blob: 298e392a404282e7bb8829dd91a80ce02d98e02d [file] [log] [blame]
Frank Barchard9c7308f2020-08-31 17:03:01 -07001// Auto-generated file. Do not edit!
2// Template: src/f32-vbinary/vop-wasmsimd.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 <wasm_simd128.h>
13
14#include <xnnpack/common.h>
15#include <xnnpack/vbinary.h>
16
17
18void xnn_f32_vsqrdiff_ukernel__wasmsimd_x16(
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 >= 16 * sizeof(float); n -= 16 * sizeof(float)) {
33 const v128_t va0123 = wasm_v128_load(a);
34 const v128_t va4567 = wasm_v128_load(a + 4);
35 const v128_t va89AB = wasm_v128_load(a + 8);
36 const v128_t vaCDEF = wasm_v128_load(a + 12);
37 a += 16;
38
39 const v128_t vb0123 = wasm_v128_load(b);
40 const v128_t vb4567 = wasm_v128_load(b + 4);
41 const v128_t vb89AB = wasm_v128_load(b + 8);
42 const v128_t vbCDEF = wasm_v128_load(b + 12);
43 b += 16;
44
45 v128_t vy0123 = wasm_f32x4_sub(va0123, vb0123);
46 v128_t vy4567 = wasm_f32x4_sub(va4567, vb4567);
47 v128_t vy89AB = wasm_f32x4_sub(va89AB, vb89AB);
48 v128_t vyCDEF = wasm_f32x4_sub(vaCDEF, vbCDEF);
49
50 vy0123 = wasm_f32x4_mul(vy0123, vy0123);
51 vy4567 = wasm_f32x4_mul(vy4567, vy4567);
52 vy89AB = wasm_f32x4_mul(vy89AB, vy89AB);
53 vyCDEF = wasm_f32x4_mul(vyCDEF, vyCDEF);
54
55
56 wasm_v128_store(y, vy0123);
57 wasm_v128_store(y + 4, vy4567);
58 wasm_v128_store(y + 8, vy89AB);
59 wasm_v128_store(y + 12, vyCDEF);
60 y += 16;
61 }
62 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
63 const v128_t va = wasm_v128_load(a);
64 a += 4;
65
66 const v128_t vb = wasm_v128_load(b);
67 b += 4;
68
69 v128_t vy = wasm_f32x4_sub(va, vb);
70 vy = wasm_f32x4_mul(vy, vy);
71
72
73 wasm_v128_store(y, vy);
74 y += 4;
75 }
76 if XNN_UNLIKELY(n != 0) {
77 const v128_t va = wasm_v128_load(a);
78 const v128_t vb = wasm_v128_load(b);
79
80 v128_t vy = wasm_f32x4_sub(va, vb);
81 vy = wasm_f32x4_mul(vy, vy);
82
83
84 if (n & (2 * sizeof(float))) {
85 *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
86 vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
87 y += 2;
88 }
89 if (n & (1 * sizeof(float))) {
90 *y = wasm_f32x4_extract_lane(vy, 0);
91 }
92 }
93}