• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vbinary/vop-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_vdiv_ukernel__wasmsimd_x8(size_t n,const float * a,const float * b,float * y,const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_vdiv_ukernel__wasmsimd_x8(
19     size_t n,
20     const float* a,
21     const float* b,
22     float* y,
23     const union xnn_f32_default_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 
32   for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
33     const v128_t va0123 = wasm_v128_load(a);
34     const v128_t va4567 = wasm_v128_load(a + 4);
35     a += 8;
36 
37     const v128_t vb0123 = wasm_v128_load(b);
38     const v128_t vb4567 = wasm_v128_load(b + 4);
39     b += 8;
40 
41     v128_t vy0123 = wasm_f32x4_div(va0123, vb0123);
42     v128_t vy4567 = wasm_f32x4_div(va4567, vb4567);
43 
44 
45 
46     wasm_v128_store(y, vy0123);
47     wasm_v128_store(y + 4, vy4567);
48     y += 8;
49   }
50   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
51     const v128_t va = wasm_v128_load(a);
52     a += 4;
53 
54     const v128_t vb = wasm_v128_load(b);
55     b += 4;
56 
57     v128_t vy = wasm_f32x4_div(va, vb);
58 
59 
60     wasm_v128_store(y, vy);
61     y += 4;
62   }
63   if XNN_UNLIKELY(n != 0) {
64     const v128_t va = wasm_v128_load(a);
65     const v128_t vb = wasm_v128_load(b);
66 
67     v128_t vy = wasm_f32x4_div(va, vb);
68 
69 
70     if (n & (2 * sizeof(float))) {
71       *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
72       vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
73       y += 2;
74     }
75     if (n & (1 * sizeof(float))) {
76       *y = wasm_f32x4_extract_lane(vy, 0);
77     }
78   }
79 }
80