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  * Based on the work of Eric Anholt <anholt@FreeBSD.org>
  29  */
  30
  31 /* FIXME: clean this entire file up */
  32
  33 #include "u_cpu_detect.h"
  34
  35 #ifdef __linux__
  36 #define OS_LINUX
  37 #endif
  38 #ifdef WIN32
  39 #define OS_WIN32
  40 #endif
  41
  42 #if defined(ARCH_POWERPC)
  43 #if defined(OS_DARWIN)
  44 #include <sys/sysctl.h>
  45 #else
  46 #include <signal.h>
  47 #include <setjmp.h>
  48 #endif
  49 #endif
  50
  51 #if defined(OS_NETBSD) || defined(OS_OPENBSD)
  52 #include <sys/param.h>
  53 #include <sys/sysctl.h>
  54 #include <machine/cpu.h>
  55 #endif
  56
  57 #if defined(OS_FREEBSD)
  58 #include <sys/types.h>
  59 #include <sys/sysctl.h>
  60 #endif
  61
  62 #if defined(OS_LINUX)
  63 #include <signal.h>
  64 #endif
  65
  66 #if defined(OS_WIN32)
  67 #include <windows.h>
  68 #endif
  69
  70 #include <stdio.h>
  71 #include <stdlib.h>
  72 #include <unistd.h>
  73 #include <string.h>
  74
  75
  76 static struct cpu_detect_caps __cpu_detect_caps;
  77 static int __cpu_detect_initialized = 0;
  78
  79 static int has_cpuid(void);
  80 static int cpuid(unsigned int ax, unsigned int *p);
  81
  82 /* The sigill handlers */
  83 #if defined(ARCH_X86) /* x86 (linux katmai handler check thing) */
  84 #if defined(OS_LINUX) && defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
  85 static void sigill_handler_sse(int signal, struct sigcontext sc)
  86 {
  87         /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
  88          * instructions are 3 bytes long.  We must increment the instruction
  89          * pointer manually to avoid repeated execution of the offending
  90          * instruction.
  91          *
  92          * If the SIGILL is caused by a divide-by-zero when unmasked
  93          * exceptions aren't supported, the SIMD FPU status and control
  94          * word will be restored at the end of the test, so we don't need
  95          * to worry about doing it here.  Besides, we may not be able to...
  96          */
  97         sc.eip += 3;
  98
  99         __cpu_detect_caps.hasSSE=0;
 100 }
 101
 102 static void sigfpe_handler_sse(int signal, struct sigcontext sc)
 103 {
 104         if (sc.fpstate->magic != 0xffff) {
 105                 /* Our signal context has the extended FPU state, so reset the
 106                  * divide-by-zero exception mask and clear the divide-by-zero
 107                  * exception bit.
 108                  */
 109                 sc.fpstate->mxcsr |= 0x00000200;
 110                 sc.fpstate->mxcsr &= 0xfffffffb;
 111         } else {
 112                 /* If we ever get here, we're completely hosed.
 113                 */
 114         }
 115 }
 116 #endif
 117 #endif /* OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
 118
 119 #if defined(OS_WIN32)
 120 LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
 121 {
 122         if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
 123                 ep->ContextRecord->Eip +=3;
 124                 __cpu_detect_caps.hasSSE=0;
 125                 return EXCEPTION_CONTINUE_EXECUTION;
 126         }
 127         return EXCEPTION_CONTINUE_SEARCH;
 128 }
 129 #endif /* OS_WIN32 */
 130
 131
 132 #if defined(ARCH_POWERPC) && !defined(OS_DARWIN)
 133 static sigjmp_buf __lv_powerpc_jmpbuf;
 134 static volatile sig_atomic_t __lv_powerpc_canjump = 0;
 135
 136 static void sigill_handler (int sig);
 137
 138 static void sigill_handler (int sig)
 139 {
 140         if (!__lv_powerpc_canjump) {
 141                 signal (sig, SIG_DFL);
 142                 raise (sig);
 143         }
 144
 145         __lv_powerpc_canjump = 0;
 146         siglongjmp(__lv_powerpc_jmpbuf, 1);
 147 }
 148
 149 static void check_os_altivec_support(void)
 150 {
 151 #if defined(OS_DARWIN)
 152         int sels[2] = {CTL_HW, HW_VECTORUNIT};
 153         int has_vu = 0;
 154         int len = sizeof (has_vu);
 155         int err;
 156
 157         err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
 158
 159         if (err == 0) {
