1 // Auto-generated file. Do not edit!
2 // Template: src/s16-rmaxabs/scalar.c.in
3 // Generator: tools/xngen
4 //
5 // Copyright 2022 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 <stddef.h>
12 #include <stdint.h>
13
14 #include <xnnpack/math.h>
15 #include <xnnpack/rmaxabs.h>
16
17
xnn_s16_rmaxabs_ukernel__scalar_x3(size_t batch,const int16_t * input,uint16_t * output)18 void xnn_s16_rmaxabs_ukernel__scalar_x3(
19 size_t batch,
20 const int16_t* input,
21 uint16_t* output) {
22
23 assert(batch > 0);
24 assert(input != NULL);
25 assert(output != NULL);
26
27 uint32_t vmax0 = 0;
28 uint32_t vmax1 = 0;
29 uint32_t vmax2 = 0;
30
31 for (; batch >= 3; batch -= 3) {
32 const int32_t vi0 = (int32_t) input[0];
33 const int32_t vi1 = (int32_t) input[1];
34 const int32_t vi2 = (int32_t) input[2];
35 input += 3;
36
37 const uint32_t vabs0 = math_abs_s32(vi0);
38 const uint32_t vabs1 = math_abs_s32(vi1);
39 const uint32_t vabs2 = math_abs_s32(vi2);
40
41 vmax0 = math_max_u32(vmax0, vabs0);
42 vmax1 = math_max_u32(vmax1, vabs1);
43 vmax2 = math_max_u32(vmax2, vabs2);
44 }
45
46 vmax0 = math_max_u32(vmax0, vmax1);
47 vmax0 = math_max_u32(vmax0, vmax2);
48
49 if (batch != 0) {
50 do {
51 const int32_t vi = (int32_t) *input++;
52 const uint32_t vabs = math_abs_s32(vi);
53 vmax0 = math_max_u32(vmax0, vabs);
54 } while (--batch != 0);
55 }
56 *output = (uint16_t) vmax0;
57 }
58