src/libFLAC/cpu.c

   1 /* libFLAC - Free Lossless Audio Codec library
   2  * Copyright (C) 2001-2009  Josh Coalson
   3  * Copyright (C) 2011-2016  Xiph.Org Foundation
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  *
   9  * - Redistributions of source code must retain the above copyright
  10  * notice, this list of conditions and the following disclaimer.
  11  *
  12  * - Redistributions in binary form must reproduce the above copyright
  13  * notice, this list of conditions and the following disclaimer in the
  14  * documentation and/or other materials provided with the distribution.
  15  *
  16  * - Neither the name of the Xiph.org Foundation nor the names of its
  17  * contributors may be used to endorse or promote products derived from
  18  * this software without specific prior written permission.
  19  *
  20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  23  * A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR
  24  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
  25  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  26  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
  27  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  28  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  29  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  30  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  31  */
  32
  33 #ifdef HAVE_CONFIG_H
  34 #  include <config.h>
  35 #endif
  36
  37 #include "private/cpu.h"
  38 #include "share/compat.h"
  39 #include <stdlib.h>
  40 #include <string.h>
  41
  42 #if defined _MSC_VER
  43 #include <intrin.h> /* for __cpuid() and _xgetbv() */
  44 #elif defined __GNUC__ && defined HAVE_CPUID_H
  45 #include <cpuid.h> /* for __get_cpuid() and __get_cpuid_max() */
  46 #endif
  47
  48 #ifndef NDEBUG
  49 #include <stdio.h>
  50 #define dfprintf fprintf
  51 #else
  52 /* This is bad practice, it should be a static void empty function */
  53 #define dfprintf(file, format, ...)
  54 #endif
  55
  56 #if defined FLAC__CPU_PPC
  57 #if defined(__linux__) || (defined(__FreeBSD__) && (__FreeBSD__ >= 12))
  58 #include <sys/auxv.h>
  59 #endif
  60 #endif
  61
  62 #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && (defined FLAC__HAS_NASM || FLAC__HAS_X86INTRIN) && !defined FLAC__NO_ASM
  63
  64 /* these are flags in EDX of CPUID AX=00000001 */
  65 static const uint32_t FLAC__CPUINFO_X86_CPUID_CMOV    = 0x00008000;
  66 static const uint32_t FLAC__CPUINFO_X86_CPUID_MMX     = 0x00800000;
  67 static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE     = 0x02000000;
  68 static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE2    = 0x04000000;
  69
  70 /* these are flags in ECX of CPUID AX=00000001 */
  71 static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE3    = 0x00000001;
  72 static const uint32_t FLAC__CPUINFO_X86_CPUID_SSSE3   = 0x00000200;
  73 static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE41   = 0x00080000;
  74 static const uint32_t FLAC__CPUINFO_X86_CPUID_SSE42   = 0x00100000;
  75 static const uint32_t FLAC__CPUINFO_X86_CPUID_OSXSAVE = 0x08000000;
  76 static const uint32_t FLAC__CPUINFO_X86_CPUID_AVX     = 0x10000000;
  77 static const uint32_t FLAC__CPUINFO_X86_CPUID_FMA     = 0x00001000;
  78
  79 /* these are flags in EBX of CPUID AX=00000007 */
  80 static const uint32_t FLAC__CPUINFO_X86_CPUID_AVX2    = 0x00000020;
  81
  82 static uint32_t
  83 cpu_xgetbv_x86(void)
  84 {
  85 #if (defined _MSC_VER || defined __INTEL_COMPILER) && FLAC__AVX_SUPPORTED
  86         return (uint32_t)_xgetbv(0);
  87 #elif defined __GNUC__
  88         uint32_t lo, hi;
  89         __asm__ volatile (".byte 0x0f, 0x01, 0xd0" : "=a"(lo), "=d"(hi) : "c" (0));
  90         return lo;
  91 #else
  92         return 0;
  93 #endif
  94 }
  95
  96 static uint32_t
  97 cpu_have_cpuid(void)
  98 {
  99 #if defined FLAC__CPU_X86_64 || defined __i686__ || defined __SSE__ || (defined _M_IX86_FP && _M_IX86_FP > 0)
 100         /* target CPU does have CPUID instruction */
 101         return 1;
 102 #elif defined FLAC__HAS_NASM
 103         return FLAC__cpu_have_cpuid_asm_ia32();
 104 #elif defined __GNUC__ && defined HAVE_CPUID_H
