src/gallium/auxiliary/util/u_cpu_detect.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2008 Dennis Smit
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * on the rights to use, copy, modify, merge, publish, distribute, sub
  10  * license, and/or sell copies of the Software, and to permit persons to whom
  11  * the Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
  20  * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  21  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  22  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  23  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  **************************************************************************/
  26
  27 /**
  28  * @file
  29  * CPU feature detection.
  30  *
  31  * @author Dennis Smit
  32  * @author Based on the work of Eric Anholt <anholt@FreeBSD.org>
  33  */
  34
  35 #include "pipe/p_config.h"
  36
  37 #include "u_debug.h"
  38 #include "u_cpu_detect.h"
  39
  40 #if defined(PIPE_ARCH_PPC)
  41 #if defined(PIPE_OS_APPLE)
  42 #include <sys/sysctl.h>
  43 #else
  44 #include <signal.h>
  45 #include <setjmp.h>
  46 #endif
  47 #endif
  48
  49 #if defined(PIPE_OS_NETBSD) || defined(PIPE_OS_OPENBSD)
  50 #include <sys/param.h>
  51 #include <sys/sysctl.h>
  52 #include <machine/cpu.h>
  53 #endif
  54
  55 #if defined(PIPE_OS_FREEBSD)
  56 #include <sys/types.h>
  57 #include <sys/sysctl.h>
  58 #endif
  59
  60 #if defined(PIPE_OS_LINUX)
  61 #include <signal.h>
  62 #endif
  63
  64 #ifdef PIPE_OS_UNIX
  65 #include <unistd.h>
  66 #endif
  67
  68 #if defined(PIPE_OS_WINDOWS)
  69 #include <windows.h>
  70 #if defined(MSVC)
  71 #include <intrin.h>
  72 #endif
  73 #endif
  74
  75
  76 DEBUG_GET_ONCE_BOOL_OPTION(dump_cpu, "GALLIUM_DUMP_CPU", FALSE)
  77
  78
  79 struct util_cpu_caps util_cpu_caps;
  80
  81 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
  82 static int has_cpuid(void);
  83 #endif
  84
  85
  86 #if defined(PIPE_ARCH_X86)
  87
  88 /* The sigill handlers */
  89 #if defined(PIPE_OS_LINUX) /*&& defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)*/
  90 static void
  91 sigill_handler_sse(int signal, struct sigcontext sc)
  92 {
  93    /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
  94     * instructions are 3 bytes long.  We must increment the instruction
  95     * pointer manually to avoid repeated execution of the offending
  96     * instruction.
  97     *
  98     * If the SIGILL is caused by a divide-by-zero when unmasked
  99     * exceptions aren't supported, the SIMD FPU status and control
 100     * word will be restored at the end of the test, so we don't need
 101     * to worry about doing it here.  Besides, we may not be able to...
 102     */
 103    sc.eip += 3;
 104
 105    util_cpu_caps.has_sse=0;
 106 }
 107
 108 static void
 109 sigfpe_handler_sse(int signal, struct sigcontext sc)
 110 {
 111    if (sc.fpstate->magic != 0xffff) {
 112       /* Our signal context has the extended FPU state, so reset the
 113        * divide-by-zero exception mask and clear the divide-by-zero
 114        * exception bit.
 115        */
 116       sc.fpstate->mxcsr |= 0x00000200;
 117       sc.fpstate->mxcsr &= 0xfffffffb;
 118    } else {
 119       /* If we ever get here, we're completely hosed.
