1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vsqrt/neon-sqrt.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 #include <math.h>
12
13 #include <arm_neon.h>
14
15 #include <xnnpack/common.h>
16 #include <xnnpack/vunary.h>
17
18
xnn_f32_vsqrt_ukernel__neon_sqrt_x8(size_t n,const float * x,float * y,const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS (1)])19 void xnn_f32_vsqrt_ukernel__neon_sqrt_x8(
20 size_t n,
21 const float* x,
22 float* y,
23 const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
24 {
25 assert(n != 0);
26 assert(n % sizeof(float) == 0);
27
28 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
29 const float32x4_t vx0123 = vld1q_f32(x); x += 4;
30 const float32x4_t vx4567 = vld1q_f32(x); x += 4;
31
32 const float32x4_t vy0123 = vsqrtq_f32(vx0123);
33 const float32x4_t vy4567 = vsqrtq_f32(vx4567);
34
35 vst1q_f32(y, vy0123); y += 4;
36 vst1q_f32(y, vy4567); y += 4;
37 }
38 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
39 const float32x4_t vx = vld1q_f32(x); x += 4;
40 const float32x4_t vy = vsqrtq_f32(vx);
41 vst1q_f32(y, vy); y += 4;
42 }
43 if XNN_UNLIKELY(n != 0) {
44 const float32x4_t vx = vld1q_f32(x);
45
46 float32x2_t vy_lo = vsqrt_f32(vget_low_f32(vx));
47 const float32x2_t vx_hi = vget_high_f32(vx);
48 if (n & (2 * sizeof(float))) {
49 vst1_f32(y, vy_lo); y += 2;
50 vy_lo = vsqrt_f32(vx_hi);
51 }
52 if (n & (1 * sizeof(float))) {
53 vst1_lane_f32(y, vy_lo, 0);
54 }
55 }
56 }
57