Merge branch 'feature-openmp' into 'master'

Add OpenMP support for shared memory systems

See merge request !108

.gitlab-ci.yml

@@ -93,6 +93,18 @@ build_parsec:
     BUILD_OPTIONS: "-DCHAMELEON_SCHED=PARSEC"
     VERSION: parsec
 
+build_openmp:
+  <<: *build_script
+  artifacts:
+    name: build_openmp
+    expire_in: 48 hours
+    paths:
+      - build
+      - chameleon_openmp.log
+  variables:
+    BUILD_OPTIONS: "-DCHAMELEON_SCHED=OPENMP"
+    VERSION: openmp
+
 test_starpu_branches:
   <<: *test_script
   variables:
@@ -239,6 +251,42 @@ test_parsec_master:
   only:
     - master@solverstack/chameleon
 
+test_openmp_branches:
+  <<: *test_script
+  variables:
+    TESTS_RESTRICTION: "-R \"test_shm_s|test_mpi_s\""
+    VERSION: openmp
+  dependencies:
+    - build_openmp
+  artifacts:
+    name: test_openmp
+    expire_in: 48 hours
+    paths:
+      - build
+      - chameleon_openmp.log
+      - chameleon_openmp.lcov
+  only:
+    - branches
+  except:
+    - master
+
+test_openmp_master:
+  <<: *test_script
+  variables:
+    TESTS_RESTRICTION: "-R \"_shm_|_mpi_\""
+    VERSION: openmp
+  dependencies:
+    - build_openmp
+  artifacts:
+    name: test_openmp
+    expire_in: 48 hours
+    paths:
+      - build
+      - chameleon_openmp.log
+      - chameleon_openmp.lcov
+  only:
+    - master@solverstack/chameleon
+
 coverage:
   stage: analyse
   dependencies:
@@ -246,11 +294,13 @@ coverage:
     - test_starpu_simgrid_branches
     - test_quark_branches
     - test_parsec_branches
+    - test_openmp_branches
   script:
       - lcov -a chameleon_starpu.lcov
              -a chameleon_starpu_simgrid.lcov
              -a chameleon_quark.lcov
              -a chameleon_parsec.lcov
+             -a chameleon_openmp.lcov
              -o chameleon.lcov
       - lcov --summary chameleon.lcov
   only:
@@ -289,6 +339,7 @@ sonarqube:
     - test_starpu_simgrid_master
     - test_quark_master
     - test_parsec_master
+    - test_openmp_master
   artifacts:
     name: chameleon_analysis
     expire_in: 1 week

CMakeLists.txt

@@ -108,7 +108,7 @@ option(CHAMELEON_USE_MIGRATE
 # -----------------------------
 
 # Create a list of possible runtime
-set(CHAMELEON_SCHED_list PARSEC STARPU QUARK
+set(CHAMELEON_SCHED_list PARSEC STARPU QUARK OPENMP
   CACHE INTERNAL "List of available runtimes" )
 
 set( CHAMELEON_SCHED_PARSEC OFF CACHE INTERNAL
@@ -120,6 +120,9 @@ set(CHAMELEON_SCHED_STARPU OFF CACHE INTERNAL
 set(CHAMELEON_SCHED_QUARK OFF CACHE INTERNAL
   "Enable Quark scheduler as the default runtime
     (Conflict with other CHAMELEON_SCHED_* options)")
+set(CHAMELEON_SCHED_OPENMP OFF CACHE INTERNAL
+  "Enable OpenMP scheduler as the default runtime
+    (Conflict with other CHAMELEON_SCHED_* options)")
 
 set( CHAMELEON_SCHED STARPU
   CACHE STRING "Choose the chameleon internal runtime from ${CHAMELEON_SCHED_list}")
@@ -892,6 +895,22 @@ if( CHAMELEON_SCHED_QUARK )
 
 endif()
 
