1 // Auto-generated file. Do not edit!
2 // Template: src/cs16-fftr/scalar.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2022 Google LLC
6 //
7 // This source code is licensed under the BSD-style license found in the
8 // LICENSE file in the root directory of this source tree.
9
10 #include <assert.h>
11 #include <stddef.h>
12 #include <stdint.h>
13
14 #include <xnnpack/math.h>
15 #include <xnnpack/fft.h>
16
17
xnn_cs16_fftr_ukernel__scalar_x2(size_t samples,int16_t * data,const int16_t * twiddle)18 void xnn_cs16_fftr_ukernel__scalar_x2(
19 size_t samples,
20 int16_t* data,
21 const int16_t* twiddle)
22 {
23 assert(samples >= 2);
24 assert(samples % 2 == 0);
25 assert(data != NULL);
26 assert(data != NULL);
27 assert(twiddle != NULL);
28
29 int16_t* dl = data;
30 int16_t* dr = data + samples * 2;
31 int32_t vdcr = (int32_t) dl[0];
32 int32_t vdci = (int32_t) dl[1];
33
34 vdcr = math_asr_s32(vdcr * 16383 + 16384, 15);
35 vdci = math_asr_s32(vdci * 16383 + 16384, 15);
36
37 dl[0] = vdcr + vdci;
38 dl[1] = 0;
39 dl += 2;
40 dr[0] = vdcr - vdci;
41 dr[1] = 0;
42
43 samples >>= 1;
44
45 for (; samples >= 2; samples -= 2) {
46 dr -= 2 * 2;
47 int32_t vilr0 = dl[0];
48 int32_t vili0 = dl[1];
49 int32_t vilr1 = dl[2];
50 int32_t vili1 = dl[3];
51 int32_t virr0 = (int32_t) dr[2];
52 int32_t viri0 = -(int32_t) dr[3];
53 int32_t virr1 = (int32_t) dr[0];
54 int32_t viri1 = -(int32_t) dr[1];
55 const int32_t vtwr0 = twiddle[0];
56 const int32_t vtwi0 = twiddle[1];
57 const int32_t vtwr1 = twiddle[2];
58 const int32_t vtwi1 = twiddle[3];
59 twiddle += 2 * 2;
60
61 vilr0 = math_asr_s32(vilr0 * 16383 + 16384, 15);
62 virr0 = math_asr_s32(virr0 * 16383 + 16384, 15);
63 vilr1 = math_asr_s32(vilr1 * 16383 + 16384, 15);
64 virr1 = math_asr_s32(virr1 * 16383 + 16384, 15);
65 vili0 = math_asr_s32(vili0 * 16383 + 16384, 15);
66 viri0 = math_asr_s32(viri0 * 16383 + 16384, 15);
67 vili1 = math_asr_s32(vili1 * 16383 + 16384, 15);
68 viri1 = math_asr_s32(viri1 * 16383 + 16384, 15);
69 const int32_t vacc1r0 = vilr0 + virr0;
70 const int32_t vacc2r0 = vilr0 - virr0;
71 const int32_t vacc1r1 = vilr1 + virr1;
72 const int32_t vacc2r1 = vilr1 - virr1;
73 const int32_t vacc1i0 = vili0 + viri0;
74 const int32_t vacc2i0 = vili0 - viri0;
75 const int32_t vacc1i1 = vili1 + viri1;
76 const int32_t vacc2i1 = vili1 - viri1;
77
78 const int32_t twr0 = math_asr_s32(vacc2r0 * vtwr0 - vacc2i0 * vtwi0 + 16384, 15);
79 const int32_t twr1 = math_asr_s32(vacc2r1 * vtwr1 - vacc2i1 * vtwi1 + 16384, 15);
80 const int32_t twi0 = math_asr_s32(vacc2r0 * vtwi0 + vacc2i0 * vtwr0 + 16384, 15);
81 const int32_t twi1 = math_asr_s32(vacc2r1 * vtwi1 + vacc2i1 * vtwr1 + 16384, 15);
82
83 dl[0] = math_asr_s32(vacc1r0 + twr0, 1);
84 dl[1] = math_asr_s32(vacc1i0 + twi0, 1);
85 dl[2] = math_asr_s32(vacc1r1 + twr1, 1);
86 dl[3] = math_asr_s32(vacc1i1 + twi1, 1);
87 dr[2] = math_asr_s32(vacc1r0 - twr0, 1);
88 dr[3] = math_asr_s32(twi0 - vacc1i0, 1);
89 dr[0] = math_asr_s32(vacc1r1 - twr1, 1);
90 dr[1] = math_asr_s32(twi1 - vacc1i1, 1);
91 dl += 2 * 2;
92 }
93
94 if XNN_UNLIKELY(samples != 0) {
95 do {
96 dr -= 2;
97 int32_t vilr = dl[0];
98 int32_t vili = dl[1];
99 int32_t virr = (int32_t) dr[0];
100 int32_t viri = -(int32_t) dr[1];
101 const int32_t vtwr = twiddle[0];
102 const int32_t vtwi = twiddle[1];
103 twiddle += 2;
104
105 vilr = math_asr_s32(vilr * 16383 + 16384, 15);
106 vili = math_asr_s32(vili * 16383 + 16384, 15);
107 virr = math_asr_s32(virr * 16383 + 16384, 15);
108 viri = math_asr_s32(viri * 16383 + 16384, 15);
109 const int32_t vacc1r = vilr + virr;
110 const int32_t vacc1i = vili + viri;
111 const int32_t vacc2r = vilr - virr;
112 const int32_t vacc2i = vili - viri;
113
114 const int32_t twr = math_asr_s32(vacc2r * vtwr - vacc2i * vtwi + 16384, 15);
115 const int32_t twi = math_asr_s32(vacc2r * vtwi + vacc2i * vtwr + 16384, 15);
116
117 dl[0] = math_asr_s32(vacc1r + twr, 1);
118 dl[1] = math_asr_s32(vacc1i + twi, 1);
119 dr[0] = math_asr_s32(vacc1r - twr, 1);
120 dr[1] = math_asr_s32(twi - vacc1i, 1);
121 dl += 2;
122 } while (--samples != 0);
123 }
124 }
125