• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vrnd/neonv8.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 <arm_neon.h>
13 
14 #include <xnnpack/common.h>
15 #include <xnnpack/math.h>
16 #include <xnnpack/vunary.h>
17 
18 
xnn_f32_vrndne_ukernel__neonv8_x4(size_t n,const float * x,float * y,const union xnn_f32_rnd_params params[restrict XNN_MIN_ELEMENTS (1)])19 void xnn_f32_vrndne_ukernel__neonv8_x4(
20     size_t n,
21     const float* x,
22     float* y,
23     const union xnn_f32_rnd_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN
24 {
25   assert(n != 0);
26   assert(n % sizeof(float) == 0);
27 
28   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
29     const float32x4_t vx0123 = vld1q_f32(x); x += 4;
30 
31     const float32x4_t vy0123 = vrndnq_f32(vx0123);
32 
33     vst1q_f32(y, vy0123); y += 4;
34   }
35   if XNN_UNLIKELY(n != 0) {
36     const float32x4_t vx = vld1q_f32(x);
37     const float32x4_t vy = vrndnq_f32(vx);
38     float32x2_t vy_lo = vget_low_f32(vy);
39     if (n & (2 * sizeof(float))) {
40       vst1_f32(y, vy_lo); y += 2;
41       vy_lo = vget_high_f32(vy);
42     }
43     if (n & (1 * sizeof(float))) {
44       vst1_lane_f32(y, vy_lo, 0);
45     }
46   }
47 }
48