sub/OSD: move some related files to sub/
[mplayer/greg.git] / cpudetect.c
blob343d2ac25b698663c611c49bc1046fe450842893
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 #if defined(__linux__) && defined(_POSIX_SOURCE) && !ARCH_X86_64
58 static void sigill_handler_sse( int signal, struct sigcontext sc )
60 mp_msg(MSGT_CPUDETECT,MSGL_V, "SIGILL, " );
62 /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
63 * instructions are 3 bytes long. We must increment the instruction
64 * pointer manually to avoid repeated execution of the offending
65 * instruction.
67 * If the SIGILL is caused by a divide-by-zero when unmasked
68 * exceptions aren't supported, the SIMD FPU status and control
69 * word will be restored at the end of the test, so we don't need
70 * to worry about doing it here. Besides, we may not be able to...
72 sc.eip += 3;
74 gCpuCaps.hasSSE=0;
76 #endif /* __linux__ && _POSIX_SOURCE */
78 #if (defined(__MINGW32__) || defined(__CYGWIN__)) && !ARCH_X86_64
79 LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
81 if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
82 mp_msg(MSGT_CPUDETECT,MSGL_V, "SIGILL, " );
83 ep->ContextRecord->Eip +=3;
84 gCpuCaps.hasSSE=0;
85 return EXCEPTION_CONTINUE_EXECUTION;
87 return EXCEPTION_CONTINUE_SEARCH;
89 #endif /* defined(__MINGW32__) || defined(__CYGWIN__) */
91 #ifdef __OS2__
92 ULONG _System os2_sig_handler_sse(PEXCEPTIONREPORTRECORD p1,
93 PEXCEPTIONREGISTRATIONRECORD p2,
94 PCONTEXTRECORD p3,
95 PVOID p4)
97 if(p1->ExceptionNum == XCPT_ILLEGAL_INSTRUCTION){
98 mp_msg(MSGT_CPUDETECT, MSGL_V, "SIGILL, ");
100 p3->ctx_RegEip += 3;
101 gCpuCaps.hasSSE = 0;
103 return XCPT_CONTINUE_EXECUTION;
105 return XCPT_CONTINUE_SEARCH;
107 #endif
109 /* If we're running on a processor that can do SSE, let's see if we
110 * are allowed to or not. This will catch 2.4.0 or later kernels that
111 * haven't been configured for a Pentium III but are running on one,
112 * and RedHat patched 2.2 kernels that have broken exception handling
113 * support for user space apps that do SSE.
116 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__)
117 #define SSE_SYSCTL_NAME "hw.instruction_sse"
118 #elif defined(__APPLE__)
119 #define SSE_SYSCTL_NAME "hw.optional.sse"
120 #endif
122 static void check_os_katmai_support( void )
124 #if ARCH_X86_64
125 gCpuCaps.hasSSE=1;
126 gCpuCaps.hasSSE2=1;
127 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__) || defined(__APPLE__)
128 int has_sse=0, ret;
129 size_t len=sizeof(has_sse);
131 ret = sysctlbyname(SSE_SYSCTL_NAME, &has_sse, &len, NULL, 0);
132 if (ret || !has_sse)
133 gCpuCaps.hasSSE=0;
135 #elif defined(__NetBSD__) || defined (__OpenBSD__)
136 #if __NetBSD_Version__ >= 105250000 || (defined __OpenBSD__)
137 int has_sse, has_sse2, ret, mib[2];
138 size_t varlen;
140 mib[0] = CTL_MACHDEP;
141 mib[1] = CPU_SSE;
142 varlen = sizeof(has_sse);
144 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
145 ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
146 gCpuCaps.hasSSE = ret >= 0 && has_sse;
147 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
149 mib[1] = CPU_SSE2;
150 varlen = sizeof(has_sse2);
151 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE2... " );
152 ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
153 gCpuCaps.hasSSE2 = ret >= 0 && has_sse2;
154 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE2 ? "yes.\n" : "no!\n" );
155 #else
156 gCpuCaps.hasSSE = 0;
157 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "No OS support for SSE, disabling to be safe.\n" );
158 #endif
159 #elif defined(__MINGW32__) || defined(__CYGWIN__)
160 LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
161 if ( gCpuCaps.hasSSE ) {
162 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
163 exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
164 __asm__ volatile ("xorps %xmm0, %xmm0");
165 SetUnhandledExceptionFilter(exc_fil);
166 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
168 #elif defined(__OS2__)
169 EXCEPTIONREGISTRATIONRECORD RegRec = { 0, &os2_sig_handler_sse };
170 if ( gCpuCaps.hasSSE ) {
171 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
172 DosSetExceptionHandler( &RegRec );
173 __asm__ volatile ("xorps %xmm0, %xmm0");
174 DosUnsetExceptionHandler( &RegRec );
175 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
177 #elif defined(__linux__)
178 #if defined(_POSIX_SOURCE)
179 struct sigaction saved_sigill;
181 /* Save the original signal handlers.
