diff --git a/CMakeLists.txt b/CMakeLists.txt
index c3eed33..d8d9e5a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -45,7 +45,6 @@ include(cmake/vars.cmake)
 
 # Add subdirectories
 add_subdirectory(ext)
-add_subdirectory(include)
 add_subdirectory(src)
 add_subdirectory(demo)
 add_subdirectory(share)
diff --git a/README.md b/README.md
index caa609c..8e1f0e7 100644
--- a/README.md
+++ b/README.md
@@ -7,7 +7,7 @@ Integrated Singular Value Decomposition (iSVD)
 * https://github.com/emfomy/isvd
 
 ### Documentation
-* http://emfomy.github.io/isvd
+* Please download from https://github.com/emfomy/isvd/releases or build using **make doc** (see below) on your own computer.
 
 ### Author
 * Mu Yang <<emfomy@gmail.com>>
@@ -33,9 +33,10 @@ Integrated Singular Value Decomposition (iSVD)
 Please use the following commands to create Makefiles
 
 ```
+cd <isvd-source-folder>
 mkdir build
 cd build
-cmake <path-to-source>
+cmake ..
 ```
 
 ### Options
@@ -43,7 +44,7 @@ cmake <path-to-source>
 Use the following command to set options
 
 ```
-ccmake <path-to-source>
+ccmake ..
 ```
 
 The following table are the main options
@@ -67,20 +68,28 @@ The following table are the main options
 
 ### Makefile
 
-The following table are the main make rules
+The following table are the main Makefile rules
 
-| Command        | Detail                         |
-|----------------|--------------------------------|
-| `make all`     | build all libraries            |
-| `make install` | install package                |
-| `make check`   | build and run unit tests       |
-| `make doc`     | build documentation            |
-| `make help`    | display make-rules             |
+| Command        | Detail                         | Options                        |
+|----------------|--------------------------------|--------------------------------|
+| `make all`     | build all libraries            |                                |
+| `make install` | install package                |                                |
+| `make check`   | build and run unit tests       | Require `ISVD_BUILD_TEST`      |
+| `make doc`     | build documentation            | Require `ISVD_BUILD_DOC`       |
+| `make help`    | display make-rules             |                                |
+
+### Test installation
+
+* Set `ISVD_BUILD_TEST` using **ccmake**. (Also recommended to unset `ISVD_VERBOSE_TEST` if GPU is enabled).
+* Run **make check**
+  * Known issue: RealSingle_WenYinIntegration.Test.#/s_integrate_wen_yin_# always fail the test.
 
 ## Usage
 
 * Define `ISVD_USE_ILP64` before include `isvd.h` to use 64-bit integer.
 * All 64bit libraries and executables are named with suffix "`_64`".
+* The header files are located in `build/include`
+* The libraries are located in `build/lib`
 
 ### Libraries
 
diff --git a/check/CMakeLists.txt b/check/CMakeLists.txt
index e22c4d4..a47dafb 100644
--- a/check/CMakeLists.txt
+++ b/check/CMakeLists.txt
@@ -35,7 +35,7 @@ function(ISVD_SET_TARGET_CHECK_CPU target)
   isvd_set_target_mpi(${target} CXX)
   isvd_set_target_blas(${target})
   isvd_set_target_gtest(${target})
-  target_link_libraries(${target} checkisvd extmmio)
+  target_link_libraries(${target} checkisvd_gpu_none checkisvd_core_la extmmio)
   target_compile_definitions(${target} PUBLIC "-DISVD_DATA_PATH=\"${PROJECT_SOURCE_DIR}/data\"")
 endfunction()
 
@@ -45,18 +45,18 @@ function(ISVD_SET_TARGET_CHECK_GPU target)
   isvd_set_target_blas(${target})
   isvd_set_target_gtest(${target})
   isvd_set_target_gpu(${target})
-  target_link_libraries(${target} checkisvd_gpu extmmio)
+  target_link_libraries(${target} checkisvd_gpu_magma checkisvd_core_la extmmio)
   target_compile_definitions(${target} PUBLIC "-DISVD_DATA_PATH=\"${PROJECT_SOURCE_DIR}/data\"")
 endfunction()
 
 # Set include paths
-include_directories("${PROJECT_CONFIG_DIR}/include/c"
-                    "${CMAKE_CURRENT_CONFIG_DIR}/src"
-                    "${PROJECT_CONFIG_DIR}/src"
-             SYSTEM "${PROJECT_SOURCE_DIR}/ext")
+include_directories("${CMAKE_CURRENT_CONFIG_DIR}/lib"
+                    "${PROJECT_BINARY_DIR}/include/c"
+                    "${PROJECT_BINARY_DIR}/lib"
+             SYSTEM "${PROJECT_SOURCE_DIR}/ext/mmio")
 
 # Add subdirectories
-add_subdirectory(src)
+add_subdirectory(lib)
 
 # Configure files
 isvd_configure_x_fn("${CMAKE_CURRENT_SOURCE_DIR}/check" "${CMAKE_CURRENT_CONFIG_DIR}/check" "${ISVD_S_TYPES}")
diff --git a/check/check/libisvd/core/stage/@x@_integrate_hierarchical_reduction.cxx b/check/check/libisvd/core/stage/@x@_integrate_hierarchical_reduction.cxx
index d592a5a..33e9f1b 100644
--- a/check/check/libisvd/core/stage/@x@_integrate_hierarchical_reduction.cxx
+++ b/check/check/libisvd/core/stage/@x@_integrate_hierarchical_reduction.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -10,7 +10,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 TEST(@XStr@_HierarchicalReductionIntegration, Test) {
 
@@ -26,7 +26,7 @@ TEST(@XStr@_HierarchicalReductionIntegration, Test) {
 
   // Read Qs
   file = fopen(QS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -47,7 +47,7 @@ TEST(@XStr@_HierarchicalReductionIntegration, Test) {
 
   // Read Qbar
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -92,12 +92,12 @@ TEST(@XStr@_HierarchicalReductionIntegration, Test) {
   isvd_int_t ldqt = l;
 
   // Run stage
-  isvd_@x@IntegrateHierarchicalReduction(param, nullptr, 0, nullptr, 0, qst, ldqst, qt, ldqt);
+  isvd_@x@IntegrateHierarchicalReduction(param, NULL, 0, NULL, 0, qst, ldqst, qt, ldqt);
 
   // Gather results
   isvd_val_t *qt_ = isvd_@x@malloc(l * Pmb);
   isvd_int_t ldqt_ = l;
-  MPI_Gather(qt, mb*ldqt, MPI_@X_TYPE@, qt_, mb*ldqt, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+  MPI_Gather(qt, mb*ldqt, MPI_@XTYPE@, qt_, mb*ldqt, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
   if ( mpi_rank == mpi_root ) {
     // Compute space
diff --git a/check/check/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.cxx b/check/check/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.cxx
index a8f1cfd..e120dd7 100644
--- a/check/check/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.cxx
+++ b/check/check/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -15,7 +15,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 TEST(@XStr@_KolmogorovNagumoIntegration, Test) {
 
@@ -31,7 +31,7 @@ TEST(@XStr@_KolmogorovNagumoIntegration, Test) {
 
   // Read Qs
   file = fopen(QS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -52,7 +52,7 @@ TEST(@XStr@_KolmogorovNagumoIntegration, Test) {
 
   // Read Qbar
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -107,7 +107,7 @@ TEST(@XStr@_KolmogorovNagumoIntegration, Test) {
   // Gather results
   isvd_val_t *qt_ = isvd_@x@malloc(l * Pmb);
   isvd_int_t ldqt_ = l;
-  MPI_Gather(qt, mb*ldqt, MPI_@X_TYPE@, qt_, mb*ldqt, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+  MPI_Gather(qt, mb*ldqt, MPI_@XTYPE@, qt_, mb*ldqt, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
   // Check results
   if ( mpi_rank == mpi_root ) {
diff --git a/check/check/libisvd/core/stage/@x@_integrate_wen_yin.cxx b/check/check/libisvd/core/stage/@x@_integrate_wen_yin.cxx
index 44f0729..a0a753b 100644
--- a/check/check/libisvd/core/stage/@x@_integrate_wen_yin.cxx
+++ b/check/check/libisvd/core/stage/@x@_integrate_wen_yin.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -15,7 +15,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 TEST(@XStr@_WenYinIntegration, Test) {
 
@@ -31,7 +31,7 @@ TEST(@XStr@_WenYinIntegration, Test) {
 
   // Read Qs
   file = fopen(QS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -52,7 +52,7 @@ TEST(@XStr@_WenYinIntegration, Test) {
 
   // Read Qbar
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -107,7 +107,7 @@ TEST(@XStr@_WenYinIntegration, Test) {
   // Gather results
   isvd_val_t *qt_ = isvd_@x@malloc(l * Pmb);
   isvd_int_t ldqt_ = l;
-  MPI_Gather(qt, mb*ldqt, MPI_@X_TYPE@, qt_, mb*ldqt, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+  MPI_Gather(qt, mb*ldqt, MPI_@XTYPE@, qt_, mb*ldqt, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
   // Check results
   if ( mpi_rank == mpi_root ) {
diff --git a/check/check/libisvd/core/stage/@x@_orthogonalize_gramian.cxx b/check/check/libisvd/core/stage/@x@_orthogonalize_gramian.cxx
index 26c7ef0..d72eaca 100644
--- a/check/check/libisvd/core/stage/@x@_orthogonalize_gramian.cxx
+++ b/check/check/libisvd/core/stage/@x@_orthogonalize_gramian.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -10,7 +10,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 TEST(@XStr@_GramianOrthogonalization, Test) {
 
@@ -25,7 +25,7 @@ TEST(@XStr@_GramianOrthogonalization, Test) {
 
   // Read Ys
   file = fopen(YS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -46,7 +46,7 @@ TEST(@XStr@_GramianOrthogonalization, Test) {
 
   // Read Qs
   file = fopen(QS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -90,12 +90,12 @@ TEST(@XStr@_GramianOrthogonalization, Test) {
   isvd_@x@Omatcopy('N', Nl, mj, 1.0, yst0 + param.rowidxbegin * ldyst0, ldyst0, yst, ldyst);
 
   // Run
-  isvd_@x@OrthogonalizeGramian(param, nullptr, 0, nullptr, 0, yst, ldyst);
+  isvd_@x@OrthogonalizeGramian(param, NULL, 0, NULL, 0, yst, ldyst);
 
   // Gather results
   isvd_val_t *qst_ = isvd_@x@malloc(Pmb * ldyst);
   isvd_int_t ldqst_ = ldyst;
-  MPI_Gather(yst, mb*ldyst, MPI_@X_TYPE@, qst_, mb*ldyst, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+  MPI_Gather(yst, mb*ldyst, MPI_@XTYPE@, qst_, mb*ldyst, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
   if ( mpi_rank == mpi_root ) {
     // Compute space
diff --git a/check/check/libisvd/core/stage/@x@_postprocess_gramian.cxx b/check/check/libisvd/core/stage/@x@_postprocess_gramian.cxx
index ce9152a..3262efb 100644
--- a/check/check/libisvd/core/stage/@x@_postprocess_gramian.cxx
+++ b/check/check/libisvd/core/stage/@x@_postprocess_gramian.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -13,7 +13,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 typedef enum {
   GatherUV,
@@ -34,14 +34,14 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
   MM_typecode matcode;
 
   // Check arguments
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   ASSERT_NE(dista_,  '\0');
   ASSERT_NE(ordera_, '\0');
 
   // Read A
   file = fopen(A_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -72,7 +72,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read Q
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -97,7 +97,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read S
   file = fopen(S_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -120,7 +120,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read U
   file = fopen(U_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -146,7 +146,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read V
   file = fopen(V_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -216,7 +216,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case GatherUV: {
 
       // Run stage
-      isvd_@x@PostprocessGramian(param, nullptr, 0, nullptr, 0, dista_, ordera_,
+      isvd_@x@PostprocessGramian(param, NULL, 0, NULL, 0, dista_, ordera_,
                                  a, lda, qt, ldqt, s, ut_, ldut_, vt_, ldvt_, mpi_root, mpi_root);
 
       break;
@@ -232,12 +232,12 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
       isvd_int_t ldvt = l;
 
       // Run stage
-      isvd_@x@PostprocessGramian(param, nullptr, 0, nullptr, 0, dista_, ordera_,
+      isvd_@x@PostprocessGramian(param, NULL, 0, NULL, 0, dista_, ordera_,
                                  a, lda, qt, ldqt, s, ut, ldut, vt, ldvt, -1, -1);
 
       // Gather results
-      MPI_Gather(ut, mb*ldut, MPI_@X_TYPE@, ut_, mb*ldut, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
-      MPI_Gather(vt, nb*ldvt, MPI_@X_TYPE@, vt_, nb*ldvt, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+      MPI_Gather(ut, mb*ldut, MPI_@XTYPE@, ut_, mb*ldut, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
+      MPI_Gather(vt, nb*ldvt, MPI_@XTYPE@, vt_, nb*ldvt, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
       // Deallocate memory
       isvd_free(ut);
@@ -249,8 +249,8 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case NoUV: {
 
       // Run stage
-      isvd_@x@PostprocessGramian(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                 a, lda, qt, ldqt, s, nullptr, 0, nullptr, 0, -2, -2);
+      isvd_@x@PostprocessGramian(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                 a, lda, qt, ldqt, s, NULL, 0, NULL, 0, -2, -2);
 
       break;
     }
diff --git a/check/check/libisvd/core/stage/@x@_postprocess_symmetric.cxx b/check/check/libisvd/core/stage/@x@_postprocess_symmetric.cxx
index acfe9e0..e5ac46d 100644
--- a/check/check/libisvd/core/stage/@x@_postprocess_symmetric.cxx
+++ b/check/check/libisvd/core/stage/@x@_postprocess_symmetric.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -12,7 +12,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 typedef enum {
   GatherUV,
@@ -33,14 +33,14 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
   MM_typecode matcode;
 
   // Check arguments
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   ASSERT_NE(dista_,  '\0');
   ASSERT_NE(ordera_, '\0');
 
   // Read A
   file = fopen(A_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -71,7 +71,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read Q
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -96,7 +96,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read S
   file = fopen(S_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -119,7 +119,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read U
   file = fopen(U_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -186,8 +186,8 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case GatherUV: {
 
       // Run stage
-      isvd_@x@PostprocessSymmetric(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                   a, lda, qt, ldqt, s, ut_, ldut_, nullptr, 0, mpi_root, -2);
+      isvd_@x@PostprocessSymmetric(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                   a, lda, qt, ldqt, s, ut_, ldut_, NULL, 0, mpi_root, -2);
 
       break;
     }
@@ -199,11 +199,11 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
       isvd_int_t ldut = l;
 
       // Run stage
-      isvd_@x@PostprocessSymmetric(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                   a, lda, qt, ldqt, s, ut, ldut, nullptr, 0, -1, -2);
+      isvd_@x@PostprocessSymmetric(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                   a, lda, qt, ldqt, s, ut, ldut, NULL, 0, -1, -2);
 
       // Gather results
-      MPI_Gather(ut, mb*ldut, MPI_@X_TYPE@, ut_, mb*ldut, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+      MPI_Gather(ut, mb*ldut, MPI_@XTYPE@, ut_, mb*ldut, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
       // Deallocate memory
       isvd_free(ut);
@@ -214,8 +214,8 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case NoUV: {
 
       // Run stage
-      isvd_@x@PostprocessSymmetric(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                   a, lda, qt, ldqt, s, nullptr, 0, nullptr, 0, -2, -2);
+      isvd_@x@PostprocessSymmetric(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                   a, lda, qt, ldqt, s, NULL, 0, NULL, 0, -2, -2);
 
       break;
     }
diff --git a/check/check/libisvd/core/stage/@x@_sketch_gaussian_projection.cxx b/check/check/libisvd/core/stage/@x@_sketch_gaussian_projection.cxx
index 4b7941e..86c2dea 100644
--- a/check/check/libisvd/core/stage/@x@_sketch_gaussian_projection.cxx
+++ b/check/check/libisvd/core/stage/@x@_sketch_gaussian_projection.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -10,7 +10,7 @@
 #define serr 1e-1
 #define derr 1e-6
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 static void test( char dista, char ordera ) {
 
@@ -25,14 +25,14 @@ static void test( char dista, char ordera ) {
   MM_typecode matcode;
 
   // Check arguments
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   ASSERT_NE(dista_,  '\0');
   ASSERT_NE(ordera_, '\0');
 
   // Read A
   file = fopen(A_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -63,7 +63,7 @@ static void test( char dista, char ordera ) {
 
   // Read Ys
   file = fopen(YS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -119,13 +119,13 @@ static void test( char dista, char ordera ) {
   isvd_int_t ldyst = Nl;
 
   // Run stage
-  isvd_@x@SketchGaussianProjection(param, nullptr, 0, nullptr, 0, dista_, ordera_, a, lda, yst, ldyst, seed, mpi_root);
+  isvd_@x@SketchGaussianProjection(param, NULL, 0, NULL, 0, dista_, ordera_, a, lda, yst, ldyst, seed, mpi_root);
 
 #if defined(ISVD_USE_MKL)
   // Gather results
   isvd_val_t *yst_ = isvd_@x@malloc(Nl * Pmb);
   isvd_int_t ldyst_ = Nl;
-  MPI_Gather(yst, mb*ldyst, MPI_@X_TYPE@, yst_, mb*ldyst, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+  MPI_Gather(yst, mb*ldyst, MPI_@XTYPE@, yst_, mb*ldyst, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
   // Check results
   if ( mpi_rank == mpi_root ) {
diff --git a/check/check/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.cxx b/check/check/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.cxx
index a208e06..2f6cfb1 100644
--- a/check/check/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.cxx
+++ b/check/check/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -13,7 +13,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 typedef enum {
   GatherUV,
@@ -34,14 +34,14 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
   MM_typecode matcode;
 
   // Check arguments
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   ASSERT_NE(dista_,  '\0');
   ASSERT_NE(ordera_, '\0');
 
   // Read A
   file = fopen(A_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -72,7 +72,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read Q
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -97,7 +97,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read S
   file = fopen(S_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -120,7 +120,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read U
   file = fopen(U_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -146,7 +146,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read V
   file = fopen(V_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -216,7 +216,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case GatherUV: {
 
       // Run stage
-      isvd_@x@PostprocessGramian_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_,
+      isvd_@x@PostprocessGramian_gpu(param, NULL, 0, NULL, 0, dista_, ordera_,
                                      a, lda, qt, ldqt, s, ut_, ldut_, vt_, ldvt_, mpi_root, mpi_root);
 
       break;
@@ -232,12 +232,12 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
       isvd_int_t ldvt = l;
 
       // Run stage
-      isvd_@x@PostprocessGramian_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_,
+      isvd_@x@PostprocessGramian_gpu(param, NULL, 0, NULL, 0, dista_, ordera_,
                                      a, lda, qt, ldqt, s, ut, ldut, vt, ldvt, -1, -1);
 
       // Gather results
-      MPI_Gather(ut, mb*ldut, MPI_@X_TYPE@, ut_, mb*ldut, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
-      MPI_Gather(vt, nb*ldvt, MPI_@X_TYPE@, vt_, nb*ldvt, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+      MPI_Gather(ut, mb*ldut, MPI_@XTYPE@, ut_, mb*ldut, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
+      MPI_Gather(vt, nb*ldvt, MPI_@XTYPE@, vt_, nb*ldvt, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
       // Deallocate memory
       isvd_free(ut);
@@ -249,8 +249,8 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case NoUV: {
 
       // Run stage
-      isvd_@x@PostprocessGramian_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                     a, lda, qt, ldqt, s, nullptr, 0, nullptr, 0, -2, -2);
+      isvd_@x@PostprocessGramian_gpu(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                     a, lda, qt, ldqt, s, NULL, 0, NULL, 0, -2, -2);
 
       break;
     }
diff --git a/check/check/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.cxx b/check/check/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.cxx
index 5737672..d041385 100644
--- a/check/check/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.cxx
+++ b/check/check/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -12,7 +12,7 @@
 #define serr 1e-3
 #define derr 1e-8
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 typedef enum {
   GatherUV,
@@ -33,14 +33,14 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
   MM_typecode matcode;
 
   // Check arguments
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   ASSERT_NE(dista_,  '\0');
   ASSERT_NE(ordera_, '\0');
 
   // Read A
   file = fopen(A_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -71,7 +71,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read Q
   file = fopen(Q_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -96,7 +96,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read S
   file = fopen(S_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -119,7 +119,7 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
 
   // Read U
   file = fopen(U_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -186,8 +186,8 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case GatherUV: {
 
       // Run stage
-      isvd_@x@PostprocessSymmetric_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                       a, lda, qt, ldqt, s, ut_, ldut_, nullptr, 0, mpi_root, -2);
+      isvd_@x@PostprocessSymmetric_gpu(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                       a, lda, qt, ldqt, s, ut_, ldut_, NULL, 0, mpi_root, -2);
 
       break;
     }
@@ -199,11 +199,11 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
       isvd_int_t ldut = l;
 
       // Run stage
-      isvd_@x@PostprocessSymmetric_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                       a, lda, qt, ldqt, s, ut, ldut, nullptr, 0, -1, -2);
+      isvd_@x@PostprocessSymmetric_gpu(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                       a, lda, qt, ldqt, s, ut, ldut, NULL, 0, -1, -2);
 
       // Gather results
-      MPI_Gather(ut, mb*ldut, MPI_@X_TYPE@, ut_, mb*ldut, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+      MPI_Gather(ut, mb*ldut, MPI_@XTYPE@, ut_, mb*ldut, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
       // Deallocate memory
       isvd_free(ut);
@@ -214,8 +214,8 @@ static void test( char dista, char ordera, const JobUV jobuv ) {
     case NoUV: {
 
       // Run stage
-      isvd_@x@PostprocessSymmetric_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_,
-                                       a, lda, qt, ldqt, s, nullptr, 0, nullptr, 0, -2, -2);
+      isvd_@x@PostprocessSymmetric_gpu(param, NULL, 0, NULL, 0, dista_, ordera_,
+                                       a, lda, qt, ldqt, s, NULL, 0, NULL, 0, -2, -2);
 
       break;
     }
diff --git a/check/check/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.cxx b/check/check/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.cxx
index 31edc73..609de5a 100644
--- a/check/check/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.cxx
+++ b/check/check/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.cxx
@@ -1,6 +1,6 @@
 #include <gtest/gtest.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 #include <isvd.h>
 #include <libisvd.h>
 
@@ -10,7 +10,7 @@
 #define serr 1e-1
 #define derr 1e-6
 
-typedef @xtype@ isvd_val_t;
+typedef @xtype_____@ isvd_val_t;
 
 static void test( char dista, char ordera ) {
 
@@ -25,14 +25,14 @@ static void test( char dista, char ordera ) {
   MM_typecode matcode;
 
   // Check arguments
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   ASSERT_NE(dista_,  '\0');
   ASSERT_NE(ordera_, '\0');
 
   // Read A
   file = fopen(A_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -63,7 +63,7 @@ static void test( char dista, char ordera ) {
 
   // Read Ys
   file = fopen(YS_PATH, "r");
-  ASSERT_NE(file, (void*)(nullptr));
+  ASSERT_NE(file, nullptr);
   ASSERT_EQ(mm_read_banner(file, &matcode), 0);
   EXPECT_TRUE(mm_is_array(matcode))   << mm_typecode_to_str(matcode);
   EXPECT_TRUE(mm_is_real(matcode))    << mm_typecode_to_str(matcode);
@@ -119,13 +119,13 @@ static void test( char dista, char ordera ) {
   isvd_int_t ldyst = Nl;
 
   // Run stage
-  isvd_@x@SketchGaussianProjection_gpu(param, nullptr, 0, nullptr, 0, dista_, ordera_, a, lda, yst, ldyst, seed, mpi_root);
+  isvd_@x@SketchGaussianProjection_gpu(param, NULL, 0, NULL, 0, dista_, ordera_, a, lda, yst, ldyst, seed, mpi_root);
 
 #if defined(ISVD_USE_MKL)
   // Gather results
   isvd_val_t *yst_ = isvd_@x@malloc(Nl * Pmb);
   isvd_int_t ldyst_ = Nl;
-  MPI_Gather(yst, mb*ldyst, MPI_@X_TYPE@, yst_, mb*ldyst, MPI_@X_TYPE@, mpi_root, MPI_COMM_WORLD);
+  MPI_Gather(yst, mb*ldyst, MPI_@XTYPE@, yst_, mb*ldyst, MPI_@XTYPE@, mpi_root, MPI_COMM_WORLD);
 
   // Check results
   if ( mpi_rank == mpi_root ) {
diff --git a/check/lib/CMakeLists.txt b/check/lib/CMakeLists.txt
new file mode 100644
index 0000000..59e4584
--- /dev/null
+++ b/check/lib/CMakeLists.txt
@@ -0,0 +1,42 @@
+# The CMake setting of 'check/lib'
+
+isvd_set_config_var()
+
+# Configure files
+file(COPY "${PROJECT_CONFIG_DIR}/src/lib/" DESTINATION "${CMAKE_CURRENT_CONFIG_DIR}")
+file(GLOB_RECURSE cfiles "${CMAKE_CURRENT_CONFIG_DIR}/*.c")
+foreach(cfile ${cfiles})
+  string(REGEX REPLACE "\\.c$" ".cxx" cxxfile ${cfile})
+  file(RENAME ${cfile} ${cxxfile})
+endforeach()
+
+file(RENAME "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/param.cxx"
+            "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/param.c")
+
+# checkisvd_core_la
+file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/*"
+                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/la/*"
+                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/util/*")
+add_library(checkisvd_core_la EXCLUDE_FROM_ALL ${files})
+isvd_set_target(checkisvd_core_la)
+isvd_set_target_omp(checkisvd_core_la CXX)
+isvd_set_target_mpi(checkisvd_core_la CXX)
+isvd_set_target_blas(checkisvd_core_la)
+isvd_set_target_gtest(checkisvd_core_la)
+
+# checkisvd_gpu_none
+file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/nogpu/*")
+add_library(checkisvd_gpu_none EXCLUDE_FROM_ALL ${files})
+isvd_set_target(checkisvd_gpu_none)
+isvd_set_target_mpi(checkisvd_gpu_none CXX)
+isvd_set_target_gtest(checkisvd_gpu_none)
+
+# checkisvd_gpu_magma
+if(ISVD_USE_GPU)
+  file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/gpu/*")
+  add_library(checkisvd_gpu_magma EXCLUDE_FROM_ALL ${files})
+  isvd_set_target(checkisvd_gpu_magma)
+  isvd_set_target_mpi(checkisvd_gpu_magma CXX)
+  isvd_set_target_gpu(checkisvd_gpu_magma)
+  isvd_set_target_gtest(checkisvd_gpu_magma)
+endif()
diff --git a/check/src/CMakeLists.txt b/check/src/CMakeLists.txt
deleted file mode 100644
index bebedf5..0000000
--- a/check/src/CMakeLists.txt
+++ /dev/null
@@ -1,41 +0,0 @@
-# The CMake setting of 'check/src'
-
-isvd_set_config_var()
-
-# Configure files
-file(COPY "${PROJECT_CONFIG_DIR}/src/" DESTINATION "${CMAKE_CURRENT_CONFIG_DIR}")
-file(GLOB_RECURSE cfiles "${CMAKE_CURRENT_CONFIG_DIR}/*.c")
-foreach(cfile ${cfiles})
-  string(REGEX REPLACE "\\.c$" ".cxx" cxxfile ${cfile})
-  file(RENAME ${cfile} ${cxxfile})
-endforeach()
-
-file(RENAME "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/param.cxx"
-            "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/param.c")
-
-# checkisvd
-file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/*"
-                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/la/*"
-                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/util/*"
-                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/nogpu/*")
-add_library(checkisvd EXCLUDE_FROM_ALL ${files})
-isvd_set_target(checkisvd)
-isvd_set_target_omp(checkisvd CXX)
-isvd_set_target_mpi(checkisvd CXX)
-isvd_set_target_blas(checkisvd)
-isvd_set_target_gtest(checkisvd)
-
-# checkisvd_gpu
-if(ISVD_USE_GPU)
-  file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/*"
-                          "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/la/*"
-                          "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/util/*"
-                          "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/gpu/*")
-  add_library(checkisvd_gpu EXCLUDE_FROM_ALL ${files})
-  isvd_set_target(checkisvd_gpu)
-  isvd_set_target_omp(checkisvd_gpu CXX)
-  isvd_set_target_mpi(checkisvd_gpu CXX)
-  isvd_set_target_blas(checkisvd_gpu)
-  isvd_set_target_gpu(checkisvd_gpu)
-  isvd_set_target_gtest(checkisvd_gpu)
-endif()
diff --git a/cmake/display.cmake b/cmake/display.cmake
index d125268..3e635cd 100644
--- a/cmake/display.cmake
+++ b/cmake/display.cmake
@@ -153,4 +153,8 @@ if(ISVD_BUILD_BIN)
   if(ISVD_BLAS STREQUAL "MKL" AND NOT ISVD_OMP)
     message(${DEPRECATION} "${Esc}[1;33mOpenMP is not enabled. Recommended to use it for better performance.${Esc}[0m")
   endif()
+
+  if(CMAKE_COMPILER_IS_GNUCC AND NOT CMAKE_C_COMPILER_VERSION VERSION_GREATER 7 AND ISVD_USE_GPU AND CMAKE_BUILD_TYPE STREQUAL "Debug")
+    message(${DEPRECATION} "${Esc}[1;33mGPU routines with GCC under 7.0 might crash in debug mode.${Esc}[0m")
+  endif()
 endif()
diff --git a/cmake/options.cmake b/cmake/options.cmake
index 3a3158c..e5ea5f7 100644
--- a/cmake/options.cmake
+++ b/cmake/options.cmake
@@ -9,7 +9,7 @@ endif()
 option(ISVD_BUILD_LIB     "Build libraries."      "ON")
 option(ISVD_BUILD_DEMO    "Build demo codes."     "ON")
 option(ISVD_BUILD_TEST    "Build unit tests."     "OFF")
-option(ISVD_BUILD_DOC     "Build documentation."  "OFF")
+option(ISVD_BUILD_DOC     "Build documentation."  "ON")
 option(ISVD_USE_GPU       "Enable GPU support."   "OFF")
 option(ISVD_VERBOSE_TEST  "Verbose unit tests."   "ON")
 
