1 // Auto-generated file. Do not edit!
2 // Template: src/s8-ibilinear/scalar.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2021 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/ibilinear.h>
13 #include <xnnpack/math.h>
14
15
xnn_u8_ibilinear_ukernel__scalar_c1(size_t output_pixels,size_t channels,const uint8_t ** restrict input,size_t input_offset,const int16_t * restrict weights,uint8_t * restrict output,size_t output_increment)16 void xnn_u8_ibilinear_ukernel__scalar_c1(
17 size_t output_pixels,
18 size_t channels,
19 const uint8_t**restrict input,
20 size_t input_offset,
21 const int16_t*restrict weights,
22 uint8_t*restrict output,
23 size_t output_increment)
24 {
25 assert(output_pixels != 0);
26 assert(channels != 0);
27
28 do {
29 const uint8_t* i0 = (const uint8_t*) ((uintptr_t) input[0] + input_offset);
30 const uint8_t* i1 = (const uint8_t*) ((uintptr_t) input[1] + input_offset);
31 const uint8_t* i2 = (const uint8_t*) ((uintptr_t) input[2] + input_offset);
32 const uint8_t* i3 = (const uint8_t*) ((uintptr_t) input[3] + input_offset);
33 input += 4;
34
35 const int32_t valphah = (int32_t) (uint32_t) (uint16_t) weights[0];
36 const int32_t valphav = (int32_t) (uint32_t) (uint16_t) weights[1];
37 weights += 2;
38
39 const int32_t vrounding = INT32_C(0x00200000);
40
41 size_t c = channels;
42 do {
43 const int32_t vtl = (int32_t) *i0++;
44 const int32_t vtr = (int32_t) *i1++;
45 const int32_t vbl = (int32_t) *i2++;
46 const int32_t vbr = (int32_t) *i3++;
47
48 const int32_t vtd = vtr - vtl;
49 const int32_t vbd = vbr - vbl;
50
51 const int32_t vt = (int32_t) ((uint32_t) vtl << 11) + vtd * valphah;
52 const int32_t vb = (int32_t) ((uint32_t) vbl << 11) + vbd * valphah;
53
54 const int32_t vd = vb - vt;
55
56 const int32_t vacc = (int32_t) ((uint32_t) vt << 11) + vd * valphav;
57
58 const int32_t vo = math_asr_s32(vacc + vrounding, 22);
59
60 *output++ = vo;
61
62 c -= sizeof(uint8_t);
63 } while (c != 0);
64
65 output = (uint8_t*) ((uintptr_t) output + output_increment);
66 } while (--output_pixels != 0);
67 }
68