示例#1
0
/*
 *             Automatically Tuned Linear Algebra Software v3.10.2
 *                    (C) Copyright 1999 R. Clint Whaley
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *   1. Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *   2. Redistributions in binary form must reproduce the above copyright
 *      notice, this list of conditions, and the following disclaimer in the
 *      documentation and/or other materials provided with the distribution.
 *   3. The name of the ATLAS group or the names of its contributers may
 *      not be used to endorse or promote products derived from this
 *      software without specific written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE ATLAS GROUP OR ITS CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 *
 */
#ifdef TCPLX
static int ATL_potrf2(const int N, TYPE *A, const int lda)
{
   int j, k;
   static const TYPE one[2] = {ATL_rone, ATL_rzero};
   static const TYPE none[2] = {ATL_rnone, ATL_rzero};
   const size_t lda2 = lda+lda;
   TYPE Ajj, *Ac=A, *An=A+lda2;

   for (k=j=0; j != N; j++, k += 2)
   {
      Ajj = Ac[k] - llt_dot(k, Ac, 1, Ac, 1);
      Ac[k+1] = ATL_rzero;
      if (Ajj > ATL_rzero)
      {
         Ac[k] = Ajj = sqrt(Ajj);
         if (j != N-1)
         {
            llt_scal(j, ATL_rnone, Ac+1, 2);
            cblas_gemv(CblasColMajor, CblasTrans, j, N-j-1, none,
                       An, lda, Ac, 1, one, An+k, lda);
            llt_scal(j, ATL_rnone, Ac+1, 2);
            llt_rscal(N-j-1, ATL_rone/Ajj, An+k, lda);
            Ac = An;
            An += lda2;
         }
      }
      else
      {
         Ac[k] = Ajj;
         return(j+1);
      }
   }
   return(0);
}
#else  /* real version */
static int ATL_potrf2(const int N, TYPE *A, const int lda)
{
   int j;
   TYPE Ajj, *Ac=A, *An=A+lda;

   for (j=0; j != N; j++)
   {
      Ajj = Ac[j] - cblas_dot(j, Ac, 1, Ac, 1);
      if (Ajj > ATL_rzero)
      {
         Ac[j] = Ajj = sqrt(Ajj);
         if (j != N-1)
         {
            cblas_gemv(CblasColMajor, CblasTrans, j, N-j-1, ATL_rnone,
                       An, lda, Ac, 1, ATL_rone, An+j, lda);
            cblas_scal(N-j-1, ATL_rone/Ajj, An+j, lda);
            Ac = An;
            An += lda;
         }
      }
      else
      {
         Ac[j] = Ajj;
         return(j+1);
      }
   }
   return(0);
}
示例#2
0
/*
 *             Automatically Tuned Linear Algebra Software v3.11.28
 *                    (C) Copyright 1999 R. Clint Whaley
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *   1. Redistributions of source code must retain the above copyright
 *      notice, this list of conditions and the following disclaimer.
 *   2. Redistributions in binary form must reproduce the above copyright
 *      notice, this list of conditions, and the following disclaimer in the
 *      documentation and/or other materials provided with the distribution.
 *   3. The name of the ATLAS group or the names of its contributers may
 *      not be used to endorse or promote products derived from this
 *      software without specific written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE ATLAS GROUP OR ITS CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 *
 */
static int ATL_potrf2(const int N, TYPE *A, const int lda)
{
#ifdef TREAL
    TYPE Ajj, *Ac=A;
    const int N_1 = N-1;
    int j;

    for (j=0; j != N_1; j++, Ac += lda)
    {
       Ajj = Ac[j] - cblas_dot(j, A+j, lda, A+j, lda);

       if (Ajj > ATL_rzero)
       {
          Ac[j] = Ajj = sqrt(Ajj);
          cblas_gemv(CblasColMajor, CblasNoTrans, N-j-1, j, ATL_rnone,
                     A+j+1, lda, A+j, lda, ATL_rone, Ac+j+1, 1);
          cblas_scal(N-j-1, ATL_rone/Ajj, Ac+j+1, 1);
       }
       else
       {
          Ac[j] = Ajj;
          return(j+1);
       }
    }
    Ajj = Ac[j] - cblas_dot(j, A+j, lda, A+j, lda);
    if (Ajj > ATL_rzero) Ac[j] = Ajj = sqrt(Ajj);
    else
    {
       Ac[j] = Ajj;
       return(N);
    }
#else
    TYPE Ajj, *Ac=A;
    TYPE one[2] = {ATL_rone, ATL_rzero};
    TYPE none[2] = {ATL_rnone, ATL_rzero};
    const int N_1 = N-1, lda2 = lda<<1;
    int j, j2;

