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_PPC) && !defined(PIPE_OS_APPLE)
  87 static jmp_buf  __lv_powerpc_jmpbuf;
  88 static volatile sig_atomic_t __lv_powerpc_canjump = 0;
  89
  90 static void
  91 sigill_handler(int sig)
  92 {
  93    if (!__lv_powerpc_canjump) {
  94       signal (sig, SIG_DFL);
  95       raise (sig);
  96    }
  97
  98    __lv_powerpc_canjump = 0;
  99    longjmp(__lv_powerpc_jmpbuf, 1);
 100 }
 101 #endif
 102
 103 #if defined(PIPE_ARCH_PPC)
 104 static void
 105 check_os_altivec_support(void)
 106 {
 107 #if defined(PIPE_OS_APPLE)
 108    int sels[2] = {CTL_HW, HW_VECTORUNIT};
 109    int has_vu = 0;
 110    int len = sizeof (has_vu);
 111    int err;
 112
 113    err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
 114
 115    if (err == 0) {
 116       if (has_vu != 0) {
 117          util_cpu_caps.has_altivec = 1;
 118       }
 119    }
 120 #else /* !PIPE_OS_APPLE */
 121    /* not on Apple/Darwin, do it the brute-force way */
 122    /* this is borrowed from the libmpeg2 library */
 123    signal(SIGILL, sigill_handler);
 124    if (setjmp(__lv_powerpc_jmpbuf)) {
 125       signal(SIGILL, SIG_DFL);
 126    } else {
 127       __lv_powerpc_canjump = 1;
 128
 129       __asm __volatile
 130          ("mtspr 256, %0\n\t"
 131           "vand %%v0, %%v0, %%v0"
 132           :
 133           : "r" (-1));
 134
 135       signal(SIGILL, SIG_DFL);
 136       util_cpu_caps.has_altivec = 1;
 137    }
 138 #endif /* !PIPE_OS_APPLE */
 139 }
 140 #endif /* PIPE_ARCH_PPC */
 141
 142
 143 #if defined(PIPE_ARCH_X86) || defined (PIPE_ARCH_X86_64)
 144 static int has_cpuid(void)
 145 {
 146 #if defined(PIPE_ARCH_X86)
 147 #if defined(PIPE_OS_GCC)
 148    int a, c;
 149
 150    __asm __volatile
 151       ("pushf\n"
 152        "popl %0\n"
 153        "movl %0, %1\n"
 154        "xorl $0x200000, %0\n"
 155        "push %0\n"
 156        "popf\n"
 157        "pushf\n"
 158        "popl %0\n"
 159        : "=a" (a), "=c" (c)
 160        :
 161        : "cc");
 162
 163    return a != c;
 164 #else
 165    /* FIXME */
 166    return 1;
 167 #endif
 168 #elif defined(PIPE_ARCH_X86_64)
 169    return 1;
 170 #else
 171    return 0;
 172 #endif
 173 }
 174
 175
 176 /**
 177  * @sa cpuid.h included in gcc-4.3 onwards.
 178  * @sa http://msdn.microsoft.com/en-us/library/hskdteyh.aspx
 179  */
 180 static INLINE void
 181 cpuid(uint32_t ax, uint32_t *p)
 182 {
 183 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
 184    __asm __volatile (
 185      "xchgl %%ebx, %1\n\t"
 186      "cpuid\n\t"
 187      "xchgl %%ebx, %1"
 188      : "=a" (p[0]),
 189        "=S" (p[1]),
 190        "=c" (p[2]),
 191        "=d" (p[3])
 192      : "0" (ax)
 193    );
 194 #elif defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86_64)
 195    __asm __volatile (
 196      "cpuid\n\t"
 197      : "=a" (p[0]),
 198        "=b" (p[1]),
 199        "=c" (p[2]),
 200        "=d" (p[3])
 201      : "0" (ax)
 202    );
 203 #elif defined(PIPE_CC_MSVC)
 204    __cpuid(p, ax);
 205 #else
 206    p[0] = 0;
 207    p[1] = 0;
 208    p[2] = 0;
 209    p[3] = 0;
 210 #endif
 211 }
 212 #endif /* X86 or X86_64 */
 213
 214 void
 215 util_cpu_detect(void)
 216 {
 217    static boolean util_cpu_detect_initialized = FALSE;
 218
 219    if(util_cpu_detect_initialized)
 220       return;
 221
 222    memset(&util_cpu_caps, 0, sizeof util_cpu_caps);
 223
 224    /* Count the number of CPUs in system */
 225 #if defined(PIPE_OS_WINDOWS)
 226    {
 227       SYSTEM_INFO system_info;
 228       GetSystemInfo(&system_info);
 229       util_cpu_caps.nr_cpus = system_info.dwNumberOfProcessors;
 230    }
 231 #elif defined(PIPE_OS_UNIX) && defined(_SC_NPROCESSORS_ONLN)
 232    util_cpu_caps.nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 233    if (util_cpu_caps.nr_cpus == -1)
 234       util_cpu_caps.nr_cpus = 1;
 235 #elif defined(PIPE_OS_BSD)
 236    {
 237       int mib[2], ncpu;
 238       int len;
 239
 240       mib[0] = CTL_HW;
 241       mib[1] = HW_NCPU;
 242
 243       len = sizeof (ncpu);
