1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vbinary/vop-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/common.h>
13 #include <xnnpack/math.h>
14 #include <xnnpack/vbinary.h>
15 
16 
xnn_f32_vsub_relu_ukernel__scalar_x8(size_t n,const float * a,const float * b,float * y,const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS (1)])17 void xnn_f32_vsub_relu_ukernel__scalar_x8(
18     size_t n,
19     const float* a,
20     const float* b,
21     float* y,
22     const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS(1)])
23 {
24   assert(n != 0);
25   assert(n % sizeof(float) == 0);
26   assert(a != NULL);
27   assert(b != NULL);
28   assert(y != NULL);
29 
30 
31   for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
32     const float va0 = a[0];
33     const float va1 = a[1];
34     const float va2 = a[2];
35     const float va3 = a[3];
36     const float va4 = a[4];
37     const float va5 = a[5];
38     const float va6 = a[6];
39     const float va7 = a[7];
40     a += 8;
41 
42     const float vb0 = b[0];
43     const float vb1 = b[1];
44     const float vb2 = b[2];
45     const float vb3 = b[3];
46     const float vb4 = b[4];
47     const float vb5 = b[5];
48     const float vb6 = b[6];
49     const float vb7 = b[7];
50     b += 8;
51 
52     float vy0 = va0 - vb0;
53     float vy1 = va1 - vb1;
54     float vy2 = va2 - vb2;
55     float vy3 = va3 - vb3;
56     float vy4 = va4 - vb4;
57     float vy5 = va5 - vb5;
58     float vy6 = va6 - vb6;
59     float vy7 = va7 - vb7;
60 
61 
62     vy0 = math_max_f32(vy0, 0.0f);
63     vy1 = math_max_f32(vy1, 0.0f);
64     vy2 = math_max_f32(vy2, 0.0f);
65     vy3 = math_max_f32(vy3, 0.0f);
66     vy4 = math_max_f32(vy4, 0.0f);
67     vy5 = math_max_f32(vy5, 0.0f);
68     vy6 = math_max_f32(vy6, 0.0f);
69     vy7 = math_max_f32(vy7, 0.0f);
70 
71     y[0] = vy0;
72     y[1] = vy1;
73     y[2] = vy2;
74     y[3] = vy3;
75     y[4] = vy4;
76     y[5] = vy5;
77     y[6] = vy6;
78     y[7] = vy7;
79     y += 8;
80   }
81   if XNN_UNLIKELY(n != 0) {
82     do {
83       const float va = *a++;
84       const float vb = *b++;
85       float vy = va - vb;
86       vy = math_max_f32(vy, 0.0f);
87       *y++ = vy;
88       n -= sizeof(float);
89     } while (n != 0);
90   }
91 }
92