Line data    Source code

       1              : // Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
       2              : // SPDX-License-Identifier: Apache-2.0
       3              : 
       4              : #ifndef PALACE_LIBCEED_HDIV_MASS_BUILD_32_QF_H
       5              : #define PALACE_LIBCEED_HDIV_MASS_BUILD_32_QF_H
       6              : 
       7              : #include "../coeff/coeff_1_qf.h"
       8              : #include "../coeff/coeff_3_qf.h"
       9              : #include "utils_32_qf.h"
      10              : 
      11            0 : CEED_QFUNCTION(f_build_hdivmass_32)(void *__restrict__ ctx, CeedInt Q,
      12              :                                     const CeedScalar *const *in, CeedScalar *const *out)
      13              : {
      14            0 :   const CeedScalar *attr = in[0], *wdetJ = in[0] + Q, *adjJt = in[0] + 2 * Q, *qw = in[1];
      15            0 :   CeedScalar *__restrict__ qd1 = out[0], *__restrict__ qd2 = out[0] + 3 * Q;
      16              : 
      17            0 :   CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
      18              :   {
      19              :     {
      20              :       CeedScalar coeff[9], adjJt_loc[6], qd_loc[4];
      21            0 :       CoeffUnpack3((const CeedIntScalar *)ctx, (CeedInt)attr[i], coeff);
      22              :       MatUnpack32(adjJt + i, Q, adjJt_loc);
      23              :       MultAtBA32(adjJt_loc, coeff, qd_loc);
      24              : 
      25            0 :       qd1[i + Q * 0] = wdetJ[i] * qd_loc[0];
      26            0 :       qd1[i + Q * 1] = wdetJ[i] * qd_loc[1];
      27            0 :       qd1[i + Q * 2] = wdetJ[i] * qd_loc[2];
      28            0 :       qd1[i + Q * 3] = wdetJ[i] * qd_loc[3];
      29              :     }
      30              :     {
      31              :       const CeedScalar coeff =
      32            0 :           CoeffUnpack1(CoeffPairSecond<3>((const CeedIntScalar *)ctx), (CeedInt)attr[i]);
      33              : 
      34            0 :       qd2[i] = coeff * qw[i] * qw[i] / wdetJ[i];
      35              :     }
      36              :   }
      37            0 :   return 0;
      38              : }
      39              : 
      40              : #endif  // PALACE_LIBCEED_HDIV_MASS_BUILD_32_QF_H