xref: /aosp_15_r20/external/XNNPACK/src/f32-vsqrt/gen/wasmsimd-sqrt-x8.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vsqrt/wasmsimd-sqrt.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 #include <math.h>
12 
13 #include <wasm_simd128.h>
14 
15 #include <xnnpack/common.h>
16 #include <xnnpack/vunary.h>
17 
18 
xnn_f32_vsqrt_ukernel__wasmsimd_sqrt_x8(size_t n,const float * x,float * y,const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS (1)])19 void xnn_f32_vsqrt_ukernel__wasmsimd_sqrt_x8(
20     size_t n,
21     const float* x,
22     float* y,
23     const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
24 {
25   assert(n != 0);
26   assert(n % sizeof(float) == 0);
27 
28   for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
29     const v128_t vx0123 = wasm_v128_load(x);
30     const v128_t vx4567 = wasm_v128_load(x + 4);
31     x += 8;
32 
33     const v128_t vy0123 = wasm_f32x4_sqrt(vx0123);
34     const v128_t vy4567 = wasm_f32x4_sqrt(vx4567);
35 
36     wasm_v128_store(y, vy0123);
37     wasm_v128_store(y + 4, vy4567);
38     y += 8;
39   }
40   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
41     const v128_t vx = wasm_v128_load(x);
42     x += 4;
43     const v128_t vy = wasm_f32x4_sqrt(vx);
44     wasm_v128_store(y, vy);
45     y += 4;
46   }
47   if XNN_UNLIKELY(n != 0) {
48     const v128_t vx = wasm_v128_load(x);
49     v128_t vy = wasm_f32x4_sqrt(vx);
50     if (n & (2 * sizeof(float))) {
51       *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
52       vy = wasm_v64x2_shuffle(vy, vy, 1, 1);
53       y += 2;
54     }
55     if (n & (1 * sizeof(float))) {
56       *y = wasm_f32x4_extract_lane(vy, 0);
57     }
58   }
59 }
60