diff --git a/CMakeLists.txt b/CMakeLists.txt
index 909272b9d870330487d2bf2ae6bf4211aaf0a60e..8e88df1d441cb1cb847d01de13b0ca352f61ffae 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -6,6 +6,10 @@ SET(CMAKE_INCLUDE_CURRENT_DIR ON)
 SET(CMAKE_CXX_STANDARD 11)
 SET(CMAKE_C_STANDARD 11)
 
+# get system architecture
+EXECUTE_PROCESS(COMMAND uname -m COMMAND tr -d '\n' OUTPUT_VARIABLE ARCHITECTURE)
+set(CMAKE_LIBRARY_ARCHITECTURE ${ARCHITECTURE})
+
 # fix win stack size issue
 if(WIN32)
     SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wl,--stack,10000000")
@@ -52,68 +56,99 @@ IF(NOT BOOST_LIB)
 ENDIF()
 INCLUDE_DIRECTORIES("${BOOST_LIB}")
 
-SET(MKLROOT "$ENV{MKLROOT}")
-IF(NOT MKLROOT)
-    MESSAGE(FATAL_ERROR "Specify environment variable MKLROOT to the location of Intel MKL package")
-ENDIF()
-INCLUDE_DIRECTORIES("${MKLROOT}/include")
-
 # PLINK2.0 pgen library;
 INCLUDE_DIRECTORIES("submods/plink-ng/2.0")
 
 if(WIN32)
     set(CMAKE_FIND_LIBRARY_SUFFIXES ".lib")
 else()
-    set(CMAKE_FIND_LIBRARY_SUFFIXES ".a")
+    set(CMAKE_FIND_LIBRARY_SUFFIXES ".a" ".so")
 endif()
 
-find_library(lib_lp64 NAMES mkl_intel_lp64 PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
-if(CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR APPLE)
-    find_library(lib_thread NAMES mkl_intel_thread PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
-else()
-    find_library(lib_thread NAMES mkl_gnu_thread PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
-endif()
-find_library(lib_core NAMES mkl_core PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
-set(MKL_LIB ${lib_lp64} ${lib_thread} ${lib_core})
-
 IF(NOT CMAKE_BUILD_TYPE)
     set(CMAKE_BUILD_TYPE RELEASE)
     message("Setting to release build")
 ENDIF()
-IF(NOT SSE)
-    set(SSE "-msse2")
-    message("SSE build")
-ENDIF()
 
