Chris@10
|
1 /*
|
Chris@10
|
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
|
Chris@10
|
3 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
|
Chris@10
|
4 *
|
Chris@10
|
5 * This program is free software; you can redistribute it and/or modify
|
Chris@10
|
6 * it under the terms of the GNU General Public License as published by
|
Chris@10
|
7 * the Free Software Foundation; either version 2 of the License, or
|
Chris@10
|
8 * (at your option) any later version.
|
Chris@10
|
9 *
|
Chris@10
|
10 * This program is distributed in the hope that it will be useful,
|
Chris@10
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@10
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@10
|
13 * GNU General Public License for more details.
|
Chris@10
|
14 *
|
Chris@10
|
15 * You should have received a copy of the GNU General Public License
|
Chris@10
|
16 * along with this program; if not, write to the Free Software
|
Chris@10
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@10
|
18 *
|
Chris@10
|
19 */
|
Chris@10
|
20
|
Chris@10
|
21 /* This file was automatically generated --- DO NOT EDIT */
|
Chris@10
|
22 /* Generated on Sun Nov 25 07:39:45 EST 2012 */
|
Chris@10
|
23
|
Chris@10
|
24 #include "codelet-rdft.h"
|
Chris@10
|
25
|
Chris@10
|
26 #ifdef HAVE_FMA
|
Chris@10
|
27
|
Chris@10
|
28 /* Generated by: ../../../genfft/gen_r2cf.native -fma -reorder-insns -schedule-for-pipeline -compact -variables 4 -pipeline-latency 4 -n 4 -name r2cf_4 -include r2cf.h */
|
Chris@10
|
29
|
Chris@10
|
30 /*
|
Chris@10
|
31 * This function contains 6 FP additions, 0 FP multiplications,
|
Chris@10
|
32 * (or, 6 additions, 0 multiplications, 0 fused multiply/add),
|
Chris@10
|
33 * 7 stack variables, 0 constants, and 8 memory accesses
|
Chris@10
|
34 */
|
Chris@10
|
35 #include "r2cf.h"
|
Chris@10
|
36
|
Chris@10
|
37 static void r2cf_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
|
Chris@10
|
38 {
|
Chris@10
|
39 {
|
Chris@10
|
40 INT i;
|
Chris@10
|
41 for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
|
Chris@10
|
42 E T1, T2, T4, T5, T3, T6;
|
Chris@10
|
43 T1 = R0[0];
|
Chris@10
|
44 T2 = R0[WS(rs, 1)];
|
Chris@10
|
45 T4 = R1[0];
|
Chris@10
|
46 T5 = R1[WS(rs, 1)];
|
Chris@10
|
47 Cr[WS(csr, 1)] = T1 - T2;
|
Chris@10
|
48 T3 = T1 + T2;
|
Chris@10
|
49 Ci[WS(csi, 1)] = T5 - T4;
|
Chris@10
|
50 T6 = T4 + T5;
|
Chris@10
|
51 Cr[0] = T3 + T6;
|
Chris@10
|
52 Cr[WS(csr, 2)] = T3 - T6;
|
Chris@10
|
53 }
|
Chris@10
|
54 }
|
Chris@10
|
55 }
|
Chris@10
|
56
|
Chris@10
|
57 static const kr2c_desc desc = { 4, "r2cf_4", {6, 0, 0, 0}, &GENUS };
|
Chris@10
|
58
|
Chris@10
|
59 void X(codelet_r2cf_4) (planner *p) {
|
Chris@10
|
60 X(kr2c_register) (p, r2cf_4, &desc);
|
Chris@10
|
61 }
|
Chris@10
|
62
|
Chris@10
|
63 #else /* HAVE_FMA */
|
Chris@10
|
64
|
Chris@10
|
65 /* Generated by: ../../../genfft/gen_r2cf.native -compact -variables 4 -pipeline-latency 4 -n 4 -name r2cf_4 -include r2cf.h */
|
Chris@10
|
66
|
Chris@10
|
67 /*
|
Chris@10
|
68 * This function contains 6 FP additions, 0 FP multiplications,
|
Chris@10
|
69 * (or, 6 additions, 0 multiplications, 0 fused multiply/add),
|
Chris@10
|
70 * 7 stack variables, 0 constants, and 8 memory accesses
|
Chris@10
|
71 */
|
Chris@10
|
72 #include "r2cf.h"
|
Chris@10
|
73
|
Chris@10
|
74 static void r2cf_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
|
Chris@10
|
75 {
|
Chris@10
|
76 {
|
Chris@10
|
77 INT i;
|
Chris@10
|
78 for (i = v; i > 0; i = i - 1, R0 = R0 + ivs, R1 = R1 + ivs, Cr = Cr + ovs, Ci = Ci + ovs, MAKE_VOLATILE_STRIDE(16, rs), MAKE_VOLATILE_STRIDE(16, csr), MAKE_VOLATILE_STRIDE(16, csi)) {
|
Chris@10
|
79 E T1, T2, T3, T4, T5, T6;
|
Chris@10
|
80 T1 = R0[0];
|
Chris@10
|
81 T2 = R0[WS(rs, 1)];
|
Chris@10
|
82 T3 = T1 + T2;
|
Chris@10
|
83 T4 = R1[0];
|
Chris@10
|
84 T5 = R1[WS(rs, 1)];
|
Chris@10
|
85 T6 = T4 + T5;
|
Chris@10
|
86 Cr[WS(csr, 1)] = T1 - T2;
|
Chris@10
|
87 Ci[WS(csi, 1)] = T5 - T4;
|
Chris@10
|
88 Cr[WS(csr, 2)] = T3 - T6;
|
Chris@10
|
89 Cr[0] = T3 + T6;
|
Chris@10
|
90 }
|
Chris@10
|
91 }
|
Chris@10
|
92 }
|
Chris@10
|
93
|
Chris@10
|
94 static const kr2c_desc desc = { 4, "r2cf_4", {6, 0, 0, 0}, &GENUS };
|
Chris@10
|
95
|
Chris@10
|
96 void X(codelet_r2cf_4) (planner *p) {
|
Chris@10
|
97 X(kr2c_register) (p, r2cf_4, &desc);
|
Chris@10
|
98 }
|
Chris@10
|
99
|
Chris@10
|
100 #endif /* HAVE_FMA */
|