1 // Auto-generated file. Do not edit!
2 // Template: src/f32-prelu/wasm.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2020 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/math.h>
13 #include <xnnpack/prelu.h>
14
15
xnn_f32_prelu_ukernel__wasm_2x1(size_t rows,size_t channels,const float * restrict input,size_t input_stride,const float * restrict weights,float * restrict output,size_t output_stride)16 void xnn_f32_prelu_ukernel__wasm_2x1(
17 size_t rows,
18 size_t channels,
19 const float*restrict input,
20 size_t input_stride,
21 const float*restrict weights,
22 float*restrict output,
23 size_t output_stride)
24 {
25 assert(rows != 0);
26 assert(channels != 0);
27 assert(channels % sizeof(float) == 0);
28
29 const float* i0 = input;
30 float* o0 = output;
31 const float* i1 = (const float*) ((uintptr_t) i0 + input_stride);
32 float* o1 = (float*) ((uintptr_t) o0 + output_stride);
33
34 const size_t input_increment = input_stride * 2 - channels;
35 const size_t output_increment = output_stride * 2 - channels;
36
37 const float vzero = 0.0f;
38 do {
39 if XNN_UNPREDICTABLE(rows < 2) {
40 i1 = i0;
41 o1 = o0;
42 }
43
44 const float* w = weights;
45 size_t c = channels;
46 do {
47 const float vw = *w++;
48
49 float vi0 = *i0++;
50 float vi1 = *i1++;
51
52 float vacc0 = __builtin_wasm_max_f32(vi0, vzero);
53 vi0 = __builtin_wasm_min_f32(vi0, vzero);
54 float vacc1 = __builtin_wasm_max_f32(vi1, vzero);
55 vi1 = __builtin_wasm_min_f32(vi1, vzero);
56
57 vacc0 += vi0 * vw;
58 vacc1 += vi1 * vw;
59
60 *o0++ = vacc0;
61 *o1++ = vacc1;
62
63 c -= sizeof(float);
64 } while (c != 0);
65 i0 = (const float*) ((uintptr_t) i0 + input_increment);
66 o0 = (float*) ((uintptr_t) o0 + output_increment);
67 i1 = (const float*) ((uintptr_t) i1 + input_increment);
68 o1 = (float*) ((uintptr_t) o1 + output_increment);
69 rows = doz(rows, 2);
70 } while (rows != 0);
71 }
72