 160                 if (has_vu != 0) {
 161                         __cpu_detect_caps.hasAltiVec = 1;
 162                 }
 163         }
 164 #else /* !OS_DARWIN */
 165         /* no Darwin, do it the brute-force way */
 166         /* this is borrowed from the libmpeg2 library */
 167         signal(SIGILL, sigill_handler);
 168         if (sigsetjmp(__lv_powerpc_jmpbuf, 1)) {
 169                 signal(SIGILL, SIG_DFL);
 170         } else {
 171                 __lv_powerpc_canjump = 1;
 172
 173                 __asm __volatile
 174                         ("mtspr 256, %0\n\t"
 175                          "vand %%v0, %%v0, %%v0"
 176                          :
 177                          : "r" (-1));
 178
 179                 signal(SIGILL, SIG_DFL);
 180                 __cpu_detect_caps.hasAltiVec = 1;
 181         }
 182 #endif
 183 }
 184 #endif
 185
 186 /* If we're running on a processor that can do SSE, let's see if we
 187  * are allowed to or not.  This will catch 2.4.0 or later kernels that
 188  * haven't been configured for a Pentium III but are running on one,
 189  * and RedHat patched 2.2 kernels that have broken exception handling
 190  * support for user space apps that do SSE.
 191  */
 192 static void check_os_katmai_support(void)
 193 {
 194 #if defined(ARCH_X86)
 195 #if defined(OS_FREEBSD)
 196         int has_sse=0, ret;
 197         int len = sizeof (has_sse);
 198
 199         ret = sysctlbyname("hw.instruction_sse", &has_sse, &len, NULL, 0);
 200         if (ret || !has_sse)
 201                 __cpu_detect_caps.hasSSE=0;
 202
 203 #elif defined(OS_NETBSD) || defined(OS_OPENBSD)
 204         int has_sse, has_sse2, ret, mib[2];
 205         int varlen;
 206
 207         mib[0] = CTL_MACHDEP;
 208         mib[1] = CPU_SSE;
 209         varlen = sizeof (has_sse);
 210
 211         ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
 212         if (ret < 0 || !has_sse) {
 213                 __cpu_detect_caps.hasSSE = 0;
 214         } else {
 215                 __cpu_detect_caps.hasSSE = 1;
 216         }
 217
 218         mib[1] = CPU_SSE2;
 219         varlen = sizeof (has_sse2);
 220         ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
 221         if (ret < 0 || !has_sse2) {
 222                 __cpu_detect_caps.hasSSE2 = 0;
 223         } else {
 224                 __cpu_detect_caps.hasSSE2 = 1;
 225         }
 226         __cpu_detect_caps.hasSSE = 0; /* FIXME ?!?!? */
 227
 228 #elif defined(OS_WIN32)
 229         LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
 230         if (__cpu_detect_caps.hasSSE) {
 231                 exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
 232                 __asm __volatile ("xorps %xmm0, %xmm0");
 233                 SetUnhandledExceptionFilter(exc_fil);
 234         }
 235 #elif defined(OS_LINUX)
 236         struct sigaction saved_sigill;
 237         struct sigaction saved_sigfpe;
 238
 239         /* Save the original signal handlers.
 240         */
 241         sigaction(SIGILL, NULL, &saved_sigill);
 242         sigaction(SIGFPE, NULL, &saved_sigfpe);
 243
 244         signal(SIGILL, (void (*)(int))sigill_handler_sse);
 245         signal(SIGFPE, (void (*)(int))sigfpe_handler_sse);
 246
 247         /* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
 248          * supports the extended FPU save and restore required for SSE.  If
 249          * we execute an SSE instruction on a PIII and get a SIGILL, the OS
 250          * doesn't support Streaming SIMD Exceptions, even if the processor
 251          * does.
 252          */
 253         if (__cpu_detect_caps.hasSSE) {
 254                 __asm __volatile ("xorps %xmm1, %xmm0");
 255         }
 256
 257         /* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
 258          * it supports unmasked SIMD FPU exceptions.  If we unmask the
 259          * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
 260          * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
 261          * as expected, we're okay but we need to clean up after it.
