• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-relu/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/vunary.h>
15 #include <xnnpack/common.h>
16 
17 
xnn_f32_relu_ukernel__wasmsimd_x4(size_t n,const float * x,float * y,const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_relu_ukernel__wasmsimd_x4(
19     size_t n,
20     const float* x,
21     float* y,
22     const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN
23 {
24   assert(n != 0);
25   assert(n % sizeof(float) == 0);
26   assert(x != NULL);
27   assert(y != NULL);
28 
29   const v128_t vzero = wasm_f32x4_splat(0.0f);
30 
31   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
32     v128_t vacc = wasm_v128_load(x);
33     x += 4;
34 
35     vacc = wasm_i32x4_max(vacc, vzero);
36 
37     wasm_v128_store(y, vacc);
38     y += 4;
39   }
40   if XNN_UNLIKELY(n != 0) {
41     v128_t vacc = wasm_v128_load(x);
42 
43     vacc = wasm_i32x4_max(vacc, vzero);
44 
45     if (n & (2 * sizeof(float))) {
46       *((double*) y) = wasm_f64x2_extract_lane(vacc, 0);
47       vacc = wasm_v32x4_shuffle(vacc, vacc, 2, 3, 2, 3);
48       y += 2;
49     }
50     if (n & (1 * sizeof(float))) {
51       *y = wasm_f32x4_extract_lane(vacc, 0);
52     }
53   }
54 }
55