1 // 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
xnn_f32_vsub_relu_ukernel__wasmsimd_x8(size_t n,const float * a,const float * b,float * y,const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_vsub_relu_ukernel__wasmsimd_x8(
19 size_t n,
20 const float* a,
21 const float* b,
22 float* y,
23 const union xnn_f32_relu_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 const v128_t vzero = wasm_f32x4_splat(0.0f);
32
33 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
34 const v128_t va0123 = wasm_v128_load(a);
35 const v128_t va4567 = wasm_v128_load(a + 4);
36 a += 8;
37
38 const v128_t vb0123 = wasm_v128_load(b);
39 const v128_t vb4567 = wasm_v128_load(b + 4);
40 b += 8;
41
42 v128_t vy0123 = wasm_f32x4_sub(va0123, vb0123);
43 v128_t vy4567 = wasm_f32x4_sub(va4567, vb4567);
44
45
46 vy0123 = wasm_i32x4_max(vy0123, vzero);
47 vy4567 = wasm_i32x4_max(vy4567, vzero);
48
49 wasm_v128_store(y, vy0123);
50 wasm_v128_store(y + 4, vy4567);
51 y += 8;
52 }
53 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
54 const v128_t va = wasm_v128_load(a);
55 a += 4;
56
57 const v128_t vb = wasm_v128_load(b);
58 b += 4;
59
60 v128_t vy = wasm_f32x4_sub(va, vb);
61
62 vy = wasm_i32x4_max(vy, vzero);
63
64 wasm_v128_store(y, vy);
65 y += 4;
66 }
67 if XNN_UNLIKELY(n != 0) {
68 const v128_t va = wasm_v128_load(a);
69 const v128_t vb = wasm_v128_load(b);
70
71 v128_t vy = wasm_f32x4_sub(va, vb);
72
73 vy = wasm_i32x4_max(vy, vzero);
74
75 if (n & (2 * sizeof(float))) {
76 *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
77 vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
78 y += 2;
79 }
80 if (n & (1 * sizeof(float))) {
81 *y = wasm_f32x4_extract_lane(vy, 0);
82 }
83 }
84 }
85