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_up1x9__wasm_acc2(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_up1x9__wasm_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 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 const float* i4 = input[4];
41 assert(i4 != NULL);
42 const float* i5 = input[5];
43 assert(i5 != NULL);
44 const float* i6 = input[6];
45 assert(i6 != NULL);
46 const float* i7 = input[7];
47 assert(i7 != NULL);
48 const float* i8 = input[8];
49 assert(i8 != NULL);
50 input = (const float**) ((uintptr_t) input + input_stride);
51
52 size_t c = channels;
53 const float* w = weights;
54 do {
55 float vacc0p0 = w[0];
56
57 const float vi0 = *i0++;
58 const float vk0 = w[1];
59 vacc0p0 += vi0 * vk0;
60
61 const float vi1 = *i1++;
62 const float vk1 = w[2];
63 float vacc0p1 = vi1 * vk1;
64
65 const float vi2 = *i2++;
66 const float vk2 = w[3];
67 vacc0p0 += vi2 * vk2;
68
69 const float vi3 = *i3++;
70 const float vk3 = w[4];
71 vacc0p1 += vi3 * vk3;
72
73 const float vi4 = *i4++;
74 const float vk4 = w[5];
75 vacc0p0 += vi4 * vk4;
76
77 const float vi5 = *i5++;
78 const float vk5 = w[6];
79 vacc0p1 += vi5 * vk5;
80
81 const float vi6 = *i6++;
82 const float vk6 = w[7];
83 vacc0p0 += vi6 * vk6;
84
85 const float vi7 = *i7++;
86 const float vk7 = w[8];
87 vacc0p1 += vi7 * vk7;
88
89 const float vi8 = *i8++;
90 const float vk8 = w[9];
91 vacc0p0 += vi8 * vk8;
92
93 w += 10;
94
95 vacc0p0 += vacc0p1;
96
97 float vacc0 = __builtin_wasm_max_f32(vacc0p0, vmin);
98 vacc0 = __builtin_wasm_min_f32(vacc0, vmax);
99
100 *output++ = vacc0;
101 } while (--c != 0);
102
103 output = (float*) ((uintptr_t) output + output_increment);
104 } while (--output_width != 0);
105 }
106