 262          *
 263          * Are we being too stringent in our requirement that the OS support
 264          * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
 265          * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
 266          * doesn't even support them.  We at least know the user-space SSE
 267          * support is good in kernels that do support unmasked exceptions,
 268          * and therefore to be safe I'm going to leave this test in here.
 269          */
 270         if (__cpu_detect_caps.hasSSE) {
 271                 //      test_os_katmai_exception_support();
 272         }
 273
 274         /* Restore the original signal handlers.
 275         */
 276         sigaction(SIGILL, &saved_sigill, NULL);
 277         sigaction(SIGFPE, &saved_sigfpe, NULL);
 278
 279 #else
 280         /* We can't use POSIX signal handling to test the availability of
 281          * SSE, so we disable it by default.
 282          */
 283         __cpu_detect_caps.hasSSE = 0;
 284 #endif /* __linux__ */
 285 #endif
 286 }
 287
 288
 289 static int has_cpuid(void)
 290 {
 291 #if defined(ARCH_X86)
 292         int a, c;
 293
 294         __asm __volatile
 295                 ("pushf\n"
 296                  "popl %0\n"
 297                  "movl %0, %1\n"
 298                  "xorl $0x200000, %0\n"
 299                  "push %0\n"
 300                  "popf\n"
 301                  "pushf\n"
 302                  "popl %0\n"
 303                  : "=a" (a), "=c" (c)
 304                  :
 305                  : "cc");
 306
 307         return a != c;
 308 #else
 309         return 0;
 310 #endif
 311 }
 312
 313 static int cpuid(unsigned int ax, unsigned int *p)
 314 {
 315 #if defined(ARCH_X86)
 316         unsigned int flags;
 317
 318         __asm __volatile
 319                 ("movl %%ebx, %%esi\n\t"
 320                  "cpuid\n\t"
 321                  "xchgl %%ebx, %%esi"
 322                  : "=a" (p[0]), "=S" (p[1]),
 323                  "=c" (p[2]), "=d" (p[3])
 324                  : "0" (ax));
 325
 326         return 0;
 327 #else
 328         return -1;
 329 #endif
 330 }
 331
 332 void cpu_detect_initialize()
 333 {
 334         unsigned int regs[4];
 335         unsigned int regs2[4];
 336
 337         int mib[2], ncpu;
 338         int len;
 339
 340         memset(&__cpu_detect_caps, 0, sizeof (struct cpu_detect_caps));
 341
 342         /* Check for arch type */
 343 #if defined(ARCH_MIPS)
 344         __cpu_detect_caps.type = CPU_DETECT_TYPE_MIPS;
 345 #elif defined(ARCH_ALPHA)
 346         __cpu_detect_caps.type = CPU_DETECT_TYPE_ALPHA;
 347 #elif defined(ARCH_SPARC)
 348         __cpu_detect_caps.type = CPU_DETECT_TYPE_SPARC;
 349 #elif defined(ARCH_X86)
 350         __cpu_detect_caps.type = CPU_DETECT_TYPE_X86;
 351 #elif defined(ARCH_POWERPC)
 352         __cpu_detect_caps.type = CPU_DETECT_TYPE_POWERPC;
 353 #else
 354         __cpu_detect_caps.type = CPU_DETECT_TYPE_OTHER;
 355 #endif
 356
 357         /* Count the number of CPUs in system */
 358 #if !defined(OS_WIN32) && !defined(OS_UNKNOWN) && defined(_SC_NPROCESSORS_ONLN)
 359         __cpu_detect_caps.nrcpu = sysconf(_SC_NPROCESSORS_ONLN);
 360         if (__cpu_detect_caps.nrcpu == -1)
 361                 __cpu_detect_caps.nrcpu = 1;
 362
 363 #elif defined(OS_NETBSD) || defined(OS_FREEBSD) || defined(OS_OPENBSD)
 364
 365         mib[0] = CTL_HW;
 366         mib[1] = HW_NCPU;
 367
 368         len = sizeof (ncpu);
 369         sysctl(mib, 2, &ncpu, &len, NULL, 0);
 370         __cpu_detect_caps.nrcpu = ncpu;
 371
 372 #else
 373         __cpu_detect_caps.nrcpu = 1;
 374 #endif
 375
 376 #if defined(ARCH_X86)
 377         /* No cpuid, old 486 or lower */
 378         if (has_cpuid() == 0)
 379                 return;
 380
 381         __cpu_detect_caps.cacheline = 32;
 382
 383         /* Get max cpuid level */
 384         cpuid(0x00000000, regs);
 385
 386         if (regs[0] >= 0x00000001) {
