blob: f8d1df7fdc3d75dc22b9cd717c9aa30c713660cb [file] [log] [blame]
Marat Dukhan13bafb02020-06-05 00:43:11 -07001// 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
17void xnn_f32_vsqrdiff_ukernel__scalar_x4(
18 size_t n,
19 const float* a,
20 const float* b,
21 float* y,
22 const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS(1)])
23{
24 assert(n != 0);
25 assert(n % sizeof(float) == 0);
Frank Barchard0822dde2020-07-04 12:47:24 -070026 assert(a != NULL);
27 assert(b != NULL);
28 assert(y != NULL);
Marat Dukhan13bafb02020-06-05 00:43:11 -070029
30
31 for (; n >= 4 * sizeof(float); n -= 4 * 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 a += 4;
37
38 const float vb0 = b[0];
39 const float vb1 = b[1];
40 const float vb2 = b[2];
41 const float vb3 = b[3];
42 b += 4;
43
44 float vy0 = va0 - vb0;
45 float vy1 = va1 - vb1;
46 float vy2 = va2 - vb2;
47 float vy3 = va3 - vb3;
48
49 vy0 = vy0 * vy0;
50 vy1 = vy1 * vy1;
51 vy2 = vy2 * vy2;
52 vy3 = vy3 * vy3;
53
54
55 y[0] = vy0;
56 y[1] = vy1;
57 y[2] = vy2;
58 y[3] = vy3;
59 y += 4;
60 }
61 if XNN_UNLIKELY(n != 0) {
62 do {
63 const float va = *a++;
64 const float vb = *b++;
65 float vy = va - vb;
66 vy = vy * vy;
67 *y++ = vy;
68 n -= sizeof(float);
69 } while (n != 0);
70 }
71}