xref: /aosp_15_r20/external/XNNPACK/src/f32-vunary/gen/vsqr-sse-x8.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
1 // Auto-generated file. Do not edit!
2 //   Template: src/f32-vunary/sse.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/math.h>
16 #include <xnnpack/vunary.h>
17 
18 
xnn_f32_vsqr_ukernel__sse_x8(size_t n,const float * x,float * y,const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS (1)])19 void xnn_f32_vsqr_ukernel__sse_x8(
20     size_t n,
21     const float* x,
22     float* y,
23     const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
24 {
25   assert(n != 0);
26   assert(n % sizeof(float) == 0);
27   assert(x != NULL);
28   assert(y != NULL);
29 
30   for (; n >= 8 * sizeof(float); n -= 8 * sizeof(float)) {
31     const __m128 vx0123 = _mm_loadu_ps(x);
32     const __m128 vx4567 = _mm_loadu_ps(x + 4);
33     x += 8;
34 
35     const __m128 vy0123 = _mm_mul_ps(vx0123, vx0123);
36     const __m128 vy4567 = _mm_mul_ps(vx4567, vx4567);
37 
38     _mm_storeu_ps(y, vy0123);
39     _mm_storeu_ps(y + 4, vy4567);
40     y += 8;
41   }
42   for (; n >= 4 * sizeof(float); n -= 4 * sizeof(float)) {
43     const __m128 vx = _mm_loadu_ps(x);
44     x += 4;
45     const __m128 vy = _mm_mul_ps(vx, vx);
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_mul_ps(vx, vx);
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