Rework the MPI_Op support.

Remove all ops with 3 buffers, we ended up not using them anywhere in
the code.
Change the loop order in the base MPI_Op to allow for more
optimizations, as discussed in #9717.

Fixes #9717.

Signed-off-by: George Bosilca <[email protected]>

Author: bosilca

ompi/debuggers/predefined_gap_test.c

@@ -110,7 +110,6 @@ int main(int argc, char **argv) {
     GAP_CHECK("o_flags", test_op, o_flags, o_name, 1)
     GAP_CHECK("o_f_to_c_index", test_op, o_f_to_c_index, o_flags, 1)
     GAP_CHECK("o_func", test_op, o_func, o_f_to_c_index, 1)
-    GAP_CHECK("o_3buff_instrinsic", test_op, o_3buff_intrinsic, o_func, 1)
 
     /* Test Predefined datatype sizes */
     printf("=============================================\n");

ompi/mca/op/avx/op_avx_component.c

@@ -237,15 +237,12 @@ avx_component_init_query(bool enable_progress_threads,
 
 #if OMPI_MCA_OP_HAVE_AVX512
  extern ompi_op_base_handler_fn_t ompi_op_avx_functions_avx512[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
- extern ompi_op_base_3buff_handler_fn_t ompi_op_avx_3buff_functions_avx512[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
 #endif
 #if OMPI_MCA_OP_HAVE_AVX2
  extern ompi_op_base_handler_fn_t ompi_op_avx_functions_avx2[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
- extern ompi_op_base_3buff_handler_fn_t ompi_op_avx_3buff_functions_avx2[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
 #endif
 #if OMPI_MCA_OP_HAVE_AVX
  extern ompi_op_base_handler_fn_t ompi_op_avx_functions_avx[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
- extern ompi_op_base_3buff_handler_fn_t ompi_op_avx_3buff_functions_avx[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
 #endif
 /*
  * Query whether this component can be used for a specific op
@@ -274,35 +271,25 @@ avx_component_op_query(struct ompi_op_t *op, int *priority)
 #if OMPI_MCA_OP_HAVE_AVX512
             if( mca_op_avx_component.flags & OMPI_OP_AVX_HAS_AVX512F_FLAG ) {
                 module->opm_fns[i] = ompi_op_avx_functions_avx512[op->o_f_to_c_index][i];
-                module->opm_3buff_fns[i] = ompi_op_avx_3buff_functions_avx512[op->o_f_to_c_index][i];
             }
 #endif
 #if OMPI_MCA_OP_HAVE_AVX2
             if( mca_op_avx_component.flags & OMPI_OP_AVX_HAS_AVX2_FLAG ) {
                 if( NULL == module->opm_fns[i] ) {
                     module->opm_fns[i] = ompi_op_avx_functions_avx2[op->o_f_to_c_index][i];
                 }
-                if( NULL == module->opm_3buff_fns[i] ) {
-                    module->opm_3buff_fns[i] = ompi_op_avx_3buff_functions_avx2[op->o_f_to_c_index][i];
-                }
             }
 #endif
 #if OMPI_MCA_OP_HAVE_AVX
             if( mca_op_avx_component.flags & OMPI_OP_AVX_HAS_AVX_FLAG ) {
                 if( NULL == module->opm_fns[i] ) {
                     module->opm_fns[i] = ompi_op_avx_functions_avx[op->o_f_to_c_index][i];
                 }
-                if( NULL == module->opm_3buff_fns[i] ) {
-                    module->opm_3buff_fns[i] = ompi_op_avx_3buff_functions_avx[op->o_f_to_c_index][i];
-                }
             }
 #endif
             if( NULL != module->opm_fns[i] ) {
                 OBJ_RETAIN(module);
             }
-            if( NULL != module->opm_3buff_fns[i] ) {
-                OBJ_RETAIN(module);
-            }
         }
         break;
     case OMPI_OP_BASE_FORTRAN_LAND:

ompi/mca/op/avx/op_avx_functions.c

@@ -35,8 +35,6 @@ BEGIN_C_DECLS
  */
 OMPI_DECLSPEC extern ompi_op_base_handler_fn_t
     ompi_op_base_functions[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
-OMPI_DECLSPEC extern ompi_op_base_3buff_handler_fn_t
-    ompi_op_base_3buff_functions[OMPI_OP_BASE_FORTRAN_OP_MAX][OMPI_OP_BASE_TYPE_MAX];
 
 END_C_DECLS
 

ompi/mca/op/base/functions.h

@@ -43,15 +43,13 @@ static void module_constructor(ompi_op_base_module_t *m)
     m->opm_enable = NULL;
     m->opm_op = NULL;
     memset(&(m->opm_fns), 0, sizeof(m->opm_fns));
-    memset(&(m->opm_3buff_fns), 0, sizeof(m->opm_3buff_fns));
 }
 
 static void module_constructor_1_0_0(ompi_op_base_module_1_0_0_t *m)
 {
     m->opm_enable = NULL;
     m->opm_op = NULL;
     memset(&(m->opm_fns), 0, sizeof(m->opm_fns));
-    memset(&(m->opm_3buff_fns), 0, sizeof(m->opm_3buff_fns));
 }
 
 OBJ_CLASS_INSTANCE(ompi_op_base_module_t, opal_object_t,