Fix vf_tcdump's compilation
[mplayer/kovensky.git] / cpudetect.c
blobe0f99f25367d02c33a87fdf0f4ede7abebae8666
1 /*
2 * This file is part of MPlayer.
4 * MPlayer is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * MPlayer is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License along
15 * with MPlayer; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 #include "config.h"
20 #include "cpudetect.h"
21 #include "mp_msg.h"
23 CpuCaps gCpuCaps;
25 #include <stdlib.h>
27 #if ARCH_X86
29 #include <stdio.h>
30 #include <string.h>
32 #if defined (__NetBSD__) || defined(__OpenBSD__)
33 #include <sys/param.h>
34 #include <sys/sysctl.h>
35 #include <machine/cpu.h>
36 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__) || defined(__APPLE__)
37 #include <sys/types.h>
38 #include <sys/sysctl.h>
39 #elif defined(__linux__)
40 #include <signal.h>
41 #elif defined(__MINGW32__) || defined(__CYGWIN__)
42 #include <windows.h>
43 #elif defined(__OS2__)
44 #define INCL_DOS
45 #include <os2.h>
46 #elif defined(__AMIGAOS4__)
47 #include <proto/exec.h>
48 #endif
50 /* Thanks to the FreeBSD project for some of this cpuid code, and
51 * help understanding how to use it. Thanks to the Mesa
52 * team for SSE support detection and more cpu detect code.
55 /* I believe this code works. However, it has only been used on a PII and PIII */
57 static void check_os_katmai_support( void );
59 // return TRUE if cpuid supported
60 static int has_cpuid(void)
62 // code from libavcodec:
63 #if ARCH_X86_64
64 return 1;
65 #else
66 long a, c;
67 __asm__ volatile (
68 /* See if CPUID instruction is supported ... */
69 /* ... Get copies of EFLAGS into eax and ecx */
70 "pushfl\n\t"
71 "pop %0\n\t"
72 "mov %0, %1\n\t"
74 /* ... Toggle the ID bit in one copy and store */
75 /* to the EFLAGS reg */
76 "xor $0x200000, %0\n\t"
77 "push %0\n\t"
78 "popfl\n\t"
80 /* ... Get the (hopefully modified) EFLAGS */
81 "pushfl\n\t"
82 "pop %0\n\t"
83 : "=a" (a), "=c" (c)
85 : "cc"
88 return a != c;
89 #endif
92 static void
93 do_cpuid(unsigned int ax, unsigned int *p)
95 #if 0
96 __asm__ volatile(
97 "cpuid;"
98 : "=a" (p[0]), "=b" (p[1]), "=c" (p[2]), "=d" (p[3])
99 : "0" (ax)
101 #else
102 // code from libavcodec:
103 __asm__ volatile
104 ("mov %%"REG_b", %%"REG_S"\n\t"
105 "cpuid\n\t"
106 "xchg %%"REG_b", %%"REG_S
107 : "=a" (p[0]), "=S" (p[1]),
108 "=c" (p[2]), "=d" (p[3])
109 : "0" (ax));
110 #endif
114 void GetCpuCaps( CpuCaps *caps)
116 unsigned int regs[4];
117 unsigned int regs2[4];
119 memset(caps, 0, sizeof(*caps));
120 caps->isX86=1;
121 caps->cl_size=32; /* default */
122 if (!has_cpuid()) {
123 mp_msg(MSGT_CPUDETECT,MSGL_WARN,"CPUID not supported!??? (maybe an old 486?)\n");
124 return;
126 do_cpuid(0x00000000, regs); // get _max_ cpuid level and vendor name
127 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU vendor name: %.4s%.4s%.4s max cpuid level: %d\n",
128 (char*) (regs+1),(char*) (regs+3),(char*) (regs+2), regs[0]);
129 if (regs[0]>=0x00000001)
131 char *tmpstr, *ptmpstr;
132 unsigned cl_size;
134 do_cpuid(0x00000001, regs2);
136 caps->cpuType=(regs2[0] >> 8)&0xf;
137 caps->cpuModel=(regs2[0] >> 4)&0xf;
139 // see AMD64 Architecture Programmer's Manual, Volume 3: General-purpose and
140 // System Instructions, Table 3-2: Effective family computation, page 120.