-if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 ${SSE} -DMKL_LP64 -pthread")
-    set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -DNDEBUG -DEIGEN_USE_MKL_ALL")
-    set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -g")
-    add_compile_options(-fopenmp)
-    set(CMAKE_EXE_LINKER_FLAGS "-Wl,-no_implicit_dylibs")
-    if(NOT APPLE)
-        # don't delete this it will result in _implibs
+message(STATUS "ARCHITECTURE:${CMAKE_LIBRARY_ARCHITECTURE}")
+if(CMAKE_LIBRARY_ARCHITECTURE STREQUAL "aarch64")
+    # Use blas library of KML and lapack of openblas
+    # TODO, connect lapack library of kml when kml_lapck completed
+    SET(KMLROOT "$ENV{KMLROOT}")
+    IF(NOT KMLROOT)
+        MESSAGE(FATAL_ERROR "Specify environment variable KMLROOT to the location of KML package")
+    ENDIF()
+    INCLUDE_DIRECTORIES("${KMLROOT}/include")
+
+    SET(OPENBLAS "$ENV{OPENBLAS}")
+    IF(NOT OPENBLAS)
+        MESSAGE(FATAL_ERROR "Specify environment variable OPENBLAS to the location of OPENBLAS package")
+    ENDIF()
+    INCLUDE_DIRECTORIES("${OPENBLAS}/include")
+
+    find_library(lib_blas64 NAMES kblas PATHS "${KMLROOT}/lib/kblas/nolocking" "${KMLROOT}/lib")
+    find_library(lib_openblas NAMES openblas PATHS "${OPENBLAS}" "${OPENBLAS}/lib")
+    set(KML_LIB ${lib_blas64} ${lib_openblas})
+
+    if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -I ${CMAKE_CURRENT_SOURCE_DIR} -pthread -fvisibility=hidden -fvisibility-inlines-hidden")
+        set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -DNDEBUG") #-flto 
+        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -fno-inline -fno-implicit-inline-templates -g3")
+        add_compile_options(-fopenmp)
         set(CMAKE_EXE_LINKER_FLAGS "-static-libgcc -static-libstdc++")
+    else()
+        message(FATAL_ERROR "In aarch64 architecture, please use GNU compiler")
+    endif()
+else()
+    SET(MKLROOT "$ENV{MKLROOT}")
+    IF(NOT MKLROOT)
+        MESSAGE(FATAL_ERROR "Specify environment variable MKLROOT to the location of Intel MKL package")
+    ENDIF()
+    INCLUDE_DIRECTORIES("${MKLROOT}/include")
+
+    find_library(lib_lp64 NAMES mkl_intel_lp64 PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
+    if(CMAKE_CXX_COMPILER_ID MATCHES "Clang" OR APPLE)
+        find_library(lib_thread NAMES mkl_intel_thread PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
+    else()
+        find_library(lib_thread NAMES mkl_gnu_thread PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
+    endif()
+    find_library(lib_core NAMES mkl_core PATHS "${MKLROOT}/lib/intel64" "${MKLROOT}/lib")
+    set(MKL_LIB ${lib_lp64} ${lib_thread} ${lib_core})
+
+    IF(NOT SSE)
+        set(SSE "-msse2")
+        message("SSE build")
+    ENDIF()
+
+    if("${CMAKE_CXX_COMPILER_ID}" MATCHES "Clang")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 ${SSE} -DMKL_LP64 -pthread")
+        set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -DNDEBUG -DEIGEN_USE_MKL_ALL")
+        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -g")
+        add_compile_options(-fopenmp)
+        set(CMAKE_EXE_LINKER_FLAGS "-Wl,-no_implicit_dylibs")
+        if(NOT APPLE)
+            # don't delete this it will result in _implibs
+            set(CMAKE_EXE_LINKER_FLAGS "-static-libgcc -static-libstdc++")
+        endif()
+    elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 ${SSE} -pthread -DMKL_LP64 -fvisibility=hidden -fvisibility-inlines-hidden")
+        set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -DNDEBUG -DEIGEN_USE_MKL_ALL") #-flto 
+        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -fno-inline -fno-implicit-inline-templates -g3")
+        add_compile_options(-fopenmp)
+        set(CMAKE_EXE_LINKER_FLAGS "-static-libgcc -static-libstdc++")
+    elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Intel")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -openmp -axAVX2 -pthread")
+        set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -ipo -DNDEBUG")
+        set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -g")
+    elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
+        message(FALTAL_ERROR "GCTA can not compile by Microsft C++ compiler in CMake currently, please use gcta_win64 folder to build")
     endif()
-elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 ${SSE} -pthread -DMKL_LP64 -fvisibility=hidden -fvisibility-inlines-hidden")
-    set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -DNDEBUG -DEIGEN_USE_MKL_ALL") #-flto 
-    set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -fno-inline -fno-implicit-inline-templates -g3")
-    add_compile_options(-fopenmp)
-    set(CMAKE_EXE_LINKER_FLAGS "-static-libgcc -static-libstdc++")
-elseif("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Intel")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -openmp -axAVX2 -pthread")
-    set(CMAKE_CXX_FLAGS_RELEASE  "${CMAKE_CXX_FLAGS_RELEASE} -O3 -ipo -DNDEBUG")
-    set(CMAKE_CXX_FLAGS_DEBUG  "${CMAKE_CXX_FLAGS_DEBUG} -O0 -g")
-elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
-    message(FALTAL_ERROR "GCTA can not compile by Microsft C++ compiler in CMake currently, please use gcta_win64 folder to build")
 endif()
 
 set(COMMON_INCLUDES ${PROJECT_SOURCE_DIR}/include)
 set(MAIN_SOURCE "${PROJECT_SOURCE_DIR}/src/main.cpp")
 include_directories(${COMMON_INCLUDES})
 
-
 file(GLOB SRCS "${PROJECT_SOURCE_DIR}/src/*.cpp")
 list(REMOVE_ITEM SRCS "${MAIN_SOURCE}")
 
@@ -132,7 +167,7 @@ if(APPLE)
     link_directories(${MKLROOT}/lib)
     link_directories(/usr/local/lib)
 else(CMAKE_CXX_COMPILER_ID MATCHES "Clang")
-    #linux
+    #linux for amd64
     link_directories(${MKLROOT}/../../../lib/intel64)
 endif()
 
