (i>=j)   return data_[j*rows_ - T(j) + i];  // Lower, use our notation
+            else        return data_[i*rows_ - T(i) + j];  // Upper, use opposite index
          }
          else {
             // In row major we use the upper triangle of the matrix
-            if (i<=j)   return m_[i*cols_ - T(i) + j];  // Upper, use our notation
-            else        return m_[j*cols_ - T(j) + i];  // Lower, use opposite index
+            if (i<=j)   return data_[i*cols_ - T(i) + j];  // Upper, use our notation
+            else        return data_[j*cols_ - T(j) + i];  // Lower, use opposite index
          }
       }
       else {
          if constexpr (Order == MatrixOrder::COLMAJOR)
-            return m_[i + j*rows_];
+            return data_[i + j*rows_];
          else
-            return m_[i*cols_ + j];
+            return data_[i*cols_ + j];
       }
    }
 
+   /*!
+    * \fn DataType set(DataType, IndexType, IndexType)
+    * \param v
+    * \param i
+    * \param j
+    * \return
+    */
    DataType set (DataType v, IndexType i, IndexType j) {
       if constexpr (Symmetric) {
          auto T = [](size_t i)->size_t { return i*(i+1)/2; };          // Triangular number of i
          if constexpr (Order == MatrixOrder::COLMAJOR) {
             // In column major we use the lower triangle of the matrix
-            if (i>=j)   return m_[j*rows_ - T(j) + i] = v;  // Lower, use our notation
-            else        return m_[i*rows_ - T(i) + j] = v;  // Upper, use opposite index
+            if (i>=j)   return data_[j*rows_ - T(j) + i] = v;  // Lower, use our notation
+            else        return data_[i*rows_ - T(i) + j] = v;  // Upper, use opposite index
          }
          else {
             // In row major we use the upper triangle of the matrix
-            if (i<=j)   return m_[i*cols_ - T(i) + j] = v;  // Upper, use our notation
-            else        return m_[j*cols_ - T(j) + i] = v;  // Lower, use opposite index
+            if (i<=j)   return data_[i*cols_ - T(i) + j] = v;  // Upper, use our notation
+            else        return data_[j*cols_ - T(j) + i] = v;  // Lower, use opposite index
          }
       }
       else {
          if constexpr (Order == MatrixOrder::COLMAJOR)
-            return m_[i + j*rows_] = v;
+            return data_[i + j*rows_] = v;
          else
-            return m_[i*cols_ + j] = v;
+            return data_[i*cols_ + j] = v;
       }
    }
 //   DataType operator()(IndexType i, IndexType j) { return get(i, j); }
@@ -198,11 +232,11 @@ struct Matrix {
    }
 
    // a basic serial iterator support
-   DataType* data() noexcept { return m_.data(); }
+   DataType* data() noexcept { return data_; }
 //   IndexType begin_idx() noexcept { return 0; }
 //   IndexType end_idx()   noexcept { return capacity(rows_, cols_); }
 
-   const DataType* data() const noexcept { return m_.data(); }
+   const DataType* data() const noexcept { return data_; }
    const IndexType begin_idx() const noexcept { return 0; }
    const IndexType end_idx()   const noexcept { return capacity(rows_, cols_); }
    //! @}
@@ -224,19 +258,29 @@ struct Matrix {
 
    //
    void swap(Matrix& src) noexcept {
-      std::swap(m_, src.m_);
+      std::swap(vector_storage_, src.vector_storage_);
+      std::swap(raw_storage_, src.raw_storage_);
+      std::swap(data_, src.data_);
+      std::swap(use_vector_, src.use_vector_);
       std::swap(rows_, src.rows_);
       std::swap(cols_, src.cols_);
    }
 private:
    //! move helper
    void moves(Matrix&& src) noexcept {
-      m_    = std::move(src.m_);
+      data_ = std::move(src.vector_storage_);
+      data_ = std::move(src.raw_storage_);
+      data_ = std::move(src.data_);
+      data_ = std::move(src.use_vector_);
       rows_ = std::move(src.rows_);
       cols_ = std::move(src.cols_);
    }
 
-   std::vector<DataType> m_ {};  //!< Pointer to actual data.
+   std::vector<DataType>
+               vector_storage_;  //!< Internal storage (if used).
+   DataType*   raw_storage_;     //!< External storage (if used).
+   DataType*   data_;            //!< Pointer to active storage.
+   bool        use_vector_;      //!< True if using vector storage, false for raw pointer.
    IndexType   rows_{};          //!< the virtual size of rows.
    IndexType   cols_{};          //!< the virtual size of columns.
 };
@@ -484,6 +528,125 @@ private:
 };
 
