2 Copyright (C) 2012 Paul Davis
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 #include "libpbd-config.h"
22 #define _XOPEN_SOURCE 600
23 #include <cstring> // for memset
28 #ifdef PLATFORM_WINDOWS
33 #include "pbd/error.h"
42 unsigned long cpuflags = 0;
46 #if !( (defined __x86_64__) || (defined __i386__) ) // !ARCH_X86
50 #ifdef PLATFORM_WINDOWS
52 // Get CPU flags using Microsoft function
53 // It works for both 64 and 32 bit systems
54 // no need to use assembler for getting info from register, this function does this for us
57 cpuflags = cpuInfo[3];
61 #ifndef _LP64 /* *nix; 32 bit version. This odd macro constant is required because we need something that identifies this as a 32 bit
62 build on Linux and on OS X. Anything that serves this purpose will do, but this is the best thing we've identified
74 : "%eax", "%ecx", "%edx"
77 #else /* *nix; 64 bit version */
79 /* asm notes: although we explicitly save&restore rbx, we must tell
80 gcc that ebx,rbx is clobbered so that it doesn't try to use it as an intermediate
81 register when storing rbx. gcc 4.3 didn't make this "mistake", but gcc 4.4
82 does, at least on x86_64.
93 : "%rax", "%rbx", "%rcx", "%rdx"
97 #endif /* PLATFORM_WINDOWS */
99 if (cpuflags & (1<<25)) {
100 _flags = Flags (_flags | (HasSSE|HasFlushToZero));
103 if (cpuflags & (1<<26)) {
104 _flags = Flags (_flags | HasSSE2);
107 if (cpuflags & (1 << 24)) {
111 /* DAZ wasn't available in the first version of SSE. Since
112 setting a reserved bit in MXCSR causes a general protection
113 fault, we need to be able to check the availability of this
114 feature without causing problems. To do this, one needs to
115 set up a 512-byte area of memory to save the SSE state to,
116 using fxsave, and then one needs to inspect bytes 28 through
117 31 for the MXCSR_MASK value. If bit 6 is set, DAZ is
118 supported, otherwise, it isn't.
121 #ifndef HAVE_POSIX_MEMALIGN
122 # ifdef PLATFORM_WINDOWS
123 fxbuf = (char **) _aligned_malloc (sizeof (char *), 16);
125 *fxbuf = (char *) _aligned_malloc (512, 16);
128 # warning using default malloc for aligned memory
129 fxbuf = (char **) malloc (sizeof (char *));
131 *fxbuf = (char *) malloc (512);
135 (void) posix_memalign ((void **) &fxbuf, 16, sizeof (char *));
137 (void) posix_memalign ((void **) fxbuf, 16, 512);
141 memset (*fxbuf, 0, 512);
157 uint32_t mxcsr_mask = *((uint32_t*) &((*fxbuf)[28]));
159 /* if the mask is zero, set its default value (from intel specs) */
161 if (mxcsr_mask == 0) {
165 if (mxcsr_mask & (1<<6)) {
166 _flags = Flags (_flags | HasDenormalsAreZero);
169 #if !defined HAVE_POSIX_MEMALIGN && defined PLATFORM_WINDOWS
170 _aligned_free (*fxbuf);
171 _aligned_free (fxbuf);