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_22_QF_H
5 : #define PALACE_LIBCEED_HDIV_22_QF_H
6 :
7 : #include "../coeff/coeff_2_qf.h"
8 : #include "utils_22_qf.h"
9 :
10 8015 : CEED_QFUNCTION(f_apply_hdiv_22)(void *__restrict__ ctx, CeedInt Q,
11 : const CeedScalar *const *in, CeedScalar *const *out)
12 : {
13 8015 : const CeedScalar *attr = in[0], *wdetJ = in[0] + Q, *adjJt = in[0] + 2 * Q, *u = in[1];
14 8015 : CeedScalar *v = out[0];
15 :
16 647015 : CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
17 : {
18 639000 : const CeedScalar u_loc[2] = {u[i + Q * 0], u[i + Q * 1]};
19 : CeedScalar coeff[4], adjJt_loc[4], J_loc[4], v_loc[2];
20 639000 : CoeffUnpack2((const CeedIntScalar *)ctx, (CeedInt)attr[i], coeff);
21 639000 : MatUnpack22(adjJt + i, Q, adjJt_loc);
22 : AdjJt22(adjJt_loc, J_loc);
23 : MultAtBCx22(J_loc, coeff, J_loc, u_loc, v_loc);
24 :
25 639000 : v[i + Q * 0] = wdetJ[i] * v_loc[0];
26 639000 : v[i + Q * 1] = wdetJ[i] * v_loc[1];
27 : }
28 8015 : return 0;
29 : }
30 :
31 : #endif // PALACE_LIBCEED_HDIV_22_QF_H
|