1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vrelu/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_vrelu_ukernel__wasmsimd_x8(size_t n,const float * x,float * y,const union xnn_f32_relu_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_vrelu_ukernel__wasmsimd_x8(
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_OOB_READS
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_i32x4_const_splat(0);
30
31 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
32 v128_t vacc0123 = wasm_v128_load(x);
33 v128_t vacc4567 = wasm_v128_load(x + 4);
34 x += 8;
35
36 vacc0123 = wasm_i32x4_max(vacc0123, vzero);
37 vacc4567 = wasm_i32x4_max(vacc4567, vzero);
38
39 wasm_v128_store(y, vacc0123);
40 wasm_v128_store(y + 4, vacc4567);
41 y += 8;
42 }
43 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
44 v128_t vacc = wasm_v128_load(x);
45 x += 4;
46
47 vacc = wasm_i32x4_max(vacc, vzero);
48
49 wasm_v128_store(y, vacc);
50 y += 4;
51 }
52 if XNN_UNLIKELY(n != 0) {
53 v128_t vacc = wasm_v128_load(x);
54
55 vacc = wasm_i32x4_max(vacc, vzero);
56
57 if (n & (2 * sizeof(float))) {
58 *((double*) y) = wasm_f64x2_extract_lane(vacc, 0);
59 vacc = wasm_v32x4_shuffle(vacc, vacc, 2, 3, 2, 3);
60 y += 2;
61 }
62 if (n & (1 * sizeof(float))) {
63 *y = wasm_f32x4_extract_lane(vacc, 0);
64 }
65 }
66 }
67