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_MASS_BUILD_32_QF_H
       5              : #define PALACE_LIBCEED_HCURL_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_hcurlmass_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;
      15            0 :   CeedScalar *__restrict__ qd1 = out[0], *__restrict__ qd2 = out[0] + Q;
      16              : 
      17            0 :   CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
      18              :   {
      19              :     {
      20            0 :       const CeedScalar coeff = CoeffUnpack1((const CeedIntScalar *)ctx, (CeedInt)attr[i]);
      21              : 
      22            0 :       qd1[i + Q * 0] = coeff * wdetJ[i];
      23              :     }
      24              :     {
      25              :       CeedScalar coeff[9], adjJt_loc[6], qd_loc[4];
      26            0 :       CoeffUnpack3(CoeffPairSecond<1>((const CeedIntScalar *)ctx), (CeedInt)attr[i], coeff);
      27            0 :       MatUnpack32(adjJt + i, Q, adjJt_loc);
      28              :       MultAtBA32(adjJt_loc, coeff, qd_loc);
      29              : 
      30            0 :       qd2[i + Q * 0] = wdetJ[i] * qd_loc[0];
      31            0 :       qd2[i + Q * 1] = wdetJ[i] * qd_loc[1];
      32            0 :       qd2[i + Q * 2] = wdetJ[i] * qd_loc[2];
      33            0 :       qd2[i + Q * 3] = wdetJ[i] * qd_loc[3];
      34              :     }
      35              :   }
      36            0 :   return 0;
      37              : }
      38              : 
      39              : #endif  // PALACE_LIBCEED_HCURL_MASS_BUILD_32_QF_H