HW2: A local only version of the distbitonic

homework_2/Makefile

@@ -20,12 +20,17 @@
 # ============== Project settings ==============
 # Project's name
 PROJECT         := PDS_homework_2
+
 # Excecutable's name
-TARGET          := distbitonic
+TARGET          := bitonic
+
 # Source directories list(space seperated). Makefile-relative path, UNDER current directory.
-SRC_DIR_LIST    := src
+SRC_DIR_LIST    := src test test/gtest
+
 # Include directories list(space seperated). Makefile-relative path.
 INC_DIR_LIST    := include \
+				   test \
+				   test/gtest/ \
 				   /usr/lib/x86_64-linux-gnu/openmpi/include/ \
                    src
 
@@ -148,6 +153,14 @@ $(BUILD_DIR)/$(TARGET): $(OBJ)
 	@$(CSIZE) $(@D)/$(TARGET)
 	@echo Done
 
+
+#
+# ================ Default local build rules =================
+# example:
+# make debug
+
+.DEFAULT_GOAL := all
+
 .PHONY: clean
 clean:
 	@echo Cleaning build directories
@@ -155,67 +168,53 @@ clean:
 	@rm -rf $(DEP_DIR)
 	@rm -rf $(BUILD_DIR)
 
-#
-# ================ Local build rules =================
-# example:
-# make debug
-
 debug: CFLAGS := $(DEB_CFLAGS)
 debug: $(BUILD_DIR)/$(TARGET)
 
 release: CFLAGS := $(REL_CFLAGS)
 release: $(BUILD_DIR)/$(TARGET)
 
-all: release
-
-hpc-results/post:
-	$(CXX) $(CFLAGS) -o $@ hpc-results/main.cpp 
-
-hpc-clean:
-	rm hpc-results/post
-
 #
-# ================ Local (and/or) via docker build rules =================
+# ================ Build rules =================
-#
-# examples:
-# make IMAGE=hpcimage v0
-# make IMAGE=hpcimage v1_cilk
-#
-dist_v05: CC := mpicc
-dist_v05: CXX := mpic++
-dist_v05: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=50
-dist_v05: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=50
-dist_v05: TARGET := dist_v05
-dist_v05: $(BUILD_DIR)/$(TARGET)
-
-dist_v1: CC := mpicc
-dist_v1: CXX := mpic++
-dist_v1: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=100
-dist_v1: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=100
-dist_v1: TARGET := dist_v1
-dist_v1: $(BUILD_DIR)/$(TARGET)
-
-#
-# ========= Inside CSAL Image build rules ===========
-#
-# 1) first jump into image (make sure you are in the directory where Makefile is):
-#    > docker run -it -v ${PWD}:/usr/src/exercise_1 -w /usr/src/exercise_1/ hpcimage
-# 2) Clean binaries first **important**
-#    > make clean
-# 3) for v4 cilk for example:
-#    > make csal_v4_cilk
-# 4) run executables from `bin/`. Examples:
-#    > ./bin/tcount_ompv3 -i mtx/NACA0015.mtx --timing -r 3 -o /dev/null
-#    > ./bin/tcount_pthv4 -i mtx/com_Youtube.mtx --timing --dynamic --print_count
-
-
-
-#
-# ======== Run from container =========
-#
-# examples:
-#
-# make IMAGE=hpcimage EXEC=knnsearch_v1 run
-# make IMAGE=hpcimage EXEC=knnsearch_v1 run
 #
 