@@ -79,4 +79,4 @@ endif()
 
 # Set environment variables
 list(APPEND ENVS "OMP_NUM_THREADS=${OMP_THRDS}")
-list(APPEND ENVS "ASAN_OPTIONS=color=always:protect_shadow_gap=0:replace_intrin=0:detect_leaks=0")
+list(APPEND ENVS "ASAN_OPTIONS=color=always:protect_shadow_gap=0")
diff --git a/cmake/vars.cmake b/cmake/vars.cmake
index 97d3464..c69fbcd 100644
--- a/cmake/vars.cmake
+++ b/cmake/vars.cmake
@@ -1,11 +1,11 @@
 # Types definitions
-macro(ISVD_SET_TYPES x_ xtype_ x_type_ XName_ XStr_)
+macro(ISVD_SET_TYPES x_ xtype_ XName_ XStr_)
   string(TOLOWER "${x_}" x)
   string(TOUPPER "${x_}" X)
 
-  set(xtype ${xtype_})
+  set(xtype_____ ${xtype_})
 
-  string(TOUPPER "${x_type_}" X_TYPE)
+  string(TOUPPER "${xtype_}" XTYPE)
 
   string(TOLOWER "${XName_}" xname)
   set(XName "${XName_}")
@@ -17,25 +17,25 @@ unset(ISVD_S_TYPES)
 unset(ISVD_D_TYPES)
 unset(ISVD_C_TYPES)
 unset(ISVD_Z_TYPES)
-list(APPEND ISVD_S_TYPES "s" "float"           "float"          "Real Single"    "RealSingle")
-list(APPEND ISVD_D_TYPES "d" "double"          "double"         "Real Double"    "RealDouble")
-list(APPEND ISVD_C_TYPES "c" "_Complex float"  "complex_float"  "Complex Single" "ComplexSingle")
-list(APPEND ISVD_Z_TYPES "z" "_Complex double" "complex_double" "Complex Double" "ComplexDouble")
+list(APPEND ISVD_S_TYPES "s" "isvd_s_val_t" "Real Single"    "RealSingle")
+list(APPEND ISVD_D_TYPES "d" "isvd_d_val_t" "Real Double"    "RealDouble")
+list(APPEND ISVD_C_TYPES "c" "isvd_c_val_t" "Complex Single" "ComplexSingle")
+list(APPEND ISVD_Z_TYPES "z" "isvd_z_val_t" "Complex Double" "ComplexDouble")
 
 # BLAS definitions
 set(
-  ISVD_LA_BLAS_TYPE_DEFINE
+  ISVD_TYPE_MACRO_DEFINE
   "#define CHAR1 char"
   "#define INT   isvd_int_t"
-  "#define REAL4 float"
-  "#define REAL8 double"
-  "#define COMP4 _Complex float"
-  "#define COMP8 _Complex double"
+  "#define REAL4 isvd_s_val_t"
+  "#define REAL8 isvd_d_val_t"
+  "#define COMP4 isvd_c_val_t"
+  "#define COMP8 isvd_z_val_t"
 )
-string(REPLACE ";" "\n" ISVD_LA_BLAS_TYPE_DEFINE "${ISVD_LA_BLAS_TYPE_DEFINE}")
+string(REPLACE ";" "\n" ISVD_TYPE_MACRO_DEFINE "${ISVD_TYPE_MACRO_DEFINE}")
 
 set(
-  ISVD_LA_BLAS_TYPE_UNDEF
+  ISVD_TYPE_MACRO_UNDEF
   "#undef CHAR1"
   "#undef INT"
   "#undef REAL4"
@@ -43,4 +43,4 @@ set(
   "#undef COMP4"
   "#undef COMP8"
 )
-string(REPLACE ";" "\n" ISVD_LA_BLAS_TYPE_UNDEF "${ISVD_LA_BLAS_TYPE_UNDEF}")
+string(REPLACE ";" "\n" ISVD_TYPE_MACRO_UNDEF "${ISVD_TYPE_MACRO_UNDEF}")
diff --git a/cpplint/CMakeLists.txt b/cpplint/CMakeLists.txt
index 4863902..642d100 100644
--- a/cpplint/CMakeLists.txt
+++ b/cpplint/CMakeLists.txt
@@ -5,15 +5,14 @@ find_program(CMAKE_CPPLINT cpplint)
 mark_as_advanced(CMAKE_CPPLINT)
 
 # Macro
-function(SET_CPPLINT_TARGET path)
+function(SET_CPPLINT_TARGET path linelength filter)
   string(REPLACE "/" "_" target ${path})
-  add_custom_target(
-    cpplint_${target}
-    COMMAND ${CMAKE_CPPLINT} --recursive --root=${PROJECT_CONFIG_DIR}/${path} ${ARGN} ${path}
-    WORKING_DIRECTORY ${PROJECT_CONFIG_DIR}
+  add_custom_command(
+    TARGET cpplint POST_BUILD
+    COMMAND ${CMAKE_CPPLINT} --recursive --root="${PROJECT_BINARY_DIR}/${path}" --linelength=${linelength} --filter=${filter} ${path}
+    WORKING_DIRECTORY "${PROJECT_BINARY_DIR}"
     COMMENT "Run cpplint for ${path}"
   )
-  add_dependencies(cpplint cpplint_${target})
 endfunction()
 
 # CppLint
@@ -24,5 +23,5 @@ add_custom_target(
   COMMENT "Run cpplint"
 )
 
-set_cpplint_target(include/c --linelength=128 --filter=-build/include,-build/include_order,-readability/casting,-whitespace/blank_line,-whitespace/braces,-whitespace/parens)
-set_cpplint_target(src --linelength=128 --filter=-build/include,-build/include_order,-build/include_what_you_use,-readability/casting,-whitespace/blank_line,-whitespace/braces,-whitespace/parens)
+set_cpplint_target(include/c 128 "-build/include,-build/include_order,-readability/casting,-whitespace/blank_line,-whitespace/braces,-whitespace/parens")
+set_cpplint_target(lib 128 "-build/include,-build/include_order,-build/include_what_you_use,-readability/casting,-whitespace/blank_line,-whitespace/braces,-whitespace/parens")
diff --git a/demo/CMakeLists.txt b/demo/CMakeLists.txt
index ba8b38a..5447745 100644
--- a/demo/CMakeLists.txt
+++ b/demo/CMakeLists.txt
@@ -23,9 +23,13 @@ macro(SET_TARGET name files)
 endmacro()
 
 # Set include paths
-include_directories("${PROJECT_CONFIG_DIR}/include/c")
+include_directories("${PROJECT_BINARY_DIR}/include/c")
 
 # C demo
 set_target(cdemo cdemo.c ${PROJECT_SOURCE_DIR}/data/a.mtx /dev/null /dev/null /dev/null)
 target_link_libraries(isvd_cdemo isvd extmmio)
-target_include_directories(isvd_cdemo ${SYSTEM} PUBLIC "${PROJECT_SOURCE_DIR}/ext")
+target_include_directories(isvd_cdemo ${SYSTEM} PUBLIC "${PROJECT_SOURCE_DIR}/ext/mmio")
+
+# C test
+set_target(ctest ctest.c)
+target_link_libraries(isvd_ctest isvd)
diff --git a/demo/cdemo.c b/demo/cdemo.c
index 7cae048..7060168 100644
--- a/demo/cdemo.c
+++ b/demo/cdemo.c
@@ -10,7 +10,7 @@
 #include <stdlib.h>
 #include <isvd.h>
 #include <mpi.h>
-#include <mmio/mmio.h>
+#include <mmio.h>
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// Main function
@@ -53,7 +53,7 @@ int main( int argc, char **argv ) {
   if ( !mm_is_array(matcode) || !mm_is_real(matcode) || !mm_is_general(matcode) ) {
     if ( mpi_rank == mpi_root ) {
       fprintf(stderr, "Sorry, this application does not support ");
-      fprintf(stderr, "Market Market type: [%s]\n", mm_typecode_to_str(matcode));
+      fprintf(stderr, "Matrix Market type: [%s]\n", mm_typecode_to_str(matcode));
       MPI_Abort(MPI_COMM_WORLD, 1);
     }
     MPI_Barrier(MPI_COMM_WORLD);
diff --git a/demo/ctest.c b/demo/ctest.c
new file mode 100644
index 0000000..c8e2fcb
--- /dev/null
+++ b/demo/ctest.c
@@ -0,0 +1,40 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       demo/ctest.c
+/// \brief      The C test code
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <isvd.h>
+#include <mpi.h>
+
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// Main function
+///
+int main( int argc, char **argv ) {
+
+  isvd_init(&argc, &argv, MPI_COMM_WORLD);
+
+  const mpi_int_t mpi_size = isvd_getMpiSize(MPI_COMM_WORLD);
+  const mpi_int_t mpi_rank = isvd_getMpiRank(MPI_COMM_WORLD);
+  const mpi_int_t mpi_root = 0;
+
+  if ( mpi_rank == mpi_root ) {
+    printf("iSVD C demo\n");
+    isvd_printEnvironment(MPI_COMM_WORLD);
+  }
+
+  const isvd_int_t m = 100;
+  const isvd_int_t n = 1000;
+
+  double *a = isvd_dmalloc(m * n);
+
+  isvd_vdRngGaussianDriver(0, 0, m*n, a, 0.0, 1.0);
+
+  isvd_finalize();
+
+  return 0;
+}
diff --git a/doxygen/Doxyfile b/doxygen/Doxyfile
index 5a0d8cd..abc5966 100644
--- a/doxygen/Doxyfile
+++ b/doxygen/Doxyfile
@@ -152,7 +152,8 @@ FULL_PATH_NAMES        = YES
 # will be relative from the directory where doxygen is started.
 # This tag requires that the tag FULL_PATH_NAMES is set to YES.
 
-STRIP_FROM_PATH        = @PROJECT_CONFIG_DIR@
+STRIP_FROM_PATH        = @CMAKE_CURRENT_CONFIG_DIR@ \
+                         @PROJECT_CONFIG_DIR@/src
 
 # The STRIP_FROM_INC_PATH tag can be used to strip a user-defined part of the
 # path mentioned in the documentation of a class, which tells the reader which
@@ -161,7 +162,7 @@ STRIP_FROM_PATH        = @PROJECT_CONFIG_DIR@
 # specify the list of include paths that are normally passed to the compiler
 # using the -I flag.
 
-STRIP_FROM_INC_PATH    = @PROJECT_CONFIG_DIR@
+STRIP_FROM_INC_PATH    =
 
 # If the SHORT_NAMES tag is set to YES, doxygen will generate much shorter (but
 # less readable) file names. This can be useful is your file systems doesn't
@@ -785,7 +786,6 @@ WARN_LOGFILE           =
 
 INPUT                  = @PROJECT_SOURCE_DIR@/README.md \
                          @PROJECT_CONFIG_DIR@/doxygen \
-                         @PROJECT_CONFIG_DIR@/include \
                          @PROJECT_CONFIG_DIR@/src
 
 # This tag can be used to specify the character encoding of the source files
diff --git a/doxygen/example.dox b/doxygen/example.dox
new file mode 100644
index 0000000..b4ed890
--- /dev/null
+++ b/doxygen/example.dox
@@ -0,0 +1,3 @@
+/**
+  \example demo/cdemo.c
+*/
\ No newline at end of file
diff --git a/doxygen/ext/mkl.h b/doxygen/ext/mkl.h
new file mode 100644
index 0000000..e7d83fc
--- /dev/null
+++ b/doxygen/ext/mkl.h
@@ -0,0 +1,4 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file   ext/mkl.h
+/// \brief  The Intel MKL header.
+///
diff --git a/doxygen/ext/mpi.h b/doxygen/ext/mpi.h
new file mode 100644
index 0000000..a74402d
--- /dev/null
+++ b/doxygen/ext/mpi.h
@@ -0,0 +1,4 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file   ext/mpi.h
+/// \brief  The MPI header.
+///
diff --git a/doxygen/ext/omp.h b/doxygen/ext/omp.h
new file mode 100644
index 0000000..1de9750
--- /dev/null
+++ b/doxygen/ext/omp.h
@@ -0,0 +1,4 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file   ext/omp.h
+/// \brief  The OpenMP header.
+///
diff --git a/doxygen/tutorial/core/detail.dox b/doxygen/tutorial/core/detail.dox
new file mode 100644
index 0000000..f786b2a
--- /dev/null
+++ b/doxygen/tutorial/core/detail.dox
@@ -0,0 +1,36 @@
+/**
+  @page tutorial_core_notation Notations
+
+  <h3>Format Notations</h3>
+  Name                           | Format                                 | Example
+  -------------------------------|----------------------------------------|--------------------------------
+  scalars                        | normal italic                          | \f$m, \alpha, N\f$
+  vectors                        | bold italic lowercase                  | \f$\boldsymbol{v}, \boldsymbol{\beta}\f$
+  matrices                       | bold italic uppercase                  | \f$\boldsymbol{A}, \boldsymbol{\Omega}\f$
+  combined matrices              | bold Frankfurt uppercase               | \f$\boldsymbol{\mathfrak{Q}}, \boldsymbol{\mathfrak{Y}}\f$
+  submatrices                    | sans serif / Frankfurt uppercase       | \f$\mathsf{U}, \mathfrak{Y}\f$
+  things in \f$i\f$-th sketch    | under-script bracketed                 | \f$\boldsymbol{Q}_{[i]}, \boldsymbol{Y}_{[i]}\f$
+  things in \f$j\f$-th process   | super-script parenthesized / angled    | \f$\mathsf{V}^{(j)}, \mathsf{A}^{\langle j \rangle}\f$
+
+  <h3>Parameter Notations</h3>
+
+  - \f$m\f$: the number of rows of the input matrix \f$\boldsymbol{A}\f$.
+  - \f$n\f$: the number of columns of the input matrix \f$\boldsymbol{A}\f$.
+  - \f$k\f$: the desired rank of approximate SVD.
+  - \f$p\f$: the oversampling parameter.
+  - \f$l\f$: the dimension of randomized sketches, i.e., \f$l=k+p\f$.
+  - \f$N\f$: the number of random sketches.
+  - \f$P\f$: the number of MPI processors.
+  - \f$m_b\f$: the row dimensions of a row-block, i.e., \f$m_b = \lceil \frac{m}{P} \rceil\f$.
+  - \f$n_b\f$: the column dimensions of a column-block, i.e., \f$n_b = \lceil \frac{n}{P} \rceil\f$.
+  - \f$m^{(j)}\f$: the exact row dimensions of the \f$j\f$-th row-block.
+    - For \f$j < P-1\f$, \f$m^{(j)} = m_b\f$.
+    - For \f$j = P-1\f$, \f$m^{(j)} = m - m_b(P-1)\f$.
+    - \f$m^{(0)} + m^{(1)} + \dotsb + m^{(P-1)} = m\f$.
+  - \f$n^{(j)}\f$: the exact column dimensions of the \f$j\f$-th column-block.
+    - For \f$j < P-1\f$, \f$n^{(j)} = n_b\f$.
+    - For \f$j = P-1\f$, \f$n^{(j)} = n - n_b(P-1)\f$.
+    - \f$n^{(0)} + n^{(1)} + \dotsb + n^{(P-1)} = n\f$.
+
+  \see  isvd_Param
+*/
diff --git a/doxygen/tutorial/core/example.dox b/doxygen/tutorial/core/example.dox
index b1dcde7..60ca091 100644
--- a/doxygen/tutorial/core/example.dox
+++ b/doxygen/tutorial/core/example.dox
@@ -1,52 +1,72 @@
 /**
-  \page tutorial_core_example iSVD Example
+  \page tutorial_core_example Example Code
 
-  <h3>iSVD</h3>
+  \ref isvd_dIsvd "iSVD driver" solves an approximate low-rank singular value decomposition of matrix 𝑨. \ref demo/cdemo.c is an example code. In the following, we describe the details of each part of this example.
 
-  iSVD solves an approximate low-rank singular value decomposition of matrix `a`.
-
-  <h3>Example</h3>
-
-  The following is an example of using \ref isvd_dIsvd "iSVD driver". The code is also available in `<ISVD-DIR>/demo/cdemo.c`.
-
-  \includelineno demo/cdemo.c
-
-  <h4>Initialize Environment</h4>
+  <h3>Initialize Environment</h3>
 
   The iSVD environment and random seed should be set up before calling the driver. \ref isvd_init also initialize the MPI environment and the MAGMA environment, and \ref isvd_printEnvironment display the the MPI environment and the MAGMA environment.
 
-  If one needs CPU only, one may use \ref isvd_init_cpu, \ref isvd_printEnvironment_cpu, and \ref isvd_finalize_cpu.
+  If one needs CPU only, one may use \ref isvd_init_cpu, \ref isvd_printEnvironment_cpu, and \ref isvd_finalize_cpu instead.
 
   \snippetlineno demo/cdemo.c init-isvd
 
-  <h4>Load Data</h4>
+  <h3>Load Data</h3>
 
-  In this example, we load the data from a Matrix Market format file. Note the all the MPI nodes should load the matrix.
+  In this example, we load the data from a Matrix Market format file. Note the all the MPI processors should load the matrix.
 
   \snippetlineno demo/cdemo.c load-data
 
-  <h4>Allocate matrices</h4>
+  <h3>Allocate matrices</h3>
 
   The output matrices should be allocated before calling the driver. See \ref isvd_dIsvd for the detail of matrix sizes.
 
   \snippetlineno demo/cdemo.c allocate-matrix
 
-  <h4>Run iSVD</h4>
+  <h3>Run iSVD</h3>
 
-  The iSVD driver compute the approximate low-rank singular value decomposition. See \ref isvd_dIsvd for the detail of the arguments.
+  The iSVD driver compute the approximate low-rank singular value decomposition. See \ref isvd_dIsvd for the detail of the arguments. The 𝑗-th MPI-rank stores the 𝑗-th block of the input matrix 𝑨 (see \ref tutorial_core_storage for detail).
 
   \snippetlineno demo/cdemo.c run-isvd
 
-  <h4>Get Executing Time</h4>
+  <h3>Get Executing Time</h3>
 
-  iSVD also records the executing time of each stage.
+  The iSVD driver also records the executing time of each stage.
 
   \snippetlineno demo/cdemo.c display-time
 
-  <h4>Finalize</h4>
+  <h3>Finalize</h3>
 
   Remember to \ref isvd_finalize "finalize" the environment.
 
   \snippetlineno demo/cdemo.c final-isvd
 
+  <h2>Possible Result</h2>
+
+  One may use **make run_cdemo** to run \ref demo/cdemo.c. If you want to compile it directly, please refer the usage guide in the main page. The following are the possible results of the example code.
+
+  \code{.txt}
+  iSVD C demo
+  iSVD 1.2.0, 32-bit isvd_int_t, 64-bit pointer
+  4 MPI processors, 4 OpenMP threads per process
+
+  No GPU support
+
+  Using Real Double Precision Gaussian Projection Sketching
+  Using Real Double Precision Gramian Orthogonalization
+  Using Real Double Precision Kolmogorov-Nagumo Integration
+  Using Real Double Precision Gramian Postprocessing
+
+  Sketching ...................... done
+  Orthogonalizing ................ done
+  Integrating .................... done
+  Postprocessing ................. done
+
+  Average total computing time:   0.085510 seconds.
+  Average sketching time:         0.030245 seconds.
+  Average orthogonalizing time:   0.028071 seconds.
+  Average integrating time:       0.026233 seconds.
+  Average postprocessing time:    0.000962 seconds.
+  \endcode
+
 */
diff --git a/doxygen/tutorial/core/stage.dox b/doxygen/tutorial/core/stage.dox
index f94ae11..a19f483 100644
--- a/doxygen/tutorial/core/stage.dox
+++ b/doxygen/tutorial/core/stage.dox
@@ -1,9 +1,31 @@
 /**
-  \page tutorial_core_stage iSVD Stages
+  @page tutorial_core_stage Stages
+
+  The iSVD algorithm contains four stages: sketching, orthogonalization, integration, and postprocessing.
+
+  Stage               | Input                                              | Output
+  --------------------|----------------------------------------------------|--------------------------------
+  sketching           | \f$\boldsymbol{A}\f$                               | \f$\boldsymbol{\mathfrak{Y}}\f$
+  orthogonalization   | \f$\boldsymbol{\mathfrak{Y}}\f$                    | \f$\boldsymbol{\mathfrak{Q}}\f$
+  integration         | \f$\boldsymbol{\mathfrak{Q}}\f$                    | \f$\overline{\boldsymbol{Q}}\f$
+  postprocessing      | \f$\boldsymbol{A}, \overline{\boldsymbol{Q}}\f$    | \f$\boldsymbol{U}, \boldsymbol{V}, \boldsymbol{\sigma}\f$
+
+  Note that the Frankfurt matrices are defined as
+  @f[
+    \boldsymbol{\mathfrak{Y}} =
+      \begin{bmatrix}
+        \boldsymbol{Y}_{[0]} & \boldsymbol{Y}_{[1]} & \cdots & \boldsymbol{Y}_{[N-1]}
+      \end{bmatrix},
+    \qquad
+    \boldsymbol{\mathfrak{Q}} =
+      \begin{bmatrix}
+        \boldsymbol{Q}_{[0]} & \boldsymbol{Q}_{[1]} & \cdots & \boldsymbol{Q}_{[N-1]}
+      \end{bmatrix}.
+  @f]
 
   <h3>Sketching</h3>
 
-  The sketching stage randomly sketches \ref isvd_Param::num_sketch "𝑁" rank-\ref isvd_Param::dim_sketch "𝑙" column subspaces \f$\boldsymbol{Y}_{[i]}\f$ of the input matrix \f$\boldsymbol{A}\f$.
+  The sketching stage randomly sketches 𝑁 rank-𝑙 column subspaces \f$\boldsymbol{Y}_{[i]}\f$ of the input matrix \f$\boldsymbol{A}\f$.
 
   - The \ref isvd_dSketchGaussianProjection "Gramian Projection Sketching" multiples \f$\boldsymbol{A}\f$ by some random matrices using Gaussian normal distribution.
 
@@ -23,9 +45,8 @@
 
   <h3>Postprocessing</h3>
 
-  The postprocessing stage forms a rank-\ref isvd_Param::rank "𝑘" approximate SVD of \f$\boldsymbol{A}\f$ in the range of \f$\overline{\boldsymbol{Q}}\f$.
+  The postprocessing stage forms a rank-𝑘 approximate SVD of \f$\boldsymbol{A}\f$ in the range of \f$\overline{\boldsymbol{Q}}\f$.
 
   - The \ref isvd_dPostprocessGramian "Gramian Postprocessing" finds approximate SVD using the eigen-decomposition of \f$\overline{\boldsymbol{Q}}^\top \boldsymbol{A} \boldsymbol{A}^\top \overline{\boldsymbol{Q}}\f$.
   - The \ref isvd_dPostprocessSymmetric "Symmetric Postprocessing" finds approximate SVD using the eigen-decomposition of \f$\overline{\boldsymbol{Q}}^\top \boldsymbol{A} \overline{\boldsymbol{Q}}\f$ for symmetric input matrix \f$\boldsymbol{A}\f$.
-
 */
diff --git a/doxygen/tutorial/core/storage.dox b/doxygen/tutorial/core/storage.dox
new file mode 100644
index 0000000..875354c
--- /dev/null
+++ b/doxygen/tutorial/core/storage.dox
@@ -0,0 +1,58 @@
+/**
+  \page tutorial_core_storage Parallelism Storage
+
+  <h3>Block-Row Storage</h3>
+
+  In Block-Row Parallelism, we storage the matrices in row-blocks. The \f$j\f$-th block is stored in the \f$j\f$-th MPI-rank.
+
+  @f[
+    \boldsymbol{A} =
+      \begin{bmatrix}
+        \mathsf{A}^{(0)} \\ \mathsf{A}^{(1)} \\ \vdots \\ \mathsf{A}^{(P-1)}
+      \end{bmatrix},
+    \qquad
+    \boldsymbol{\mathfrak{Y}} =
+      \begin{bmatrix}
+        \mathfrak{Y}^{(0)} \\ \mathfrak{Y}^{(1)} \\ \vdots \\ \mathfrak{Y}^{(P-1)}
+      \end{bmatrix},
+    \qquad
+    \boldsymbol{\mathfrak{Q}} =
+      \begin{bmatrix}
+        \mathfrak{Q}^{(0)} \\ \mathfrak{Q}^{(1)} \\ \vdots \\ \mathfrak{Q}^{(P-1)}
+      \end{bmatrix},
+    \qquad
+    \overline{\boldsymbol{Q}} =
+      \begin{bmatrix}
+        \overline{\mathsf{Q}}^{(0)} \\ \overline{\mathsf{Q}}^{(1)} \\ \vdots \\ \overline{\mathsf{Q}}^{(P-1)}
+      \end{bmatrix},
+    \qquad
+    \boldsymbol{U} =
+      \begin{bmatrix}
+        \mathsf{U}^{(0)} \\ \mathsf{U}^{(1)} \\ \vdots \\ \mathsf{U}^{(P-1)}
+      \end{bmatrix},
+    \qquad
+    \boldsymbol{V} =
+      \begin{bmatrix}
+        \mathsf{V}^{(0)} \\ \mathsf{V}^{(1)} \\ \vdots \\ \mathsf{V}^{(P-1)}
+      \end{bmatrix}.
+  @f]
+
+  The \f$j\f$-th block of the input matrices should contain \f$m^{(j)}\f$ rows, and the \f$j\f$-th block of the input matrices should contain \f$m_b\f$ rows.
+
+  For example, let \f$\boldsymbol{A}\f$ be a matrix with 11 rows and \f$P\f$ is 4. Then \f$\mathsf{A}^{(0)}\f$ are the (0-2)-th rows, \f$\mathsf{A}^{(1)}\f$ are the (0-5)-th rows, \f$\mathsf{A}^{(2)}\f$ are the (6-8)-th rows, \f$\mathsf{A}^{(3)}\f$ are the (9-10)-th rows.
+
+  <h3>Block-Column Storage</h3>
+
+  In Block-Column Parallelism, we storage the input matrix \f$\boldsymbol{A}\f$ in row-blocks. The \f$j\f$-th block is stored in the \f$j\f$-th MPI-rank.
+
+  @f[
+    \boldsymbol{A} =
+      \begin{bmatrix}
+        \mathsf{A}^{\langle 0 \rangle} & \mathsf{A}^{\langle 1 \rangle} & \cdots & \mathsf{A}^{\langle P-1 \rangle}
+      \end{bmatrix}.
+  @f]
+
+  The \f$j\f$-th block of the input matrices should contain \f$n^{(j)}\f$ columns, and the \f$j\f$-th block of the input matrices should contain \f$n_b\f$ columns.
+
+  For example, let \f$\boldsymbol{A}\f$ be a matrix with 11 columns and \f$P\f$ is 4. Then \f$\mathsf{A}^{\langle 0 \rangle}\f$ are the (0-2)-th columns, \f$\mathsf{A}^{\langle 1 \rangle}\f$ are the (0-5)-th columns, \f$\mathsf{A}^{\langle 2 \rangle}\f$ are the (6-8)-th columns, \f$\mathsf{A}^{\langle 3 \rangle}\f$ are the (9-10)-th columns.
+*/
diff --git a/doxygen/tutorial/main.dox b/doxygen/tutorial/main.dox
index b01f899..b0d8a87 100644
--- a/doxygen/tutorial/main.dox
+++ b/doxygen/tutorial/main.dox
@@ -1,8 +1,10 @@
 /**
   \page tutorial_main Tutorial
 