+if( CHAMELEON_SCHED_OPENMP )
+    find_package(OpenMP REQUIRED)
+
+    if ( OPENMP_FOUND )
+        message("-- ${Blue}Add definition CHAMELEON_SCHED_OPENMP"
+        " - Activate OpenMP in Chameleon${ColourReset}")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
+        set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} ${OpenMP_Fortran_FLAGS}")
+        set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_C_FLAGS}")
+        # TODO: check for compiler support for "_Pragma"
+    else ( OPENMP_FOUND )
+      message(FATAL_ERROR "Something went wrong when finding OpenMP")
+    endif ( OPENMP_FOUND )
+
+endif( CHAMELEON_SCHED_OPENMP )
+
 # getopt
 check_include_files(getopt.h CHAMELEON_HAVE_GETOPT_H)
 if (CHAMELEON_HAVE_GETOPT_H)
@@ -937,6 +956,9 @@ endif()
 if (CHAMELEON_SCHED_STARPU)
   list(APPEND CHAMELEON_LIBRARIES chameleon_starpu)
 endif()
+if (CHAMELEON_SCHED_OPENMP)
+  list(APPEND CHAMELEON_LIBRARIES chameleon_openmp)
+endif()
 list(APPEND CHAMELEON_LIBRARIES hqr)
 set(CHAMELEON_LIBRARIES_DEP ${CHAMELEON_LIBRARIES} ${CHAMELEON_DEP})
 

compute/CMakeLists.txt

@@ -29,7 +29,7 @@ if ( CHAMELEON_USE_CUDA )
   set( CHAMELEON_COPY_DIAG ON )
 else()
   option(CHAMELEON_COPY_DIAG
-    "This options enables the duplication of the diagonal tiles in some algorithm to avoid anti-dependencies on lower/upper triangular parts (Might be useful to StarPU)" ON)
+    "This options enables the duplication of the diagonal tiles in some algorithm to avoid anti-dependencies on lower/upper triangular parts (Might be useful to StarPU and OpenMP)" ON)
 endif()
 
 if ( CHAMELEON_SCHED_QUARK )
@@ -296,6 +296,8 @@ elseif(CHAMELEON_SCHED_PARSEC)
   target_link_libraries(chameleon chameleon_parsec)
 elseif(CHAMELEON_SCHED_QUARK)
   target_link_libraries(chameleon chameleon_quark)
+elseif(CHAMELEON_SCHED_OPENMP)
+  target_link_libraries(chameleon chameleon_openmp)
 endif()
 if (NOT CHAMELEON_SIMULATION)
   # Depends on coreblas only for set_coreblas_gemm3m_enabled() (Maybe we should change that)

control/chameleon_f77.c

@@ -30,11 +30,11 @@ extern "C" {
     /**
      *  FORTRAN API - auxiliary function prototypes
      */
-    void CHAMELEON_INIT(int *CORES, int *NGPUS, int *INFO)
-    {   *INFO = CHAMELEON_Init(*CORES, *NGPUS); }
+    void __CHAMELEON_INIT(int *CORES, int *NGPUS, int *INFO)
+    {   *INFO = __chameleon_init(*CORES, *NGPUS); }
 
-    void CHAMELEON_FINALIZE(int *INFO)
-    {   *INFO = CHAMELEON_Finalize(); }
+    void __CHAMELEON_FINALIZE(int *INFO)
+    {   *INFO = __chameleon_finalize(); }
 
     void CHAMELEON_ENABLE(int *lever, int *INFO)
     {   *INFO = CHAMELEON_Enable(*lever); }

control/chameleon_f77.h

@@ -30,8 +30,8 @@
 #define CHAMELEON_WS_FNAME(lcname, UCNAME) CHAMELEON_GLOBAL(chameleon_alloc_workspace_##lcname, CHAMELEON_ALLOC_WORKSPACE_##UCNAME)
 #define CHAMELEON_WST_FNAME(lcname, UCNAME) CHAMELEON_GLOBAL(chameleon_alloc_workspace_##lcname##_tile, CHAMELEON_ALLOC_WORKSPACE_##UCNAME##_TILE)
 
-#define CHAMELEON_INIT CHAMELEON_GLOBAL(chameleon_init, CHAMELEON_INIT)
-#define CHAMELEON_FINALIZE CHAMELEON_GLOBAL(chameleon_finalize, CHAMELEON_FINALIZE)
+#define __CHAMELEON_INIT CHAMELEON_GLOBAL(__chameleon_init, __CHAMELEON_INIT)
+#define __CHAMELEON_FINALIZE CHAMELEON_GLOBAL(__chameleon_finalize, __CHAMELEON_FINALIZE)
 #define CHAMELEON_ENABLE CHAMELEON_GLOBAL(chameleon_enable, CHAMELEON_ENABLE)
 #define CHAMELEON_DISABLE CHAMELEON_GLOBAL(chameleon_disable, CHAMELEON_DISABLE)
 #define CHAMELEON_SET CHAMELEON_GLOBAL(chameleon_set, CHAMELEON_SET)
@@ -46,4 +46,20 @@
 #define CHAMELEON_LAPACK_TO_TILE CHAMELEON_GLOBAL(chameleon_lapack_to_tile, CHAMELEON_LAPACK_TO_TILE)
 #define CHAMELEON_TILE_TO_LAPACK CHAMELEON_GLOBAL(chameleon_tile_to_lapack, CHAMELEON_TILE_TO_LAPACK)
 
