HW2: RC3 - Batch scripts added

HW2: RC3 - A small exchange optimization
HW2: RC2 - Parallel Full Sort
2025-01-04 19:25:13 +02:00 · 2025-01-04 16:56:34 +02:00 · 2025-01-03 16:15:13 +02:00 · 2025-01-03 14:45:16 +02:00 · 2024-12-31 14:26:53 +02:00 · 2024-12-31 12:54:58 +02:00
44 changed files with 31547 additions and 955 deletions
@@ -1,5 +1,5 @@
 #
-# PDS homework_1 Makefile
+# PDS HW2 Makefile
 #
 # Copyright (C) 2024 Christos Choutouridis <christos@choutouridis.net>
 #
@@ -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

@@ -40,18 +45,18 @@ DEP_DIR         := $(BUILD_DIR)/.dep

 # ========== Compiler settings ==========
 # Compiler flags for debug and release
-DEB_CFLAGS      := -DDEBUG -g3 -Wall -Wextra -std=c11
-REL_CFLAGS      := -Wall -Wextra -O3 -std=c11
-DEB_CXXFLAGS    := -DDEBUG -g3 -Wall -Wextra -std=c++17
-REL_CXXFLAGS    := -Wall -Wextra -O3 -std=c++17
+DEB_CFLAGS      := -DDEBUG -g3 -Wall -Wextra -std=c11 -fopenmp
+REL_CFLAGS      := -Wall -Wextra -O3 -std=c11 -fopenmp
+DEB_CXXFLAGS    := -DDEBUG -g3 -Wall -Wextra -std=c++17 -fopenmp
+REL_CXXFLAGS    := -Wall -Wextra -O3 -std=c++17 -fopenmp

 # Pre-defines
 # PRE_DEFS := MYCAB=1729 SUPER_MODE
-PRE_DEFS		:=
+PRE_DEFS        := _GLIBCXX_PARALLEL

 # ============== Linker settings ==============
 # Linker flags (example: -pthread -lm)
-LDFLAGS         := -pthread
+LDFLAGS         := -pthread -fopenmp

 # Map output file
 MAP_FILE        := output.map
@@ -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 +168,83 @@ 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 =================
-#
-# 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
+# ================ Build rules =================
 #

+# 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)
+	@mkdir -p out
+	cp $(BUILD_DIR)/$(TARGET) out/$(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)
+	@mkdir -p out
+	cp $(BUILD_DIR)/$(TARGET) out/$(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)
+	@mkdir -p out
+	cp $(BUILD_DIR)/$(TARGET) out/$(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)
+	@mkdir -p out
+	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
+
+tests: CC := mpicc
+tests: CXX := mpic++
+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)
+	@mkdir -p out
+	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
+
+perfbitonic: CC := mpicc
+perfbitonic: CXX := mpic++
+perfbitonic: CFLAGS := $(REL_CFLAGS) -g -DCODE_VERSION=BITONIC
+perfbitonic: CXXFLAGS := $(REL_CXXFLAGS) -g -DCODE_VERSION=BITONIC
+perfbitonic: TARGET := perfbitonic
+perfbitonic: $(BUILD_DIR)/$(TARGET)
+	@mkdir -p out
+	cp $(BUILD_DIR)/$(TARGET) out/$(TARGET)
+
+
+hpc-build:
+	make clean
+	make distbubbletonic
+	make clean
+	make distbitonic
+	make clean
+	make tests
+
+
+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.
+
@@ -0,0 +1,15 @@
+#! /usr/bin/env bash
+
+if [[ $# -lt 2 ]]; then
+   echo "Error: You must pass the directory with the scripts and partition"
+   echo "you want to run to "
+   echo "example $ enqueueAll.sh ./hpc/Q23 batch"
+   echo "example $ enqueueAll.sh ./hpc/Q20 rome"
+   exit 1;
+fi
+
+# Enqueue
+for file in $(ls $1); do
+  echo "sbatch -p $2 --qos=small $file";
+  eval "sbatch -p $2 --qos=small $file";
+done
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=2
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=16
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=32
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=8
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 20 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=2
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=16
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=32
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=8
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 23 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=2
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=16
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=32
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=8
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 25 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=2
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=1
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=1:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=2
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=16
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=32
+#SBATCH --cpus-per-task=4
+#SBATCH --time=5:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=4
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -0,0 +1,28 @@
+#! /usr/bin/env bash
+
+#SBATCH --nodes=4
+#SBATCH --ntasks-per-node=8
+#SBATCH --cpus-per-task=4
+#SBATCH --time=2:00
+
+# Use this as following
+#   $> sbatch -p batch|rome <this file>
+#
+# NOTE:
+#   First compile in aristotle with
+#   $> module load gcc/9.2.0 openmpi/4.0.3
+#   $> make -j hpc-build
+#
+
+module load gcc/9.2.0 openmpi/4.0.3
+# Note:
+#   The above versions are matching w/ my system's
+#   versions, thus making compiling/debugging easier.
+
+# Suppress unused UCX_ROOT warning
+export UCX_WARN_UNUSED_ENV_VARS=n
+
+# Suppress CUDA-aware support is disabled warning
+export OMPI_MCA_opal_warn_on_missing_libcuda=0
+
+srun ./out/distbitonic -q 27 --perf --validation
@@ -1,5 +1,5 @@
 /*!
- * \file    config,h
+ * \file
 * \brief   Build configuration file.
 *
 * \author
@@ -10,30 +10,52 @@
 #ifndef CONFIG_H_
 #define CONFIG_H_

-#include <iostream>
-#include <string>
+#include <cstdint>

 /*
 * Defines for different version of the exercise
 */
-#define  V50     (50)
-#define  V100    (100)
+#define  BITONIC        (1)
+#define  BUBBLETONIC    (2)


 // Fail-safe version selection
 #if !defined CODE_VERSION
-#define CODE_VERSION   V1
+#define CODE_VERSION   BITONIC
 #endif

+// Default Data size (in case -q <N> is not present)
+#define DEFAULT_DATA_SIZE (1 << 16)
+
+/*!
+ * Value type selection
+ *
+ * We support the following compiler types or the <cstdint> that translate to them:
+ *  char       -  unsigned char
+ *  short      -  unsigned short
+ *  int        -  unsigned int
+ *  long       -  unsigned long
+ *  long long  -  unsigned long long
+ *  float
+ *  double
+ */
+using   distValue_t = uint32_t;

 /*!
 * Session option for each invocation of the executable
 */
-struct session_t {
-    bool        timing          {false};
-    bool        verbose         {false};    //!< Flag to enable verbose output to stdout
+struct config_t {
+    size_t  arraySize{DEFAULT_DATA_SIZE};   //!< The array size of the local data to sort.
+    bool    validation{false};              //!< Request a full validation at the end, performed by process rank 0.
+    bool    ndebug{false};                  //!< Skips debug trap on DEBUG builds.
+    bool    perf{false};                    //!< Enable performance timing measurements and prints.
+    bool    verbose{false};                 //!< Flag to enable verbose output to stdout.
 };

-extern session_t session;
+/*
+ * Exported data types
+ */
+extern config_t config;
+

