1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
|
//
// Copyright 2011-2012 Ettus Research LLC
// Copyright 2018 Ettus Research, a National Instruments Company
//
// SPDX-License-Identifier: GPL-3.0-or-later
//
#include "convert_common.hpp"
#include <uhd/utils/byteswap.hpp>
#include <emmintrin.h>
using namespace uhd::convert;
DECLARE_CONVERTER(fc64, 1, sc16_item32_le, 1, PRIORITY_SIMD)
{
const fc64_t* input = reinterpret_cast<const fc64_t*>(inputs[0]);
item32_t* output = reinterpret_cast<item32_t*>(outputs[0]);
const __m128d scalar = _mm_set1_pd(scale_factor);
#define convert_fc64_1_to_item32_1_nswap_guts(_al_) \
for (; i + 3 < nsamps; i += 4) { \
/* load from input */ \
__m128d tmp0 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 0)); \
__m128d tmp1 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 1)); \
__m128d tmp2 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 2)); \
__m128d tmp3 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 3)); \
\
/* convert and scale */ \
__m128i tmpi0 = _mm_cvttpd_epi32(_mm_mul_pd(tmp0, scalar)); \
__m128i tmpi1 = _mm_cvttpd_epi32(_mm_mul_pd(tmp1, scalar)); \
__m128i tmpilo = _mm_unpacklo_epi64(tmpi0, tmpi1); \
__m128i tmpi2 = _mm_cvttpd_epi32(_mm_mul_pd(tmp2, scalar)); \
__m128i tmpi3 = _mm_cvttpd_epi32(_mm_mul_pd(tmp3, scalar)); \
__m128i tmpihi = _mm_unpacklo_epi64(tmpi2, tmpi3); \
\
/* pack + swap 16-bit pairs */ \
__m128i tmpi = _mm_packs_epi32(tmpilo, tmpihi); \
tmpi = _mm_shufflelo_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
tmpi = _mm_shufflehi_epi16(tmpi, _MM_SHUFFLE(2, 3, 0, 1)); \
\
/* store to output */ \
_mm_storeu_si128(reinterpret_cast<__m128i*>(output + i), tmpi); \
}
size_t i = 0;
// dispatch according to alignment
if ((size_t(input) & 0xf) == 0) {
convert_fc64_1_to_item32_1_nswap_guts(_)
} else {
convert_fc64_1_to_item32_1_nswap_guts(u_)
}
// convert remainder
xx_to_item32_sc16<uhd::htowx>(input + i, output + i, nsamps - i, scale_factor);
}
DECLARE_CONVERTER(fc64, 1, sc16_item32_be, 1, PRIORITY_SIMD)
{
const fc64_t* input = reinterpret_cast<const fc64_t*>(inputs[0]);
item32_t* output = reinterpret_cast<item32_t*>(outputs[0]);
const __m128d scalar = _mm_set1_pd(scale_factor);
#define convert_fc64_1_to_item32_1_bswap_guts(_al_) \
for (; i + 3 < nsamps; i += 4) { \
/* load from input */ \
__m128d tmp0 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 0)); \
__m128d tmp1 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 1)); \
__m128d tmp2 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 2)); \
__m128d tmp3 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 3)); \
\
/* convert and scale */ \
__m128i tmpi0 = _mm_cvttpd_epi32(_mm_mul_pd(tmp0, scalar)); \
__m128i tmpi1 = _mm_cvttpd_epi32(_mm_mul_pd(tmp1, scalar)); \
__m128i tmpilo = _mm_unpacklo_epi64(tmpi0, tmpi1); \
__m128i tmpi2 = _mm_cvttpd_epi32(_mm_mul_pd(tmp2, scalar)); \
__m128i tmpi3 = _mm_cvttpd_epi32(_mm_mul_pd(tmp3, scalar)); \
__m128i tmpihi = _mm_unpacklo_epi64(tmpi2, tmpi3); \
\
/* pack + byteswap -> byteswap 16 bit words */ \
__m128i tmpi = _mm_packs_epi32(tmpilo, tmpihi); \
tmpi = _mm_or_si128(_mm_srli_epi16(tmpi, 8), _mm_slli_epi16(tmpi, 8)); \
\
/* store to output */ \
_mm_storeu_si128(reinterpret_cast<__m128i*>(output + i), tmpi); \
}
size_t i = 0;
// dispatch according to alignment
if ((size_t(input) & 0xf) == 0) {
convert_fc64_1_to_item32_1_bswap_guts(_)
} else {
convert_fc64_1_to_item32_1_bswap_guts(u_)
}
// convert remainder
xx_to_item32_sc16<uhd::htonx>(input + i, output + i, nsamps - i, scale_factor);
}
DECLARE_CONVERTER(fc64, 1, sc16_chdr, 1, PRIORITY_SIMD)
{
const fc64_t* input = reinterpret_cast<const fc64_t*>(inputs[0]);
sc16_t* output = reinterpret_cast<sc16_t*>(outputs[0]);
const __m128d scalar = _mm_set1_pd(scale_factor);
#define convert_fc64_1_to_chdr_1_guts(_al_) \
for (; i + 3 < nsamps; i += 4) { \
/* load from input */ \
__m128d tmp0 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 0)); \
__m128d tmp1 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 1)); \
__m128d tmp2 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 2)); \
__m128d tmp3 = \
_mm_load##_al_##pd(reinterpret_cast<const double*>(input + i + 3)); \
\
/* convert and scale */ \
__m128i tmpi0 = _mm_cvttpd_epi32(_mm_mul_pd(tmp0, scalar)); \
__m128i tmpi1 = _mm_cvttpd_epi32(_mm_mul_pd(tmp1, scalar)); \
__m128i tmpilo = _mm_unpacklo_epi64(tmpi0, tmpi1); \
__m128i tmpi2 = _mm_cvttpd_epi32(_mm_mul_pd(tmp2, scalar)); \
__m128i tmpi3 = _mm_cvttpd_epi32(_mm_mul_pd(tmp3, scalar)); \
__m128i tmpihi = _mm_unpacklo_epi64(tmpi2, tmpi3); \
\
/* store to output */ \
_mm_storeu_si128( \
reinterpret_cast<__m128i*>(output + i), _mm_packs_epi32(tmpilo, tmpihi)); \
}
size_t i = 0;
// dispatch according to alignment
if ((size_t(input) & 0xf) == 0) {
convert_fc64_1_to_chdr_1_guts(_)
} else {
convert_fc64_1_to_chdr_1_guts(u_)
}
// convert remainder
xx_to_chdr_sc16(input + i, output + i, nsamps - i, scale_factor);
}
|
