Chris@42
|
1 /*
|
Chris@42
|
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
|
Chris@42
|
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
Chris@42
|
4 *
|
Chris@42
|
5 * This program is free software; you can redistribute it and/or modify
|
Chris@42
|
6 * it under the terms of the GNU General Public License as published by
|
Chris@42
|
7 * the Free Software Foundation; either version 2 of the License, or
|
Chris@42
|
8 * (at your option) any later version.
|
Chris@42
|
9 *
|
Chris@42
|
10 * This program is distributed in the hope that it will be useful,
|
Chris@42
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@42
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@42
|
13 * GNU General Public License for more details.
|
Chris@42
|
14 *
|
Chris@42
|
15 * You should have received a copy of the GNU General Public License
|
Chris@42
|
16 * along with this program; if not, write to the Free Software
|
Chris@42
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@42
|
18 *
|
Chris@42
|
19 */
|
Chris@42
|
20
|
Chris@42
|
21 /* This file was automatically generated --- DO NOT EDIT */
|
Chris@42
|
22 /* Generated on Sat Jul 30 16:50:34 EDT 2016 */
|
Chris@42
|
23
|
Chris@42
|
24 #include "codelet-rdft.h"
|
Chris@42
|
25
|
Chris@42
|
26 #ifdef HAVE_FMA
|
Chris@42
|
27
|
Chris@42
|
28 /* Generated by: ../../../genfft/gen_r2cb.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include r2cbIII.h */
|
Chris@42
|
29
|
Chris@42
|
30 /*
|
Chris@42
|
31 * This function contains 6 FP additions, 4 FP multiplications,
|
Chris@42
|
32 * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
|
Chris@42
|
33 * 9 stack variables, 2 constants, and 8 memory accesses
|
Chris@42
|
34 */
|
Chris@42
|
35 #include "r2cbIII.h"
|
Chris@42
|
36
|
Chris@42
|
37 static void r2cbIII_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
|
Chris@42
|
38 {
|
Chris@42
|
39 DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
|
Chris@42
|
40 DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
|
Chris@42
|
41 {
|
Chris@42
|
42 INT i;
|
Chris@42
|
43 for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
|
Chris@42
|
44 E T1, T2, T4, T5, T3, T6;
|
Chris@42
|
45 T1 = Cr[0];
|
Chris@42
|
46 T2 = Cr[WS(csr, 1)];
|
Chris@42
|
47 T4 = Ci[0];
|
Chris@42
|
48 T5 = Ci[WS(csi, 1)];
|
Chris@42
|
49 R0[0] = KP2_000000000 * (T1 + T2);
|
Chris@42
|
50 T3 = T1 - T2;
|
Chris@42
|
51 R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
|
Chris@42
|
52 T6 = T4 + T5;
|
Chris@42
|
53 R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
|
Chris@42
|
54 R1[0] = KP1_414213562 * (T3 - T6);
|
Chris@42
|
55 }
|
Chris@42
|
56 }
|
Chris@42
|
57 }
|
Chris@42
|
58
|
Chris@42
|
59 static const kr2c_desc desc = { 4, "r2cbIII_4", {6, 4, 0, 0}, &GENUS };
|
Chris@42
|
60
|
Chris@42
|
61 void X(codelet_r2cbIII_4) (planner *p) {
|
Chris@42
|
62 X(kr2c_register) (p, r2cbIII_4, &desc);
|
Chris@42
|
63 }
|
Chris@42
|
64
|
Chris@42
|
65 #else /* HAVE_FMA */
|
Chris@42
|
66
|
Chris@42
|
67 /* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include r2cbIII.h */
|
Chris@42
|
68
|
Chris@42
|
69 /*
|
Chris@42
|
70 * This function contains 6 FP additions, 4 FP multiplications,
|
Chris@42
|
71 * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
|
Chris@42
|
72 * 9 stack variables, 2 constants, and 8 memory accesses
|
Chris@42
|
73 */
|
Chris@42
|
74 #include "r2cbIII.h"
|
Chris@42
|
75
|
Chris@42
|
76 static void r2cbIII_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
|
Chris@42
|
77 {
|
Chris@42
|
78 DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
|
Chris@42
|
79 DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
|
Chris@42
|
80 {
|
Chris@42
|
81 INT i;
|
Chris@42
|
82 for (i = v; i > 0; i = i - 1, R0 = R0 + ovs, R1 = R1 + ovs, Cr = Cr + ivs, Ci = Ci + ivs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
|
Chris@42
|
83 E T1, T2, T3, T4, T5, T6;
|
Chris@42
|
84 T1 = Cr[0];
|
Chris@42
|
85 T2 = Cr[WS(csr, 1)];
|
Chris@42
|
86 T3 = T1 - T2;
|
Chris@42
|
87 T4 = Ci[0];
|
Chris@42
|
88 T5 = Ci[WS(csi, 1)];
|
Chris@42
|
89 T6 = T4 + T5;
|
Chris@42
|
90 R0[0] = KP2_000000000 * (T1 + T2);
|
Chris@42
|
91 R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
|
Chris@42
|
92 R1[0] = KP1_414213562 * (T3 - T6);
|
Chris@42
|
93 R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
|
Chris@42
|
94 }
|
Chris@42
|
95 }
|
Chris@42
|
96 }
|
Chris@42
|
97
|
Chris@42
|
98 static const kr2c_desc desc = { 4, "r2cbIII_4", {6, 4, 0, 0}, &GENUS };
|
Chris@42
|
99
|
Chris@42
|
100 void X(codelet_r2cbIII_4) (planner *p) {
|
Chris@42
|
101 X(kr2c_register) (p, r2cbIII_4, &desc);
|
Chris@42
|
102 }
|
Chris@42
|
103
|
Chris@42
|
104 #endif /* HAVE_FMA */
|