 #endif /* CONFIG_H_ */
@@ -0,0 +1,380 @@
+/*!
+ * \file
+ * \brief   Distributed sort implementation header
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#ifndef DISTBITONIC_H_
+#define DISTBITONIC_H_
+
+#include <vector>
+#include <algorithm>
+#include <parallel/algorithm>
+#include <cmath>
+#include <cstdint>
+#if !defined DEBUG
+#define NDEBUG
+#endif
+#include <cassert>
+
+#include "utils.hpp"
+
+extern Timing TfullSort, Texchange, Tminmax, TelbowSort;    // make timers public
+
+/*!
+ * Enumerator for the different versions of the sorting method
+ */
+enum class SortMode {
+    Bubbletonic,    //!< The v0.5 of the algorithm where we use a bubble-sort like approach
+    Bitonic         //!< The v1.0 of the algorithm where we use the bitonic data-exchange approach
+};
+
+/*
+ * ============================== Sort utilities ==============================
+ */
+
+/*!
+ * The primary function template of ascending(). It is DISABLED since , it is explicitly specialized
+ * for each of the \c SortMode
+ */
+template <SortMode Mode> inline bool ascending(mpi_id_t, [[maybe_unused]] size_t) noexcept = delete;
+
+/*!
+ * 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      [mpi_id_t] The current node (MPI process)
+ * @return          [bool]     True if we need ascending configuration, false otherwise
+ */
+template <> inline
+bool ascending<SortMode::Bubbletonic>(mpi_id_t node, [[maybe_unused]] size_t depth) noexcept {
+    return (node % 2) == 0;
+}
+
+/*!
+ * 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      [mpi_id_t] The current node (MPI process)
+ * @param depth     [size_t]   The total depth of the sorting network (same for each step for a given network)
+ * @return          [bool]     True if we need ascending configuration, false otherwise
+ */
+template <> inline
+bool ascending<SortMode::Bitonic>(mpi_id_t node, size_t depth) noexcept {
+    return !(node & (1 << depth));
+}
+
+/*!
+ * The primary function template of partner(). It is DISABLED since , it is explicitly specialized
+ * for each of the \c SortMode
+ */
+template <SortMode Mode> inline mpi_id_t partner(mpi_id_t, size_t) noexcept = delete;
+
+/*!
+ * Returns the node's partner for data exchange during the sorting network iterations
+ * of Bubbletonic
+ *
+ * @param node      [mpi_id_t] The current node
+ * @param step      [size_t]   The step of the sorting network
+ * @return          [mpi_id_t] The node id of the partner for data exchange
+ */
+template <> inline
+mpi_id_t partner<SortMode::Bubbletonic>(mpi_id_t node, size_t step) noexcept {
+    //return (node % 2 == step % 2) ? node + 1 : node - 1;
+    return (((node+step) % 2) == 0) ? node + 1 : node - 1;
+}
+
+/*!
+ * Returns the node's partner for data exchange during the sorting network iterations
+ * of Bitonic
+ *
+ * @param node      [mpi_id_t] The current node
+ * @param step      [size_t]   The step of the sorting network
+ * @return          [mpi_id_t] The node id of the partner for data exchange
+ */
+template <> inline
+mpi_id_t partner<SortMode::Bitonic>(mpi_id_t node, size_t step) noexcept {
+    return (node ^ (1 << step));
+}
+
+
+/*!
+ * The primary function template of keepSmall(). It is DISABLED since , it is explicitly specialized
+ * for each of the \c SortMode
+ */
+template<SortMode Mode> inline bool keepSmall(mpi_id_t, mpi_id_t, [[maybe_unused]] size_t) = delete;
+
+/*!
+ * Predicate to check if a node keeps the small numbers during the bubbletonic sort network exchange.
+ *
+ * @param node      [mpi_id_t] The node for which we check
+ * @param partner   [mpi_id_t] The partner of the data exchange
+ * @return          [bool]     True if the node should keep the small values, false otherwise
+ */
+template <> inline
+bool keepSmall<SortMode::Bubbletonic>(mpi_id_t node, mpi_id_t partner, [[maybe_unused]] size_t depth)  {
+    if (node == partner)
+        throw std::runtime_error("(keepSmall) Node and Partner can not be the same\n");
+    return (node < partner);
+}
+
+/*!
+ * Predicate to check if a node keeps the small numbers during the bitonic sort network exchange.
+ *
+ * @param node      [mpi_id_t] The node for which we check
+ * @param partner   [mpi_id_t] The partner of the data exchange
+ * @param depth     [size_t]   The total depth of the sorting network (same for each step for a given network)
+ * @return          [bool]     True if the node should keep the small values, false otherwise
+ */
+template <> inline
+bool keepSmall<SortMode::Bitonic>(mpi_id_t node, mpi_id_t partner, size_t depth) {
+    if (node == partner)
+        throw std::runtime_error("(keepSmall) Node and Partner can not be the same\n");
+    return ascending<SortMode::Bitonic>(node, depth) == (node < partner);
+}
+
+/*!
+ * Predicate to check if the node is active in the current iteration of the bubbletonic
+ * sort exchange.
+ *
+ * @param node      [mpi_id_t] The node to check
+ * @param nodes     [size_t]   The total number of nodes
+ * @return          [bool]     True if the node is active, false otherwise
+ */
+bool isActive(mpi_id_t node, size_t nodes);
+
+/*
+ * ============================== Data utilities ==============================
+ */
+
+/*!
+ * Sort a range using the build-in O(Nlog(N)) algorithm
+ *
+ * @tparam RangeT   A range type with random access iterator
+ *
+ * @param data      [RangeT] The data to be sorted
+ * @param ascending [bool]   Flag to indicate the sorting order
+ */
+template<typename RangeT>
+void fullSort(RangeT& data, bool ascending) noexcept {
+    // Use introsort from stdlib++ here, unless ... __gnu_parallel
+    if (ascending) {
+        __gnu_parallel::sort(data.begin(), data.end(), std::less<>());
+    }
+    else {
+        __gnu_parallel::sort(data.begin(), data.end(), std::greater<>());
+    }
+}
+
+/*!
+ * Core functionality of sort for shadowed buffer types using
+ * the "elbow sort" algorithm.
+ *
+ * @note:
+ *  This algorithm can not work "in place".
+ *  We use the active buffer as source and the shadow as target.
+ *  At the end we switch which buffer is active and which is the shadow.
+ * @note
+ *  This is the core functionality. Use the elbowSort() function instead
+ *
+ * @tparam ShadowedDataT    A Shadowed buffer type with random access iterator.
+ * @tparam CompT            A Comparison type for binary operation comparisons
+ *
+ * @param data          [ShadowedDataT] The data to sort
+ * @param ascending     [bool]          Flag to indicate the sorting order
+ * @param comp          [CompT]         The binary operator object
+ */
+template<typename ShadowedDataT, typename CompT>
+void elbowSortCore(ShadowedDataT& data, bool ascending, CompT comp) noexcept {
+    auto& active = data.getActive(); // Get the source vector (the data to sort)
+    auto& shadow = data.getShadow(); // Get the target vector (the sorted data)
+
+    size_t N = data.size();         // The total size is the same or both vectors
+    size_t left = std::distance(
+            active.begin(),
+            (ascending) ?
+                std::min_element(active.begin(), active.end()) :
+                std::max_element(active.begin(), active.end())
+    );                              // start 'left' from elbow of the bitonic
+    size_t right = (left == N-1) ? 0 : left + 1;
+
+    // Walk in opposite directions from elbow and insert-sort to target vector
+    for (size_t i = 0 ; i<N ; ++i) {
+        if (comp(active[left], active[right])) {
+            shadow[i] = active[left];
+            left = (left == 0) ? N-1 : left -1; // cycle decrease
+        }
+        else {
+            shadow[i] = active[right];
+            right = (right + 1) % N;            // cycle increase
+        }
+    }
+    data.switch_active();           // Switch active-shadow buffers
+}
+
+/*!
+ * Sort a shadowed buffer using the "elbow sort" algorithm.
+ *
+ * @tparam ShadowedDataT    A Shadowed buffer type with random access iterator.
+ *
+ * @param data          [ShadowedDataT] The data to sort
+ * @param ascending     [bool]          Flag to indicate the sorting order
+ */
+template<typename ShadowedDataT>
+void elbowSort(ShadowedDataT& data, bool ascending) noexcept {
+    if (ascending)
+        elbowSortCore(data, ascending, std::less<>());
+    else
+        elbowSortCore(data, ascending, std::greater<>());
+}
+
+/*!
+ * Predicate for exchange optimization. Returns true only if an exchange between partners is needed.
+ * In order to do that we exchange min and max statistics of the partner's data.
+ *
+ * @tparam StatT    Statistics data type (for min-max)
+ *
+ * @param lstat     [const StatT]   Reference to the local statistic data
+ * @param rstat     [StatT]         Reference to the remote statistic data to fill
+ * @param part      [mpi_id_t]      The partner for the exchange
+ * @param tag       [int]           The tag to use for the exchange of stats
+ * @param keepSmall [bool]          Flag to indicate if the local thread keeps the small ro the large values
+ * @return          True if we need data exchange, false otherwise
+ */
+template<typename StatT>
+bool needsExchange(const StatT& lstat, StatT& rstat, mpi_id_t part, int tag, bool keepSmall) {
+    timeCall(Texchange, mpi.exchange_it, lstat, rstat, part, tag);
+    return (keepSmall) ?
+        rstat.min < lstat.max   // Lmin: rstat.min - Smax: lstat.max
+      : lstat.min < rstat.max;  // Lmin: lstat.min - Smax: rstat.max
+}
+
+/*!
+ * Update stats utility
+ *
+ * @tparam RangeT   A range type with random access iterator
+ * @tparam StatT    Statistics data type (for min-max)
+ *
+ * @param stat      [StatT]         Reference to the statistic data to update
+ * @param data      [const RangeT]  Reference to the sequence to extract stats from
+ */
+template<typename RangeT, typename StatT>
+void updateMinMax(StatT& stat, const RangeT& data) noexcept {
+    auto [min, max] = std::minmax_element(data.begin(), data.end());
+    stat.min = *min;
+    stat.max = *max;
+}
+
+/*!
+ * Takes two sorted sequences where one is in increasing and the other is in decreasing order
+ * and selects either the larger or the smaller items in one-to-one comparison between them.
+ * The result is a bitonic sequence.
+ *
+ * @tparam RangeT   A range type with random access iterator
+ *
+ * @param local     [RangeT]        Reference to the local sequence
+ * @param remote    [const RangeT]  Reference to the remote sequence (copied locally by MPI)
+ * @param keepSmall [bool]          Flag to indicate if we keep the small items in local sequence
+ */
+template<typename RangeT>
+void keepMinOrMax(RangeT& local, const RangeT& remote, bool keepSmall) noexcept {
+    using value_t = typename RangeT::value_type;
+    std::transform(
+            local.begin(), local.end(),
+            remote.begin(),
+            local.begin(),
+            [&keepSmall](const value_t& a, const value_t& b){
+                return (keepSmall) ? std::min(a, b) : std::max(a, b);
+            });
+}
+
+/*
+ * ============================== Sort algorithms ==============================
+ */
+
+/*!
+ * A distributed version of the Bubbletonic sort algorithm.
+ *
+ * @note
+ *  Each MPI process should run an instance of this function.
+ *
+ * @tparam ShadowedDataT    A Shadowed buffer type with random access iterator.
+ *
+ * @param data          [ShadowedDataT] The local to MPI process data to sort
+ * @param Processes     [mpi_id_t]      The total number of MPI processes
+ * @param rank          [mpi_id_t]      The current process id
+ */
+template<typename ShadowedDataT>
+void distBubbletonic(ShadowedDataT& data, mpi_id_t Processes, mpi_id_t rank) {
+    // Initially sort to create a half part of a bitonic sequence
+    timeCall(TfullSort,  fullSort, data, ascending<SortMode::Bubbletonic>(rank, 0));
+    updateMinMax(localStat, data);
+
+    // Sort network (O(N) iterations)
+    for (size_t step = 0; step < static_cast<size_t>(Processes); ++step) {
+        // Find out exchange configuration
+        auto part = partner<SortMode::Bubbletonic>(rank, step);
+        auto ks = keepSmall<SortMode::Bubbletonic>(rank, part, Processes);
+        if ( isActive(rank, Processes) &&
+             isActive(part, Processes) ) {
+            // Exchange with partner, keep nim-or-max and sort - O(N)
+            int tag = static_cast<int>(2 * step);
+            if (needsExchange(localStat, remoteStat, part, tag, ks)) {
+                timeCall(Texchange, mpi.exchange_data, data.getActive(), data.getShadow(), part, ++tag);
+                timeCall(Tminmax, keepMinOrMax, data.getActive(), data.getShadow(), ks);
+                updateMinMax(localStat, data);
+            }
+            timeCall(TelbowSort, elbowSort, data, ascending<SortMode::Bubbletonic>(rank, Processes));
+        }
+    }
+
+    // Invert if the node was descending.
+    if (!ascending<SortMode::Bubbletonic>(rank, 0))
+        elbowSort(data, true);
+
+}
+
+
+/*!
+ * A distributed version of the Bitonic sort algorithm.
+ *
+ * @note
+ *  Each MPI process should run an instance of this function.
+ *
+ * @tparam ShadowedDataT    A Shadowed buffer type with random access iterator.
+ *
+ * @param data          [ShadowedDataT] The local to MPI process data to sort
+ * @param Processes     [mpi_id_t]      The total number of MPI processes
+ * @param rank          [mpi_id_t]      The current process id
+ */
+template<typename ShadowedDataT>
+void distBitonic(ShadowedDataT& data, mpi_id_t Processes, mpi_id_t rank) {
+    // Initially sort to create a half part of a bitonic sequence
+    timeCall(TfullSort, fullSort, data, ascending<SortMode::Bitonic>(rank, 0));
+    updateMinMax(localStat, data);
+
+    // Run through sort network using elbow-sort ( O(LogN * LogN) iterations )
+    auto p = static_cast<uint32_t>(std::log2(Processes));
+    for (size_t depth = 1; depth <= p; ++depth) {
+        for (size_t step = depth; step > 0;) {
+            --step;
+            // Find out exchange configuration
+            auto part = partner<SortMode::Bitonic>(rank, step);
+            auto ks = keepSmall<SortMode::Bitonic>(rank, part, depth);
+            // Exchange with partner, keep nim-or-max
+            int tag = static_cast<int>( (2*p*depth) + (2*step) );
+            if (needsExchange(localStat, remoteStat, part, tag, ks)) {
+                timeCall(Texchange, mpi.exchange_data, data.getActive(), data.getShadow(), part, tag);
+                timeCall(Tminmax, keepMinOrMax, data.getActive(), data.getShadow(), ks);
+                updateMinMax(localStat, data);
+            }
+        }
+        // sort - O(N)
+        timeCall(TelbowSort, elbowSort, data, ascending<SortMode::Bitonic>(rank, depth));
+    }
+}
+
+#endif //DISTBITONIC_H_
@@ -1,804 +0,0 @@
-/**
- * \file    matrix.hpp
- * \brief   A matrix abstraction implementation
- *
- * \author
- *    Christos Choutouridis AEM:8997
- *    <cchoutou@ece.auth.gr>
- */
-#ifndef MATRIX_HPP_
-#define MATRIX_HPP_
-
-#include <type_traits>
-#include <utility>
-#include <algorithm>
-#include <vector>
-#include <tuple>
-
-namespace mtx {
-
-using std::size_t;
-
-/*
- * Small helper to strip types
- */
-template<typename T>
-struct remove_cvref {
-    typedef std::remove_cv_t<std::remove_reference_t<T>> type;
-};
-template<typename T>
-using remove_cvref_t = typename remove_cvref<T>::type;
-
-/*!
- * Enumerator to denote the storage type of the array to use.
- */
-enum class MatrixType {
-   DENSE,      /*!< Matrix is dense */
-   SPARSE,     /*!< Matrix is sparse */
-};
-
-/*!
- * Enumerator to denote the storage type of the array to use.
- */
-enum class MatrixOrder {
-   COLMAJOR,     /*!< Matrix is column major */
-   ROWMAJOR,     /*!< Matrix is row major */
-};
-
-/*
- * Forward type declarations
- */
-
-template<typename MatrixType> struct MatCol;
-template<typename MatrixType> struct MatRow;
-template<typename MatrixType> struct MatVal;
-
- /*!
- * A 2-D matrix functionality over a 1-D array
- *
- * This is a very thin abstraction layer over a native array.
- * This is tested using compiler explorer and our template produce
- * almost identical assembly.
- *
- * The penalty hit we have is due to the fact that we use a one dimension array
- * and we have to calculate the actual position from an (i,j) pair.
- * The use of 1D array was our intention from the beginning, so the penalty
- * was pretty much unavoidable.
- *
- * \tparam DataType  The underling data type of the array
- * \tparam IndexType The underling type for the index variables and sizes
- * \tparam Type      The storage type of the array
- *    \arg  FULL              For full matrix
- *    \arg  SYMMETRIC         For symmetric matrix (we use only the lower part)
- */
-template<typename DataType,
-         typename IndexType = size_t,
-         MatrixType Type    = MatrixType::DENSE,
-         MatrixOrder Order  = MatrixOrder::ROWMAJOR,
-         bool Symmetric     = false>
-struct Matrix {
-
-   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
-   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
-       : 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 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)
-      : 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.");
-      }
-      data_ = vector_storage_.data();
-   }
-
-   //! move ctor
-   Matrix(Matrix&& m) noexcept { moves(std::move(m)); }
-   //! move
-   Matrix& operator=(Matrix&& m) noexcept { moves(std::move(m)); return *this; }
-   Matrix(const Matrix& m)             = delete;  //!< No copy ctor
-   Matrix& operator=(const Matrix& m)  = delete;  //!< No copy
-   //Matrix(const Matrix& m);
-   //Matrix& operator=(const Matrix& m) { copy(m); }
-
-   //! @}
-
-   //! \name Data exposure
-   //! @{
-
-
-   //! Get/Set the size of each dimension
-   IndexType rows() const noexcept { return rows_; }
-   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_;
-   }
-   //! Set the interface size of the Matrix (what appears to be the size)
-   IndexType resize(IndexType rows, IndexType columns) {
-      if (use_vector_) {
-         rows_ = rows;
-         cols_ = columns;
-         vector_storage_.resize(capacity(rows_, cols_));
-         data_ = vector_storage_.data();
-      }
-      return capacity(rows_, cols_);
-   }
-
-   //! Actual memory capacity of the symmetric matrix
-   static constexpr IndexType capacity(IndexType M, IndexType N) {
-      if constexpr (Symmetric)
-         return (M+1)*N/2;
-      else
-         return M*N;
-   }
-
-   /*
-    * virtual 2D accessors
-    */
-   DataType get (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 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 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 data_[i + j*rows_];
-         else
-            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 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 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 data_[i + j*rows_] = v;
-         else
-            return data_[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> operator()(IndexType i, IndexType j) noexcept {
-      return MatVal<Matrix>(this, get(i, j), i, j);
-   }
-
-   // a basic serial iterator support
-   DataType* data() noexcept { return data_; }
-   DataType* begin() noexcept { return data_; }
-   const DataType* begin() const noexcept { return data_; }
-   DataType* end() noexcept { return data_ + capacity(rows_, cols_); }
-   const DataType* end() const noexcept { return data_ + capacity(rows_, cols_); }
-
-//   IndexType begin_idx() noexcept { return 0; }
-//   IndexType end_idx()   noexcept { return capacity(rows_, cols_); }
-
