|
Chris@82
|
1 /*
|
|
Chris@82
|
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
|
|
Chris@82
|
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
|
Chris@82
|
4 *
|
|
Chris@82
|
5 * This program is free software; you can redistribute it and/or modify
|
|
Chris@82
|
6 * it under the terms of the GNU General Public License as published by
|
|
Chris@82
|
7 * the Free Software Foundation; either version 2 of the License, or
|
|
Chris@82
|
8 * (at your option) any later version.
|
|
Chris@82
|
9 *
|
|
Chris@82
|
10 * This program is distributed in the hope that it will be useful,
|
|
Chris@82
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
Chris@82
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
Chris@82
|
13 * GNU General Public License for more details.
|
|
Chris@82
|
14 *
|
|
Chris@82
|
15 * You should have received a copy of the GNU General Public License
|
|
Chris@82
|
16 * along with this program; if not, write to the Free Software
|
|
Chris@82
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
Chris@82
|
18 *
|
|
Chris@82
|
19 */
|
|
Chris@82
|
20
|
|
Chris@82
|
21
|
|
Chris@82
|
22 #include "kernel/ifftw.h"
|
|
Chris@82
|
23
|
|
Chris@82
|
24 #if HAVE_AVX2
|
|
Chris@82
|
25
|
|
Chris@82
|
26 #if defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64)
|
|
Chris@82
|
27 # include "amd64-cpuid.h"
|
|
Chris@82
|
28 #else
|
|
Chris@82
|
29 # include "x86-cpuid.h"
|
|
Chris@82
|
30 #endif
|
|
Chris@82
|
31
|
|
Chris@82
|
32 int X(have_simd_avx2_128)(void)
|
|
Chris@82
|
33 {
|
|
Chris@82
|
34 static int init = 0, res;
|
|
Chris@82
|
35 int max_stdfn, eax, ebx, ecx, edx;
|
|
Chris@82
|
36
|
|
Chris@82
|
37 if (!init) {
|
|
Chris@82
|
38 cpuid_all(0,0,&eax,&ebx,&ecx,&edx);
|
|
Chris@82
|
39 max_stdfn = eax;
|
|
Chris@82
|
40 if (max_stdfn >= 0x1) {
|
|
Chris@82
|
41 /* have AVX and OSXSAVE? (implies XGETBV exists) */
|
|
Chris@82
|
42 cpuid_all(0x1, 0, &eax, &ebx, &ecx, &edx);
|
|
Chris@82
|
43 if ((ecx & 0x18000000) == 0x18000000) {
|
|
Chris@82
|
44 /* have AVX2? */
|
|
Chris@82
|
45 cpuid_all(7,0,&eax,&ebx,&ecx,&edx);
|
|
Chris@82
|
46 if (ebx & (1 << 5)) {
|
|
Chris@82
|
47 /* have OS support for XMM, YMM? */
|
|
Chris@82
|
48 res = ((xgetbv_eax(0) & 0x6) == 0x6);
|
|
Chris@82
|
49 }
|
|
Chris@82
|
50 }
|
|
Chris@82
|
51 }
|
|
Chris@82
|
52 init = 1;
|
|
Chris@82
|
53 }
|
|
Chris@82
|
54 return res;
|
|
Chris@82
|
55 }
|
|
Chris@82
|
56
|
|
Chris@82
|
57 int X(have_simd_avx2)(void)
|
|
Chris@82
|
58 {
|
|
Chris@82
|
59 /*
|
|
Chris@82
|
60 * For now 256-bit AVX2 support is identical to 128-bit.
|
|
Chris@82
|
61 * This might change in the future if AMD released AVX2-capable
|
|
Chris@82
|
62 * chips that work better with the 128-bit flavor, but since AMD
|
|
Chris@82
|
63 * might actually change it to implement 256-bit AVX2 efficiently
|
|
Chris@82
|
64 * by then we don't want to disable it before we know.
|
|
Chris@82
|
65 */
|
|
Chris@82
|
66 return X(have_simd_avx2_128)();
|
|
Chris@82
|
67 }
|
|
Chris@82
|
68 #endif
|
