1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vrnd/sse41.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 <smmintrin.h>
13
14 #include <xnnpack/common.h>
15 #include <xnnpack/math.h>
16 #include <xnnpack/vunary.h>
17
18
xnn_f32_vrndu_ukernel__sse41_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_vrndu_ukernel__sse41_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_DISABLE_TSAN
24 {
25 assert(n != 0);
26 assert(n % sizeof(float) == 0);
27
28 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
29 const __m128 vx0123 = _mm_loadu_ps(x);
30 const __m128 vx4567 = _mm_loadu_ps(x + 4);
31 x += 8;
32
33 const __m128 vy0123 = _mm_round_ps(vx0123, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
34 const __m128 vy4567 = _mm_round_ps(vx4567, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
35
36 _mm_storeu_ps(y, vy0123);
37 _mm_storeu_ps(y + 4, vy4567);
38 y += 8;
39 }
40 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
41 const __m128 vx = _mm_loadu_ps(x);
42 x += 4;
43
44 const __m128 vy = _mm_round_ps(vx, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
45
46 _mm_storeu_ps(y, vy);
47 y += 4;
48 }
49 if XNN_UNLIKELY(n != 0) {
50 const __m128 vx = _mm_loadu_ps(x);
51 __m128 vy = _mm_round_ps(vx, _MM_FROUND_TO_POS_INF | _MM_FROUND_NO_EXC);
52 if (n & (2 * sizeof(float))) {
53 _mm_storel_pi((__m64*) y, vy);
54 vy = _mm_movehl_ps(vy, vy);
55 y += 2;
56 }
57 if (n & (1 * sizeof(float))) {
58 _mm_store_ss(y, vy);
59 }
60 }
61 }
62