+#if defined(CHAMELEON_SCHED_OPENMP)
+#define CHAMELEON_INIT(nworkers, ncudas)\
+    CALL __CHAMELEON_INIT(nworkers, ncudas)\
+    !$omp parallel\
+    !$omp master
+#define CHAMELEON_FINALIZE()\
+    !$omp end master\
+    !$omp end parallel\
+    CALL __CHAMELEON_FINALIZE()
+#else
+#define CHAMELEON_INIT(nworkers, ncudas)\
+    CALL __CHAMELEON_INIT(nworkers, ncudas)
+#define CHAMELEON_FINALIZE()\
+    CALL __CHAMELEON_FINALIZE()
+#endif
+
 #endif /* _chameleon_f77_h_ */

control/control.c

@@ -34,7 +34,7 @@
  *
  * @ingroup Control
  *
- *  CHAMELEON_Init - Initialize CHAMELEON.
+ * @brief Initialize CHAMELEON.
  *
  ******************************************************************************
  *
@@ -46,20 +46,19 @@
  *
  ******************************************************************************
  *
- * @return
- *          \retval CHAMELEON_SUCCESS successful exit
+ * @retval CHAMELEON_SUCCESS successful exit
  *
  */
-int CHAMELEON_Init(int cores, int gpus)
+int __chameleon_init(int cores, int gpus)
 {
-    return CHAMELEON_InitPar(cores, gpus, -1);
+   return __chameleon_initpar(cores, gpus, -1);
 }
 
 /**
  *
  * @ingroup Control
  *
- *  CHAMELEON_InitPar - Initialize CHAMELEON.
+ * @brief Initialize CHAMELEON.
  *
  ******************************************************************************
  *
@@ -74,11 +73,10 @@ int CHAMELEON_Init(int cores, int gpus)
  *
  ******************************************************************************
  *
- * @return
- *          \retval CHAMELEON_SUCCESS successful exit
+ * @retval CHAMELEON_SUCCESS successful exit
  *
  */
-int CHAMELEON_InitPar(int ncpus, int ncudas, int nthreads_per_worker)
+int __chameleon_initpar(int ncpus, int ncudas, int nthreads_per_worker)
 {
     CHAM_context_t *chamctxt;
 
@@ -119,15 +117,14 @@ int CHAMELEON_InitPar(int ncpus, int ncudas, int nthreads_per_worker)
  *
  * @ingroup Control
  *
- *  CHAMELEON_Finalize - Finalize CHAMELEON.
+ * @brief Finalize CHAMELEON.
  *
  ******************************************************************************
  *
- * @return
- *          \retval CHAMELEON_SUCCESS successful exit
+ * @retval CHAMELEON_SUCCESS successful exit
  *
  */
-int CHAMELEON_Finalize(void)
+int __chameleon_finalize(void)
 {
     CHAM_context_t *chamctxt = chameleon_context_self();
     if (chamctxt == NULL) {

example/lapack_to_chameleon/step1.c

@@ -79,9 +79,9 @@ int main(int argc, char *argv[]) {
     print_header( argv[0], iparam);
 
     /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
     /*
@@ -170,8 +170,14 @@ int main(int argc, char *argv[]) {
     free(B);
     free(X);
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 
-    return EXIT_SUCCESS;
+    return rc;
 }

example/lapack_to_chameleon/step2.c

@@ -74,9 +74,9 @@ int main(int argc, char *argv[]) {
     print_header( argv[0], iparam);
 
     /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
     /* Question chameleon to get the block (tile) size (number of columns) */
@@ -220,8 +220,14 @@ int main(int argc, char *argv[]) {
     CHAMELEON_Desc_Destroy( &descX );
     CHAMELEON_Desc_Destroy( &descAC );
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 
-    return EXIT_SUCCESS;
+    return rc;
 }

example/lapack_to_chameleon/step3.c

@@ -75,11 +75,12 @@ int main(int argc, char *argv[]) {
     print_header( argv[0], iparam);
 
