Blame cblas/source_trmm_c.h

Packit 67cb25
/* blas/source_trmm_c.h
Packit 67cb25
 *
Packit 67cb25
 * Copyright (C) 2001, 2007 Brian Gough
Packit 67cb25
 *
Packit 67cb25
 * This program is free software; you can redistribute it and/or modify
Packit 67cb25
 * it under the terms of the GNU General Public License as published by
Packit 67cb25
 * the Free Software Foundation; either version 3 of the License, or (at
Packit 67cb25
 * your option) any later version.
Packit 67cb25
 *
Packit 67cb25
 * This program is distributed in the hope that it will be useful, but
Packit 67cb25
 * WITHOUT ANY WARRANTY; without even the implied warranty of
Packit 67cb25
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
Packit 67cb25
 * General Public License for more details.
Packit 67cb25
 *
Packit 67cb25
 * You should have received a copy of the GNU General Public License
Packit 67cb25
 * along with this program; if not, write to the Free Software
Packit 67cb25
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
Packit 67cb25
 */
Packit 67cb25
Packit 67cb25
{
Packit 67cb25
  INDEX i, j, k;
Packit 67cb25
  INDEX n1, n2;
Packit 67cb25
Packit 67cb25
  const int nonunit = (Diag == CblasNonUnit);
Packit 67cb25
  const int conj = (TransA == CblasConjTrans) ? -1 : 1;
Packit 67cb25
  int side, uplo, trans;
Packit 67cb25
Packit 67cb25
  CHECK_ARGS12(TRMM,Order,Side,Uplo,TransA,Diag,M,N,alpha,A,lda,B,ldb);
Packit 67cb25
Packit 67cb25
  {
Packit 67cb25
    const BASE alpha_real = CONST_REAL0(alpha);
Packit 67cb25
    const BASE alpha_imag = CONST_IMAG0(alpha);
Packit 67cb25
Packit 67cb25
    if (Order == CblasRowMajor) {
Packit 67cb25
      n1 = M;
Packit 67cb25
      n2 = N;
Packit 67cb25
      side = Side;
Packit 67cb25
      uplo = Uplo;
Packit 67cb25
      trans = (TransA == CblasNoTrans) ? CblasNoTrans : CblasTrans;
Packit 67cb25
    } else {
Packit 67cb25
      n1 = N;
Packit 67cb25
      n2 = M;
Packit 67cb25
      side = (Side == CblasLeft) ? CblasRight : CblasLeft;        /* exchanged */
Packit 67cb25
      uplo = (Uplo == CblasUpper) ? CblasLower : CblasUpper;      /* exchanged */
Packit 67cb25
      trans = (TransA == CblasNoTrans) ? CblasNoTrans : CblasTrans;       /* same */
Packit 67cb25
    }
Packit 67cb25
Packit 67cb25
    if (side == CblasLeft && uplo == CblasUpper && trans == CblasNoTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * TriU(A)*B */
Packit 67cb25
Packit 67cb25
      for (i = 0; i < n1; i++) {
Packit 67cb25
        for (j = 0; j < n2; j++) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Aii_real = CONST_REAL(A, i * lda + i);
Packit 67cb25
            const BASE Aii_imag = conj * CONST_IMAG(A, i * lda + i);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real = Aii_real * Bij_real - Aii_imag * Bij_imag;
Packit 67cb25
            temp_imag = Aii_real * Bij_imag + Aii_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real = REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag = IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          for (k = i + 1; k < n1; k++) {
Packit 67cb25
            const BASE Aik_real = CONST_REAL(A, i * lda + k);
Packit 67cb25
            const BASE Aik_imag = conj * CONST_IMAG(A, i * lda + k);
Packit 67cb25
            const BASE Bkj_real = REAL(B, k * ldb + j);
Packit 67cb25
            const BASE Bkj_imag = IMAG(B, k * ldb + j);
Packit 67cb25
            temp_real += Aik_real * Bkj_real - Aik_imag * Bkj_imag;
Packit 67cb25
            temp_imag += Aik_real * Bkj_imag + Aik_imag * Bkj_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else if (side == CblasLeft && uplo == CblasUpper && trans == CblasTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * (TriU(A))' *B */
Packit 67cb25
Packit 67cb25
      for (i = n1; i > 0 && i--;) {
Packit 67cb25
        for (j = 0; j < n2; j++) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          for (k = 0; k < i; k++) {
Packit 67cb25
            const BASE Aki_real = CONST_REAL(A, k * lda + i);
Packit 67cb25
            const BASE Aki_imag = conj * CONST_IMAG(A, k * lda + i);
Packit 67cb25
            const BASE Bkj_real = REAL(B, k * ldb + j);
Packit 67cb25
            const BASE Bkj_imag = IMAG(B, k * ldb + j);
Packit 67cb25
            temp_real += Aki_real * Bkj_real - Aki_imag * Bkj_imag;
Packit 67cb25
            temp_imag += Aki_real * Bkj_imag + Aki_imag * Bkj_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Aii_real = CONST_REAL(A, i * lda + i);
Packit 67cb25
