1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vbinary/vopc-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_vminc_ukernel__wasmsimd_x86_x16(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_vminc_ukernel__wasmsimd_x86_x16(
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 const v128_t vb = wasm_v32x4_load_splat(b);
32 for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) {
33 const v128_t va0123 = wasm_v128_load(a);
34 const v128_t va4567 = wasm_v128_load(a + 4);
35 const v128_t va89AB = wasm_v128_load(a + 8);
36 const v128_t vaCDEF = wasm_v128_load(a + 12);
37 a += 16;
38
39 const v128_t vm0123 = wasm_f32x4_lt(va0123, vb);
40 const v128_t vm4567 = wasm_f32x4_lt(va4567, vb);
41 const v128_t vm89AB = wasm_f32x4_lt(va89AB, vb);
42 const v128_t vmCDEF = wasm_f32x4_lt(vaCDEF, vb);
43
44 v128_t vy0123 = wasm_v128_bitselect(va0123, vb, vm0123);
45 v128_t vy4567 = wasm_v128_bitselect(va4567, vb, vm4567);
46 v128_t vy89AB = wasm_v128_bitselect(va89AB, vb, vm89AB);
47 v128_t vyCDEF = wasm_v128_bitselect(vaCDEF, vb, vmCDEF);
48
49
50 wasm_v128_store(y, vy0123);
51 wasm_v128_store(y + 4, vy4567);
52 wasm_v128_store(y + 8, vy89AB);
53 wasm_v128_store(y + 12, vyCDEF);
54 y += 16;
55 }
56 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
57 const v128_t va = wasm_v128_load(a);
58 a += 4;
59
60 const v128_t vm = wasm_f32x4_lt(va, vb);
61 v128_t vy = wasm_v128_bitselect(va, vb, vm);
62
63
64 wasm_v128_store(y, vy);
65 y += 4;
66 }
67 if XNN_UNLIKELY(n != 0) {
68 const v128_t va = wasm_v128_load(a);
69
70 const v128_t vm = wasm_f32x4_lt(va, vb);
71 v128_t vy = wasm_v128_bitselect(va, vb, vm);
72
73
74 if (n & (2 * sizeof(float))) {
75 *((double*) y) = wasm_f64x2_extract_lane(vy, 0);
76 vy = wasm_v32x4_shuffle(vy, vy, 2, 3, 2, 3);
77 y += 2;
78 }
79 if (n & (1 * sizeof(float))) {
80 *y = wasm_f32x4_extract_lane(vy, 0);
81 }
82 }
83 }
84