fpu.h revision 276133
1/*- 2 * Copyright (c) 1990 The Regents of the University of California. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * William Jolitz. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 4. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * from: @(#)npx.h 5.3 (Berkeley) 1/18/91 33 * $FreeBSD: stable/10/sys/x86/include/fpu.h 276133 2014-12-23 12:00:41Z kib $ 34 */ 35 36/* 37 * Floating Point Data Structures and Constants 38 * W. Jolitz 1/90 39 */ 40 41#ifndef _X86_FPU_H_ 42#define _X86_FPU_H_ 43 44/* Environment information of floating point unit. */ 45struct env87 { 46 int32_t en_cw; /* control word (16bits) */ 47 int32_t en_sw; /* status word (16bits) */ 48 int32_t en_tw; /* tag word (16bits) */ 49 int32_t en_fip; /* fp instruction pointer */ 50 uint16_t en_fcs; /* fp code segment selector */ 51 uint16_t en_opcode; /* opcode last executed (11 bits) */ 52 int32_t en_foo; /* fp operand offset */ 53 int32_t en_fos; /* fp operand segment selector */ 54}; 55 56/* Contents of each x87 floating point accumulator. */ 57struct fpacc87 { 58 uint8_t fp_bytes[10]; 59}; 60 61/* Floating point context. (i386 fnsave/frstor) */ 62struct save87 { 63 struct env87 sv_env; /* floating point control/status */ 64 struct fpacc87 sv_ac[8]; /* accumulator contents, 0-7 */ 65 uint8_t sv_pad0[4]; /* saved status word (now unused) */ 66 /* 67 * Bogus padding for emulators. Emulators should use their own 68 * struct and arrange to store into this struct (ending here) 69 * before it is inspected for ptracing or for core dumps. Some 70 * emulators overwrite the whole struct. We have no good way of 71 * knowing how much padding to leave. Leave just enough for the 72 * GPL emulator's i387_union (176 bytes total). 73 */ 74 uint8_t sv_pad[64]; /* padding; used by emulators */ 75}; 76 77/* Contents of each SSE extended accumulator. */ 78struct xmmacc { 79 uint8_t xmm_bytes[16]; 80}; 81 82/* Contents of the upper 16 bytes of each AVX extended accumulator. */ 83struct ymmacc { 84 uint8_t ymm_bytes[16]; 85}; 86 87/* Rename structs below depending on machine architecture. */ 88#ifdef __i386__ 89#define __envxmm32 envxmm 90#else 91#define __envxmm32 envxmm32 92#define __envxmm64 envxmm 93#endif 94 95struct __envxmm32 { 96 uint16_t en_cw; /* control word (16bits) */ 97 uint16_t en_sw; /* status word (16bits) */ 98 uint16_t en_tw; /* tag word (16bits) */ 99 uint16_t en_opcode; /* opcode last executed (11 bits) */ 100 uint32_t en_fip; /* fp instruction pointer */ 101 uint16_t en_fcs; /* fp code segment selector */ 102 uint16_t en_pad0; /* padding */ 103 uint32_t en_foo; /* fp operand offset */ 104 uint16_t en_fos; /* fp operand segment selector */ 105 uint16_t en_pad1; /* padding */ 106 uint32_t en_mxcsr; /* SSE control/status register */ 107 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 108}; 109 110struct __envxmm64 { 111 uint16_t en_cw; /* control word (16bits) */ 112 uint16_t en_sw; /* status word (16bits) */ 113 uint8_t en_tw; /* tag word (8bits) */ 114 uint8_t en_zero; 115 uint16_t en_opcode; /* opcode last executed (11 bits ) */ 116 uint64_t en_rip; /* fp instruction pointer */ 117 uint64_t en_rdp; /* fp operand pointer */ 118 uint32_t en_mxcsr; /* SSE control/status register */ 119 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 120}; 121 122/* Floating point context. (i386 fxsave/fxrstor) */ 123struct savexmm { 124 struct __envxmm32 sv_env; 125 struct { 126 struct fpacc87 fp_acc; 127 uint8_t fp_pad[6]; /* padding */ 128 } sv_fp[8]; 129 struct xmmacc sv_xmm[8]; 130 uint8_t sv_pad[224]; 131} __aligned(16); 132 133#ifdef __i386__ 134union savefpu { 135 struct save87 sv_87; 136 struct savexmm sv_xmm; 137}; 138#else 139/* Floating point context. (amd64 fxsave/fxrstor) */ 140struct savefpu { 141 struct __envxmm64 sv_env; 142 struct { 143 struct fpacc87 fp_acc; 144 uint8_t fp_pad[6]; /* padding */ 145 } sv_fp[8]; 146 struct xmmacc sv_xmm[16]; 147 uint8_t sv_pad[96]; 148} __aligned(16); 149#endif 150 151struct xstate_hdr { 152 uint64_t xstate_bv; 153 uint64_t xstate_xcomp_bv; 154 uint8_t xstate_rsrv0[8]; 155 uint8_t xstate_rsrv[40]; 156}; 157#define XSTATE_XCOMP_BV_COMPACT (1ULL << 63) 158 159struct savexmm_xstate { 160 struct xstate_hdr sx_hd; 161 struct ymmacc sx_ymm[16]; 162}; 163 164struct savexmm_ymm { 165 struct __envxmm32 sv_env; 166 struct { 167 struct fpacc87 fp_acc; 168 int8_t fp_pad[6]; /* padding */ 169 } sv_fp[8]; 170 struct xmmacc sv_xmm[16]; 171 uint8_t sv_pad[96]; 172 struct savexmm_xstate sv_xstate; 173} __aligned(64); 174 175struct savefpu_xstate { 176 struct xstate_hdr sx_hd; 177 struct ymmacc sx_ymm[16]; 178}; 179 180struct savefpu_ymm { 181 struct __envxmm64 sv_env; 182 struct { 183 struct fpacc87 fp_acc; 184 int8_t fp_pad[6]; /* padding */ 185 } sv_fp[8]; 186 struct xmmacc sv_xmm[16]; 187 uint8_t sv_pad[96]; 188 struct savefpu_xstate sv_xstate; 189} __aligned(64); 190 191#undef __envxmm32 192#undef __envxmm64 193 194/* 195 * The hardware default control word for i387's and later coprocessors is 196 * 0x37F, giving: 197 * 198 * round to nearest 199 * 64-bit precision 200 * all exceptions masked. 201 * 202 * FreeBSD/i386 uses 53 bit precision for things like fadd/fsub/fsqrt etc 203 * because of the difference between memory and fpu register stack arguments. 204 * If its using an intermediate fpu register, it has 80/64 bits to work 205 * with. If it uses memory, it has 64/53 bits to work with. However, 206 * gcc is aware of this and goes to a fair bit of trouble to make the 207 * best use of it. 208 * 209 * This is mostly academic for AMD64, because the ABI prefers the use 210 * SSE2 based math. For FreeBSD/amd64, we go with the default settings. 211 */ 212#define __INITIAL_FPUCW__ 0x037F 213#define __INITIAL_FPUCW_I386__ 0x127F 214#define __INITIAL_NPXCW__ __INITIAL_FPUCW_I386__ 215#define __INITIAL_MXCSR__ 0x1F80 216#define __INITIAL_MXCSR_MASK__ 0xFFBF 217 218#endif /* !_X86_FPU_H_ */ 219