-   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_); }
-   //! @}
-
-   /*!
-    * \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);
-      }
-   }
-   //! @}
-
-   //
-   void swap(Matrix& src) noexcept {
-      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 {
-      vector_storage_ = std::move(src.vector_storage_);
-      raw_storage_    = std::move(src.raw_storage_);
-      data_           = std::move(src.data_);
-      use_vector_     = std::move(src.use_vector_);
-      rows_           = std::move(src.rows_);
-      cols_           = std::move(src.cols_);
-   }
-
-   // Storage
-   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.
-};
-
-
-/**
- * A simple sparse matrix specialization.
- *
- * We use CSC format and provide get/set functionalities for each (i,j) item
- * on the matrix. We also provide a () overload using a proxy MatVal object.
- * This way the user can:
- * \code
- *    auto v = A(3,4);
- *    A(3, 4) = 7;
- * \endcode
- *
- * We also provide getCol() and getRow() functions witch return a viewer/iterator to rows and
- * columns of the matrix. In the case of a symmetric matrix instead of a row we return the
- * equivalent column. This way we gain speed due to CSC format nature.
- *
- * @tparam DataType  The type for values
- * @tparam IndexType The type for indexes
- * @tparam Type      The Matrix type (FULL or SYMMETRIC)
- */
-template<typename DataType, typename IndexType,
-         MatrixOrder Order,
-         bool Symmetric>
-struct Matrix<DataType, IndexType, MatrixType::SPARSE, Order, Symmetric> {
-
-   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 = MatrixType::SPARSE;  //!< meta:export of array type
-   static constexpr bool symmetric = Symmetric;    //!< meta:export symmetric flag
-
-   friend struct MatCol<Matrix>;
-   friend struct MatRow<Matrix>;
-   friend struct MatVal<Matrix>;
-
-   /*!
-    * \name Obj lifetime
-    */
-   //! @{
-
-   //! Default ctor with optional memory allocations
-   Matrix(IndexType n=IndexType{}) noexcept:
-      values{},
-      rows{},
-      col_ptr((n)? n+1:2, IndexType{}),
-      N(n),
-      NNZ(0) { }
-
-   //! A ctor using csc array data
-   Matrix(IndexType n, IndexType nnz, const IndexType* row, const IndexType* col) noexcept:
-      values(nnz, 1),
-      rows(row, row+nnz),
-      col_ptr(col, col+n+1),
-      N(n),
-      NNZ(nnz) { }
-
-   //! ctor using csc array data with value array
-   Matrix(IndexType n, IndexType nnz, const DataType* v, const IndexType* row, const IndexType* col) noexcept:
-      values(v, v+nnz),
-      rows(row, row+nnz),
-      col_ptr(col, col+n+1),
-      N(n),
-      NNZ(nnz) { }
-
-   //! ctor vectors of row/col and default value for values array
-   Matrix(IndexType n, IndexType nnz, const DataType v,
-          const std::vector<IndexType>& row, const std::vector<IndexType>& col) noexcept:
-      values(nnz, v),
-      rows (row),
-      col_ptr(col),
-      N(n),
-      NNZ(nnz) { }
-
-   //! move ctor
-   Matrix(Matrix&& m) noexcept { moves(std::move(m)); }
-   //! move
-   Matrix& operator=(Matrix&& m) noexcept { moves(std::move(m)); return *this; }
-   Matrix(const Matrix& m)             = delete;  //!< make sure there are no copies
-   Matrix& operator=(const Matrix& m)  = delete;  //!< make sure there are no copies
-   //! @}
-
-   //! \name Data exposure
-   //! @{
-
-   //! \return the dimension of the matrix
-   IndexType size()     noexcept { return N; }
-   //! After construction size configuration tool
-   IndexType resize(IndexType n) {
-      col_ptr.resize(n+1);
-      return N = n;
-   }
-   //! \return the NNZ of the matrix
-   IndexType capacity() noexcept { return NNZ; }
-   //! After construction NNZ size configuration tool
-   IndexType capacity(IndexType nnz) noexcept {
-      values.reserve(nnz);
-      rows.reserve(nnz);
-      return NNZ;
-   }
-   // getters for row arrays of the struct (unused)
-   std::vector<DataType>&  getValues() noexcept { return values; }
-   std::vector<IndexType>& getRows() noexcept { return rows; }
-   std::vector<IndexType>& getCols() noexcept { return col_ptr; }
-
-   /*!
-    * 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> operator()(IndexType i, IndexType j) noexcept {
-      return MatVal<Matrix>(this, get(i, j), i, j);
-   }
-
-   /*!
-    * A read item functionality using binary search to find the correct row
-    *
-    * @param i    The row number
-    * @param j    The column number
-    * @return     The value of the item or DataType{} if is not present.
-    */
-   DataType get(IndexType i, IndexType j) noexcept {
-      IndexType idx; bool found;
-      std::tie(idx, found) =find_idx(rows, col_ptr[j], col_ptr[j+1], i);
-      return (found) ? values[idx] : 0;
-   }
-
-   /*!
-    * A write item functionality.
-    *
-    * First we search if the matrix has already a value in (i, j) position.
-    * If so we just change it to a new value. If not we add the item on the matrix.
-    *
-    * @note
-    *    When change a value, we don't increase the NNZ value of the struct. We expect the user has already
-    *    change the NNZ value to the right one using @see capacity() function. When adding a value we
-    *    increase the NNZ.
-    *
-    * @param i    The row number
-    * @param j    The column number
-    * @return     The new value of the item .
-    */
-   DataType set(DataType v, IndexType i, IndexType j) {
-      IndexType idx; bool found;
-      std::tie(idx, found) = find_idx(rows, col_ptr[j], col_ptr[j+1], i);
-      if (found)
-         return values[idx] = v;    // we don't change NNZ even if we write "0"
-      else {
-         values.insert(values.begin()+idx, v);
-         rows.insert(rows.begin()+idx, i);
-         std::transform(col_ptr.begin()+j+1, col_ptr.end(), col_ptr.begin()+j+1, [](IndexType it) {
-            return ++it;
-         });
-         ++NNZ;                     // we increase the NNZ even if we write "0"
-         return v;
-      }
-   }
-
-   /*!
-    * Get a view of a CSC column
-    * @param j    The column to get
-    * @return     The MatCol object @see MatCol
-    */
-   MatCol<Matrix> getCol(IndexType j) noexcept {
-      return MatCol<Matrix>(this, col_ptr[j], col_ptr[j+1]);
-   }
-
-   /*!
-    * Get a view of a CSC row
-    *
-    * In case of a SYMMETRIC matrix we can return a column instead.
-    *
-    * @param j    The row to get
-    * @return     On symmetric matrix MatCol otherwise a MatRow
-    */
-
-   MatCol<Matrix> getRow(IndexType i) noexcept {
-      if constexpr (Symmetric)
-         return getCol(i);
-      else
-         return MatRow<Matrix>(this, i);
-   }
-
-   // values only iterator support
-   DataType* begin() noexcept { return values.begin(); }
-   DataType* end()   noexcept { return values.end(); }
-   //! @}
-
-   //! A small iteration helper
-   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:
-   /*!
-    * A small binary search implementation using index for begin-end instead of iterators.
-    *
-    * \param   v     Reference to vector to search
-    * \param   begin The vector's index to begin
-    * \param   end   The vector's index to end
-    * \param   match What to search
-    * \return  An <index, status> pair.
-    *                index    is the index of the item or end if not found
-    *                status   is true if found, false otherwise
-    */
-   std::pair<IndexType, bool> find_idx(const std::vector<IndexType>& v, IndexType begin, IndexType end, IndexType match) {
-      if (v.capacity() != 0 && begin < end) {
-         IndexType b = begin, e = end-1;
-         while (b <= e) {
-            IndexType m = (b+e)/2;
-            if       (v[m] == match)   return  std::make_pair(m, true);
-            else if  (b >= e)          return  std::make_pair(end, false);
-            else {
-               if    (v[m] <  match)   b = m +1;
-               else                    e = m -1;
-            }
-         }
-      }
-      return std::make_pair(end, false);
-   }
-
-   // move helper
-   void moves(Matrix&& src) noexcept {
-      values  = std::move(src.values);
-      rows    = std::move(src.rows);
-      col_ptr = std::move(src.col_ptr);
-      N       = std::move(src.N);   // redundant for primitives
-      NNZ     = std::move(src.NNZ); //
-   }
-   //! \name Data
-   //! @{
-   std::vector<DataType>   values {};     //!< vector to store the values of the matrix
-   std::vector<IndexType>  rows{};        //!< vector to store the row information
-   std::vector<IndexType>  col_ptr{1,0};  //!< vector to store the column pointers
-   IndexType   N{0};                      //!< The dimension of the matrix (square)
-   IndexType   NNZ{0};                    //!< The NNZ (capacity of the matrix)
-   //! @}
-};
-
-
-/*!
- * A view/iterator hybrid object for Matrix columns.
- *
- * This object provides access to a column of a Matrix. The public functionalities
- * allow data access using indexes instead of iterators. We prefer indexes over iterators
- * because we can apply the same index to different inner vector of Matrix without conversion.
- *
- * @tparam DataType
- * @tparam IndexType
- */
-template<typename MatrixType>
-struct MatCol {
-   using owner_t = MatrixType;
-
-   using DataType  = typename MatrixType::dataType;
-   using IndexType = typename MatrixType::indexType;
-
-   /*!
-    * ctor using column pointers for begin-end. own is pointer to Matrix.
-    */
-   MatCol(owner_t* own, const IndexType begin, const IndexType end) noexcept :
-      owner_(own), index_(begin), begin_(begin), end_(end) {
-      vindex_ = vIndexCalc(index_);
-   }
-   MatCol()                        = default;
-   MatCol(const MatCol&)           = delete;   //!< make sure there are no copies
-   MatCol& operator=(const MatCol&)= delete;   //!< make sure there are no copies
-   MatCol(MatCol&&)                = default;
-   MatCol& operator=(MatCol&&)     = default;
-
-   //! a simple dereference operator, like an iterator
-   DataType operator* () {
-      return get();
-   }
-   //! Increment operator acts on index(), like an iterator
-   MatCol& operator++ ()    { advance(); return *this; }
-   MatCol& operator++ (int) { MatCol& p = *this; advance(); return p; }
-
-   //! () operator acts as member access (like a view)
-   DataType operator()(IndexType x) {
-      return (x == index())? get() : DataType{};
-   }
-   //! = operator acts as member assignment (like a view)
-   DataType operator= (DataType v) { return owner_->values[index_] = v; }
-   // iterator like handlers
-   // these return a virtual index value based on the items position on the full matrix
-   // but the move of the index is just a ++ away.
-   IndexType       index()       noexcept { return vindex_; }
-   const IndexType index() const noexcept { return vindex_; }
-   IndexType       begin()       noexcept { return vIndexCalc(begin_); }
-   const IndexType begin() const noexcept { return vIndexCalc(begin_); }
-   IndexType       end()         noexcept { return owner_->N; }
-   const IndexType end()   const noexcept { return owner_->N; }
-
-   /*!
-    * Multiplication operator
-    *
-    * We follow only the non-zero values and multiply only the common indexes.
-    *
-    * @tparam C   Universal reference for the type right half site column
-    *
-    * @param c    The right hand site matrix
-    * @return     The value of the inner product of two vectors
-    * @note       The time complexity is \$ O(nnz1+nnz2) \$.
-    *             Where the nnz is the max NNZ elements of the column of the matrix
-    */
-   template <typename C>
-   DataType operator* (C&& c) {
-      static_assert(std::is_same<remove_cvref_t<C>, MatCol<MatrixType>>(), "");
-      DataType v{};
-      while (index() != end() && c.index() != c.end()) {
-         if      (index() < c.index())  advance(); // advance me
-         else if (index() > c.index())  ++c;       // advance other
-         else { //index() == c.index()
-            v += get() * *c;                       // multiply and advance both
-            ++c;
-            advance();
-         }
-      }
-      return v;
-   }
-
-private:
-   //! small tool to increase the index pointers to Matrix
-   void advance() noexcept {
-      ++index_;
-      vindex_ = vIndexCalc(index_);
-   }
-   //! tool to translate between col_ptr indexes and Matrix "virtual" full matrix indexes
-   IndexType vIndexCalc(IndexType idx) {
-      return (idx < end_) ? owner_->rows[idx] : end();
-   }
-   //! small get tool
-   DataType get() { return owner_->values[index_]; }
-
-   owner_t*    owner_   {nullptr};     //!< Pointer to owner Matrix. MatCol is just a view
-   IndexType   vindex_  {IndexType{}}; //!< Virtual index of full matrix
-   IndexType   index_   {IndexType{}}; //!< index to Matrix::rows
-   IndexType   begin_   {IndexType{}}; //!< beginning index of the column in Matrix::rows
-   IndexType   end_     {IndexType{}}; //!< ending index of the column in Matrix::rows
-};
-
-/*!
- * A view/iterator hybrid object for Matrix rows.
- *
- * This object provides access to a column of a Matrix. The public functionalities
- * allow data access using indexes instead of iterators. We prefer indexes over iterators
- * because we can apply the same index to different inner vector of Matrix without conversion.
- *
- * @tparam DataType
- * @tparam IndexType
- */
-template<typename MatrixType>
-struct MatRow {
-   using owner_t = MatrixType;
-
-   using DataType  = typename MatrixType::dataType;
-   using IndexType = typename MatrixType::indexType;
-
-   /*!
-    * ctor using virtual full matrix row index. own is pointer to Matrix.
-    */
-   MatRow(owner_t* own, const IndexType row) noexcept :
-      owner_(own), vindex_(IndexType{}), row_(row), index_(IndexType{}),
-      begin_(IndexType{}), end_(owner_->NNZ) {
-      // place begin
-      while(begin_ != end_ && owner_->rows[begin_] != row_)
-         ++begin_;
-      // place index_ and vindex_
-      if (owner_->rows[index_] != row_)
-         advance();
-   }
-   MatRow()                        = default;
-   MatRow(const MatRow&)           = delete;   //!< make sure there are no copies
-   MatRow& operator=(const MatRow&)= delete;   //!< make sure there are no copies
-   MatRow(MatRow&&)                = default;
-   MatRow& operator=(MatRow&&)     = default;
-
-   //! a simple dereference operator, like an iterator
-   DataType operator* () {
-      return get();
-   }
-   //! Increment operator acts on index(), like an iterator
-   //! here the increment is a O(N) process.
-   MatRow& operator++ ()    { advance(); return *this; }
-   MatRow& operator++ (int) { MatRow& p = *this; advance(); return p; }
-
-   //! () operator acts as member access (like a view)
-   DataType operator()(IndexType x) {
-      return (x == index())? get() : DataType{};
-   }
-   //! = operator acts as member assignment (like a view)
-   DataType operator= (DataType v) { return owner_->values[index_] = v; }
-   // iterator like handlers
-   // these return a virtual index value based on the items position on the full matrix
-   // but the move of the index is just a ++ away.
-   IndexType       index()       noexcept { return vindex_; }
-   const IndexType index() const noexcept { return vindex_; }
-   IndexType       begin()       noexcept { return vIndexCalc(begin_); }
-   const IndexType begin() const noexcept { return vIndexCalc(begin_); }
-   IndexType       end()         noexcept { return owner_->N; }
-   const IndexType end()   const noexcept { return owner_->N; }
-
-   /*!
-    * Multiplication operator
-    *
-    * We follow only the non-zero values and multiply only the common indexes.
-    *
-    * @tparam C   Universal reference for the type right half site column
-    *
-    * @param c    The right hand site matrix
-    * @return     The value of the inner product of two vectors
-    * @note       The time complexity is \$ O(N+nnz2) \$ and way heavier the ColxCol multiplication.
-    *             Where the nnz is the max NNZ elements of the column of the matrix
-    */
-   template <typename C>
-   DataType operator* (C&& c) {
-      static_assert(std::is_same<remove_cvref_t<C>, MatCol<MatrixType>>(), "");
-      DataType v{};
-      while (index() != end() && c.index() != c.end()) {
-         if      (index() < c.index())  advance(); // advance me
-         else if (index() > c.index())  ++c;       // advance other
-         else { //index() == c.index()
-            v += get() * *c;                       // multiply and advance both
-            ++c;
-            advance();
-         }
-      }
-      return v;
-   }
-private:
-   //! small tool to increase the index pointers to Matrix matrix
-   //! We have to search the entire rows vector in Matrix to find the next
-   //! virtual row position.
-   //! time complexity O(N)
-   void advance() noexcept {
-      do
-         ++index_;
-      while(index_ != end_ && owner_->rows[index_] != row_);
-      vindex_ = vIndexCalc(index_);
-   }
-   //! tool to translate between col_ptr indexes and Matrix "virtual" full matrix indexes
-   IndexType vIndexCalc(IndexType idx) {
-      for(IndexType i =0 ; i<(owner_->N+1) ; ++i)
-         if (idx < owner_->col_ptr[i])
-            return i-1;
-      return end();
-   }
-   //! small get tool
-   DataType get() { return owner_->values[index_]; }
-
-   owner_t*    owner_   {nullptr};     //!< Pointer to owner Matrix. MatCol is just a view
-   IndexType   vindex_  {IndexType{}}; //!< Virtual index of full matrix
-   IndexType   row_     {IndexType{}}; //!< The virtual full matrix row of the object
-   IndexType   index_   {IndexType{}}; //!< index to Matrix::rows
-   IndexType   begin_   {IndexType{}}; //!< beginning index of the column in Matrix::rows
-   IndexType   end_     {IndexType{}}; //!< ending index of the column in Matrix::rows
-};
-
-/*!
- * A proxy Matrix value object/view.
- *
- * This object acts as proxy to provide read/write access to an Matrix item.
- *
- * @tparam DataType     The type of the values of the Matrix matrix
- * @tparam IndexType    The type of the indexes of the Matrix matrix
- */
-template<typename MatrixType>
-struct MatVal {
-   using owner_t = MatrixType;
-
-   using DataType  = typename MatrixType::dataType;
-   using IndexType = typename MatrixType::indexType;
-
-   //!< ctor using all value-row-column data, plus a pointer to owner Matrix object
-   MatVal(owner_t* own, DataType v, IndexType i, IndexType j) :
-      owner_(own), v_(v), i_(i), j_(j) { }
-   MatVal()                         = default;
-   MatVal(const MatVal&)            = delete;  //!< make sure there are no copies
-   MatVal& operator=(const MatVal&) = delete;  //!< make sure there are no copies
-   MatVal(MatVal&&)                 = default;
-   MatVal& operator=(MatVal&&)      = default;
-
-   //! Operator to return the DataType value implicitly
-   operator DataType() { return v_; }
-   //! Operator to write back to owner the assigned value
-   //! for ex: A(2,3) = 5;
-   MatVal& operator=(DataType v) {
-      v_ = v;
-      owner_->set(v_, i_, j_);
-      return *this;
-   }
-private:
-   owner_t*    owner_{nullptr};  //!< Pointer to owner Matrix. MatVal is just a view.
-   DataType    v_{DataType{}};   //!< The value of the row-column pair (for speed)
-   IndexType   i_{IndexType{}};  //!< The row
-   IndexType   j_{IndexType{}};  //!< the column
-};
-
-
-} // namespace mtx
-
-
-#endif /* MATRIX_HPP_ */
@@ -1,5 +1,5 @@
 /**
- * \file    utils.hpp
+ * \file
 * \brief   Utilities header
 *
 * \author
@@ -9,13 +9,322 @@
 #ifndef UTILS_HPP_
 #define UTILS_HPP_

