|
Chris@42
|
1 /*
|
|
Chris@42
|
2 * Copyright (c) 2003, 2007-11 Matteo Frigo
|
|
Chris@42
|
3 * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
|
|
Chris@42
|
4 * Copyright (c) 2012-2013 Romain Dolbeau
|
|
Chris@42
|
5 *
|
|
Chris@42
|
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
Chris@42
|
7 * of this software and associated documentation files (the "Software"), to deal
|
|
Chris@42
|
8 * in the Software without restriction, including without limitation the rights
|
|
Chris@42
|
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
Chris@42
|
10 * copies of the Software, and to permit persons to whom the Software is
|
|
Chris@42
|
11 * furnished to do so, subject to the following conditions:
|
|
Chris@42
|
12 *
|
|
Chris@42
|
13 * The above copyright notice and this permission notice shall be included in
|
|
Chris@42
|
14 * all copies or substantial portions of the Software.
|
|
Chris@42
|
15 *
|
|
Chris@42
|
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
Chris@42
|
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
Chris@42
|
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
Chris@42
|
19 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
Chris@42
|
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
Chris@42
|
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
Chris@42
|
22 * THE SOFTWARE.
|
|
Chris@42
|
23 *
|
|
Chris@42
|
24 */
|
|
Chris@42
|
25
|
|
Chris@42
|
26 #include "ifftw.h"
|
|
Chris@42
|
27
|
|
Chris@42
|
28 #if HAVE_AVX512
|
|
Chris@42
|
29
|
|
Chris@42
|
30 #if defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64)
|
|
Chris@42
|
31
|
|
Chris@42
|
32 #include "amd64-cpuid.h"
|
|
Chris@42
|
33
|
|
Chris@42
|
34 int X(have_simd_avx512)(void)
|
|
Chris@42
|
35 {
|
|
Chris@42
|
36 static int init = 0, res;
|
|
Chris@42
|
37 int max_stdfn, eax, ebx, ecx, edx;
|
|
Chris@42
|
38
|
|
Chris@42
|
39 /* NOTE: this code is a total guess. I don't have an avx512
|
|
Chris@42
|
40 machine available. The code contributed by Erik Lindahl would
|
|
Chris@42
|
41 crash on a machine without XGETBV, so I had to guess a fix. */
|
|
Chris@42
|
42 if (!init) {
|
|
Chris@42
|
43 cpuid_all(0,0,&eax,&ebx,&ecx,&edx);
|
|
Chris@42
|
44 max_stdfn = eax;
|
|
Chris@42
|
45 if (max_stdfn >= 0x1) {
|
|
Chris@42
|
46 /* have OSXSAVE? (implies XGETBV exists) */
|
|
Chris@42
|
47 cpuid_all(0x1, 0, &eax, &ebx, &ecx, &edx);
|
|
Chris@42
|
48 if ((ecx & 0x08000000) == 0x08000000) {
|
|
Chris@42
|
49 /* have AVX512? */
|
|
Chris@42
|
50 cpuid_all(7,0,&eax,&ebx,&ecx,&edx);
|
|
Chris@42
|
51 if (ebx & (1 << 16)) {
|
|
Chris@42
|
52 /* have OS support for XMM, YMM, ZMM */
|
|
Chris@42
|
53 int zmm_ymm_xmm = (7 << 5) | (1 << 2) | (1 << 1);
|
|
Chris@42
|
54 res = ((xgetbv_eax(0) & zmm_ymm_xmm) == zmm_ymm_xmm);
|
|
Chris@42
|
55 }
|
|
Chris@42
|
56 }
|
|
Chris@42
|
57 }
|
|
Chris@42
|
58 init = 1;
|
|
Chris@42
|
59 }
|
|
Chris@42
|
60
|
|
Chris@42
|
61 return res;
|
|
Chris@42
|
62 }
|
|
Chris@42
|
63
|
|
Chris@42
|
64 #else /* 32-bit code */
|
|
Chris@42
|
65
|
|
Chris@42
|
66 #error "Avx512 is 64 bits only"
|
|
Chris@42
|
67
|
|
Chris@42
|
68 #endif
|
|
Chris@42
|
69
|
|
Chris@42
|
70 #endif
|