-  <h4>Subpages</h4>
+  <h2>Subpages</h2>
 
   - \subpage tutorial_core_example
+  - \subpage tutorial_core_notation
   - \subpage tutorial_core_stage
+  - \subpage tutorial_core_storage
 */
diff --git a/include/CMakeLists.txt b/include/CMakeLists.txt
deleted file mode 100644
index 24c4f82..0000000
--- a/include/CMakeLists.txt
+++ /dev/null
@@ -1,11 +0,0 @@
-# The CMake setting of 'include/'
-
-isvd_set_config_var()
-
-# Configure files
-isvd_configure_x_fn("${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_CONFIG_DIR}" "${ISVD_S_TYPES}")
-isvd_configure_x_fn("${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_CONFIG_DIR}" "${ISVD_D_TYPES}")
-isvd_configure_fn("${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_CONFIG_DIR}")
-
-# Set install rule
-install(DIRECTORY "${CMAKE_CURRENT_CONFIG_DIR}/c/" DESTINATION include/c)
diff --git a/include/c/isvd/core/@x@_stage.h b/include/c/isvd/core/@x@_stage.h
deleted file mode 100644
index 4c2b9ff..0000000
--- a/include/c/isvd/core/@x@_stage.h
+++ /dev/null
@@ -1,79 +0,0 @@
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       include/c/isvd/core/@x@_stage.h
-/// \brief      The iSVD stages.
-///
-/// \author     Mu Yang <<emfomy@gmail.com>>
-/// \copyright  MIT License
-///
-
-#ifndef ISVD_CORE_@X@_STAGE_H_
-#define ISVD_CORE_@X@_STAGE_H_
-
-#include <isvd/def.h>
-#include <isvd/core/param.h>
-
-#if defined(__cplusplus)
-extern "C" {
-#endif  // __cplusplus
-
-// Sketching
-void isvd_@x@SketchGaussianProjection(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda,
-    @xtype@ *yst, const isvd_int_t ldyst, const isvd_int_t seed, const mpi_int_t mpi_root
-);
-
-// Orthogonalization
-void isvd_@x@OrthogonalizeTallSkinnyQr(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    @xtype@ *yst, const isvd_int_t ldyst
-);
-
-void isvd_@x@OrthogonalizeGramian(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    @xtype@ *yst, const isvd_int_t ldyst
-);
-
-// Integration
-void isvd_@x@IntegrateKolmogorovNagumo(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const @xtype@ *yst, const isvd_int_t ldyst, @xtype@ *qt, const isvd_int_t ldqt
-);
-
-void isvd_@x@IntegrateWenYin(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const @xtype@ *yst, const isvd_int_t ldyst, @xtype@ *qt, const isvd_int_t ldqt
-);
-
-void isvd_@x@IntegrateHierarchicalReduction(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-          @xtype@ *yst, const isvd_int_t ldyst, @xtype@ *qt, const isvd_int_t ldqt
-);
-
-// Postprocessing
-void isvd_@x@PostprocessTallSkinnyQr(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda, const @xtype@ *qt, const isvd_int_t ldqt,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
-    const mpi_int_t ut_root, const mpi_int_t vt_root
-);
-
-void isvd_@x@PostprocessGramian(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda, const @xtype@ *qt, const isvd_int_t ldqt,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
-    const mpi_int_t ut_root, const mpi_int_t vt_root
-);
-
-void isvd_@x@PostprocessSymmetric(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda, const @xtype@ *qt, const isvd_int_t ldqt,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
-    const mpi_int_t ut_root, const mpi_int_t vt_root
-);
-
-#if defined(__cplusplus)
-}
-#endif  // __cplusplus
-
-#endif  // ISVD_CORE_@X@_STAGE_H_
diff --git a/include/c/isvd/util/memory.h b/include/c/isvd/util/memory.h
deleted file mode 100644
index c238b9e..0000000
--- a/include/c/isvd/util/memory.h
+++ /dev/null
@@ -1,42 +0,0 @@
-////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       include/c/isvd/util/memory.h
-/// \brief      The memory utilities.
-///
-/// \author     Mu Yang <<emfomy@gmail.com>>
-/// \copyright  MIT License
-///
-
-#ifndef ISVD_UTIL_MEMORY_H_
-#define ISVD_UTIL_MEMORY_H_
-
-#include <isvd/def.h>
-
-#if defined(__cplusplus)
-extern "C" {
-#endif  // __cplusplus
-
-isvd_int_t* isvd_imalloc( const size_t num );
-float* isvd_smalloc( const size_t num );
-double* isvd_dmalloc( const size_t num );
-_Complex float* isvd_cmalloc( const size_t num );
-_Complex double* isvd_zmalloc( const size_t num );
-
-void isvd_free( void *ptr );
-
-void isvd_imemset0( isvd_int_t *ptr, const size_t num );
-void isvd_smemset0( float *ptr, const size_t num );
-void isvd_dmemset0( double *ptr, const size_t num );
-void isvd_cmemset0( _Complex float *ptr, const size_t num );
-void isvd_zmemset0( _Complex double *ptr, const size_t num );
-
-void isvd_imemcpy( isvd_int_t *dst, const isvd_int_t *src, const size_t num );
-void isvd_smemcpy( float *dst, const float *src, const size_t num );
-void isvd_dmemcpy( double *dst, const double *src, const size_t num );
-void isvd_cmemcpy( _Complex float *dst, const _Complex float *src, const size_t num );
-void isvd_zmemcpy( _Complex double *dst, const _Complex double *src, const size_t num );
-
-#if defined(__cplusplus)
-}
-#endif  // __cplusplus
-
-#endif  // ISVD_UTIL_MEMORY_H_
diff --git a/share/isvd/isvd-config.cmake b/share/isvd/isvd-config.cmake
deleted file mode 100644
index bfcc580..0000000
--- a/share/isvd/isvd-config.cmake
+++ /dev/null
@@ -1,3 +0,0 @@
-# Compute paths
-set(ISVD_INCLUDES  "@CMAKE_INSTALL_PREFIX@/include")
-set(ISVD_LIBRARIES "@CMAKE_INSTALL_PREFIX@/@LIB_FOLDER@/libisvd.so")
diff --git a/src/CMakeLists.txt b/src/CMakeLists.txt
index 369cf18..8a9d2a0 100644
--- a/src/CMakeLists.txt
+++ b/src/CMakeLists.txt
@@ -2,83 +2,18 @@
 
 isvd_set_config_var()
 
+# Shared library
 SET(BUILD_SHARED_LIBS ON)
-
-# RPath
 SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${LIB_FOLDER}")
 set(CMAKE_INSTALL_RPATH_USE_LINK_PATH ON)
 
-# Macro
-macro(SET_TARGET target files)
-  # Set library
-  add_library(${target} SHARED ${files})
-  isvd_set_target(${target})
-  set_property(TARGET ${target} PROPERTY VERSION ${ISVD_VERSION})
-  install(TARGETS ${target} LIBRARY DESTINATION ${LIB_FOLDER})
-endmacro()
-
-# Set include paths
-include_directories("${PROJECT_CONFIG_DIR}/include/c"
-                    "${CMAKE_CURRENT_CONFIG_DIR}")
-
 # Configure files
 isvd_configure_x_fn("${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_CONFIG_DIR}" "${ISVD_S_TYPES}")
 isvd_configure_x_fn("${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_CONFIG_DIR}" "${ISVD_D_TYPES}")
 isvd_configure_fn("${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_CONFIG_DIR}")
 
-# Break
-if(NOT ISVD_BUILD_LIB)
-  return()
-endif()
-
-# libisvd
-set_target(isvd /dev/null)
-set_property(TARGET isvd PROPERTY LINKER_LANGUAGE C)
-
-# libisvd_core
-file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/*"
-                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/util/*")
-set_target(isvd_core "${files}")
-isvd_set_target_omp(isvd_core C)
-isvd_set_target_mpi(isvd_core C)
-isvd_set_target_blas(isvd_core)
-target_link_libraries(isvd isvd_core)
-
-# libisvd_la
-if(ISVD_OMP)
-  string(TOLOWER ${ISVD_OMP} isvd_omp)
-  set(isvd_omp "_${isvd_omp}")
-else()
-  set(isvd_omp "")
-endif()
-if(ISVD_USE_MKL)
-  set(isvd_la isvd_la_mkl${isvd_omp})
-else()
-  set(isvd_la isvd_la_blas${isvd_omp})
-endif(ISVD_USE_MKL)
-file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/la/*")
-set_target(${isvd_la} "${files}")
-isvd_set_target_omp(${isvd_la} C)
-isvd_set_target_mpi(${isvd_la} C)
-isvd_set_target_blas(${isvd_la})
-target_link_libraries(isvd ${isvd_la})
-
-# libisvd_gpu_none
-file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/nogpu/*")
-set_target(isvd_gpu_none "${files}")
-isvd_set_target_omp(isvd_gpu_none C)
-isvd_set_target_mpi(isvd_gpu_none C)
-isvd_set_target_blas(isvd_gpu_none)
+add_subdirectory(include)
+execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink "${CMAKE_CURRENT_CONFIG_DIR}/include" "${PROJECT_BINARY_DIR}/include")
 
-# libisvd_gpu
-if(ISVD_USE_GPU)
-  file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/gpu/*")
-  set_target(isvd_gpu_magma "${files}")
-  isvd_set_target_omp(isvd_gpu_magma C)
-  isvd_set_target_mpi(isvd_gpu_magma C)
-  isvd_set_target_blas(isvd_gpu_magma)
-  isvd_set_target_gpu(isvd_gpu_magma)
-  target_link_libraries(isvd isvd_gpu_magma)
-else()
-  target_link_libraries(isvd isvd_gpu_none)
-endif(ISVD_USE_GPU)
+add_subdirectory(lib)
+execute_process(COMMAND ${CMAKE_COMMAND} -E create_symlink "${CMAKE_CURRENT_CONFIG_DIR}/lib" "${PROJECT_BINARY_DIR}/lib")
diff --git a/src/include/CMakeLists.txt b/src/include/CMakeLists.txt
new file mode 100644
index 0000000..c9c9494
--- /dev/null
+++ b/src/include/CMakeLists.txt
@@ -0,0 +1,4 @@
+# The CMake setting of 'src/include/'
+
+# Set install rule
+install(DIRECTORY "${CMAKE_CURRENT_CONFIG_DIR}/c/" DESTINATION include/c)
diff --git a/include/c/isvd.h b/src/include/c/isvd.h
similarity index 98%
rename from include/c/isvd.h
rename to src/include/c/isvd.h
index 49e5fc7..19656dc 100644
--- a/include/c/isvd.h
+++ b/src/include/c/isvd.h
@@ -9,7 +9,6 @@
 #ifndef ISVD_H_
 #define ISVD_H_
 
-#include <isvd/def.h>
 #include <isvd/core.h>
 #include <isvd/gpu.h>
 #include <isvd/la.h>
diff --git a/include/c/isvd/config.h b/src/include/c/isvd/config.h
similarity index 100%
rename from include/c/isvd/config.h
rename to src/include/c/isvd/config.h
diff --git a/include/c/isvd/core.h b/src/include/c/isvd/core.h
similarity index 100%
rename from include/c/isvd/core.h
rename to src/include/c/isvd/core.h
diff --git a/include/c/isvd/core/@x@_driver.h b/src/include/c/isvd/core/@x@_driver.h
similarity index 76%
rename from include/c/isvd/core/@x@_driver.h
rename to src/include/c/isvd/core/@x@_driver.h
index 616537e..69d7942 100644
--- a/include/c/isvd/core/@x@_driver.h
+++ b/src/include/c/isvd/core/@x@_driver.h
@@ -19,10 +19,10 @@ extern "C" {
 void isvd_@x@Isvd(
     const char *alg_s, const char *alg_o, const char *alg_i, const char *alg_p,
     const isvd_int_t m, const isvd_int_t n, const isvd_int_t k, const isvd_int_t p, const isvd_int_t N,
-    const @xtype@ *argv[4], const isvd_int_t argc[4], @xtype@ *retv[4], const isvd_int_t retc[4],
+    const @xtype_____@ *argv[4], const isvd_int_t argc[4], @xtype_____@ *retv[4], const isvd_int_t retc[4],
     double time[4], FILE *stream,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
+    const char dista, const char ordera, const @xtype_____@ *a, const isvd_int_t lda,
+    @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
     const isvd_int_t seed, const mpi_int_t ut_root, const mpi_int_t vt_root,
     const mpi_int_t mpi_root, const isvd_MpiComm mpi_comm
 );
diff --git a/src/include/c/isvd/core/@x@_stage.h b/src/include/c/isvd/core/@x@_stage.h
new file mode 100644
index 0000000..c54cdb8
--- /dev/null
+++ b/src/include/c/isvd/core/@x@_stage.h
@@ -0,0 +1,82 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       include/c/isvd/core/@x@_stage.h
+/// \brief      The iSVD stages.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#ifndef ISVD_CORE_@X@_STAGE_H_
+#define ISVD_CORE_@X@_STAGE_H_
+
+#include <isvd/def.h>
+#include <isvd/core/param.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif  // __cplusplus
+
+// Sketching
+void isvd_@x@SketchGaussianProjection(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera, const @xtype_____@ *a, const isvd_int_t lda,
+    @xtype_____@ *yst, const isvd_int_t ldyst, const isvd_int_t seed, const mpi_int_t mpi_root
+);
+
+// Orthogonalization
+void isvd_@x@OrthogonalizeTallSkinnyQr(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    @xtype_____@ *yst, const isvd_int_t ldyst
+);
+
+void isvd_@x@OrthogonalizeGramian(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    @xtype_____@ *yst, const isvd_int_t ldyst
+);
+
+// Integration
+void isvd_@x@IntegrateKolmogorovNagumo(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const @xtype_____@ *yst, const isvd_int_t ldyst, @xtype_____@ *qt, const isvd_int_t ldqt
+);
+
+void isvd_@x@IntegrateWenYin(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const @xtype_____@ *yst, const isvd_int_t ldyst, @xtype_____@ *qt, const isvd_int_t ldqt
+);
+
+void isvd_@x@IntegrateHierarchicalReduction(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+          @xtype_____@ *yst, const isvd_int_t ldyst, @xtype_____@ *qt, const isvd_int_t ldqt
+);
+
+// Postprocessing
+void isvd_@x@PostprocessTallSkinnyQr(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera,
+    const @xtype_____@ *a, const isvd_int_t lda, const @xtype_____@ *qt, const isvd_int_t ldqt,
+    @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
+    const mpi_int_t ut_root, const mpi_int_t vt_root
+);
+
+void isvd_@x@PostprocessGramian(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera,
+    const @xtype_____@ *a, const isvd_int_t lda, const @xtype_____@ *qt, const isvd_int_t ldqt,
+          @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
+    const mpi_int_t ut_root, const mpi_int_t vt_root
+);
+
+void isvd_@x@PostprocessSymmetric(
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera,
+    const @xtype_____@ *a, const isvd_int_t lda, const @xtype_____@ *qt, const isvd_int_t ldqt,
+          @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
+    const mpi_int_t ut_root, const mpi_int_t vt_root
+);
+
+#if defined(__cplusplus)
+}
+#endif  // __cplusplus
+
+#endif  // ISVD_CORE_@X@_STAGE_H_
diff --git a/include/c/isvd/core/env.h b/src/include/c/isvd/core/env.h
similarity index 100%
rename from include/c/isvd/core/env.h
rename to src/include/c/isvd/core/env.h
diff --git a/include/c/isvd/core/param.h b/src/include/c/isvd/core/param.h
similarity index 94%
rename from include/c/isvd/core/param.h
rename to src/include/c/isvd/core/param.h
index 59d849b..56e38fc 100644
--- a/include/c/isvd/core/param.h
+++ b/src/include/c/isvd/core/param.h
@@ -19,6 +19,8 @@ extern "C" {
 /// \ingroup  c_core_module
 /// \brief  The parameters.
 ///
+/// \see  \ref tutorial_core_notation
+///
 typedef struct {
 
   /// The MPI communicator.
@@ -36,10 +38,10 @@ typedef struct {
   /// \f$n\f$ The number of columns of the matrix.
   const isvd_int_t ncol;
 
-  /// \f$m_j\f$ The number of rows of current MPI process.
+  /// \f$m^{(j)}\f$ The number of rows of current MPI process.
   const isvd_int_t nrow_proc;
 
-  /// \f$n_j\f$ The number of columns of current MPI process.
+  /// \f$n^{(j)}\f$ The number of columns of current MPI process.
   const isvd_int_t ncol_proc;
 
   /// \f$m_b = \lceil \frac{m}{P} \rceil\f$ The number of rows per MPI process.
diff --git a/include/c/isvd/def.h b/src/include/c/isvd/def.h
similarity index 50%
rename from include/c/isvd/def.h
rename to src/include/c/isvd/def.h
index f4f9a2c..a346232 100644
--- a/include/c/isvd/def.h
+++ b/src/include/c/isvd/def.h
@@ -9,9 +9,11 @@
 #ifndef ISVD_DEF_H_
 #define ISVD_DEF_H_
 
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 #if !defined(__STDC_FORMAT_MACROS)
   #define __STDC_FORMAT_MACROS
-#endif
+#endif  // __STDC_FORMAT_MACROS
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
 
 #include <assert.h>
 #include <stdio.h>
@@ -26,31 +28,57 @@
 #undef imaginary
 #undef complex
 
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 #if !defined(__cplusplus)
   #define ISVD_UNKNOWN
 #else  // __cplusplus
   #define ISVD_UNKNOWN ...
 #endif  // __cplusplus
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
 
 #if defined(__cplusplus)
 extern "C" {
 #endif  // __cplusplus
 
 /// \ingroup  c_core_module
-/// \brief  The type of index.
+/// \brief  The type of index. (N = 32/64)
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 #if !defined(ISVD_USE_ILP64)
-typedef int32_t  isvd_int_t;
+typedef int32_t isvd_int_t;
 #else  // ISVD_USE_ILP64
-typedef int64_t  isvd_int_t;
+typedef int64_t isvd_int_t;
 #endif  // ISVD_USE_ILP64
+#else  // DOXYGEN_SHOULD_SKIP_THIS)
+typedef intN_t  isvd_int_t;
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
 
 /// \ingroup  c_core_module
-/// \brief  The type of unsigned index.
+/// \brief  The type of unsigned index. (N = 32/64)
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 #if !defined(ISVD_USE_ILP64)
-typedef uint32_t  isvd_uint_t;
+typedef uint32_t isvd_uint_t;
 #else  // ISVD_USE_ILP64
-typedef uint64_t  isvd_uint_t;
+typedef uint64_t isvd_uint_t;
 #endif  // ISVD_USE_ILP64
+#else  // DOXYGEN_SHOULD_SKIP_THIS)
+typedef uintN_t  isvd_uint_t;
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
+
+/// \ingroup  c_core_module
+/// \brief  The type of real single floating point.
+typedef float           isvd_s_val_t;
+
+/// \ingroup  c_core_module
+/// \brief  The type of real double floating point.
+typedef double          isvd_d_val_t;
+
+/// \ingroup  c_core_module
+/// \brief  The type of complex single floating point.
+typedef _Complex float  isvd_c_val_t;
+
+/// \ingroup  c_core_module
+/// \brief  The type of complex double floating point.
+typedef _Complex double isvd_z_val_t;
 
 /// \ingroup  c_core_module
 /// \brief  The type of MPI index.
@@ -62,7 +90,11 @@ typedef int omp_int_t;
 
 /// \ingroup  c_core_module
 /// \brief  The MPI communicator type.
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 typedef int isvd_MpiComm;
+#else  // DOXYGEN_SHOULD_SKIP_THIS)
+typedef MPI_Comm isvd_MpiComm;
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/gpu.h b/src/include/c/isvd/gpu.h
similarity index 100%
rename from include/c/isvd/gpu.h
rename to src/include/c/isvd/gpu.h
diff --git a/include/c/isvd/gpu/@x@_stage.h b/src/include/c/isvd/gpu/@x@_stage.h
similarity index 59%
rename from include/c/isvd/gpu/@x@_stage.h
rename to src/include/c/isvd/gpu/@x@_stage.h
index 5607219..62cefab 100644
--- a/include/c/isvd/gpu/@x@_stage.h
+++ b/src/include/c/isvd/gpu/@x@_stage.h
@@ -18,60 +18,63 @@ extern "C" {
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_gpu_@x@_stage_module
-/// \brief  GPU Gaussian Projection Sketching (@xname@ precision)
+/// \brief  GPU Gaussian Projection Sketching (@xname@ precision).
 ///
 /// \copydetails  isvd_@x@SketchGaussianProjection
 ///
 /// \attention  Set \ref isvd_gpu_memory_limit as the limit of GPU memory usage.
 ///
 void isvd_@x@SketchGaussianProjection_gpu(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda,
-    @xtype@ *yst, const isvd_int_t ldyst, const isvd_int_t seed, const mpi_int_t mpi_root
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera, const @xtype_____@ *a, const isvd_int_t lda,
+    @xtype_____@ *yst, const isvd_int_t ldyst, const isvd_int_t seed, const mpi_int_t mpi_root
 );
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_gpu_@x@_stage_module
-/// \brief  GPU Tall-Skinny QR Postprocessing (@xname@ precision)
+/// \brief  GPU Tall-Skinny QR Postprocessing (@xname@ precision).
 ///
 /// \copydetails  isvd_@x@PostprocessTallSkinnyQr
 ///
 /// \attention  Set \ref isvd_gpu_memory_limit as the limit of GPU memory usage.
 ///
 void isvd_@x@PostprocessTallSkinnyQr_gpu(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda, const @xtype@ *qt, const isvd_int_t ldqt,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera,
+    const @xtype_____@ *a, const isvd_int_t lda, const @xtype_____@ *qt, const isvd_int_t ldqt,
+          @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
     const mpi_int_t ut_root, const mpi_int_t vt_root
 );
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_gpu_@x@_stage_module
-/// \brief  GPU Gramian Postprocessing (@xname@ precision)
+/// \brief  GPU Gramian Postprocessing (@xname@ precision).
 ///
 /// \copydetails  isvd_@x@PostprocessGramian
 ///
 /// \attention  Set \ref isvd_gpu_memory_limit as the limit of GPU memory usage.
 ///
 void isvd_@x@PostprocessGramian_gpu(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda, const @xtype@ *qt, const isvd_int_t ldqt,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera,
+    const @xtype_____@ *a, const isvd_int_t lda, const @xtype_____@ *qt, const isvd_int_t ldqt,
+          @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
     const mpi_int_t ut_root, const mpi_int_t vt_root
 );
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_gpu_@x@_stage_module
-/// \brief  GPU Symmetric Postprocessing (@xname@ precision)
+/// \brief  GPU Symmetric Postprocessing (@xname@ precision).
 ///
 /// \copydetails  isvd_@x@PostprocessSymmetric
 ///
 /// \attention  Set \ref isvd_gpu_memory_limit as the limit of GPU memory usage.
 ///
 void isvd_@x@PostprocessSymmetric_gpu(
-    const isvd_Param param, const @xtype@ *argv, const isvd_int_t argc, @xtype@ *retv, const isvd_int_t retc,
-    const char dista, const char ordera, const @xtype@ *a, const isvd_int_t lda, const @xtype@ *qt, const isvd_int_t ldqt,
-    @xtype@ *s, @xtype@ *ut, const isvd_int_t ldut, @xtype@ *vt, const isvd_int_t ldvt,
+    const isvd_Param param, const @xtype_____@ *argv, const isvd_int_t argc, @xtype_____@ *retv, const isvd_int_t retc,
+    const char dista, const char ordera,
+    const @xtype_____@ *a, const isvd_int_t lda, const @xtype_____@ *qt, const isvd_int_t ldqt,
+          @xtype_____@ *s, @xtype_____@ *ut, const isvd_int_t ldut, @xtype_____@ *vt, const isvd_int_t ldvt,
     const mpi_int_t ut_root, const mpi_int_t vt_root
 );
 
diff --git a/include/c/isvd/gpu/env.h b/src/include/c/isvd/gpu/env.h
similarity index 100%
rename from include/c/isvd/gpu/env.h
rename to src/include/c/isvd/gpu/env.h
diff --git a/include/c/isvd/la.h b/src/include/c/isvd/la.h
similarity index 100%
rename from include/c/isvd/la.h
rename to src/include/c/isvd/la.h
diff --git a/include/c/isvd/la/blas.h b/src/include/c/isvd/la/blas.h
similarity index 100%
rename from include/c/isvd/la/blas.h
rename to src/include/c/isvd/la/blas.h
diff --git a/include/c/isvd/la/blas/blas1.h b/src/include/c/isvd/la/blas/blas1.h
similarity index 99%
rename from include/c/isvd/la/blas/blas1.h
rename to src/include/c/isvd/la/blas/blas1.h
index f0e6b67..ec24ebf 100644
--- a/include/c/isvd/la/blas/blas1.h
+++ b/src/include/c/isvd/la/blas/blas1.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -169,7 +169,7 @@ static inline void isvd_zdScal(
 ) { zdscal_(&n, &alpha, x, &incx); }
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/blas/blas2.h b/src/include/c/isvd/la/blas/blas2.h
similarity index 91%
rename from include/c/isvd/la/blas/blas2.h
rename to src/include/c/isvd/la/blas/blas2.h
index ae4d7e2..96ffe77 100644
--- a/include/c/isvd/la/blas/blas2.h
+++ b/src/include/c/isvd/la/blas/blas2.h
@@ -15,13 +15,13 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
 #endif  // DOXYGEN_SHOULD_SKIP_THIS
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/blas/blas3.h b/src/include/c/isvd/la/blas/blas3.h
similarity index 99%
rename from include/c/isvd/la/blas/blas3.h
rename to src/include/c/isvd/la/blas/blas3.h
index 0769a96..9c89688 100644
--- a/include/c/isvd/la/blas/blas3.h
+++ b/src/include/c/isvd/la/blas/blas3.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -102,7 +102,7 @@ static inline void isvd_zSyrk(
 ) { zherk_(&uplo, &trans, &n, &k, &alpha, a, &lda, &beta, c, &ldc); }
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/blas/blas_like.h b/src/include/c/isvd/la/blas/blas_like.h
similarity index 98%
rename from include/c/isvd/la/blas/blas_like.h
rename to src/include/c/isvd/la/blas/blas_like.h
index 39f25c6..978172c 100644
--- a/include/c/isvd/la/blas/blas_like.h
+++ b/src/include/c/isvd/la/blas/blas_like.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_la_blas_like_module
@@ -96,7 +96,7 @@ void isvd_zGemmt(
 );
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/lapack.h b/src/include/c/isvd/la/lapack.h
similarity index 100%
rename from include/c/isvd/la/lapack.h
rename to src/include/c/isvd/la/lapack.h
diff --git a/include/c/isvd/la/lapack/auxiliary.h b/src/include/c/isvd/la/lapack/auxiliary.h
similarity index 97%
rename from include/c/isvd/la/lapack/auxiliary.h
rename to src/include/c/isvd/la/lapack/auxiliary.h
index 66f04d9..8bd65e1 100644
--- a/include/c/isvd/la/lapack/auxiliary.h
+++ b/src/include/c/isvd/la/lapack/auxiliary.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -54,7 +54,7 @@ bool isvd_Lsame( const CHAR1 ca, const CHAR1 cb );
 /// \brief  Tests two character strings for equality regardless of the case.
 bool isvd_Lsamen( const INT n, const CHAR1 *sa, const CHAR1 *sb );
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/lapack/least_square.h b/src/include/c/isvd/la/lapack/least_square.h
similarity index 97%
rename from include/c/isvd/la/lapack/least_square.h
rename to src/include/c/isvd/la/lapack/least_square.h
index 15f17d6..98dd62e 100644
--- a/include/c/isvd/la/lapack/least_square.h
+++ b/src/include/c/isvd/la/lapack/least_square.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_la_lapack_ls_module
@@ -41,7 +41,7 @@ void isvd_zGesvd( const CHAR1 jobu, const CHAR1 jobvt, const INT m, const INT n,
                   const INT ldu, COMP8 *v, const INT ldvt );
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/lapack/linear_equation.h b/src/include/c/isvd/la/lapack/linear_equation.h
similarity index 95%
rename from include/c/isvd/la/lapack/linear_equation.h
rename to src/include/c/isvd/la/lapack/linear_equation.h
index a339cd7..83fdd1f 100644
--- a/include/c/isvd/la/lapack/linear_equation.h
+++ b/src/include/c/isvd/la/lapack/linear_equation.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_la_lapack_le_module
@@ -27,7 +27,7 @@ void isvd_cGeinv( const INT n, COMP4 *a, const INT lda );
 void isvd_zGeinv( const INT n, COMP8 *a, const INT lda );
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/vml.h b/src/include/c/isvd/la/vml.h
similarity index 100%
rename from include/c/isvd/la/vml.h
rename to src/include/c/isvd/la/vml.h
diff --git a/include/c/isvd/la/vml/mathematical.h b/src/include/c/isvd/la/vml/mathematical.h
similarity index 97%
rename from include/c/isvd/la/vml/mathematical.h
rename to src/include/c/isvd/la/vml/mathematical.h
index ca43144..1401138 100644
--- a/include/c/isvd/la/vml/mathematical.h
+++ b/src/include/c/isvd/la/vml/mathematical.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_la_vml_math_module
@@ -47,7 +47,7 @@ void isvd_vcSqrt( const INT n, const COMP4 *a, COMP4 *y );
 void isvd_vzSqrt( const INT n, const COMP8 *a, COMP8 *y );
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/vml/power_root.h b/src/include/c/isvd/la/vml/power_root.h
similarity index 95%
rename from include/c/isvd/la/vml/power_root.h
rename to src/include/c/isvd/la/vml/power_root.h
index d41dab7..e3dca13 100644
--- a/include/c/isvd/la/vml/power_root.h
+++ b/src/include/c/isvd/la/vml/power_root.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_la_vml_math_module
@@ -27,7 +27,7 @@ void isvd_vcDiv( const INT n, const COMP4 *a, const COMP4 *b, COMP4 *y );
 void isvd_vzDiv( const INT n, const COMP8 *a, const COMP8 *b, COMP8 *y );
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/la/vsl.h b/src/include/c/isvd/la/vsl.h
similarity index 77%
rename from include/c/isvd/la/vsl.h
rename to src/include/c/isvd/la/vsl.h
index 80d28f3..f4e5d60 100644
--- a/include/c/isvd/la/vsl.h
+++ b/src/include/c/isvd/la/vsl.h
@@ -11,6 +11,7 @@
 
