Blame cblas/source_gemv_c.h

Packit 67cb25
/* blas/source_gemv_c.h
Packit 67cb25
 *
Packit 67cb25
 * Copyright (C) 1996, 1997, 1998, 1999, 2000 Gerard Jungman
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;
Packit 67cb25
  INDEX lenX, lenY;
Packit 67cb25
Packit 67cb25
  const BASE alpha_real = CONST_REAL0(alpha);
Packit 67cb25
  const BASE alpha_imag = CONST_IMAG0(alpha);
Packit 67cb25
Packit 67cb25
  const BASE beta_real = CONST_REAL0(beta);
Packit 67cb25
  const BASE beta_imag = CONST_IMAG0(beta);
Packit 67cb25
Packit 67cb25
  CHECK_ARGS12(GEMV,order,TransA,M,N,alpha,A,lda,X,incX,beta,Y,incY);
Packit 67cb25
Packit 67cb25
  if (M == 0 || N == 0)
Packit 67cb25
    return;
Packit 67cb25
Packit 67cb25
  if ((alpha_real == 0.0 && alpha_imag == 0.0)
Packit 67cb25
      && (beta_real == 1.0 && beta_imag == 0.0))
Packit 67cb25
    return;
Packit 67cb25
Packit 67cb25
  if (TransA == CblasNoTrans) {
Packit 67cb25
    lenX = N;
Packit 67cb25
    lenY = M;
Packit 67cb25
  } else {
Packit 67cb25
    lenX = M;
Packit 67cb25
    lenY = N;
Packit 67cb25
  }
Packit 67cb25
Packit 67cb25
  /* form  y := beta*y */
Packit 67cb25
Packit 67cb25
  if (beta_real == 0.0 && beta_imag == 0.0) {
Packit 67cb25
    INDEX iy = OFFSET(lenY, incY);
Packit 67cb25
    for (i = 0; i < lenY; i++) {
Packit 67cb25
      REAL(Y, iy) = 0.0;
Packit 67cb25
      IMAG(Y, iy) = 0.0;
Packit 67cb25
      iy += incY;
Packit 67cb25
    }
Packit 67cb25
  } else if (!(beta_real == 1.0 && beta_imag == 0.0)) {
Packit 67cb25
    INDEX iy = OFFSET(lenY, incY);
Packit 67cb25
    for (i = 0; i < lenY; i++) {
Packit 67cb25
      const BASE y_real = REAL(Y, iy);
Packit 67cb25
      const BASE y_imag = IMAG(Y, iy);
Packit 67cb25
      const BASE tmpR = y_real * beta_real - y_imag * beta_imag;
Packit 67cb25
      const BASE tmpI = y_real * beta_imag + y_imag * beta_real;
Packit 67cb25
      REAL(Y, iy) = tmpR;
Packit 67cb25
      IMAG(Y, iy) = tmpI;
Packit 67cb25
      iy += incY;
Packit 67cb25
    }
Packit 67cb25
  }
Packit 67cb25
Packit 67cb25
  if (alpha_real == 0.0 && alpha_imag == 0.0)
Packit 67cb25
    return;
Packit 67cb25
Packit 67cb25
  if ((order == CblasRowMajor && TransA == CblasNoTrans)
Packit 67cb25
      || (order == CblasColMajor && TransA == CblasTrans)) {
Packit 67cb25
    /* form  y := alpha*A*x + y */
Packit 67cb25
    INDEX iy = OFFSET(lenY, incY);
Packit 67cb25
    for (i = 0; i < lenY; i++) {
Packit 67cb25
      BASE dotR = 0.0;
Packit 67cb25
      BASE dotI = 0.0;
Packit 67cb25
      INDEX ix = OFFSET(lenX, incX);
Packit 67cb25
      for (j = 0; j < lenX; j++) {
Packit 67cb25
        const BASE x_real = CONST_REAL(X, ix);
Packit 67cb25
        const BASE x_imag = CONST_IMAG(X, ix);
Packit 67cb25
        const BASE A_real = CONST_REAL(A, lda * i + j);
Packit 67cb25
        const BASE A_imag = CONST_IMAG(A, lda * i + j);
Packit 67cb25
Packit 67cb25
        dotR += A_real * x_real - A_imag * x_imag;
Packit 67cb25
        dotI += A_real * x_imag + A_imag * x_real;
Packit 67cb25
        ix += incX;
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
      REAL(Y, iy) += alpha_real * dotR - alpha_imag * dotI;
Packit 67cb25
      IMAG(Y, iy) += alpha_real * dotI + alpha_imag * dotR;
Packit 67cb25
      iy += incY;
Packit 67cb25
    }