 105         if (__get_cpuid_max(0, 0) != 0)
 106                 return 1;
 107         else
 108                 return 0;
 109 #elif defined _MSC_VER
 110         FLAC__uint32 flags1, flags2;
 111         __asm {
 112                 pushfd
 113                 pushfd
 114                 pop             eax
 115                 mov             flags1, eax
 116                 xor             eax, 0x200000
 117                 push    eax
 118                 popfd
 119                 pushfd
 120                 pop             eax
 121                 mov             flags2, eax
 122                 popfd
 123         }
 124         if (((flags1^flags2) & 0x200000) != 0)
 125                 return 1;
 126         else
 127                 return 0;
 128 #else
 129         return 0;
 130 #endif
 131 }
 132
 133 static void
 134 cpuinfo_x86(FLAC__uint32 level, FLAC__uint32 *eax, FLAC__uint32 *ebx, FLAC__uint32 *ecx, FLAC__uint32 *edx)
 135 {
 136 #if defined _MSC_VER
 137         int cpuinfo[4];
 138         int ext = level & 0x80000000;
 139         __cpuid(cpuinfo, ext);
 140         if ((uint32_t)cpuinfo[0] >= level) {
 141 #if FLAC__AVX_SUPPORTED
 142                 __cpuidex(cpuinfo, level, 0); /* for AVX2 detection */
 143 #else
 144                 __cpuid(cpuinfo, level); /* some old compilers don't support __cpuidex */
 145 #endif
 146                 *eax = cpuinfo[0]; *ebx = cpuinfo[1]; *ecx = cpuinfo[2]; *edx = cpuinfo[3];
 147                 return;
 148         }
 149 #elif defined __GNUC__ && defined HAVE_CPUID_H
 150         FLAC__uint32 ext = level & 0x80000000;
 151         __cpuid(ext, *eax, *ebx, *ecx, *edx);
 152         if (*eax >= level) {
 153                 __cpuid_count(level, 0, *eax, *ebx, *ecx, *edx);
 154                 return;
 155         }
 156 #elif defined FLAC__HAS_NASM && defined FLAC__CPU_IA32
 157         FLAC__cpu_info_asm_ia32(level, eax, ebx, ecx, edx);
 158         return;
 159 #endif
 160         *eax = *ebx = *ecx = *edx = 0;
 161 }
 162
 163 #endif
 164
 165 static void
 166 x86_cpu_info (FLAC__CPUInfo *info)
 167 {
 168 #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && (defined FLAC__HAS_NASM || FLAC__HAS_X86INTRIN) && !defined FLAC__NO_ASM
 169         FLAC__bool x86_osxsave = false;
 170         FLAC__bool os_avx = false;
 171         FLAC__uint32 flags_eax, flags_ebx, flags_ecx, flags_edx;
 172
 173         info->use_asm = true; /* we assume a minimum of 80386 */
 174         if (!cpu_have_cpuid())
 175                 return;
 176
 177         cpuinfo_x86(0, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
 178         info->x86.intel = (flags_ebx == 0x756E6547 && flags_edx == 0x49656E69 && flags_ecx == 0x6C65746E) ? true : false; /* GenuineIntel */
 179         cpuinfo_x86(1, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
 180
 181         info->x86.cmov  = (flags_edx & FLAC__CPUINFO_X86_CPUID_CMOV ) ? true : false;
 182         info->x86.mmx   = (flags_edx & FLAC__CPUINFO_X86_CPUID_MMX  ) ? true : false;
 183         info->x86.sse   = (flags_edx & FLAC__CPUINFO_X86_CPUID_SSE  ) ? true : false;
 184         info->x86.sse2  = (flags_edx & FLAC__CPUINFO_X86_CPUID_SSE2 ) ? true : false;
 185         info->x86.sse3  = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSE3 ) ? true : false;
 186         info->x86.ssse3 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSSE3) ? true : false;
 187         info->x86.sse41 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSE41) ? true : false;
 188         info->x86.sse42 = (flags_ecx & FLAC__CPUINFO_X86_CPUID_SSE42) ? true : false;
 189
 190         if (FLAC__AVX_SUPPORTED) {
 191                 x86_osxsave     = (flags_ecx & FLAC__CPUINFO_X86_CPUID_OSXSAVE) ? true : false;
 192                 info->x86.avx   = (flags_ecx & FLAC__CPUINFO_X86_CPUID_AVX    ) ? true : false;
 193                 info->x86.fma   = (flags_ecx & FLAC__CPUINFO_X86_CPUID_FMA    ) ? true : false;
 194                 cpuinfo_x86(7, &flags_eax, &flags_ebx, &flags_ecx, &flags_edx);
 195                 info->x86.avx2  = (flags_ebx & FLAC__CPUINFO_X86_CPUID_AVX2   ) ? true : false;
 196         }
