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_minmax_ukernel_up1x3__scalar_acc2(size_t channels,size_t output_width,const float ** input,const float * weights,float * output,size_t input_stride,size_t output_increment,size_t input_offset,const float * zero,const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS (1)])16 void xnn_f32_dwconv_minmax_ukernel_up1x3__scalar_acc2(
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 size_t input_offset,
25 const float* zero,
26 const union xnn_f32_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
27 {
28 assert(channels != 0);
29 assert(output_width != 0);
30
31 const float vmin = params->scalar.min;
32 const float vmax = params->scalar.max;
33 do {
34 const float* i0 = input[0];
35 assert(i0 != NULL);
36 if XNN_UNPREDICTABLE(i0 != zero) {
37 i0 = (const float*) ((uintptr_t) i0 + input_offset);
38 }
39 const float* i1 = input[1];
40 assert(i1 != NULL);
41 if XNN_UNPREDICTABLE(i1 != zero) {
42 i1 = (const float*) ((uintptr_t) i1 + input_offset);
43 }
44 const float* i2 = input[2];
45 assert(i2 != NULL);
46 if XNN_UNPREDICTABLE(i2 != zero) {
47 i2 = (const float*) ((uintptr_t) i2 + input_offset);
48 }
49 input = (const float**) ((uintptr_t) input + input_stride);
50
51 size_t c = channels;
52 const float* w = weights;
53 do {
54 float vacc0p0 = w[0];
55
56 const float vi0 = *i0++;
57 const float vk0 = w[1];
58 vacc0p0 = math_muladd_f32(vi0, vk0, vacc0p0);
59
60 const float vi1 = *i1++;
61 const float vk1 = w[2];
62 float vacc0p1 = vi1 * vk1;
63
64 const float vi2 = *i2++;
65 const float vk2 = w[3];
66 vacc0p0 = math_muladd_f32(vi2, vk2, vacc0p0);
67
68 w += 4;
69
70 vacc0p0 += vacc0p1;
71
72 float vacc0 = math_max_f32(vacc0p0, vmin);
73 vacc0 = math_min_f32(vacc0, vmax);
74 *output++ = vacc0;
75 } while (--c != 0);
76
77 output = (float*) ((uintptr_t) output + output_increment);
78 } while (--output_width != 0);
79 }
80