Packit 67cb25
  } else if ((order == CblasRowMajor && TransA == CblasTrans)
Packit 67cb25
             || (order == CblasColMajor && TransA == CblasNoTrans)) {
Packit 67cb25
    /* form  y := alpha*A'*x + y */
Packit 67cb25
    INDEX ix = OFFSET(lenX, incX);
Packit 67cb25
    for (j = 0; j < lenX; j++) {
Packit 67cb25
      BASE x_real = CONST_REAL(X, ix);
Packit 67cb25
      BASE x_imag = CONST_IMAG(X, ix);
Packit 67cb25
      BASE tmpR = alpha_real * x_real - alpha_imag * x_imag;
Packit 67cb25
      BASE tmpI = alpha_real * x_imag + alpha_imag * x_real;
Packit 67cb25
Packit 67cb25
      INDEX iy = OFFSET(lenY, incY);
Packit 67cb25
      for (i = 0; i < lenY; i++) {
Packit 67cb25
        const BASE A_real = CONST_REAL(A, lda * j + i);
Packit 67cb25
        const BASE A_imag = CONST_IMAG(A, lda * j + i);
Packit 67cb25
        REAL(Y, iy) += A_real * tmpR - A_imag * tmpI;
Packit 67cb25
        IMAG(Y, iy) += A_real * tmpI + A_imag * tmpR;
Packit 67cb25
        iy += incY;
Packit 67cb25
      }
Packit 67cb25
      ix += incX;
Packit 67cb25
    }
Packit 67cb25
  } else if (order == CblasRowMajor && TransA == CblasConjTrans) {
Packit 67cb25
    /* form  y := alpha*A^H*x + y */
Packit 67cb25
    INDEX ix = OFFSET(lenX, incX);
Packit 67cb25
    for (j = 0; j < lenX; j++) {
Packit 67cb25
      BASE x_real = CONST_REAL(X, ix);
Packit 67cb25
      BASE x_imag = CONST_IMAG(X, ix);
Packit 67cb25
      BASE tmpR = alpha_real * x_real - alpha_imag * x_imag;
Packit 67cb25
      BASE tmpI = alpha_real * x_imag + alpha_imag * x_real;
Packit 67cb25
Packit 67cb25
      INDEX iy = OFFSET(lenY, incY);
Packit 67cb25
      for (i = 0; i < lenY; i++) {
Packit 67cb25
        const BASE A_real = CONST_REAL(A, lda * j + i);
Packit 67cb25
        const BASE A_imag = CONST_IMAG(A, lda * j + i);
Packit 67cb25
        REAL(Y, iy) += A_real * tmpR - (-A_imag) * tmpI;
Packit 67cb25
        IMAG(Y, iy) += A_real * tmpI + (-A_imag) * tmpR;
Packit 67cb25
        iy += incY;
Packit 67cb25
      }
Packit 67cb25
      ix += incX;
Packit 67cb25
    }
Packit 67cb25
  } else if (order == CblasColMajor && TransA == CblasConjTrans) {
Packit 67cb25
    /* form  y := alpha*A^H*x + y */
Packit 67cb25
    INDEX iy = OFFSET(lenY, incY);
Packit 67cb25
    for (i = 0; i < lenY; i++) {
Packit 67cb25
      BASE dotR = 0.0;
Packit 67cb25
      BASE dotI = 0.0;
Packit 67cb25
      INDEX ix = OFFSET(lenX, incX);
Packit 67cb25
      for (j = 0; j < lenX; j++) {
Packit 67cb25
        const BASE x_real = CONST_REAL(X, ix);
Packit 67cb25
        const BASE x_imag = CONST_IMAG(X, ix);
Packit 67cb25
        const BASE A_real = CONST_REAL(A, lda * i + j);
Packit 67cb25
        const BASE A_imag = CONST_IMAG(A, lda * i + j);
Packit 67cb25
Packit 67cb25
        dotR += A_real * x_real - (-A_imag) * x_imag;
Packit 67cb25
        dotI += A_real * x_imag + (-A_imag) * x_real;
Packit 67cb25
        ix += incX;
Packit 67cb25
      }
Packit 67cb25
Packit 67cb25
      REAL(Y, iy) += alpha_real * dotR - alpha_imag * dotI;
Packit 67cb25
      IMAG(Y, iy) += alpha_real * dotI + alpha_imag * dotR;
Packit 67cb25
      iy += incY;
Packit 67cb25
    }
Packit 67cb25
  } else {
Packit 67cb25
    BLAS_ERROR("unrecognized operation");
Packit 67cb25
  }
Packit 67cb25
Packit 67cb25
}