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
155
|
#ifndef INCLUDED_volk_32f_s32f_convert_8i_a_H
#define INCLUDED_volk_32f_s32f_convert_8i_a_H
#include <volk/volk_common.h>
#include <inttypes.h>
#include <stdio.h>
#ifdef LV_HAVE_SSE2
#include <emmintrin.h>
/*!
\brief Multiplies each point in the input buffer by the scalar value, then converts the result into a 8 bit integer value
\param inputVector The floating point input data buffer
\param outputVector The 8 bit output data buffer
\param scalar The value multiplied against each point in the input buffer
\param num_points The number of data values to be converted
*/
static inline void volk_32f_s32f_convert_8i_a_sse2(int8_t* outputVector, const float* inputVector, const float scalar, unsigned int num_points){
unsigned int number = 0;
const unsigned int sixteenthPoints = num_points / 16;
const float* inputVectorPtr = (const float*)inputVector;
int8_t* outputVectorPtr = outputVector;
float min_val = -128;
float max_val = 127;
float r;
__m128 vScalar = _mm_set_ps1(scalar);
__m128 inputVal1, inputVal2, inputVal3, inputVal4;
__m128i intInputVal1, intInputVal2, intInputVal3, intInputVal4;
__m128 vmin_val = _mm_set_ps1(min_val);
__m128 vmax_val = _mm_set_ps1(max_val);
for(;number < sixteenthPoints; number++){
inputVal1 = _mm_load_ps(inputVectorPtr); inputVectorPtr += 4;
inputVal2 = _mm_load_ps(inputVectorPtr); inputVectorPtr += 4;
inputVal3 = _mm_load_ps(inputVectorPtr); inputVectorPtr += 4;
inputVal4 = _mm_load_ps(inputVectorPtr); inputVectorPtr += 4;
inputVal1 = _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal1, vScalar), vmax_val), vmin_val);
inputVal2 = _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal2, vScalar), vmax_val), vmin_val);
inputVal3 = _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal3, vScalar), vmax_val), vmin_val);
inputVal4 = _mm_max_ps(_mm_min_ps(_mm_mul_ps(inputVal4, vScalar), vmax_val), vmin_val);
intInputVal1 = _mm_cvtps_epi32(inputVal1);
intInputVal2 = _mm_cvtps_epi32(inputVal2);
intInputVal3 = _mm_cvtps_epi32(inputVal3);
intInputVal4 = _mm_cvtps_epi32(inputVal4);
intInputVal1 = _mm_packs_epi32(intInputVal1, intInputVal2);
intInputVal3 = _mm_packs_epi32(intInputVal3, intInputVal4);
intInputVal1 = _mm_packs_epi16(intInputVal1, intInputVal3);
_mm_store_si128((__m128i*)outputVectorPtr, intInputVal1);
outputVectorPtr += 16;
}
number = sixteenthPoints * 16;
for(; number < num_points; number++){
r = inputVector[number] * scalar;
if(r > max_val)
r = max_val;
else if(r < min_val)
r = min_val;
outputVector[number] = (int8_t)(r);
}
}
#endif /* LV_HAVE_SSE2 */
#ifdef LV_HAVE_SSE
#include <xmmintrin.h>
/*!
\brief Multiplies each point in the input buffer by the scalar value, then converts the result into a 8 bit integer value
\param inputVector The floating point input data buffer
\param outputVector The 8 bit output data buffer
\param scalar The value multiplied against each point in the input buffer
\param num_points The number of data values to be converted
*/
static inline void volk_32f_s32f_convert_8i_a_sse(int8_t* outputVector, const float* inputVector, const float scalar, unsigned int num_points){
unsigned int number = 0;
const unsigned int quarterPoints = num_points / 4;
const float* inputVectorPtr = (const float*)inputVector;
float min_val = -128;
float max_val = 127;
float r;
int8_t* outputVectorPtr = outputVector;
__m128 vScalar = _mm_set_ps1(scalar);
__m128 ret;
__m128 vmin_val = _mm_set_ps1(min_val);
__m128 vmax_val = _mm_set_ps1(max_val);
__VOLK_ATTR_ALIGNED(16) float outputFloatBuffer[4];
for(;number < quarterPoints; number++){
ret = _mm_load_ps(inputVectorPtr);
inputVectorPtr += 4;
ret = _mm_max_ps(_mm_min_ps(_mm_mul_ps(ret, vScalar), vmax_val), vmin_val);
_mm_store_ps(outputFloatBuffer, ret);
*outputVectorPtr++ = (int8_t)(outputFloatBuffer[0]);
*outputVectorPtr++ = (int8_t)(outputFloatBuffer[1]);
*outputVectorPtr++ = (int8_t)(outputFloatBuffer[2]);
*outputVectorPtr++ = (int8_t)(outputFloatBuffer[3]);
}
number = quarterPoints * 4;
for(; number < num_points; number++){
r = inputVector[number] * scalar;
if(r > max_val)
r = max_val;
else if(r < min_val)
r = min_val;
outputVector[number] = (int8_t)(r);
}
}
#endif /* LV_HAVE_SSE */
#ifdef LV_HAVE_GENERIC
/*!
\brief Multiplies each point in the input buffer by the scalar value, then converts the result into a 8 bit integer value
\param inputVector The floating point input data buffer
\param outputVector The 8 bit output data buffer
\param scalar The value multiplied against each point in the input buffer
\param num_points The number of data values to be converted
*/
static inline void volk_32f_s32f_convert_8i_a_generic(int8_t* outputVector, const float* inputVector, const float scalar, unsigned int num_points){
int8_t* outputVectorPtr = outputVector;
const float* inputVectorPtr = inputVector;
unsigned int number = 0;
float min_val = -128;
float max_val = 127;
float r;
for(number = 0; number < num_points; number++){
r = *inputVectorPtr++ * scalar;
if(r > max_val)
r = max_val;
else if(r < min_val)
r = min_val;
*outputVectorPtr++ = (int8_t)(r);
}
}
#endif /* LV_HAVE_GENERIC */
#endif /* INCLUDED_volk_32f_s32f_convert_8i_a_H */
|