@@ -150,8 +185,12 @@ else()
         #target_link_libraries(gcta64 mainV1 ${libs_list} -Wl,-Bstatic z -Wl,--start-group ${MKL_LIB} -Wl,--end-group iomp5 -Wl,-Bdynamic pthread m dl)
         target_link_libraries(gcta64 mainV1 ${libs_list} Pgenlib -static z sqlite3 zstd -Wl,--start-group ${MKL_LIB} -Wl,--end-group iomp5 -Wl,--whole-archive -lpthread -Wl,--no-whole-archive m dl)
     else()
-        #target_link_libraries(gcta64 mainV1 ${libs_list} -Wl,-Bstatic z -Wl,--start-group ${MKL_LIB} -Wl,--end-group gomp -Wl,-Bdynamic pthread m dl)
-        target_link_libraries(gcta64 mainV1 ${libs_list} Pgenlib -static z sqlite3 zstd -Wl,--start-group ${MKL_LIB} -Wl,--end-group gomp -Wl,--whole-archive -lpthread -Wl,--no-whole-archive m dl)
+        if(CMAKE_LIBRARY_ARCHITECTURE STREQUAL "aarch64")
+            target_link_libraries(gcta64 mainV1 ${libs_list} Pgenlib -static z sqlite3 zstd -Wl,--start-group ${KML_LIB} -lgfortran -Wl,--end-group gomp -Wl,--whole-archive -lpthread -Wl,--no-whole-archive m dl)
+        else()
+            #target_link_libraries(gcta64 mainV1 ${libs_list} -Wl,-Bstatic z -Wl,--start-group ${MKL_LIB} -Wl,--end-group gomp -Wl,-Bdynamic pthread m dl)
+            target_link_libraries(gcta64 mainV1 ${libs_list} Pgenlib -static z sqlite3 zstd -Wl,--start-group ${MKL_LIB} -Wl,--end-group gomp -Wl,--whole-archive -lpthread -Wl,--no-whole-archive m dl)
+        endif()
     endif()
 endif()
 
diff --git a/include/Matrix.hpp b/include/Matrix.hpp
index f31aa83369efd5033ad5c172cd46522aa6a1f96c..3e3038be646544c64f81b6bd4b5e475276ad5711 100644
--- a/include/Matrix.hpp
+++ b/include/Matrix.hpp
@@ -6,6 +6,12 @@
 #include <iostream>
 #include <Logger.h>
 
+#if defined(__aarch64__)
+    #include <lapack.h>
+#else
+    #include <mkl.h>
+#endif
+
 static_assert(std::numeric_limits<double>::is_iec559, "Not a supported compiler");
 
 // two step to inverse the matrix
@@ -29,12 +35,20 @@ bool _LLT(MatrixType &A, double &logdet){
     int info, cols = (int)A.cols();
     char uplo = 'L';
     LOGGER.ts("LLT");
-    dpotrf(&uplo, &cols, vi, &cols, &info);
+    #if defined(__aarch64__)
+        dpotrf_(&uplo, &cols, vi, &cols, &info);
+    #else
+        dpotrf(&uplo, &cols, vi, &cols, &info);
+    #endif
     //LOGGER << "  LLT time: " << LOGGER.tp("LLT") << std::endl;
     if(info == 0){
         logdet = A.diagonal().array().square().log().sum();
         //LOGGER.ts("LLT_INV");
-        dpotri(&uplo, &cols, vi, &cols, &info);
+        #if defined(__aarch64__)
+            dpotri_(&uplo, &cols, vi, &cols, &info);
+        #else
+            dpotri(&uplo, &cols, vi, &cols, &info);
+        #endif
         //LOGGER << "  LLT inverse time: " << LOGGER.tp("LLT_INV") << std::endl;
         if(info == 0){
             A.template triangularView<Eigen::Upper>() = A.transpose();
diff --git a/main/gcta.h b/main/gcta.h
index a6049e8453dbf07b7ff122755faa211edc62f6d7..2a589d957f0777a848abf0bca01ff1c9abf54ec6 100644
--- a/main/gcta.h
+++ b/main/gcta.h
@@ -37,11 +37,17 @@
 #include <unsupported/Eigen/SparseExtra>
 #include <unsupported/Eigen/IterativeSolvers>
 #include <omp.h>
-#include <mkl_cblas.h>
-#include <mkl_lapack.h>
 #include "Logger.h"
 #include "Matrix.hpp"
 