            const BASE Aii_imag = conj * CONST_IMAG(A, i * lda + i);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real += Aii_real * Bij_real - Aii_imag * Bij_imag;
Packit 67cb25
            temp_imag += Aii_real * Bij_imag + Aii_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real += REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag += IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else if (side == CblasLeft && uplo == CblasLower && trans == CblasNoTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * TriL(A)*B */
Packit 67cb25
Packit 67cb25
Packit 67cb25
      for (i = n1; i > 0 && i--;) {
Packit 67cb25
        for (j = 0; j < n2; j++) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          for (k = 0; k < i; k++) {
Packit 67cb25
            const BASE Aik_real = CONST_REAL(A, i * lda + k);
Packit 67cb25
            const BASE Aik_imag = conj * CONST_IMAG(A, i * lda + k);
Packit 67cb25
            const BASE Bkj_real = REAL(B, k * ldb + j);
Packit 67cb25
            const BASE Bkj_imag = IMAG(B, k * ldb + j);
Packit 67cb25
            temp_real += Aik_real * Bkj_real - Aik_imag * Bkj_imag;
Packit 67cb25
            temp_imag += Aik_real * Bkj_imag + Aik_imag * Bkj_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Aii_real = CONST_REAL(A, i * lda + i);
Packit 67cb25
            const BASE Aii_imag = conj * CONST_IMAG(A, i * lda + i);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real += Aii_real * Bij_real - Aii_imag * Bij_imag;
Packit 67cb25
            temp_imag += Aii_real * Bij_imag + Aii_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real += REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag += IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
Packit 67cb25
Packit 67cb25
    } else if (side == CblasLeft && uplo == CblasLower && trans == CblasTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * TriL(A)' *B */
Packit 67cb25
Packit 67cb25
      for (i = 0; i < n1; i++) {
Packit 67cb25
        for (j = 0; j < n2; j++) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Aii_real = CONST_REAL(A, i * lda + i);
Packit 67cb25
            const BASE Aii_imag = conj * CONST_IMAG(A, i * lda + i);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real = Aii_real * Bij_real - Aii_imag * Bij_imag;
Packit 67cb25
            temp_imag = Aii_real * Bij_imag + Aii_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real = REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag = IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          for (k = i + 1; k < n1; k++) {
Packit 67cb25
            const BASE Aki_real = CONST_REAL(A, k * lda + i);
Packit 67cb25
            const BASE Aki_imag = conj * CONST_IMAG(A, k * lda + i);
Packit 67cb25
            const BASE Bkj_real = REAL(B, k * ldb + j);
Packit 67cb25
            const BASE Bkj_imag = IMAG(B, k * ldb + j);
Packit 67cb25
            temp_real += Aki_real * Bkj_real - Aki_imag * Bkj_imag;
Packit 67cb25
            temp_imag += Aki_real * Bkj_imag + Aki_imag * Bkj_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else if (side == CblasRight && uplo == CblasUpper && trans == CblasNoTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * B * TriU(A) */
Packit 67cb25
Packit 67cb25
      for (i = 0; i < n1; i++) {
Packit 67cb25
        for (j = n2; j > 0 && j--;) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          for (k = 0; k < j; k++) {
Packit 67cb25
            const BASE Akj_real = CONST_REAL(A, k * lda + j);
Packit 67cb25
            const BASE Akj_imag = conj * CONST_IMAG(A, k * lda + j);
Packit 67cb25
            const BASE Bik_real = REAL(B, i * ldb + k);
Packit 67cb25
            const BASE Bik_imag = IMAG(B, i * ldb + k);
Packit 67cb25
            temp_real += Akj_real * Bik_real - Akj_imag * Bik_imag;
Packit 67cb25
            temp_imag += Akj_real * Bik_imag + Akj_imag * Bik_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Ajj_real = CONST_REAL(A, j * lda + j);
Packit 67cb25
            const BASE Ajj_imag = conj * CONST_IMAG(A, j * lda + j);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real += Ajj_real * Bij_real - Ajj_imag * Bij_imag;
Packit 67cb25
            temp_imag += Ajj_real * Bij_imag + Ajj_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real += REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag += IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else if (side == CblasRight && uplo == CblasUpper && trans == CblasTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * B * (TriU(A))' */