 #include <isvd/la/vsl/service.h>
 #include <isvd/la/vsl/distribution.h>
+#include <isvd/la/vsl/driver.h>
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \defgroup  c_la_vsl_service_module  Service Routines
@@ -24,4 +25,10 @@
 /// \brief     The VSL Distribution Generators
 ///
 
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \defgroup  c_la_vsl_driver_module  Drivers
+/// \ingroup   c_la_vsl_module
+/// \brief     The VSL Drivers
+///
+
 #endif  // ISVD_LA_VSL_H_
diff --git a/include/c/isvd/la/vsl/distribution.h b/src/include/c/isvd/la/vsl/distribution.h
similarity index 95%
rename from include/c/isvd/la/vsl/distribution.h
rename to src/include/c/isvd/la/vsl/distribution.h
index f405a80..24ee155 100644
--- a/include/c/isvd/la/vsl/distribution.h
+++ b/src/include/c/isvd/la/vsl/distribution.h
@@ -16,7 +16,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_la_vsl_distribution_module
@@ -26,7 +26,7 @@ void isvd_vsRngGaussian( isvd_VSLStreamStatePtr stream, const INT n, REAL4 *r, c
 void isvd_vdRngGaussian( isvd_VSLStreamStatePtr stream, const INT n, REAL8 *r, const REAL8 a, const REAL8 sigma );
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/src/include/c/isvd/la/vsl/driver.h b/src/include/c/isvd/la/vsl/driver.h
new file mode 100644
index 0000000..8d05a6a
--- /dev/null
+++ b/src/include/c/isvd/la/vsl/driver.h
@@ -0,0 +1,35 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       include/c/isvd/la/vsl/driver.h
+/// \brief      The VSL driver header.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#ifndef ISVD_LA_VSL_DRIVER_H_
+#define ISVD_LA_VSL_DRIVER_H_
+
+#include <isvd/def.h>
+#include <isvd/la/vsl/service.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif  // __cplusplus
+
+@ISVD_TYPE_MACRO_DEFINE@
+
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \ingroup  c_la_vsl_driver_module
+/// \brief  Generates normally distributed random numbers in parallel.
+//\{
+void isvd_vsRngGaussianDriver( const INT seed, const INT nskip, const INT n, REAL4 *r, const REAL4 a, const REAL4 sigma );
+void isvd_vdRngGaussianDriver( const INT seed, const INT nskip, const INT n, REAL8 *r, const REAL8 a, const REAL8 sigma );
+//\}
+
+@ISVD_TYPE_MACRO_UNDEF@
+
+#if defined(__cplusplus)
+}
+#endif  // __cplusplus
+
+#endif  // ISVD_LA_VSL_DRIVER_H_
diff --git a/include/c/isvd/la/vsl/service.h b/src/include/c/isvd/la/vsl/service.h
similarity index 96%
rename from include/c/isvd/la/vsl/service.h
rename to src/include/c/isvd/la/vsl/service.h
index 05139de..134c34d 100644
--- a/include/c/isvd/la/vsl/service.h
+++ b/src/include/c/isvd/la/vsl/service.h
@@ -15,7 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 typedef INT* isvd_VSLStreamStatePtr;
 
@@ -34,7 +34,7 @@ void isvd_vslDeleteStream( isvd_VSLStreamStatePtr *streamp );
 /// \brief  Initializes the stream by the skip-ahead method.
 void isvd_vslSkipAheadStream( isvd_VSLStreamStatePtr stream, const INT nskip );
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/include/c/isvd/util.h b/src/include/c/isvd/util.h
similarity index 100%
rename from include/c/isvd/util.h
rename to src/include/c/isvd/util.h
diff --git a/include/c/isvd/util/io.h b/src/include/c/isvd/util/io.h
similarity index 72%
rename from include/c/isvd/util/io.h
rename to src/include/c/isvd/util/io.h
index bb270ac..892ebc7 100644
--- a/include/c/isvd/util/io.h
+++ b/src/include/c/isvd/util/io.h
@@ -15,9 +15,13 @@
 extern "C" {
 #endif  // __cplusplus
 
-void isvd_ifget( FILE *stream, isvd_int_t *varp );
-void isvd_sfget( FILE *stream, float *varp );
-void isvd_dfget( FILE *stream, double *varp );
+@ISVD_TYPE_MACRO_DEFINE@
+
+void isvd_ifget( FILE *stream, INT   *varp );
+void isvd_sfget( FILE *stream, REAL4 *varp );
+void isvd_dfget( FILE *stream, REAL8 *varp );
+
+@ISVD_TYPE_MACRO_UNDEF@
 
 #if defined(__cplusplus)
 }
diff --git a/src/include/c/isvd/util/memory.h b/src/include/c/isvd/util/memory.h
new file mode 100644
index 0000000..cd1fa37
--- /dev/null
+++ b/src/include/c/isvd/util/memory.h
@@ -0,0 +1,46 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       include/c/isvd/util/memory.h
+/// \brief      The memory utilities.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#ifndef ISVD_UTIL_MEMORY_H_
+#define ISVD_UTIL_MEMORY_H_
+
+#include <isvd/def.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif  // __cplusplus
+
+@ISVD_TYPE_MACRO_DEFINE@
+
+INT*   isvd_imalloc( const size_t num );
+REAL4* isvd_smalloc( const size_t num );
+REAL8* isvd_dmalloc( const size_t num );
+COMP4* isvd_cmalloc( const size_t num );
+COMP8* isvd_zmalloc( const size_t num );
+
+void isvd_free( void *ptr );
+
+void isvd_imemset0( INT   *ptr, const size_t num );
+void isvd_smemset0( REAL4 *ptr, const size_t num );
+void isvd_dmemset0( REAL8 *ptr, const size_t num );
+void isvd_cmemset0( COMP4 *ptr, const size_t num );
+void isvd_zmemset0( COMP8 *ptr, const size_t num );
+
+void isvd_imemcpy( INT   *dst, const INT   *src, const size_t num );
+void isvd_smemcpy( REAL4 *dst, const REAL4 *src, const size_t num );
+void isvd_dmemcpy( REAL8 *dst, const REAL8 *src, const size_t num );
+void isvd_cmemcpy( COMP4 *dst, const COMP4 *src, const size_t num );
+void isvd_zmemcpy( COMP8 *dst, const COMP8 *src, const size_t num );
+
+@ISVD_TYPE_MACRO_UNDEF@
+
+#if defined(__cplusplus)
+}
+#endif  // __cplusplus
+
+#endif  // ISVD_UTIL_MEMORY_H_
diff --git a/include/c/isvd/util/mpi.h b/src/include/c/isvd/util/mpi.h
similarity index 61%
rename from include/c/isvd/util/mpi.h
rename to src/include/c/isvd/util/mpi.h
index 8655673..54da529 100644
--- a/include/c/isvd/util/mpi.h
+++ b/src/include/c/isvd/util/mpi.h
@@ -15,6 +15,18 @@
 extern "C" {
 #endif  // __cplusplus
 
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
+#if !defined(ISVD_USE_ILP64)
+#define MPI_ISVD_INT_T    MPI_INTEGER4
+#else  // ISVD_USE_ILP64
+#define MPI_ISVD_INT_T    MPI_INTEGER8
+#endif  // ISVD_USE_ILP64
+#define MPI_ISVD_S_VAL_T  MPI_REAL4
+#define MPI_ISVD_D_VAL_T  MPI_REAL8
+#define MPI_ISVD_C_VAL_T  MPI_COMPLEX8
+#define MPI_ISVD_Z_VAL_T  MPI_COMPLEX16
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
+
 mpi_int_t isvd_getMpiSize( const isvd_MpiComm comm );
 mpi_int_t isvd_getMpiRank( const isvd_MpiComm comm );
 
diff --git a/include/c/isvd/util/omp.h b/src/include/c/isvd/util/omp.h
similarity index 93%
rename from include/c/isvd/util/omp.h
rename to src/include/c/isvd/util/omp.h
index ebbea8c..2bf865b 100644
--- a/include/c/isvd/util/omp.h
+++ b/src/include/c/isvd/util/omp.h
@@ -15,6 +15,7 @@
 extern "C" {
 #endif  // __cplusplus
 
+omp_int_t isvd_getOmpMaxSize( void );
 omp_int_t isvd_getOmpSize( void );
 omp_int_t isvd_getOmpRank( void );
 
diff --git a/src/lib/CMakeLists.txt b/src/lib/CMakeLists.txt
new file mode 100644
index 0000000..00643ce
--- /dev/null
+++ b/src/lib/CMakeLists.txt
@@ -0,0 +1,73 @@
+# The CMake setting of 'src/lib'
+
+isvd_set_config_var()
+
+# Shared library
+SET(BUILD_SHARED_LIBS ON)
+SET(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/${LIB_FOLDER}")
+set(CMAKE_INSTALL_RPATH_USE_LINK_PATH ON)
+
+# Macro
+macro(SET_TARGET target files)
+  # Set library
+  add_library(${target} SHARED ${files})
+  isvd_set_target(${target})
+  set_property(TARGET ${target} PROPERTY VERSION ${ISVD_VERSION})
+  set_property(TARGET ${target} PROPERTY LIBRARY_OUTPUT_DIRECTORY "${CMAKE_CURRENT_CONFIG_DIR}")
+  install(TARGETS ${target} LIBRARY DESTINATION ${LIB_FOLDER})
+endmacro()
+
+# Set include paths
+include_directories("${CMAKE_CURRENT_CONFIG_DIR}/../include/c"
+                    "${CMAKE_CURRENT_CONFIG_DIR}")
+
+# Break
+if(NOT ISVD_BUILD_LIB)
+  return()
+endif()
+
+# libisvd
+set_target(isvd /dev/null)
+set_property(TARGET isvd PROPERTY LINKER_LANGUAGE C)
+
+# libisvd_core
+file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/core/*"
+                        "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/util/*")
+set_target(isvd_core "${files}")
+isvd_set_target_mpi(isvd_core C)
+target_link_libraries(isvd isvd_core)
+
+# libisvd_la
+if(ISVD_OMP)
+  string(TOLOWER ${ISVD_OMP} isvd_omp)
+  set(isvd_omp "_${isvd_omp}")
+else()
+  set(isvd_omp "")
+endif()
+if(ISVD_USE_MKL)
+  set(isvd_la isvd_la_mkl${isvd_omp})
+else()
+  set(isvd_la isvd_la_blas${isvd_omp})
+endif(ISVD_USE_MKL)
+file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/la/*")
+set_target(${isvd_la} "${files}")
+isvd_set_target_omp(${isvd_la} C)
+isvd_set_target_mpi(${isvd_la} C)
+isvd_set_target_blas(${isvd_la})
+target_link_libraries(isvd ${isvd_la})
+
+# libisvd_gpu_none
+file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/nogpu/*")
+set_target(isvd_gpu_none "${files}")
+isvd_set_target_mpi(isvd_gpu_none C)
+
+# libisvd_gpu
+if(ISVD_USE_GPU)
+  file(GLOB_RECURSE files "${CMAKE_CURRENT_CONFIG_DIR}/libisvd/gpu/*")
+  set_target(isvd_gpu_magma "${files}")
+  isvd_set_target_mpi(isvd_gpu_magma C)
+  isvd_set_target_gpu(isvd_gpu_magma)
+  target_link_libraries(isvd isvd_gpu_magma)
+else()
+  target_link_libraries(isvd isvd_gpu_none)
+endif(ISVD_USE_GPU)
diff --git a/src/libisvd.h b/src/lib/libisvd.h
similarity index 88%
rename from src/libisvd.h
rename to src/lib/libisvd.h
index 1353227..49a6c7c 100644
--- a/src/libisvd.h
+++ b/src/lib/libisvd.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd.h
+/// \file       lib/libisvd.h
 /// \brief      The main library header.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -9,8 +9,8 @@
 #ifndef LIBISVD_H_
 #define LIBISVD_H_
 
-#include <libisvd/def.h>
 #include <libisvd/core.h>
+#include <libisvd/la.h>
 #include <libisvd/util.h>
 
 #if defined(ISVD_USE_GPU)
diff --git a/src/libisvd/core.h b/src/lib/libisvd/core.h
similarity index 91%
rename from src/libisvd/core.h
rename to src/lib/libisvd/core.h
index 3dceb78..6342534 100644
--- a/src/libisvd/core.h
+++ b/src/lib/libisvd/core.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core.h
+/// \file       lib/libisvd/core.h
 /// \brief      The core header.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/core/@x@_arg.h b/src/lib/libisvd/core/@x@_arg.h
similarity index 95%
rename from src/libisvd/core/@x@_arg.h
rename to src/lib/libisvd/core/@x@_arg.h
index 0124a68..53a143c 100644
--- a/src/libisvd/core/@x@_arg.h
+++ b/src/lib/libisvd/core/@x@_arg.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/@x@_arg.h
+/// \file       lib/libisvd/core/@x@_arg.h
 /// \brief      The @xname@ precision drivers argument utilities.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -24,14 +24,14 @@ static inline isvd_fun_t isvd_arg2@x@AlgS( const char *arg ) {
   if ( !strcmp(arg, "GP")     ) return (isvd_fun_t) isvd_@x@SketchGaussianProjection;
   if ( !strcmp(arg, "GP_gpu") ) return (isvd_fun_t) isvd_@x@SketchGaussianProjection_gpu;
   fprintf(stderr, "Unknown sketching abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline isvd_fun_t isvd_arg2@x@AlgO( const char *arg ) {
   if ( !strcmp(arg, "TS")     ) return (isvd_fun_t) isvd_@x@OrthogonalizeTallSkinnyQr;
   if ( !strcmp(arg, "GR")     ) return (isvd_fun_t) isvd_@x@OrthogonalizeGramian;
   fprintf(stderr, "Unknown orthogonalization abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline isvd_fun_t isvd_arg2@x@AlgI( const char *arg ) {
@@ -39,7 +39,7 @@ static inline isvd_fun_t isvd_arg2@x@AlgI( const char *arg ) {
   if ( !strcmp(arg, "WY")     ) return (isvd_fun_t) isvd_@x@IntegrateWenYin;
   if ( !strcmp(arg, "HR")     ) return (isvd_fun_t) isvd_@x@IntegrateHierarchicalReduction;
   fprintf(stderr, "Unknown integration abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline isvd_fun_t isvd_arg2@x@AlgP( const char *arg ) {
@@ -50,21 +50,21 @@ static inline isvd_fun_t isvd_arg2@x@AlgP( const char *arg ) {
   if ( !strcmp(arg, "GR_gpu") ) return (isvd_fun_t) isvd_@x@PostprocessGramian_gpu;
   if ( !strcmp(arg, "SY_gpu") ) return (isvd_fun_t) isvd_@x@PostprocessSymmetric_gpu;
   fprintf(stderr, "Unknown postprocess abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline const char* isvd_arg2@x@AlgNameS( const char *arg ) {
   if ( !strcmp(arg, "GP")     ) return "@XName@ Precision Gaussian Projection Sketching";
   if ( !strcmp(arg, "GP_gpu") ) return "@XName@ Precision GPU Gaussian Projection Sketching";
   fprintf(stderr, "Unknown sketching abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline const char* isvd_arg2@x@AlgNameO( const char *arg ) {
   if ( !strcmp(arg, "TS")     ) return "@XName@ Precision Tall-Skinny QR Orthogonalization";
   if ( !strcmp(arg, "GR")     ) return "@XName@ Precision Gramian Orthogonalization";
   fprintf(stderr, "Unknown orthogonalization abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline const char* isvd_arg2@x@AlgNameI( const char *arg ) {
@@ -72,7 +72,7 @@ static inline const char* isvd_arg2@x@AlgNameI( const char *arg ) {
   if ( !strcmp(arg, "WY")     ) return "@XName@ Precision Wen-Yin Integration";
   if ( !strcmp(arg, "HR")     ) return "@XName@ Precision Hierarchical Reduction Integration";
   fprintf(stderr, "Unknown integration abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 static inline const char* isvd_arg2@x@AlgNameP( const char *arg ) {
@@ -83,7 +83,7 @@ static inline const char* isvd_arg2@x@AlgNameP( const char *arg ) {
   if ( !strcmp(arg, "GR_gpu") ) return "@XName@ Precision GPU Gramian Postprocessing";
   if ( !strcmp(arg, "SY_gpu") ) return "@XName@ Precision GPU Symmetric Postprocessing";
   fprintf(stderr, "Unknown postprocess abbreviation \"%s\"!\n", arg);
-  return nullptr;
+  return NULL;
 }
 
 #if defined(__cplusplus)
diff --git a/src/libisvd/core/driver/@x@_isvd.c b/src/lib/libisvd/core/driver/@x@_isvd.c
similarity index 65%
rename from src/libisvd/core/driver/@x@_isvd.c
rename to src/lib/libisvd/core/driver/@x@_isvd.c
index 489da9d..e1174f3 100644
--- a/src/libisvd/core/driver/@x@_isvd.c
+++ b/src/lib/libisvd/core/driver/@x@_isvd.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/driver/@x@_isvd.c
+/// \file       lib/libisvd/core/driver/@x@_isvd.c
 /// \brief      The iSVD driver (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -54,9 +54,9 @@
 /// \param[in]   ordera        The storage ordering of 𝑨. <br>
 ///                            `'C'`: column-major ordering. <br>
 ///                            `'R'`: row-major ordering.
-/// \param[in]   a, lda        The column/row-block 𝑨 (\f$m \times n_j\f$) and its leading dimension. <br>
-///                            \b dista = `'C'`: the size must be \f$m \times n_j\f$. <br>
-///                            \b dista = `'R'`: the size must be \f$m_j \times n\f$.
+/// \param[in]   a, lda        The column/row-block 𝑨 (\f$m \times n^{(j)}\f$) and its leading dimension. <br>
+///                            \b dista = `'C'`: the size must be \f$m \times n^{(j)}\f$. <br>
+///                            \b dista = `'R'`: the size must be \f$m^{(j)} \times n\f$.
 /// \param[in]   s             The vector 𝝈 (\f$k \times 1\f$).
 /// \param[in]   ut, ldut      The matrix 𝑼 (row-major) and its leading dimension. <br>
 ///                            \b ut_root ≥  0: the size must be \f$Pm_b \times k\f$, and \b ldut must be \f$l\f$. <br>
@@ -94,36 +94,41 @@
 /// \param[out]  ut            Replaced by the left singular vectors 𝑼 (row-major).
 /// \param[out]  vt            Replaced by the right singular vectors 𝑽 (row-major).
 ///
+/// \note  If \b argc ≠ `NULL` and \b argc[i] < 0, then a default argument query is assumed on the i-th stage;
+///        the routine only returns the first \b retc[i] default arguments in \b retv[i].
+///
+/// \see isvd_Param
+///
 void isvd_@x@Isvd(
-    const char       *alg_s,
-    const char       *alg_o,
-    const char       *alg_i,
-    const char       *alg_p,
-    const isvd_int_t  m,
-    const isvd_int_t  n,
-    const isvd_int_t  k,
-    const isvd_int_t  p,
-    const isvd_int_t  N,
-    const @xtype@    *argv[4],
-    const isvd_int_t  argc[4],
-          @xtype@    *retv[4],
-    const isvd_int_t  retc[4],
-          double      time[4],
-          FILE       *stream,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const isvd_int_t  seed,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root,
-    const mpi_int_t   mpi_root,
-    const isvd_MpiComm    mpi_comm
+    const char         *alg_s,
+    const char         *alg_o,
+    const char         *alg_i,
+    const char         *alg_p,
+    const isvd_int_t    m,
+    const isvd_int_t    n,
+    const isvd_int_t    k,
+    const isvd_int_t    p,
+    const isvd_int_t    N,
+    const @xtype_____@ *argv[4],
+    const isvd_int_t    argc[4],
+          @xtype_____@ *retv[4],
+    const isvd_int_t    retc[4],
+          double        time[4],
+          FILE        *stream,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const isvd_int_t    seed,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root,
+    const mpi_int_t     mpi_root,
+    const isvd_MpiComm  mpi_comm
 ) {
 
   const mpi_int_t mpi_rank = isvd_getMpiRank(MPI_COMM_WORLD);
@@ -136,10 +141,10 @@ void isvd_@x@Isvd(
   const char *opts_i[] = {"KN", "WY", "HR"};
   const char *opts_p[] = {"TS", "GR", "SY", "TS_gpu", "GR_gpu", "SY_gpu"};
 
-  const char *alg_s_ = isvd_arg2str("ALG_S", alg_s, opts_s, nullptr, lenof(opts_s));
-  const char *alg_o_ = isvd_arg2str("ALG_O", alg_o, opts_o, nullptr, lenof(opts_o));
-  const char *alg_i_ = isvd_arg2str("ALG_I", alg_i, opts_i, nullptr, lenof(opts_i));
-  const char *alg_p_ = isvd_arg2str("ALG_P", alg_p, opts_p, nullptr, lenof(opts_p));
+  const char *alg_s_ = isvd_arg2str("ALG_S", alg_s, opts_s, NULL, lenof(opts_s));
+  const char *alg_o_ = isvd_arg2str("ALG_O", alg_o, opts_o, NULL, lenof(opts_o));
+  const char *alg_i_ = isvd_arg2str("ALG_I", alg_i, opts_i, NULL, lenof(opts_i));
+  const char *alg_p_ = isvd_arg2str("ALG_P", alg_p, opts_p, NULL, lenof(opts_p));
   if ( !alg_s_ || !alg_o_ || !alg_i_ || !alg_p_ ) return;
 
   isvd_fun_t fun_s = isvd_arg2@x@AlgS(alg_s_);
@@ -147,14 +152,6 @@ void isvd_@x@Isvd(
   isvd_fun_t fun_i = isvd_arg2@x@AlgI(alg_i_);
   isvd_fun_t fun_p = isvd_arg2@x@AlgP(alg_p_);
 
-  if ( stream != nullptr && mpi_rank == mpi_root ) {
-    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameS(alg_s_));
-    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameO(alg_o_));
-    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameI(alg_i_));
-    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameP(alg_p_));
-    fprintf(stream, "\n");
-  }
-
   // ====================================================================================================================== //
   // Create parameters
 
@@ -167,65 +164,112 @@ void isvd_@x@Isvd(
   // ====================================================================================================================== //
   // Gets arguments and return values of stages
 
-  const @xtype@ *argv_s = (argv != nullptr) ? argv[0] : nullptr;
-  const @xtype@ *argv_o = (argv != nullptr) ? argv[1] : nullptr;
-  const @xtype@ *argv_i = (argv != nullptr) ? argv[2] : nullptr;
-  const @xtype@ *argv_p = (argv != nullptr) ? argv[3] : nullptr;
+  const @xtype_____@ *argv_s = (argv != NULL) ? argv[0] : NULL;
+  const @xtype_____@ *argv_o = (argv != NULL) ? argv[1] : NULL;
+  const @xtype_____@ *argv_i = (argv != NULL) ? argv[2] : NULL;
+  const @xtype_____@ *argv_p = (argv != NULL) ? argv[3] : NULL;
+
+  const isvd_int_t    argc_s = (argv != NULL) ? argc[0] : 0;
+  const isvd_int_t    argc_o = (argv != NULL) ? argc[1] : 0;
+  const isvd_int_t    argc_i = (argv != NULL) ? argc[2] : 0;
+  const isvd_int_t    argc_p = (argv != NULL) ? argc[3] : 0;
 
-  const isvd_int_t argc_s = (argv != nullptr) ? argc[0] : 0;
-  const isvd_int_t argc_o = (argv != nullptr) ? argc[1] : 0;
-  const isvd_int_t argc_i = (argv != nullptr) ? argc[2] : 0;
-  const isvd_int_t argc_p = (argv != nullptr) ? argc[3] : 0;
+        @xtype_____@ *retv_s = (retv != NULL) ? retv[0] : NULL;
+        @xtype_____@ *retv_o = (retv != NULL) ? retv[1] : NULL;
+        @xtype_____@ *retv_i = (retv != NULL) ? retv[2] : NULL;
+        @xtype_____@ *retv_p = (retv != NULL) ? retv[3] : NULL;
 
-  @xtype@ *retv_s = (retv != nullptr) ? retv[0] : nullptr;
-  @xtype@ *retv_o = (retv != nullptr) ? retv[1] : nullptr;
-  @xtype@ *retv_i = (retv != nullptr) ? retv[2] : nullptr;
-  @xtype@ *retv_p = (retv != nullptr) ? retv[3] : nullptr;
+  const isvd_int_t    retc_s = (retv != NULL) ? retc[0] : 0;
+  const isvd_int_t    retc_o = (retv != NULL) ? retc[1] : 0;
+  const isvd_int_t    retc_i = (retv != NULL) ? retc[2] : 0;
+  const isvd_int_t    retc_p = (retv != NULL) ? retc[3] : 0;
+
+  // ====================================================================================================================== //
+  // Check stage arguments
+
+  if ( argc_s > 0 ) { isvd_assert_ne(argv_s, nullptr); }
+  if ( argc_o > 0 ) { isvd_assert_ne(argv_o, nullptr); }
+  if ( argc_i > 0 ) { isvd_assert_ne(argv_i, nullptr); }
+  if ( argc_p > 0 ) { isvd_assert_ne(argv_p, nullptr); }
+  if ( retc_s > 0 ) { isvd_assert_ne(retv_s, nullptr); }
+  if ( retc_o > 0 ) { isvd_assert_ne(retv_o, nullptr); }
+  if ( retc_i > 0 ) { isvd_assert_ne(retv_i, nullptr); }
+  if ( retc_p > 0 ) { isvd_assert_ne(retv_p, nullptr); }
+
+  // ====================================================================================================================== //
+  // Query stage arguments
 
-  const isvd_int_t retc_s = (retv != nullptr) ? retc[0] : 0;
-  const isvd_int_t retc_o = (retv != nullptr) ? retc[1] : 0;
-  const isvd_int_t retc_i = (retv != nullptr) ? retc[2] : 0;
-  const isvd_int_t retc_p = (retv != nullptr) ? retc[3] : 0;
+  bool query = false;
+  if ( argc_s < 0 ) {
+    fun_s(param, argv_s, argc_s, retv_s, retc_s, dista, ordera, NULL, 1, NULL, 1, seed, mpi_root);
+    query = true;
+  }
+  if ( argc_o < 0 ) {
+    fun_o(param, argv_o, argc_o, retv_o, retc_o, NULL, 1);
+    query = true;
+  }
+  if ( argc_i < 0 ) {
+    fun_i(param, argv_i, argc_i, retv_i, retc_i, NULL, 1, NULL, 1);
+    query = true;
+  }
+  if ( argc_p < 0 ) {
+    fun_p(param, argv_p, argc_p, retv_p, retc_p, dista, ordera, NULL, 1, NULL, 1, NULL, NULL, 1, NULL, 1, ut_root, vt_root);
+    query = true;
+  }
+  if ( query ) {
+    return;
+  }
+
+  // ====================================================================================================================== //
+  // Print arguments
+
+  if ( stream != NULL && mpi_rank == mpi_root ) {
+    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameS(alg_s_));
+    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameO(alg_o_));
+    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameI(alg_i_));
+    fprintf(stream, "Using %s\n", isvd_arg2@x@AlgNameP(alg_p_));
+    fprintf(stream, "\n");
+  }
 
