Addendum Cholesky Benchmark and GPU AMD with Specx

src/Benchmark/Cholesky/Cholesky_GPU_HIP/CMakeLists.txt

@@ -0,0 +1,190 @@
+
+
+project(main)
+
+cmake_minimum_required(VERSION 3.22)
+
+#set(CMAKE_BUILD_TYPE Debug)
+set(CMAKE_BUILD_TYPE Release)
+
+
+
+#set(CMAKE_CXX_COMPILER "/usr/local/bin/g++")
+#set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD 20)
+#set(CMAKE_CXX_STANDARD 23)
+
+#set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MD /02")
+#string(REGEX REPLACE "/MD" "/02" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
+
+
+if(FALSE)
+    find_package(MPI REQUIRED)
+    if (MPI_FOUND)
+        MESSAGE("{MPI_CXX_LIBRARIES}")
+        else (MPI_FOUND)
+        MESSAGE (SEND_ERROR "This application cannot compile without MPI")
+    endif(MPI_FOUND)
+endif()
+
+if(FALSE)
+    find_package(OpenMP)
+    if (OpenMP_CXX_FOUND)
+        MESSAGE("{OpenMP_CXX_LIBRARIES}")
+        else (OpenMP_CXX_FOUND)
+        MESSAGE (SEND_ERROR "This application cannot compile without OpenMP")
+    endif(OpenMP_CXX_FOUND)
+endif()
+
+if(FALSE)
+    find_package(CUDA REQUIRED)
+    if (CUDA_FOUND)
+            MESSAGE("{CUDA_CXX_LIBRARIES}")
+            MESSAGE(STATUS "Found headers CUDA : ${CUDA_INCLUDE_DIRS}")
+            MESSAGE(STATUS "Found lib CUDA : ${CUDA_LIBRARIES}")
+            MESSAGE(STATUS "Found CUDA nvcc : ${CUDA_NVCC_EXECUTABLE}")
+            else (CUDA_FOUND)
+            MESSAGE (SEND_ERROR "This application cannot compile without CUDA")
+        endif(CUDA_FOUND)
+
+    add_definitions(-D_FORCE_INLINES)
+    #set (CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS} --gpu-architecture sm_21 -std=c++11)
+    set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS}; -O3 )
+endif()
+
+
+if(TRUE)
+    find_package(Eigen3 REQUIRED)
+    include_directories(${EIGEN3_INCLUDE_DIR})
+    message("${CMAKE_MODULE_PATH}")
+    message("${EIGEN3_INCLUDE_DIR}")
+endif()
+
+
+#set(THREADS_PREFER_PTHREAD_FLAG ON)
+#find_package(Threads REQUIRED)
+find_package(Threads)
+
+
+#find_package(TBB REQUIRED)
+#if (TBB_FOUND)
+#    MESSAGE("{TBB_CXX_LIBRARIES}")
+#endif(TBB_FOUND)
+
+
+MESSAGE("{COMPILE_WITH_HIP}")
+#include_directories(/opt/rocm/hip/include)
+include_directories(/opt/rocm-5.7.2/hip/include)
+#list(APPEND CMAKE_PREFIX_PATH /opt/rocm/hip /opt/rocm)
+list(APPEND CMAKE_PREFIX_PATH /opt/rocm-5.7.2 /opt/rocm-5.7.2)
+# Find hip
+#find_package(hip)
+# Set the GPU to compile for
+#set(GPU_TARGETS "gfx900:gfx906:gfx1032;gfx1035" CACHE STRING "GPU targets to compile for")
+
+set(GPU_RUNTIME "HIP" CACHE STRING "Switches between HIP and CUDA")
+set(GPU_RUNTIMES "HIP" "CUDA")
+set_property(CACHE GPU_RUNTIME PROPERTY STRINGS ${GPU_RUNTIMES})
+
+if(NOT "${GPU_RUNTIME}" IN_LIST GPU_RUNTIMES)
+    set(ERROR_MESSAGE "GPU_RUNTIME is set to \"${GPU_RUNTIME}\".\nGPU_RUNTIME must be either HIP or CUDA.")
+    message(FATAL_ERROR ${ERROR_MESSAGE})
+endif()
+
+enable_language(${GPU_RUNTIME})
+
+set(CMAKE_${GPU_RUNTIME}_STANDARD 17)
+set(CMAKE_${GPU_RUNTIME}_EXTENSIONS OFF)
+set(CMAKE_${GPU_RUNTIME}_STANDARD_REQUIRED ON)
+
+#set(ROCM_ROOT "/opt/rocm" CACHE PATH "Root directory of the ROCm installation")
+set(ROCM_ROOT "/opt/rocm-5.7.2" CACHE PATH "Root directory of the ROCm installation")
+list(APPEND CMAKE_PREFIX_PATH "${ROCM_ROOT}")
+
+
+message(CURRENT DIR= [${CMAKE_CURRENT_SOURCE_DIR}])
+
+#BEGIN::SPECX
+
+file(GLOB_RECURSE Specx_list 
+    #"${CMAKE_CURRENT_SOURCE_DIR}/../../specxpp/*.cpp"
+    "${CMAKE_CURRENT_SOURCE_DIR}/../../../Specx20/specx/*.hpp"
+)
+
+if (TRUE)
+    foreach(file_path ${Specx_list})
+        message(${file_path})
+    endforeach()
+endif()
+
+
+MACRO(HEADER_DIRECTORIES return_list)
+    FILE(GLOB_RECURSE new_list ${CMAKE_CURRENT_SOURCE_DIR}/../../../Specx20/specx/Src/*.hpp)
+    SET(dir_list "")
+    FOREACH(file_path ${new_list})
+        GET_FILENAME_COMPONENT(dir_path ${file_path} PATH)
+        SET(dir_list ${dir_list} ${dir_path})
+    ENDFOREACH()
+    LIST(REMOVE_DUPLICATES dir_list)
+    SET(${return_list} ${dir_list})
+ENDMACRO()
+
+HEADER_DIRECTORIES(header_specx_dir_list)
+
+list(LENGTH header_specx_dir_list header_dir_list_count)
+message(STATUS "[INFO] Found ${header_dir_list_count} header directories.")
+
+if (TRUE)
+    foreach(dir_path ${header_specx_dir_list})
+        message(${dir_path})
+    endforeach()
+endif()
+
+#include_directories("/home/u2/lemoine/SpecxProjects/specxpp/build/Src") 
+
+include_directories(${CMAKE_CURRENT_SOURCE_DIR}/../../../Specx20/specx/build/Src) 
+include_directories(${CMAKE_CURRENT_SOURCE_DIR}/../../../Specx20/specx/Src) 
+
+#END::SPECX
+
+include_directories(${CMAKE_CURRENT_SOURCE_DIR}/../Test13)
+
+set(include_dirs "../Test13")
+
+file(GLOB SOURCES
+    "*.cpp"
+)
+
+add_executable(main ${SOURCES} ${Specx_list})
+
+#/home/u2/lemoine/Specx20/specx/build/Src/libspecx.a
+
+target_link_libraries(main
+    ${PTHREADS}
+    PRIVATE
+        ${CMAKE_CURRENT_SOURCE_DIR}/../../../Specx20/specx/build/Src/libspecx.a 
+    #PUBLIC
+        #TBB::tbb
+    #PRIVATE
+    #    pthread
+)
+
+#target_include_directories(MiniTest PRIVATE "/home/u2/lemoine/SpecxProjects/specxpp")
+
+#if(GPU_RUNTIME STREQUAL "CUDA")
+#    list(APPEND include_dirs "${ROCM_ROOT}/include")
+#endif()
+
+if(GPU_RUNTIME STREQUAL "HIP")
+    list(APPEND include_dirs "${ROCM_ROOT}/include")
+    MESSAGE(${ROCM_ROOT}/include)
+endif()
+
+set_source_files_properties(main.cpp PROPERTIES LANGUAGE ${GPU_RUNTIME})
+
+target_include_directories(
+    main
+    PRIVATE
+        ${header_specx_dir_list}
+)
+