    for (j2=j=0; j != N_1; j++, j2 += 2, Ac += lda2)
    {
       Ajj = Ac[j2];
       cblas_dotc_sub(j, A+j2, lda, A+j2, lda, Ac+j2);
       Ajj -= Ac[j2];

       if (Ajj > ATL_rzero)
       {
          Ac[j2] = Ajj = sqrt(Ajj);
          llt_scal(j, ATL_rnone, A+j2+1, lda2);
          cblas_gemv(CblasColMajor, CblasNoTrans, N-j-1, j, none,
                     A+j2+2, lda, A+j2, lda, one, Ac+j2+2, 1);
          llt_scal(j, ATL_rnone, A+j2+1, lda2);
          llt_scal((N-j-1)<<1, ATL_rone/Ajj, Ac+j2+2, 1);
       }
       else
       {
          Ac[j2] = Ajj;
          return(j+1);
       }
    }
    Ajj = Ac[j2];
    cblas_dotc_sub(j, A+j2, lda, A+j2, lda, Ac+j2);
    Ajj -= Ac[j2];
    if (Ajj > ATL_rzero) Ac[j2] = sqrt(Ajj);
    else
    {
       Ac[j2] = Ajj;
       return(N);
    }
#endif
   return(0);
}
示例#3
0
template <typename IndexType, typename ValueType> void proj_to_lp_cpu(csr_matrix<IndexType, ValueType> &inputSet,
        ValueType * minNormVector, IndexType inSetDim, IndexType vectorDim, ValueType tollerance) {
    ValueType lpShift = (ValueType) 10; //Установим начальное смещение, оно должно быть очень большим
    //переместим наш политоп в соответствии с этим смещением, это будет означать
    //что мы переместили начало координат в эту точку, что соответствует тому что мы отнимим
    //от вектора b из уравнения Ax <=b следующую величину b + t * A * c (с = [1])
    ValueType* vshift = (ValueType *) malloc(inSetDim * sizeof (ValueType));
    ValueType* t_C = (ValueType *) malloc((vectorDim - 1) * sizeof (ValueType));
    for (IndexType i = 0; i < vectorDim - 1; i++) {
        t_C[i] = -lpShift;
    }

    IndexType maxApprIters = 10;
    ValueType *x_s = new_host_array<ValueType > (vectorDim);
    ValueType *delta_x = new_host_array<ValueType > (vectorDim);
    ValueType* v_t = new_host_array<ValueType > (vectorDim); //- 1); //(ValueType *) malloc((vectorDim -1) * sizeof (ValueType));

    IndexType *kvec = new_host_array<IndexType>(2 * inputSet.num_cols + 1 );
    IndexType *kvec_old = new_host_array<IndexType>(2 * inputSet.num_cols + 1 );
    IndexType *basisVecInx = new_host_array<IndexType>(2 * inputSet.num_cols + 1);
    IndexType baselen = 0;

    for (IndexType i = 0; i < 2 * inputSet.num_cols + 1; i++) {
        kvec[i] = 0;
        kvec_old[i] = 0;
    }

    //Начальный x_s это ноль
    cblas_scal(vectorDim, 0.0, x_s, 1);
    for (IndexType apprIt = 0; apprIt < maxApprIters; apprIt++) {
        //Получаем новое B
        cblas_copy(vectorDim - 1, t_C, 1, v_t, 1);
        cblas_axpy(vectorDim - 1, -1.0, x_s, 1, v_t, 1);
        //std::cout << "Vector V_t\n";
        //printMatrixCPU(vectorDim - 1, 1, v_t);
        cblas_copy(vectorDim, x_s, 1, delta_x, 1);
        inputSet.num_rows--;
        spmv_csr_t_serial_host(inputSet, v_t, vshift);
        inputSet.num_rows++;