+#if defined(__aarch64__)
+    #include <kblas.h>
+    #include <lapack.h>
+#else
+    #include <mkl_cblas.h>
+    #include <mkl_lapack.h>
+#endif
+
 #ifdef SINGLE_PRECISION
 typedef Eigen::SparseMatrix<float, Eigen::ColMajor, long long> eigenSparseMat;
 #else
diff --git a/main/mkl.cpp b/main/mkl.cpp
index dbf043574e55dd7f422662c56fc74e8c842e0aac..c126fc71851b439dbe98320ea038034d109125bd 100644
--- a/main/mkl.cpp
+++ b/main/mkl.cpp
@@ -362,7 +362,11 @@ bool gcta::comput_inverse_logdet_LDLT_mkl(eigenMatrix &Vi, double &logdet)
     // MKL's Cholesky decomposition
     int info = 0, int_n = (int) n;
     char uplo = 'L';
-    dpotrf(&uplo, &int_n, Vi_mkl, &int_n, &info);
+    #if defined(__aarch64__)
+        dpotrf_(&uplo, &int_n, Vi_mkl, &int_n, &info);
+    #else
+        dpotrf(&uplo, &int_n, Vi_mkl, &int_n, &info);
+    #endif
     //LOGGER << "Finished decompose" << endl;
     //spotrf( &uplo, &n, Vi_mkl, &n, &info );
     if (info < 0){
@@ -379,7 +383,11 @@ bool gcta::comput_inverse_logdet_LDLT_mkl(eigenMatrix &Vi, double &logdet)
 
         //LOGGER << "start inverse" << endl;
         // Calcualte V inverse
-        dpotri(&uplo, &int_n, Vi_mkl, &int_n, &info);
+        #if defined(__aarch64__)
+            dpotri_(&uplo, &int_n, Vi_mkl, &int_n, &info);
+        #else
+            dpotri(&uplo, &int_n, Vi_mkl, &int_n, &info);
+        #endif
         //LOGGER << "Inverse finished" << endl;
         //spotri( &uplo, &n, Vi_mkl, &n, &info );
         if (info < 0){
@@ -420,7 +428,11 @@ bool gcta::comput_inverse_logdet_LU_mkl(eigenMatrix &Vi, double &logdet)
     int LWORK = N*N;
     double *WORK = new double[n * n];
     int INFO;
-    dgetrf(&N, &N, Vi_mkl, &N, IPIV, &INFO);
+    #if defined(__aarch64__)
+        dgetrf_(&N, &N, Vi_mkl, &N, IPIV, &INFO);
+    #else
+        dgetrf(&N, &N, Vi_mkl, &N, IPIV, &INFO);
+    #endif 
     if (INFO < 0) LOGGER.e(0, "LU decomposition failed. Invalid values found in the matrix.\n");
     else if (INFO > 0) {
         delete[] Vi_mkl;
@@ -435,7 +447,11 @@ bool gcta::comput_inverse_logdet_LU_mkl(eigenMatrix &Vi, double &logdet)
         }
 
         // Calcualte V inverse
-        dgetri(&N, Vi_mkl, &N, IPIV, WORK, &LWORK, &INFO);
+        #if defined(__aarch64__)
+            dgetri_(&N, Vi_mkl, &N, IPIV, WORK, &LWORK, &INFO);
+        #else
+            dgetri(&N, Vi_mkl, &N, IPIV, WORK, &LWORK, &INFO);
+        #endif
         if (INFO < 0){
             LOGGER.e(0, "invalid values found in the varaince-covaraince (V) matrix.\n");
         }else if (INFO > 0){
@@ -474,7 +490,11 @@ bool gcta::comput_inverse_logdet_LU_mkl_array(int n, float *Vi, double &logdet)
     int LWORK = N*N;
     double *WORK = new double[n * n];
     int INFO;
-    dgetrf(&N, &N, Vi_mkl, &N, IPIV, &INFO);
+    #if defined(__aarch64__)
+        dgetrf_(&N, &N, Vi_mkl, &N, IPIV, &INFO);
+    #else
+        dgetrf(&N, &N, Vi_mkl, &N, IPIV, &INFO);
+    #endif
     if (INFO < 0) LOGGER.e(0, "LU decomposition failed. Invalid values found in the matrix.\n");
     else if (INFO > 0) {
         // free memory
@@ -491,7 +511,11 @@ bool gcta::comput_inverse_logdet_LU_mkl_array(int n, float *Vi, double &logdet)
         }
 