183 sigaction( SIGILL, NULL, &saved_sigill );
185 signal( SIGILL, (void (*)(int))sigill_handler_sse );
187 /* Emulate test for OSFXSR in CR4. The OS will set this bit if it
188 * supports the extended FPU save and restore required for SSE. If
189 * we execute an SSE instruction on a PIII and get a SIGILL, the OS
190 * doesn't support Streaming SIMD Exceptions, even if the processor
191 * does.
193 if ( gCpuCaps.hasSSE ) {
194 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
196 // __asm__ volatile ("xorps %%xmm0, %%xmm0");
197 __asm__ volatile ("xorps %xmm0, %xmm0");
199 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
202 /* Restore the original signal handlers.
204 sigaction( SIGILL, &saved_sigill, NULL );
206 /* If we've gotten to here and the XMM CPUID bit is still set, we're
207 * safe to go ahead and hook out the SSE code throughout Mesa.
209 mp_msg(MSGT_CPUDETECT,MSGL_V, "Tests of OS support for SSE %s\n", gCpuCaps.hasSSE ? "passed." : "failed!" );
210 #else
211 /* We can't use POSIX signal handling to test the availability of
212 * SSE, so we disable it by default.
214 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "Cannot test OS support for SSE, disabling to be safe.\n" );
215 gCpuCaps.hasSSE=0;
216 #endif /* _POSIX_SOURCE */
217 #else
218 /* Do nothing on other platforms for now.
220 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "Cannot test OS support for SSE, leaving disabled.\n" );
221 gCpuCaps.hasSSE=0;
222 #endif /* __linux__ */
226 // return TRUE if cpuid supported
227 static int has_cpuid(void)
229 // code from libavcodec:
230 #if ARCH_X86_64
231 return 1;
232 #else
233 long a, c;
234 __asm__ volatile (
235 /* See if CPUID instruction is supported ... */
236 /* ... Get copies of EFLAGS into eax and ecx */
237 "pushfl\n\t"
238 "pop %0\n\t"
239 "mov %0, %1\n\t"
241 /* ... Toggle the ID bit in one copy and store */
242 /* to the EFLAGS reg */
243 "xor $0x200000, %0\n\t"
244 "push %0\n\t"
245 "popfl\n\t"
247 /* ... Get the (hopefully modified) EFLAGS */
248 "pushfl\n\t"
249 "pop %0\n\t"
250 : "=a" (a), "=c" (c)
252 : "cc"
255 return a != c;
256 #endif
259 void
260 do_cpuid(unsigned int ax, unsigned int *p)
262 // code from libavcodec:
263 __asm__ volatile
264 ("mov %%"REG_b", %%"REG_S"\n\t"
265 "cpuid\n\t"
266 "xchg %%"REG_b", %%"REG_S
267 : "=a" (p[0]), "=S" (p[1]),
268 "=c" (p[2]), "=d" (p[3])
269 : "0" (ax));
272 void GetCpuCaps( CpuCaps *caps)
274 unsigned int regs[4];
275 unsigned int regs2[4];
277 memset(caps, 0, sizeof(*caps));
278 caps->isX86=1;
279 caps->cl_size=32; /* default */
280 if (!has_cpuid()) {
281 mp_msg(MSGT_CPUDETECT,MSGL_WARN,"CPUID not supported!??? (maybe an old 486?)\n");
282 return;
284 do_cpuid(0x00000000, regs); // get _max_ cpuid level and vendor name
285 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU vendor name: %.4s%.4s%.4s max cpuid level: %d\n",
286 (char*) (regs+1),(char*) (regs+3),(char*) (regs+2), regs[0]);
287 if (regs[0]>=0x00000001)
289 char *tmpstr, *ptmpstr;
290 unsigned cl_size;
292 do_cpuid(0x00000001, regs2);
294 caps->cpuType=(regs2[0] >> 8)&0xf;
295 caps->cpuModel=(regs2[0] >> 4)&0xf;
297 // see AMD64 Architecture Programmer's Manual, Volume 3: General-purpose and
298 // System Instructions, Table 3-2: Effective family computation, page 120.
