xref: /aosp_15_r20/external/XNNPACK/src/f32-vrnd/gen/vrndd-wasmsimd-x8.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vrnd/wasmsimd.c.in
3 //   Generator: tools/xngen
4 //
5 // Copyright 2021 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_vrndd_ukernel__wasmsimd_x8(size_t n,const float * x,float * y,const union xnn_f32_rnd_params params[restrict XNN_MIN_ELEMENTS (1)])19 void xnn_f32_vrndd_ukernel__wasmsimd_x8(
20     size_t n,
21     const float* x,
22     float* y,
23     const union xnn_f32_rnd_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); x += 4;
30     const v128_t vx4567 = wasm_v128_load(x); x += 4;
31 
32     const v128_t vy0123 = wasm_f32x4_floor(vx0123);
33     const v128_t vy4567 = wasm_f32x4_floor(vx4567);
34 
35     wasm_v128_store(y, vy0123); y += 4;
36     wasm_v128_store(y, vy4567); y += 4;
37   }
38   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
39     const v128_t vx = wasm_v128_load(x); x += 4;
40     const v128_t vy = wasm_f32x4_floor(vx);
41     wasm_v128_store(y, vy); y += 4;
42   }
43   if XNN_UNLIKELY(n != 0) {
44     const v128_t vx = wasm_v128_load(x);
45     v128_t vy = wasm_f32x4_floor(vx);
46 
47     if (n & (2 * sizeof(float))) {
48       *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
49       vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
50       y += 2;
51     }
52     if (n & (1 * sizeof(float))) {
53       *y = wasm_f32x4_extract_lane(vy, 0);
54     }
55   }
56 }
57