+# Local or inside HPC rules
+distbubbletonic: CC := mpicc
+distbubbletonic: CXX := mpic++
+distbubbletonic: CFLAGS := $(REL_CFLAGS) -DCODE_VERSION=BUBBLETONIC
+distbubbletonic: CXXFLAGS := $(REL_CXXFLAGS) -DCODE_VERSION=BUBBLETONIC
+distbubbletonic: TARGET := distbubbletonic
+distbubbletonic: $(BUILD_DIR)/$(TARGET)
+
+distbitonic: CC := mpicc
+distbitonic: CXX := mpic++
+distbitonic: CFLAGS := $(REL_CFLAGS) -DCODE_VERSION=BITONIC
+distbitonic: CXXFLAGS := $(REL_CXXFLAGS) -DCODE_VERSION=BITONIC
+distbitonic: TARGET := distbitonic
+distbitonic: $(BUILD_DIR)/$(TARGET)
+
+deb_distbubbletonic: CC := mpicc
+deb_distbubbletonic: CXX := mpic++
+deb_distbubbletonic: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=BUBBLETONIC -DDEBUG
+deb_distbubbletonic: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=BUBBLETONIC -DDEBUG
+deb_distbubbletonic: TARGET := deb_distbubbletonic
+deb_distbubbletonic: $(BUILD_DIR)/$(TARGET)
+
+deb_distbitonic: CC := mpicc
+deb_distbitonic: CXX := mpic++
+deb_distbitonic: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=BITONIC -DDEBUG
+deb_distbitonic: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=BITONIC -DDEBUG
+deb_distbitonic: TARGET := deb_distbitonic
+deb_distbitonic: $(BUILD_DIR)/$(TARGET)
+
+tests: CFLAGS := $(DEB_CFLAGS) -DCODE_VERSION=BITONIC -DDEBUG -DTESTING
+tests: CXXFLAGS := $(DEB_CXXFLAGS) -DCODE_VERSION=BITONIC -DDEBUG -DTESTING
+tests: TARGET := tests
+tests: $(BUILD_DIR)/$(TARGET)
+
+
+all: debug distbubbletonic distbitonic
+# Note:
+#	Add a gcc based make rule here in order for clangd to successfully scan the project files.
+#	Otherwise we do not need the gcc build.
+

homework_2/include/config.h

@@ -1,5 +1,5 @@
 /*!
- * \file    config,h
+ * \file    config.h
  * \brief   Build configuration file.
  *
  * \author
@@ -16,13 +16,13 @@
 /*
  * Defines for different version of the exercise
  */
-#define  V50     (50)
+#define  BITONIC        (1)
-#define  V100    (100)
+#define  BUBBLETONIC    (2)
 
 
 // Fail-safe version selection
 #if !defined CODE_VERSION
-#define CODE_VERSION   V1
+#define CODE_VERSION   BITONIC
 #endif
 
 
@@ -36,4 +36,5 @@ struct session_t {
 
 extern session_t session;
 
+
 #endif /* CONFIG_H_ */

homework_2/include/distbitonic.hpp

@@ -0,0 +1,44 @@
+/*!
+ * \file
+ * \brief   Distributed bitonic implementation header
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#ifndef DISTBITONIC_H_
+#define DISTBITONIC_H_
+
+#if !defined DEBUG
+#define NDEBUG
+#endif
+#include <cassert>
+
+#include <vector>
+#if !defined TESTING
+  #include <mpi.h>
+#endif
+
+using Data_t    = std::vector<uint8_t>;
+using AllData_t = std::vector<Data_t>;
+
+struct mpi_t {
+    size_t world_size{};
+    size_t world_rank{};
+    std::string processor_name {};
+};
+
+extern mpi_t mpi;
+
+
+bool ascending(size_t node, size_t depth) noexcept;
+size_t partner(size_t node, size_t step) noexcept;
+bool keepsmall(size_t node, size_t partner, size_t depth) noexcept;
+
+void exchange(size_t node, size_t partner);
+void minmax(AllData_t& data, size_t node, size_t partner, bool keepsmall);
+void sort_network(AllData_t& data, size_t nodes, size_t depth);
+void distbitonic(size_t P, AllData_t& data);
+
+#endif //DISTBITONIC_H_

homework_2/include/impl.hpp

@@ -0,0 +1,14 @@
+/*!
+ * \file
+ * \brief   The distributed bitonic implementation header
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#ifndef IMPL_H_
+#define IMPL_H_
+
+
+#endif //IMPL_H_

homework_2/include/utils.hpp

@@ -19,8 +19,10 @@
 /*!
  * A Logger for entire program.
  */
+
 struct Log {
-   struct Endl {} endl;    //!< a tag object to to use it as a new line request.
+    struct Endl {
+    } endl;    //!< a tag object to to use it as a new line request.
 