 
+template<typename ...> struct Matrix_view { };
+
+/*!
+ * @struct Matrix_view
+ * @tparam MatrixType
+ */
+template<template <typename, typename, MatrixType, MatrixOrder, bool> class Matrix,
+         typename DataType,
+         typename IndexType,
+         MatrixType Type,
+         MatrixOrder Order>
+struct Matrix_view<Matrix<DataType, IndexType, Type, Order, false>> {
+   using owner_t = Matrix<DataType, IndexType, Type, Order, false>;
+
+   using dataType    = DataType;                      //!< meta:export of underling data type
+   using indexType   = IndexType;                     //!< meta:export of underling index type
+   static constexpr MatrixOrder matrixOrder = Order;  //!< meta:export of array order
+   static constexpr MatrixType matrixType = Type;     //!< meta:export of array type
+
+   /*!
+    * \name Obj lifetime
+    */
+   //! @{
+
+   //! Construct a matrix view to entire matrix
+   Matrix_view(const owner_t* owner) noexcept :
+      owner_(owner), m_(owner->data()), rows_(owner->rows()), cols_(owner->columns()) { }
+
+   Matrix_view(const owner_t* owner, IndexType begin, IndexType end) noexcept :
+      owner_(owner) {
+         if constexpr (Order == MatrixOrder::ROWMAJOR) {
+            m_ = owner->data() + begin * owner->columns();
+            rows_ = end - begin;
+            cols_ = owner->columns();
+         } else if (Order == MatrixOrder::COLMAJOR) {
+            m_ = owner->data() + begin * owner->rows();
+            rows_ = owner->rows();
+            cols_ = end - begin;
+         }
+   }
+
+   Matrix_view(Matrix_view&& m)                 = delete; //! No move
+   Matrix_view& operator=(Matrix_view&& m)      = delete;
+   Matrix_view(const Matrix_view& m)            = delete; //!< No copy
+   Matrix_view& operator=(const Matrix_view& m) = delete;
+   //! @}
+
+   //! Get/Set the size of each dimension
+   const IndexType rows() const noexcept { return rows_; }
+   const IndexType columns() const noexcept { return cols_; }
+
+   //! Get the interface size of the Matrix (what appears to be the size)
+   IndexType size() const {
+      return rows_ * cols_;
+   }
+
+   //! Actual memory capacity of the symmetric matrix
+   static constexpr IndexType capacity(IndexType M, IndexType N) {
+      return M*N;
+   }
+   /*
+    * virtual 2D accessors
+    */
+   const DataType get (IndexType i, IndexType j) const {
+      if constexpr (Order == MatrixOrder::COLMAJOR)
+         return m_[i + j*rows_];
+      else
+         return m_[i*cols_ + j];
+   }
+
+   DataType set (DataType v, IndexType i, IndexType j) {
+      if constexpr (Order == MatrixOrder::COLMAJOR)
+         return m_[i + j*rows_] = v;
+      else
+         return m_[i*cols_ + j] = v;
+   }
+//   DataType operator()(IndexType i, IndexType j) { return get(i, j); }
+   /*!
+    * Return a proxy MatVal object with read and write capabilities.
+    * @param i    The row number
+    * @param j    The column number
+    * @return     tHE MatVal object
+    */
+   MatVal<Matrix_view> operator()(IndexType i, IndexType j) noexcept {
+      return MatVal<Matrix_view>(this, get(i, j), i, j);
+   }
+
+   // a basic serial iterator support
+   DataType* data() noexcept { return m_.data(); }
+//   IndexType begin_idx() noexcept { return 0; }
+//   IndexType end_idx()   noexcept { return capacity(rows_, cols_); }
+
+   const DataType* data() const noexcept { return m_; }
+   const IndexType begin_idx() const noexcept { return 0; }
+   const IndexType end_idx()   const noexcept { return capacity(rows_, cols_); }
+   //! @}
+
+   /*!
+    * \name Safe iteration API
+    *
+    * This api automates the iteration over the array based on
+    * MatrixType
+    */
+   //! @{
+   template<typename F, typename... Args>
+   void for_each_in (IndexType begin, IndexType end, F&& lambda, Args&&... args) {
+      for (IndexType it=begin ; it<end ; ++it) {
+         std::forward<F>(lambda)(std::forward<Args>(args)..., it);
+      }
+   }
+   //! @}
+   //!
+private:
+   const owner_t* owner_ {nullptr};     //!< Pointer to Matrix
+   DataType*   m_       {nullptr};     //!< Starting address of the slice/view
+   IndexType   rows_{};                //!< the virtual size of rows.
+   IndexType   cols_{};                //!< the virtual size of columns.
+};
+
 /*!
  * A view/iterator hybrid object for Matrix columns.
  *

homework_1/inc/utils.hpp

@@ -14,8 +14,8 @@
 #include <unistd.h>
 #include <hdf5.h>
 
-#include <matrix.hpp>
-#include <config.h>
+#include "matrix.hpp"
+#include "config.h"
 
 /*!
  * A Logger for entire program.

