1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vlrelu/wasmsimd-minmax.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/vunary.h>
16
17
xnn_f32_vlrelu_ukernel__wasmsimd_minmax_x4(size_t n,const float * x,float * y,const union xnn_f32_lrelu_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_vlrelu_ukernel__wasmsimd_minmax_x4(
19 size_t n,
20 const float* x,
21 float* y,
22 const union xnn_f32_lrelu_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN
23 {
24 assert(n != 0);
25 assert(n % sizeof(float) == 0);
26
27 const v128_t vslope = wasm_v32x4_load_splat(¶ms->scalar.slope);
28 const v128_t vzero = wasm_f32x4_splat(0.0f);
29 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
30 v128_t vx = wasm_v128_load(x);
31 x += 4;
32 v128_t vacc = wasm_i32x4_max(vx, vzero);
33 vx = wasm_i32x4_min(vx, vzero);
34 vacc = wasm_f32x4_add(vacc, wasm_f32x4_mul(vx, vslope));
35 wasm_v128_store(y, vacc);
36 y += 4;
37 }
38 if XNN_UNLIKELY(n != 0) {
39 v128_t vx = wasm_v128_load(x);
40 v128_t vacc = wasm_i32x4_max(vx, vzero);
41 vx = wasm_i32x4_min(vx, vzero);
42 vacc = wasm_f32x4_add(vacc, wasm_f32x4_mul(vx, vslope));
43
44 if (n & (2 * sizeof(float))) {
45 *((double*) y) = wasm_f64x2_extract_lane(vacc, 0);
46 vacc = wasm_v32x4_shuffle(vacc, vacc, 2, 3, 2, 3);
47 y += 2;
48 }
49 if (n & (1 * sizeof(float))) {
50 *y = wasm_f32x4_extract_lane(vacc, 0);
51 }
52 }
53 }
54