Coverage Report

Created: 2025-06-22 06:45

/src/fftw3/rdft/scalar/r2cb/r2cbIII_4.c
Line
Count
Source (jump to first uncovered line)
1
/*
2
 * Copyright (c) 2003, 2007-14 Matteo Frigo
3
 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
4
 *
5
 * This program is free software; you can redistribute it and/or modify
6
 * it under the terms of the GNU General Public License as published by
7
 * the Free Software Foundation; either version 2 of the License, or
8
 * (at your option) any later version.
9
 *
10
 * This program is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13
 * GNU General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU General Public License
16
 * along with this program; if not, write to the Free Software
17
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
18
 *
19
 */
20
21
/* This file was automatically generated --- DO NOT EDIT */
22
/* Generated on Sun Jun 22 06:44:45 UTC 2025 */
23
24
#include "rdft/codelet-rdft.h"
25
26
#if defined(ARCH_PREFERS_FMA) || defined(ISA_EXTENSION_PREFERS_FMA)
27
28
/* Generated by: ../../../genfft/gen_r2cb.native -fma -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include rdft/scalar/r2cbIII.h */
29
30
/*
31
 * This function contains 6 FP additions, 4 FP multiplications,
32
 * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
33
 * 9 stack variables, 2 constants, and 8 memory accesses
34
 */
35
#include "rdft/scalar/r2cbIII.h"
36
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)
38
{
39
     DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
40
     DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
41
     {
42
    INT i;
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)) {
44
         E T1, T2, T3, T4, T5, T6;
45
         T1 = Cr[0];
46
         T2 = Cr[WS(csr, 1)];
47
         T3 = T1 - T2;
48
         T4 = Ci[0];
49
         T5 = Ci[WS(csi, 1)];
50
         T6 = T4 + T5;
51
         R0[0] = KP2_000000000 * (T1 + T2);
52
         R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
53
         R1[0] = KP1_414213562 * (T3 - T6);
54
         R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
55
    }
56
     }
57
}
58
59
static const kr2c_desc desc = { 4, "r2cbIII_4", { 6, 4, 0, 0 }, &GENUS };
60
61
void X(codelet_r2cbIII_4) (planner *p) { X(kr2c_register) (p, r2cbIII_4, &desc);
62
}
63
64
#else
65
66
/* Generated by: ../../../genfft/gen_r2cb.native -compact -variables 4 -pipeline-latency 4 -sign 1 -n 4 -name r2cbIII_4 -dft-III -include rdft/scalar/r2cbIII.h */
67
68
/*
69
 * This function contains 6 FP additions, 4 FP multiplications,
70
 * (or, 6 additions, 4 multiplications, 0 fused multiply/add),
71
 * 9 stack variables, 2 constants, and 8 memory accesses
72
 */
73
#include "rdft/scalar/r2cbIII.h"
74
75
static void r2cbIII_4(R *R0, R *R1, R *Cr, R *Ci, stride rs, stride csr, stride csi, INT v, INT ivs, INT ovs)
76
0
{
77
0
     DK(KP1_414213562, +1.414213562373095048801688724209698078569671875);
78
0
     DK(KP2_000000000, +2.000000000000000000000000000000000000000000000);
79
0
     {
80
0
    INT i;
81
0
    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)) {
82
0
         E T1, T2, T3, T4, T5, T6;
83
0
         T1 = Cr[0];
84
0
         T2 = Cr[WS(csr, 1)];
85
0
         T3 = T1 - T2;
86
0
         T4 = Ci[0];
87
0
         T5 = Ci[WS(csi, 1)];
88
0
         T6 = T4 + T5;
89
0
         R0[0] = KP2_000000000 * (T1 + T2);
90
0
         R0[WS(rs, 1)] = KP2_000000000 * (T5 - T4);
91
0
         R1[0] = KP1_414213562 * (T3 - T6);
92
0
         R1[WS(rs, 1)] = -(KP1_414213562 * (T3 + T6));
93
0
    }
94
0
     }
95
0
}
96
97
static const kr2c_desc desc = { 4, "r2cbIII_4", { 6, 4, 0, 0 }, &GENUS };
98
99
1
void X(codelet_r2cbIII_4) (planner *p) { X(kr2c_register) (p, r2cbIII_4, &desc);
100
1
}
101
102
#endif