141 if(caps->cpuType==0xf){
142 // use extended family (P4, IA64, K8)
143 caps->cpuType=0xf+((regs2[0]>>20)&255);
145 if(caps->cpuType==0xf || caps->cpuType==6)
146 caps->cpuModel |= ((regs2[0]>>16)&0xf) << 4;
148 caps->cpuStepping=regs2[0] & 0xf;
150 // general feature flags:
151 caps->hasTSC = (regs2[3] & (1 << 8 )) >> 8; // 0x0000010
152 caps->hasMMX = (regs2[3] & (1 << 23 )) >> 23; // 0x0800000
153 caps->hasSSE = (regs2[3] & (1 << 25 )) >> 25; // 0x2000000
154 caps->hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; // 0x4000000
155 caps->hasSSE3 = (regs2[2] & 1); // 0x0000001
156 caps->hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9; // 0x0000200
157 caps->hasMMX2 = caps->hasSSE; // SSE cpus supports mmxext too
158 cl_size = ((regs2[1] >> 8) & 0xFF)*8;
159 if(cl_size) caps->cl_size = cl_size;
161 ptmpstr=tmpstr=GetCpuFriendlyName(regs, regs2);
162 while(*ptmpstr == ' ') // strip leading spaces
163 ptmpstr++;
164 mp_msg(MSGT_CPUDETECT,MSGL_INFO,"CPU: %s ", ptmpstr);
165 free(tmpstr);
166 mp_msg(MSGT_CPUDETECT,MSGL_INFO,"(Family: %d, Model: %d, Stepping: %d)\n",
167 caps->cpuType, caps->cpuModel, caps->cpuStepping);
170 do_cpuid(0x80000000, regs);
171 if (regs[0]>=0x80000001) {
172 mp_msg(MSGT_CPUDETECT,MSGL_V,"extended cpuid-level: %d\n",regs[0]&0x7FFFFFFF);
173 do_cpuid(0x80000001, regs2);
174 caps->hasMMX |= (regs2[3] & (1 << 23 )) >> 23; // 0x0800000
175 caps->hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; // 0x400000
176 caps->has3DNow = (regs2[3] & (1 << 31 )) >> 31; //0x80000000
177 caps->has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
178 caps->hasSSE4a = (regs2[2] & (1 << 6 )) >> 6; // 0x0000040
180 if(regs[0]>=0x80000006)
182 do_cpuid(0x80000006, regs2);
183 mp_msg(MSGT_CPUDETECT,MSGL_V,"extended cache-info: %d\n",regs2[2]&0x7FFFFFFF);
184 caps->cl_size = regs2[2] & 0xFF;
186 mp_msg(MSGT_CPUDETECT,MSGL_V,"Detected cache-line size is %u bytes\n",caps->cl_size);
187 #if 0
188 mp_msg(MSGT_CPUDETECT,MSGL_INFO,"cpudetect: MMX=%d MMX2=%d SSE=%d SSE2=%d 3DNow=%d 3DNowExt=%d\n",
189 gCpuCaps.hasMMX,
190 gCpuCaps.hasMMX2,
191 gCpuCaps.hasSSE,
192 gCpuCaps.hasSSE2,
193 gCpuCaps.has3DNow,
194 gCpuCaps.has3DNowExt );
195 #endif
197 /* FIXME: Does SSE2 need more OS support, too? */
198 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__) \
199 || defined(__NetBSD__) || defined(__OpenBSD__) || defined(__DragonFly__) \
200 || defined(__APPLE__) || defined(__CYGWIN__) || defined(__MINGW32__) \
201 || defined(__OS2__)
202 if (caps->hasSSE)
203 check_os_katmai_support();
204 if (!caps->hasSSE)
205 caps->hasSSE2 = 0;
206 #else
207 caps->hasSSE=0;
208 caps->hasSSE2 = 0;
209 #endif
210 // caps->has3DNow=1;
211 // caps->hasMMX2 = 0;
212 // caps->hasMMX = 0;
214 #if !CONFIG_RUNTIME_CPUDETECT
215 #if !HAVE_MMX
216 if(caps->hasMMX) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"MMX supported but disabled\n");
217 caps->hasMMX=0;
218 #endif
219 #if !HAVE_MMX2
220 if(caps->hasMMX2) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"MMX2 supported but disabled\n");
221 caps->hasMMX2=0;
222 #endif
223 #if !HAVE_SSE