 120       */
 121    }
 122 }
 123 #endif /* PIPE_OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
 124
 125 #if defined(PIPE_OS_WINDOWS)
 126 static LONG CALLBACK
 127 win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
 128 {
 129    if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
 130       ep->ContextRecord->Eip +=3;
 131       util_cpu_caps.has_sse=0;
 132       return EXCEPTION_CONTINUE_EXECUTION;
 133    }
 134    return EXCEPTION_CONTINUE_SEARCH;
 135 }
 136 #endif /* PIPE_OS_WINDOWS */
 137
 138 #endif /* PIPE_ARCH_X86 */
 139
 140
 141 #if defined(PIPE_ARCH_PPC) && !defined(PIPE_OS_APPLE)
 142 static jmp_buf  __lv_powerpc_jmpbuf;
 143 static volatile sig_atomic_t __lv_powerpc_canjump = 0;
 144
 145 static void
 146 sigill_handler(int sig)
 147 {
 148    if (!__lv_powerpc_canjump) {
 149       signal (sig, SIG_DFL);
 150       raise (sig);
 151    }
 152
 153    __lv_powerpc_canjump = 0;
 154    longjmp(__lv_powerpc_jmpbuf, 1);
 155 }
 156 #endif
 157
 158 #if defined(PIPE_ARCH_PPC)
 159 static void
 160 check_os_altivec_support(void)
 161 {
 162 #if defined(PIPE_OS_APPLE)
 163    int sels[2] = {CTL_HW, HW_VECTORUNIT};
 164    int has_vu = 0;
 165    int len = sizeof (has_vu);
 166    int err;
 167
 168    err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
 169
 170    if (err == 0) {
 171       if (has_vu != 0) {
 172          util_cpu_caps.has_altivec = 1;
 173       }
 174    }
 175 #else /* !PIPE_OS_APPLE */
 176    /* not on Apple/Darwin, do it the brute-force way */
 177    /* this is borrowed from the libmpeg2 library */
 178    signal(SIGILL, sigill_handler);
 179    if (setjmp(__lv_powerpc_jmpbuf)) {
 180       signal(SIGILL, SIG_DFL);
 181    } else {
 182       __lv_powerpc_canjump = 1;
 183
 184       __asm __volatile
 185          ("mtspr 256, %0\n\t"
 186           "vand %%v0, %%v0, %%v0"
 187           :
 188           : "r" (-1));
 189
 190       signal(SIGILL, SIG_DFL);
 191       util_cpu_caps.has_altivec = 1;
 192    }
 193 #endif /* !PIPE_OS_APPLE */
 194 }
 195 #endif /* PIPE_ARCH_PPC */
 196
 197
 198 #if defined(PIPE_ARCH_X86) || defined (PIPE_ARCH_X86_64)
 199 static int has_cpuid(void)
 200 {
 201 #if defined(PIPE_ARCH_X86)
 202 #if defined(PIPE_OS_GCC)
 203    int a, c;
 204
 205    __asm __volatile
 206       ("pushf\n"
 207        "popl %0\n"
 208        "movl %0, %1\n"
 209        "xorl $0x200000, %0\n"
 210        "push %0\n"
 211        "popf\n"
 212        "pushf\n"
 213        "popl %0\n"
 214        : "=a" (a), "=c" (c)
 215        :
 216        : "cc");
 217
 218    return a != c;
 219 #else
 220    /* FIXME */
 221    return 1;
 222 #endif
 223 #elif defined(PIPE_ARCH_X86_64)
 224    return 1;
 225 #else
 226    return 0;
 227 #endif
 228 }
 229
 230
 231 /**
 232  * @sa cpuid.h included in gcc-4.3 onwards.
 233  * @sa http://msdn.microsoft.com/en-us/library/hskdteyh.aspx
 234  */
 235 static INLINE void
 236 cpuid(uint32_t ax, uint32_t *p)
 237 {
 238 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
 239    __asm __volatile (
 240      "xchgl %%ebx, %1\n\t"
 241      "cpuid\n\t"
 242      "xchgl %%ebx, %1"
 243      : "=a" (p[0]),
 244        "=S" (p[1]),
 245        "=c" (p[2]),
 246        "=d" (p[3])
 247      : "0" (ax)
 248    );
 249 #elif defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86_64)
 250    __asm __volatile (
 251      "cpuid\n\t"
 252      : "=a" (p[0]),
 253        "=b" (p[1]),
 254        "=c" (p[2]),
 255        "=d" (p[3])
 256      : "0" (ax)
 257    );
 258 #elif defined(PIPE_CC_MSVC)
 259    __cpuid(p, ax);
 260 #else
 261    p[0] = 0;
 262    p[1] = 0;
 263    p[2] = 0;
 264    p[3] = 0;
 265 #endif
 266 }
 267 #endif /* X86 or X86_64 */
 268
 269 void
 270 util_cpu_detect(void)
 271 {
 272    static boolean util_cpu_detect_initialized = FALSE;
 273
 274    if(util_cpu_detect_initialized)
 275       return;
 276
 277    memset(&util_cpu_caps, 0, sizeof util_cpu_caps);
 278
 279    /* Count the number of CPUs in system */
 280 #if defined(PIPE_OS_WINDOWS)
 281    {
 282       SYSTEM_INFO system_info;
 283       GetSystemInfo(&system_info);
 284       util_cpu_caps.nr_cpus = system_info.dwNumberOfProcessors;
 285    }
 286 #elif defined(PIPE_OS_UNIX) && defined(_SC_NPROCESSORS_ONLN)
 287    util_cpu_caps.nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 288    if (util_cpu_caps.nr_cpus == -1)
