xref: /aosp_15_r20/external/XNNPACK/src/f32-dwconv/gen/up1x3-scalar-acc2.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
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_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_default_params params[restrict XNN_MIN_ELEMENTS (1)])16 void xnn_f32_dwconv_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_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     input = (const float**) ((uintptr_t) input + input_stride);
48 
49     size_t c = channels;
50     const float* w = weights;
51     do {
52       float vacc0p0 = w[0];
53 
54       const float vi0 = *i0++;
55       const float vk0 = w[1];
56       vacc0p0 = math_muladd_f32(vi0, vk0, vacc0p0);
57 
58       const float vi1 = *i1++;
59       const float vk1 = w[2];
60       float vacc0p1 = vi1 * vk1;
61 
62       const float vi2 = *i2++;
63       const float vk2 = w[3];
64       vacc0p0 = math_muladd_f32(vi2, vk2, vacc0p0);
65 
66       w += 4;
67 
68       vacc0p0 += vacc0p1;
69 
70       *output++ = vacc0p0;
71     } while (--c != 0);
72 
73     output = (float*) ((uintptr_t) output + output_increment);
74   } while (--output_width != 0);
75 }
76