homework_1/inc/v0.hpp

@@ -14,8 +14,8 @@
 #include <vector>
 #include <algorithm>
 
-#include <matrix.hpp>
-#include <config.h>
+#include "matrix.hpp"
+#include "config.h"
 
 namespace v0 {
 
@@ -30,8 +30,10 @@ namespace v0 {
  * \param n    number of rows in Y
  * \param d    number of columns in both X and Y
  */
-template<typename DataType>
-void pdist2(const mtx::Matrix<DataType>& X, const mtx::Matrix<DataType>& Y, mtx::Matrix<DataType>& D2) {
+template<typename Matrix>
+void pdist2(const Matrix& X, const Matrix& Y, Matrix& D2) {
+   using DataType = typename Matrix::dataType;
+
    int M = X.rows();
    int N = Y.rows();
    int d = X.columns();
@@ -79,26 +81,24 @@ void quickselect(std::vector<std::pair<DataType, IndexType>>& vec, int k) {
  * \param dst  Is the Nxk matrix with the k distances to the C points of the nth
  *             point of Q
  */
-template<typename DataType, typename IndexType>
-void knnsearch(const mtx::Matrix<DataType>& C, const mtx::Matrix<DataType>& Q, int k,
-               mtx::Matrix<IndexType>& idx,
-               mtx::Matrix<DataType>& dst) {
+template<typename MatrixD, typename MatrixI>
+void knnsearch(const MatrixD& C, const MatrixD& Q, size_t idx_offset, size_t k, size_t m, MatrixI& idx, MatrixD& dst) {
 
-   int M = C.rows();
-   int N = Q.rows();
+   using DstType = typename MatrixD::dataType;
+   using IdxType = typename MatrixI::dataType;
 
-   mtx::Matrix<DataType> D(M, N);
+   size_t M = C.rows();
+   size_t N = Q.rows();
 
-   pdist2(C, Q, D);
+   mtx::Matrix<DstType> D(M, N);
 
-   idx.resize(N, k);
-   dst.resize(N, k);
+   pdist2(C, Q, D);
 
-   for (int j = 0; j < N; ++j) {
+   for (size_t j = 0; j < N; ++j) {
       // Create a vector of pairs (distance, index) for the j-th query
-      std::vector<std::pair<DataType, IndexType>> dst_idx(M);
-      for (int i = 0; i < M; ++i) {
-        dst_idx[i] = {D.data()[i * N + j], i};
+      std::vector<std::pair<DstType, IdxType>> dst_idx(M);
+      for (size_t i = 0; i < M; ++i) {
+         dst_idx[i] = {D.data()[i * N + j], i};
       }
       // Find the k smallest distances using quickSelectKSmallest
       quickselect(dst_idx, k);
@@ -107,9 +107,9 @@ void knnsearch(const mtx::Matrix<DataType>& C, const mtx::Matrix<DataType>& Q, i
       std::sort(dst_idx.begin(), dst_idx.end());
 
       // Store the indices and distances
-      for (int i = 0; i < k; ++i) {
-         idx(j, i) = dst_idx[i].second;
-         dst(j, i) = dst_idx[i].first;
+      for (size_t i = 0; i < k; ++i) {
+         dst.set(dst_idx[i].first, j, i);
+         idx.set(dst_idx[i].second + idx_offset, j, i);
       }
    }
 }

homework_1/inc/v1.hpp

@@ -9,8 +9,100 @@
 #ifndef V1_HPP_
 #define V1_HPP_
 
+#include <vector>
+#include <algorithm>
 
+#include "matrix.hpp"
+#include "v0.hpp"
+#include "config.h"
 
