1 // Copyright 2019 Google LLC
2 //
3 // This source code is licensed under the BSD-style license found in the
4 // LICENSE file in the root directory of this source tree.
5
6 #include <assert.h>
7
8 #include <psimd.h>
9
10 #include <xnnpack/clamp.h>
11
12
xnn_f32_clamp_ukernel__psimd(size_t n,const float * x,float * y,const union xnn_f32_output_params params[restrict static1])13 void xnn_f32_clamp_ukernel__psimd(
14 size_t n,
15 const float* x,
16 float* y,
17 const union xnn_f32_output_params params[restrict static 1])
18 {
19 assert(n != 0);
20 assert(n % sizeof(float) == 0);
21
22 const psimd_f32 voutput_max = psimd_load_splat_f32(¶ms->scalar.max);
23 const psimd_f32 voutput_min = psimd_load_splat_f32(¶ms->scalar.min);
24
25 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
26 const psimd_f32 vx = psimd_load_f32(x);
27 x += 4;
28
29 const psimd_f32 vy = psimd_min_f32(psimd_max_f32(vx, voutput_min), voutput_max);
30
31 psimd_store_f32(y, vy);
32 y += 4;
33 }
34 if (n != 0) {
35 const psimd_f32 vx = psimd_load_f32(x);
36
37 psimd_f32 vy = psimd_min_f32(psimd_max_f32(vx, voutput_min), voutput_max);
38
39 if (n & 2 * sizeof(float)) {
40 psimd_store2_f32(y, vy);
41 vy = psimd_concat_hi_f32(vy, vy);
42 y += 2;
43 }
44 if (n & 1 * sizeof(float)) {
45 psimd_store1_f32(y, vy);
46 }
47 }
48 }
49