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__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_default_params params[restrict XNN_MIN_ELEMENTS (1)])16 void xnn_f32_dwconv_ukernel_up1x9__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_default_params params[restrict XNN_MIN_ELEMENTS(1)])
27 {
28 assert(channels != 0);
29 assert(output_width != 0);
30
31 do {
32 const float* i0 = input[0];
33 assert(i0 != NULL);
34 if XNN_UNPREDICTABLE(i0 != zero) {
35 i0 = (const float*) ((uintptr_t) i0 + input_offset);
36 }
37 const float* i1 = input[1];
38 assert(i1 != NULL);
39 if XNN_UNPREDICTABLE(i1 != zero) {
40 i1 = (const float*) ((uintptr_t) i1 + input_offset);
41 }
42 const float* i2 = input[2];
43 assert(i2 != NULL);
44 if XNN_UNPREDICTABLE(i2 != zero) {
45 i2 = (const float*) ((uintptr_t) i2 + input_offset);
46 }
47 const float* i3 = input[3];
48 assert(i3 != NULL);
49 if XNN_UNPREDICTABLE(i3 != zero) {
50 i3 = (const float*) ((uintptr_t) i3 + input_offset);
51 }
52 const float* i4 = input[4];
53 assert(i4 != NULL);
54 if XNN_UNPREDICTABLE(i4 != zero) {
55 i4 = (const float*) ((uintptr_t) i4 + input_offset);
56 }
57 const float* i5 = input[5];
58 assert(i5 != NULL);
59 if XNN_UNPREDICTABLE(i5 != zero) {
60 i5 = (const float*) ((uintptr_t) i5 + input_offset);
61 }
62 const float* i6 = input[6];
63 assert(i6 != NULL);
64 if XNN_UNPREDICTABLE(i6 != zero) {
65 i6 = (const float*) ((uintptr_t) i6 + input_offset);
66 }
67 const float* i7 = input[7];
68 assert(i7 != NULL);
69 if XNN_UNPREDICTABLE(i7 != zero) {
70 i7 = (const float*) ((uintptr_t) i7 + input_offset);
71 }
72 const float* i8 = input[8];
73 assert(i8 != NULL);
74 if XNN_UNPREDICTABLE(i8 != zero) {
75 i8 = (const float*) ((uintptr_t) i8 + input_offset);
76 }
77 input = (const float**) ((uintptr_t) input + input_stride);
78
79 size_t c = channels;
80 const float* w = weights;
81 do {
82 float vacc0p0 = w[0];
83
84 const float vi0 = *i0++;
85 const float vk0 = w[1];
86 vacc0p0 = math_muladd_f32(vi0, vk0, vacc0p0);
87
88 const float vi1 = *i1++;
89 const float vk1 = w[2];
90 float vacc0p1 = vi1 * vk1;
91
92 const float vi2 = *i2++;
93 const float vk2 = w[3];
94 vacc0p0 = math_muladd_f32(vi2, vk2, vacc0p0);
95
96 const float vi3 = *i3++;
97 const float vk3 = w[4];
98 vacc0p1 = math_muladd_f32(vi3, vk3, vacc0p1);
99
100 const float vi4 = *i4++;
101 const float vk4 = w[5];
102 vacc0p0 = math_muladd_f32(vi4, vk4, vacc0p0);
103
104 const float vi5 = *i5++;
105 const float vk5 = w[6];
106 vacc0p1 = math_muladd_f32(vi5, vk5, vacc0p1);
107
108 const float vi6 = *i6++;
109 const float vk6 = w[7];
110 vacc0p0 = math_muladd_f32(vi6, vk6, vacc0p0);
111
112 const float vi7 = *i7++;
113 const float vk7 = w[8];
114 vacc0p1 = math_muladd_f32(vi7, vk7, vacc0p1);
115
116 const float vi8 = *i8++;
117 const float vk8 = w[9];
118 vacc0p0 = math_muladd_f32(vi8, vk8, vacc0p0);
119
120 w += 10;
121
122 vacc0p0 += vacc0p1;
123
124 *output++ = vacc0p0;
125 } while (--c != 0);
126
127 output = (float*) ((uintptr_t) output + output_increment);
128 } while (--output_width != 0);
129 }
130