1 // Auto-generated file. Do not edit!
2 // Template: src/qs8-vcvt/scalar.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/vcvt.h>
14
15
xnn_qu8_vcvt_ukernel__scalar_x4(size_t n,const uint8_t * x,uint8_t * y,const union xnn_qu8_cvt_params params[restrict XNN_MIN_ELEMENTS (1)])16 void xnn_qu8_vcvt_ukernel__scalar_x4(
17 size_t n,
18 const uint8_t* x,
19 uint8_t* y,
20 const union xnn_qu8_cvt_params params[restrict XNN_MIN_ELEMENTS(1)])
21 {
22 const int32_t vbias = params->scalar.bias;
23 const int32_t vmultiplier = params->scalar.multiplier;
24 for (; n >= 4 * sizeof(uint8_t); n -= 4 * sizeof(uint8_t)) {
25 int32_t vacc0 = x[0];
26 int32_t vacc1 = x[1];
27 int32_t vacc2 = x[2];
28 int32_t vacc3 = x[3];
29 x += 4;
30
31 vacc0 = vbias + vacc0 * vmultiplier;
32 vacc1 = vbias + vacc1 * vmultiplier;
33 vacc2 = vbias + vacc2 * vmultiplier;
34 vacc3 = vbias + vacc3 * vmultiplier;
35
36 int32_t vout0 = math_asr_s32(vacc0, 8);
37 int32_t vout1 = math_asr_s32(vacc1, 8);
38 int32_t vout2 = math_asr_s32(vacc2, 8);
39 int32_t vout3 = math_asr_s32(vacc3, 8);
40
41 vout0 = math_max_s32(vout0, 0);
42 vout1 = math_max_s32(vout1, 0);
43 vout2 = math_max_s32(vout2, 0);
44 vout3 = math_max_s32(vout3, 0);
45
46 vout0 = math_min_s32(vout0, 255);
47 vout1 = math_min_s32(vout1, 255);
48 vout2 = math_min_s32(vout2, 255);
49 vout3 = math_min_s32(vout3, 255);
50
51 y[0] = (uint8_t) vout0;
52 y[1] = (uint8_t) vout1;
53 y[2] = (uint8_t) vout2;
54 y[3] = (uint8_t) vout3;
55 y += 4;
56 }
57 if XNN_UNLIKELY(n != 0) {
58 do {
59 int32_t vacc = *x++;
60 vacc = vbias + vacc * vmultiplier;
61
62 int32_t vout = math_asr_s32(vacc, 8);
63 vout = math_max_s32(vout, 0);
64 vout = math_min_s32(vout, 255);
65 *y++ = (uint8_t) vout;
66
67 n -= sizeof(uint8_t);
68 } while (n != 0);
69 }
70 }
71