• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vunary/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/math.h>
16 #include <xnnpack/vunary.h>
17 
18 
xnn_f32_vneg_ukernel__wasmsimd_x8(size_t n,const float * x,float * y,const union xnn_f32_neg_params params[restrict XNN_MIN_ELEMENTS (1)])19 void xnn_f32_vneg_ukernel__wasmsimd_x8(
20     size_t n,
21     const float* x,
22     float* y,
23     const union xnn_f32_neg_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_DISABLE_TSAN
24 {
25   assert(n != 0);
26   assert(n % sizeof(float) == 0);
27   assert(x != NULL);
28   assert(y != NULL);
29 
30   const v128_t vsign_mask = wasm_v32x4_load_splat(&params->wasmsimd.sign_mask);
31   for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
32     const v128_t vx0123 = wasm_v128_load(x);
33     const v128_t vx4567 = wasm_v128_load(x + 4);
34     x += 8;
35 
36     const v128_t vy0123 = wasm_v128_xor(vx0123, vsign_mask);
37     const v128_t vy4567 = wasm_v128_xor(vx4567, vsign_mask);
38 
39     wasm_v128_store(y, vy0123);
40     wasm_v128_store(y + 4, vy4567);
41     y += 8;
42   }
43   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
44     const v128_t vx = wasm_v128_load(x);
45     x += 4;
46 
47     const v128_t vy = wasm_v128_xor(vx, vsign_mask);
48 
49     wasm_v128_store(y, vy);
50     y += 4;
51   }
52   if XNN_UNLIKELY(n != 0) {
53     const v128_t vx = wasm_v128_load(x);
54     v128_t vy = wasm_v128_xor(vx, vsign_mask);
55     if (n & (2 * sizeof(float))) {
56       *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
57       vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
58       y += 2;
59     }
60     if (n & (1 * sizeof(float))) {
61       *y = wasm_f32x4_extract_lane(vy, 0);
62     }
63   }
64 }
65