     //! We provide logging via << operator
     template<typename T>
@@ -29,20 +31,22 @@ struct Log {
             if (line_) {
                 std::cout << "[Log]: " << t;
                 line_ = false;
-         }
+            } else
-         else
                 std::cout << t;
         }
         return *this;
     }
+
     // overload for special end line handling
-   Log& operator<< (Endl e) { (void)e;
+    Log &operator<<(Endl e) {
+        (void) e;
         if (session.verbose) {
             std::cout << '\n';
             line_ = true;
         }
         return *this;
     }
+
 private:
     bool line_{true};
 };
@@ -60,28 +64,33 @@ struct Timing{
 
     //! tool to mark the starting point
     Tpoint start() noexcept { return start_ = std::chrono::steady_clock::now(); }
+
     //! tool to mark the ending point
     Tpoint stop() noexcept { return stop_ = std::chrono::steady_clock::now(); }
 
     auto dt() noexcept {
         return std::chrono::duration_cast<std::chrono::microseconds>(stop_ - start_).count();
     }
+
     //! tool to print the time interval
     void print_dt(const char *what) noexcept {
         if (session.timing) {
             auto t = stop_ - start_;
             if (std::chrono::duration_cast<microseconds>(t).count() < 10000)
-            std::cout << "[Timing]: " << what << ": " << std::to_string(std::chrono::duration_cast<microseconds>(t).count()) << " [usec]\n";
+                std::cout << "[Timing]: " << what << ": "
+                          << std::to_string(std::chrono::duration_cast<microseconds>(t).count()) << " [usec]\n";
             else if (std::chrono::duration_cast<milliseconds>(t).count() < 10000)
-            std::cout << "[Timing]: " << what << ": " << std::to_string(std::chrono::duration_cast<milliseconds>(t).count()) << " [msec]\n";
+                std::cout << "[Timing]: " << what << ": "
+                          << std::to_string(std::chrono::duration_cast<milliseconds>(t).count()) << " [msec]\n";
             else
-            std::cout << "[Timing]: " << what << ": " << std::to_string(std::chrono::duration_cast<seconds>(t).count()) << " [sec]\n";
+                std::cout << "[Timing]: " << what << ": "
+                          << std::to_string(std::chrono::duration_cast<seconds>(t).count()) << " [sec]\n";
         }
     }
+
 private:
     Tpoint start_;
     Tpoint stop_;
 };
 
-
 #endif /* UTILS_HPP_ */

homework_2/src/distbitonic.cpp

@@ -0,0 +1,106 @@
+/*!
+ * \file
+ * \brief   Distributed bitonic implementation.
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#include <vector>
+#include <algorithm>
+#include <cmath>
+#include <cassert>
+#include "distbitonic.hpp"
+
+/*!
+ * Returns the ascending or descending configuration of the node's sequence based on
+ * the current node (MPI process) and the depth of the sorting network
+ *
+ * @param node      The current node (MPI process)
+ * @param depth     The total depth of the sorting network (same for each step for a given network)
+ *
+ * @return          True if we need ascending configuration, false otherwise
+ */
+bool ascending(size_t node, size_t depth) noexcept {
+    return !(node & (1 << depth));
+}
+
+/*!
+ * Returns the node's partner for data exchange during the sorting network iterations
+ *
+ * @param node      The current node
+ * @param step      The step of the sorting network
+ * @return          The node id of the partner for data exchange
+ */
+size_t partner(size_t node, size_t step) noexcept {
+    return (node ^ (1 << step));
+}
+
+/*!
+ * Predicate to check if a node keeps the small numbers during the bitonic
+ * sort network exchange.
+ *
+ * @param node      The node for which we check
+ * @param partner   The partner of the data exchange
+ * @param depth     The total depth of the sorting network (same for each step for a given network)
+ * @return          True if the node should keep the small values, false otherwise
+ */
+bool keepsmall(size_t node, size_t partner, size_t depth) noexcept {
+    assert(node != partner);
+    return ascending(node, depth) == (node < partner);
+}
+
+void exchange(size_t node, size_t partner) {
+    assert(node != partner);
+
+}
+
+void minmax(AllData_t& data, size_t node, size_t partner, bool keepsmall) {
+    for (size_t i = 0; i < data[node].size(); ++i) {
+        if (keepsmall && data[node][i] > data[partner][i])
+            std::swap(data[node][i], data[partner][i]);
+        if (!keepsmall && data[node][i] < data[partner][i])
+            std::swap(data[node][i], data[partner][i]);
+    }
+}
+
+void sort_network(AllData_t& data, size_t nodes, size_t depth) {
+    for (size_t step = depth; step > 0;) {
+        --step;
+        for (size_t node = 0; node < nodes; ++node) {
+            auto part = partner(node, step);
+            auto ks = keepsmall(node, part, depth);
+            if (node < part) {
+                exchange(node, part);
+                minmax(data, node, part, ks);
+            }
+        }
+    }
+}
+
+void distbitonic(size_t P, AllData_t& data) {
+    auto p = static_cast<uint32_t>(std::log2(P));
+
+    for (size_t node = 0 ; node < P ; ++node) { // Currently we do all nodes here!
+        // Initially sort to create the half part of a bitonic
+        if (ascending(node, 0))
+            std::sort(data[node].begin(), data[node].end(), std::less<>());
+        else
+            std::sort(data[node].begin(), data[node].end(), std::greater<>());
+    }
+
+    // Run through sort network using elbow-sort
+    for (size_t depth = 1; depth <= p; ++depth) {
+        sort_network(data, P, depth);
+
+        for (size_t node = 0 ; node < P ; ++node) { // Currently we do all nodes here!
+            // elbow-sort here
+            if (ascending(node, depth))
+                std::sort(data[node].begin(), data[node].end(), std::less<>());
+            else
+                std::sort(data[node].begin(), data[node].end(), std::greater<>());
+        }
+
+    }
+}

