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_32_QF_H
       5              : #define PALACE_LIBCEED_HDIV_MASS_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_apply_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 :                    *u = in[2], *curlu = in[3];
      16            0 :   CeedScalar *__restrict__ v = out[0], *__restrict__ curlv = out[1];
      17              : 
      18            0 :   CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
      19              :   {
      20              :     {
      21            0 :       const CeedScalar u_loc[2] = {u[i + Q * 0], u[i + Q * 1]};
      22              :       CeedScalar coeff[9], adjJt_loc[6], v_loc[2];
      23            0 :       CoeffUnpack3((const CeedIntScalar *)ctx, (CeedInt)attr[i], coeff);
      24            0 :       MatUnpack32(adjJt + i, Q, adjJt_loc);
      25              :       MultAtBCx32(adjJt_loc, coeff, adjJt_loc, u_loc, v_loc);
      26              : 
      27            0 :       v[i + Q * 0] = wdetJ[i] * v_loc[0];
      28            0 :       v[i + Q * 1] = wdetJ[i] * v_loc[1];
      29              :     }
      30              :     {
      31              :       const CeedScalar coeff =
      32            0 :           CoeffUnpack1(CoeffPairSecond<3>((const CeedIntScalar *)ctx), (CeedInt)attr[i]);
      33              : 
      34            0 :       curlv[i] = (coeff * qw[i] * qw[i] / wdetJ[i]) * curlu[i];
      35              :     }
      36              :   }
      37            0 :   return 0;
      38              : }
      39              : 
      40              : #endif  // PALACE_LIBCEED_HDIV_MASS_32_QF_H