cannam@95: /*
cannam@95:  * Copyright (c) 2003, 2007-11 Matteo Frigo
cannam@95:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
cannam@95:  *
cannam@95:  * This program is free software; you can redistribute it and/or modify
cannam@95:  * it under the terms of the GNU General Public License as published by
cannam@95:  * the Free Software Foundation; either version 2 of the License, or
cannam@95:  * (at your option) any later version.
cannam@95:  *
cannam@95:  * This program is distributed in the hope that it will be useful,
cannam@95:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
cannam@95:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
cannam@95:  * GNU General Public License for more details.
cannam@95:  *
cannam@95:  * You should have received a copy of the GNU General Public License
cannam@95:  * along with this program; if not, write to the Free Software
cannam@95:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
cannam@95:  *
cannam@95:  */
cannam@95: 
cannam@95: 
cannam@95: #include "ifftw.h"
cannam@95: 
cannam@95: #ifdef FFTW_SINGLE
cannam@95: #  define DS(d,s) s /* single-precision option */
cannam@95: #else
cannam@95: #  define DS(d,s) d /* double-precision option */
cannam@95: #endif
cannam@95: 
cannam@95: #if HAVE_SSE2
cannam@95: 
cannam@95: # if defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64)
cannam@95: 
cannam@95:   int X(have_simd_sse2)(void)
cannam@95:   {
cannam@95:        return 1;
cannam@95:   }
cannam@95: 
cannam@95: # else /* !x86_64 */
cannam@95: 
cannam@95: # include <signal.h>
cannam@95: # include <setjmp.h>
cannam@95: # include "x86-cpuid.h"
cannam@95: 
cannam@95:   static jmp_buf jb;
cannam@95: 
cannam@95:   static void sighandler(int x)
cannam@95:   {
cannam@95:        UNUSED(x);
cannam@95:        longjmp(jb, 1);
cannam@95:   }
cannam@95: 
cannam@95:   static int sse2_works(void)
cannam@95:   {
cannam@95:        void (*oldsig)(int);
cannam@95:        oldsig = signal(SIGILL, sighandler);
cannam@95:        if (setjmp(jb)) {
cannam@95: 	    signal(SIGILL, oldsig);
cannam@95: 	    return 0;
cannam@95:        } else {
cannam@95: #         ifdef _MSC_VER
cannam@95: 	    _asm { DS(xorpd,xorps) xmm0,xmm0 }
cannam@95: #         else
cannam@95: 	    /* asm volatile ("xorpd/s %xmm0, %xmm0"); */
cannam@95: 	    asm volatile(DS(".byte 0x66; .byte 0x0f; .byte 0x57; .byte 0xc0",
cannam@95: 			                ".byte 0x0f; .byte 0x57; .byte 0xc0"));
cannam@95: #         endif
cannam@95: 	    signal(SIGILL, oldsig);
cannam@95: 	    return 1;
cannam@95:        }
cannam@95:   }
cannam@95: 
cannam@95:   extern void X(check_alignment_of_sse2_pm)(void);
cannam@95: 
cannam@95:   int X(have_simd_sse2)(void)
cannam@95:   {
cannam@95:        static int init = 0, res;
cannam@95: 
cannam@95:        if (!init) {
cannam@95: 	    res =   !is_386() 
cannam@95: 		 && has_cpuid()
cannam@95: 		 && (cpuid_edx(1) & (1 << DS(26,25)))
cannam@95: 		 && sse2_works();
cannam@95: 	    init = 1;
cannam@95: 	    X(check_alignment_of_sse2_pm)();
cannam@95:        }
cannam@95:        return res;
cannam@95:   }
cannam@95: 
cannam@95: # endif
cannam@95: 
cannam@95: #endif