homework_2/src/main.cpp

@@ -1,22 +1,43 @@
 /*!
- * \file    main.cpp
+ * \file
  * \brief   Main application file for PDS HW2 (MPI)
  *
  * \author
  *    Christos Choutouridis AEM:8997
  *    <cchoutou@ece.auth.gr>
  */
+
 #include <exception>
 #include <iostream>
-
+#include <algorithm>  // rand/srand
+//#include <ctime>      // rand/srand
+#if !defined TESTING
   #include <mpi.h>
+#endif
 
-#include "matrix.hpp"
+#include "distbitonic.hpp"
 #include "utils.hpp"
 #include "config.h"
+//#include "matrix.hpp"
 
 // Global session data
 session_t  session;
+mpi_t      mpi;
+
+/*
+ * Sorting data for up to 8 processes
+ */
+AllData_t Data {
+        Data_t (8),
+        Data_t (8),
+        Data_t (8),
+        Data_t (8),
+        Data_t (8),
+        Data_t (8),
+        Data_t (8),
+        Data_t (8)
+};
+
 
 /*!
  * A small command line argument parser
@@ -64,6 +85,9 @@ bool get_options(int argc, char* argv[]){
     return status;
 }
 
+
+
+#if !defined TESTING
 int main(int argc, char* argv[]) try {
     // try to read command line
     if (!get_options(argc, argv))
@@ -72,22 +96,49 @@ int main(int argc, char* argv[]) try {
     // Initialize the MPI environment
     MPI_Init(NULL, NULL);
 
+    #if defined DEBUG
+        /*
+         * In case of a debug build we will wait here until sleep_wait
+         * will reset via debugger. In order to do that the user must attach
+         * debugger to all processes. For example:
+         *  $> mpirun -np 2 ./<program path>
+         *  $> ps aux | grep <program>
+         *  $> gdb <program> <PID1>
+         *  $> gdb <program> <PID2>
+         */
+    #if defined TESTING
+        volatile bool sleep_wait = false;
+    #else
+        volatile bool sleep_wait = true;
+    #endif
+        while (sleep_wait)
+            sleep(1);
+    #endif
+
     // Get the number of processes
-    int world_size;
+    MPI_Comm_size(MPI_COMM_WORLD, reinterpret_cast<int *>(&mpi.world_size));
-    MPI_Comm_size(MPI_COMM_WORLD, &world_size);
 
     // Get the rank of the process
-    int world_rank;
+    MPI_Comm_rank(MPI_COMM_WORLD, reinterpret_cast<int *>(&mpi.world_rank));
-    MPI_Comm_rank(MPI_COMM_WORLD, &world_rank);
 
     // Get the name of the processor
     char processor_name[MPI_MAX_PROCESSOR_NAME];
     int name_len;
     MPI_Get_processor_name(processor_name, &name_len);
+    mpi.processor_name = std::string (processor_name, name_len);
 