+namespace v1 {
 
+template <typename DataType, typename IndexType>
+void mergeResultsWithM(mtx::Matrix<IndexType>& N1, mtx::Matrix<DataType>& D1,
+                       mtx::Matrix<IndexType>& N2, mtx::Matrix<DataType>& D2,
+                       size_t k, size_t m,
+                       mtx::Matrix<IndexType>& N, mtx::Matrix<DataType>& D) {
+   size_t numQueries = N1.rows();
+   size_t maxCandidates = std::min((IndexType)m, (IndexType)(N1.columns() + N2.columns()));
+
+   for (size_t q = 0; q < numQueries; ++q) {
+      // Combine distances and neighbors
+      std::vector<std::pair<DataType, IndexType>> candidates(N1.columns() + N2.columns());
+
+      // Concatenate N1 and N2 rows
+      for (size_t i = 0; i < N1.columns(); ++i) {
+         candidates[i] = {D1.get(q, i), N1.get(q, i)};
+      }
+      for (size_t i = 0; i < N2.columns(); ++i) {
+         candidates[i + N1.columns()] = {D2.get(q, i), N2.get(q, i)};
+      }
+
+      // Keep only the top-m candidates
+      v0::quickselect(candidates, maxCandidates);
+
+      // Sort the top-m candidates
+      std::sort(candidates.begin(), candidates.begin() + maxCandidates);
+
+      // If m < k, pad the remaining slots with invalid values
+      for (size_t i = 0; i < k; ++i) {
+         if (i < maxCandidates) {
+            D.set(candidates[i].first, q, i);
+            N.set(candidates[i].second, q, i);
+         } else {
+            D.set(std::numeric_limits<DataType>::infinity(), q, i);
+            N.set(static_cast<IndexType>(-1), q, i); // Invalid index (end)
+         }
+      }
+    }
+}
+
+
+
+template<typename MatrixD, typename MatrixI>
+void knnsearch(const MatrixD& C, const MatrixD& Q, size_t idx_offset, size_t k, size_t m, MatrixI& idx, MatrixD& dst) {
+
+   using DstType = typename MatrixD::dataType;
+   using IdxType = typename MatrixI::dataType;
+
+    if (C.rows() <= 8 || Q.rows() <= 4) {
+        // Base case: Call knnsearch directly
+        v0::knnsearch(C, Q, idx_offset, k, m, idx, dst);
+        return;
+    }
+
+    // Divide Corpus and Query into subsets
+    IdxType midC = C.rows() / 2;
+    IdxType midQ = Q.rows() / 2;
+
+    // Slice corpus and query matrixes
+    MatrixD C1((DstType*)C.data(), 0,    midC, C.columns());
+    MatrixD C2((DstType*)C.data(), midC, midC, C.columns());
+    MatrixD Q1((DstType*)Q.data(), 0,    midQ, Q.columns());
+    MatrixD Q2((DstType*)Q.data(), midQ, midQ, Q.columns());
+
+    // Allocate temporary matrixes for all permutations
+    MatrixI N1_1(midQ, k), N1_2(midQ, k), N2_1(midQ, k), N2_2(midQ, k);
+    MatrixD D1_1(midQ, k), D1_2(midQ, k), D2_1(midQ, k), D2_2(midQ, k);
+
+    // Recursive calls
+    knnsearch(C1, Q1, idx_offset,        k, m, N1_1, D1_1);
+    knnsearch(C2, Q1, idx_offset + midC, k, m, N1_2, D1_2);
+    knnsearch(C1, Q2, idx_offset,        k, m, N2_1, D2_1);
+    knnsearch(C2, Q2, idx_offset + midC, k, m, N2_2, D2_2);
+
+    // slice output matrixes
+    MatrixI N1((IdxType*)idx.data(), 0,    midQ, k);
+    MatrixI N2((IdxType*)idx.data(), midQ, midQ, k);
+    MatrixD D1((DstType*)dst.data(), 0,    midQ, k);
+    MatrixD D2((DstType*)dst.data(), midQ, midQ, k);
+
+    // Merge results in place
+    mergeResultsWithM(N1_1, D1_1, N1_2, D1_2, k, m, N1, D1);
+    mergeResultsWithM(N2_1, D2_1, N2_2, D2_2, k, m, N2, D2);
+
+}
+
+} // namespace v1
 
 #endif /* V1_HPP_ */

homework_1/matlab/mergeReducedResultsWithMink.m

@@ -0,0 +1,11 @@
+function [N, D] = mergeReducedResultsWithMink(N_sub, D_sub, C_sub, k, ~)
+    % Merge reduced results for one subset of queries using mink
+    numQueries = size(N_sub, 1);  % Number of queries
+
+    % Find the k smallest distances and their indices using mink
+    [D, idx] = mink(D_sub, k, 2);
+
+    % Select the corresponding neighbors based on the indices
+    N = N_sub(sub2ind(size(N_sub), ...
+            repmat((1:numQueries)', 1, k), idx));
+end

homework_1/matlab/mergeResults.m

@@ -0,0 +1,19 @@
+function [N, D] = mergeResults(N1, D1, N2, D2, C1, C2, k, ~)
+    % Merge neighbors from two sources for a specific subset of queries
+    numQueries = size(N1, 1); % Queries corresponding to N1 (or N2)
+    N_combined = [N1, N2 + size(C1, 1)]; % Adjust indices for C2
+    D_combined = [D1, D2];
+    
+    % Sort distances and select top-k for each query
+    [D_sorted, idx] = sort(D_combined, 2); % Sort by distance for each query
+    idx = idx(:, 1:k); % Top-k indices
+
+    % Select corresponding neighbors
+    N = zeros(numQueries, k);
+    for i = 1:numQueries
+        N(i, :) = N_combined(i, idx(i, :));
+    end
+
+    % Select corresponding distances
+    D = D_sorted(:, 1:k);
+end

homework_1/matlab/mergeResultsWithM.m

@@ -0,0 +1,46 @@
+function [N, D] = mergeResultsWithM(N1, D1, N2, D2, C1, C2, k, m)
+    % Merge neighbors from two sources with a limit on candidate neighbors (m)
+    
+    numQueries = size(N1, 1);  % Number of queries in this subset
+    maxCandidates = min(m, size(N1, 2) + size(N2, 2));  % Maximum candidates to consider
+
+    % Combine distances and neighbors
+    N_combined = [N1, N2 + size(C1, 1)];  % Adjust indices for C2
+    D_combined = [D1, D2];
+
+    % Sort distances and keep only top-m candidates for each query
+    [D_sorted, idx] = sort(D_combined, 2, 'ascend');
+    D_sorted = D_sorted(:, 1:maxCandidates);  % Keep only top-m distances
+    idx = idx(:, 1:maxCandidates);  % Keep indices corresponding to top-m distances
+
+    % Select the corresponding neighbors
+    %N_sorted = N_combined(sub2ind(size(N_combined), ...
+    %                  repmat((1:numQueries)', 1, maxCandidates), idx));
+    N_sorted = zeros(numQueries, maxCandidates);  % Initialize output
+    for i = 1:numQueries
+        for j = 1:maxCandidates
+            N_sorted(i, j) = N_combined(i, idx(i, j));
+        end
+    end
+
+    % Handle cases where m < k
+    if maxCandidates < k
+        % Pad with Inf distances and invalid indices
+        D_sorted = [D_sorted, Inf(numQueries, k - maxCandidates)];
+        N_sorted = [N_sorted, zeros(numQueries, k - maxCandidates)];
+    end
+
+    % Extract top-k from the reduced set of candidates
+    [D, idx_final] = sort(D_sorted, 2, 'ascend');
+    D = D(:, 1:k);  % Final top-k distances
+    %N = N_sorted(sub2ind(size(N_sorted), ...
+    %                 repmat((1:numQueries)', 1, k), idx_final(:, 1:k)));
+    % Extract top-k neighbors using a loop
+    N = zeros(numQueries, k);  % Initialize output
+    for i = 1:numQueries
+        for j = 1:k
+            N(i, j) = N_sorted(i, idx_final(i, j));
+        end
+    end
+
+end