     /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
+
     /* Question chameleon to get the block (tile) size (number of columns) */
     CHAMELEON_Get( CHAMELEON_TILE_SIZE, &NB );
 
@@ -201,8 +202,14 @@ int main(int argc, char *argv[]) {
     CHAMELEON_Desc_Destroy( &descX );
     CHAMELEON_Desc_Destroy( &descAC );
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 
-    return EXIT_SUCCESS;
+    return rc;
 }

example/lapack_to_chameleon/step4.c

@@ -53,7 +53,6 @@ int main(int argc, char *argv[]) {
     RUNTIME_sequence_t *sequence = NULL;
     /* CHAMELEON request uniquely identifies each asynchronous function call */
     RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
-    int status;
 
     /* initialize some parameters with default values */
     int iparam[IPARAM_SIZEOF];
@@ -83,9 +82,9 @@ int main(int argc, char *argv[]) {
     print_header( argv[0], iparam);
 
     /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
     /* Question chameleon to get the block (tile) size (number of columns) */
@@ -141,10 +140,10 @@ int main(int argc, char *argv[]) {
      * have been terminated */
     CHAMELEON_Sequence_Wait(sequence);
 
-    status = sequence->status;
-    if ( status != 0 ) {
-        fprintf(stderr, "Error in computation (%d)\n", status);
-        return EXIT_FAILURE;
+    rc = sequence->status;
+    if ( rc != CHAMELEON_SUCCESS ) {
+        fprintf(stderr, "Error in computation (%d)\n", rc);
+        goto finalize;
     }
     CHAMELEON_Sequence_Destroy(sequence);
 
@@ -195,8 +194,14 @@ int main(int argc, char *argv[]) {
     CHAMELEON_Desc_Destroy( &descX );
     CHAMELEON_Desc_Destroy( &descAC );
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 
-    return EXIT_SUCCESS;
+    return rc;
 }

example/lapack_to_chameleon/step5.c

@@ -53,7 +53,6 @@ int main(int argc, char *argv[]) {
     RUNTIME_sequence_t *sequence = NULL;
     /* CHAMELEON request uniquely identifies each asynchronous function call */
     RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
-    int status;
 
     /* initialize some parameters with default values */
     int iparam[IPARAM_SIZEOF];
@@ -86,9 +85,9 @@ int main(int argc, char *argv[]) {
     print_header( argv[0], iparam);
 
      /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
     /* set some specific parameters related to CHAMELEON: blocks size and inner-blocking size */
@@ -145,10 +144,10 @@ int main(int argc, char *argv[]) {
      * have been terminated */
     CHAMELEON_Sequence_Wait(sequence);
 
-    status = sequence->status;
-    if ( status != 0 ) {
-        fprintf(stderr, "Error in computation (%d)\n", status);
-        return EXIT_FAILURE;
+    rc = sequence->status;
+    if ( rc != CHAMELEON_SUCCESS ) {
+        fprintf(stderr, "Error in computation (%d)\n", rc);
+        goto finalize;
     }
     CHAMELEON_Sequence_Destroy(sequence);
 
@@ -199,6 +198,12 @@ int main(int argc, char *argv[]) {
     CHAMELEON_Desc_Destroy( &descX );
     CHAMELEON_Desc_Destroy( &descAC );
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 

example/lapack_to_chameleon/step6.c

@@ -54,7 +54,6 @@ int main(int argc, char *argv[]) {
     RUNTIME_sequence_t *sequence = NULL;
     /* CHAMELEON request uniquely identifies each asynchronous function call */
     RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
-    int status;
 
     /* initialize some parameters with default values */
     int iparam[IPARAM_SIZEOF];
@@ -84,9 +83,9 @@ int main(int argc, char *argv[]) {
     NGPU = iparam[IPARAM_NCUDAS];
 
      /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
     /* set some specific parameters related to CHAMELEON: blocks size and inner-blocking size */
@@ -167,10 +166,10 @@ int main(int argc, char *argv[]) {
      * have been terminated */
     CHAMELEON_Sequence_Wait(sequence);
 
-    status = sequence->status;
-    if ( status != 0 ) {
-        fprintf(stderr, "Error in computation (%d)\n", status);
-        return EXIT_FAILURE;
+    rc = sequence->status;
+    if ( rc != CHAMELEON_SUCCESS ) {
+        fprintf(stderr, "Error in computation (%d)\n", rc);
+        goto finalize;
     }
     CHAMELEON_Sequence_Destroy(sequence);
 