         // Calcualte V inverse
-        dgetri(&N, Vi_mkl, &N, IPIV, WORK, &LWORK, &INFO);
+        #if defined(__aarch64__)
+            dgetri_(&N, Vi_mkl, &N, IPIV, WORK, &LWORK, &INFO);
+        #else
+            dgetri(&N, Vi_mkl, &N, IPIV, WORK, &LWORK, &INFO);
+        #endif
         if (INFO < 0) LOGGER.e(0, "invalid values found in the varaince-covaraince (V) matrix.\n");
         else if (INFO > 0) {
             // free memory
diff --git a/src/GRM.cpp b/src/GRM.cpp
index 15cfedc03e99a7f8e82476db8f6497d1fc225d5d..02c3866f3c2dfd870d290b779b7afef84e7c11aa 100644
--- a/src/GRM.cpp
+++ b/src/GRM.cpp
@@ -33,9 +33,34 @@
 #include <boost/algorithm/string.hpp>
 #include <boost/algorithm/string/join.hpp>
 #include <sstream>
-#include <mkl.h>
 #include <csignal>
 
+#if defined( __i386__ ) || defined(i386) || defined(_M_IX86)
+    /*
+     * __i386__ is defined by gcc and Intel compiler on Linux,
+     * _M_IX86 by VS compiler,
+     * i386 by Sun compilers on opensolaris at least
+     */
+    #define CPU_X86
+#elif defined(__x86_64__) || defined(__amd64__) || defined(__x86_64) || defined(_M_AMD64)
+    /*
+     * both __x86_64__ and __amd64__ are defined by gcc
+     * __x86_64 defined by sun compiler on opensolaris at least
+     * _M_AMD64 defined by MS compiler
+     */
+    #define CPU_AMD64
+#elif defined(__arm__) || defined(__aarch64__)
+    #define CPU_ARM
+#else
+    #error Unknown CPU
+#endif
+
+#if defined(CPU_ARM)
+    #include <f77blas.h>
+#else
+    #include <mkl.h>
+#endif
+
 using std::to_string;
 
 map<string, string> GRM::options;
@@ -919,14 +944,14 @@ void flip64(uint64_t a[64]) {
 //#pragma message("multiple target of N thread")
 //__attribute__((target_clones("popcnt","default")))
 //#endif
-#ifdef __linux__
+#if defined(__linux__) && !defined(CPU_ARM)
 __attribute__((target("default")))
 #endif
 uint32_t popcounts(uint64_t dw){
     return popcount(dw);
 }
 
-#ifdef __linux__
+#if defined(__linux__) && !defined(CPU_ARM)
 __attribute__((target("popcnt")))
 uint32_t popcounts(uint64_t dw){
     return popcount(dw);
@@ -977,12 +1002,24 @@ void GRM::calculate_GRM_blas(uintptr_t *buf, const vector<uint32_t> &markerIndex
     static const char uplo='L';
    // A * At 
     if(part_keep_indices.first == 0){
-        dsyrk(&uplo, &notrans, &n, &curNumValidMarkers, &alpha, stdGeno, &n_sample, &beta, grm, &m);
+        #if defined(CPU_ARM)
+            dsyrk_(&uplo, &notrans, &n, &curNumValidMarkers, &alpha, stdGeno, &n_sample, &beta, grm, &m);
+        #else
+            dsyrk(&uplo, &notrans, &n, &curNumValidMarkers, &alpha, stdGeno, &n_sample, &beta, grm, &m);
+        #endif
     }else{
-        //dgemm(&notrans, &trans, &m, &n, &num_marker, &alpha, stdGeno + part_keep_indices.first, &n_sample, stdGeno, &n_sample, &beta, grm, &m);
-        dgemm(&notrans, &trans, &m, &s_n, &curNumValidMarkers, &alpha, stdGeno + part_keep_indices.first, &n_sample, stdGeno, &n_sample, &beta, grm, &m);
+        #if defined(CPU_ARM)
+            dgemm_(&notrans, &trans, &m, &s_n, &curNumValidMarkers, &alpha, stdGeno + part_keep_indices.first, &n_sample, stdGeno, &n_sample, &beta, grm, &m);
+        #else
+            //dgemm(&notrans, &trans, &m, &n, &num_marker, &alpha, stdGeno + part_keep_indices.first, &n_sample, stdGeno, &n_sample, &beta, grm, &m);
+            dgemm(&notrans, &trans, &m, &s_n, &curNumValidMarkers, &alpha, stdGeno + part_keep_indices.first, &n_sample, stdGeno, &n_sample, &beta, grm, &m);
+        #endif
         double * grm_start = grm + ((uint64_t)s_n) * m;
-        dsyrk(&uplo, &notrans, &m, &curNumValidMarkers, &alpha, stdGeno + part_keep_indices.first, &n_sample, &beta, grm_start, &m); 
+        #if defined(CPU_ARM)
+            dsyrk_(&uplo, &notrans, &m, &curNumValidMarkers, &alpha, stdGeno + part_keep_indices.first, &n_sample, &beta, grm_start, &m); 
+        #else
+            dsyrk(&uplo, &notrans, &m, &curNumValidMarkers, &alpha, stdGeno + part_keep_indices.first, &n_sample, &beta, grm_start, &m); 
+        #endif
     }
 
