Chris@42
|
1 /*
|
Chris@42
|
2 * Copyright (c) 2003, 2007-14 Matteo Frigo
|
Chris@42
|
3 * Copyright (c) 2003, 2007-14 Massachusetts Institute of Technology
|
Chris@42
|
4 *
|
Chris@42
|
5 * This program is free software; you can redistribute it and/or modify
|
Chris@42
|
6 * it under the terms of the GNU General Public License as published by
|
Chris@42
|
7 * the Free Software Foundation; either version 2 of the License, or
|
Chris@42
|
8 * (at your option) any later version.
|
Chris@42
|
9 *
|
Chris@42
|
10 * This program is distributed in the hope that it will be useful,
|
Chris@42
|
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
Chris@42
|
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
Chris@42
|
13 * GNU General Public License for more details.
|
Chris@42
|
14 *
|
Chris@42
|
15 * You should have received a copy of the GNU General Public License
|
Chris@42
|
16 * along with this program; if not, write to the Free Software
|
Chris@42
|
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
Chris@42
|
18 *
|
Chris@42
|
19 */
|
Chris@42
|
20
|
Chris@42
|
21 /* detection of alignment. This is complicated because a machine may
|
Chris@42
|
22 support multiple SIMD extensions (e.g. SSE2 and AVX) but only one
|
Chris@42
|
23 set of alignment contraints. So this alignment stuff cannot be
|
Chris@42
|
24 defined in the SIMD header files. Rather than defining a separate
|
Chris@42
|
25 set of "machine" header files, we just do this ugly ifdef here. */
|
Chris@42
|
26 #if defined(HAVE_AVX512)
|
Chris@42
|
27 # if defined(FFTW_SINGLE)
|
Chris@42
|
28 # define ALIGNMENT 8 /* Alignment for the LD/ST macros */
|
Chris@42
|
29 # define ALIGNMENTA 64 /* Alignment for the LDA/STA macros */
|
Chris@42
|
30 # else
|
Chris@42
|
31 # define ALIGNMENT 16 /* Alignment for the LD/ST macros */
|
Chris@42
|
32 # define ALIGNMENTA 64 /* Alignment for the LDA/STA macros */
|
Chris@42
|
33 # endif
|
Chris@42
|
34 #elif defined(HAVE_SSE2) || defined(HAVE_AVX) || defined(HAVE_AVX2) || (HAVE_AVX_128_FMA)
|
Chris@42
|
35 # if defined(FFTW_SINGLE)
|
Chris@42
|
36 # define ALIGNMENT 8 /* Alignment for the LD/ST macros */
|
Chris@42
|
37 # define ALIGNMENTA 16 /* Alignment for the LDA/STA macros */
|
Chris@42
|
38 # else
|
Chris@42
|
39 # define ALIGNMENT 16 /* Alignment for the LD/ST macros */
|
Chris@42
|
40 # define ALIGNMENTA 16 /* Alignment for the LDA/STA macros */
|
Chris@42
|
41 # endif
|
Chris@42
|
42 #elif defined(HAVE_ALTIVEC)
|
Chris@42
|
43 # define ALIGNMENT 8 /* Alignment for the LD/ST macros */
|
Chris@42
|
44 # define ALIGNMENTA 16 /* Alignment for the LDA/STA macros */
|
Chris@42
|
45 #elif defined(HAVE_NEON) || defined(HAVE_VSX)
|
Chris@42
|
46 # define ALIGNMENT 8 /* Alignment for the LD/ST macros */
|
Chris@42
|
47 # define ALIGNMENTA 8 /* Alignment for the LDA/STA macros */
|
Chris@42
|
48 #elif defined(HAVE_KCVI)
|
Chris@42
|
49 # if defined(FFTW_SINGLE)
|
Chris@42
|
50 # define ALIGNMENT 8 /* Alignment for the LD/ST macros */
|
Chris@42
|
51 # else
|
Chris@42
|
52 # define ALIGNMENT 16 /* Alignment for the LD/ST macros */
|
Chris@42
|
53 # endif
|
Chris@42
|
54 # define ALIGNMENTA 64 /* Alignment for the LDA/STA macros */
|
Chris@42
|
55 #elif defined(HAVE_GENERIC_SIMD256)
|
Chris@42
|
56 # if defined(FFTW_SINGLE)
|
Chris@42
|
57 # define ALIGNMENT 8
|
Chris@42
|
58 # define ALIGNMENTA 32
|
Chris@42
|
59 # else
|
Chris@42
|
60 # define ALIGNMENT 16
|
Chris@42
|
61 # define ALIGNMENTA 32
|
Chris@42
|
62 # endif
|
Chris@42
|
63 #elif defined(HAVE_GENERIC_SIMD128)
|
Chris@42
|
64 # if defined(FFTW_SINGLE)
|
Chris@42
|
65 # define ALIGNMENT 8
|
Chris@42
|
66 # define ALIGNMENTA 16
|
Chris@42
|
67 # else
|
Chris@42
|
68 # define ALIGNMENT 16
|
Chris@42
|
69 # define ALIGNMENTA 16
|
Chris@42
|
70 # endif
|
Chris@42
|
71 #endif
|
Chris@42
|
72
|
Chris@42
|
73 #if HAVE_SIMD
|
Chris@42
|
74 # ifndef ALIGNMENT
|
Chris@42
|
75 # error "ALIGNMENT not defined"
|
Chris@42
|
76 # endif
|
Chris@42
|
77 # ifndef ALIGNMENTA
|
Chris@42
|
78 # error "ALIGNMENTA not defined"
|
Chris@42
|
79 # endif
|
Chris@42
|
80 #endif
|
Chris@42
|
81
|
Chris@42
|
82 /* rename for precision and for SIMD extensions */
|
Chris@42
|
83 #define XSIMD0(name, suffix) CONCAT(name, suffix)
|
Chris@42
|
84 #define XSIMD(name) XSIMD0(X(name), SIMD_SUFFIX)
|
Chris@42
|
85 #define XSIMD_STRING(x) x STRINGIZE(SIMD_SUFFIX)
|
Chris@42
|
86
|
Chris@42
|
87 /* TAINT_BIT is set if pointers are not guaranteed to be multiples of
|
Chris@42
|
88 ALIGNMENT */
|
Chris@42
|
89 #define TAINT_BIT 1
|
Chris@42
|
90
|
Chris@42
|
91 /* TAINT_BITA is set if pointers are not guaranteed to be multiples of
|
Chris@42
|
92 ALIGNMENTA */
|
Chris@42
|
93 #define TAINT_BITA 2
|
Chris@42
|
94
|
Chris@42
|
95 #define PTRINT(p) ((uintptr_t)(p))
|
Chris@42
|
96
|
Chris@42
|
97 #define ALIGNED(p) \
|
Chris@42
|
98 (((PTRINT(UNTAINT(p)) % ALIGNMENT) == 0) && !(PTRINT(p) & TAINT_BIT))
|
Chris@42
|
99
|
Chris@42
|
100 #define ALIGNEDA(p) \
|
Chris@42
|
101 (((PTRINT(UNTAINT(p)) % ALIGNMENTA) == 0) && !(PTRINT(p) & TAINT_BITA))
|
Chris@42
|
102
|
Chris@42
|
103 #define SIMD_STRIDE_OK(x) (!(((x) * sizeof(R)) % ALIGNMENT))
|
Chris@42
|
104 #define SIMD_STRIDE_OKA(x) (!(((x) * sizeof(R)) % ALIGNMENTA))
|
Chris@42
|
105 #define SIMD_VSTRIDE_OK SIMD_STRIDE_OK
|
Chris@42
|
106
|