1/* 2 * Copyright (C) 2005 Free Software Foundation, Inc. 3 * 4 * This file is free software; you can redistribute it and/or modify it 5 * under the terms of the GNU General Public License as published by the 6 * Free Software Foundation; either version 2, or (at your option) any 7 * later version. 8 * 9 * In addition to the permissions in the GNU General Public License, the 10 * Free Software Foundation gives you unlimited permission to link the 11 * compiled version of this file with other programs, and to distribute 12 * those programs without any restriction coming from the use of this 13 * file. (The General Public License restrictions do apply in other 14 * respects; for example, they cover modification of the file, and 15 * distribution when not linked into another program.) 16 * 17 * This file is distributed in the hope that it will be useful, but 18 * WITHOUT ANY WARRANTY; without even the implied warranty of 19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 20 * General Public License for more details. 21 * 22 * You should have received a copy of the GNU General Public License 23 * along with this program; see the file COPYING. If not, write to 24 * the Free Software Foundation, 51 Franklin Street, Fifth Floor, 25 * Boston, MA 02110-1301, USA. 26 * 27 * As a special exception, if you link this library with files 28 * compiled with GCC to produce an executable, this does not cause 29 * the resulting executable to be covered by the GNU General Public License. 30 * This exception does not however invalidate any other reasons why 31 * the executable file might be covered by the GNU General Public License. 32 */ 33 34#define MXCSR_DAZ (1 << 6) /* Enable denormals are zero mode */ 35#define MXCSR_FTZ (1 << 15) /* Enable flush to zero mode */ 36 37#define FXSAVE (1 << 24) 38#define SSE (1 << 25) 39 40static void __attribute__((constructor)) 41set_fast_math (void) 42{ 43#ifndef __x86_64__ 44 /* All 64-bit targets have SSE and DAZ; only check them explicitly 45 for 32-bit ones. */ 46 unsigned int eax, ebx, ecx, edx; 47 48 /* See if we can use cpuid. */ 49 asm volatile ("pushfl; pushfl; popl %0; movl %0,%1; xorl %2,%0;" 50 "pushl %0; popfl; pushfl; popl %0; popfl" 51 : "=&r" (eax), "=&r" (ebx) 52 : "i" (0x00200000)); 53 54 if (((eax ^ ebx) & 0x00200000) == 0) 55 return; 56 57 /* Check the highest input value for eax. */ 58 asm volatile ("xchgl %%ebx, %1; cpuid; xchgl %%ebx, %1" 59 : "=a" (eax), "=r" (ebx), "=c" (ecx), "=d" (edx) 60 : "0" (0)); 61 62 if (eax == 0) 63 return; 64 65 asm volatile ("xchgl %%ebx, %1; cpuid; xchgl %%ebx, %1" 66 : "=a" (eax), "=r" (ebx), "=c" (ecx), "=d" (edx) 67 : "0" (1)); 68 69 if (edx & SSE) 70 { 71 unsigned int mxcsr = __builtin_ia32_stmxcsr (); 72 73 mxcsr |= MXCSR_FTZ; 74 75 if (edx & FXSAVE) 76 { 77 /* Check if DAZ is available. */ 78 struct 79 { 80 unsigned short int cwd; 81 unsigned short int swd; 82 unsigned short int twd; 83 unsigned short int fop; 84 long int fip; 85 long int fcs; 86 long int foo; 87 long int fos; 88 long int mxcsr; 89 long int mxcsr_mask; 90 long int st_space[32]; 91 long int xmm_space[32]; 92 long int padding[56]; 93 } __attribute__ ((aligned (16))) fxsave; 94 95 __builtin_memset (&fxsave, 0, sizeof (fxsave)); 96 97 asm volatile ("fxsave %0" : "=m" (fxsave) : "m" (fxsave)); 98 99 if (fxsave.mxcsr_mask & MXCSR_DAZ) 100 mxcsr |= MXCSR_DAZ; 101 } 102 103 __builtin_ia32_ldmxcsr (mxcsr); 104 } 105#else 106 unsigned int mxcsr = __builtin_ia32_stmxcsr (); 107 mxcsr |= MXCSR_DAZ | MXCSR_FTZ; 108 __builtin_ia32_ldmxcsr (mxcsr); 109#endif 110} 111