crtfastmath.c revision 1.1.1.11
1 1.1 mrg /* 2 1.1.1.11 mrg * Copyright (C) 2005-2024 Free Software Foundation, Inc. 3 1.1 mrg * 4 1.1 mrg * This file is free software; you can redistribute it and/or modify it 5 1.1 mrg * under the terms of the GNU General Public License as published by the 6 1.1 mrg * Free Software Foundation; either version 3, or (at your option) any 7 1.1 mrg * later version. 8 1.1 mrg * 9 1.1 mrg * This file is distributed in the hope that it will be useful, but 10 1.1 mrg * WITHOUT ANY WARRANTY; without even the implied warranty of 11 1.1 mrg * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 12 1.1 mrg * General Public License for more details. 13 1.1 mrg * 14 1.1 mrg * Under Section 7 of GPL version 3, you are granted additional 15 1.1 mrg * permissions described in the GCC Runtime Library Exception, version 16 1.1 mrg * 3.1, as published by the Free Software Foundation. 17 1.1 mrg * 18 1.1 mrg * You should have received a copy of the GNU General Public License and 19 1.1 mrg * a copy of the GCC Runtime Library Exception along with this program; 20 1.1 mrg * see the files COPYING3 and COPYING.RUNTIME respectively. If not, see 21 1.1 mrg * <http://www.gnu.org/licenses/>. 22 1.1 mrg */ 23 1.1 mrg 24 1.1.1.2 mrg #ifndef _SOFT_FLOAT 25 1.1 mrg #define MXCSR_DAZ (1 << 6) /* Enable denormals are zero mode */ 26 1.1 mrg #define MXCSR_FTZ (1 << 15) /* Enable flush to zero mode */ 27 1.1 mrg 28 1.1 mrg #ifndef __x86_64__ 29 1.1 mrg /* All 64-bit targets have SSE and DAZ; 30 1.1 mrg only check them explicitly for 32-bit ones. */ 31 1.1 mrg #include "cpuid.h" 32 1.1 mrg 33 1.1.1.3 mrg __attribute__ ((target("fxsr,sse"))) 34 1.1.1.3 mrg static void 35 1.1 mrg /* The i386 ABI only requires 4-byte stack alignment, so this is necessary 36 1.1 mrg to make sure the fxsave struct gets correct alignment. 37 1.1 mrg See PR27537 and PR28621. */ 38 1.1 mrg __attribute__ ((force_align_arg_pointer)) 39 1.1.1.3 mrg set_fast_math_sse (unsigned int edx) 40 1.1.1.3 mrg { 41 1.1.1.3 mrg unsigned int mxcsr; 42 1.1.1.3 mrg 43 1.1.1.3 mrg if (edx & bit_FXSAVE) 44 1.1.1.3 mrg { 45 1.1.1.3 mrg /* Check if DAZ is available. */ 46 1.1.1.3 mrg struct 47 1.1.1.3 mrg { 48 1.1.1.3 mrg unsigned short cwd; 49 1.1.1.3 mrg unsigned short swd; 50 1.1.1.3 mrg unsigned short twd; 51 1.1.1.3 mrg unsigned short fop; 52 1.1.1.3 mrg unsigned int fip; 53 1.1.1.3 mrg unsigned int fcs; 54 1.1.1.3 mrg unsigned int foo; 55 1.1.1.3 mrg unsigned int fos; 56 1.1.1.3 mrg unsigned int mxcsr; 57 1.1.1.3 mrg unsigned int mxcsr_mask; 58 1.1.1.3 mrg unsigned int st_space[32]; 59 1.1.1.3 mrg unsigned int xmm_space[32]; 60 1.1.1.3 mrg unsigned int padding[56]; 61 1.1.1.3 mrg } __attribute__ ((aligned (16))) fxsave; 62 1.1.1.3 mrg 63 1.1.1.3 mrg /* This is necessary since some implementations of FXSAVE 64 1.1.1.3 mrg do not modify reserved areas within the image. */ 65 1.1.1.3 mrg fxsave.mxcsr_mask = 0; 66 1.1.1.3 mrg 67 1.1.1.3 mrg __builtin_ia32_fxsave (&fxsave); 68 1.1.1.3 mrg 69 1.1.1.3 mrg mxcsr = fxsave.mxcsr; 70 1.1.1.3 mrg 71 1.1.1.3 mrg if (fxsave.mxcsr_mask & MXCSR_DAZ) 72 1.1.1.3 mrg mxcsr |= MXCSR_DAZ; 73 1.1.1.3 mrg } 74 1.1.1.3 mrg else 75 1.1.1.3 mrg mxcsr = __builtin_ia32_stmxcsr (); 76 1.1.1.3 mrg 77 1.1.1.3 mrg mxcsr |= MXCSR_FTZ; 78 1.1.1.3 mrg __builtin_ia32_ldmxcsr (mxcsr); 79 1.1.1.3 mrg } 80 1.1 mrg #endif 81 1.1.1.3 mrg 82 1.1.1.3 mrg static void __attribute__((constructor)) 83 1.1 mrg set_fast_math (void) 84 1.1 mrg { 85 1.1 mrg #ifndef __x86_64__ 86 1.1 mrg unsigned int eax, ebx, ecx, edx; 87 1.1 mrg 88 1.1 mrg if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx)) 89 1.1 mrg return; 90 1.1 mrg 91 1.1 mrg if (edx & bit_SSE) 92 1.1.1.3 mrg set_fast_math_sse (edx); 93 1.1 mrg #else 94 1.1 mrg unsigned int mxcsr = __builtin_ia32_stmxcsr (); 95 1.1 mrg mxcsr |= MXCSR_DAZ | MXCSR_FTZ; 96 1.1 mrg __builtin_ia32_ldmxcsr (mxcsr); 97 1.1 mrg #endif 98 1.1 mrg } 99 1.1.1.2 mrg #endif 100
Indexes created Sat Feb 28 05:31:39 UTC 2026