+#include <vector>
 #include <iostream>
 #include <chrono>
 #include <unistd.h>
+#include <mpi.h>

-#include "matrix.hpp"
 #include "config.h"

+/*!
+ * Min-Max statistics data for exchange optimization
+ * @tparam Value_t  The underlying data type of the sequence data
+ */
+template <typename Value_t>
+struct Stat_t {
+    using value_type = Value_t;     //!< meta-export the type
+
+    Value_t min{};                  //!< The minimum value of the sequence
+    Value_t max{};                  //!< The maximum value of the sequence
+};
+
+//! Application data selection alias
+using distStat_t = Stat_t<distValue_t>;
+extern distStat_t localStat, remoteStat;    // Make stats public
+
+/*
+ * MPI_<type> dispatcher mechanism
+ */
+template <typename T> struct MPI_TypeMapper { };
+
+template <> struct MPI_TypeMapper<char>          { static MPI_Datatype getType() { return MPI_CHAR; } };
+template <> struct MPI_TypeMapper<short>         { static MPI_Datatype getType() { return MPI_SHORT; } };
+template <> struct MPI_TypeMapper<int>           { static MPI_Datatype getType() { return MPI_INT; } };
+template <> struct MPI_TypeMapper<long>          { static MPI_Datatype getType() { return MPI_LONG; } };
+template <> struct MPI_TypeMapper<long long>     { static MPI_Datatype getType() { return MPI_LONG_LONG; } };
+template <> struct MPI_TypeMapper<unsigned char> { static MPI_Datatype getType() { return MPI_UNSIGNED_CHAR; } };
+template <> struct MPI_TypeMapper<unsigned short>{ static MPI_Datatype getType() { return MPI_UNSIGNED_SHORT; } };
+template <> struct MPI_TypeMapper<unsigned int>  { static MPI_Datatype getType() { return MPI_UNSIGNED; } };
+template <> struct MPI_TypeMapper<unsigned long> { static MPI_Datatype getType() { return MPI_UNSIGNED_LONG; } };
+template <> struct MPI_TypeMapper<unsigned long long> { static MPI_Datatype getType() { return MPI_UNSIGNED_LONG_LONG; } };
+template <> struct MPI_TypeMapper<float>         { static MPI_Datatype getType() { return MPI_FLOAT; } };
+template <> struct MPI_TypeMapper<double>        { static MPI_Datatype getType() { return MPI_DOUBLE; } };
+
+/*!
+ * MPI wrapper type to provide MPI functionality and RAII to MPI as a resource
+ *
+ * @tparam TID  The MPI type for process id [default: int]
+ */
+template<typename TID = int>
+struct MPI_t {
+    using ID_t = TID; // Export TID type (currently int defined by the standard)
+
+    /*!
+     * Initializes the MPI environment, must called from each process
+     *
+     * @param argc  [int*]      POINTER to main's argc argument
+     * @param argv  [char***]   POINTER to main's argv argument
+     */
+    void init(int* argc, char*** argv) {
+        // Initialize the MPI environment
+        int err;
+        if ((err = MPI_Init(argc, argv)) != MPI_SUCCESS)
+            mpi_throw(err, "(MPI) MPI_Init() - ");
+        initialized_ = true;
+
+        // Get the number of processes
+        int size_value, rank_value;
+        if ((err = MPI_Comm_size(MPI_COMM_WORLD, &size_value)) != MPI_SUCCESS)
+            mpi_throw(err, "(MPI) MPI_Comm_size() - ");
+        if ((err = MPI_Comm_rank(MPI_COMM_WORLD, &rank_value)) != MPI_SUCCESS)
+            mpi_throw(err, "(MPI) MPI_Comm_rank() - ");
+        size_ = static_cast<ID_t>(size_value);
+        rank_ = static_cast<ID_t>(rank_value);
+
+        // Get the name of the processor
+        char processor_name[MPI_MAX_PROCESSOR_NAME];
+        int name_len;
+        if ((err = MPI_Get_processor_name(processor_name, &name_len)) != MPI_SUCCESS)
+            mpi_throw(err, "(MPI) MPI_Get_processor_name() - ");
+        name_ = std::string (processor_name, name_len);
+    }
+
+    /*!
+     * Exchange data with partner as part of the sorting network of both bubbletonic or bitonic
+     * sorting algorithms.
+     *
+     * This function matches a transmit and a receive in order for fully exchanged data between
+     * current node and partner.
+     *
+     * @tparam T        The inner valur type used in buffer
+     *
+     * @param ldata     [std::vector<T>]    Reference to local data to send
+     * @param rdata     [std::vector<T>]    Reference to buffer to receive data from partner
+     * @param partner   [mpi_id_t]          The partner for the exchange
+     * @param tag       [int]               The tag to use for the MPI communication
+     */
+    template<typename T>
+    void exchange_data(const std::vector<T>& ldata, std::vector<T>& rdata, ID_t partner, int tag) {
+        if (tag < 0)
+            throw std::runtime_error("(MPI) exchange_data() [tag] - Out of bound");
+
+        MPI_Datatype datatype = MPI_TypeMapper<T>::getType();
+        int count = static_cast<int>(ldata.size());
+        MPI_Status status;
+        int err;
+        if ((err = MPI_Sendrecv(
+                ldata.data(), count, datatype, partner, tag,
+                rdata.data(), count, datatype, partner, tag,
+                 MPI_COMM_WORLD, &status
+        )) != MPI_SUCCESS)
+            mpi_throw(err, "(MPI) MPI_Sendrecv() [data] - ");
+    }
+
+    /*!
+     * Exchange a data object with partner as part of the sorting network of both bubbletonic
+     * or bitonic sorting algorithms.
+     *
+     * This function matches a transmit and a receive in order for fully exchanged the data object
+     * between current node and partner.
+     *
+     * @tparam T        The object type
+     *
+     * @param local     [const T&]  Reference to the local object to send
+     * @param remote    [T&]        Reference to the object to receive data from partner
+     * @param partner   [mpi_id_t]  The partner for the exchange
+     * @param tag       [int]       The tag to use for the MPI communication
+     */
+    template<typename T>
+    void exchange_it(const T& local, T& remote, ID_t partner, int tag) {
+        if (tag < 0)
+            throw std::runtime_error("(MPI) exchange_it() [tag] - Out of bound");
+        MPI_Status status;
+        int err;
+        if ((err = MPI_Sendrecv(
+                &local, sizeof(T), MPI_BYTE, partner, tag,
+                &remote, sizeof(T), MPI_BYTE, partner, tag,
+                MPI_COMM_WORLD, &status
+        )) != MPI_SUCCESS)
+            mpi_throw(err, "(MPI) MPI_Sendrecv() [item] - ");
+    }
+
+    // Accessors
+    [[nodiscard]] ID_t rank() const noexcept { return rank_; }
+    [[nodiscard]] ID_t size() const noexcept { return size_; }
+    [[nodiscard]] const std::string& name() const noexcept { return name_; }
+
+    // Mutators
+    ID_t rank(ID_t rank) noexcept { return rank_ = rank; }
+    ID_t size(ID_t size) noexcept { return size_ = size; }
+    std::string& name(const std::string& name) noexcept { return name_ = name; }
+
+    /*!
+     * Finalized the MPI
+     */
+    void finalize() {
+        // Finalize the MPI environment
+        initialized_ = false;
+        MPI_Finalize();
+    }
+
+    //! RAII MPI finalization
+    ~MPI_t() {
+        // Finalize the MPI environment even on unexpected errors
+        if (initialized_)
+            MPI_Finalize();
+    }
+
+
+    // Local functionality
+private:
+    /*!
+     * Throw exception helper. It bundles the prefix msg with the MPI error string retrieved by
+     * MPI API.
+     *
+     * @param err           The MPI error code
+     * @param prefixMsg     The prefix text for the exception error message
+     */
+    void mpi_throw(int err, const char* prefixMsg) {
+        char err_msg[MPI_MAX_ERROR_STRING];
+        int msg_len;
+        MPI_Error_string(err, err_msg, &msg_len);
+        throw std::runtime_error(prefixMsg + std::string (err_msg) + '\n');
+    }
+
+private:
+    ID_t rank_{};           //!< MPI rank of the process
+    ID_t size_{};           //!< MPI total size of the execution
+    std::string name_{};    //!< The name of the local machine
+    bool initialized_{};    //!< RAII helper flag
+};
+
+/*
+ * Exported data types
+ */
+extern MPI_t<>  mpi;
+using mpi_id_t = MPI_t<>::ID_t;
+
+
+
+/*!
+ * @brief A std::vector wrapper with 2 vectors, an active and a shadow.
+ *
+ * This type exposes the standard vector
+ * functionality of the active vector. The shadow can be used when we need to use the vector as mutable
+ * data in algorithms that can not support "in-place" editing (like elbow-sort for example)
+ *
+ * @tparam Value_t  the underlying data type of the vectors
+ */
+template <typename Value_t>
+struct ShadowedVec_t {
+    // STL requirements
+    using value_type     = Value_t;
+    using iterator       = typename std::vector<Value_t>::iterator;
+    using const_iterator = typename std::vector<Value_t>::const_iterator;
+    using size_type      = typename std::vector<Value_t>::size_type;
+
+    // Default constructor
+    ShadowedVec_t() = default;
+
+    // Constructor from an std::vector
+    explicit ShadowedVec_t(const std::vector<Value_t>& vec)
+            : North(vec), South(), active(north) {
+        South.resize(North.size());
+    }
+
+    explicit ShadowedVec_t(std::vector<Value_t>&& vec)
+            : North(std::move(vec)), South(), active(north) {
+        South.resize(North.size());
+    }
+
+    // Copy assignment operator
+    ShadowedVec_t& operator=(const ShadowedVec_t& other) {
+        if (this != &other) { // Avoid self-assignment
+            North = other.North;
+            South = other.South;
+            active = other.active;
+        }
+        return *this;
+    }
+
+    // Move assignment operator
+    ShadowedVec_t& operator=(ShadowedVec_t&& other) noexcept {
+        if (this != &other) { // Avoid self-assignment
+            North = std::move(other.North);
+            South = std::move(other.South);
+            active = other.active;
+
+            // There is no need to zero out other since it is valid but in a non-defined state
+        }
+        return *this;
+    }
+
+    // Type accessors
+    std::vector<Value_t>& getActive() { return (active == north) ? North : South; }
+    std::vector<Value_t>& getShadow() { return (active == north) ? South : North; }
+    const std::vector<Value_t>& getActive() const { return (active == north) ? North : South; }
+    const std::vector<Value_t>& getShadow() const { return (active == north) ? South : North; }
+
+    // Swap vectors
+    void switch_active() { active = (active == north) ? south : north; }
+
+    // Dispatch vector functionality to active vector
+    Value_t& operator[](size_type index) { return getActive()[index]; }
+    const Value_t& operator[](size_type index) const { return getActive()[index]; }
+
+    Value_t& at(size_type index) { return getActive().at(index); }
+    const Value_t& at(size_type index) const { return getActive().at(index); }
+
+    void push_back(const Value_t& value) { getActive().push_back(value); }
+    void push_back(Value_t&& value)      { getActive().push_back(std::move(value)); }
+    void pop_back()                      { getActive().pop_back(); }
+    Value_t& front() { return getActive().front(); }
+    Value_t& back()  { return getActive().back(); }
+    const Value_t& front() const { return getActive().front(); }
+    const Value_t& back()  const { return getActive().back(); }
+
+    iterator begin() { return getActive().begin(); }
+    const_iterator begin() const { return getActive().begin(); }
+    iterator end() { return getActive().end(); }
+    const_iterator end() const { return getActive().end(); }
+
+    size_type size() const { return getActive().size(); }
+    void resize(size_t new_size) {
+        North.resize(new_size);
+        South.resize(new_size);
+    }
+
+    void reserve(size_t new_capacity) {
+        North.reserve(new_capacity);
+        South.reserve(new_capacity);
+    }
+    [[nodiscard]] size_t capacity() const { return getActive().capacity(); }
+    [[nodiscard]] bool empty() const { return getActive().empty(); }
+
+    void clear() { getActive().clear(); }
+    void swap(std::vector<Value_t>& other) { getActive().swap(other); }
+
+    // Comparisons
+    bool operator== (const ShadowedVec_t& other) { return getActive() == other.getActive(); }
+    bool operator!= (const ShadowedVec_t& other) { return getActive() != other.getActive(); }
+    bool operator== (const std::vector<value_type>& other) { return getActive() == other; }
+    bool operator!= (const std::vector<value_type>& other) { return getActive() != other; }
+
+private:
+    std::vector<Value_t> North{};       //!< Actual buffer to be used either as active or shadow
+    std::vector<Value_t> South{};       //!< Actual buffer to be used either as active or shadow
+    enum {
+        north, south
+    } active{north};                    //!< Flag to select between North and South buffer
+};
+
+/*
+ * Exported data types
+ */
+using distBuffer_t = ShadowedVec_t<distValue_t>;
+extern distBuffer_t Data;
+
 /*!
 * A Logger for entire program.
 */
@@ -25,24 +334,26 @@ struct Log {
    //! We provide logging via << operator
    template<typename T>
    Log &operator<<(T &&t) {
-      if (session.verbose) {
+        if (config.verbose) {
            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;
-      if (session.verbose) {
+    Log &operator<<(Endl e) {
+        (void) e;
+        if (config.verbose) {
            std::cout << '\n';
            line_ = true;
        }
        return *this;
    }
+
 private:
    bool line_{true};
 };
@@ -54,34 +365,57 @@ extern Log logger;
 */
 struct Timing {
    using Tpoint = std::chrono::steady_clock::time_point;
+    using Tduration = std::chrono::microseconds;
    using microseconds = std::chrono::microseconds;
    using milliseconds = std::chrono::milliseconds;
    using seconds = std::chrono::seconds;