     // Print off a hello world message
-    printf("Hello world from processor %s, rank %d out of %d processors\n",
+    std::cout   << "Hello world from processor: " << mpi.processor_name
-           processor_name, world_rank, world_size);
+                << " rank " << mpi.world_rank
+                << " out of " << mpi.world_size << " processors\n";
+
+//    std::srand(unsigned(std::time(nullptr)));
+//    for (auto& v : Data) {
+//        std::generate(v.begin(), v.end(), std::rand);
+//    }
+//
+//    distbitonic (2, Data);
+//    distbitonic (4, Data);
 
     // Finalize the MPI environment.
     MPI_Finalize();
@@ -98,3 +149,18 @@ catch (std::exception& e) {
     std::cerr << "Error: " << e.what() << '\n';
     exit(1);
 }
+
+#else
+
+#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_2/test/tests.cpp

@@ -0,0 +1,384 @@
+/**
+ * \file
+ * \brief   PDS HW2 tests
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#include <gtest/gtest.h>
+
+#include <algorithm>  // rand/srand
+#include <ctime>      // rand/srand
+#include "distbitonic.hpp"
+
+
+/*
+ * ==========================================
+ */
+
+/*
+ * bool ascending(size_t node, size_t depth);
+ * depth 0 (the initial ascending pattern)
+ */
+TEST(Tdistbitonic_UT, ascending_test1) {
+
+    EXPECT_EQ(ascending(0, 0), true);
+    EXPECT_EQ(ascending(1, 0), false);
+    EXPECT_EQ(ascending(2, 0), true);
+    EXPECT_EQ(ascending(3, 0), false);
+    EXPECT_EQ(ascending(4, 0), true);
+    EXPECT_EQ(ascending(5, 0), false);
+    EXPECT_EQ(ascending(6, 0), true);
+    EXPECT_EQ(ascending(7, 0), false);
+
+    for (size_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(ascending(node, 0), ((node % 2) ? false : true) );
+    }
+}
+
+/*
+ * bool ascending(size_t node, size_t depth);
+ * depth 1
+ */
+TEST(Tdistbitonic_UT, ascending_test2) {
+
+    EXPECT_EQ(ascending(0, 1), true);
+    EXPECT_EQ(ascending(1, 1), true);
+    EXPECT_EQ(ascending(2, 1), false);
+    EXPECT_EQ(ascending(3, 1), false);
+    EXPECT_EQ(ascending(4, 1), true);
+    EXPECT_EQ(ascending(5, 1), true);
+    EXPECT_EQ(ascending(6, 1), false);
+    EXPECT_EQ(ascending(7, 1), false);
+
+    for (size_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(ascending(2*node,   1), ((node % 2) ? false:true));
+        EXPECT_EQ(ascending(2*node+1, 1), ((node % 2) ? false:true));
+    }
+}
+
+/*
+ * bool ascending(size_t node, size_t depth);
+ * various depths
+ */
+TEST(Tdistbitonic_UT, ascending_test3) {
+
+    // Depth = 3
+    size_t ts_depth = 3;
+
+    for (size_t n = 0UL ; n < (1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), true);
+    for (size_t n = (1UL<<(ts_depth)) ; n < 2*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), false);
+    for (size_t n = 2*(1UL<<(ts_depth)) ; n < 3*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), true);
+    for (size_t n = 3*(1UL<<(ts_depth)) ; n < 4*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), false);
+
+    // Depth = 4
+    ts_depth = 4;
+
+    for (size_t n = 0UL ; n < (1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), true);
+    for (size_t n = (1UL<<(ts_depth)) ; n < 2*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), false);