224 if(caps->hasSSE) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"SSE supported but disabled\n");
225 caps->hasSSE=0;
226 #endif
227 #if !HAVE_SSE2
228 if(caps->hasSSE2) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"SSE2 supported but disabled\n");
229 caps->hasSSE2=0;
230 #endif
231 #if !HAVE_AMD3DNOW
232 if(caps->has3DNow) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"3DNow supported but disabled\n");
233 caps->has3DNow=0;
234 #endif
235 #if !HAVE_AMD3DNOWEXT
236 if(caps->has3DNowExt) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"3DNowExt supported but disabled\n");
237 caps->has3DNowExt=0;
238 #endif
239 #endif // CONFIG_RUNTIME_CPUDETECT
242 char *GetCpuFriendlyName(unsigned int regs[], unsigned int regs2[]){
243 char vendor[13];
244 char *retname;
245 int i;
247 if (NULL==(retname=malloc(256))) {
248 mp_msg(MSGT_CPUDETECT,MSGL_FATAL,"Error: GetCpuFriendlyName() not enough memory\n");
249 exit(1);
251 retname[0] = '\0';
253 sprintf(vendor,"%.4s%.4s%.4s",(char*)(regs+1),(char*)(regs+3),(char*)(regs+2));
255 do_cpuid(0x80000000,regs);
256 if (regs[0] >= 0x80000004)
258 // CPU has built-in namestring
259 for (i = 0x80000002; i <= 0x80000004; i++)
261 do_cpuid(i, regs);
262 strncat(retname, (char*)regs, 16);
265 return retname;
268 #if defined(__linux__) && defined(_POSIX_SOURCE) && !ARCH_X86_64
269 static void sigill_handler_sse( int signal, struct sigcontext sc )
271 mp_msg(MSGT_CPUDETECT,MSGL_V, "SIGILL, " );
273 /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
274 * instructions are 3 bytes long. We must increment the instruction
275 * pointer manually to avoid repeated execution of the offending
276 * instruction.
278 * If the SIGILL is caused by a divide-by-zero when unmasked
279 * exceptions aren't supported, the SIMD FPU status and control
280 * word will be restored at the end of the test, so we don't need
281 * to worry about doing it here. Besides, we may not be able to...
283 sc.eip += 3;
285 gCpuCaps.hasSSE=0;
287 #endif /* __linux__ && _POSIX_SOURCE */
289 #if (defined(__MINGW32__) || defined(__CYGWIN__)) && !ARCH_X86_64
290 LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
292 if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
293 mp_msg(MSGT_CPUDETECT,MSGL_V, "SIGILL, " );
294 ep->ContextRecord->Eip +=3;
295 gCpuCaps.hasSSE=0;
296 return EXCEPTION_CONTINUE_EXECUTION;
298 return EXCEPTION_CONTINUE_SEARCH;
300 #endif /* defined(__MINGW32__) || defined(__CYGWIN__) */
302 #ifdef __OS2__
303 ULONG _System os2_sig_handler_sse( PEXCEPTIONREPORTRECORD p1,
304 PEXCEPTIONREGISTRATIONRECORD p2,
305 PCONTEXTRECORD p3,
306 PVOID p4 )
308 if(p1->ExceptionNum == XCPT_ILLEGAL_INSTRUCTION){
309 mp_msg(MSGT_CPUDETECT, MSGL_V, "SIGILL, ");
311 p3->ctx_RegEip += 3;
312 gCpuCaps.hasSSE = 0;
314 return XCPT_CONTINUE_EXECUTION;
316 return XCPT_CONTINUE_SEARCH;
318 #endif
320 /* If we're running on a processor that can do SSE, let's see if we
321 * are allowed to or not. This will catch 2.4.0 or later kernels that
322 * haven't been configured for a Pentium III but are running on one,
323 * and RedHat patched 2.2 kernels that have broken exception handling
324 * support for user space apps that do SSE.
