1 // Auto-generated file. Do not edit!
2 // Template: src/f32-dwconv/up-scalar.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2019 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/dwconv.h>
13 #include <xnnpack/math.h>
14
15
xnn_f32_dwconv_ukernel_up1x4__scalar(size_t channels,size_t output_width,const float ** input,const float * weights,float * output,size_t input_stride,size_t output_increment,const union xnn_f32_output_params params[restrict static1])16 void xnn_f32_dwconv_ukernel_up1x4__scalar(
17 size_t channels,
18 size_t output_width,
19 const float** input,
20 const float* weights,
21 float* output,
22 size_t input_stride,
23 size_t output_increment,
24 const union xnn_f32_output_params params[restrict static 1])
25 {
26 assert(channels != 0);
27 assert(output_width != 0);
28
29 const float vmin = params->scalar.min;
30 const float vmax = params->scalar.max;
31 do {
32 const float* i0 = input[0];
33 assert(i0 != NULL);
34 const float* i1 = input[1];
35 assert(i1 != NULL);
36 const float* i2 = input[2];
37 assert(i2 != NULL);
38 const float* i3 = input[3];
39 assert(i3 != NULL);
40 input = (const float**) ((uintptr_t) input + input_stride);
41
42 size_t c = channels;
43 const float* w = weights;
44 do {
45 float vacc0p0 = w[0];
46
47 const float vi0 = *i0++;
48 const float vk0 = w[1];
49 vacc0p0 += vi0 * vk0;
50
51 const float vi1 = *i1++;
52 const float vk1 = w[2];
53 vacc0p0 += vi1 * vk1;
54
55 const float vi2 = *i2++;
56 const float vk2 = w[3];
57 vacc0p0 += vi2 * vk2;
58
59 const float vi3 = *i3++;
60 const float vk3 = w[4];
61 vacc0p0 += vi3 * vk3;
62
63 w += 5;
64
65
66 float vacc0 = math_max_f32(vacc0p0, vmin);
67 vacc0 = math_min_f32(vacc0, vmax);
68
69 *output++ = vacc0;
70 } while (--c != 0);
71
72 output = (float*) ((uintptr_t) output + output_increment);
73 } while (--output_width != 0);
74 }
75