homework_1/matlab/recursiveKNN.m

@@ -0,0 +1,32 @@
+function [neighbors, distances] = recursiveKNN(C, Q, k, m)
+    % Recursive break, call knnsearch
+    if size(C, 1) <= 1000 || size(Q, 1) <= 500  % Adjastable limit
+        [neighbors, distances] = knnsearch(C, Q, 'K', k);
+        return;
+    end
+
+    % Divide into subsets
+    midC = floor(size(C, 1) / 2);
+    midQ = floor(size(Q, 1) / 2);
+    C1 = C(1:midC, :);
+    C2 = C(midC+1:end, :);
+    Q1 = Q(1:midQ, :);
+    Q2 = Q(midQ+1:end, :);
+
+    % Recursive calls
+    [N1_1, D1_1] = recursiveKNN(C1, Q1, k, m);
+    [N1_2, D1_2] = recursiveKNN(C2, Q1, k, m);
+    [N2_1, D2_1] = recursiveKNN(C1, Q2, k, m);
+    [N2_2, D2_2] = recursiveKNN(C2, Q2, k, m);
+
+    % Merge
+    [N1, D1] = mergeResultsWithM(N1_1, D1_1, N1_2, D1_2, C1, C2, k, m);
+    [N2, D2] = mergeResultsWithM(N2_1, D2_1, N2_2, D2_2, C1, C2, k, m);
+
+    % Combine results for Q1 and Q2
+    neighbors = [N1; N2];
+    distances = [D1; D2];
+
+end
+
+

homework_1/matlab/run_test.m

@@ -0,0 +1,23 @@
+%
+%
+%
+
+%C = rand(10000, 2);  % Corpus
+%Q = rand(10000, 2);   % Queries
+C = rand(20000, 2);  % Δύο clusters
+Q = C;
+%Q = rand(10000, 2);  % Queries κοντά στο μέσο
+k = 100;             % Number of neighbors
+m = 100;             % Max candidates per query
+global count;
+
+count =0;
+tic; [pi, pd] = recursiveKNN(C, Q, k, m); toc
+count
+tic; [si, sd] = knnsearch(C, Q, 'k', k); toc
+
+[a,b] = size(si);
+all_neighbors = a*b
+accuracy = 1 - nnz(si-pi)/all_neighbors
+false_neighbors = nnz(sd-pd)
+

homework_1/matlab/test.m

@@ -1,24 +0,0 @@
-%
-%
-%
-
-C = rand(40000,4);
-Q = rand(4000,4);
-
-disp ('C-Q');
-disp ('build-in')
-tic; [i1, d1] = knnsearch(C, Q, 'k', 4); toc
-disp ('mine')
-tic; [i2, d2] = knnsearch2(C, Q, 4); toc
-sum(i1-i2)
-sum(d1-d2)
-
-disp (' ');
-disp (' ');
-disp ('C-C');
-disp ('build-in')
-tic; [i1, d1] = knnsearch(C, C, 'k', 4); toc
-disp ('mine')
-tic; [i2, d2] = knnsearch2(C, C, 4); toc
-sum(i1-i2)
-sum(d1-d2)

homework_1/src/main.cpp

@@ -13,11 +13,12 @@
 #include <unistd.h>
 #include <cstdio>
 
-#include <v0.hpp>
-#include <v1.hpp>
-#include <matrix.hpp>
-#include <utils.hpp>
-#include <config.h>
+#include "matrix.hpp"
+
+#include "v0.hpp"
+#include "v1.hpp"
+#include "utils.hpp"
+#include "config.h"
 
 // Global session data
 session_t   session;
@@ -108,8 +109,7 @@ bool get_options(int argc, char* argv[]){
    return status;
 }
 
-NAMESPACE_VERSION;
-
+#ifndef TESTING
 int main(int argc, char* argv[]) try {
    // Instantiate matrixes
    MatrixDst Corpus;
@@ -117,6 +117,12 @@ int main(int argc, char* argv[]) try {
    MatrixIdx Idx;
    MatrixDst Dst;
 
