xref: /aosp_15_r20/external/XNNPACK/src/f32-vunary/gen/vsqr-avx512f-x16.c (revision 4bdc94577ba0e567308109d787f7fec7b531ce36)
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_x16(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__avx512f_x16(
20     size_t n,
21     const float* x,
22     float* y,
23     const union xnn_f32_default_params params[restrict XNN_MIN_ELEMENTS(1)])
24 {
25   assert(n != 0);
26   assert(n % sizeof(float) == 0);
27   assert(x != NULL);
28   assert(y != NULL);
29 
30   for (; n >= 16 * sizeof(float); n -= 16 * sizeof(float)) {
31     const __m512 vx0123456789ABCDEF = _mm512_loadu_ps(x);
32     x += 16;
33 
34     const __m512 vy0123456789ABCDEF = _mm512_mul_ps(vx0123456789ABCDEF, vx0123456789ABCDEF);
35 
36     _mm512_storeu_ps(y, vy0123456789ABCDEF);
37     y += 16;
38   }
39   if XNN_UNLIKELY(n != 0) {
40     assert(n >= 1 * sizeof(float));
41     assert(n <= 15 * sizeof(float));
42     // Prepare mask for valid 32-bit elements (depends on n).
43     n >>= 2 /* log2(sizeof(float)) */;
44     const __mmask16 vmask = _cvtu32_mask16((uint16_t) ((uint32_t) (UINT32_C(1) << n) - UINT32_C(1)));
45 
46     const __m512 vx = _mm512_maskz_loadu_ps(vmask, x);
47     const __m512 vy = _mm512_mul_ps(vx, vx);
48     _mm512_mask_storeu_ps(y, vmask, vy);
49   }
50 }
51