1 // Auto-generated file. Do not edit!
2 // Template: src/qs8-vlrelu/scalar-select.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
12 #include <xnnpack/math.h>
13 #include <xnnpack/vlrelu.h>
14
15
xnn_qu8_vlrelu_ukernel__scalar_select_x2(size_t n,const uint8_t * x,uint8_t * y,const union xnn_qu8_lrelu_params params[restrict XNN_MIN_ELEMENTS (1)])16 void xnn_qu8_vlrelu_ukernel__scalar_select_x2(
17 size_t n,
18 const uint8_t* x,
19 uint8_t* y,
20 const union xnn_qu8_lrelu_params params[restrict XNN_MIN_ELEMENTS(1)])
21 {
22 const int32_t vinput_zero_point = params->scalar_select.input_zero_point;
23 const int32_t vpositive_multiplier = params->scalar_select.positive_multiplier;
24 const int32_t vnegative_multiplier = params->scalar_select.negative_multiplier;
25 const int32_t vbias = params->scalar_select.bias;
26 for (; n >= 2 * sizeof(uint8_t); n -= 2 * sizeof(uint8_t)) {
27 int32_t vacc0 = (int32_t) x[0];
28 int32_t vacc1 = (int32_t) x[1];
29 x += 2;
30
31 vacc0 -= vinput_zero_point;
32 vacc1 -= vinput_zero_point;
33
34 const int32_t vmultiplier0 = XNN_UNPREDICTABLE(vacc0 >= 0) ? vpositive_multiplier : vnegative_multiplier;
35 const int32_t vmultiplier1 = XNN_UNPREDICTABLE(vacc1 >= 0) ? vpositive_multiplier : vnegative_multiplier;
36
37 vacc0 = vbias + vacc0 * vmultiplier0;
38 vacc1 = vbias + vacc1 * vmultiplier1;
39
40 int32_t vout0 = math_asr_s32(vacc0, 8);
41 int32_t vout1 = math_asr_s32(vacc1, 8);
42
43 vout0 = math_max_s32(vout0, 0);
44 vout1 = math_max_s32(vout1, 0);
45
46 vout0 = math_min_s32(vout0, 255);
47 vout1 = math_min_s32(vout1, 255);
48
49 y[0] = (uint8_t) vout0;
50 y[1] = (uint8_t) vout1;
51 y += 2;
52 }
53 if XNN_UNLIKELY(n != 0) {
54 int32_t vacc = (int32_t) *x++ - vinput_zero_point;
55 const int32_t vmultiplier = XNN_UNPREDICTABLE(vacc >= 0) ? vpositive_multiplier : vnegative_multiplier;
56 vacc = vbias + vacc * vmultiplier;
57
58 int32_t vout = math_asr_s32(vacc, 8);
59 vout = math_max_s32(vout, 0);
60 vout = math_min_s32(vout, 255);
61 *y = (uint8_t) vout;
62 }
63 }
64