1 // Auto-generated file. Do not edit!
2 // Template: src/f32-vunary/avx512f.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 <immintrin.h>
13
14 #include <xnnpack/common.h>
15 #include <xnnpack/intrinsics-polyfill.h>
16 #include <xnnpack/vunary.h>
17
18
xnn_f32_vsqr_ukernel__avx512f_x32(size_t n,const float * x,float * y,const void * params)19 void xnn_f32_vsqr_ukernel__avx512f_x32(
20 size_t n,
21 const float* x,
22 float* y,
23 const void* params)
24 {
25 assert(n != 0);
26 assert(n % sizeof(float) == 0);
27 assert(x != NULL);
28 assert(y != NULL);
29
30 for (; n >= 32 * sizeof(float); n -= 32 * sizeof(float)) {
31 const __m512 vx0123456789ABCDEF = _mm512_loadu_ps(x);
32 const __m512 vxGHIJKLMNOPQRSTUV = _mm512_loadu_ps(x + 16);
33 x += 32;
34
35 const __m512 vy0123456789ABCDEF = _mm512_mul_ps(vx0123456789ABCDEF, vx0123456789ABCDEF);
36 const __m512 vyGHIJKLMNOPQRSTUV = _mm512_mul_ps(vxGHIJKLMNOPQRSTUV, vxGHIJKLMNOPQRSTUV);
37
38 _mm512_storeu_ps(y, vy0123456789ABCDEF);
39 _mm512_storeu_ps(y + 16, vyGHIJKLMNOPQRSTUV);
40 y += 32;
41 }
42 for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) {
43 const __m512 vx = _mm512_loadu_ps(x);
44 x += 16;
45
46 const __m512 vy = _mm512_mul_ps(vx, vx);
47
48 _mm512_storeu_ps(y, vy);
49 y += 16;
50 }
51 if XNN_UNLIKELY(n != 0) {
52 assert(n >= 1 * sizeof(float));
53 assert(n <= 15 * sizeof(float));
54 // Prepare mask for valid 32-bit elements (depends on n).
55 n >>= 2 /* log2(sizeof(float)) */;
56 const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1)));
57
58 const __m512 vx = _mm512_maskz_loadu_ps(vmask, x);
59 const __m512 vy = _mm512_mul_ps(vx, vx);
60 _mm512_mask_storeu_ps(y, vmask, vy);
61 }
62 }
63