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
|
#ifndef HVX_SCALE_H
#define HVX_SCALE_H
#include <assert.h>
#include <stddef.h>
#include <stdint.h>
#include "hvx-base.h"
#define hvx_scale_f32_loop_body(dst_type, src_type, vec_store) \
do { \
dst_type * restrict vdst = (dst_type *) dst; \
src_type * restrict vsrc = (src_type *) src; \
\
HVX_Vector vs = hvx_vec_splat_f32(scale); \
\
const uint32_t elem_size = sizeof(float); \
const uint32_t epv = 128 / elem_size; \
const uint32_t nvec = n / epv; \
const uint32_t nloe = n % epv; \
\
uint32_t i = 0; \
\
_Pragma("unroll(4)") \
for (; i < nvec; ++i) { \
HVX_Vector v = Q6_Vqf32_vmpy_VsfVsf(vsrc[i], vs); \
vdst[i] = Q6_Vsf_equals_Vqf32(v); \
} \
if (nloe) { \
HVX_Vector v = Q6_Vqf32_vmpy_VsfVsf(vsrc[i], vs); \
vec_store((void *) &vdst[i], nloe * elem_size, Q6_Vsf_equals_Vqf32(v)); \
} \
} while(0)
static inline void hvx_scale_f32_aa(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale) {
assert((size_t) dst % 128 == 0);
assert((size_t) src % 128 == 0);
hvx_scale_f32_loop_body(HVX_Vector, HVX_Vector, hvx_vec_store_a);
}
static inline void hvx_scale_f32_au(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale) {
assert((size_t) dst % 128 == 0);
hvx_scale_f32_loop_body(HVX_Vector, HVX_UVector, hvx_vec_store_a);
}
static inline void hvx_scale_f32_ua(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale) {
assert((size_t) src % 128 == 0);
hvx_scale_f32_loop_body(HVX_UVector, HVX_Vector, hvx_vec_store_u);
}
static inline void hvx_scale_f32_uu(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale) {
hvx_scale_f32_loop_body(HVX_UVector, HVX_UVector, hvx_vec_store_u);
}
static inline void hvx_scale_f32(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale) {
if (((size_t) dst & 127) == 0) {
if (((size_t) src & 127) == 0) {
hvx_scale_f32_aa(dst, src, n, scale);
} else {
hvx_scale_f32_au(dst, src, n, scale);
}
} else {
if (((size_t) src & 127) == 0) {
hvx_scale_f32_ua(dst, src, n, scale);
} else {
hvx_scale_f32_uu(dst, src, n, scale);
}
}
}
#define hvx_scale_offset_f32_loop_body(dst_type, src_type, vec_store) \
do { \
dst_type * restrict vdst = (dst_type *) dst; \
src_type * restrict vsrc = (src_type *) src; \
\
HVX_Vector vs = hvx_vec_splat_f32(scale); \
HVX_Vector vo = hvx_vec_splat_f32(offset); \
\
const uint32_t elem_size = sizeof(float); \
const uint32_t epv = 128 / elem_size; \
const uint32_t nvec = n / epv; \
const uint32_t nloe = n % epv; \
\
uint32_t i = 0; \
\
_Pragma("unroll(4)") \
for (; i < nvec; ++i) { \
HVX_Vector v = Q6_Vqf32_vadd_Vqf32Vsf(Q6_Vqf32_vmpy_VsfVsf(vsrc[i], vs), vo); \
vdst[i] = Q6_Vsf_equals_Vqf32(v); \
} \
if (nloe) { \
HVX_Vector v = Q6_Vqf32_vadd_Vqf32Vsf(Q6_Vqf32_vmpy_VsfVsf(vsrc[i], vs), vo); \
vec_store((void *) &vdst[i], nloe * elem_size, Q6_Vsf_equals_Vqf32(v)); \
} \
} while(0)
static inline void hvx_scale_offset_f32_aa(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale, const float offset) {
assert((size_t) dst % 128 == 0);
assert((size_t) src % 128 == 0);
hvx_scale_offset_f32_loop_body(HVX_Vector, HVX_Vector, hvx_vec_store_a);
}
static inline void hvx_scale_offset_f32_au(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale, const float offset) {
assert((size_t) dst % 128 == 0);
hvx_scale_offset_f32_loop_body(HVX_Vector, HVX_UVector, hvx_vec_store_a);
}
static inline void hvx_scale_offset_f32_ua(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale, const float offset) {
assert((size_t) src % 128 == 0);
hvx_scale_offset_f32_loop_body(HVX_UVector, HVX_Vector, hvx_vec_store_u);
}
static inline void hvx_scale_offset_f32_uu(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale, const float offset) {
hvx_scale_offset_f32_loop_body(HVX_UVector, HVX_UVector, hvx_vec_store_u);
}
static inline void hvx_scale_offset_f32(uint8_t * restrict dst, const uint8_t * restrict src, const int n, const float scale, const float offset) {
if (((size_t) dst & 127) == 0) {
if (((size_t) src & 127) == 0) {
hvx_scale_offset_f32_aa(dst, src, n, scale, offset);
} else {
hvx_scale_offset_f32_au(dst, src, n, scale, offset);
}
} else {
if (((size_t) src & 127) == 0) {
hvx_scale_offset_f32_ua(dst, src, n, scale, offset);
} else {
hvx_scale_offset_f32_uu(dst, src, n, scale, offset);
}
}
}
#endif // HVX_SCALE_H
|