+#if CODE_VERSION == V0
+   using namespace v0;
+#else
+   using namespace v1;
+#endif
+
    // try to read command line
    if (!get_options(argc, argv))
       exit(1);
@@ -132,17 +138,21 @@ int main(int argc, char* argv[]) try {
    timer.stop();
    timer.print_dt("Load hdf5 files");
 
+   // Prepare output memory
+   Idx.resize(Query.rows(), session.k);
+   Dst.resize(Query.rows(), session.k);
+
+   // Do the search
    logger << "Start knnsearch ...";
    timer.start();
    if (session.queryMtx)
-      knnsearch(Corpus, Query, session.k, Idx, Dst);
+      knnsearch(Corpus, Query, 0, session.k, session.k, Idx, Dst);
    else
-      knnsearch(Corpus, Corpus, session.k, Idx, Dst);
+      knnsearch(Corpus, Corpus, 0, session.k, session.k, Idx, Dst);
    timer.stop();
    logger << " Done" << logger.endl;
    timer.print_dt("knnsearch");
 
-
    // Store data
    timer.start();
    Mtx::store<MatrixIdx, IdxHDF5Type>(session.outMtxFile, session.outMtxIdxDataSet, Idx);
@@ -158,4 +168,17 @@ catch (std::exception& e) {
    exit(1);
 }
 
+#elif defined TESTING
+
+#include <gtest/gtest.h>
+#include <exception>
+
+GTEST_API_ int main(int argc, char **argv) try {
+   testing::InitGoogleTest(&argc, argv);
+   return RUN_ALL_TESTS();
+}
+catch (std::exception& e) {
+    std::cout << "Exception: " << e.what() << '\n';
+}
 