327 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__)
328 #define SSE_SYSCTL_NAME "hw.instruction_sse"
329 #elif defined(__APPLE__)
330 #define SSE_SYSCTL_NAME "hw.optional.sse"
331 #endif
333 static void check_os_katmai_support( void )
335 #if ARCH_X86_64
336 gCpuCaps.hasSSE=1;
337 gCpuCaps.hasSSE2=1;
338 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__) || defined(__APPLE__)
339 int has_sse=0, ret;
340 size_t len=sizeof(has_sse);
342 ret = sysctlbyname(SSE_SYSCTL_NAME, &has_sse, &len, NULL, 0);
343 if (ret || !has_sse)
344 gCpuCaps.hasSSE=0;
346 #elif defined(__NetBSD__) || defined (__OpenBSD__)
347 #if __NetBSD_Version__ >= 105250000 || (defined __OpenBSD__)
348 int has_sse, has_sse2, ret, mib[2];
349 size_t varlen;
351 mib[0] = CTL_MACHDEP;
352 mib[1] = CPU_SSE;
353 varlen = sizeof(has_sse);
355 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
356 ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
357 gCpuCaps.hasSSE = ret >= 0 && has_sse;
358 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
360 mib[1] = CPU_SSE2;
361 varlen = sizeof(has_sse2);
362 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE2... " );
363 ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
364 gCpuCaps.hasSSE2 = ret >= 0 && has_sse2;
365 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE2 ? "yes.\n" : "no!\n" );
366 #else
367 gCpuCaps.hasSSE = 0;
368 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "No OS support for SSE, disabling to be safe.\n" );
369 #endif
370 #elif defined(__MINGW32__) || defined(__CYGWIN__)
371 LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
372 if ( gCpuCaps.hasSSE ) {
373 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
374 exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
375 __asm__ volatile ("xorps %xmm0, %xmm0");
376 SetUnhandledExceptionFilter(exc_fil);
377 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
379 #elif defined(__OS2__)
380 EXCEPTIONREGISTRATIONRECORD RegRec = { 0, &os2_sig_handler_sse };
381 if ( gCpuCaps.hasSSE ) {
382 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
383 DosSetExceptionHandler( &RegRec );
384 __asm__ volatile ("xorps %xmm0, %xmm0");
385 DosUnsetExceptionHandler( &RegRec );
386 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
388 #elif defined(__linux__)
389 #if defined(_POSIX_SOURCE)
390 struct sigaction saved_sigill;
392 /* Save the original signal handlers.
394 sigaction( SIGILL, NULL, &saved_sigill );
396 signal( SIGILL, (void (*)(int))sigill_handler_sse );
398 /* Emulate test for OSFXSR in CR4. The OS will set this bit if it
399 * supports the extended FPU save and restore required for SSE. If
400 * we execute an SSE instruction on a PIII and get a SIGILL, the OS
401 * doesn't support Streaming SIMD Exceptions, even if the processor
402 * does.
404 if ( gCpuCaps.hasSSE ) {
405 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
407 // __asm__ volatile ("xorps %%xmm0, %%xmm0");
408 __asm__ volatile ("xorps %xmm0, %xmm0");
410 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
413 /* Restore the original signal handlers.
415 sigaction( SIGILL, &saved_sigill, NULL );
417 /* If we've gotten to here and the XMM CPUID bit is still set, we're
418 * safe to go ahead and hook out the SSE code throughout Mesa.
420 mp_msg(MSGT_CPUDETECT,MSGL_V, "Tests of OS support for SSE %s\n", gCpuCaps.hasSSE ? "passed." : "failed!" );
421 #else
422 /* We can't use POSIX signal handling to test the availability of
423 * SSE, so we disable it by default.