 289       util_cpu_caps.nr_cpus = 1;
 290 #elif defined(PIPE_OS_BSD)
 291    {
 292       int mib[2], ncpu;
 293       int len;
 294
 295       mib[0] = CTL_HW;
 296       mib[1] = HW_NCPU;
 297
 298       len = sizeof (ncpu);
 299       sysctl(mib, 2, &ncpu, &len, NULL, 0);
 300       util_cpu_caps.nr_cpus = ncpu;
 301    }
 302 #else
 303    util_cpu_caps.nr_cpus = 1;
 304 #endif
 305
 306 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
 307    if (has_cpuid()) {
 308       uint32_t regs[4];
 309       uint32_t regs2[4];
 310
 311       util_cpu_caps.cacheline = 32;
 312
 313       /* Get max cpuid level */
 314       cpuid(0x00000000, regs);
 315
 316       if (regs[0] >= 0x00000001) {
 317          unsigned int cacheline;
 318
 319          cpuid (0x00000001, regs2);
 320
 321          util_cpu_caps.x86_cpu_type = (regs2[0] >> 8) & 0xf;
 322          if (util_cpu_caps.x86_cpu_type == 0xf)
 323              util_cpu_caps.x86_cpu_type = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
 324
 325          /* general feature flags */
 326          util_cpu_caps.has_tsc    = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
 327          util_cpu_caps.has_mmx    = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
 328          util_cpu_caps.has_sse    = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
 329          util_cpu_caps.has_sse2   = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
 330          util_cpu_caps.has_sse3   = (regs2[2] & (1));          /* 0x0000001 */
 331          util_cpu_caps.has_ssse3  = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
 332          util_cpu_caps.has_sse4_1 = (regs2[2] & (1 << 19)) >> 19;
 333          util_cpu_caps.has_mmx2   = util_cpu_caps.has_sse; /* SSE cpus supports mmxext too */
 334
 335          cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
 336          if (cacheline > 0)
 337             util_cpu_caps.cacheline = cacheline;
 338       }
 339
 340       cpuid(0x80000000, regs);
 341
 342       if (regs[0] >= 0x80000001) {
 343
 344          cpuid(0x80000001, regs2);
 345
 346          util_cpu_caps.has_mmx  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
 347          util_cpu_caps.has_mmx2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
 348          util_cpu_caps.has_3dnow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
 349          util_cpu_caps.has_3dnow_ext = (regs2[3] & (1 << 30 )) >> 30;
 350       }
 351
 352       if (regs[0] >= 0x80000006) {
 353          cpuid(0x80000006, regs2);
 354          util_cpu_caps.cacheline = regs2[2] & 0xFF;
 355       }
 356
 357       if (!util_cpu_caps.has_sse) {
 358          util_cpu_caps.has_sse2 = 0;
 359          util_cpu_caps.has_sse3 = 0;
 360          util_cpu_caps.has_ssse3 = 0;
 361          util_cpu_caps.has_sse4_1 = 0;
 362       }
 363    }
 364 #endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
 365
 366 #if defined(PIPE_ARCH_PPC)
 367    check_os_altivec_support();
 368 #endif /* PIPE_ARCH_PPC */
 369
 370 #ifdef DEBUG
 371    if (debug_get_option_dump_cpu()) {
 372       debug_printf("util_cpu_caps.nr_cpus = %u\n", util_cpu_caps.nr_cpus);
 373
 374       debug_printf("util_cpu_caps.x86_cpu_type = %u\n", util_cpu_caps.x86_cpu_type);
 375       debug_printf("util_cpu_caps.cacheline = %u\n", util_cpu_caps.cacheline);
 376
 377       debug_printf("util_cpu_caps.has_tsc = %u\n", util_cpu_caps.has_tsc);
 378       debug_printf("util_cpu_caps.has_mmx = %u\n", util_cpu_caps.has_mmx);
 379       debug_printf("util_cpu_caps.has_mmx2 = %u\n", util_cpu_caps.has_mmx2);
 380       debug_printf("util_cpu_caps.has_sse = %u\n", util_cpu_caps.has_sse);
 381       debug_printf("util_cpu_caps.has_sse2 = %u\n", util_cpu_caps.has_sse2);
 382       debug_printf("util_cpu_caps.has_sse3 = %u\n", util_cpu_caps.has_sse3);
 383       debug_printf("util_cpu_caps.has_ssse3 = %u\n", util_cpu_caps.has_ssse3);
 384       debug_printf("util_cpu_caps.has_sse4_1 = %u\n", util_cpu_caps.has_sse4_1);
 385       debug_printf("util_cpu_caps.has_3dnow = %u\n", util_cpu_caps.has_3dnow);
 386       debug_printf("util_cpu_caps.has_3dnow_ext = %u\n", util_cpu_caps.has_3dnow_ext);
 387       debug_printf("util_cpu_caps.has_altivec = %u\n", util_cpu_caps.has_altivec);
 388    }
 389 #endif
 390
 391    util_cpu_detect_initialized = TRUE;
 392 }