Packit 67cb25
Packit 67cb25
      for (i = 0; i < n1; i++) {
Packit 67cb25
        for (j = 0; j < n2; j++) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Ajj_real = CONST_REAL(A, j * lda + j);
Packit 67cb25
            const BASE Ajj_imag = conj * CONST_IMAG(A, j * lda + j);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real = Ajj_real * Bij_real - Ajj_imag * Bij_imag;
Packit 67cb25
            temp_imag = Ajj_real * Bij_imag + Ajj_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real = REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag = IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          for (k = j + 1; k < n2; k++) {
Packit 67cb25
            const BASE Ajk_real = CONST_REAL(A, j * lda + k);
Packit 67cb25
            const BASE Ajk_imag = conj * CONST_IMAG(A, j * lda + k);
Packit 67cb25
            const BASE Bik_real = REAL(B, i * ldb + k);
Packit 67cb25
            const BASE Bik_imag = IMAG(B, i * ldb + k);
Packit 67cb25
            temp_real += Ajk_real * Bik_real - Ajk_imag * Bik_imag;
Packit 67cb25
            temp_imag += Ajk_real * Bik_imag + Ajk_imag * Bik_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else if (side == CblasRight && uplo == CblasLower && trans == CblasNoTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha *B * TriL(A) */
Packit 67cb25
Packit 67cb25
      for (i = 0; i < n1; i++) {
Packit 67cb25
        for (j = 0; j < n2; j++) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Ajj_real = CONST_REAL(A, j * lda + j);
Packit 67cb25
            const BASE Ajj_imag = conj * CONST_IMAG(A, j * lda + j);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real = Ajj_real * Bij_real - Ajj_imag * Bij_imag;
Packit 67cb25
            temp_imag = Ajj_real * Bij_imag + Ajj_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real = REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag = IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          for (k = j + 1; k < n2; k++) {
Packit 67cb25
            const BASE Akj_real = CONST_REAL(A, k * lda + j);
Packit 67cb25
            const BASE Akj_imag = conj * CONST_IMAG(A, k * lda + j);
Packit 67cb25
            const BASE Bik_real = REAL(B, i * ldb + k);
Packit 67cb25
            const BASE Bik_imag = IMAG(B, i * ldb + k);
Packit 67cb25
            temp_real += Akj_real * Bik_real - Akj_imag * Bik_imag;
Packit 67cb25
            temp_imag += Akj_real * Bik_imag + Akj_imag * Bik_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else if (side == CblasRight && uplo == CblasLower && trans == CblasTrans) {
Packit 67cb25
Packit 67cb25
      /* form  B := alpha * B * TriL(A)' */
Packit 67cb25
Packit 67cb25
      for (i = 0; i < n1; i++) {
Packit 67cb25
        for (j = n2; j > 0 && j--;) {
Packit 67cb25
          BASE temp_real = 0.0;
Packit 67cb25
          BASE temp_imag = 0.0;
Packit 67cb25
Packit 67cb25
          for (k = 0; k < j; k++) {
Packit 67cb25
            const BASE Ajk_real = CONST_REAL(A, j * lda + k);
Packit 67cb25
            const BASE Ajk_imag = conj * CONST_IMAG(A, j * lda + k);
Packit 67cb25
            const BASE Bik_real = REAL(B, i * ldb + k);
Packit 67cb25
            const BASE Bik_imag = IMAG(B, i * ldb + k);
Packit 67cb25
            temp_real += Ajk_real * Bik_real - Ajk_imag * Bik_imag;
Packit 67cb25
            temp_imag += Ajk_real * Bik_imag + Ajk_imag * Bik_real;
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          if (nonunit) {
Packit 67cb25
            const BASE Ajj_real = CONST_REAL(A, j * lda + j);
Packit 67cb25
            const BASE Ajj_imag = conj * CONST_IMAG(A, j * lda + j);
Packit 67cb25
            const BASE Bij_real = REAL(B, i * ldb + j);
Packit 67cb25
            const BASE Bij_imag = IMAG(B, i * ldb + j);
Packit 67cb25
            temp_real += Ajj_real * Bij_real - Ajj_imag * Bij_imag;
Packit 67cb25
            temp_imag += Ajj_real * Bij_imag + Ajj_imag * Bij_real;
Packit 67cb25
          } else {
Packit 67cb25
            temp_real += REAL(B, i * ldb + j);
Packit 67cb25
            temp_imag += IMAG(B, i * ldb + j);
Packit 67cb25
          }
Packit 67cb25
Packit 67cb25
          REAL(B, ldb * i + j) = alpha_real * temp_real - alpha_imag * temp_imag;
Packit 67cb25
          IMAG(B, ldb * i + j) = alpha_real * temp_imag + alpha_imag * temp_real;
Packit 67cb25
        }
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
    } else {
Packit 67cb25
      BLAS_ERROR("unrecognized operation");
Packit 67cb25
    }
Packit 67cb25
  }
Packit 67cb25
}