    //! 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(); }
+    Tpoint start() noexcept { return mark_ = std::chrono::steady_clock::now(); }

-   auto dt () noexcept {
-      return std::chrono::duration_cast<std::chrono::microseconds>(stop_ - start_).count();
+    //! tool to mark the ending point
+    Tpoint stop() noexcept {
+        Tpoint now = std::chrono::steady_clock::now();
+        duration_ += dt(now, mark_);
+        return now;
    }
-   //! 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";
-         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";
+
+    //! A duration calculation utility
+    static Tduration dt(Tpoint t2, Tpoint t1) noexcept {
+        return std::chrono::duration_cast<Tduration>(t2 - t1);
+    }
+
+    //! Tool to print the time interval
+    void print_duration(const char *what, mpi_id_t rank) noexcept {
+        if (std::chrono::duration_cast<microseconds>(duration_).count() < 10000)
+            std::cout << "[Timing] (Rank " << rank << ") " << what << ": "
+                      << std::to_string(std::chrono::duration_cast<microseconds>(duration_).count()) << " [usec]\n";
+        else if (std::chrono::duration_cast<milliseconds>(duration_).count() < 10000)
+            std::cout << "[Timing] (Rank " << rank << ") " << what << ": "
+                      << std::to_string(std::chrono::duration_cast<milliseconds>(duration_).count()) << " [msec]\n";
        else
-            std::cout << "[Timing]: " << what << ": " << std::to_string(std::chrono::duration_cast<seconds>(t).count()) << " [sec]\n";
-      }
+            std::cout << "[Timing] (Rank " << rank << ") " << what << ": "
+                      << std::to_string(std::chrono::duration_cast<seconds>(duration_).count()) << " [sec]\n";
+
    }
+
 private:
-   Tpoint start_;
-   Tpoint stop_;
+    Tpoint mark_{};
+    Tduration duration_{};
 };

+/*!
+ * Utility "high level function"-like macro to forward a function call
+ * and accumulate the execution time to the corresponding timing object.
+ *
+ * @param   Tim     The Timing object [Needs to have methods start() and stop()]
+ * @param   Func    The function name
+ * @param   ...     The arguments to pass to function (the preprocessor way)
+ */
+#define timeCall(Tim, Func, ...)    \
+    Tim.start();                    \
+    Func(__VA_ARGS__);              \
+    Tim.stop();                     \
+

 #endif /* UTILS_HPP_ */
@@ -50,7 +50,7 @@ function distbubletonic!(p, data)
    for i in 1:p
        sort!(view(data, i, :), rev = !ascending[i])
    end
-    for step in 0:p-2
+    for step in 0:p-1
        direction = [true for x = 1:p]
        partnerid = partner.(pid, step)
        activeids = active.(partnerid, p)
@@ -4,9 +4,9 @@
 #

 function exchange(localid,  remoteid)
-    if verbose
-        println("Exchange local data from $localid with partner $remoteid")
-    end
+    # if verbose
+    #     println("Exchange local data from $localid with partner $remoteid")
+    # end
    nothing # We have all data here ;)
 end

@@ -23,9 +23,89 @@ function minmax(data, localid, remoteid, keepsmall)
 end


+function is_bitonic(arr)
+    n = length(arr)
+    if n <= 2
+        return true # Any sequence of length <= 2 is bitonic
+    end
+
+    # State for state machine. 1: inc, -1: dec, 0: z-state
+    state = 0
+    inc_count = 0
+    dec_count = 0
+    ret = false
+
+    for i in 1:n-1
+        # Find the first order
+        if state == 0
+            if arr[i] > arr[i+1]
+                state = -1
+                dec_count += 1
+            elseif arr[i] < arr[i+1]
+                state = 1
+                inc_count += 1
+            end
+        elseif state == -1   # decreasing
+            if arr[i] < arr[i + 1]
+                state = 1
+                inc_count += 1
+            end
+        elseif state == 1   # increasing
+            if arr[i] > arr[i+1]
+                state = -1
+                dec_count += 1
+            end
+        end
+    end
+
+    if inc_count <= 1 && dec_count <= 1
+        ret =  true # Sequence is bitonic
+    elseif inc_count == 2 && dec_count == 1
+        ret =  (arr[1] >= arr[n])
+    elseif inc_count == 1 && dec_count == 2
+        ret = (arr[1] <= arr[n])
+    end
+
+    ret
+end
+
+function is_sort(arr)
+    # State for state machine. 1: inc, -1: dec, 0: z-state
+    state = 0
+    inc_count = 0
+    dec_count = 0
+
+    for i in 1:length(arr)-1
+        # Find the first order
+        if state == 0
+            if arr[i] > arr[i+1]
+                state = -1
+                dec_count += 1
+            elseif arr[i] < arr[i+1]
+                state = 1
+                inc_count += 1
+            end
+        elseif state == -1   # decreasing
+            if arr[i] < arr[i + 1]
+                state = 1
+                inc_count += 1
+            end
+        elseif state == 1   # increasing
+            if arr[i] > arr[i+1]
+                state = -1
+                dec_count += 1
+            end
+        end
+    end
+
+    ret =  ((inc_count + dec_count) == 1) ? state : 0
+    ret
+end

 function sort_network!(data, n, depth)
    nodes = 0:n-1
+    bitonicFlag = zeros(Int8, size(data, 1))
+    sortFlag    = zeros(Int8, size(data, 1))
    for step = depth-1:-1:0
        partnerid = nodes .⊻ (1 << step)
        direction = (nodes .& (1 << depth)) .== 0 .& (nodes .< partnerid)
@@ -40,6 +120,13 @@ function sort_network!(data, n, depth)
                minmax(data, i, partnerid[i+1], keepsmall[i+1])
            end
        end
+        if verbose
+            for i in 1:size(data, 1)
+                bitonicFlag[i] = is_bitonic(data[i, :])
+                sortFlag[i] = is_sort(data[i, :])
+            end
+            println("depth: $depth | step: $step | bitonicFlag: $bitonicFlag | sorfFlag: $sortFlag")
+        end
    end
 end

@@ -0,0 +1,27 @@
+/*!
+ * \file
+ * \brief   Distributed sort implementation
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+#include "utils.hpp"
+#include "distsort.hpp"
+
+//! Statistic variables for exchange optimization
+distStat_t localStat, remoteStat;
+
+//! Performance timers for each one of the "costly" functions
+Timing TfullSort, Texchange, Tminmax, TelbowSort;
+
+
+bool isActive(mpi_id_t node, size_t nodes) {
+    if (!((nodes > 0) &&
+          (nodes <= std::numeric_limits<mpi_id_t>::max()) ))
+        throw std::runtime_error("(isActive) Non-acceptable value of MPI Nodes\n");
+    // ^ Assert that mpi_id_t can hold nodes, and thus we can cast without data loss!
+
+    return (node >= 0) && (node < static_cast<mpi_id_t>(nodes));
+}
+
@@ -1,22 +1,28 @@
 /*!
- * \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>
+#include <random>

-#include <mpi.h>
-
-#include "matrix.hpp"
 #include "utils.hpp"
 #include "config.h"
+#include "distsort.hpp"

-// Global session data
-session_t   session;
+
+// Global config data
+config_t        config;
+MPI_t<>         mpi;
+distBuffer_t    Data;
+Log             logger;
+Timing          Ttotal;

 /*!
 * A small command line argument parser
@@ -29,29 +35,53 @@ bool get_options(int argc, char* argv[]){
    for (int i=1 ; i<argc ; ++i) {
        std::string arg(argv[i]);     // get current argument