 244       sysctl(mib, 2, &ncpu, &len, NULL, 0);
 245       util_cpu_caps.nr_cpus = ncpu;
 246    }
 247 #else
 248    util_cpu_caps.nr_cpus = 1;
 249 #endif
 250
 251 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
 252    if (has_cpuid()) {
 253       uint32_t regs[4];
 254       uint32_t regs2[4];
 255
 256       util_cpu_caps.cacheline = 32;
 257
 258       /* Get max cpuid level */
 259       cpuid(0x00000000, regs);
 260
 261       if (regs[0] >= 0x00000001) {
 262          unsigned int cacheline;
 263
 264          cpuid (0x00000001, regs2);
 265
 266          util_cpu_caps.x86_cpu_type = (regs2[0] >> 8) & 0xf;
 267          if (util_cpu_caps.x86_cpu_type == 0xf)
 268              util_cpu_caps.x86_cpu_type = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
 269
 270          /* general feature flags */
 271          util_cpu_caps.has_tsc    = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
 272          util_cpu_caps.has_mmx    = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
 273          util_cpu_caps.has_sse    = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
 274          util_cpu_caps.has_sse2   = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
 275          util_cpu_caps.has_sse3   = (regs2[2] & (1));          /* 0x0000001 */
 276          util_cpu_caps.has_ssse3  = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
 277          util_cpu_caps.has_sse4_1 = (regs2[2] & (1 << 19)) >> 19;
 278          util_cpu_caps.has_mmx2   = util_cpu_caps.has_sse; /* SSE cpus supports mmxext too */
 279
 280          cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
 281          if (cacheline > 0)
 282             util_cpu_caps.cacheline = cacheline;
 283       }
 284
 285       cpuid(0x80000000, regs);
 286
 287       if (regs[0] >= 0x80000001) {
 288
 289          cpuid(0x80000001, regs2);
 290
 291          util_cpu_caps.has_mmx  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
 292          util_cpu_caps.has_mmx2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
 293          util_cpu_caps.has_3dnow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
 294          util_cpu_caps.has_3dnow_ext = (regs2[3] & (1 << 30 )) >> 30;
 295       }
 296
 297       if (regs[0] >= 0x80000006) {
 298          cpuid(0x80000006, regs2);
 299          util_cpu_caps.cacheline = regs2[2] & 0xFF;
 300       }
 301
 302       if (!util_cpu_caps.has_sse) {
 303          util_cpu_caps.has_sse2 = 0;
 304          util_cpu_caps.has_sse3 = 0;
 305          util_cpu_caps.has_ssse3 = 0;
 306          util_cpu_caps.has_sse4_1 = 0;
 307       }
 308    }
 309 #endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
 310
 311 #if defined(PIPE_ARCH_PPC)
 312    check_os_altivec_support();
 313 #endif /* PIPE_ARCH_PPC */
 314
 315 #ifdef DEBUG
 316    if (debug_get_option_dump_cpu()) {
 317       debug_printf("util_cpu_caps.nr_cpus = %u\n", util_cpu_caps.nr_cpus);
 318
 319       debug_printf("util_cpu_caps.x86_cpu_type = %u\n", util_cpu_caps.x86_cpu_type);
 320       debug_printf("util_cpu_caps.cacheline = %u\n", util_cpu_caps.cacheline);
 321
 322       debug_printf("util_cpu_caps.has_tsc = %u\n", util_cpu_caps.has_tsc);
 323       debug_printf("util_cpu_caps.has_mmx = %u\n", util_cpu_caps.has_mmx);
 324       debug_printf("util_cpu_caps.has_mmx2 = %u\n", util_cpu_caps.has_mmx2);
 325       debug_printf("util_cpu_caps.has_sse = %u\n", util_cpu_caps.has_sse);
 326       debug_printf("util_cpu_caps.has_sse2 = %u\n", util_cpu_caps.has_sse2);
 327       debug_printf("util_cpu_caps.has_sse3 = %u\n", util_cpu_caps.has_sse3);
 328       debug_printf("util_cpu_caps.has_ssse3 = %u\n", util_cpu_caps.has_ssse3);
 329       debug_printf("util_cpu_caps.has_sse4_1 = %u\n", util_cpu_caps.has_sse4_1);
 330       debug_printf("util_cpu_caps.has_3dnow = %u\n", util_cpu_caps.has_3dnow);
 331       debug_printf("util_cpu_caps.has_3dnow_ext = %u\n", util_cpu_caps.has_3dnow_ext);
 332       debug_printf("util_cpu_caps.has_altivec = %u\n", util_cpu_caps.has_altivec);
 333    }
 334 #endif
 335
 336    util_cpu_detect_initialized = TRUE;
 337 }