+    for (size_t n = 2*(1UL<<(ts_depth)) ; n < 3*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), true);
+    for (size_t n = 3*(1UL<<(ts_depth)) ; n < 4*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), false);
+
+    // Depth = 8
+    ts_depth = 8;
+
+    for (size_t n = 0UL ; n < (1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), true);
+    for (size_t n = (1UL<<(ts_depth)) ; n < 2*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), false);
+    for (size_t n = 2*(1UL<<(ts_depth)) ; n < 3*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), true);
+    for (size_t n = 3*(1UL<<(ts_depth)) ; n < 4*(1UL<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending(n,   ts_depth), false);
+}
+
+/*
+ * size_t partner(size_t node, size_t step);
+ * step = 0
+ */
+TEST(Tdistbitonic_UT, partner_test1) {
+
+    EXPECT_EQ(partner(0, 0), 1UL);
+    EXPECT_EQ(partner(1, 0), 0UL);
+    EXPECT_EQ(partner(2, 0), 3UL);
+    EXPECT_EQ(partner(3, 0), 2UL);
+    EXPECT_EQ(partner(4, 0), 5UL);
+    EXPECT_EQ(partner(5, 0), 4UL);
+    EXPECT_EQ(partner(6, 0), 7UL);
+    EXPECT_EQ(partner(7, 0), 6UL);
+
+    for (size_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(partner(node, 0), (node % 2) ? node-1 : node+1);
+    }
+}
+
+/*
+ * size_t partner(size_t node, size_t step);
+ * step = 1
+ */
+TEST(Tdistbitonic_UT, partner_test2) {
+
+    EXPECT_EQ(partner(0, 1), 2UL);
+    EXPECT_EQ(partner(1, 1), 3UL);
+    EXPECT_EQ(partner(2, 1), 0UL);
+    EXPECT_EQ(partner(3, 1), 1UL);
+    EXPECT_EQ(partner(4, 1), 6UL);
+    EXPECT_EQ(partner(5, 1), 7UL);
+    EXPECT_EQ(partner(6, 1), 4UL);
+    EXPECT_EQ(partner(7, 1), 5UL);
+
+    for (size_t n1 = 0 ; n1 < 256 ; n1 += 2) {
+        auto n2 = n1 + 1UL;
+        EXPECT_EQ(partner(n1, 1), ((n1 % 4) ? n1-2 : n1+2));
+        EXPECT_EQ(partner(n2, 1), ((n1 % 4) ? n2-2 : n2+2));
+    }
+}
+
+/*
+ * size_t partner(size_t node, size_t step);
+ * various steps
+ */
+TEST(Tdistbitonic_UT, partner_test3) {
+    // step = 2
+    size_t ts_step = 2;
+
+    for (size_t n1 = 0 ; n1 < 256 ; n1 += 4) {
+        auto n2 = n1 + 1UL;
+        auto n3 = n1 + 2UL;
+        auto n4 = n1 + 3UL;
+        EXPECT_EQ(partner(n1, ts_step), ((n1 % 8) ? n1-4 : n1+4));
+        EXPECT_EQ(partner(n2, ts_step), ((n1 % 8) ? n2-4 : n2+4));
+        EXPECT_EQ(partner(n3, ts_step), ((n1 % 8) ? n3-4 : n3+4));
+        EXPECT_EQ(partner(n4, ts_step), ((n1 % 8) ? n4-4 : n4+4));
+    }
+
+    // step = 3
+    ts_step = 3;
+
+    for (size_t n1 = 0 ; n1 < 256 ; n1 += 8) {
+        auto n2 = n1 + 1UL;
+        auto n3 = n1 + 2UL;
+        auto n4 = n1 + 3UL;
+        auto n5 = n1 + 4UL;
+        auto n6 = n1 + 5UL;
+        auto n7 = n1 + 6UL;
+        auto n8 = n1 + 7UL;
+        EXPECT_EQ(partner(n1, ts_step), ((n1 % 16) ? n1-8 : n1+8));
+        EXPECT_EQ(partner(n2, ts_step), ((n1 % 16) ? n2-8 : n2+8));
+        EXPECT_EQ(partner(n3, ts_step), ((n1 % 16) ? n3-8 : n3+8));
+        EXPECT_EQ(partner(n4, ts_step), ((n1 % 16) ? n4-8 : n4+8));
+        EXPECT_EQ(partner(n5, ts_step), ((n1 % 16) ? n5-8 : n5+8));
+        EXPECT_EQ(partner(n6, ts_step), ((n1 % 16) ? n6-8 : n6+8));
+        EXPECT_EQ(partner(n7, ts_step), ((n1 % 16) ? n7-8 : n7+8));
+        EXPECT_EQ(partner(n8, ts_step), ((n1 % 16) ? n8-8 : n8+8));