-        if (arg == "-x" || arg == "--xxxxx") {
-            if (i+2 < argc) {
-//                session.corpusMtxFile = std::string(argv[++i]);
-//                session.corpusDataSet = std::string(argv[++i]);
+        if (arg == "-q" || arg == "--array-size") {
+            if (i+1 < argc) {
+                config.arraySize = 1 << atoi(argv[++i]);
            }
-            else
+            else {
                status = false;
            }
-
-        else if (arg == "-v" || arg == "--verbose")
-            session.verbose = true;
+        }
+        else if (arg == "--validation") {
+            config.validation = true;
+        }
+        else if (arg == "--ndebug") {
+            config.ndebug = true;
+        }
+        else if (arg == "--perf") {
+            config.perf = true;
+        }
+        else if (arg == "-v" || arg == "--verbose") {
+            config.verbose = true;
+        }
        else if (arg == "-h" || arg == "--help") {
-            std::cout << "distbitonic - A distributed bitonic sort\n\n";
-            std::cout << "distbitonic -x <> [-v]\n";
+            std::cout << "distbitonic/distbubbletonic - A distributed bitonic sort\n\n";
+            std::cout << "distbitonic -q <N> [--validation] [--ndebug] [-v]\n";
+            std::cout << "distbitonic -h\n";
+            std::cout << "distbubbletonic -q <N> [--validation] [--ndebug] [-v]\n";
+            std::cout << "distbubbletonic -h\n";
            std::cout << '\n';
            std::cout << "Options:\n\n";
+            std::cout << "   -q | --array-size <N>\n";
+            std::cout << "      Selects the array size according to size = 2^N\n\n";
+            std::cout << "   --par-sort\n";
+            std::cout << "      Request a parallel full sorting algorithm\n\n";
+            std::cout << "   --validation\n";
+            std::cout << "      Request a full validation at the end, performed by process rank 0\n\n";
+            std::cout << "   --ndebug\n";
+            std::cout << "      Skip debug breakpoint when on debug build.\n\n";
+            std::cout << "   -t | --timing\n";
+            std::cout << "      Request timing measurements output to stdout.\n\n";
            std::cout << "   -v | --verbose\n";
            std::cout << "      Request a more verbose output to stdout.\n\n";
            std::cout << "   -h | --help\n";
            std::cout << "      Prints this and exit.\n\n";
            std::cout << "Examples:\n\n";
-            std::cout << "   ...Example case...:\n";
-            std::cout << "   > distbitonic -x <xxxxx>  \n\n";
+            std::cout << "   mpirun -np 4 distbitonic -q 24\n";
+            std::cout << "      Runs distbitonic in 4 MPI processes with 2^24 array points each\n\n";
+            std::cout << "   mpirun -np 16 distbubbletonic -q 20\n";
+            std::cout << "      Runs distbubbletonic in 16 MPI processes with 2^20 array points each\n\n";

            exit(0);
        }
@@ -64,11 +94,128 @@ bool get_options(int argc, char* argv[]){
    return status;
 }

+/*!
+ * A simple validator for the entire distributed process
+ *
+ * @tparam ShadowedDataT    A Shadowed buffer type with random access iterator.
+ *
+ * @param data          [ShadowedDataT] The local to MPI process
+ * @param Processes     [mpi_id_t]      The total number of MPI processes
+ * @param rank          [mpi_id_t]      The current process id
+ *
+ * @return              [bool]          True if all are sorted and in total ascending order
+ */
+template<typename ShadowedDataT>
+bool validator(ShadowedDataT& data, mpi_id_t Processes, mpi_id_t rank) {
+    using value_t = typename ShadowedDataT::value_type;
+    bool ret = true;    // Have faith!
+
+    // Local results
+    value_t lmin   = data.front();
+    value_t lmax   = data.back();
+    value_t lsort  = static_cast<value_t>(std::is_sorted(data.begin(), data.end()));
+
+    // Gather min/max/sort to rank 0
+    std::vector<value_t> mins(Processes);
+    std::vector<value_t> maxes(Processes);
+    std::vector<value_t> sorts(Processes);
+
+    MPI_Datatype datatype = MPI_TypeMapper<value_t>::getType();
+    MPI_Gather(&lmin,  1, datatype, mins.data(),  1, datatype, 0, MPI_COMM_WORLD);
+    MPI_Gather(&lmax,  1, datatype, maxes.data(), 1, datatype, 0, MPI_COMM_WORLD);
+    MPI_Gather(&lsort, 1, datatype, sorts.data(), 1, datatype, 0, MPI_COMM_WORLD);
+
+    // Check all results
+    if (rank == 0) {
+        for (mpi_id_t r = 1; r < Processes; ++r) {
+            if (sorts[r] == 0)
+                ret = false;
+            if (maxes[r - 1] > mins[r])
+                ret = false;
+        }
+    }
+    return ret;
+}
+
+#if !defined TESTING
+/*!
+ * @return Returns 0, but.... we may throw or exit(1)
+ */
 int main(int argc, char* argv[]) try {
-    // try to read command line
+    // Initialize MPI environment
+    mpi.init(&argc, &argv);
+
+    // try to read command line (after MPI parsing)
    if (!get_options(argc, argv))
        exit(1);

