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_HCURL_HDIV_33_QF_H
       5              : #define PALACE_LIBCEED_HCURL_HDIV_33_QF_H
       6              : 
       7              : #include "../coeff/coeff_3_qf.h"
       8              : #include "utils_33_qf.h"
       9              : 
      10        47339 : CEED_QFUNCTION(f_apply_hcurlhdiv_33)(void *__restrict__ ctx, CeedInt Q,
      11              :                                      const CeedScalar *const *in, CeedScalar *const *out)
      12              : {
      13        47339 :   const CeedScalar *attr = in[0], *wdetJ = in[0] + Q, *adjJt = in[0] + 2 * Q, *u = in[1];
      14        47339 :   CeedScalar *v = out[0];
      15              : 
      16     14847067 :   CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
      17              :   {
      18     14799728 :     const CeedScalar u_loc[3] = {u[i + Q * 0], u[i + Q * 1], u[i + Q * 2]};
      19              :     CeedScalar coeff[9], adjJt_loc[9], J_loc[9], v_loc[3];
      20     14799728 :     CoeffUnpack3((const CeedIntScalar *)ctx, (CeedInt)attr[i], coeff);
      21     14799728 :     MatUnpack33(adjJt + i, Q, adjJt_loc);
      22              :     AdjJt33(adjJt_loc, J_loc);
      23              :     MultAtBCx33(J_loc, coeff, adjJt_loc, u_loc, v_loc);
      24              : 
      25     14799728 :     v[i + Q * 0] = wdetJ[i] * v_loc[0];
      26     14799728 :     v[i + Q * 1] = wdetJ[i] * v_loc[1];
      27     14799728 :     v[i + Q * 2] = wdetJ[i] * v_loc[2];
      28              :   }
      29        47339 :   return 0;
      30              : }
      31              : 
      32        35431 : CEED_QFUNCTION(f_apply_hdivhcurl_33)(void *__restrict__ ctx, CeedInt Q,
      33              :                                      const CeedScalar *const *in, CeedScalar *const *out)
      34              : {
      35        35431 :   const CeedScalar *attr = in[0], *wdetJ = in[0] + Q, *adjJt = in[0] + 2 * Q, *u = in[1];
      36        35431 :   CeedScalar *v = out[0];
      37              : 
      38     11029831 :   CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
      39              :   {
      40     10994400 :     const CeedScalar u_loc[3] = {u[i + Q * 0], u[i + Q * 1], u[i + Q * 2]};
      41              :     CeedScalar coeff[9], adjJt_loc[9], J_loc[9], v_loc[3];
      42     10994400 :     CoeffUnpack3((const CeedIntScalar *)ctx, (CeedInt)attr[i], coeff);
      43     10994400 :     MatUnpack33(adjJt + i, Q, adjJt_loc);
      44              :     AdjJt33(adjJt_loc, J_loc);
      45              :     MultAtBCx33(adjJt_loc, coeff, J_loc, u_loc, v_loc);
      46              : 
      47     10994400 :     v[i + Q * 0] = wdetJ[i] * v_loc[0];
      48     10994400 :     v[i + Q * 1] = wdetJ[i] * v_loc[1];
      49     10994400 :     v[i + Q * 2] = wdetJ[i] * v_loc[2];
      50              :   }
      51        35431 :   return 0;
      52              : }
      53              : 
      54              : #endif  // PALACE_LIBCEED_HCURL_HDIV_33_QF_H