|
RTOpPack: Extra C/C++ Code for Vector Reduction/Transformation Operators
Version of the Day
|
00001 /* 00002 // @HEADER 00003 // *********************************************************************** 00004 // 00005 // Moocho: Multi-functional Object-Oriented arCHitecture for Optimization 00006 // Copyright (2003) Sandia Corporation 00007 // 00008 // Under terms of Contract DE-AC04-94AL85000, there is a non-exclusive 00009 // license for use of this work by or on behalf of the U.S. Government. 00010 // 00011 // Redistribution and use in source and binary forms, with or without 00012 // modification, are permitted provided that the following conditions are 00013 // met: 00014 // 00015 // 1. Redistributions of source code must retain the above copyright 00016 // notice, this list of conditions and the following disclaimer. 00017 // 00018 // 2. Redistributions in binary form must reproduce the above copyright 00019 // notice, this list of conditions and the following disclaimer in the 00020 // documentation and/or other materials provided with the distribution. 00021 // 00022 // 3. Neither the name of the Corporation nor the names of the 00023 // contributors may be used to endorse or promote products derived from 00024 // this software without specific prior written permission. 00025 // 00026 // THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY 00027 // EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 00028 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 00029 // PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE 00030 // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 00031 // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 00032 // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 00033 // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 00034 // LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 00035 // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 00036 // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 00037 // 00038 // Questions? Contact Roscoe A. Bartlett (rabartl@sandia.gov) 00039 // 00040 // *********************************************************************** 00041 // @HEADER 00042 */ 00043 00044 #include "RTOp_TOp_axpy.h" 00045 #include "RTOp_obj_value_vtbl.h" 00046 #include "RTOp_obj_null_vtbl.h" 00047 00048 /* Implementation functions for RTOp_RTOp */ 00049 00050 static int RTOp_TOp_axpy_apply_op( 00051 const struct RTOp_RTOp_vtbl_t* vtbl, const void* obj_data 00052 , const int num_vecs, const struct RTOp_SubVector vecs[] 00053 , const int num_targ_vecs, const struct RTOp_MutableSubVector targ_vecs[] 00054 , RTOp_ReductTarget targ_obj ) 00055 { 00056 RTOp_value_type alpha; 00057 RTOp_index_type z_sub_dim; 00058 RTOp_value_type *z_val; 00059 ptrdiff_t z_val_s; 00060 RTOp_index_type v0_sub_dim; 00061 const RTOp_value_type *v0_val; 00062 ptrdiff_t v0_val_s; 00063 register RTOp_index_type k; 00064 RTOp_value_type *z_val_tmp = NULL; 00065 #ifdef RTOp_DEBUG 00066 RTOp_index_type i; 00067 #endif 00068 00069 /* */ 00070 /* Validate the input */ 00071 /* */ 00072 if( num_vecs != 1 || vecs == NULL ) 00073 return RTOp_ERR_INVALID_NUM_VECS; 00074 if( num_targ_vecs != 1 || targ_vecs == NULL ) 00075 return RTOp_ERR_INVALID_NUM_TARG_VECS; 00076 if( targ_vecs[0].sub_dim != vecs[0].sub_dim ) 00077 return RTOp_ERR_INCOMPATIBLE_VECS; 00078 00079 /* */ 00080 /* Get pointers to data */ 00081 /* */ 00082 00083 /* alpha */ 00084 alpha = *((RTOp_value_type*)obj_data); 00085 00086 /* z */ 00087 z_sub_dim = targ_vecs[0].sub_dim; 00088 z_val = targ_vecs[0].values; 00089 z_val_s = targ_vecs[0].values_stride; 00090 00091 /* v1 */ 00092 v0_sub_dim = vecs[0].sub_dim; 00093 v0_val = vecs[0].values; 00094 v0_val_s = vecs[0].values_stride; 00095 00096 z_val_tmp = z_val; 00097 00098 /* */ 00099 /* Perform the operation */ 00100 /* */ 00101 00102 /* Dense vector v[0] */ 00103 if( z_val_s == 1 && v0_val_s == 1 ) { 00104 /* Slightly faster loop for unit stride vectors */ 00105 for( k = 0; k < z_sub_dim; ++k ) 00106 *z_val++ += alpha * (*v0_val++); 00107 } 00108 else { 00109 /* More general implementation for one or both non-unit strides */ 00110 for( k = 0; k < z_sub_dim; ++k, z_val += z_val_s, v0_val += v0_val_s ) 00111 *z_val += alpha * (*v0_val); 00112 } 00113 00114 return 0; /* success? */ 00115 } 00116 00117 /* Virtual function table */ 00118 const struct RTOp_RTOp_vtbl_t RTOp_TOp_axpy_vtbl = 00119 { 00120 &RTOp_obj_value_vtbl /* use simple scalar value type for object instance data */ 00121 ,&RTOp_obj_null_vtbl /* use null type for target object */ 00122 ,"TOp_axpy" 00123 ,NULL /* use default from reduct_vtbl */ 00124 ,RTOp_TOp_axpy_apply_op 00125 ,NULL 00126 ,NULL 00127 }; 00128 00129 /* Class specific functions */ 00130 00131 int RTOp_TOp_axpy_construct( RTOp_value_type alpha, struct RTOp_RTOp* op ) 00132 { 00133 op->vtbl = &RTOp_TOp_axpy_vtbl; 00134 op->vtbl->obj_data_vtbl->obj_create(NULL,NULL,&op->obj_data); 00135 *((RTOp_value_type*)op->obj_data) = alpha; 00136 return 0; /* success? */ 00137 } 00138 00139 int RTOp_TOp_axpy_destroy( struct RTOp_RTOp* op ) 00140 { 00141 op->vtbl->obj_data_vtbl->obj_free(NULL,NULL,&op->obj_data); 00142 op->vtbl = NULL; 00143 return 0; /* success? */ 00144 } 00145 00146 int RTOp_TOp_axpy_set_alpha( RTOp_value_type alpha, struct RTOp_RTOp* op ) 00147 { 00148 *((RTOp_value_type*)op->obj_data) = alpha; 00149 return 0; /* success? */ 00150 }
1.7.6.1