1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vbinary/vop-neon.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 <arm_neon.h>
13
14 #include <xnnpack/common.h>
15 #include <xnnpack/vbinary.h>
16
17
xnn_f32_vsqrdiff_ukernel__neon_x4(size_t n,const float * a,const float * b,float * y,const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_vsqrdiff_ukernel__neon_x4(
19 size_t n,
20 const float* a,
21 const float* b,
22 float* y,
23 const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN
24 {
25 assert(n != 0);
26 assert(n % sizeof(float) == 0);
27 assert(a != NULL);
28 assert(b != NULL);
29 assert(y != NULL);
30
31
32 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
33 const float32x4_t va0123 = vld1q_f32(a); a += 4;
34 const float32x4_t vb0123 = vld1q_f32(b); b += 4;
35
36 float32x4_t vy0123 = vsubq_f32(va0123, vb0123);
37
38 vy0123 = vmulq_f32(vy0123, vy0123);
39
40
41 vst1q_f32(y, vy0123); y += 4;
42 }
43 if XNN_UNLIKELY(n != 0) {
44 const float32x4_t va0123 = vld1q_f32(a);
45 const float32x4_t vb0123 = vld1q_f32(b);
46
47 float32x4_t vy0123 = vsubq_f32(va0123, vb0123);
48 vy0123 = vmulq_f32(vy0123, vy0123);
49
50 float32x2_t vy01 = vget_low_f32(vy0123);
51 if (n & (2 * sizeof(float))) {
52 vst1_f32(y, vy01); y += 2;
53 vy01 = vget_high_f32(vy0123);
54 }
55 if (n & (1 * sizeof(float))) {
56 vst1_lane_f32(y, vy01, 0);
57 }
58 }
59 }
60