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *yst = isvd_@x@malloc(Nl * mb);
+  @xtype_____@ *yst = isvd_@x@malloc(Nl * mb);
   isvd_int_t ldyst = Nl;
 
-  @xtype@ *qt = isvd_@x@malloc(l * mb);
+  @xtype_____@ *qt = isvd_@x@malloc(l * mb);
   isvd_int_t ldqt = l;
 
   // ====================================================================================================================== //
   // Run
 
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "Sketching ...................... "); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "Sketching ...................... "); fflush(stream); }
   double time_s = MPI_Wtime();
   fun_s(param, argv_s, argc_s, retv_s, retc_s, dista, ordera, a, lda, yst, ldyst, seed, mpi_root);
   time_s = MPI_Wtime() - time_s;
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
 
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "Orthogonalizing ................ "); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "Orthogonalizing ................ "); fflush(stream); }
   double time_i = MPI_Wtime();
   fun_o(param, argv_o, argc_o, retv_o, retc_o, yst, ldyst);
   time_i = MPI_Wtime() - time_i;
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
 
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "Integrating .................... "); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "Integrating .................... "); fflush(stream); }
   double time_o = MPI_Wtime();
   fun_i(param, argv_i, argc_i, retv_i, retc_i, yst, ldyst, qt, ldqt);
   time_o = MPI_Wtime() - time_o;
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
 
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "Postprocessing ................. "); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "Postprocessing ................. "); fflush(stream); }
   double time_p = MPI_Wtime();
   fun_p(param, argv_p, argc_p, retv_p, retc_p, dista, ordera, a, lda, qt, ldqt, s, ut, ldut, vt, ldvt, ut_root, vt_root);
   time_p = MPI_Wtime() - time_p;
-  if ( stream != nullptr && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
+  if ( stream != NULL && mpi_rank == mpi_root ) { fprintf(stream, "done\n"); fflush(stream); }
 
   // ====================================================================================================================== //
   // Gets executing times
-  if ( time != nullptr ) {
+  if ( time != NULL ) {
     time[0] = time_s;
     time[1] = time_i;
     time[2] = time_o;
diff --git a/src/libisvd/core/env.c b/src/lib/libisvd/core/env.c
similarity index 86%
rename from src/libisvd/core/env.c
rename to src/lib/libisvd/core/env.c
index fe1f5d4..40101f2 100644
--- a/src/libisvd/core/env.c
+++ b/src/lib/libisvd/core/env.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/env.c
+/// \file       lib/libisvd/core/env.c
 /// \brief      The iSVD environment routines (CPU only).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -39,12 +39,9 @@ void isvd_finalize_cpu( void ) {
 /// \note  This routines displays the MPI environment.
 ///
 void isvd_printEnvironment_cpu( const isvd_MpiComm mpi_comm ) {
-  mpi_int_t mpi_size = isvd_getMpiSize(mpi_comm), omp_size;
-  ISVD_OMP_PARALLEL
-  {
-    omp_size = isvd_getOmpSize();
-  }
+  mpi_int_t mpi_size = isvd_getMpiSize(mpi_comm);
+  omp_int_t omp_size = isvd_getOmpMaxSize();
 
   printf("iSVD %s, %lu-bit isvd_int_t, %lu-bit pointer\n", ISVD_VERSION, sizeof(isvd_int_t) * 8, sizeof(void*) * 8);
-  printf("%d MPI nodes, %d OpenMP threads per node\n\n", mpi_size, omp_size);
+  printf("%d MPI processors, %d OpenMP threads per process\n\n", mpi_size, omp_size);
 }
diff --git a/src/libisvd/core/param.c b/src/lib/libisvd/core/param.c
similarity index 98%
rename from src/libisvd/core/param.c
rename to src/lib/libisvd/core/param.c
index e58f70f..503e274 100644
--- a/src/libisvd/core/param.c
+++ b/src/lib/libisvd/core/param.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/param.c
+/// \file       lib/libisvd/core/param.c
 /// \brief      The parameter structure.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c b/src/lib/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
similarity index 79%
rename from src/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
rename to src/lib/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
index ca78300..9488538 100644
--- a/src/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
+++ b/src/lib/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
-/// \brief      The Hierarchical Reduction Integration (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_integrate_hierarchical_reduction.c
+/// \brief      The Hierarchical Reduction Integration (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -10,10 +10,11 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Hierarchical Reduction Integration (@xname@ precision)
+/// \brief  Hierarchical Reduction Integration (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. (not using)
@@ -30,24 +31,20 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@IntegrateHierarchicalReduction(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-          @xtype@    *qt,
-    const isvd_int_t  ldqt
+    const isvd_Param   param,
+    const @xtype_____@ *argv,
+    const isvd_int_t   argc,
+          @xtype_____@ *retv,
+    const isvd_int_t   retc,
+          @xtype_____@ *yst,
+    const isvd_int_t   ldyst,
+          @xtype_____@ *qt,
+    const isvd_int_t   ldqt
 ) {
 
-  if ( argc < 0 ) {
-    return;
-  }
-
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
   // ====================================================================================================================== //
   // Get parameters
@@ -65,21 +62,21 @@ void isvd_@x@IntegrateHierarchicalReduction(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *qst = yst;
+  @xtype_____@ *qst = yst;
   isvd_int_t ldqst = ldyst;
 
   // matrix B
-  @xtype@ *bs = isvd_@x@malloc(l * l * (N+1)/2);
+  @xtype_____@ *bs = isvd_@x@malloc(l * l * (N+1)/2);
   isvd_int_t ldbs = l;
 
   // matrix T
-  @xtype@ *tt = isvd_@x@malloc(l * l);
+  @xtype_____@ *tt = isvd_@x@malloc(l * l);
   isvd_int_t ldtt = l;
 
   // vector s
-  @xtype@ *s = isvd_@x@malloc(l);
+  @xtype_____@ *s = isvd_@x@malloc(l);
 
-  @xtype@ *tmpt = qt;
+  @xtype_____@ *tmpt = qt;
   isvd_int_t ldtmpt = ldqt;
 
   // ====================================================================================================================== //
@@ -91,24 +88,24 @@ void isvd_@x@IntegrateHierarchicalReduction(
     for ( isvd_int_t i = 0; i < h; ++i ) {
       isvd_@x@Gemm('N', 'T', l, l, mj, 1.0, qst + i*l, ldqst, qst + (i+h)*l, ldqst, 0.0, bs + i*ldbs*l, ldbs);
     }
-    MPI_Allreduce(MPI_IN_PLACE, bs, ldbs*l*h, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+    MPI_Allreduce(MPI_IN_PLACE, bs, ldbs*l*h, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
     for ( isvd_int_t i = 0; i < h; ++i ) {
 
       // matrix W
-      @xtype@ *w = bs + i*ldbs*l;
+      @xtype_____@ *w = bs + i*ldbs*l;
       isvd_int_t ldw = ldbs;
 
       // matrix Q(i)
-      @xtype@ *qit = qst + i*l;
+      @xtype_____@ *qit = qst + i*l;
       isvd_int_t ldqit = ldqst;
 
       // matrix Q(i+h)
-      @xtype@ *qiht = qst + (i+h)*l;
+      @xtype_____@ *qiht = qst + (i+h)*l;
       isvd_int_t ldqiht = ldqst;
 
       // svd(B(i)) = W * S * T'
-      isvd_@x@Gesvd('O', 'S', l, l, w, ldw, s, nullptr, 1, tt, ldtt);
+      isvd_@x@Gesvd('O', 'S', l, l, w, ldw, s, NULL, 1, tt, ldtt);
 
       // Q(i) := Q(i) * W + Q(i+h) * T (Q(i)' := W' * Q(i)' + T' * Q(i+h)')
       isvd_@x@Omatcopy('N', l, mj, 1.0, qit, ldqit, tmpt, ldtmpt);
diff --git a/src/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c b/src/lib/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
similarity index 83%
rename from src/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
rename to src/lib/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
index 94e665d..c5a1a0a 100644
--- a/src/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
+++ b/src/lib/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
-/// \brief      The Kolmogorov-Nagumo Integration (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_integrate_kolmogorov_nagumo.c
+/// \brief      The Kolmogorov-Nagumo Integration (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -10,13 +10,14 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 #define kMaxit 256
 #define kTol   1e-4
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Kolmogorov-Nagumo Integration (@xname@ precision)
+/// \brief  Kolmogorov-Nagumo Integration (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. <br>
@@ -36,17 +37,20 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@IntegrateKolmogorovNagumo(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const @xtype@    *yst,
-    const isvd_int_t  ldyst,
-          @xtype@    *qt,
-    const isvd_int_t  ldqt
+    const isvd_Param   param,
+    const @xtype_____@ *argv,
+    const isvd_int_t   argc,
+          @xtype_____@ *retv,
+    const isvd_int_t   retc,
+    const @xtype_____@ *yst,
+    const isvd_int_t   ldyst,
+          @xtype_____@ *qt,
+    const isvd_int_t   ldqt
 ) {
 
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+
   // ====================================================================================================================== //
   // Query arguments
 
@@ -62,7 +66,7 @@ void isvd_@x@IntegrateKolmogorovNagumo(
 
   isvd_int_t argi = -1;
   const isvd_int_t maxit = ( argc > ++argi ) ? argv[argi] : kMaxit;
-  const @xtype@ tol      = ( argc > ++argi ) ? argv[argi] : kTol;
+  const @xtype_____@ tol = ( argc > ++argi ) ? argv[argi] : kTol;
 
   // ====================================================================================================================== //
   // Get parameters
@@ -82,64 +86,64 @@ void isvd_@x@IntegrateKolmogorovNagumo(
   // ====================================================================================================================== //
   // Allocate memory
 
-  const @xtype@ *qst = yst;
+  const @xtype_____@ *qst = yst;
   isvd_int_t ldqst = ldyst;
 
   // matrix Qc'
-  @xtype@ *qct = isvd_@x@malloc(l * mj);
+  @xtype_____@ *qct = isvd_@x@malloc(l * mj);
   isvd_int_t ldqct = l;
 
   // matrix Q+'
-  @xtype@ *qpt = isvd_@x@malloc(l * mj);
+  @xtype_____@ *qpt = isvd_@x@malloc(l * mj);
   isvd_int_t ldqpt = l;
 
   // matrix Gc'
-  @xtype@ *gct = isvd_@x@malloc(l * mj);
+  @xtype_____@ *gct = isvd_@x@malloc(l * mj);
   isvd_int_t ldgct = l;
 
   // matrix Bc
-  @xtype@ *bc = isvd_@x@malloc(Nl * l);
+  @xtype_____@ *bc = isvd_@x@malloc(Nl * l);
   isvd_int_t ldbc = Nl;
 
   // matrix B+
-  @xtype@ *bp = isvd_@x@malloc(Nl * l);
+  @xtype_____@ *bp = isvd_@x@malloc(Nl * l);
   isvd_int_t ldbp = Nl;
 
   // matrix Bgc
-  @xtype@ *bgc = isvd_@x@malloc(Nl * l);
+  @xtype_____@ *bgc = isvd_@x@malloc(Nl * l);
   isvd_int_t ldbgc = Nl;
 
   // matrix Dc
-  @xtype@ *dc = isvd_@x@malloc(l * l);
+  @xtype_____@ *dc = isvd_@x@malloc(l * l);
   isvd_int_t lddc = l;
 
   // matrix Z
-  @xtype@ *z = isvd_@x@malloc(l * l);
+  @xtype_____@ *z = isvd_@x@malloc(l * l);
   isvd_int_t ldz = l;
 
   // matrix C
-  @xtype@ *c = isvd_@x@malloc(l * l);
+  @xtype_____@ *c = isvd_@x@malloc(l * l);
   isvd_int_t ldc = l;
 
   // matrix inv(C)
-  @xtype@ *cinv = isvd_@x@malloc(l * l);
+  @xtype_____@ *cinv = isvd_@x@malloc(l * l);
   isvd_int_t ldcinv = l;
 
   // vector s
-  @xtype@ *s = isvd_@x@malloc(l * 2);
+  @xtype_____@ *s = isvd_@x@malloc(l * 2);
 
   // matrix Z * sqrt(S)
-  @xtype@ *zs     = cinv;
-  isvd_int_t ldzs    = ldcinv;
+  @xtype_____@ *zs = cinv;
+  isvd_int_t ldzs = ldcinv;
 
   // matrix Z / sqrt(S)
-  @xtype@ *zinvs  = z;
+  @xtype_____@ *zinvs = z;
   isvd_int_t ldzinvs = ldz;
 
   // matrix sqrt(S)
-  @xtype@ *ss = s + l;
+  @xtype_____@ *ss = s + l;
 
-  @xtype@ *tmp;
+  @xtype_____@ *tmp;
 
   // ====================================================================================================================== //
   // Initializing
@@ -149,13 +153,13 @@ void isvd_@x@IntegrateKolmogorovNagumo(
 
   // Bc := Qs' * Qc
   isvd_@x@Gemm('N', 'T', Nl, l, mj, 1.0, qst, ldqst, qct, ldqct, 0.0, bc, ldbc);
-  MPI_Allreduce(MPI_IN_PLACE, bc, ldbc*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Allreduce(MPI_IN_PLACE, bc, ldbc*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Iterating
 
   isvd_int_t iter;
-  @xtype@ error = -1.0/0.0;
+  @xtype_____@ error = -1.0/0.0;
 
   for ( iter = 1; ; ++iter ) {
 
@@ -167,7 +171,7 @@ void isvd_@x@IntegrateKolmogorovNagumo(
 
     // Bgc := Qs' * Gc
     isvd_@x@Gemm('N', 'T', Nl, l, mj, 1.0, qst, ldqst, gct, ldgct, 0.0, bgc, ldbgc);
-    MPI_Allreduce(MPI_IN_PLACE, bgc, ldbgc*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+    MPI_Allreduce(MPI_IN_PLACE, bgc, ldbgc*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
     // Dc := 1/N * Bc' * Bc
     isvd_@x@Gemm('T', 'N', l, l, Nl, 1.0/N, bc, ldbc, bc, ldbc, 0.0, dc, lddc);
diff --git a/src/libisvd/core/stage/@x@_integrate_wen_yin.c b/src/lib/libisvd/core/stage/@x@_integrate_wen_yin.c
similarity index 76%
rename from src/libisvd/core/stage/@x@_integrate_wen_yin.c
rename to src/lib/libisvd/core/stage/@x@_integrate_wen_yin.c
index e2fae04..6af8cce 100644
--- a/src/libisvd/core/stage/@x@_integrate_wen_yin.c
+++ b/src/lib/libisvd/core/stage/@x@_integrate_wen_yin.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_integrate_wen_yin.c
-/// \brief      The Wen-Yin Integration (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_integrate_wen_yin.c
+/// \brief      The Wen-Yin Integration (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -10,6 +10,7 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 #define kMaxit    256
 #define kTol      1e-3
@@ -23,12 +24,19 @@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Wen-Yin Integration (@xname@ precision)
+/// \brief  Wen-Yin Integration (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. <br>
 ///                          \b argv[0]: The maximum number of iteration. <br>
-///                          \b argv[1]: The tolerance of convergence condition.
+///                          \b argv[1]: The tolerance of convergence condition. <br>
+///                          \b argv[2]: The initial step size \f$ \tau_0 \f$. <br>
+///                          \b argv[3]: The maximum predicting step size \f$ \tau_\max \f$. <br>
+///                          \b argv[4]: The minimum predicting step size \f$ \tau_\min \f$. <br>
+///                          \b argv[5]: The maximum number of iteration in predicting step size. <br>
+///                          \b argv[6]: The scaling parameter for step size searching \f$ \beta \f$. <br>
+///                          \b argv[7]: The parameter for step size searching \f$ \sigma \f$. <br>
+///                          \b argv[8]: The parameter for next step searching \f$ \eta \f$.
 /// \param[in]   retv, retc  The return values and its length.
 /// <hr>
 /// \param[in]   yst, ldyst  The row-block 𝕼 (\f$ m_b \times Nl \f$, row-major) and its leading dimension.
@@ -43,17 +51,20 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@IntegrateWenYin(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const @xtype@    *yst,
-    const isvd_int_t  ldyst,
-          @xtype@    *qt,
-    const isvd_int_t  ldqt
+    const isvd_Param   param,
+    const @xtype_____@ *argv,
+    const isvd_int_t   argc,
+          @xtype_____@ *retv,
+    const isvd_int_t   retc,
+    const @xtype_____@ *yst,
+    const isvd_int_t   ldyst,
+          @xtype_____@ *qt,
+    const isvd_int_t   ldqt
 ) {
 
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+
   // ====================================================================================================================== //
   // Query arguments
 
@@ -75,15 +86,15 @@ void isvd_@x@IntegrateWenYin(
   // Get arguments
 
   isvd_int_t argi = -1;
-  const isvd_int_t maxit    = ( argc > ++argi ) ? argv[argi] : kMaxit;
-  const @xtype@ tol         = ( argc > ++argi ) ? argv[argi] : kTol;
-  const @xtype@ tau0        = ( argc > ++argi ) ? argv[argi] : kTau0;
-  const @xtype@ taumax      = ( argc > ++argi ) ? argv[argi] : kTaumax;
-  const @xtype@ taumin      = ( argc > ++argi ) ? argv[argi] : kTaumin;
-  const isvd_int_t taumaxit = ( argc > ++argi ) ? argv[argi] : kTauMaxit;
-  const @xtype@ beta        = ( argc > ++argi ) ? argv[argi] : kBeta;
-  const @xtype@ sigma       = ( argc > ++argi ) ? argv[argi] : kSigma;
-  const @xtype@ eta         = ( argc > ++argi ) ? argv[argi] : kEta;
+  const isvd_int_t   maxit    = ( argc > ++argi ) ? argv[argi] : kMaxit;
+  const @xtype_____@ tol      = ( argc > ++argi ) ? argv[argi] : kTol;
+  const @xtype_____@ tau0     = ( argc > ++argi ) ? argv[argi] : kTau0;
+  const @xtype_____@ taumax   = ( argc > ++argi ) ? argv[argi] : kTaumax;
+  const @xtype_____@ taumin   = ( argc > ++argi ) ? argv[argi] : kTaumin;
+  const isvd_int_t   taumaxit = ( argc > ++argi ) ? argv[argi] : kTauMaxit;
+  const @xtype_____@ beta     = ( argc > ++argi ) ? argv[argi] : kBeta;
+  const @xtype_____@ sigma    = ( argc > ++argi ) ? argv[argi] : kSigma;
+  const @xtype_____@ eta      = ( argc > ++argi ) ? argv[argi] : kEta;
 
   // ====================================================================================================================== //
   // Get parameters
@@ -111,74 +122,74 @@ void isvd_@x@IntegrateWenYin(
   // ====================================================================================================================== //
   // Allocate memory
 
-  const @xtype@ *qst = yst;
+  const @xtype_____@ *qst = yst;
   isvd_int_t ldqst = ldyst;
 
   // matrix Qc'
-  @xtype@ *qct = isvd_@x@malloc(l * mj);
+  @xtype_____@ *qct = isvd_@x@malloc(l * mj);
   isvd_int_t ldqct = l;
 
   // matrix Q+'
-  @xtype@ *qpt = isvd_@x@malloc(l * mj);
+  @xtype_____@ *qpt = isvd_@x@malloc(l * mj);
   isvd_int_t ldqpt = l;
 
   // matrix Gc'
-  @xtype@ *gct = isvd_@x@malloc(l * mj);
+  @xtype_____@ *gct = isvd_@x@malloc(l * mj);
   isvd_int_t ldgct = l;
 
   // matrix Xc'
-  @xtype@ *xct = isvd_@x@malloc(l * mj);
+  @xtype_____@ *xct = isvd_@x@malloc(l * mj);
   isvd_int_t ldxct = l;
 
   // matrix X+'
-  @xtype@ *xpt = isvd_@x@malloc(l * mj);
+  @xtype_____@ *xpt = isvd_@x@malloc(l * mj);
   isvd_int_t ldxpt = l;
 
   // matrix Bc
-  @xtype@ *bc = isvd_@x@malloc(Nl * l);
+  @xtype_____@ *bc = isvd_@x@malloc(Nl * l);
   isvd_int_t ldbc = Nl;
 
   // matrix B+
-  @xtype@ *bp = isvd_@x@malloc(Nl * l);
+  @xtype_____@ *bp = isvd_@x@malloc(Nl * l);
   isvd_int_t ldbp = Nl;
 
   // matrix Bgc
-  @xtype@ *bgc = isvd_@x@malloc(Nl * l);
+  @xtype_____@ *bgc = isvd_@x@malloc(Nl * l);
   isvd_int_t ldbgc = Nl;
 
   // matrix Dc
-  @xtype@ *dc = isvd_@x@malloc(l * l);
+  @xtype_____@ *dc = isvd_@x@malloc(l * l);
   isvd_int_t lddc = l;
 
   // matrix Dgc
-  @xtype@ *dgc = isvd_@x@malloc(l * l);
+  @xtype_____@ *dgc = isvd_@x@malloc(l * l);
   isvd_int_t lddgc = l;
 
   // matrix C & inv(C)
-  @xtype@ *c = isvd_@x@malloc(l2 * l2);
+  @xtype_____@ *c = isvd_@x@malloc(l2 * l2);
   isvd_int_t ldc = l2;
 
   // matrix inv(C)_??
-  @xtype@ *c11 = c;
-  @xtype@ *c21 = c + l;
-  @xtype@ *c12 = c + ldc*l;
-  @xtype@ *c22 = c + l + ldc*l;
+  @xtype_____@ *c11 = c;
+  @xtype_____@ *c21 = c + l;
+  @xtype_____@ *c12 = c + ldc*l;
+  @xtype_____@ *c22 = c + l + ldc*l;
 
   // matrix inv(C)_#?
-  @xtype@ *cs1 = c11;
-  @xtype@ *cs2 = c12;
+  @xtype_____@ *cs1 = c11;
+  @xtype_____@ *cs2 = c12;
 
   // matrix Fc
-  @xtype@ *fc = c21;
+  @xtype_____@ *fc = c21;
   isvd_int_t ldfc = ldc;
 
   // matrix Fgc
-  @xtype@ *fgc = c11;
+  @xtype_____@ *fgc = c11;
   isvd_int_t ldfgc = ldc;
 
-  @xtype@ *tmp;
+  @xtype_____@ *tmp;
 
-  @xtype@ taug, zeta, phi, mu;
+  @xtype_____@ taug, zeta, phi, mu;
 
   // ====================================================================================================================== //
   // Initializing
@@ -188,7 +199,7 @@ void isvd_@x@IntegrateWenYin(
 
   // Bc := Qs' * Qc
   isvd_@x@Gemm('N', 'T', Nl, l, mj, 1.0, qst, ldqst, qct, ldqct, 0.0, bc, ldbc);
-  MPI_Allreduce(MPI_IN_PLACE, bc, ldbc*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Allreduce(MPI_IN_PLACE, bc, ldbc*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // Dc := 1/N * Bc' * Bc
   isvd_@x@Gemm('T', 'N', l, l, Nl, 1.0/N, bc, ldbc, bc, ldbc, 0.0, dc, lddc);
@@ -198,7 +209,7 @@ void isvd_@x@IntegrateWenYin(
 
   // Bgc := Qs' * Gc
   isvd_@x@Gemm('N', 'T', Nl, l, mj, 1.0, qst, ldqst, gct, ldgct, 0.0, bgc, ldbgc);
-  MPI_Allreduce(MPI_IN_PLACE, bgc, ldbgc*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Allreduce(MPI_IN_PLACE, bgc, ldbgc*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // Dgc := 1/N * Bc' * Bgc
   isvd_@x@Gemm('T', 'N', l, l, Nl, 1.0/N, bc, ldbc, bgc, ldbgc, 0.0, dgc, lddgc);
@@ -224,7 +235,7 @@ void isvd_@x@IntegrateWenYin(
 
     // ================================================================================================================== //
     // Find step size
-    @xtype@ tau = taug, phit = phi;
+    @xtype_____@ tau = taug, phit = phi;
     for ( isvd_int_t tauiter = 1; tauiter <= taumaxit; ++tauiter, tau *= beta ) {
 
       // C := [ Dc/2 - I/tau , I/2          ;
@@ -285,7 +296,7 @@ void isvd_@x@IntegrateWenYin(
 
     // Bg+ [in Bgc] := Qs' * G+ [in Gc]
     isvd_@x@Gemm('N', 'T', Nl, l, mj, 1.0, qst, ldqst, gct, ldgct, 0.0, bgc, ldbgc);
-    MPI_Allreduce(MPI_IN_PLACE, bgc, ldbgc*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+    MPI_Allreduce(MPI_IN_PLACE, bgc, ldbgc*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
     // Dg+ [in Dgc] := 1/N * B+' * Bg+ [in Bgc]
     isvd_@x@Gemm('T', 'N', l, l, Nl, 1.0/N, bp, ldbp, bgc, ldbgc, 0.0, dgc, lddgc);
@@ -317,7 +328,7 @@ void isvd_@x@IntegrateWenYin(
     isvd_v@x@Sub(mj*ldxct, xct, xpt, xct);
 
     // Update taug
-    @xtype@ t[2];
+    @xtype_____@ t[2];
 
     if ( iter % 2 ) {
       t[0] = isvd_@x@Dot(mj*ldqct, qct, 1, qct, 1);
@@ -327,7 +338,7 @@ void isvd_@x@IntegrateWenYin(
       t[1] = isvd_@x@Dot(mj*ldxct, xct, 1, xct, 1);
     }
 
-    MPI_Allreduce(MPI_IN_PLACE, t, 2, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+    MPI_Allreduce(MPI_IN_PLACE, t, 2, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
     taug = fabs(t[0]/t[1]);
     if ( taug < taumin ) { taug = taumin; }
diff --git a/src/libisvd/core/stage/@x@_orthogonalize_gramian.c b/src/lib/libisvd/core/stage/@x@_orthogonalize_gramian.c
similarity index 78%
rename from src/libisvd/core/stage/@x@_orthogonalize_gramian.c
rename to src/lib/libisvd/core/stage/@x@_orthogonalize_gramian.c
index 64ae42f..07d8591 100644
--- a/src/libisvd/core/stage/@x@_orthogonalize_gramian.c
+++ b/src/lib/libisvd/core/stage/@x@_orthogonalize_gramian.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_orthogonalize_gramian.c
-/// \brief      The Gramian Orthogonalization (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_orthogonalize_gramian.c
+/// \brief      The Gramian Orthogonalization (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -10,10 +10,11 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Gramian Orthogonalization (@xname@ precision)
+/// \brief  Gramian Orthogonalization (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. (not using)
@@ -27,22 +28,18 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@OrthogonalizeGramian(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst
+    const isvd_Param   param,
+    const @xtype_____@ *argv,
+    const isvd_int_t   argc,
+          @xtype_____@ *retv,
+    const isvd_int_t   retc,
+          @xtype_____@ *yst,
+    const isvd_int_t   ldyst
 ) {
 
-  if ( argc < 0 ) {
-    return;
-  }
-
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
   // ====================================================================================================================== //
   // Get parameters
@@ -60,13 +57,13 @@ void isvd_@x@OrthogonalizeGramian(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *yst_ = isvd_@x@malloc(ldyst * mj);
+  @xtype_____@ *yst_ = isvd_@x@malloc(ldyst * mj);
   isvd_int_t ldyst_ = ldyst;
 
-  @xtype@ *w = isvd_@x@malloc(l * Nl);
+  @xtype_____@ *w = isvd_@x@malloc(l * Nl);
   isvd_int_t ldw = l;
 
-  @xtype@ *s = isvd_@x@malloc(l * N);
+  @xtype_____@ *s = isvd_@x@malloc(l * N);
   isvd_int_t lds = l;
 
   // ====================================================================================================================== //
@@ -76,11 +73,11 @@ void isvd_@x@OrthogonalizeGramian(
   for ( isvd_int_t i = 0; i < N; ++i ) {
     isvd_@x@Gemm('N', 'T', l, l, mj, 1.0, yst + i*l, ldyst, yst + i*l, ldyst, 0.0, w + i*ldw*l, ldw);
   }
-  MPI_Allreduce(MPI_IN_PLACE, w, ldw*Nl, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Allreduce(MPI_IN_PLACE, w, ldw*Nl, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // eig(Wi) = Wi * Si^2 * Wi'
   for ( isvd_int_t i = 0; i < N; ++i ) {
-    isvd_@x@Gesvd('O', 'N', l, l, w + i*ldw*l, ldw, s + i*lds, nullptr, 1, nullptr, 1);
+    isvd_@x@Gesvd('O', 'N', l, l, w + i*ldw*l, ldw, s + i*lds, NULL, 1, NULL, 1);
   }
   isvd_v@x@Sqrt(lds*N, s, s);
 
diff --git a/src/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c b/src/lib/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
similarity index 69%
rename from src/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
rename to src/lib/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
index 6a065ea..6df2d46 100644
--- a/src/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
+++ b/src/lib/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
-/// \brief      The Tall-Skinny QR Orthogonalization (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_orthogonalize_tall_skinny_qr.c
+/// \brief      The Tall-Skinny QR Orthogonalization (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -10,10 +10,11 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Tall-Skinny QR Orthogonalization (@xname@ precision)
+/// \brief  Tall-Skinny QR Orthogonalization (@xname@ precision).
 ///
 /// \attention  Not implemented!
 ///
@@ -21,25 +22,22 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@OrthogonalizeTallSkinnyQr(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst
+    const isvd_Param   param,
+    const @xtype_____@ *argv,
+    const isvd_int_t   argc,
+          @xtype_____@ *retv,
+    const isvd_int_t   retc,
+          @xtype_____@ *yst,
+    const isvd_int_t   ldyst
 ) {
 
   fprintf(stderr, "Tall-Skinny QR Orthogonalization is not implemented!\n");
 
