Chris@10: /*
Chris@10:  * Copyright (c) 2003, 2007-11 Matteo Frigo
Chris@10:  * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology
Chris@10:  *
Chris@10:  * This program is free software; you can redistribute it and/or modify
Chris@10:  * it under the terms of the GNU General Public License as published by
Chris@10:  * the Free Software Foundation; either version 2 of the License, or
Chris@10:  * (at your option) any later version.
Chris@10:  *
Chris@10:  * This program is distributed in the hope that it will be useful,
Chris@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
Chris@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
Chris@10:  * GNU General Public License for more details.
Chris@10:  *
Chris@10:  * You should have received a copy of the GNU General Public License
Chris@10:  * along with this program; if not, write to the Free Software
Chris@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
Chris@10:  *
Chris@10:  */
Chris@10: 
Chris@10: 
Chris@10: #include "ifftw.h"
Chris@10: 
Chris@10: #ifdef FFTW_SINGLE
Chris@10: #  define DS(d,s) s /* single-precision option */
Chris@10: #else
Chris@10: #  define DS(d,s) d /* double-precision option */
Chris@10: #endif
Chris@10: 
Chris@10: #if HAVE_SSE2
Chris@10: 
Chris@10: # if defined(__x86_64__) || defined(_M_X64) || defined(_M_AMD64)
Chris@10: 
Chris@10:   int X(have_simd_sse2)(void)
Chris@10:   {
Chris@10:        return 1;
Chris@10:   }
Chris@10: 
Chris@10: # else /* !x86_64 */
Chris@10: 
Chris@10: # include <signal.h>
Chris@10: # include <setjmp.h>
Chris@10: # include "x86-cpuid.h"
Chris@10: 
Chris@10:   static jmp_buf jb;
Chris@10: 
Chris@10:   static void sighandler(int x)
Chris@10:   {
Chris@10:        UNUSED(x);
Chris@10:        longjmp(jb, 1);
Chris@10:   }
Chris@10: 
Chris@10:   static int sse2_works(void)
Chris@10:   {
Chris@10:        void (*oldsig)(int);
Chris@10:        oldsig = signal(SIGILL, sighandler);
Chris@10:        if (setjmp(jb)) {
Chris@10: 	    signal(SIGILL, oldsig);
Chris@10: 	    return 0;
Chris@10:        } else {
Chris@10: #         ifdef _MSC_VER
Chris@10: 	    _asm { DS(xorpd,xorps) xmm0,xmm0 }
Chris@10: #         else
Chris@10: 	    /* asm volatile ("xorpd/s %xmm0, %xmm0"); */
Chris@10: 	    asm volatile(DS(".byte 0x66; .byte 0x0f; .byte 0x57; .byte 0xc0",
Chris@10: 			                ".byte 0x0f; .byte 0x57; .byte 0xc0"));
Chris@10: #         endif
Chris@10: 	    signal(SIGILL, oldsig);
Chris@10: 	    return 1;
Chris@10:        }
Chris@10:   }
Chris@10: 
Chris@10:   extern void X(check_alignment_of_sse2_pm)(void);
Chris@10: 
Chris@10:   int X(have_simd_sse2)(void)
Chris@10:   {
Chris@10:        static int init = 0, res;
Chris@10: 
Chris@10:        if (!init) {
Chris@10: 	    res =   !is_386() 
Chris@10: 		 && has_cpuid()
Chris@10: 		 && (cpuid_edx(1) & (1 << DS(26,25)))
Chris@10: 		 && sse2_works();
Chris@10: 	    init = 1;
Chris@10: 	    X(check_alignment_of_sse2_pm)();
Chris@10:        }
Chris@10:        return res;
Chris@10:   }
Chris@10: 
Chris@10: # endif
Chris@10: 
Chris@10: #endif