1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vsqrt/sse-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
12 #include <xmmintrin.h>
13
14 #include <xnnpack/common.h>
15 #include <xnnpack/vunary.h>
16
17
xnn_f32_vsqrt_ukernel__sse_sqrt_x8(size_t n,const float * x,float * y,const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS (1)])18 void xnn_f32_vsqrt_ukernel__sse_sqrt_x8(
19 size_t n,
20 const float* x,
21 float* y,
22 const union xnn_f32_sqrt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
23 {
24 assert(n != 0);
25 assert(n % sizeof(float) == 0);
26
27 for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
28 const __m128 vx0123 = _mm_loadu_ps(x);
29 const __m128 vx4567 = _mm_loadu_ps(x + 4);
30 x += 8;
31
32 const __m128 vy0123 = _mm_sqrt_ps(vx0123);
33 const __m128 vy4567 = _mm_sqrt_ps(vx4567);
34
35 _mm_storeu_ps(y, vy0123);
36 _mm_storeu_ps(y + 4, vy4567);
37 y += 8;
38 }
39 for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
40 const __m128 vx = _mm_loadu_ps(x);
41 x += 4;
42 const __m128 vy = _mm_sqrt_ps(vx);
43 _mm_storeu_ps(y, vy);
44 y += 4;
45 }
46 if XNN_UNLIKELY(n != 0) {
47 const __m128 vx = _mm_loadu_ps(x);
48 __m128 vy = _mm_sqrt_ps(vx);
49 if (n & (2 * sizeof(float))) {
50 _mm_storel_pi((__m64*) y, vy);
51 vy = _mm_movehl_ps(vy, vy);
52 y += 2;
53 }
54 if (n & (1 * sizeof(float))) {
55 _mm_store_ss(y, vy);
56 }
57 }
58 }
59