 387                 unsigned int cacheline;
 388
 389                 cpuid (0x00000001, regs2);
 390
 391                 __cpu_detect_caps.x86cpuType = (regs2[0] >> 8) & 0xf;
 392                 if (__cpu_detect_caps.x86cpuType == 0xf)
 393                     __cpu_detect_caps.x86cpuType = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
 394
 395                 /* general feature flags */
 396                 __cpu_detect_caps.hasTSC  = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
 397                 __cpu_detect_caps.hasMMX  = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
 398                 __cpu_detect_caps.hasSSE  = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
 399                 __cpu_detect_caps.hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
 400                 __cpu_detect_caps.hasSSE3 = (regs2[2] & (1));          /* 0x0000001 */
 401                 __cpu_detect_caps.hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
 402                 __cpu_detect_caps.hasMMX2 = __cpu_detect_caps.hasSSE; /* SSE cpus supports mmxext too */
 403
 404                 cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
 405                 if (cacheline > 0)
 406                         __cpu_detect_caps.cacheline = cacheline;
 407         }
 408
 409         cpuid(0x80000000, regs);
 410
 411         if (regs[0] >= 0x80000001) {
 412
 413                 cpuid(0x80000001, regs2);
 414
 415                 __cpu_detect_caps.hasMMX  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
 416                 __cpu_detect_caps.hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
 417                 __cpu_detect_caps.has3DNow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
 418                 __cpu_detect_caps.has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
 419         }
 420
 421         if (regs[0] >= 0x80000006) {
 422                 cpuid(0x80000006, regs2);
 423                 __cpu_detect_caps.cacheline = regs2[2] & 0xFF;
 424         }
 425
 426
 427 #if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_CYGWIN) || defined(OS_OPENBSD)
 428         if (__cpu_detect_caps.hasSSE)
 429                 check_os_katmai_support();
 430
 431         if (!__cpu_detect_caps.hasSSE) {
 432                 __cpu_detect_caps.hasSSE2 = 0;
 433                 __cpu_detect_caps.hasSSE3 = 0;
 434                 __cpu_detect_caps.hasSSSE3 = 0;
 435         }
 436 #else
 437         __cpu_detect_caps.hasSSE = 0;
 438         __cpu_detect_caps.hasSSE2 = 0;
 439         __cpu_detect_caps.hasSSE3 = 0;
 440         __cpu_detect_caps.hasSSSE3 = 0;
 441 #endif
 442 #endif /* ARCH_X86 */
 443
 444 #if defined(ARCH_POWERPC)
 445         check_os_altivec_support();
 446 #endif /* ARCH_POWERPC */
 447
 448         __cpu_detect_initialized = 1;
 449 }
 450
 451 struct cpu_detect_caps *cpu_detect_get_caps()
 452 {
 453         return &__cpu_detect_caps;
 454 }
 455
 456 /* The getters and setters for feature flags */
 457 int cpu_detect_get_tsc()
 458 {
 459         return __cpu_detect_caps.hasTSC;
 460 }
 461
 462 int cpu_detect_get_mmx()
 463 {
 464         return __cpu_detect_caps.hasMMX;
 465 }
 466
 467 int cpu_detect_get_mmx2()
 468 {
 469         return __cpu_detect_caps.hasMMX2;
 470 }
 471
 472 int cpu_detect_get_sse()
 473 {
 474         return __cpu_detect_caps.hasSSE;
 475 }
 476
 477 int cpu_detect_get_sse2()
 478 {
 479         return __cpu_detect_caps.hasSSE2;
 480 }
 481
 482 int cpu_detect_get_sse3()
 483 {
 484         return __cpu_detect_caps.hasSSE3;
 485 }
 486
 487 int cpu_detect_get_ssse3()
 488 {
 489         return __cpu_detect_caps.hasSSSE3;
 490 }
 491
 492 int cpu_detect_get_3dnow()
 493 {
 494         return __cpu_detect_caps.has3DNow;
 495 }
 496
 497 int cpu_detect_get_3dnow2()
 498 {
 499         return __cpu_detect_caps.has3DNowExt;
 500 }
 501
 502 int cpu_detect_get_altivec()
 503 {
 504         return __cpu_detect_caps.hasAltiVec;
 505 }
 506