-  if ( argc < 0 ) {
-    return;
-  }
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
   ISVD_UNUSED(param);
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
   ISVD_UNUSED(yst);
   ISVD_UNUSED(ldyst);
 
diff --git a/src/libisvd/core/stage/@x@_postprocess.h b/src/lib/libisvd/core/stage/@x@_postprocess.h
similarity index 80%
rename from src/libisvd/core/stage/@x@_postprocess.h
rename to src/lib/libisvd/core/stage/@x@_postprocess.h
index a42cb74..ffc9229 100644
--- a/src/libisvd/core/stage/@x@_postprocess.h
+++ b/src/lib/libisvd/core/stage/@x@_postprocess.h
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_postprocess.h
-/// \brief      The Postprocessing utilities (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_postprocess.h
+/// \brief      The Postprocessing utilities (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -12,21 +12,22 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 static void projectBlockCol(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *zt,
-    const isvd_int_t  ldzt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-    const mpi_int_t   ut_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *zt,
+    const isvd_int_t    ldzt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+    const mpi_int_t     ut_root
 ) {
 
   ISVD_UNUSED(s);
@@ -61,7 +62,7 @@ static void projectBlockCol(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *qt_;
+  @xtype_____@ *qt_;
   if ( use_ut ) {
     qt_ = ut;
   } else {
@@ -72,7 +73,7 @@ static void projectBlockCol(
   // ====================================================================================================================== //
   // Rearrange
 
-  MPI_Allgather(qt, mb*ldqt, MPI_@X_TYPE@, qt_, mb*ldqt, MPI_@X_TYPE@, param.mpi_comm);
+  MPI_Allgather(qt, mb*ldqt, MPI_@XTYPE@, qt_, mb*ldqt, MPI_@XTYPE@, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Project
@@ -91,18 +92,18 @@ static void projectBlockCol(
 }
 
 static void projectBlockRow(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *zt,
-    const isvd_int_t  ldzt,
-          @xtype@    *s,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *zt,
+    const isvd_int_t    ldzt,
+          @xtype_____@ *s,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     vt_root
 ) {
 
   ISVD_UNUSED(s);
@@ -137,7 +138,7 @@ static void projectBlockRow(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *zt_;
+  @xtype_____@ *zt_;
   if ( use_vt ) {
     zt_ = vt;
   } else {
@@ -155,7 +156,7 @@ static void projectBlockRow(
   // ====================================================================================================================== //
   // Rearrange
 
-  MPI_Reduce_scatter_block(zt_, zt, nb*ldzt, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Reduce_scatter_block(zt_, zt, nb*ldzt, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Deallocate memory
diff --git a/src/libisvd/core/stage/@x@_postprocess_gramian.c b/src/lib/libisvd/core/stage/@x@_postprocess_gramian.c
similarity index 77%
rename from src/libisvd/core/stage/@x@_postprocess_gramian.c
rename to src/lib/libisvd/core/stage/@x@_postprocess_gramian.c
index 70c0528..f3ebb3d 100644
--- a/src/libisvd/core/stage/@x@_postprocess_gramian.c
+++ b/src/lib/libisvd/core/stage/@x@_postprocess_gramian.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_postprocess_gramian.c
-/// \brief      The Gramian Postprocessing (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_postprocess_gramian.c
+/// \brief      The Gramian Postprocessing (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -15,7 +15,7 @@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Gramian Postprocessing (@xname@ precision)
+/// \brief  Gramian Postprocessing (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. (not using)
@@ -27,9 +27,9 @@
 /// \param[in]   ordera      The storage ordering of 𝑨. <br>
 ///                          `'C'`: column-major ordering. <br>
 ///                          `'R'`: row-major ordering.
-/// \param[in]   a, lda      The column/row-block 𝑨 (\f$m \times n_j\f$) and its leading dimension. <br>
-///                          \b dista = `'C'`: the size must be \f$m \times n_j\f$. <br>
-///                          \b dista = `'R'`: the size must be \f$m_j \times n\f$.
+/// \param[in]   a, lda      The column/row-block 𝑨 (\f$m \times n^{(j)}\f$) and its leading dimension. <br>
+///                          \b dista = `'C'`: the size must be \f$m \times n^{(j)}\f$. <br>
+///                          \b dista = `'R'`: the size must be \f$m^{(j)} \times n\f$.
 /// \param[in]   qt, ldqt    The row-block 𝑸 (\f$ m_b \times l \f$, row-major) and its leading dimension.
 /// \param[in]   s           The vector 𝝈 (\f$k \times 1\f$).
 /// \param[in]   ut, ldut    The matrix 𝑼 (row-major) and its leading dimension. <br>
@@ -57,33 +57,29 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@PostprocessGramian(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root
 ) {
 
-  if ( argc < 0 ) {
-    return;
-  }
-
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
   // ====================================================================================================================== //
   // Get parameters
@@ -98,8 +94,8 @@ void isvd_@x@PostprocessGramian(
   // ====================================================================================================================== //
   // Check arguments
 
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   if ( !dista_ || !ordera_ ) return;
 
   if ( ut_root >= 0 ) {
@@ -117,10 +113,10 @@ void isvd_@x@PostprocessGramian(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *zt = isvd_@x@malloc(l * nb);
+  @xtype_____@ *zt = isvd_@x@malloc(l * nb);
   isvd_int_t ldzt = l;
 
-  @xtype@ *w = isvd_@x@malloc(l * l);
+  @xtype_____@ *w = isvd_@x@malloc(l * l);
   isvd_int_t ldw = l;
 
   // ====================================================================================================================== //
@@ -137,11 +133,11 @@ void isvd_@x@PostprocessGramian(
 
   // W := Z' * Z
   isvd_@x@Gemm('N', 'T', l, l, nj, 1.0, zt, ldzt, zt, ldzt, 0.0, w, ldw);
-  MPI_Allreduce(MPI_IN_PLACE, w, ldw*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Allreduce(MPI_IN_PLACE, w, ldw*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // eig(W) = W * S^2 * W'
   const char jobw_ = (ut_root >= -1 || vt_root >= -1) ? 'O' : 'N';
-  isvd_@x@Gesvd(jobw_, 'N', l, l, w, ldw, s, nullptr, 1, nullptr, 1);
+  isvd_@x@Gesvd(jobw_, 'N', l, l, w, ldw, s, NULL, 1, NULL, 1);
   isvd_v@x@Sqrt(l, s, s);
 
   // ====================================================================================================================== //
@@ -153,9 +149,9 @@ void isvd_@x@PostprocessGramian(
 
     if ( ut_root >= 0 ) {
       if ( param.mpi_rank == ut_root ) {
-        MPI_Gather(MPI_IN_PLACE, mb*ldut, MPI_@X_TYPE@, ut, mb*ldut, MPI_@X_TYPE@, ut_root, param.mpi_comm);
+        MPI_Gather(MPI_IN_PLACE, mb*ldut, MPI_@XTYPE@, ut, mb*ldut, MPI_@XTYPE@, ut_root, param.mpi_comm);
       } else {
-        MPI_Gather(ut, mb*ldut, MPI_@X_TYPE@, nullptr, mb*ldut, MPI_@X_TYPE@, ut_root, param.mpi_comm);
+        MPI_Gather(ut, mb*ldut, MPI_@XTYPE@, NULL, mb*ldut, MPI_@XTYPE@, ut_root, param.mpi_comm);
       }
     }
   }
@@ -167,9 +163,9 @@ void isvd_@x@PostprocessGramian(
 
     if ( vt_root >= 0 ) {
       if ( param.mpi_rank == vt_root ) {
-        MPI_Gather(MPI_IN_PLACE, nb*ldvt, MPI_@X_TYPE@, vt, nb*ldvt, MPI_@X_TYPE@, vt_root, param.mpi_comm);
+        MPI_Gather(MPI_IN_PLACE, nb*ldvt, MPI_@XTYPE@, vt, nb*ldvt, MPI_@XTYPE@, vt_root, param.mpi_comm);
       } else {
-        MPI_Gather(vt, nb*ldvt, MPI_@X_TYPE@, nullptr, nb*ldvt, MPI_@X_TYPE@, vt_root, param.mpi_comm);
+        MPI_Gather(vt, nb*ldvt, MPI_@XTYPE@, NULL, nb*ldvt, MPI_@XTYPE@, vt_root, param.mpi_comm);
       }
     }
   }
diff --git a/src/libisvd/core/stage/@x@_postprocess_symmetric.c b/src/lib/libisvd/core/stage/@x@_postprocess_symmetric.c
similarity index 80%
rename from src/libisvd/core/stage/@x@_postprocess_symmetric.c
rename to src/lib/libisvd/core/stage/@x@_postprocess_symmetric.c
index a862634..c34828c 100644
--- a/src/libisvd/core/stage/@x@_postprocess_symmetric.c
+++ b/src/lib/libisvd/core/stage/@x@_postprocess_symmetric.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_postprocess_symmetric.c
-/// \brief      The Symmetric Postprocessing (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_postprocess_symmetric.c
+/// \brief      The Symmetric Postprocessing (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -15,7 +15,7 @@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Symmetric Postprocessing (@xname@ precision)
+/// \brief  Symmetric Postprocessing (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. (not using)
@@ -27,9 +27,9 @@
 /// \param[in]   ordera      The storage ordering of 𝑨. <br>
 ///                          `'C'`: column-major ordering. <br>
 ///                          `'R'`: row-major ordering.
-/// \param[in]   a, lda      The column/row-block 𝑨 (\f$m \times n_j\f$) and its leading dimension. <br>
-///                          \b dista = `'C'`: the size must be \f$m \times n_j\f$. <br>
-///                          \b dista = `'R'`: the size must be \f$m_j \times n\f$.
+/// \param[in]   a, lda      The column/row-block 𝑨 (\f$m \times n^{(j)}\f$) and its leading dimension. <br>
+///                          \b dista = `'C'`: the size must be \f$m \times n^{(j)}\f$. <br>
+///                          \b dista = `'R'`: the size must be \f$m^{(j)} \times n\f$.
 /// \param[in]   qt, ldqt    The row-block 𝑸 (\f$ m_b \times l \f$, row-major) and its leading dimension.
 /// \param[in]   s           The vector 𝝈 (\f$k \times 1\f$).
 /// \param[in]   ut, ldut    The matrix 𝑼 (row-major) and its leading dimension. <br>
@@ -57,33 +57,30 @@
 /// \note  The result of 𝑼 and 𝑽 are the same.
 ///
 void isvd_@x@PostprocessSymmetric(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root
 ) {
 
-  if ( argc < 0 ) {
-    return;
-  }
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
   ISVD_UNUSED(vt);
   ISVD_UNUSED(ldvt);
 
@@ -100,8 +97,8 @@ void isvd_@x@PostprocessSymmetric(
   // ====================================================================================================================== //
   // Check arguments
 
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   if ( vt_root >= -1 ) {
     fprintf(stderr, "VT_ROOT must not be set!");
     return;
@@ -120,10 +117,10 @@ void isvd_@x@PostprocessSymmetric(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *zt = isvd_@x@malloc(l * nb);
+  @xtype_____@ *zt = isvd_@x@malloc(l * nb);
   isvd_int_t ldzt = l;
 
-  @xtype@ *w = isvd_@x@malloc(l * l);
+  @xtype_____@ *w = isvd_@x@malloc(l * l);
   isvd_int_t ldw = l;
 
   // ====================================================================================================================== //
@@ -140,7 +137,7 @@ void isvd_@x@PostprocessSymmetric(
 
   // W := Z' * Q
   isvd_@x@Gemmt('U', 'N', 'T', l, nj, 1.0, zt, ldzt, qt, ldqt, 0.0, w, ldw);
-  MPI_Allreduce(MPI_IN_PLACE, w, ldw*l, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Allreduce(MPI_IN_PLACE, w, ldw*l, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // eig(W) = W * S * W'
   const char jobw_ = (ut_root >= -1) ? 'V' : 'N';
@@ -155,9 +152,9 @@ void isvd_@x@PostprocessSymmetric(
 
     if ( ut_root >= 0 ) {
       if ( param.mpi_rank == ut_root ) {
-        MPI_Gather(MPI_IN_PLACE, mb*ldut, MPI_@X_TYPE@, ut, mb*ldut, MPI_@X_TYPE@, ut_root, param.mpi_comm);
+        MPI_Gather(MPI_IN_PLACE, mb*ldut, MPI_@XTYPE@, ut, mb*ldut, MPI_@XTYPE@, ut_root, param.mpi_comm);
       } else {
-        MPI_Gather(ut, mb*ldut, MPI_@X_TYPE@, nullptr, mb*ldut, MPI_@X_TYPE@, ut_root, param.mpi_comm);
+        MPI_Gather(ut, mb*ldut, MPI_@XTYPE@, NULL, mb*ldut, MPI_@XTYPE@, ut_root, param.mpi_comm);
       }
     }
   }
diff --git a/src/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c b/src/lib/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
similarity index 62%
rename from src/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
rename to src/lib/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
index da5bbe9..3276561 100644
--- a/src/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
+++ b/src/lib/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
-/// \brief      The Tall-Skinny QR Postprocessing (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_postprocess_tall_skinny_qr.c
+/// \brief      The Tall-Skinny QR Postprocessing (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -15,7 +15,7 @@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Tall-Skinny QR Postprocessing (@xname@ precision)
+/// \brief  Tall-Skinny QR Postprocessing (@xname@ precision).
 ///
 /// \attention  Not implemented!
 ///
@@ -23,36 +23,33 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@PostprocessTallSkinnyQr(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root
 ) {
 
   fprintf(stderr, "Tall-Skinny QR Postprocessing is not implemented!\n");
 
-  if ( argc < 0 ) {
-    return;
-  }
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
   ISVD_UNUSED(param);
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
   ISVD_UNUSED(dista);
   ISVD_UNUSED(ordera);
   ISVD_UNUSED(a);
diff --git a/src/libisvd/core/stage/@x@_sketch.h b/src/lib/libisvd/core/stage/@x@_sketch.h
similarity index 69%
rename from src/libisvd/core/stage/@x@_sketch.h
rename to src/lib/libisvd/core/stage/@x@_sketch.h
index d594b76..a8a61ba 100644
--- a/src/libisvd/core/stage/@x@_sketch.h
+++ b/src/lib/libisvd/core/stage/@x@_sketch.h
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_sketch.h
-/// \brief      The Sketching utilities (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_sketch.h
+/// \brief      The Sketching utilities (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -13,18 +13,19 @@
 #include <libisvd/def.h>
 #include <isvd/la.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 #include <isvd/util/omp.h>
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 static void sketchBlockCol(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-    const isvd_int_t  seed,
-    const mpi_int_t   mpi_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *yst,
+    const isvd_int_t    ldyst,
+    const isvd_int_t    seed,
+    const mpi_int_t     mpi_root
 ) {
 
   ISVD_UNUSED(ldyst);
@@ -52,10 +53,10 @@ static void sketchBlockCol(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *omegat = isvd_@x@malloc(Nl * nj);
+  @xtype_____@ *omegat = isvd_@x@malloc(Nl * nj);
   isvd_int_t ldomegat = Nl;
 
-  @xtype@ *yst_ = isvd_@x@malloc(Nl * Pmb);
+  @xtype_____@ *yst_ = isvd_@x@malloc(Nl * Pmb);
   isvd_int_t ldyst_ = Nl;
 
   // ====================================================================================================================== //
@@ -63,26 +64,7 @@ static void sketchBlockCol(
 
   isvd_int_t seed_ = seed;
   MPI_Bcast(&seed_, sizeof(seed_), MPI_BYTE, mpi_root, param.mpi_comm);
-
-
-  ISVD_OMP_PARALLEL
-  {
-    omp_int_t omp_size = isvd_getOmpSize();
-    omp_int_t omp_rank = isvd_getOmpRank();
-
-    isvd_int_t len   = nj * Nl / omp_size;
-    isvd_int_t start = len * omp_rank;
-    if ( omp_rank == omp_size-1 ) {
-      len = nj * Nl - start;
-    }
-
-    isvd_VSLStreamStatePtr stream;
-    isvd_vslNewStream(&stream, seed_);
-    isvd_vslSkipAheadStream(stream, (Nl * nb * param.mpi_rank + start) * 2);
-    isvd_v@x@RngGaussian(stream, len, omegat + start, 0.0, 1.0);
-
-    isvd_vslDeleteStream(&stream);
-  }
+  isvd_v@x@RngGaussianDriver(seed_, Nl * nb * param.mpi_rank, nj * Nl, omegat, 0.0, 1.0);
 
   // ====================================================================================================================== //
   // Project
@@ -94,7 +76,7 @@ static void sketchBlockCol(
   // ====================================================================================================================== //
   // Rearrange
 
-  MPI_Reduce_scatter_block(yst_, yst, mb*ldyst_, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Reduce_scatter_block(yst_, yst, mb*ldyst_, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Deallocate memory
@@ -105,14 +87,14 @@ static void sketchBlockCol(
 }
 
 static void sketchBlockRow(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-    const isvd_int_t  seed,
-    const mpi_int_t   mpi_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *yst,
+    const isvd_int_t    ldyst,
+    const isvd_int_t    seed,
+    const mpi_int_t     mpi_root
 ) {
 
   // ====================================================================================================================== //
@@ -135,7 +117,7 @@ static void sketchBlockRow(
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *omegat = isvd_@x@malloc(Nl * n);
+  @xtype_____@ *omegat = isvd_@x@malloc(Nl * n);
   isvd_int_t ldomegat = Nl;
 
   // ====================================================================================================================== //
@@ -143,25 +125,7 @@ static void sketchBlockRow(
 
   isvd_int_t seed_ = seed;
   MPI_Bcast(&seed_, sizeof(isvd_VSLStreamStatePtr), MPI_BYTE, mpi_root, param.mpi_comm);
-
-  ISVD_OMP_PARALLEL
-  {
-    omp_int_t omp_size = isvd_getOmpSize();
-    omp_int_t omp_rank = isvd_getOmpRank();
-
-    isvd_int_t len   = n * Nl / omp_size;
-    isvd_int_t start = len * omp_rank;
-    if ( omp_rank == omp_size-1 ) {
-      len = n * Nl - start;
-    }
-
-    isvd_VSLStreamStatePtr stream;
-    isvd_vslNewStream(&stream, seed_);
-    isvd_vslSkipAheadStream(stream, start * 2);
-    isvd_v@x@RngGaussian(stream, len, omegat + start, 0.0, 1.0);
-
-    isvd_vslDeleteStream(&stream);
-  }
+  isvd_v@x@RngGaussianDriver(seed_, 0, n * Nl, omegat, 0.0, 1.0);
 
   // ====================================================================================================================== //
   // Project
diff --git a/src/libisvd/core/stage/@x@_sketch_gaussian_projection.c b/src/lib/libisvd/core/stage/@x@_sketch_gaussian_projection.c
similarity index 74%
rename from src/libisvd/core/stage/@x@_sketch_gaussian_projection.c
rename to src/lib/libisvd/core/stage/@x@_sketch_gaussian_projection.c
index 44feacf..8e9ffc1 100644
--- a/src/libisvd/core/stage/@x@_sketch_gaussian_projection.c
+++ b/src/lib/libisvd/core/stage/@x@_sketch_gaussian_projection.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/core/stage/@x@_sketch_gaussian_projection.c
-/// \brief      The Gaussian Projection Sketching (@xname@ precision)
+/// \file       lib/libisvd/core/stage/@x@_sketch_gaussian_projection.c
+/// \brief      The Gaussian Projection Sketching (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -14,7 +14,7 @@
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_core_@x@_stage_module
-/// \brief  Gaussian Projection Sketching (@xname@ precision)
+/// \brief  Gaussian Projection Sketching (@xname@ precision).
 ///
 /// \param[in]   param       The \ref isvd_Param "parameters".
 /// \param[in]   argv, argc  The arguments and its length. (not using)
@@ -26,9 +26,9 @@
 /// \param[in]   ordera      The storage ordering of 𝑨. <br>
 ///                          `'C'`: column-major ordering. <br>
 ///                          `'R'`: row-major ordering.
-/// \param[in]   a, lda      The column/row-block 𝑨 (\f$m \times n_j\f$) and its leading dimension. <br>
-///                          \b dista = `'C'`: the size must be \f$m \times n_j\f$. <br>
-///                          \b dista = `'R'`: the size must be \f$m_j \times n\f$.
+/// \param[in]   a, lda      The column/row-block 𝑨 (\f$m \times n^{(j)}\f$) and its leading dimension. <br>
+///                          \b dista = `'C'`: the size must be \f$m \times n^{(j)}\f$. <br>
+///                          \b dista = `'R'`: the size must be \f$m^{(j)} \times n\f$.
 /// \param[in]   yst, ldyst  The row-block 𝖄 (\f$m_b \times Nl\f$, row-major) and its leading dimension. <br>
 ///                          \b dista = `'C'`: \b ldyst must be \f$Nl\f$. <br>
 ///                          \b dista = `'R'`: no condition.
@@ -41,34 +41,30 @@
 ///        the routine only returns the first \b retc default arguments in \b retv.
 ///
 void isvd_@x@SketchGaussianProjection(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-    const isvd_int_t  seed,
-    const mpi_int_t   mpi_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *yst,
+    const isvd_int_t    ldyst,
+    const isvd_int_t    seed,
+    const mpi_int_t     mpi_root
 ) {
 
-  if ( argc < 0 ) {
-    return;
-  }
-
-  ISVD_UNUSED(argv);
-  ISVD_UNUSED(retv);
-  ISVD_UNUSED(retc);
+  if ( argc > 0 ) { isvd_assert_ne(argv, nullptr); }
+  if ( retc > 0 ) { isvd_assert_ne(retv, nullptr); }
+  if ( argc < 0 ) return;
 
   // ====================================================================================================================== //
   // Check arguments
 
-  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", nullptr);
-  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", nullptr);
+  const char dista_  = isvd_arg2char("DISTA",  dista,  "CR", NULL);
+  const char ordera_ = isvd_arg2char("ORDERA", ordera, "CR", NULL);
   if ( !dista_ || !ordera_ ) return;
 
   // ====================================================================================================================== //
diff --git a/src/libisvd/def.h b/src/lib/libisvd/def.h
similarity index 81%
rename from src/libisvd/def.h
rename to src/lib/libisvd/def.h
index 1333429..cc33244 100644
--- a/src/libisvd/def.h
+++ b/src/lib/libisvd/def.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/def.h
+/// \file       lib/libisvd/def.h
 /// \brief      The definitions.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -10,55 +10,18 @@
 #define LIBISVD_DEF_H_
 
 #include <isvd/def.h>
-
-#if defined(ISVD_USE_ILP64) && !defined(MKL_ILP64)
-  #define MKL_ILP64
-#else  // ISVD_USE_ILP64
-  #undef MKL_ILP64
-#endif  // ISVD_USE_ILP64
-
-#if defined(_OPENMP)
-  #if !defined(ISVD_USE_OMP)
-    #define ISVD_USE_OMP
-  #endif  // ISVD_USE_OMP
-#else  // _OPENMP
-  #undef ISVD_USE_OMP
-#endif  // _OPENMP
-
-#if defined(ISVD_USE_OMP)
-  #include <omp.h>
-  #define ISVD_OMP_PARALLEL _Pragma("omp parallel")
-#else
-  #define ISVD_OMP_PARALLEL
-#endif  // ISVD_USE_OMP
-
 #include <mpi.h>
 
 #if defined(ISVD_USE_GTEST) && defined(__cplusplus)
   #include <gtest/gtest.h>
 #endif  // ISVD_USE_GTEST
 
-
 #define ISVD_UNUSED( x ) (void)(x)
 
 #if !defined(__cplusplus) || (__cplusplus < 201103L)
-  #define nullptr NULL
+  #define nullptr (void*)(NULL)
 #endif
 
-#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
-#if defined(ISVD_USE_MKL)
-  #define MKL_INT       isvd_int_t
-  #define MKL_UINT      isvd_uint_t
-  #define MKL_Complex8  _Complex float
-  #define MKL_Complex16 _Complex double
-#endif  // ISVD_USE_MKL
-#endif  // DOXYGEN_SHOULD_SKIP_THIS
-
-#if defined(ISVD_USE_MKL)
-  #include <mkl.h>
-#endif  // ISVD_USE_MKL
-
-
 #if defined(__cplusplus)
 extern "C" {
 #endif  // __cplusplus
diff --git a/src/libisvd/gpu.h b/src/lib/libisvd/gpu.h
similarity index 83%
rename from src/libisvd/gpu.h
rename to src/lib/libisvd/gpu.h
index a7e8367..7e137a4 100644
--- a/src/libisvd/gpu.h
+++ b/src/lib/libisvd/gpu.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu.h
+/// \file       lib/libisvd/gpu.h
 /// \brief      The GPU header.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -9,6 +9,4 @@
 #ifndef LIBISVD_GPU_H_
 #define LIBISVD_GPU_H_
 
-#include <libisvd/gpu/def.h>
-
 #endif  // LIBISVD_GPU_H_
diff --git a/src/libisvd/gpu/def.h b/src/lib/libisvd/gpu/def.h
similarity index 94%
rename from src/libisvd/gpu/def.h
rename to src/lib/libisvd/gpu/def.h
index 48b33cd..e6efe21 100644
--- a/src/libisvd/gpu/def.h
+++ b/src/lib/libisvd/gpu/def.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/def.h
+/// \file       lib/libisvd/gpu/def.h
 /// \brief      The GPU definitions.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/gpu/env.c b/src/lib/libisvd/gpu/env.c
similarity index 98%
rename from src/libisvd/gpu/env.c
rename to src/lib/libisvd/gpu/env.c
index 71d8337..687b1b0 100644
--- a/src/libisvd/gpu/env.c
+++ b/src/lib/libisvd/gpu/env.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/env.c
+/// \file       lib/libisvd/gpu/env.c
 /// \brief      The iSVD environment routines (GPU only).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/gpu/stage/@x@_postprocess_gpu.h b/src/lib/libisvd/gpu/stage/@x@_postprocess_gpu.h
similarity index 82%
rename from src/libisvd/gpu/stage/@x@_postprocess_gpu.h
rename to src/lib/libisvd/gpu/stage/@x@_postprocess_gpu.h
index ce11b3a..9120fd6 100644
--- a/src/libisvd/gpu/stage/@x@_postprocess_gpu.h
+++ b/src/lib/libisvd/gpu/stage/@x@_postprocess_gpu.h
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/stage/@x@_postprocess_gpu.h
-/// \brief      The GPU Postprocessing utilities (@xname@ precision)
+/// \file       lib/libisvd/gpu/stage/@x@_postprocess_gpu.h
+/// \brief      The GPU Postprocessing utilities (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -21,21 +21,22 @@
 #include <isvd/la.h>
 #include <libisvd/util/function.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 static void projectBlockCol(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *zt,
-    const isvd_int_t  ldzt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-    const mpi_int_t   ut_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *zt,
+    const isvd_int_t    ldzt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+    const mpi_int_t     ut_root
 ) {
 
   ISVD_UNUSED(s);
@@ -73,21 +74,21 @@ static void projectBlockCol(
   size_t free_byte, total_byte;
   cudaMemGetInfo(&free_byte, &total_byte);
   if ( isvd_gpu_memory_limit > 0 ) free_byte = minl(free_byte, isvd_gpu_memory_limit);
-  size_t melem = free_byte / sizeof(@xtype@);
+  size_t melem = free_byte / sizeof(@xtype_____@);
   size_t nelem_used = m * l;
   isvd_int_t n_gpu = (melem - nelem_used) / (m + l);
   if ( n_gpu > (isvd_int_t)isvd_kBlockSizeGpu ) n_gpu = (n_gpu / isvd_kBlockSizeGpu) * isvd_kBlockSizeGpu;
   n_gpu = min(n_gpu, nj);
   if ( n_gpu <= 0 ) {
     fprintf(stderr, "No enough GPU memory. (Request at least %" PRId64 " bytes. Only %" PRId64 " bytes free.",
-            nelem_used * sizeof(@xtype@), melem * sizeof(@xtype@));
+            nelem_used * sizeof(@xtype_____@), melem * sizeof(@xtype_____@));
     isvd_assert_fail();
   }
 