425 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "Cannot test OS support for SSE, disabling to be safe.\n" );
426 gCpuCaps.hasSSE=0;
427 #endif /* _POSIX_SOURCE */
428 #else
429 /* Do nothing on other platforms for now.
431 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "Cannot test OS support for SSE, leaving disabled.\n" );
432 gCpuCaps.hasSSE=0;
433 #endif /* __linux__ */
435 #else /* ARCH_X86 */
437 #ifdef __APPLE__
438 #include <sys/sysctl.h>
439 #elif defined(__AMIGAOS4__)
440 /* nothing */
441 #else
442 #include <signal.h>
443 #include <setjmp.h>
445 static sigjmp_buf jmpbuf;
446 static volatile sig_atomic_t canjump = 0;
448 static void sigill_handler (int sig)
450 if (!canjump) {
451 signal (sig, SIG_DFL);
452 raise (sig);
455 canjump = 0;
456 siglongjmp (jmpbuf, 1);
458 #endif /* __APPLE__ */
460 void GetCpuCaps( CpuCaps *caps)
462 caps->cpuType=0;
463 caps->cpuModel=0;
464 caps->cpuStepping=0;
465 caps->hasMMX=0;
466 caps->hasMMX2=0;
467 caps->has3DNow=0;
468 caps->has3DNowExt=0;
469 caps->hasSSE=0;
470 caps->hasSSE2=0;
471 caps->hasSSE3=0;
472 caps->hasSSSE3=0;
473 caps->hasSSE4a=0;
474 caps->isX86=0;
475 caps->hasAltiVec = 0;
476 #if HAVE_ALTIVEC
477 #ifdef __APPLE__
479 rip-off from ffmpeg altivec detection code.
480 this code also appears on Apple's AltiVec pages.
483 int sels[2] = {CTL_HW, HW_VECTORUNIT};
484 int has_vu = 0;
485 size_t len = sizeof(has_vu);
486 int err;
488 err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
490 if (err == 0)
491 if (has_vu != 0)
492 caps->hasAltiVec = 1;
494 #elif defined(__AMIGAOS4__)
495 ULONG result = 0;
497 GetCPUInfoTags(GCIT_VectorUnit, &result, TAG_DONE);
498 if (result == VECTORTYPE_ALTIVEC)
499 caps->hasAltiVec = 1;
500 #else
501 /* no Darwin, do it the brute-force way */
502 /* this is borrowed from the libmpeg2 library */
504 signal (SIGILL, sigill_handler);
505 if (sigsetjmp (jmpbuf, 1)) {
506 signal (SIGILL, SIG_DFL);
507 } else {
508 canjump = 1;
510 __asm__ volatile ("mtspr 256, %0\n\t"
511 "vand %%v0, %%v0, %%v0"
513 : "r" (-1));
515 signal (SIGILL, SIG_DFL);
516 caps->hasAltiVec = 1;
519 #endif /* __APPLE__ */
520 mp_msg(MSGT_CPUDETECT,MSGL_V,"AltiVec %sfound\n", (caps->hasAltiVec ? "" : "not "));
521 #endif /* HAVE_ALTIVEC */
523 if (ARCH_IA64)
524 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Intel Itanium\n");
526 if (ARCH_SPARC)
527 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Sun Sparc\n");
529 if (ARCH_ARM)
530 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: ARM\n");
532 if (ARCH_PPC)
533 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: PowerPC\n");
535 if (ARCH_ALPHA)
536 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Digital Alpha\n");
538 if (ARCH_SGI_MIPS)
539 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: SGI MIPS\n");
541 if (ARCH_PA_RISC)
542 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Hewlett-Packard PA-RISC\n");
544 if (ARCH_S390)
545 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: IBM S/390\n");
547 if (ARCH_S390X)
548 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: IBM S/390X\n");
550 if (ARCH_VAX)
551 mp_msg(MSGT_CPUDETECT,MSGL_V, "CPU: Digital VAX\n" );
553 if (ARCH_XTENSA)
554 mp_msg(MSGT_CPUDETECT,MSGL_V, "CPU: Tensilica Xtensa\n" );
556 #endif /* !ARCH_X86 */