        //printMatrixCPU(inSetDim, 1, vshift);
        incValsInLstMtxRowVec(inputSet, inputSet.num_rows - 1, vshift);

        std::cout << "Input set on iteration\n";
        //print_csr_matrix(inputSet);
        //getProjectionOnConus(inputSet, minNormVector, inSetDim, vectorDim, tollerance);
        csr_matrix<IndexType, ValueType> inSetCopy = getCsrMtxCopy(inputSet);

        if(apprIt > 0){
            memcpy(kvec_old, kvec, (2 * inputSet.num_cols + 1) * sizeof (IndexType));
        }
        //print_csr_matrix(inSetCopy);
        //projOnFixedSimplex(inSetCopy, minNormVector, kvec, basisVecInx, baselen, tollerance );
        getMinNormElemOutRepr(in_a_csc, minNormVector, 0.0001, kvec,  basisVecInx, baselen, numberOfEqConstr);
        delete_host_matrix(inSetCopy);

        IndexType kvec_razn = 0;
        if(apprIt > 0){
            for (IndexType i = 0; i < 2 * inputSet.num_cols + 1; i++) {
                if(kvec[i] == kvec_old[i] && kvec[i] == 1){
                    kvec_razn++;
                }
            }
        }
        std::cout << "KVEC RAZN is = " << kvec_razn <<"\n";
        std::cout << "baselen is = " << baselen <<"\n";

        cblas_axpy(vectorDim, -1.0, v_t, 1, minNormVector, 1);
        //std::cout << "Min norm vector on iteration\n";
        //printMatrixCPU(vectorDim, 1, minNormVector);

        decValsInLstMtxRowVec(inputSet, inputSet.num_rows - 1, vshift);


        cblas_copy(vectorDim, minNormVector, 1, x_s, 1);

        cblas_axpy(vectorDim, -1.0, x_s, 1, delta_x, 1);
        ValueType z_summ = 0.0;        
        for (IndexType i = 0; i < vectorDim - 1; i++) {
            z_summ += minNormVector[i];
        }
        std::cout << "Summ of elements " << z_summ << " on "<< apprIt <<" iteration\n";
        ValueType dist = cblas_dot(vectorDim - 1, delta_x, 1, delta_x, 1);
        if (dist < tollerance * tollerance) {
            std::cout << "iterations count :" << apprIt << "\n";
            break;
        }       
    }

    //cblas_axpy(vectorDim, -1.0, v_t, 1, minNormVector, 1);
    ValueType dist = cblas_dot(vectorDim - 1, minNormVector, 1, minNormVector, 1);
    std::cout << "Min Vector Lengh = " << sqrt(dist) << "\n";
    ValueType z_summ = 0.0;
    IndexType nonzer_summ = 0;
    for (IndexType i = 0; i < vectorDim - 1; i++) {
        z_summ += minNormVector[i];
        if (minNormVector[i] != 0.0) {
            nonzer_summ++;
        }
    }

    std::cout << "Summ of elements " << z_summ << " \n";
    std::cout << "Count of nonzerros " << nonzer_summ << " \n";
    //printMatrixToFileForOctave(vectorDim, 1, minNormVector);
    ValueType* b_contr = new_host_array<ValueType > (inSetDim);
    inputSet.num_rows--;
    spmv_csr_t_serial_host(inputSet, minNormVector, b_contr);
    inputSet.num_rows++;
    //printMatrixCPU(inSetDim, 1, b_contr);
    IndexType b_begin = inputSet.Ap[vectorDim - 1];
    IndexType b_end = inputSet.Ap[vectorDim];
    IndexType inconsCount = 0;
    for (IndexType i = b_begin; i < b_end; i++) {
        IndexType j = inputSet.Aj[i];
        if (b_contr[j] > inputSet.Ax[i] + tollerance * tollerance) {
            //std::cout << "b_contr " << b_contr[j] << " Ax " << inputSet.Ax[i] << " \n";
            ValueType razn = b_contr[j] -  inputSet.Ax[i];
            //printf("bcontr[%i] %e Ax %e  razn %e\n", j ,b_contr[j], inputSet.Ax[i], razn);
            inconsCount++;
        }
    }
    std::cout << "Inconsistent X count: " << inconsCount << " \n";

}