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *qt_;
+  @xtype_____@ *qt_;
   if ( use_ut ) {
     qt_ = ut;
   } else {
@@ -95,22 +96,22 @@ static void projectBlockCol(
   }
   isvd_int_t ldqt_ = l;
 
-  @xtype@ *a_gpu;
+  @xtype_____@ *a_gpu;
   magma_@x@malloc(&a_gpu, m * n_gpu);
   isvd_int_t lda_gpu = (ordera == 'C') ? m : n_gpu;
 
-  @xtype@ *qt_gpu;
+  @xtype_____@ *qt_gpu;
   magma_@x@malloc(&qt_gpu, l * m);
   isvd_int_t ldqt_gpu = l;
 
-  @xtype@ *zt_gpu;
+  @xtype_____@ *zt_gpu;
   magma_@x@malloc(&zt_gpu, l * n_gpu);
   isvd_int_t ldzt_gpu = l;
 
   // ====================================================================================================================== //
   // Rearrange
 
-  MPI_Allgather(qt, mb*ldqt, MPI_@X_TYPE@, qt_, mb*ldqt, MPI_@X_TYPE@, param.mpi_comm);
+  MPI_Allgather(qt, mb*ldqt, MPI_@XTYPE@, qt_, mb*ldqt, MPI_@XTYPE@, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Send data
@@ -124,8 +125,8 @@ static void projectBlockCol(
   isvd_int_t idx;
 
   for ( idx = 0; idx < nj; idx += n_gpu ) {
-    const @xtype@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
-          @xtype@ *zt_tmp = zt + ldzt*idx;
+    const @xtype_____@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
+          @xtype_____@ *zt_tmp = zt + ldzt*idx;
     const isvd_int_t n_tmp = min(n_gpu, nj-idx);
 
     // Send A
@@ -157,18 +158,18 @@ static void projectBlockCol(
 }
 
 static void projectBlockRow(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *zt,
-    const isvd_int_t  ldzt,
-          @xtype@    *s,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *zt,
+    const isvd_int_t    ldzt,
+          @xtype_____@ *s,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     vt_root
 ) {
 
   ISVD_UNUSED(s);
@@ -205,21 +206,21 @@ static void projectBlockRow(
 
   size_t free_byte, total_byte;
   cudaMemGetInfo(&free_byte, &total_byte);
-  size_t melem = free_byte / sizeof(@xtype@);
+  size_t melem = free_byte / sizeof(@xtype_____@);
   size_t nelem_used = mj * l;
   isvd_int_t n_gpu = (melem - nelem_used) / (mj + l);
   if ( n_gpu > (isvd_int_t)isvd_kBlockSizeGpu ) n_gpu = (n_gpu / isvd_kBlockSizeGpu) * isvd_kBlockSizeGpu;
   n_gpu = min(n_gpu, n);
   if ( n_gpu <= 0 ) {
     fprintf(stderr, "No enough GPU memory. (Request at least %" PRId64 " bytes. Only %" PRId64 " bytes free.",
-            nelem_used * sizeof(@xtype@), melem * sizeof(@xtype@));
+            nelem_used * sizeof(@xtype_____@), melem * sizeof(@xtype_____@));
     isvd_assert_fail();
   }
 
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *zt_;
+  @xtype_____@ *zt_;
   if ( use_vt ) {
     zt_ = vt;
   } else {
@@ -227,15 +228,15 @@ static void projectBlockRow(
   }
   isvd_int_t ldzt_ = l;
 
-  @xtype@ *a_gpu;
+  @xtype_____@ *a_gpu;
   magma_@x@malloc(&a_gpu, mj * n_gpu);
   isvd_int_t lda_gpu = (ordera == 'C') ? mj : n_gpu;
 
-  @xtype@ *qt_gpu;
+  @xtype_____@ *qt_gpu;
   magma_@x@malloc(&qt_gpu, l * mj);
   isvd_int_t ldqt_gpu = l;
 
-  @xtype@ *zt_gpu;
+  @xtype_____@ *zt_gpu;
   magma_@x@malloc(&zt_gpu, l * n_gpu);
   isvd_int_t ldzt_gpu = l;
 
@@ -251,8 +252,8 @@ static void projectBlockRow(
   isvd_int_t idx;
 
   for ( idx = 0; idx < n; idx += n_gpu ) {
-    const @xtype@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
-          @xtype@ *zt_tmp = zt_ + ldzt*idx;
+    const @xtype_____@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
+          @xtype_____@ *zt_tmp = zt_ + ldzt*idx;
     const isvd_int_t n_tmp = min(n_gpu, n-idx);
 
     // Send A
@@ -273,7 +274,7 @@ static void projectBlockRow(
   // ====================================================================================================================== //
   // Rearrange
 
-  MPI_Reduce_scatter_block(zt_, zt, nb*ldzt, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Reduce_scatter_block(zt_, zt, nb*ldzt, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Deallocate memory
diff --git a/src/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c b/src/lib/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
similarity index 89%
rename from src/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
rename to src/lib/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
index 1126374..a8ff4b3 100644
--- a/src/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
+++ b/src/lib/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
-/// \brief      The GPU Gramian Postprocessing (@xname@ precision)
+/// \file       lib/libisvd/gpu/stage/@x@_postprocess_gramian_gpu.c
+/// \brief      The GPU Gramian Postprocessing (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
diff --git a/src/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c b/src/lib/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
similarity index 89%
rename from src/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
rename to src/lib/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
index b2fe752..05fdd82 100644
--- a/src/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
+++ b/src/lib/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
-/// \brief      The GPU Symmetric Postprocessing (@xname@ precision)
+/// \file       lib/libisvd/gpu/stage/@x@_postprocess_symmetric_gpu.c
+/// \brief      The GPU Symmetric Postprocessing (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
diff --git a/src/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c b/src/lib/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
similarity index 89%
rename from src/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
rename to src/lib/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
index adbdbda..82f68d7 100644
--- a/src/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
+++ b/src/lib/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
-/// \brief      The GPU Tall-Skinny QR Postprocessing (@xname@ precision)
+/// \file       lib/libisvd/gpu/stage/@x@_postprocess_tall_skinny_qr_gpu.c
+/// \brief      The GPU Tall-Skinny QR Postprocessing (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
diff --git a/src/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c b/src/lib/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
similarity index 89%
rename from src/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
rename to src/lib/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
index 8f476da..81b51ac 100644
--- a/src/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
+++ b/src/lib/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
-/// \brief      The GPU Gaussian Projection Sketching (@xname@ precision)
+/// \file       lib/libisvd/gpu/stage/@x@_sketch_gaussian_projection_gpu.c
+/// \brief      The GPU Gaussian Projection Sketching (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
diff --git a/src/libisvd/gpu/stage/@x@_sketch_gpu.h b/src/lib/libisvd/gpu/stage/@x@_sketch_gpu.h
similarity index 74%
rename from src/libisvd/gpu/stage/@x@_sketch_gpu.h
rename to src/lib/libisvd/gpu/stage/@x@_sketch_gpu.h
index 199ce27..10bc0a4 100644
--- a/src/libisvd/gpu/stage/@x@_sketch_gpu.h
+++ b/src/lib/libisvd/gpu/stage/@x@_sketch_gpu.h
@@ -1,6 +1,6 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/gpu/stage/@x@_sketch_gpu.h
-/// \brief      The GPU Sketching utilities (@xname@ precision)
+/// \file       lib/libisvd/gpu/stage/@x@_sketch_gpu.h
+/// \brief      The GPU Sketching utilities (@xname@ precision).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
 /// \copyright  MIT License
@@ -21,18 +21,19 @@
 #include <isvd/la.h>
 #include <libisvd/util/function.h>
 #include <isvd/util/memory.h>
+#include <isvd/util/mpi.h>
 #include <isvd/util/omp.h>
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 static void sketchBlockCol(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-    const isvd_int_t  seed,
-    const mpi_int_t   mpi_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *yst,
+    const isvd_int_t    ldyst,
+    const isvd_int_t    seed,
+    const mpi_int_t     mpi_root
 ) {
 
   ISVD_UNUSED(ldyst);
@@ -63,35 +64,35 @@ static void sketchBlockCol(
   size_t free_byte, total_byte;
   cudaMemGetInfo(&free_byte, &total_byte);
   if ( isvd_gpu_memory_limit > 0 ) free_byte = minl(free_byte, isvd_gpu_memory_limit);
-  size_t melem = free_byte / sizeof(@xtype@);
+  size_t melem = free_byte / sizeof(@xtype_____@);
   size_t nelem_used = m * Nl;
   isvd_int_t n_gpu = (melem - nelem_used) / (m + Nl);
   if ( n_gpu > (isvd_int_t)isvd_kBlockSizeGpu ) n_gpu = (n_gpu / isvd_kBlockSizeGpu) * isvd_kBlockSizeGpu;
   n_gpu = min(n_gpu, nj);
   if ( n_gpu <= 0 ) {
     fprintf(stderr, "No enough GPU memory. (Request at least %" PRId64 " bytes. Only %" PRId64 " bytes free.",
-            nelem_used * sizeof(@xtype@), melem * sizeof(@xtype@));
+            nelem_used * sizeof(@xtype_____@), melem * sizeof(@xtype_____@));
     isvd_assert_fail();
   }
 
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *omegat = isvd_@x@malloc(Nl * nj);
+  @xtype_____@ *omegat = isvd_@x@malloc(Nl * nj);
   isvd_int_t ldomegat = Nl;
 
-  @xtype@ *yst_ = isvd_@x@malloc(Nl * Pmb);
+  @xtype_____@ *yst_ = isvd_@x@malloc(Nl * Pmb);
   isvd_int_t ldyst_ = Nl;
 
-  @xtype@ *a_gpu;
+  @xtype_____@ *a_gpu;
   magma_@x@malloc(&a_gpu, m * n_gpu);
   isvd_int_t lda_gpu = (ordera == 'C') ? m : n_gpu;
 
-  @xtype@ *omegat_gpu;
+  @xtype_____@ *omegat_gpu;
   magma_@x@malloc(&omegat_gpu, Nl * n_gpu);
   isvd_int_t ldomegat_gpu = Nl;
 
-  @xtype@ *yst_gpu;
+  @xtype_____@ *yst_gpu;
   magma_@x@malloc(&yst_gpu, Nl * m);
   isvd_int_t ldyst_gpu = Nl;
 
@@ -100,36 +101,18 @@ static void sketchBlockCol(
 
   isvd_int_t seed_ = seed;
   MPI_Bcast(&seed_, sizeof(seed_), MPI_BYTE, mpi_root, param.mpi_comm);
-
-  ISVD_OMP_PARALLEL
-  {
-    omp_int_t omp_size = isvd_getOmpSize();
-    omp_int_t omp_rank = isvd_getOmpRank();
-
-    isvd_int_t len   = nj * Nl / omp_size;
-    isvd_int_t start = len * omp_rank;
-    if ( omp_rank == omp_size-1 ) {
-      len = nj * Nl - start;
-    }
-
-    isvd_VSLStreamStatePtr stream;
-    isvd_vslNewStream(&stream, seed_);
-    isvd_vslSkipAheadStream(stream, (Nl * nb * param.mpi_rank + start) * 2);
-    isvd_v@x@RngGaussian(stream, len, omegat + start, 0.0, 1.0);
-
-    isvd_vslDeleteStream(&stream);
-  }
+  isvd_v@x@RngGaussianDriver(seed_, Nl * nb * param.mpi_rank, nj * Nl, omegat, 0.0, 1.0);
 
   // ====================================================================================================================== //
   // Project
 
-  cudaMemset(yst_gpu, 0, ldyst_gpu * m * sizeof(@xtype@));
+  cudaMemset(yst_gpu, 0, ldyst_gpu * m * sizeof(@xtype_____@));
   char transa_ = (ordera == 'C') ? 'T' : 'N';
   isvd_int_t idx;
 
   for ( idx = 0; idx < nj; idx += n_gpu ) {
-    const @xtype@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
-    const @xtype@ *omegat_tmp = omegat + ldomegat*idx;
+    const @xtype_____@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
+    const @xtype_____@ *omegat_tmp = omegat + ldomegat*idx;
     const isvd_int_t n_tmp = min(n_gpu, nj-idx);
 
     // Send A and Omega
@@ -153,12 +136,13 @@ static void sketchBlockCol(
   // ====================================================================================================================== //
   // Rearrange
 
-  MPI_Reduce_scatter_block(yst_, yst, mb*ldyst_, MPI_@X_TYPE@, MPI_SUM, param.mpi_comm);
+  MPI_Reduce_scatter_block(yst_, yst, mb*ldyst_, MPI_@XTYPE@, MPI_SUM, param.mpi_comm);
 
   // ====================================================================================================================== //
   // Deallocate memory
 
   isvd_free(omegat);
+  isvd_free(yst_);
   magma_free(a_gpu);
   magma_free(omegat_gpu);
   magma_free(yst_gpu);
@@ -166,14 +150,14 @@ static void sketchBlockCol(
 }
 
 static void sketchBlockRow(
-    const isvd_Param  param,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-    const isvd_int_t  seed,
-    const mpi_int_t   mpi_root
+    const isvd_Param    param,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *yst,
+    const isvd_int_t    ldyst,
+    const isvd_int_t    seed,
+    const mpi_int_t     mpi_root
 ) {
 
   // ====================================================================================================================== //
@@ -198,32 +182,32 @@ static void sketchBlockRow(
 
   size_t free_byte, total_byte;
   cudaMemGetInfo(&free_byte, &total_byte);
-  size_t melem = free_byte / sizeof(@xtype@);
+  size_t melem = free_byte / sizeof(@xtype_____@);
   size_t nelem_used = mj * Nl;
   isvd_int_t n_gpu = (melem - nelem_used) / (mj + Nl);
   if ( n_gpu > (isvd_int_t)isvd_kBlockSizeGpu ) n_gpu = (n_gpu / isvd_kBlockSizeGpu) * isvd_kBlockSizeGpu;
   n_gpu = min(n_gpu, n);
   if ( n_gpu <= 0 ) {
     fprintf(stderr, "No enough GPU memory. (Request at least %" PRId64 " bytes. Only %" PRId64 " bytes free.",
-            nelem_used * sizeof(@xtype@), melem * sizeof(@xtype@));
+            nelem_used * sizeof(@xtype_____@), melem * sizeof(@xtype_____@));
     isvd_assert_fail();
   }
 
   // ====================================================================================================================== //
   // Allocate memory
 
-  @xtype@ *omegat = isvd_@x@malloc(n * Nl);
+  @xtype_____@ *omegat = isvd_@x@malloc(n * Nl);
   isvd_int_t ldomegat = Nl;
 
-  @xtype@ *a_gpu;
+  @xtype_____@ *a_gpu;
   magma_@x@malloc(&a_gpu, mj * n_gpu);
   isvd_int_t lda_gpu = (ordera == 'C') ? mj : n_gpu;
 
-  @xtype@ *omegat_gpu;
+  @xtype_____@ *omegat_gpu;
   magma_@x@malloc(&omegat_gpu, Nl * n_gpu);
   isvd_int_t ldomegat_gpu = Nl;
 
-  @xtype@ *yst_gpu;
+  @xtype_____@ *yst_gpu;
   magma_@x@malloc(&yst_gpu, Nl * mj);
   isvd_int_t ldyst_gpu = Nl;
 
@@ -232,36 +216,18 @@ static void sketchBlockRow(
 
   isvd_int_t seed_ = seed;
   MPI_Bcast(&seed_, sizeof(isvd_VSLStreamStatePtr), MPI_BYTE, mpi_root, param.mpi_comm);
-
-  ISVD_OMP_PARALLEL
-  {
-    omp_int_t omp_size = isvd_getOmpSize();
-    omp_int_t omp_rank = isvd_getOmpRank();
-
-    isvd_int_t len   = n * Nl / omp_size;
-    isvd_int_t start = len * omp_rank;
-    if ( omp_rank == omp_size-1 ) {
-      len = n * Nl - start;
-    }
-
-    isvd_VSLStreamStatePtr stream;
-    isvd_vslNewStream(&stream, seed_);
-    isvd_vslSkipAheadStream(stream, start * 2);
-    isvd_v@x@RngGaussian(stream, len, omegat + start, 0.0, 1.0);
-
-    isvd_vslDeleteStream(&stream);
-  }
+  isvd_v@x@RngGaussianDriver(seed_, 0, n * Nl, omegat, 0.0, 1.0);
 
   // ====================================================================================================================== //
   // Project
 
-  cudaMemset(yst_gpu, 0, ldyst_gpu * mj * sizeof(@xtype@));
+  cudaMemset(yst_gpu, 0, ldyst_gpu * mj * sizeof(@xtype_____@));
   char transa_ = (ordera == 'C') ? 'T' : 'N';
   isvd_int_t idx;
 
   for ( idx = 0; idx < n; idx += n_gpu ) {
-    const @xtype@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
-    const @xtype@ *omegat_tmp = omegat + ldomegat*idx;
+    const @xtype_____@ *a_tmp = (ordera == 'C') ? (a + lda*idx) : (a+idx);
+    const @xtype_____@ *omegat_tmp = omegat + ldomegat*idx;
     const isvd_int_t n_tmp = min(n_gpu, n-idx);
 
     // Send A and Omega
diff --git a/src/lib/libisvd/la.h b/src/lib/libisvd/la.h
new file mode 100644
index 0000000..6d0e7f9
--- /dev/null
+++ b/src/lib/libisvd/la.h
@@ -0,0 +1,12 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       lib/libisvd/la.h
+/// \brief      The linear algebra header.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#ifndef LIBISVD_LA_H_
+#define LIBISVD_LA_H_
+
+#endif  // LIBISVD_LA_H_
diff --git a/src/libisvd/la/blas/dimm.c b/src/lib/libisvd/la/blas/dimm.c
similarity index 89%
rename from src/libisvd/la/blas/dimm.c
rename to src/lib/libisvd/la/blas/dimm.c
index 9c149e3..027b903 100644
--- a/src/libisvd/la/blas/dimm.c
+++ b/src/lib/libisvd/la/blas/dimm.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/blas/dimm.c
+/// \file       lib/libisvd/la/blas/dimm.c
 /// \brief      The BLAS-Like Dimm routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,14 +7,14 @@
 ///
 
 #include <isvd/la/blas/blas_like.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <isvd/la/vml/mathematical.h>
 #include <libisvd/util/arg.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #define isvd_xDimm( side, m, n, alpha, a, b, ldb, xScal, xMul ) \
-  const char side_ = isvd_arg2char("SIDE", side, "LR", nullptr); \
+  const char side_ = isvd_arg2char("SIDE", side, "LR", NULL); \
   if ( !side_ ) return; \
   switch ( side_ ) { \
     case 'L': { \
@@ -48,4 +48,4 @@ void isvd_zDimm(
     const CHAR1 side, const INT m, const INT n, const COMP8 alpha, const COMP8 *a, COMP8 *b, const INT ldb
 ) { isvd_xDimm(side, m, n, alpha, a, b, ldb, isvd_zScal, isvd_vzMul); }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/blas/dism.c b/src/lib/libisvd/la/blas/dism.c
similarity index 89%
rename from src/libisvd/la/blas/dism.c
rename to src/lib/libisvd/la/blas/dism.c
index 8edca0e..36c34d6 100644
--- a/src/libisvd/la/blas/dism.c
+++ b/src/lib/libisvd/la/blas/dism.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/blas/dism.c
+/// \file       lib/libisvd/la/blas/dism.c
 /// \brief      The BLAS-Like Dism routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,14 +7,14 @@
 ///
 
 #include <isvd/la/blas/blas_like.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <isvd/la/vml/power_root.h>
 #include <libisvd/util/arg.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #define isvd_xDism( side, m, n, alpha, a, b, ldb, xScal, xDiv ) \
-  const char side_ = isvd_arg2char("SIDE", side, "LR", nullptr); \
+  const char side_ = isvd_arg2char("SIDE", side, "LR", NULL); \
   if ( !side_ ) return; \
   switch ( side_ ) { \
     case 'L': { \
@@ -48,4 +48,4 @@ void isvd_zDism(
     const CHAR1 side, const INT m, const INT n, const COMP8 alpha, const COMP8 *a, COMP8 *b, const INT ldb
 ) { isvd_xDism(side, m, n, alpha, a, b, ldb, isvd_zScal, isvd_vzDiv); }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/blas/gemmt.c b/src/lib/libisvd/la/blas/gemmt.c
similarity index 96%
rename from src/libisvd/la/blas/gemmt.c
rename to src/lib/libisvd/la/blas/gemmt.c
index b7011af..00fbec8 100644
--- a/src/libisvd/la/blas/gemmt.c
+++ b/src/lib/libisvd/la/blas/gemmt.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/blas/gemmt.c
+/// \file       lib/libisvd/la/blas/gemmt.c
 /// \brief      The BLAS-Like Gemmt routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,10 +7,10 @@
 ///
 
 #include <isvd/la/blas/blas_like.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <libisvd/util/arg.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -71,4 +71,4 @@ void isvd_zGemmt(
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/blas/iamax.c b/src/lib/libisvd/la/blas/iamax.c
similarity index 92%
rename from src/libisvd/la/blas/iamax.c
rename to src/lib/libisvd/la/blas/iamax.c
index eb49c68..254ed08 100644
--- a/src/libisvd/la/blas/iamax.c
+++ b/src/lib/libisvd/la/blas/iamax.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/blas/iamax.c
+/// \file       lib/libisvd/la/blas/iamax.c
 /// \brief      The BLAS-1 iAmax routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/blas/blas1.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(__cplusplus)
 extern "C" {
@@ -38,4 +38,4 @@ REAL8 isvd_dAmax( const INT n, const REAL8 *x, const INT incx ) { INT i = isvd_i
 COMP4 isvd_cAmax( const INT n, const COMP4 *x, const INT incx ) { INT i = isvd_icAmax(n, x, incx); return cabsf(x[i]); }
 COMP8 isvd_zAmax( const INT n, const COMP8 *x, const INT incx ) { INT i = isvd_izAmax(n, x, incx); return cabs(x[i]); }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/blas/iamin.c b/src/lib/libisvd/la/blas/iamin.c
similarity index 95%
rename from src/libisvd/la/blas/iamin.c
rename to src/lib/libisvd/la/blas/iamin.c
index 26a8994..158543c 100644
--- a/src/libisvd/la/blas/iamin.c
+++ b/src/lib/libisvd/la/blas/iamin.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/blas/iamin.c
+/// \file       lib/libisvd/la/blas/iamin.c
 /// \brief      The BLAS-1 iAmin routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/blas/blas1.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -62,4 +62,4 @@ REAL8 isvd_dAmin( const INT n, const REAL8 *x, const INT incx ) { INT i = isvd_i
 COMP4 isvd_cAmin( const INT n, const COMP4 *x, const INT incx ) { INT i = isvd_icAmin(n, x, incx); return cabsf(x[i]); }
 COMP8 isvd_zAmin( const INT n, const COMP8 *x, const INT incx ) { INT i = isvd_izAmin(n, x, incx); return cabs(x[i]); }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/blas/omatcopy.c b/src/lib/libisvd/la/blas/omatcopy.c
similarity index 91%
rename from src/libisvd/la/blas/omatcopy.c
rename to src/lib/libisvd/la/blas/omatcopy.c
index 502ebfe..9d07df0 100644
--- a/src/libisvd/la/blas/omatcopy.c
+++ b/src/lib/libisvd/la/blas/omatcopy.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/blas/omatcopy.c
+/// \file       lib/libisvd/la/blas/omatcopy.c
 /// \brief      The BLAS-Like Omatcopy routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,10 +7,10 @@
 ///
 
 #include <isvd/la/blas/blas_like.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <libisvd/util/arg.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -46,11 +46,11 @@ void isvd_zOmatcopy(
 
 #else  // ISVD_USE_MKL
 
-static inline float  rconjf( const float  z ) { return z; }
-static inline double rconj(  const double z ) { return z; }
+static inline REAL4 rconjf( const REAL4 z ) { return z; }
+static inline REAL8 rconj(  const REAL8 z ) { return z; }
 
 #define isvd_xOmatcopy( trans, m, n, alpha, a, lda, b, ldb, conj ) \
-  const char trans_ = isvd_arg2char("TRANS", trans, "NTRC", nullptr); \
+  const char trans_ = isvd_arg2char("TRANS", trans, "NTRC", NULL); \
   if ( !trans_ ) return; \
   switch ( trans_ ) { \
     case 'N': { \
@@ -105,4 +105,4 @@ void isvd_zOmatcopy(
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/lib/libisvd/la/def.h b/src/lib/libisvd/la/def.h
new file mode 100644
index 0000000..917a707
--- /dev/null
+++ b/src/lib/libisvd/la/def.h
@@ -0,0 +1,58 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       lib/libisvd/la/def.h
+/// \brief      The linear algebra definitions.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#ifndef LIBISVD_LA_DEF_H_
+#define LIBISVD_LA_DEF_H_
+
+#include <libisvd/def.h>
+
+#if defined(__cplusplus)
+extern "C" {
+#endif  // __cplusplus
+
+#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
+
+#if defined(_OPENMP)
+  #if !defined(ISVD_USE_OMP)
+    #define ISVD_USE_OMP
+  #endif  // ISVD_USE_OMP
+#else  // _OPENMP
+  #undef ISVD_USE_OMP
+#endif  // _OPENMP
+
+#if defined(ISVD_USE_OMP)
+  #include <omp.h>
+  #define ISVD_OMP_PARALLEL _Pragma("omp parallel")
+#else
+  #define ISVD_OMP_PARALLEL
+#endif  // ISVD_USE_OMP
+
+#if defined(ISVD_USE_ILP64) && !defined(MKL_ILP64)
+  #define MKL_ILP64
+#else  // ISVD_USE_ILP64
+  #undef MKL_ILP64
+#endif  // ISVD_USE_ILP64
+
+#if defined(ISVD_USE_MKL)
+  #define MKL_INT       isvd_int_t
+  #define MKL_UINT      isvd_uint_t
+  #define MKL_Complex8  isvd_c_val_t
+  #define MKL_Complex16 isvd_z_val_t
+#endif  // ISVD_USE_MKL
+
+#endif  // DOXYGEN_SHOULD_SKIP_THIS
+
+#if defined(ISVD_USE_MKL)
+  #include <mkl.h>
+#endif  // ISVD_USE_MKL
+
+#if defined(__cplusplus)
+}
+#endif  // __cplusplus
+
+#endif  // LIBISVD_LA_DEF_H_
diff --git a/src/libisvd/la/lapack/geinv.c b/src/lib/libisvd/la/lapack/geinv.c
similarity index 84%
rename from src/libisvd/la/lapack/geinv.c
rename to src/lib/libisvd/la/lapack/geinv.c
index 0b1f670..5050a11 100644
--- a/src/libisvd/la/lapack/geinv.c
+++ b/src/lib/libisvd/la/lapack/geinv.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/lapack/geinv.c
+/// \file       lib/libisvd/la/lapack/geinv.c
 /// \brief      The LAPACK Getrf+Getri routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,10 +7,10 @@
 ///
 
 #include <isvd/la/lapack/linear_equation.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <isvd/util/memory.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -42,7 +42,7 @@ void isvd_sGeinv(
     const INT n, REAL4 *a, const INT lda
 ) {
   REAL4 qwork; INT lwork = -1, info;
-  sgetri_(&n, a, &lda, nullptr, &qwork, &lwork, &info); isvd_assert_pass(info);
+  sgetri_(&n, a, &lda, NULL, &qwork, &lwork, &info); isvd_assert_pass(info);
   lwork = qwork;
   INT *ipiv = isvd_imalloc(n);
   REAL4 *work = isvd_smalloc(lwork);
@@ -55,7 +55,7 @@ void isvd_dGeinv(
     const INT n, REAL8 *a, const INT lda
 ) {
   REAL8 qwork; INT lwork = -1, info;
-  dgetri_(&n, a, &lda, nullptr, &qwork, &lwork, &info); isvd_assert_pass(info);
+  dgetri_(&n, a, &lda, NULL, &qwork, &lwork, &info); isvd_assert_pass(info);
   lwork = qwork;
   INT *ipiv = isvd_imalloc(n);
   REAL8 *work = isvd_dmalloc(lwork);
@@ -68,7 +68,7 @@ void isvd_cGeinv(
     const INT n, COMP4 *a, const INT lda
 ) {
   COMP4 qwork; INT lwork = -1, info;
-  cgetri_(&n, a, &lda, nullptr, &qwork, &lwork, &info); isvd_assert_pass(info);
+  cgetri_(&n, a, &lda, NULL, &qwork, &lwork, &info); isvd_assert_pass(info);
   lwork = creal(qwork);
   INT *ipiv = isvd_imalloc(n);
   COMP4 *work  = isvd_cmalloc(lwork);
@@ -81,7 +81,7 @@ void isvd_zGeinv(
     const INT n, COMP8 *a, const INT lda
 ) {
   COMP8 qwork; INT lwork = -1, info;
-  zgetri_(&n, a, &lda, nullptr, &qwork, &lwork, &info); isvd_assert_pass(info);
+  zgetri_(&n, a, &lda, NULL, &qwork, &lwork, &info); isvd_assert_pass(info);
   lwork = creal(qwork);
   INT *ipiv = isvd_imalloc(n);
   COMP8 *work  = isvd_zmalloc(lwork);
@@ -91,4 +91,4 @@ void isvd_zGeinv(
   isvd_free(work);
 }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/lapack/gesvd.c b/src/lib/libisvd/la/lapack/gesvd.c
similarity index 92%
rename from src/libisvd/la/lapack/gesvd.c
rename to src/lib/libisvd/la/lapack/gesvd.c
index b44c16f..11dad35 100644
--- a/src/libisvd/la/lapack/gesvd.c
+++ b/src/lib/libisvd/la/lapack/gesvd.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/lapack/gesvd.c
+/// \file       lib/libisvd/la/lapack/gesvd.c
 /// \brief      The LAPACK Gesvd routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,11 +7,11 @@
 ///
 