+    }
+
+    // step = 4
+    ts_step = 4;
+
+    for (size_t n1 = 0 ; n1 < 256 ; n1 += 16) {
+        auto n2 = n1 + 1UL;
+        auto n3 = n1 + 2UL;
+        auto n4 = n1 + 3UL;
+        auto n5 = n1 + 4UL;
+        auto n6 = n1 + 5UL;
+        auto n7 = n1 + 6UL;
+        auto n8 = n1 + 7UL;
+        auto n9 = n1 + 8UL;
+        auto n10 = n1 + 9UL;
+        auto n11 = n1 + 10UL;
+        auto n12 = n1 + 11UL;
+        auto n13 = n1 + 12UL;
+        auto n14 = n1 + 13UL;
+        auto n15 = n1 + 14UL;
+        auto n16 = n1 + 15UL;
+        EXPECT_EQ(partner(n1,  ts_step), ((n1 % 32) ? n1-16 : n1+16));
+        EXPECT_EQ(partner(n2,  ts_step), ((n1 % 32) ? n2-16 : n2+16));
+        EXPECT_EQ(partner(n3,  ts_step), ((n1 % 32) ? n3-16 : n3+16));
+        EXPECT_EQ(partner(n4,  ts_step), ((n1 % 32) ? n4-16 : n4+16));
+        EXPECT_EQ(partner(n5,  ts_step), ((n1 % 32) ? n5-16 : n5+16));
+        EXPECT_EQ(partner(n6,  ts_step), ((n1 % 32) ? n6-16 : n6+16));
+        EXPECT_EQ(partner(n7,  ts_step), ((n1 % 32) ? n7-16 : n7+16));
+        EXPECT_EQ(partner(n8,  ts_step), ((n1 % 32) ? n8-16 : n8+16));
+        EXPECT_EQ(partner(n9,  ts_step), ((n1 % 32) ? n9-16 : n9+16));
+        EXPECT_EQ(partner(n10, ts_step), ((n1 % 32) ? n10-16 : n10+16));
+        EXPECT_EQ(partner(n11, ts_step), ((n1 % 32) ? n11-16 : n11+16));
+        EXPECT_EQ(partner(n12, ts_step), ((n1 % 32) ? n12-16 : n12+16));
+        EXPECT_EQ(partner(n13, ts_step), ((n1 % 32) ? n13-16 : n13+16));
+        EXPECT_EQ(partner(n14, ts_step), ((n1 % 32) ? n14-16 : n14+16));
+        EXPECT_EQ(partner(n15, ts_step), ((n1 % 32) ? n15-16 : n15+16));
+        EXPECT_EQ(partner(n16, ts_step), ((n1 % 32) ? n16-16 : n16+16));
+    }
+}
+
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ * Assertion check
+ */
+TEST(Tdistbitonic_UT, keepsmall_test1) {
+    ASSERT_DEATH(keepsmall(0, 0, 0), "");
+    ASSERT_DEATH(keepsmall(1, 1, 42), "");
+    ASSERT_DEATH(keepsmall(7, 7, 42), "");
+}
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ *
+ * depth: 1 | step: 0 | partner: [1, 0, 3, 2, 5, 4, 7, 6] | keepsmall: Bool[1, 0, 0, 1, 1, 0, 0, 1]
+ */
+TEST(Tdistbitonic_UT, keepsmall_test2) {
+    size_t ts_depth   = 1UL;
+    size_t ts_partner[]  = { 1, 0, 3, 2, 5, 4, 7, 6};
+    bool   ts_expected[] = {1, 0, 0, 1, 1, 0, 0, 1};
+
+    for (size_t node = 0 ; node < 8UL ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepsmall(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ *
+ * depth: 2 | step: 1 | partner: [2, 3, 0, 1, 6, 7, 4, 5] | keepsmall: Bool[1, 1, 0, 0, 0, 0, 1, 1]
+ */
+TEST(Tdistbitonic_UT, keepsmall_test3) {
+    size_t ts_depth   = 2UL;
+    size_t ts_partner[]  = { 2, 3, 0, 1, 6, 7, 4, 5};
+    bool   ts_expected[] = {1, 1, 0, 0, 0, 0, 1, 1};
+
+    for (size_t node = 0 ; node < 8UL ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepsmall(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ *
+ * depth: 2 | step: 0 | partner: [1, 0, 3, 2, 5, 4, 7, 6] | keepsmall: Bool[1, 0, 1, 0, 0, 1, 0, 1]
+ */
+TEST(Tdistbitonic_UT, keepsmall_test4) {
+    size_t ts_depth   = 2UL;
+    size_t ts_partner[]  = { 1, 0, 3, 2, 5, 4, 7, 6};
+    bool   ts_expected[] = {1, 0, 1, 0, 0, 1, 0, 1};