+    logger << "MPI environment initialized." <<
+              " Rank: " << mpi.rank() <<
+              " Size: " << mpi.size() <<
+              logger.endl;
+
+#if defined DEBUG
+#if defined TESTING
+    /*
+     * 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>
+     */
+     volatile bool sleep_wait = false;
+#else
+    volatile bool sleep_wait = true;
+#endif
+    while (sleep_wait && !config.ndebug)
+        sleep(1);
+#endif
+
+    // Initialize local data
+    logger << "Initialize local array of " << config.arraySize << " elements" << logger.endl;
+    std::random_device rd;      // Mersenne seeded from hw if possible. range: [type_min, type_max]
+    std::mt19937 gen(rd());
+    std::uniform_int_distribution<distValue_t > dis(
+            std::numeric_limits<distValue_t>::min(),
+            std::numeric_limits<distValue_t>::max()
+        );
+    // Fill vector
+    Data.resize(config.arraySize);
+    std::generate(Data.begin(), Data.end(), [&]() { return dis(gen); });
+
+    // Run distributed sort
+    if (mpi.rank() == 0)
+        logger << "Starting distributed sorting ... ";
+    Ttotal.start();
+    #if CODE_VERSION == BUBBLETONIC
+      distBubbletonic(Data, mpi.size(), mpi.rank());
+    #else
+      distBitonic (Data, mpi.size(), mpi.rank());
+    #endif
+    Ttotal.stop();
+    if (mpi.rank() == 0)
+        logger << " Done." << logger.endl;
+
+
+    // Print-outs and validation
+    if (config.perf) {
+            Ttotal.print_duration("Total     ", mpi.rank());
+         TfullSort.print_duration("Full-Sort ", mpi.rank());
+         Texchange.print_duration("Exchange  ", mpi.rank());
+           Tminmax.print_duration("Min-Max   ", mpi.rank());
+        TelbowSort.print_duration("Elbow-Sort", mpi.rank());
+    }
+    if (config.validation) {
+        // If requested, we have the chance to fail!
+        if (mpi.rank() == 0)
+            std::cout << "Results validation ...";
+        bool val = validator(Data, mpi.size(), mpi.rank());
+        if (mpi.rank() == 0)
+            std::cout << ((val) ? "\x1B[32m [PASS] \x1B[0m\n" : " \x1B[32m [FAIL] \x1B[0m\n");
+    }
+    mpi.finalize();
    return 0;
 }
 catch (std::exception& e) {
@@ -76,3 +223,21 @@ catch (std::exception& e) {
    std::cerr << "Error: " << e.what() << '\n';
    exit(1);
 }
+
+#else
+
+#include <gtest/gtest.h>
+#include <exception>
+
+/*!
+ * The testing version of our program
+ */
+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
@@ -0,0 +1,330 @@
+/**
+ * \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 "distsort.hpp"
+
+
+
+/* ================================== ascending ================================== */
+
+/*
+ * bool ascending<SortMode::Bitonic>(mpi_id_t node, size_t depth);
+ * depth 0 (the initial ascending pattern)
+ */
+TEST(TdistBitonic_UT, ascending_test1) {
+
+    EXPECT_EQ(ascending<SortMode::Bitonic>(0, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(1, 0), false);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(2, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(3, 0), false);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(4, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(5, 0), false);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(6, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(7, 0), false);
+
+    for (mpi_id_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(ascending<SortMode::Bitonic>(node, 0), ((node % 2) ? false : true) );
+    }
+}
+
+/*
+ * bool ascending<SortMode::Bitonic>(mpi_id_t node, size_t depth);
+ * depth 1
+ */
+TEST(TdistBitonic_UT, ascending_test2) {
+
+    EXPECT_EQ(ascending<SortMode::Bitonic>(0, 1), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(1, 1), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(2, 1), false);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(3, 1), false);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(4, 1), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(5, 1), true);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(6, 1), false);
+    EXPECT_EQ(ascending<SortMode::Bitonic>(7, 1), false);
+
+    for (mpi_id_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(ascending<SortMode::Bitonic>(2*node,   1), ((node % 2) ? false:true));
+        EXPECT_EQ(ascending<SortMode::Bitonic>(2*node+1, 1), ((node % 2) ? false:true));
+    }
+}
+
+/*
+ * bool ascending<SortMode::Bitonic>(mpi_id_t node, size_t depth);
+ * various depths
+ */
+TEST(TdistBitonic_UT, ascending_test3) {
+
+    // Depth = 3
+    size_t ts_depth = 3;
+
+    for (mpi_id_t n = 0 ; n < (1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), true);
+    for (mpi_id_t n = (1<<(ts_depth)) ; n < 2*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), false);
+    for (mpi_id_t n = 2*(1<<(ts_depth)) ; n < 3*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), true);
+    for (mpi_id_t n = 3*(1<<(ts_depth)) ; n < 4*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), false);
+
+    // Depth = 4
+    ts_depth = 4;
+
+    for (mpi_id_t n = 0L ; n < (1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), true);
+    for (mpi_id_t n = (1<<(ts_depth)) ; n < 2*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), false);
+    for (mpi_id_t n = 2*(1<<(ts_depth)) ; n < 3*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), true);
+    for (mpi_id_t n = 3*(1<<(ts_depth)) ; n < 4*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), false);
+
+    // Depth = 8
+    ts_depth = 8;
+
+    for (mpi_id_t n = 0L ; n < (1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), true);
+    for (mpi_id_t n = (1<<(ts_depth)) ; n < 2*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), false);
+    for (mpi_id_t n = 2*(1<<(ts_depth)) ; n < 3*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), true);
+    for (mpi_id_t n = 3*(1<<(ts_depth)) ; n < 4*(1<<(ts_depth)) ; ++n)
+        EXPECT_EQ(ascending<SortMode::Bitonic>(n,   ts_depth), false);
+}
+
+
+
+
+/* ================================== partner ================================== */
+
+/*
+ * mpi_id_t partner<SortMode::Bitonic>(mpi_id_t node, size_t step);
+ * step = 0
+ */
+TEST(TdistBitonic_UT, partner_test1) {
+
+    EXPECT_EQ(partner<SortMode::Bitonic>(0, 0), 1);
+    EXPECT_EQ(partner<SortMode::Bitonic>(1, 0), 0);
+    EXPECT_EQ(partner<SortMode::Bitonic>(2, 0), 3);
+    EXPECT_EQ(partner<SortMode::Bitonic>(3, 0), 2);
+    EXPECT_EQ(partner<SortMode::Bitonic>(4, 0), 5);
+    EXPECT_EQ(partner<SortMode::Bitonic>(5, 0), 4);
+    EXPECT_EQ(partner<SortMode::Bitonic>(6, 0), 7);
+    EXPECT_EQ(partner<SortMode::Bitonic>(7, 0), 6);
+
+    for (mpi_id_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(partner<SortMode::Bitonic>(node, 0), (node % 2) ? node-1 : node+1);
+    }
+}
+
+/*
+ * mpi_id_t partner<SortMode::Bitonic>(mpi_id_t node, size_t step);
+ * step = 1
+ */
+TEST(TdistBitonic_UT, partner_test2) {
+
+    EXPECT_EQ(partner<SortMode::Bitonic>(0, 1), 2);
+    EXPECT_EQ(partner<SortMode::Bitonic>(1, 1), 3);
+    EXPECT_EQ(partner<SortMode::Bitonic>(2, 1), 0);
+    EXPECT_EQ(partner<SortMode::Bitonic>(3, 1), 1);
+    EXPECT_EQ(partner<SortMode::Bitonic>(4, 1), 6);
+    EXPECT_EQ(partner<SortMode::Bitonic>(5, 1), 7);
+    EXPECT_EQ(partner<SortMode::Bitonic>(6, 1), 4);
+    EXPECT_EQ(partner<SortMode::Bitonic>(7, 1), 5);
+
+    for (mpi_id_t n1 = 0 ; n1 < 256 ; n1 += 2) {
+        auto n2 = n1 + 1;
+        EXPECT_EQ(partner<SortMode::Bitonic>(n1, 1), ((n1 % 4) ? n1-2 : n1+2));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n2, 1), ((n1 % 4) ? n2-2 : n2+2));
+    }
+}
+
+/*
+ * mpi_id_t partner(mpi_id_t node, size_t step);
+ * various steps
+ */
+TEST(TdistBitonic_UT, partner_test3) {
+    // step = 2
+    size_t ts_step = 2;
+
+    for (mpi_id_t n1 = 0 ; n1 < 256 ; n1 += 4) {
+        auto n2 = n1 + 1;
+        auto n3 = n1 + 2;
+        auto n4 = n1 + 3;
+        EXPECT_EQ(partner<SortMode::Bitonic>(n1, ts_step), ((n1 % 8) ? n1-4 : n1+4));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n2, ts_step), ((n1 % 8) ? n2-4 : n2+4));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n3, ts_step), ((n1 % 8) ? n3-4 : n3+4));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n4, ts_step), ((n1 % 8) ? n4-4 : n4+4));
+    }
+
+    // step = 3
+    ts_step = 3;
+
+    for (mpi_id_t n1 = 0 ; n1 < 256 ; n1 += 8) {
+        auto n2 = n1 + 1;
+        auto n3 = n1 + 2;
+        auto n4 = n1 + 3;
+        auto n5 = n1 + 4;
+        auto n6 = n1 + 5;
+        auto n7 = n1 + 6;
+        auto n8 = n1 + 7;
+        EXPECT_EQ(partner<SortMode::Bitonic>(n1, ts_step), ((n1 % 16) ? n1-8 : n1+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n2, ts_step), ((n1 % 16) ? n2-8 : n2+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n3, ts_step), ((n1 % 16) ? n3-8 : n3+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n4, ts_step), ((n1 % 16) ? n4-8 : n4+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n5, ts_step), ((n1 % 16) ? n5-8 : n5+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n6, ts_step), ((n1 % 16) ? n6-8 : n6+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n7, ts_step), ((n1 % 16) ? n7-8 : n7+8));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n8, ts_step), ((n1 % 16) ? n8-8 : n8+8));
+    }
+
+    // step = 4
+    ts_step = 4;
+
+    for (mpi_id_t n1 = 0 ; n1 < 256 ; n1 += 16) {
+        auto n2 = n1 + 1;
+        auto n3 = n1 + 2;
+        auto n4 = n1 + 3;
+        auto n5 = n1 + 4;
+        auto n6 = n1 + 5;
+        auto n7 = n1 + 6;
+        auto n8 = n1 + 7;
+        auto n9 = n1 + 8;
+        auto n10 = n1 + 9;
+        auto n11 = n1 + 10;
+        auto n12 = n1 + 11;
+        auto n13 = n1 + 12;
+        auto n14 = n1 + 13;
+        auto n15 = n1 + 14;
+        auto n16 = n1 + 15;
+        EXPECT_EQ(partner<SortMode::Bitonic>(n1,  ts_step), ((n1 % 32) ? n1-16 : n1+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n2,  ts_step), ((n1 % 32) ? n2-16 : n2+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n3,  ts_step), ((n1 % 32) ? n3-16 : n3+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n4,  ts_step), ((n1 % 32) ? n4-16 : n4+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n5,  ts_step), ((n1 % 32) ? n5-16 : n5+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n6,  ts_step), ((n1 % 32) ? n6-16 : n6+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n7,  ts_step), ((n1 % 32) ? n7-16 : n7+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n8,  ts_step), ((n1 % 32) ? n8-16 : n8+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n9,  ts_step), ((n1 % 32) ? n9-16 : n9+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n10, ts_step), ((n1 % 32) ? n10-16 : n10+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n11, ts_step), ((n1 % 32) ? n11-16 : n11+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n12, ts_step), ((n1 % 32) ? n12-16 : n12+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n13, ts_step), ((n1 % 32) ? n13-16 : n13+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n14, ts_step), ((n1 % 32) ? n14-16 : n14+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n15, ts_step), ((n1 % 32) ? n15-16 : n15+16));
+        EXPECT_EQ(partner<SortMode::Bitonic>(n16, ts_step), ((n1 % 32) ? n16-16 : n16+16));
+    }
+}
+
+
+
+/* ================================== keepSmall ================================== */
+/*
+ * bool keepSmall(mpi_id_t node, mpi_id_t partner, size_t depth);
+ * Throw check (Not assert - ASSERT_DEATH)
+ */
+TEST(TdistBitonic_UT, keepsmall_test1) {
+    // node and partner must differ or else ...
+    EXPECT_THROW(keepSmall<SortMode::Bitonic>(0, 0, 0), std::runtime_error);
+    EXPECT_THROW(keepSmall<SortMode::Bitonic>(1, 1, 42), std::runtime_error);
+    EXPECT_THROW(keepSmall<SortMode::Bitonic>(7, 7, 42), std::runtime_error);
+}
+
+/*
+ * bool keepsmall(mpi_id_t node, mpi_id_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   = 1;
+    mpi_id_t ts_partner[] = {1, 0, 3, 2, 5, 4, 7, 6};
+    bool    ts_expected[] = {1, 0, 0, 1, 1, 0, 0, 1};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepSmall<SortMode::Bitonic>(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(mpi_id_t node, mpi_id_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   = 2;
+    mpi_id_t ts_partner[] = {2, 3, 0, 1, 6, 7, 4, 5};
+    bool    ts_expected[] = {1, 1, 0, 0, 0, 0, 1, 1};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepSmall<SortMode::Bitonic>(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepsmall(mpi_id_t node, mpi_id_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   = 2;
+    mpi_id_t ts_partner[] = {1, 0, 3, 2, 5, 4, 7, 6};
+    bool    ts_expected[] = {1, 0, 1, 0, 0, 1, 0, 1};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepSmall<SortMode::Bitonic>(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepSmall(mpi_id_t node, mpi_id_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   = 3;
+    mpi_id_t ts_partner[] = {4, 5, 6, 7, 0, 1, 2, 3};
+    bool    ts_expected[] = {1, 1, 1, 1, 0, 0, 0, 0};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepSmall<SortMode::Bitonic>(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepSmall(mpi_id_t node, mpi_id_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   = 3;
+    mpi_id_t ts_partner[] = {2, 3, 0, 1, 6, 7, 4, 5};
+    bool    ts_expected[] = {1, 1, 0, 0, 1, 1, 0, 0};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepSmall<SortMode::Bitonic>(node, ts_partner[node], ts_depth));
+    }
+}
+
+/*
+ * bool keepSmall(mpi_id_t node, mpi_id_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   = 3;
+    mpi_id_t ts_partner[] = {1, 0, 3, 2, 5, 4, 7, 6};
+    bool    ts_expected[] = {1, 0, 1, 0, 1, 0, 1, 0};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node ) {
+        EXPECT_EQ(ts_expected[node], keepSmall<SortMode::Bitonic>(node, ts_partner[node], ts_depth));
+    }
+}
@@ -0,0 +1,144 @@
+/**
+ * \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 "distsort.hpp"
+
+
+
+/* ================================== ascending ================================== */
+
+/*
+ * bool ascending<SortMode::Bubbletonic>(mpi_id_t node, size_t depth);
+ */
+TEST(TdistBubbletonic_UT, ascending_Bubbletonic_test1) {
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(0, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(1, 0), false);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(2, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(3, 0), false);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(4, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(5, 0), false);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(6, 0), true);
+    EXPECT_EQ(ascending<SortMode::Bubbletonic>(7, 0), false);
+
+    for (mpi_id_t node = 0 ; node < 256 ; ++node) {
+        EXPECT_EQ(ascending<SortMode::Bubbletonic>(node, 7), ((node % 2) ? false : true) );
+    }
+}
+
+
+/* ================================== partner ================================== */
+
+/*
+ * mpi_id_t partner<SortMode::Bubbletonic>(mpi_id_t node, size_t step);
+ * step = 0
+ */
+TEST(TdistBubbletonic_UT, partner_Bubbletonic_test1) {
+    size_t ts_step = 0;
+    mpi_id_t ts_expected[] = {1, 0, 3, 2, 5, 4, 7, 6};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node) {
+        EXPECT_EQ(partner<SortMode::Bubbletonic>(node, ts_step), ts_expected[node]);
+    }
+}
+
+/*