+#endif

homework_1/src/tests.cpp

@@ -0,0 +1,317 @@
+/**
+ * \file    tests.cpp
+ * \brief   PDS homework_1 tests
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#include <gtest/gtest.h>
+
+#include "matrix.hpp"
+
+#include "v0.hpp"
+#include "v1.hpp"
+#include "utils.hpp"
+#include "config.h"
+
+
+using matrix_t = mtx::Matrix<int>;
+
+
+// =====================================
+// C1, Q1
+mtx::Matrix<double> C1(10,2, {
+   0.8147,    0.1576,
+   0.9058,    0.9706,
+   0.1270,    0.9572,
+   0.9134,    0.4854,
+   0.6324,    0.8003,
+   0.0975,    0.1419,
+   0.2785,    0.4218,
+   0.5469,    0.9157,
+   0.9575,    0.7922,
+   0.9649,    0.9595
+});
+
+mtx::Matrix<double> Q1(5,2, {
+   0.6557,    0.7577,
+   0.0357,    0.7431,
+   0.8491,    0.3922,
+   0.9340,    0.6555,
+   0.6787,    0.1712
+});
+
+// =====================================
+// C2, Q2
+mtx::Matrix<double> C2(16,4, {
+   0.7060,    0.4456,    0.5060,    0.6160,
+   0.0318,    0.6463,    0.6991,    0.4733,
+   0.2769,    0.7094,    0.8909,    0.3517,
+   0.0462,    0.7547,    0.9593,    0.8308,
+   0.0971,    0.2760,    0.5472,    0.5853,
+   0.8235,    0.6797,    0.1386,    0.5497,
+   0.6948,    0.6551,    0.1493,    0.9172,
+   0.3171,    0.1626,    0.2575,    0.2858,
+   0.9502,    0.1190,    0.8407,    0.7572,
+   0.0344,    0.4984,    0.2543,    0.7537,
+   0.4387,    0.9597,    0.8143,    0.3804,
+   0.3816,    0.3404,    0.2435,    0.5678,
+   0.7655,    0.5853,    0.9293,    0.0759,
+   0.7952,    0.2238,    0.3500,    0.0540,
+   0.1869,    0.7513,    0.1966,    0.5308,
+   0.4898,    0.2551,    0.2511,    0.7792
+});
+
+mtx::Matrix<double> Q2(8,4, {
+   0.9340,    0.3112,    0.4505,    0.0782,
+   0.1299,    0.5285,    0.0838,    0.4427,
+   0.5688,    0.1656,    0.2290,    0.1067,
+   0.4694,    0.6020,    0.9133,    0.9619,
+   0.0119,    0.2630,    0.1524,    0.0046,
+   0.3371,    0.6541,    0.8258,    0.7749,
+   0.1622,    0.6892,    0.5383,    0.8173,
+   0.7943,    0.7482,    0.9961,    0.8687
+});
+
+
+
+/*
+ * ==========================================
+ * pdist2
+ */
+TEST(Tv0_UT, pdist2_test1) {
+
+   mtx::Matrix<double> D1_exp(10, 5, {
+      0.6208,    0.9745,    0.2371,    0.5120,    0.1367,
+      0.3284,    0.8993,    0.5811,    0.3164,    0.8310,
+      0.5651,    0.2327,    0.9169,    0.8616,    0.9603,
+      0.3749,    0.9147,    0.1132,    0.1713,    0.3921,
+      0.0485,    0.5994,    0.4621,    0.3346,    0.6308,
+      0.8312,    0.6044,    0.7922,    0.9815,    0.5819,
+      0.5052,    0.4028,    0.5714,    0.6959,    0.4722,
+      0.1919,    0.5395,    0.6045,    0.4665,    0.7561,
+      0.3037,    0.9231,    0.4144,    0.1387,    0.6807,
+      0.3692,    0.9540,    0.5790,    0.3056,    0.8386
+   });
+
+   mtx::Matrix<double> D (10,5);
+
+   v0::pdist2(C1, Q1, D);
+
+   for (size_t i = 0 ; i< D.rows() ; ++i)
+      for (size_t j = 0 ; j<D.columns() ; ++j) {
+         EXPECT_EQ (D1_exp.get(i ,j) + 0.01 > D(i, j), true);
+         EXPECT_EQ (D1_exp.get(i ,j) - 0.01 < D(i, j), true);
+      }
+}
+
+TEST(Tv0_UT, pdist2_test2) {
+
+   mtx::Matrix<double> D2_exp(16, 8, {
+      0.6020,    0.7396,    0.6583,    0.6050,    1.0070,    0.5542,    0.6298,    0.6352,
+      1.0696,    0.6348,    0.9353,    0.6914,    0.8160,    0.4475,    0.4037,    0.9145,
+      0.9268,    0.8450,    0.9376,    0.6492,    0.9671,    0.4360,    0.5956,    0.7400,
+      1.3455,    0.9876,    1.2953,    0.4709,    1.2557,    0.3402,    0.4417,    0.7500,
+      0.9839,    0.5476,    0.7517,    0.7216,    0.7074,    0.5605,    0.4784,    0.9954,
+      0.6839,    0.7200,    0.7305,    0.9495,    1.0628,    0.8718,    0.8178,    0.9179,
+      0.9850,    0.7514,    0.9585,    0.7996,    1.2054,    0.7784,    0.6680,    0.8591,
+      0.6950,    0.4730,    0.3103,    1.0504,    0.4397,    0.8967,    0.8140,    1.2066,
+      0.8065,    1.2298,    0.9722,    0.7153,    1.3933,    0.8141,    1.0204,    0.6758,
+      1.1572,    0.3686,    0.9031,    0.8232,    0.7921,    0.6656,    0.3708,    1.0970,
+      0.9432,    0.9049,    1.0320,    0.6905,    1.1167,    0.5094,    0.6455,    0.6653,
+      0.7672,    0.3740,    0.5277,    0.8247,    0.6842,    0.6945,    0.5648,    0.9968,
+      0.5768,    1.1210,    0.8403,    0.9345,    1.1316,    0.8292,    1.0380,    0.8127,
+      0.1939,    0.8703,    0.2684,    1.1794,    0.8103,    1.0683,    1.1115,    1.1646,
+      1.0106,    0.2708,    0.8184,    0.8954,    0.7402,    0.6982,    0.4509,    1.0594,
+      0.8554,    0.5878,    0.6834,    0.7699,    0.9155,    0.7161,    0.6162,    0.9481
+   });
+
+   mtx::Matrix<double> D (16,8);
+
+   v0::pdist2(C2, Q2, D);
+
+   for (size_t i = 0 ; i< D.rows() ; ++i)
+      for (size_t j = 0 ; j<D.columns() ; ++j) {
+         EXPECT_EQ (D2_exp.get(i ,j) + 0.01 > D(i, j), true);