     //memset(this->cmask_buf, 0, num_byte_cmask);
diff --git a/src/LD.cpp b/src/LD.cpp
index d34fa24c6547708fcfe677139bfc3fde3fef68d9..319b07d9a8182e87b0dd7f04684e34712b5eb10b 100644
--- a/src/LD.cpp
+++ b/src/LD.cpp
@@ -7,10 +7,15 @@
 #include <string>
 #include <functional>
 #include <omp.h>
-#include <mkl.h>
 #include <cstdio>
 #include <algorithm>
 
+#if defined(__aarch64__)
+    #include <f77blas.h>
+#else
+    #include <mkl.h>
+#endif
+
 map<string, string> LD::options;
 map<string, int> LD::options_i;
 vector<string> LD::processFunctions;
@@ -62,7 +67,11 @@ void LD::calcLD(){
     double alpha = 1.0 / (nr - 1);
     double *ptr1 = geno_buffer[cacl_index_buffer].get();
     double *res1 = new double[nc1 * nc1];
-    dsyrk(uplo, trans, &nc1, &nr, &alpha, ptr1, &nr, &zero, res1, &nc1);
+    #if defined(__aarch64__)
+        dsyrk_(uplo, trans, &nc1, &nr, &alpha, ptr1, &nr, &zero, res1, &nc1);
+    #else
+        dsyrk(uplo, trans, &nc1, &nr, &alpha, ptr1, &nr, &zero, res1, &nc1);
+    #endif
 
     double *res2 = nullptr;
     // is previous buffer active?
@@ -72,7 +81,11 @@ void LD::calcLD(){
         nc2 = cur_buffer_offset[!cacl_index_buffer] / nr;
         double *ptr2 = geno_buffer[!cacl_index_buffer].get();
         res2 = new double[nc2 * nc1];
-        dgemm(trans, notrans, &nc2, &nc1, &nr, &alpha, ptr2, &nr, ptr1, &nr, &zero, res2, &nc2);
+        #if defined(__aarch64__)
+            dgemm_(trans, notrans, &nc2, &nc1, &nr, &alpha, ptr2, &nr, ptr1, &nr, &zero, res2, &nc2);
+        #else
+            dgemm(trans, notrans, &nc2, &nc1, &nr, &alpha, ptr2, &nr, ptr1, &nr, &zero, res2, &nc2);
+        #endif
     }
     
     for(int i = 0; i < nc1; i++){
diff --git a/src/StatLib.cpp b/src/StatLib.cpp
index 4656ed2b02760a1815a8eb127bb5f0172f1c383c..d1116360a5b9df55106a3394ef38048a8e03f8ed 100644
--- a/src/StatLib.cpp
+++ b/src/StatLib.cpp
@@ -18,7 +18,12 @@
 #include <cstdio>
 #include <limits>
 #include <boost/math/distributions/chi_squared.hpp>
-#include <mkl.h>
+
+#if defined(__aarch64__)
+    #include <lapack.h>
+#else
+    #include <mkl.h>
+#endif
 
 using namespace boost::math;
 
@@ -73,8 +78,13 @@ namespace StatLib{
         char side = 'L';
         char t = 'N';
 
-        dormqr(&side, &t, &n, &n, &n, X, &lda, tau, c, 
-                &lda, work, &lwork, &info);
+        #if defined(__aarch64__)
+            dormqr_(&side, &t, &n, &n, &n, X, &lda, tau, c, 
+                    &lda, work, &lwork, &info);
+        #else
+            dormqr(&side, &t, &n, &n, &n, X, &lda, tau, c, 
+                    &lda, work, &lwork, &info);
+        #endif
 
         if(info != 0){
             return false;