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_BUILD_32_QF_H
5 : #define PALACE_LIBCEED_HDIV_BUILD_32_QF_H
6 :
7 : #include "../coeff/coeff_3_qf.h"
8 : #include "utils_32_qf.h"
9 :
10 0 : CEED_QFUNCTION(f_build_hdiv_32)(void *__restrict__ ctx, CeedInt Q,
11 : const CeedScalar *const *in, CeedScalar *const *out)
12 : {
13 0 : const CeedScalar *attr = in[0], *wdetJ = in[0] + Q, *adjJt = in[0] + 2 * Q;
14 0 : CeedScalar *qd = out[0];
15 :
16 0 : CeedPragmaSIMD for (CeedInt i = 0; i < Q; i++)
17 : {
18 : CeedScalar coeff[9], adjJt_loc[6], J_loc[6], qd_loc[4];
19 0 : CoeffUnpack3((const CeedIntScalar *)ctx, (CeedInt)attr[i], coeff);
20 0 : MatUnpack32(adjJt + i, Q, adjJt_loc);
21 : AdjJt32(adjJt_loc, J_loc);
22 : MultAtBA32(J_loc, coeff, qd_loc);
23 :
24 0 : qd[i + Q * 0] = wdetJ[i] * qd_loc[0];
25 0 : qd[i + Q * 1] = wdetJ[i] * qd_loc[1];
26 0 : qd[i + Q * 2] = wdetJ[i] * qd_loc[2];
27 0 : qd[i + Q * 3] = wdetJ[i] * qd_loc[3];
28 : }
29 0 : return 0;
30 : }
31 :
32 : #endif // PALACE_LIBCEED_HDIV_BUILD_32_QF_H
|