xref: /aosp_15_r20/external/XNNPACK/src/f32-vsqrt/gen/sse-sqrt-x4.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
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_x4(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_x4(
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 >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
28     const __m128 vx = _mm_loadu_ps(x);
29     x += 4;
30     const __m128 vy = _mm_sqrt_ps(vx);
31     _mm_storeu_ps(y, vy);
32     y += 4;
33   }
34   if XNN_UNLIKELY(n != 0) {
35     const __m128 vx = _mm_loadu_ps(x);
36     __m128 vy = _mm_sqrt_ps(vx);
37     if (n & (2 * sizeof(float))) {
38       _mm_storel_pi((__m64*) y, vy);
39       vy = _mm_movehl_ps(vy, vy);
40       y += 2;
41     }
42     if (n & (1 * sizeof(float))) {
43       _mm_store_ss(y, vy);
44     }
45   }
46 }
47