 197
 198 #if defined FLAC__CPU_IA32
 199         dfprintf(stderr, "CPU info (IA-32):\n");
 200 #else
 201         dfprintf(stderr, "CPU info (x86-64):\n");
 202 #endif
 203         dfprintf(stderr, "  CMOV ....... %c\n", info->x86.cmov    ? 'Y' : 'n');
 204         dfprintf(stderr, "  MMX ........ %c\n", info->x86.mmx     ? 'Y' : 'n');
 205         dfprintf(stderr, "  SSE ........ %c\n", info->x86.sse     ? 'Y' : 'n');
 206         dfprintf(stderr, "  SSE2 ....... %c\n", info->x86.sse2    ? 'Y' : 'n');
 207         dfprintf(stderr, "  SSE3 ....... %c\n", info->x86.sse3    ? 'Y' : 'n');
 208         dfprintf(stderr, "  SSSE3 ...... %c\n", info->x86.ssse3   ? 'Y' : 'n');
 209         dfprintf(stderr, "  SSE41 ...... %c\n", info->x86.sse41   ? 'Y' : 'n');
 210         dfprintf(stderr, "  SSE42 ...... %c\n", info->x86.sse42   ? 'Y' : 'n');
 211
 212         if (FLAC__AVX_SUPPORTED) {
 213                 dfprintf(stderr, "  AVX ........ %c\n", info->x86.avx     ? 'Y' : 'n');
 214                 dfprintf(stderr, "  FMA ........ %c\n", info->x86.fma     ? 'Y' : 'n');
 215                 dfprintf(stderr, "  AVX2 ....... %c\n", info->x86.avx2    ? 'Y' : 'n');
 216         }
 217
 218         /*
 219          * now have to check for OS support of AVX instructions
 220          */
 221         if (FLAC__AVX_SUPPORTED && info->x86.avx && x86_osxsave && (cpu_xgetbv_x86() & 0x6) == 0x6) {
 222                 os_avx = true;
 223         }
 224         if (os_avx) {
 225                 dfprintf(stderr, "  AVX OS sup . %c\n", info->x86.avx ? 'Y' : 'n');
 226         }
 227         if (!os_avx) {
 228                 /* no OS AVX support */
 229                 info->x86.avx     = false;
 230                 info->x86.avx2    = false;
 231                 info->x86.fma     = false;
 232         }
 233 #else
 234         info->use_asm = false;
 235 #endif
 236 }
 237
 238 static void
 239 ppc_cpu_info (FLAC__CPUInfo *info)
 240 {
 241 #if defined FLAC__CPU_PPC
 242 #ifndef PPC_FEATURE2_ARCH_3_00
 243 #define PPC_FEATURE2_ARCH_3_00          0x00800000
 244 #endif
 245
 246 #ifndef PPC_FEATURE2_ARCH_2_07
 247 #define PPC_FEATURE2_ARCH_2_07          0x80000000
 248 #endif
 249
 250 #ifdef __linux__
 251         if (getauxval(AT_HWCAP2) & PPC_FEATURE2_ARCH_3_00) {
 252                 info->ppc.arch_3_00 = true;
 253         } else if (getauxval(AT_HWCAP2) & PPC_FEATURE2_ARCH_2_07) {
 254                 info->ppc.arch_2_07 = true;
 255         }
 256 #elif defined(__FreeBSD__) && (__FreeBSD__ >= 12)
 257         long hwcaps;
 258         /* elf_aux_info() appeared in FreeBSD 12.0 */
 259         elf_aux_info(AT_HWCAP2, &hwcaps, sizeof(hwcaps));
 260         if (hwcaps & PPC_FEATURE2_ARCH_3_00) {
 261                 info->ppc.arch_3_00 = true;
 262         } else if (hwcaps & PPC_FEATURE2_ARCH_2_07) {
 263                 info->ppc.arch_2_07 = true;
 264         }
 265 #elif defined(__APPLE__)
 266         /* no Mac OS X version supports CPU with Power AVI v2.07 or better */
 267         info->ppc.arch_2_07 = false;
 268         info->ppc.arch_3_00 = false;
 269 #else
 270 #error Unsupported platform! Please add support for reading ppc hwcaps.
 271 #endif
 272
 273 #else
 274         info->ppc.arch_2_07 = false;
 275         info->ppc.arch_3_00 = false;
 276 #endif
 277 }
 278
 279 void FLAC__cpu_info (FLAC__CPUInfo *info)
 280 {
 281         memset(info, 0, sizeof(*info));
 282
 283 #ifdef FLAC__CPU_IA32
 284         info->type = FLAC__CPUINFO_TYPE_IA32;
 285 #elif defined FLAC__CPU_X86_64
 286         info->type = FLAC__CPUINFO_TYPE_X86_64;
 287 #elif defined FLAC__CPU_PPC
 288         info->type = FLAC__CPUINFO_TYPE_PPC;
 289 #else
 290         info->type = FLAC__CPUINFO_TYPE_UNKNOWN;
 291 #endif
 292
 293         switch (info->type) {
 294         case FLAC__CPUINFO_TYPE_IA32: /* fallthrough */
 295         case FLAC__CPUINFO_TYPE_X86_64:
 296                 x86_cpu_info (info);
 297                 break;
 298         case FLAC__CPUINFO_TYPE_PPC:
 299                 ppc_cpu_info (info);
 300                 break;
 301         default:
 302                 info->use_asm = false;
 303                 break;
 304         }
 305 }