 #include <isvd/la/lapack/least_square.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <isvd/util/memory.h>
 #include <libisvd/util/function.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -61,7 +61,7 @@ void isvd_cGesvd(
     COMP4 *v, const INT ldvt
 ) {
   COMP4 qwork; INT lwork = -1, info;
-  cgesvd_(&jobu, &jobvt, &m, &n, a, &lda, s, u, &ldu, v, &ldvt, &qwork, &lwork, nullptr, &info); isvd_assert_pass(info);
+  cgesvd_(&jobu, &jobvt, &m, &n, a, &lda, s, u, &ldu, v, &ldvt, &qwork, &lwork, NULL, &info); isvd_assert_pass(info);
   lwork = crealf(qwork);
   COMP4 *work  = isvd_cmalloc(lwork);
   REAL4 *rwork = isvd_smalloc(5*min(m, n));
@@ -74,7 +74,7 @@ void isvd_zGesvd(
     COMP8 *v, const INT ldvt
 ) {
   COMP8 qwork; INT lwork = -1, info;
-  zgesvd_(&jobu, &jobvt, &m, &n, a, &lda, s, u, &ldu, v, &ldvt, &qwork, &lwork, nullptr, &info); isvd_assert_pass(info);
+  zgesvd_(&jobu, &jobvt, &m, &n, a, &lda, s, u, &ldu, v, &ldvt, &qwork, &lwork, NULL, &info); isvd_assert_pass(info);
   lwork = creal(qwork);
   COMP8 *work  = isvd_zmalloc(lwork);
   REAL8 *rwork = isvd_dmalloc(5*min(m, n));
@@ -84,4 +84,4 @@ void isvd_zGesvd(
 }
 //\}
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/lapack/lsame.c b/src/lib/libisvd/la/lapack/lsame.c
similarity index 88%
rename from src/libisvd/la/lapack/lsame.c
rename to src/lib/libisvd/la/lapack/lsame.c
index 169b964..af68d7b 100644
--- a/src/libisvd/la/lapack/lsame.c
+++ b/src/lib/libisvd/la/lapack/lsame.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/lapack/lsame.c
+/// \file       lib/libisvd/la/lapack/lsame.c
 /// \brief      The LAPACK Lsame routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/lapack/auxiliary.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -49,4 +49,4 @@ bool isvd_Lsamen(
   return true;
 }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/lapack/syev.c b/src/lib/libisvd/la/lapack/syev.c
similarity index 87%
rename from src/libisvd/la/lapack/syev.c
rename to src/lib/libisvd/la/lapack/syev.c
index 4ba8573..2ed9691 100644
--- a/src/libisvd/la/lapack/syev.c
+++ b/src/lib/libisvd/la/lapack/syev.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/lapack/syev.c
+/// \file       lib/libisvd/la/lapack/syev.c
 /// \brief      The LAPACK Syev routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,10 +7,10 @@
 ///
 
 #include <isvd/la/lapack/least_square.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <isvd/util/memory.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if !defined(DOXYGEN_SHOULD_SKIP_THIS)
 
@@ -57,7 +57,7 @@ void isvd_cSyev(
     const CHAR1 jobz, const CHAR1 uplo, const INT n, COMP4 *a, const INT lda, REAL4 *w
 ) {
   COMP4 qwork; INT lwork = -1, info;
-  cheev_(&jobz, &uplo, &n, a, &lda, w, &qwork, &lwork, nullptr, &info); isvd_assert_pass(info);
+  cheev_(&jobz, &uplo, &n, a, &lda, w, &qwork, &lwork, NULL, &info); isvd_assert_pass(info);
   lwork = creal(qwork);
   COMP4 *work  = isvd_cmalloc(lwork);
   REAL4 *rwork = isvd_smalloc(5*n-2);
@@ -69,7 +69,7 @@ void isvd_zSyev(
     const CHAR1 jobz, const CHAR1 uplo, const INT n, COMP8 *a, const INT lda, REAL8 *w
 ) {
   COMP8 qwork; INT lwork = -1, info;
-  zheev_(&jobz, &uplo, &n, a, &lda, w, &qwork, &lwork, nullptr, &info); isvd_assert_pass(info);
+  zheev_(&jobz, &uplo, &n, a, &lda, w, &qwork, &lwork, NULL, &info); isvd_assert_pass(info);
   lwork = creal(qwork);
   COMP8 *work  = isvd_zmalloc(lwork);
   REAL8 *rwork = isvd_dmalloc(5*n-2);
@@ -78,4 +78,4 @@ void isvd_zSyev(
   isvd_free(rwork);
 }
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/util/memory.c b/src/lib/libisvd/la/util/memory.c
similarity index 57%
rename from src/libisvd/util/memory.c
rename to src/lib/libisvd/la/util/memory.c
index a8f8eea..bdba571 100644
--- a/src/libisvd/util/memory.c
+++ b/src/lib/libisvd/la/util/memory.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/util/memory.c
+/// \file       lib/libisvd/la/util/memory.c
 /// \brief      The memory utilities.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,10 +7,12 @@
 ///
 
 #include <isvd/util/memory.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <stdlib.h>
 #include <string.h>
 
+@ISVD_TYPE_MACRO_DEFINE@
+
 #if defined(ISVD_USE_MKL)
   #define isvd_xmalloc( num, type ) (type*)(mkl_malloc(num * sizeof(type), 64));
 #else  // ISVD_USE_MKL
@@ -26,24 +28,24 @@
 /// \return      The pointer to the array.
 ///
 //\{
-isvd_int_t* isvd_imalloc( const size_t num ) {
-  return isvd_xmalloc(num, isvd_int_t);
+INT*   isvd_imalloc( const size_t num ) {
+  return isvd_xmalloc(num, INT);
 }
 
-float* isvd_smalloc( const size_t num ) {
-  return isvd_xmalloc(num, float);
+REAL4* isvd_smalloc( const size_t num ) {
+  return isvd_xmalloc(num, REAL4);
 }
 
-double* isvd_dmalloc( const size_t num ) {
-  return isvd_xmalloc(num, double);
+REAL8* isvd_dmalloc( const size_t num ) {
+  return isvd_xmalloc(num, REAL8);
 }
 
-_Complex float* isvd_cmalloc( const size_t num ) {
-  return isvd_xmalloc(num, _Complex float);
+COMP4* isvd_cmalloc( const size_t num ) {
+  return isvd_xmalloc(num, COMP4);
 }
 
-_Complex double* isvd_zmalloc( const size_t num ) {
-  return isvd_xmalloc(num, _Complex double);
+COMP8* isvd_zmalloc( const size_t num ) {
+  return isvd_xmalloc(num, COMP8);
 }
 //\}
 
@@ -73,24 +75,24 @@ void isvd_free( void *ptr ) {
 /// \param  num  The number of objects.
 ///
 //\{
-void isvd_imemset0( isvd_int_t *ptr, const size_t num ) {
-  isvd_xmemset0(ptr, num, isvd_int_t);
+void isvd_imemset0( INT   *ptr, const size_t num ) {
+  isvd_xmemset0(ptr, num, INT);
 }
 
-void isvd_smemset0( float *ptr, const size_t num ) {
-  isvd_xmemset0(ptr, num, float);
+void isvd_smemset0( REAL4 *ptr, const size_t num ) {
+  isvd_xmemset0(ptr, num, REAL4);
 }
 
-void isvd_dmemset0( double *ptr, const size_t num ) {
-  isvd_xmemset0(ptr, num, double);
+void isvd_dmemset0( REAL8 *ptr, const size_t num ) {
+  isvd_xmemset0(ptr, num, REAL8);
 }
 
-void isvd_cmemset0( _Complex float *ptr, const size_t num ) {
-  isvd_xmemset0(ptr, num, _Complex float);
+void isvd_cmemset0( COMP4 *ptr, const size_t num ) {
+  isvd_xmemset0(ptr, num, COMP4);
 }
 
-void isvd_zmemset0( _Complex double *ptr, const size_t num ) {
-  isvd_xmemset0(ptr, num, _Complex double);
+void isvd_zmemset0( COMP8 *ptr, const size_t num ) {
+  isvd_xmemset0(ptr, num, COMP8);
 }
 //\}
 
@@ -105,23 +107,25 @@ void isvd_zmemset0( _Complex double *ptr, const size_t num ) {
 /// \param  num  The number of objects.
 ///
 //\{
-void isvd_imemcpy( isvd_int_t *dst, const isvd_int_t *src, const size_t num ) {
-  isvd_xmemcpy(dst, src, num, isvd_int_t);
+void isvd_imemcpy( INT   *dst, const INT   *src, const size_t num ) {
+  isvd_xmemcpy(dst, src, num, INT);
 }
 
-void isvd_smemcpy( float *dst, const float *src, const size_t num ) {
-  isvd_xmemcpy(dst, src, num, float);
+void isvd_smemcpy( REAL4 *dst, const REAL4 *src, const size_t num ) {
+  isvd_xmemcpy(dst, src, num, REAL4);
 }
 
-void isvd_dmemcpy( double *dst, const double *src, const size_t num ) {
-  isvd_xmemcpy(dst, src, num, double);
+void isvd_dmemcpy( REAL8 *dst, const REAL8 *src, const size_t num ) {
+  isvd_xmemcpy(dst, src, num, REAL8);
 }
 
-void isvd_cmemcpy( _Complex float *dst, const _Complex float *src, const size_t num ) {
-  isvd_xmemcpy(dst, src, num, _Complex float);
+void isvd_cmemcpy( COMP4 *dst, const COMP4 *src, const size_t num ) {
+  isvd_xmemcpy(dst, src, num, COMP4);
 }
 
-void isvd_zmemcpy( _Complex double *dst, const _Complex double *src, const size_t num ) {
-  isvd_xmemcpy(dst, src, num, _Complex double);
+void isvd_zmemcpy( COMP8 *dst, const COMP8 *src, const size_t num ) {
+  isvd_xmemcpy(dst, src, num, COMP8);
 }
 //\}
+
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/util/omp.c b/src/lib/libisvd/la/util/omp.c
similarity index 61%
rename from src/libisvd/util/omp.c
rename to src/lib/libisvd/la/util/omp.c
index b6cc124..0423710 100644
--- a/src/libisvd/util/omp.c
+++ b/src/lib/libisvd/la/util/omp.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       include/c/isvd/util/omp.h
+/// \file       lib/libisvd/la/util/omp.c
 /// \brief      The OpenMP utilities.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,7 +7,28 @@
 ///
 
 #include <isvd/util/omp.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
+
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \ingroup  c_util_omp_module
+/// \brief  Returns the maximum number of threads
+///
+/// \return  The maximum number of threads.
+///
+/// \attention    This routine should be called outside of an OpenMP scope.
+///
+omp_int_t isvd_getOmpMaxSize( void ) {
+  omp_int_t omp_size = 1;
+
+  ISVD_OMP_PARALLEL
+  {
+    if ( isvd_getOmpRank() == 0 ) {
+      omp_size = isvd_getOmpSize();
+    }
+  }
+
+  return omp_size;
+}
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_util_omp_module
@@ -15,6 +36,8 @@
 ///
 /// \return  The number of threads in the current team.
 ///
+/// \attention    This routine should be called within an OpenMP scope.
+///
 omp_int_t isvd_getOmpSize( void ) {
 #if defined(ISVD_USE_OMP)
   return omp_get_num_threads();
@@ -29,6 +52,8 @@ omp_int_t isvd_getOmpSize( void ) {
 ///
 /// \return  The thread identification number within the current team.
 ///
+/// \attention    This routine should be called within an OpenMP scope.
+///
 omp_int_t isvd_getOmpRank( void ) {
 #if defined(ISVD_USE_OMP)
   return omp_get_thread_num();
diff --git a/src/libisvd/la/vml/div.c b/src/lib/libisvd/la/vml/div.c
similarity index 91%
rename from src/libisvd/la/vml/div.c
rename to src/lib/libisvd/la/vml/div.c
index 02a07cd..f8bc043 100644
--- a/src/libisvd/la/vml/div.c
+++ b/src/lib/libisvd/la/vml/div.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/vml/div.c
+/// \file       lib/libisvd/la/vml/div.c
 /// \brief      The VML Div routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/vml/power_root.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -32,4 +32,4 @@ void isvd_vzDiv( const INT n, const COMP8 *a, const COMP8 *b, COMP8 *y ) { isvd_
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/vml/mul.c b/src/lib/libisvd/la/vml/mul.c
similarity index 91%
rename from src/libisvd/la/vml/mul.c
rename to src/lib/libisvd/la/vml/mul.c
index 9d0bf89..5ab7fed 100644
--- a/src/libisvd/la/vml/mul.c
+++ b/src/lib/libisvd/la/vml/mul.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/vml/mul.c
+/// \file       lib/libisvd/la/vml/mul.c
 /// \brief      The VML Mul routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/vml/mathematical.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -32,4 +32,4 @@ void isvd_vzMul( const INT n, const COMP8 *a, const COMP8 *b, COMP8 *y ) { isvd_
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/vml/sqrt.c b/src/lib/libisvd/la/vml/sqrt.c
similarity index 90%
rename from src/libisvd/la/vml/sqrt.c
rename to src/lib/libisvd/la/vml/sqrt.c
index 14325a9..1d1068c 100644
--- a/src/libisvd/la/vml/sqrt.c
+++ b/src/lib/libisvd/la/vml/sqrt.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/vml/sqrt.c
+/// \file       lib/libisvd/la/vml/sqrt.c
 /// \brief      The VML Sqrt routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/vml/mathematical.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -32,4 +32,4 @@ void isvd_vzSqrt( const INT n, const COMP8 *a, COMP8 *y ) { isvd_xSqrt(n, a, y,
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/vml/sub.c b/src/lib/libisvd/la/vml/sub.c
similarity index 91%
rename from src/libisvd/la/vml/sub.c
rename to src/lib/libisvd/la/vml/sub.c
index f8ae430..a0029df 100644
--- a/src/libisvd/la/vml/sub.c
+++ b/src/lib/libisvd/la/vml/sub.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/vml/sub.c
+/// \file       lib/libisvd/la/vml/sub.c
 /// \brief      The VML Sub routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,9 +7,9 @@
 ///
 
 #include <isvd/la/vml/mathematical.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -32,4 +32,4 @@ void isvd_vzSub( const INT n, const COMP8 *a, const COMP8 *b, COMP8 *y ) { isvd_
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/lib/libisvd/la/vsl/@x@_rng_gaussian_driver.c b/src/lib/libisvd/la/vsl/@x@_rng_gaussian_driver.c
new file mode 100644
index 0000000..5cc15b6
--- /dev/null
+++ b/src/lib/libisvd/la/vsl/@x@_rng_gaussian_driver.c
@@ -0,0 +1,40 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       lib/libisvd/la/vsl/@x@_rng_gaussian_driver.c
+/// \brief      The VSL RngGaussian driver (@xname@ precision).
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+
+#include <isvd/la/vsl/driver.h>
+#include <libisvd/la/def.h>
+#include <isvd/la/vsl/service.h>
+#include <isvd/la/vsl/distribution.h>
+#include <isvd/util/omp.h>
+
+void isvd_v@x@RngGaussianDriver(
+    const isvd_int_t seed,
+    const isvd_int_t nskip,
+    const isvd_int_t n,
+          @xtype_____@ *r,
+    const @xtype_____@ a,
+    const @xtype_____@ sigma
+) {
+  ISVD_OMP_PARALLEL
+  {
+    omp_int_t omp_size = isvd_getOmpSize();
+    omp_int_t omp_rank = isvd_getOmpRank();
+
+    isvd_int_t length = n / omp_size;
+    isvd_int_t start = length * omp_rank;
+    if ( omp_rank == omp_size-1 ) {
+      length = n - start;
+    }
+
+    isvd_VSLStreamStatePtr stream;
+    isvd_vslNewStream(&stream, seed);
+    isvd_vslSkipAheadStream(stream, (nskip + start) * 2);
+    isvd_v@x@RngGaussian(stream, length, r + start, a, sigma);
+    isvd_vslDeleteStream(&stream);
+  }
+}
diff --git a/src/libisvd/la/vsl/rng_gaussian.c b/src/lib/libisvd/la/vsl/rng_gaussian.c
similarity index 90%
rename from src/libisvd/la/vsl/rng_gaussian.c
rename to src/lib/libisvd/la/vsl/rng_gaussian.c
index 06f65f8..62ecefa 100644
--- a/src/libisvd/la/vsl/rng_gaussian.c
+++ b/src/lib/libisvd/la/vsl/rng_gaussian.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/vsl/rng_gaussian.c
+/// \file       lib/libisvd/la/vsl/rng_gaussian.c
 /// \brief      The VSL RngGaussian routine.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -7,10 +7,10 @@
 ///
 
 #include <isvd/la/vsl/distribution.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 #include <isvd/la/lapack/auxiliary.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -34,4 +34,4 @@ void isvd_vdRngGaussian( isvd_VSLStreamStatePtr stream, const INT n, REAL8 *r, c
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/la/vsl/service.c b/src/lib/libisvd/la/vsl/service.c
similarity index 92%
rename from src/libisvd/la/vsl/service.c
rename to src/lib/libisvd/la/vsl/service.c
index 07f6092..8cd93c2 100644
--- a/src/libisvd/la/vsl/service.c
+++ b/src/lib/libisvd/la/vsl/service.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/la/vsl/service.c
+/// \file       lib/libisvd/la/vsl/service.c
 /// \brief      The VSL service routines.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -12,9 +12,9 @@
 #endif  // ISVD_USE_MKL
 
 #include <isvd/la/vsl/service.h>
-#include <libisvd/def.h>
+#include <libisvd/la/def.h>
 
-@ISVD_LA_BLAS_TYPE_DEFINE@
+@ISVD_TYPE_MACRO_DEFINE@
 
 #if defined(ISVD_USE_MKL)
 
@@ -57,4 +57,4 @@ void isvd_vslSkipAheadStream( isvd_VSLStreamStatePtr stream, const INT nskip ) {
 
 #endif  // ISVD_USE_MKL
 
-@ISVD_LA_BLAS_TYPE_UNDEF@
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/libisvd/nogpu/@x@_stage.c b/src/lib/libisvd/nogpu/@x@_stage.c
similarity index 53%
rename from src/libisvd/nogpu/@x@_stage.c
rename to src/lib/libisvd/nogpu/@x@_stage.c
index bcc04a4..16b746a 100644
--- a/src/libisvd/nogpu/@x@_stage.c
+++ b/src/lib/libisvd/nogpu/@x@_stage.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/nogpu/@x@_stage.c
+/// \file       lib/libisvd/nogpu/@x@_stage.c
 /// \brief      The iSVD stages with NOGPU support.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -19,19 +19,19 @@ extern "C" {
 
 // Sketching
 void isvd_@x@SketchGaussianProjection_gpu(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-          @xtype@    *yst,
-    const isvd_int_t  ldyst,
-    const isvd_int_t  seed,
-    const mpi_int_t   mpi_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+          @xtype_____@ *yst,
+    const isvd_int_t    ldyst,
+    const isvd_int_t    seed,
+    const mpi_int_t     mpi_root
 ) {
   ISVD_UNUSED(param);
   ISVD_UNUSED(argv);
@@ -52,24 +52,24 @@ void isvd_@x@SketchGaussianProjection_gpu(
 
 // Postprocessing
 void isvd_@x@PostprocessTallSkinnyQr_gpu(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root
 ) {
   ISVD_UNUSED(param);
   ISVD_UNUSED(argv);
@@ -94,24 +94,24 @@ void isvd_@x@PostprocessTallSkinnyQr_gpu(
 }
 
 void isvd_@x@PostprocessGramian_gpu(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root
 ) {
   ISVD_UNUSED(param);
   ISVD_UNUSED(argv);
@@ -136,24 +136,24 @@ void isvd_@x@PostprocessGramian_gpu(
 }
 
 void isvd_@x@PostprocessSymmetric_gpu(
-    const isvd_Param  param,
-    const @xtype@    *argv,
-    const isvd_int_t  argc,
-          @xtype@    *retv,
-    const isvd_int_t  retc,
-    const char        dista,
-    const char        ordera,
-    const @xtype@    *a,
-    const isvd_int_t  lda,
-    const @xtype@    *qt,
-    const isvd_int_t  ldqt,
-          @xtype@    *s,
-          @xtype@    *ut,
-    const isvd_int_t  ldut,
-          @xtype@    *vt,
-    const isvd_int_t  ldvt,
-    const mpi_int_t   ut_root,
-    const mpi_int_t   vt_root
+    const isvd_Param    param,
+    const @xtype_____@ *argv,
+    const isvd_int_t    argc,
+          @xtype_____@ *retv,
+    const isvd_int_t    retc,
+    const char          dista,
+    const char          ordera,
+    const @xtype_____@ *a,
+    const isvd_int_t    lda,
+    const @xtype_____@ *qt,
+    const isvd_int_t    ldqt,
+          @xtype_____@ *s,
+          @xtype_____@ *ut,
+    const isvd_int_t    ldut,
+          @xtype_____@ *vt,
+    const isvd_int_t    ldvt,
+    const mpi_int_t     ut_root,
+    const mpi_int_t     vt_root
 ) {
   ISVD_UNUSED(param);
   ISVD_UNUSED(argv);
diff --git a/src/libisvd/nogpu/env.c b/src/lib/libisvd/nogpu/env.c
similarity index 94%
rename from src/libisvd/nogpu/env.c
rename to src/lib/libisvd/nogpu/env.c
index cbec278..0605c4d 100644
--- a/src/libisvd/nogpu/env.c
+++ b/src/lib/libisvd/nogpu/env.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/nogpu/env.c
+/// \file       lib/libisvd/nogpu/env.c
 /// \brief      The iSVD environment routines (NOGPU only).
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/util.h b/src/lib/libisvd/util.h
similarity index 91%
rename from src/libisvd/util.h
rename to src/lib/libisvd/util.h
index 8ad1edb..446e509 100644
--- a/src/libisvd/util.h
+++ b/src/lib/libisvd/util.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/util.h
+/// \file       lib/libisvd/util.h
 /// \brief      The utility header.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/util/arg.h b/src/lib/libisvd/util/arg.h
similarity index 95%
rename from src/libisvd/util/arg.h
rename to src/lib/libisvd/util/arg.h
index 5eae2a7..4668477 100644
--- a/src/libisvd/util/arg.h
+++ b/src/lib/libisvd/util/arg.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/util/arg.h
+/// \file       lib/libisvd/util/arg.h
 /// \brief      The argument utilities.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -24,7 +24,7 @@ static inline isvd_int_t isvd_arg2char(
     const char *opts,
     const char *rets
 ) {
-  if ( rets == nullptr ) {
+  if ( rets == NULL ) {
     rets = opts;
   }
 
@@ -56,7 +56,7 @@ static inline const char* isvd_arg2str(
     const char *rets[],
     const size_t nopts
 ) {
-  if ( rets == nullptr ) {
+  if ( rets == NULL ) {
     rets = opts;
   }
 
diff --git a/src/libisvd/util/function.h b/src/lib/libisvd/util/function.h
similarity index 95%
rename from src/libisvd/util/function.h
rename to src/lib/libisvd/util/function.h
index becaa0e..cd1ef41 100644
--- a/src/libisvd/util/function.h
+++ b/src/lib/libisvd/util/function.h
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/util/function.h
+/// \file       lib/libisvd/util/function.h
 /// \brief      Some useful functions.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
diff --git a/src/libisvd/util/io.c b/src/lib/libisvd/util/io.c
similarity index 68%
rename from src/libisvd/util/io.c
rename to src/lib/libisvd/util/io.c
index 5bd5f6d..11306c7 100644
--- a/src/libisvd/util/io.c
+++ b/src/lib/libisvd/util/io.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/util/io.c
+/// \file       lib/libisvd/util/io.c
 /// \brief      The I/O utilities.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -9,6 +9,8 @@
 #include <isvd/util/io.h>
 #include <libisvd/def.h>
 
+@ISVD_TYPE_MACRO_DEFINE@
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_util_io_module
 /// \brief  Get a variable from stream.
@@ -17,25 +19,27 @@
 /// \param  varp    Pointer to the variable.
 ///
 //\{
-void isvd_ifget( FILE *stream, isvd_int_t *varp ) {
+void isvd_ifget( FILE *stream, INT *varp ) {
 #if !defined(ISVD_USE_ILP64)
-  isvd_int_t info = fscanf(stream, "%"PRId32,  varp);
+  INT info = fscanf(stream, "%" PRId32, varp);
 #else  // ISVD_USE_ILP64
-  isvd_int_t info = fscanf(stream, "%" PRId64, varp);
+  INT info = fscanf(stream, "%" PRId64, varp);
 #endif  // ISVD_USE_ILP64
   ISVD_UNUSED(info);
   isvd_assert_eq(info, 1);
 }
 
-void isvd_sfget( FILE *stream, float *varp ) {
-  isvd_int_t info = fscanf(stream, "%f",  varp);
+void isvd_sfget( FILE *stream, REAL4 *varp ) {
+  INT info = fscanf(stream, "%f",  varp);
   ISVD_UNUSED(info);
   isvd_assert_eq(info, 1);
 }
 
-void isvd_dfget( FILE *stream, double *varp ) {
-  isvd_int_t info = fscanf(stream, "%lf", varp);
+void isvd_dfget( FILE *stream, REAL8 *varp ) {
+  INT info = fscanf(stream, "%lf", varp);
   ISVD_UNUSED(info);
   isvd_assert_eq(info, 1);
 }
 //\}
+
+@ISVD_TYPE_MACRO_UNDEF@
diff --git a/src/lib/libisvd/util/memory.c b/src/lib/libisvd/util/memory.c
new file mode 100644
index 0000000..f4e62ec
--- /dev/null
+++ b/src/lib/libisvd/util/memory.c
@@ -0,0 +1,11 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       lib/libisvd/util/memory.c
+/// \brief      The memory utilities.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+/// \note  Implement in \ref lib/libisvd/la/util/memory.c
+///
+
+#include <isvd/util/memory.h>
\ No newline at end of file
diff --git a/src/libisvd/util/mpi.c b/src/lib/libisvd/util/mpi.c
similarity index 86%
rename from src/libisvd/util/mpi.c
rename to src/lib/libisvd/util/mpi.c
index b6e615a..30af4b7 100644
--- a/src/libisvd/util/mpi.c
+++ b/src/lib/libisvd/util/mpi.c
@@ -1,5 +1,5 @@
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
-/// \file       src/libisvd/util/mpi.c
+/// \file       lib/libisvd/util/mpi.c
 /// \brief      The MPI utilities.
 ///
 /// \author     Mu Yang <<emfomy@gmail.com>>
@@ -9,11 +9,6 @@
 #include <isvd/util/mpi.h>
 #include <libisvd/def.h>
 
-#if !defined(DOXYGEN_SHOULD_SKIP_THIS)
-#define MPI_COMPLEX_FLOAT  MPI_COMPLEX8
-#define MPI_COMPLEX_DOUBLE MPI_COMPLEX16
-#endif  // DOXYGEN_SHOULD_SKIP_THIS
-
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 /// \ingroup  c_util_mpi_module
 /// \brief  Returns the size of the group associated with a communicator.
diff --git a/src/lib/libisvd/util/omp.c b/src/lib/libisvd/util/omp.c
new file mode 100644
index 0000000..0a6a13b
--- /dev/null
+++ b/src/lib/libisvd/util/omp.c
@@ -0,0 +1,11 @@
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+/// \file       lib/libisvd/util/omp.c
+/// \brief      The OpenMP utilities.
+///
+/// \author     Mu Yang <<emfomy@gmail.com>>
+/// \copyright  MIT License
+///
+/// \note  Implement in \ref lib/libisvd/la/util/omp.c
+///
+
+#include <isvd/util/omp.h>
\ No newline at end of file