+         EXPECT_EQ (D2_exp.get(i ,j) - 0.01 < D(i, j), true);
+      }
+}
+
+
+/*
+ * ==========================================
+ * v0::knn
+ */
+TEST(Tv0_UT, knn_test1) {
+   size_t k = 3;
+   mtx::Matrix<uint32_t> Idx_exp(5, k, {
+      5,     8,     9,
+      3,     7,     8,
+      4,     1,     9,
+      9,     4,     10,
+      1,     4,     7
+   });
+
+   mtx::Matrix<double> Dst_exp(5, k, {
+      0.0485,    0.1919,    0.3037,
+      0.2327,    0.4028,    0.5395,
+      0.1132,    0.2371,    0.4144,
+      0.1387,    0.1713,    0.3056,
+      0.1367,    0.3921,    0.4722
+   });
+
+   mtx::Matrix<uint32_t> Idx(5, k);
+   mtx::Matrix<double>   Dst(5, k);
+
+   v0::knnsearch(C1, Q1, 0, k, 0, Idx, Dst);
+
+
+   for (size_t i = 0 ; i< Idx.rows() ; ++i)
+      for (size_t j = 0 ; j<Idx.columns() ; ++j) {
+         EXPECT_EQ (Idx_exp(i ,j) == Idx(i, j) + 1, true);  // matlab starts from 1
+         EXPECT_EQ (Dst_exp.get(i ,j) + 0.01 > Dst(i, j), true);
+         EXPECT_EQ (Dst_exp.get(i ,j) - 0.01 < Dst(i, j), true);
+      }
+
+}
+
+TEST(Tv0_UT, knn_test2) {
+   size_t k = 3;
+   mtx::Matrix<uint32_t> Idx_exp(8, k, {
+      14,    13,     1,
+      15,    10,    12,
+      14,     8,    12,
+       4,     1,     3,
+       8,    12,     5,
+       4,     3,     2,
+      10,     2,     4,
+       1,    11,     9
+   });
+
+   mtx::Matrix<double> Dst_exp(8, k, {
+      0.1939,    0.5768,    0.6020,
+      0.2708,    0.3686,    0.3740,
+      0.2684,    0.3103,    0.5277,
+      0.4709,    0.6050,    0.6492,
+      0.4397,    0.6842,    0.7074,
+      0.3402,    0.4360,    0.4475,
+      0.3708,    0.4037,    0.4417,
+      0.6352,    0.6653,    0.6758
+   });
+
+   mtx::Matrix<uint32_t> Idx(8, k);
+   mtx::Matrix<double>   Dst(8, k);
+
+   v0::knnsearch(C2, Q2, 0, k, 0, Idx, Dst);
+
+
+   for (size_t i = 0 ; i< Idx.rows() ; ++i)
+      for (size_t j = 0 ; j<Idx.columns() ; ++j) {
+         EXPECT_EQ (Idx_exp(i ,j) == Idx(i, j) + 1, true);  // matlab starts from 1
+         EXPECT_EQ (Dst_exp.get(i ,j) + 0.01 > Dst(i, j), true);
+         EXPECT_EQ (Dst_exp.get(i ,j) - 0.01 < Dst(i, j), true);
+      }
+
+}
+
+/*
+ * ==========================================
+ * v1::knn
+ */
+TEST(Tv1_UT, knn_test1) {
+   size_t k = 3;
+   mtx::Matrix<uint32_t> Idx_exp(8, k, {
+      14,    13,     1,
+      15,    10,    12,
+      14,     8,    12,
+       4,     1,     3,
+       8,    12,     5,
+       4,     3,     2,
+      10,     2,     4,
+       1,    11,     9
+   });
+
+   mtx::Matrix<double> Dst_exp(8, k, {
+      0.1939,    0.5768,    0.6020,
+      0.2708,    0.3686,    0.3740,
+      0.2684,    0.3103,    0.5277,
+      0.4709,    0.6050,    0.6492,
+      0.4397,    0.6842,    0.7074,
+      0.3402,    0.4360,    0.4475,
+      0.3708,    0.4037,    0.4417,
+      0.6352,    0.6653,    0.6758
+   });
+
+   mtx::Matrix<uint32_t> Idx(8, k);
+   mtx::Matrix<double>   Dst(8, k);
+
+   v1::knnsearch(C2, Q2, 0, k, k, Idx, Dst);
+
+
+   for (size_t i = 0 ; i< Idx.rows() ; ++i)
+      for (size_t j = 0 ; j<Idx.columns() ; ++j) {
+         EXPECT_EQ (Idx_exp(i ,j) == Idx(i, j) + 1, true);  // matlab starts from 1
+         EXPECT_EQ (Dst_exp.get(i ,j) + 0.01 > Dst(i, j), true);
+         EXPECT_EQ (Dst_exp.get(i ,j) - 0.01 < Dst(i, j), true);
+      }
+
+}
+
+// all-to-all
+TEST(Tv1_UT, knn_test2) {
+   size_t k = 3;
+   mtx::Matrix<uint32_t> Idx_exp(16, k, {
+      1,    16,    12,
+      2,     3,     5,
+      3,    11,     2,
+      4,     2,     3,
+      5,    10,     2,
+      6,     7,     1,
+      7,     6,    16,
+      8,    12,     5,
+      9,     1,    16,
+     10,    15,     5,
+     11,     3,     2,
+     12,    16,     8,
+     13,     3,    11,
+     14,     8,     1,
+     15,    10,    12,
+     16,    12,     1
+   });
+
+   mtx::Matrix<double> Dst_exp(16, k, {
+      0,    0.4179,    0.4331,
+      0,    0.3401,    0.4207,
+      0,    0.3092,    0.3401,
+      0,    0.4555,    0.5381,
+      0,    0.4093,    0.4207,
+      0,    0.3903,    0.4560,
+      0,    0.3903,    0.4811,
+      0,    0.3398,    0.4846,
+      0,    0.5461,    0.7607,
+      0,    0.3745,    0.4093,
+      0,    0.3092,    0.5345,
+      0,    0.2524,    0.3398,
+      0,    0.5759,    0.5941,
+      0,    0.5428,    0.6304,
+      0,    0.3745,    0.4586,
+      0,    0.2524,    0.4179
+   });
+
+   mtx::Matrix<uint32_t> Idx(16, k);
+   mtx::Matrix<double>   Dst(16, k);
+
+   v1::knnsearch(C2, C2, 0, k, k, Idx, Dst);
+
+
+   for (size_t i = 0 ; i< Idx.rows() ; ++i)
+      for (size_t j = 0 ; j<Idx.columns() ; ++j) {
+         EXPECT_EQ (Idx_exp(i ,j) == Idx(i, j) + 1, true);  // matlab starts from 1
+         EXPECT_EQ (Dst_exp.get(i ,j) + 0.01 > Dst(i, j), true);
+         EXPECT_EQ (Dst_exp.get(i ,j) - 0.01 < Dst(i, j), true);
+      }
+
+}
+