+
+    for (size_t node = 0 ; node < 8UL ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepsmall(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ *
+ * depth: 3 | step: 2 | partner: [4, 5, 6, 7, 0, 1, 2, 3] | keepsmall: Bool[1, 1, 1, 1, 0, 0, 0, 0]
+ */
+TEST(Tdistbitonic_UT, keepsmall_test5) {
+    size_t ts_depth   = 3UL;
+    size_t ts_partner[]  = { 4, 5, 6, 7, 0, 1, 2, 3};
+    bool   ts_expected[] = {1, 1, 1, 1, 0, 0, 0, 0};
+
+    for (size_t node = 0 ; node < 8UL ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepsmall(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ *
+ * depth: 3 | step: 1 | partner: [2, 3, 0, 1, 6, 7, 4, 5] | keepsmall: Bool[1, 1, 0, 0, 1, 1, 0, 0]
+ */
+TEST(Tdistbitonic_UT, keepsmall_test6) {
+    size_t ts_depth   = 3UL;
+    size_t ts_partner[]  = { 2, 3, 0, 1, 6, 7, 4, 5};
+    bool   ts_expected[] = {1, 1, 0, 0, 1, 1, 0, 0};
+
+    for (size_t node = 0 ; node < 8UL ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepsmall(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(size_t node, size_t partner, size_t depth);
+ *
+ * depth: 3 | step: 0 | partner: [1, 0, 3, 2, 5, 4, 7, 6] | keepsmall: Bool[1, 0, 1, 0, 1, 0, 1, 0]
+ */
+TEST(Tdistbitonic_UT, keepsmall_test7) {
+    size_t ts_depth   = 3UL;
+    size_t ts_partner[]  = { 1, 0, 3, 2, 5, 4, 7, 6};
+    bool   ts_expected[] = {1, 0, 1, 0, 1, 0, 1, 0};
+
+    for (size_t node = 0 ; node < 8UL ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepsmall(node, ts_partner[node], ts_depth));
+    }
+}
+
+TEST(Tdistbitonic_UT, distbitonic_test1) {
+    AllData_t ts_Data {
+            Data_t (8), Data_t (8)
+    };
+
+    std::srand(unsigned(std::time(nullptr)));
+    for (auto& v : ts_Data) {
+        std::generate(v.begin(), v.end(), std::rand);
+    }
+
+    distbitonic(2, ts_Data);
+
+    auto max = std::numeric_limits<Data_t::value_type>::min();
+    for (auto& v : ts_Data) {
+        EXPECT_EQ((max <= v[0]), true);
+        EXPECT_EQ(std::is_sorted(v.begin(), v.end()), true);
+        max = v.back();
+    }
+}
+
+TEST(Tdistbitonic_UT, distbitonic_test2) {
+    AllData_t ts_Data {
+            Data_t (8), Data_t (8), Data_t (8), Data_t (8)
+    };
+
+    std::srand(unsigned(std::time(nullptr)));
+    for (auto& v : ts_Data) {
+        std::generate(v.begin(), v.end(), std::rand);
+    }
+
+    distbitonic(4, ts_Data);
+
+    auto max = std::numeric_limits<Data_t::value_type>::min();
+    for (auto& v : ts_Data) {
+        EXPECT_EQ((max <= v[0]), true);
+        EXPECT_EQ(std::is_sorted(v.begin(), v.end()), true);
+        max = v.back();
+    }
+}
+
+TEST(Tdistbitonic_UT, distbitonic_test3) {
+    AllData_t ts_Data {
+            Data_t (32), Data_t (32), Data_t (32), Data_t (32),
+            Data_t (32), Data_t (32), Data_t (32), Data_t (32)
+    };
+
+    std::srand(unsigned(std::time(nullptr)));
+    for (auto& v : ts_Data) {
+        std::generate(v.begin(), v.end(), std::rand);
+    }
+
+    distbitonic(8, ts_Data);
+
+    auto max = std::numeric_limits<Data_t::value_type>::min();
+    for (auto& v : ts_Data) {
+        EXPECT_EQ((max <= v[0]), true);
+        EXPECT_EQ(std::is_sorted(v.begin(), v.end()), true);
+        max = v.back();
+    }
+}