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
|
/* SPDX-License-Identifier: BSD-3-Clause
* Copyright (c) 2023 Nokia
*/
#include <odp/api/ml_quantize.h>
#include <odp_debug_internal.h>
#include <odp_macros_internal.h>
#include <odp_ml_fp16.h>
#include <math.h>
#include <stdint.h>
void odp_ml_fp32_to_uint8(uint8_t *u8, const float *fp32, uint32_t num, float scale,
uint8_t zerop)
{
float fval;
_ODP_ASSERT(scale != 0);
for (uint32_t i = 0; i < num; i++) {
/* Range mapping: map real values to signed integer */
fval = nearbyintf(fp32[i] / scale) + (float)zerop;
/* clip */
fval = _ODP_MAX(fval, 0.f);
fval = _ODP_MIN(fval, 255.f);
u8[i] = (uint8_t)(int32_t)fval;
}
}
void odp_ml_fp32_from_uint8(float *fp32, const uint8_t *u8, uint32_t num, float scale,
uint8_t zerop)
{
for (uint32_t i = 0; i < num; i++)
fp32[i] = (float)(u8[i] - zerop) * scale;
}
void odp_ml_fp32_to_int8(int8_t *i8, const float *fp32, uint32_t num, float scale, int8_t zerop)
{
float fval;
_ODP_ASSERT(scale != 0);
for (uint32_t i = 0; i < num; i++) {
/* Range mapping: map real values to signed integer */
fval = nearbyintf(fp32[i] / scale) + (float)zerop;
/* NOTE: Clamps signed quantization values to [-127,127] instead of [-128,127].
* This is to ensure that symmetric quantization results in a zero
* point of exactly 0 for signed 8 bit ints.
*/
fval = _ODP_MAX(fval, -127.f);
fval = _ODP_MIN(fval, 127.f);
i8[i] = (int8_t)(int32_t)fval;
}
}
void odp_ml_fp32_from_int8(float *fp32, const int8_t *i8, uint32_t num, float scale, int8_t zerop)
{
for (uint32_t i = 0; i < num; i++)
fp32[i] = (float)(i8[i] - zerop) * scale;
}
void odp_ml_fp32_to_fp16(uint16_t *fp16, const float *fp32, uint32_t num)
{
uint32_t i;
for (i = 0; i < num; i++)
fp16[i] = _odp_float32_to_float16(fp32[i]);
}
void odp_ml_fp32_from_fp16(float *fp32, const uint16_t *fp16, uint32_t num)
{
uint32_t i;
for (i = 0; i < num; i++)
fp32[i] = _odp_float16_to_float32(fp16[i]);
}
|