src/Benchmark/Cholesky/Cholesky_GPU_HIP/DataAnalysis.py

@@ -0,0 +1,53 @@
+import pandas as pd
+import csv
+import json
+import matplotlib.pyplot as plt
+
+df = pd.read_csv('Data.csv')
+#print(df.head())
+#print(df) 
+x=df['DimMatrix']
+nb=len(x)
+y1=df['ModeSerial']
+y2=df['ModeGpuHipAMD']
+y3=df['pthread']
+
+for i in range(0,nb):
+    sc=1000.0
+    y1[i]=y1[i]/sc
+    y2[i]=y2[i]/sc
+    y3[i]=y3[i]/sc
+
+figure1 = plt.figure()
+plt.plot(x,y1,label='CPU-Serial',marker='o', linestyle='--', color='r')  
+plt.plot(x,y2,label='HIP-AMD-GPU',marker='o', linestyle='--', color='g')
+plt.plot(x,y3,label='pthread 9',marker='o', linestyle='--', color='b')
+plt.grid()
+plt.xlabel('Dim Matrix')
+#plt.ylabel('Time in microsecond')
+plt.ylabel('Time in millisecond')
+plt.title('Cholesky factorisation resolution')
+plt.xticks(rotation=90)
+plt.legend()
+plt.show()
+plt.savefig("Data.jpg")
+
+
+ratio=[]
+for i in range(0,nb):
+    ratio.append(y1[i]/y2[i])
+
+figure2 = plt.figure()
+plt.plot(x,ratio,label='Ratio',marker='o', linestyle='--', color='r')  
+plt.grid()
+plt.xlabel('Dim Matrix')
+plt.ylabel('Ratio')
+plt.title('Cholesky factorisation resolution')
+plt.xticks(rotation=90)
+plt.legend()
+plt.show()
+plt.savefig("DataRatio.jpg")
+
+
+
+