@@ -225,6 +224,12 @@ int main(int argc, char *argv[]) {
     CHAMELEON_Desc_Destroy( &descX );
     CHAMELEON_Desc_Destroy( &descAC );
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 

example/lapack_to_chameleon/step7.c

@@ -58,7 +58,6 @@ int main(int argc, char *argv[]) {
     RUNTIME_sequence_t *sequence = NULL;
     /* CHAMELEON request uniquely identifies each asynchronous function call */
     RUNTIME_request_t request = RUNTIME_REQUEST_INITIALIZER;
-    int status;
 
     /* initialize some parameters with default values */
     int iparam[IPARAM_SIZEOF];
@@ -88,9 +87,9 @@ int main(int argc, char *argv[]) {
     NGPU = iparam[IPARAM_NCUDAS];
 
      /* Initialize CHAMELEON with main parameters */
-    if ( CHAMELEON_Init( NCPU, NGPU ) != CHAMELEON_SUCCESS ) {
-        fprintf(stderr, "Error initializing CHAMELEON library\n");
-        return EXIT_FAILURE;
+    int rc = CHAMELEON_Init( NCPU, NGPU );
+    if (rc != CHAMELEON_SUCCESS) {
+        goto finalize;
     }
 
     /* set some specific parameters related to CHAMELEON: blocks size and inner-blocking size */
@@ -175,10 +174,10 @@ int main(int argc, char *argv[]) {
      * have been terminated */
     CHAMELEON_Sequence_Wait(sequence);
 
-    status = sequence->status;
-    if ( status != 0 ) {
-        fprintf(stderr, "Error in computation (%d)\n", status);
-        return EXIT_FAILURE;
+    rc = sequence->status;
+    if ( rc != CHAMELEON_SUCCESS ) {
+        fprintf(stderr, "Error in computation (%d)\n", rc);
+        goto finalize;
     }
     CHAMELEON_Sequence_Destroy(sequence);
 
@@ -233,6 +232,12 @@ int main(int argc, char *argv[]) {
     CHAMELEON_Desc_Destroy( &descX );
     CHAMELEON_Desc_Destroy( &descAC );
 
+finalize:
+    /*
+     * Required semicolon to have at least one inst
+     * before the end of OpenMP block.
+     */
+    ;
     /* Finalize CHAMELEON */
     CHAMELEON_Finalize();
 

include/chameleon.h

@@ -80,9 +80,9 @@ BEGIN_C_DECLS
 /* Auxiliary */
 int CHAMELEON_Version           (int *ver_major, int *ver_minor, int *ver_micro);
 int CHAMELEON_My_Mpi_Rank       (void);
-int CHAMELEON_Init              (int nworkers, int ncudas);
-int CHAMELEON_InitPar           (int nworkers, int ncudas, int nthreads_per_worker);
-int CHAMELEON_Finalize          (void);
+int __chameleon_init            (int nworkers, int ncudas);
+int __chameleon_initpar         (int nworkers, int ncudas, int nthreads_per_worker);
+int __chameleon_finalize        (void);
 int CHAMELEON_Pause             (void);
 int CHAMELEON_Resume            (void);
 int CHAMELEON_Distributed_start (void);
@@ -135,6 +135,48 @@ int CHAMELEON_Sequence_Create  (RUNTIME_sequence_t **sequence);
 int CHAMELEON_Sequence_Destroy (RUNTIME_sequence_t *sequence);
 int CHAMELEON_Sequence_Wait    (RUNTIME_sequence_t *sequence);
 
+/**
+ *
+ * @ingroup Control
+ *
+ * @brief Initialize CHAMELEON.
+ *
+ ******************************************************************************
+ *
+ * @param[in] cores
+ *          Number of cores to use.
+ *
+ * @param[in] gpus
+ *          Number of cuda devices to use.
+ *
+ ******************************************************************************
+ *
+ * @retval CHAMELEON_SUCCESS successful exit
+ *
+ */
+#if defined(CHAMELEON_SCHED_OPENMP)
+#define CHAMELEON_Init(nworkers, ncudas)\
+    __chameleon_init(nworkers, ncudas);\
+    _Pragma("omp parallel")\
+    _Pragma("omp master")\
+    {
+#define CHAMELEON_InitPar(nworkers, ncudas, nthreads_per_worker)\
+    __chameleon_initpar(nworkers, ncudas, nthreads_per_worker);\
+    _Pragma("omp parallel")\