1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vbinary/vopc-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_vaddc_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_vaddc_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   const v128_t vb = wasm_v32x4_load_splat(b);
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     v128_t vy0123 = wasm_f32x4_add(va0123, vb);
39     v128_t vy4567 = wasm_f32x4_add(va4567, vb);
40 
41 
42     vy0123 = wasm_i32x4_max(vy0123, vzero);
43     vy4567 = wasm_i32x4_max(vy4567, vzero);
44 
45     wasm_v128_store(y, vy0123);
46     wasm_v128_store(y + 4, vy4567);
47     y += 8;
48   }
49   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
50     const v128_t va = wasm_v128_load(a);
51     a += 4;
52 
53     v128_t vy = wasm_f32x4_add(va, vb);
54 
55     vy = wasm_i32x4_max(vy, vzero);
56 
57     wasm_v128_store(y, vy);
58     y += 4;
59   }
60   if XNN_UNLIKELY(n != 0) {
61     const v128_t va = wasm_v128_load(a);
62 
63     v128_t vy = wasm_f32x4_add(va, vb);
64 
65     vy = wasm_i32x4_max(vy, vzero);
66 
67     if (n & (2 * sizeof(float))) {
68       *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
69       vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
70       y += 2;
71     }
72     if (n & (1 * sizeof(float))) {
73       *y = wasm_f32x4_extract_lane(vy, 0);
74     }
75   }
76 }
77