+ * mpi_id_t partner<SortMode::Bubbletonic>(mpi_id_t node, size_t step);
+ * step = 1
+ */
+TEST(TdistBubbletonic_UT, partner_Bubbletonic_test2) {
+    size_t ts_step = 1;
+    mpi_id_t ts_expected[] = {(mpi_id_t)-1, 2, 1, 4, 3, 6, 5, 8};
+
+    for (mpi_id_t node = 0 ; node < 8 ; ++node) {
+        EXPECT_EQ(partner<SortMode::Bubbletonic>(node, ts_step), ts_expected[node]);
+    }
+}
+
+/*
+ * mpi_id_t partner<SortMode::Bubbletonic>(mpi_id_t node, size_t step);
+ * various steps
+ */
+TEST(TdistBubbletonic_UT, partner_Bubbletonic_test3) {
+
+    mpi_id_t ts_even_expected[] = {
+            1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14
+    };
+    mpi_id_t ts_odd_expected[]  = {
+            (mpi_id_t)-1, 2, 1, 4, 3, 6, 5, 8, 7, 10, 9, 12, 11, 14, 13, 16
+    };
+
+    for (size_t step = 0 ; step < 32 ; ++step) {
+        if (step % 2) {
+            for (mpi_id_t node = 0; node < 16; ++node) {
+                EXPECT_EQ(partner<SortMode::Bubbletonic>(node, step), ts_odd_expected[node]);
+            }
+        }
+        else {
+            for (mpi_id_t node = 0; node < 16; ++node) {
+                EXPECT_EQ(partner<SortMode::Bubbletonic>(node, step), ts_even_expected[node]);
+            }
+        }
+    }
+}
+
+
+/* ================================== keepSmall ================================== */
+/*
+ * bool keepSmall<SortMode::Bubbletonic>(mpi_id_t node, mpi_id_t partner, size_t depth);
+ * Throw check (Not assert - ASSERT_DEATH)
+ */
+TEST(TdistBubbletonic_UT, keepsmall_test1) {
+    // node and partner must differ or else ...
+    EXPECT_THROW(keepSmall<SortMode::Bubbletonic>(0, 0, 0), std::runtime_error);
+    EXPECT_THROW(keepSmall<SortMode::Bubbletonic>(1, 1, 42), std::runtime_error);
+    EXPECT_THROW(keepSmall<SortMode::Bubbletonic>(7, 7, 42), std::runtime_error);
+}
+
+/*
+ * bool keepSmall<SortMode::Bubbletonic>(mpi_id_t node, mpi_id_t partner, size_t depth);
+ */
+TEST(TdistBubbletonic_UT, keepsmall_test2) {
+    // Check various combinations
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(0, 1, 42), true);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(0, 3, 42), true);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(2, 1, 42), false);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(7, 1, 42), false);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(0, 1, 42), true);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(7, 32, 42), true);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(7, 1, 42), false);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(4, 0, 42), false);
+    EXPECT_EQ(keepSmall<SortMode::Bubbletonic>(4, 9, 42), true);
+}
+
+/* ================================== isActive ================================== */
+/*
+ * bool isActive(mpi_id_t node, size_t nodes);
+ * Throw check
+ */
+TEST(TdistBubbletonic_UT, isActive_test1) {
+    EXPECT_THROW(isActive(0, 0), std::runtime_error);
+    EXPECT_THROW(isActive(0, static_cast<size_t>(std::numeric_limits<mpi_id_t>::max()) + 1), std::runtime_error);
+}
+
+/*
+ * bool isActive(mpi_id_t node, size_t nodes);
+ * Boundary 3 BVA
+ */
+TEST(TdistBubbletonic_UT, isActive_test2) {
+    EXPECT_EQ(isActive(-1, 8), false);
+    EXPECT_EQ(isActive(0, 8), true);
+    EXPECT_EQ(isActive(1, 8), true);
+    EXPECT_EQ(isActive(7, 8), true);
+    EXPECT_EQ(isActive(8, 8), false);
+    EXPECT_EQ(isActive(9, 8), false);
+}
@@ -0,0 +1,93 @@
+/**
+ * \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 "distsort.hpp"
+
+
+/* ================================== fullSort ================================== */
+
+/*
+ *
+ */
+TEST(TdistCommonUT, fullSort_test1) {
+    std::vector<uint8_t> ts_data     = {3, 2, 1, 4, 5, 7, 8, 6};
+    std::vector<uint8_t> ts_expected = {1, 2, 3, 4, 5, 6, 7, 8};
+    bool ts_ascending = true;
+
+    fullSort(ts_data, ts_ascending);
+    EXPECT_EQ((ts_data == ts_expected), true);
+}
+
+TEST(TdistCommonUT, fullSort_test2) {
+    std::vector<uint8_t> ts_data     = {3, 2, 1, 4, 5, 7, 8, 6};
+    std::vector<uint8_t> ts_expected = {8, 7, 6, 5, 4, 3, 2, 1};
+    bool ts_ascending = false;
+
+    fullSort(ts_data, ts_ascending);
+    EXPECT_EQ((ts_data == ts_expected), true);
+}
+
+/* ================================== elbowSort ================================== */
+
+TEST(TdistCommonUT, elbowSort_test1) {
+    ShadowedVec_t<uint8_t> ts_data1(std::vector<uint8_t>{3, 2, 1, 4, 5, 5, 7, 8});
+    ShadowedVec_t<uint8_t> ts_data2(std::vector<uint8_t>{4, 5, 7, 8, 5, 3, 2, 1});
+    ShadowedVec_t<uint8_t> ts_data3(std::vector<uint8_t>{1, 2, 3, 4, 5, 5, 7, 8});
+    ShadowedVec_t<uint8_t> ts_data4(std::vector<uint8_t>{8, 7, 5, 5, 4, 3, 2, 1});
+    std::vector<uint8_t> ts_expected = {1, 2, 3, 4, 5, 5, 7, 8};
+    bool ts_ascending = true;
+
+    elbowSort(ts_data1, ts_ascending);
+    elbowSort(ts_data2, ts_ascending);
+    elbowSort(ts_data3, ts_ascending);
+    elbowSort(ts_data4, ts_ascending);
+    EXPECT_EQ((ts_data1 == ts_expected), true);
+    EXPECT_EQ((ts_data2 == ts_expected), true);
+    EXPECT_EQ((ts_data3 == ts_expected), true);
+    EXPECT_EQ((ts_data4 == ts_expected), true);
+}
+
+TEST(TdistCommonUT, elbowSort_test2) {
+    ShadowedVec_t<uint8_t> ts_data1(std::vector<uint8_t>{3, 2, 1, 4, 5, 5, 7, 8});
+    ShadowedVec_t<uint8_t> ts_data2(std::vector<uint8_t>{4, 5, 7, 8, 5, 3, 2, 1});
+    ShadowedVec_t<uint8_t> ts_data3(std::vector<uint8_t>{1, 2, 3, 4, 5, 5, 7, 8});
+    ShadowedVec_t<uint8_t> ts_data4(std::vector<uint8_t>{8, 7, 5, 5, 4, 3, 2, 1});
+    std::vector<uint8_t> ts_expected = {8, 7, 5, 5, 4, 3, 2, 1};
+    bool ts_ascending = false;
+
+    elbowSort(ts_data1, ts_ascending);
+    elbowSort(ts_data2, ts_ascending);
+    elbowSort(ts_data3, ts_ascending);
+    elbowSort(ts_data4, ts_ascending);
+    EXPECT_EQ((ts_data1 == ts_expected), true);
+    EXPECT_EQ((ts_data2 == ts_expected), true);
+    EXPECT_EQ((ts_data3 == ts_expected), true);
+    EXPECT_EQ((ts_data4 == ts_expected), true);
+}
+
+TEST(TdistCommonUT, elbowSort_test3) {
+    ShadowedVec_t<uint8_t> ts_data(std::vector<uint8_t>{8, 7, 5, 5, 4, 3, 2, 1});
+    std::vector<uint8_t> ts_expected_asc = {1, 2, 3, 4, 5, 5, 7, 8};
+    std::vector<uint8_t> ts_expected_des = {8, 7, 5, 5, 4, 3, 2, 1};
+
+    // Check alternation for active-shadow vector inside Buffer and elbow algorithm
+    elbowSort(ts_data, true);
+    EXPECT_EQ((ts_data == ts_expected_asc), true);
+    elbowSort(ts_data, false);
+    EXPECT_EQ((ts_data == ts_expected_des), true);
+    elbowSort(ts_data, true);
+    EXPECT_EQ((ts_data == ts_expected_asc), true);
+    elbowSort(ts_data, false);
+    EXPECT_EQ((ts_data == ts_expected_des), true);
+}
+
@@ -0,0 +1,210 @@
+/**
+ * \file
+ * \brief   PDS HW2 tests
+ *
+ * To run these test execute:
+ *      make tests
+ *      mpirun -np <N> ./out/tests
+ *
+ * Note:
+ *  Yes each process runs the entire test suite!!
+ *
+ * \author
+ *    Christos Choutouridis AEM:8997
+ *    <cchoutou@ece.auth.gr>
+ */
+
+#include <gtest/gtest.h>
+#include <mpi.h>
+#include <random>
+#include "distsort.hpp"
+/*
+ * Global fixtures
+ */
+
+// MPI handler for the test session
+MPI_t<>         ts_mpi;
+
+// Mersenne seeded from hw if possible. range: [type_min, type_max]
+std::random_device rd;
+std::mt19937 gen(rd());
+
+class TMPIdistSort : public ::testing::Test {
+protected:
+    static void SetUpTestSuite() {
+        int argc = 0;
+        char** argv = nullptr;
+        ts_mpi.init(&argc, &argv);
+    }
+
+    static void TearDownTestSuite() {
+        ts_mpi.finalize();
+    }
+};
+
+
+/*
+ * MPI: SysTest (acceptance)
+ * Each process executes distBubbletonic for uin8_t [16]
+ */
+TEST_F(TMPIdistSort, distBubbletonic_test1) {
+    // Create and fill vector
+    using tsValue_t = uint8_t;      // Test parameters
+    size_t ts_buffer_size = 16;
+
+    ShadowedVec_t<tsValue_t> ts_Data;
+    std::uniform_int_distribution<tsValue_t > dis(
+            std::numeric_limits<tsValue_t>::min(),
+            std::numeric_limits<tsValue_t>::max()
+    );
+    ts_Data.resize(ts_buffer_size);
+    std::generate(ts_Data.begin(), ts_Data.end(), [&]() { return dis(gen); });
+
+    // Execute function under test in all processes
+    distBubbletonic(ts_Data, ts_mpi.size(), ts_mpi.rank());
+
+    // Local min and max
+    auto local_min = *std::min_element(ts_Data.begin(), ts_Data.end());
+    auto local_max = *std::max_element(ts_Data.begin(), ts_Data.end());
+
+    // Gather min/max to rank 0
+    std::vector<tsValue_t> global_mins(ts_mpi.size());
+    std::vector<tsValue_t> global_maxes(ts_mpi.size());
+    MPI_Datatype datatype = MPI_TypeMapper<tsValue_t>::getType();
+
+    MPI_Gather(&local_min, 1, datatype, global_mins.data(), 1, datatype, 0, MPI_COMM_WORLD);
+    MPI_Gather(&local_max, 1, datatype, global_maxes.data(), 1, datatype, 0, MPI_COMM_WORLD);
+
+    // Check results
+    EXPECT_EQ(std::is_sorted(ts_Data.begin(), ts_Data.end()), true);
+    if (ts_mpi.rank() == 0) {
+        for (size_t i = 1; i < global_mins.size(); ++i) {
+            EXPECT_LE(global_maxes[i - 1], global_mins[i]);
+        }
+    }
+}
+
+/*
+ * MPI: SysTest (acceptance)
+ * Each process executes distBubbletonic for uin32_t [1 << 16]
+ */
+TEST_F(TMPIdistSort, distBubbletonic_test2) {
+    // Create and fill vector
+    using tsValue_t = uint32_t;      // Test parameters
+    size_t ts_buffer_size = 1 << 16;
+
+    ShadowedVec_t<tsValue_t> ts_Data;
+    std::uniform_int_distribution<tsValue_t > dis(
+            std::numeric_limits<tsValue_t>::min(),
+            std::numeric_limits<tsValue_t>::max()
+    );
+    ts_Data.resize(ts_buffer_size);
+    std::generate(ts_Data.begin(), ts_Data.end(), [&]() { return dis(gen); });
+
+    // Execute function under test in all processes
+    distBubbletonic(ts_Data, ts_mpi.size(), ts_mpi.rank());
+
+    // Local min and max
+    auto local_min = *std::min_element(ts_Data.begin(), ts_Data.end());
+    auto local_max = *std::max_element(ts_Data.begin(), ts_Data.end());
+
+    // Gather min/max to rank 0
+    std::vector<tsValue_t> global_mins(ts_mpi.size());
+    std::vector<tsValue_t> global_maxes(ts_mpi.size());
+    MPI_Datatype datatype = MPI_TypeMapper<tsValue_t>::getType();
+
+    MPI_Gather(&local_min, 1, datatype, global_mins.data(), 1, datatype, 0, MPI_COMM_WORLD);
+    MPI_Gather(&local_max, 1, datatype, global_maxes.data(), 1, datatype, 0, MPI_COMM_WORLD);
+
+    // Check results
+    EXPECT_EQ(std::is_sorted(ts_Data.begin(), ts_Data.end()), true);
+    if (ts_mpi.rank() == 0) {
+        for (size_t i = 1; i < global_mins.size(); ++i) {
+            EXPECT_LE(global_maxes[i - 1], global_mins[i]);
+        }
+    }
+}
+
+
+/*
+ * MPI: SysTest (acceptance)
+ * Each process executes distBitonic for uin8_t [16]
+ */
+TEST_F(TMPIdistSort, distBitonic_test1) {
+    // Create and fill vector
+    using tsValue_t = uint8_t;      // Test parameters
+    size_t ts_buffer_size = 16;
+
+    ShadowedVec_t<tsValue_t> ts_Data;
+    std::uniform_int_distribution<tsValue_t > dis(
+            std::numeric_limits<tsValue_t>::min(),
+            std::numeric_limits<tsValue_t>::max()
+    );
+    ts_Data.resize(ts_buffer_size);
+    std::generate(ts_Data.begin(), ts_Data.end(), [&]() { return dis(gen); });
+
+    // Execute function under test in all processes
+    distBitonic(ts_Data, ts_mpi.size(), ts_mpi.rank());
+
+    // Local min and max
+    auto local_min = *std::min_element(ts_Data.begin(), ts_Data.end());
+    auto local_max = *std::max_element(ts_Data.begin(), ts_Data.end());
+
+    // Gather min/max to rank 0
+    std::vector<tsValue_t> global_mins(ts_mpi.size());
+    std::vector<tsValue_t> global_maxes(ts_mpi.size());
+    MPI_Datatype datatype = MPI_TypeMapper<tsValue_t>::getType();
+
+    MPI_Gather(&local_min, 1, datatype, global_mins.data(), 1, datatype, 0, MPI_COMM_WORLD);
+    MPI_Gather(&local_max, 1, datatype, global_maxes.data(), 1, datatype, 0, MPI_COMM_WORLD);
+
+    // Check results
+    EXPECT_EQ(std::is_sorted(ts_Data.begin(), ts_Data.end()), true);
+    if (ts_mpi.rank() == 0) {
+        for (size_t i = 1; i < global_mins.size(); ++i) {
+            EXPECT_LE(global_maxes[i - 1], global_mins[i]);
+        }
+    }
+}
+
+/*
+ * MPI: SysTest (acceptance)
+ * Each process executes distBitonic for uin32_t [1 << 16]
+ */
+TEST_F(TMPIdistSort, distBitonic_test2) {
+    // Create and fill vector
+    using tsValue_t = uint32_t;      // Test parameters
+    size_t ts_buffer_size = 1 << 16;
+
+    ShadowedVec_t<tsValue_t> ts_Data;
+    std::uniform_int_distribution<tsValue_t > dis(
+            std::numeric_limits<tsValue_t>::min(),
+            std::numeric_limits<tsValue_t>::max()
+    );
+    ts_Data.resize(ts_buffer_size);
+    std::generate(ts_Data.begin(), ts_Data.end(), [&]() { return dis(gen); });
+
+    // Execute function under test in all processes
+    distBitonic(ts_Data, ts_mpi.size(), ts_mpi.rank());
+
+    // Local min and max
+    auto local_min = *std::min_element(ts_Data.begin(), ts_Data.end());
+    auto local_max = *std::max_element(ts_Data.begin(), ts_Data.end());
+
+    // Gather min/max to rank 0
+    std::vector<tsValue_t> global_mins(ts_mpi.size());
+    std::vector<tsValue_t> global_maxes(ts_mpi.size());
+    MPI_Datatype datatype = MPI_TypeMapper<tsValue_t>::getType();
+
+    MPI_Gather(&local_min, 1, datatype, global_mins.data(), 1, datatype, 0, MPI_COMM_WORLD);
+    MPI_Gather(&local_max, 1, datatype, global_maxes.data(), 1, datatype, 0, MPI_COMM_WORLD);
+
+    // Check results
+    EXPECT_EQ(std::is_sorted(ts_Data.begin(), ts_Data.end()), true);
+    if (ts_mpi.rank() == 0) {
+        for (size_t i = 1; i < global_mins.size(); ++i) {
+            EXPECT_LE(global_maxes[i - 1], global_mins[i]);
+        }
+    }
+}
+
Author	SHA1	Message	Date
hoo2	0e44a3062c	HW2: RC3 - Batch scripts added	2025-01-04 19:25:13 +02:00
hoo2	4e7237bd4d	HW2: RC3 - A small exchange optimization	2025-01-04 16:56:34 +02:00
hoo2	bc98ef4e9f	HW2: RC2 - Parallel Full Sort	2025-01-03 16:15:13 +02:00
hoo2	1bd0cbb8d0	HW2: RC1 - Model version	2025-01-03 14:45:16 +02:00
hoo2	4dc47bb8f4	HW2: Add perf rule	2024-12-31 14:26:53 +02:00
hoo2	36de8d94fc	HW2: A validator added and small changes	2024-12-31 12:54:58 +02:00
hoo2	9dd3eb737f	HW2: Added some test and fix bubblesort <number of iteration> bug	2024-12-30 22:03:30 +02:00
hoo2	c5ffec9cca	HW2: First locally working version of the 2 variations	2024-12-30 19:12:55 +02:00
hoo2	4d1d7502aa	HW2: (WIP) Checkpoint with code re-arrangement and elbowSort	2024-12-30 14:23:13 +02:00
hoo2	1d6b271d2a	HW2: Add an MPI wrapper for the basic functionality and update all types	2024-12-28 17:36:00 +02:00
hoo2	496ee69f54	HW2: A local version of distbubbletonic added	2024-12-28 14:53:13 +02:00
hoo2	43dff95f67	HW2: A local only version of the distbitonic	2024-12-28 01:40:43 +02:00
hoo2	428b50aab7	HW2: First MPI test on AUTh's cluster	2024-12-27 11:56:26 +02:00
hoo2	058df095b0	HW2: Setup build env and basic code setup	2024-12-26 12:04:24 +02:00
hoo2	64dc18280c	HW2: Init C++ project	2024-12-25 10:02:48 +02:00
hoo2	61b63016cc	HW2: Add julia implementations for v0.5 and v1	2024-12-24 17:58:10 +02:00
hoo2	955981d177	Refresh the report and better commenting	2024-11-21 22:56:37 +02:00
hoo2	381a5737db	Copy report to the repo also (no code changes only comments)	2024-11-21 15:18:39 +02:00
hoo2	494fdddd04	v1 version and measurements	2024-11-20 21:56:04 +02:00