299 if(caps->cpuType==0xf){
300 // use extended family (P4, IA64, K8)
301 caps->cpuType=0xf+((regs2[0]>>20)&255);
303 if(caps->cpuType==0xf || caps->cpuType==6)
304 caps->cpuModel |= ((regs2[0]>>16)&0xf) << 4;
306 caps->cpuStepping=regs2[0] & 0xf;
308 // general feature flags:
309 caps->hasTSC = (regs2[3] & (1 << 8 )) >> 8; // 0x0000010
310 caps->hasMMX = (regs2[3] & (1 << 23 )) >> 23; // 0x0800000
311 caps->hasSSE = (regs2[3] & (1 << 25 )) >> 25; // 0x2000000
312 caps->hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; // 0x4000000
313 caps->hasSSE3 = (regs2[2] & 1); // 0x0000001
314 caps->hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9; // 0x0000200
315 caps->hasMMX2 = caps->hasSSE; // SSE cpus supports mmxext too
316 cl_size = ((regs2[1] >> 8) & 0xFF)*8;
317 if(cl_size) caps->cl_size = cl_size;
319 ptmpstr=tmpstr=GetCpuFriendlyName(regs, regs2);
320 while(*ptmpstr == ' ') // strip leading spaces
321 ptmpstr++;
322 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: %s ", ptmpstr);
323 free(tmpstr);
324 mp_msg(MSGT_CPUDETECT,MSGL_V,"(Family: %d, Model: %d, Stepping: %d)\n",
325 caps->cpuType, caps->cpuModel, caps->cpuStepping);
328 do_cpuid(0x80000000, regs);
329 if (regs[0]>=0x80000001) {
330 mp_msg(MSGT_CPUDETECT,MSGL_V,"extended cpuid-level: %d\n",regs[0]&0x7FFFFFFF);
331 do_cpuid(0x80000001, regs2);
332 caps->hasMMX |= (regs2[3] & (1 << 23 )) >> 23; // 0x0800000
333 caps->hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; // 0x400000
334 caps->has3DNow = (regs2[3] & (1 << 31 )) >> 31; //0x80000000
335 caps->has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
336 caps->hasSSE4a = (regs2[2] & (1 << 6 )) >> 6; // 0x0000040
338 if(regs[0]>=0x80000006)
340 do_cpuid(0x80000006, regs2);
341 mp_msg(MSGT_CPUDETECT,MSGL_V,"extended cache-info: %d\n",regs2[2]&0x7FFFFFFF);
342 caps->cl_size = regs2[2] & 0xFF;
344 mp_msg(MSGT_CPUDETECT,MSGL_V,"Detected cache-line size is %u bytes\n",caps->cl_size);
345 #if 0
346 mp_msg(MSGT_CPUDETECT,MSGL_INFO,"cpudetect: MMX=%d MMX2=%d SSE=%d SSE2=%d 3DNow=%d 3DNowExt=%d\n",
347 gCpuCaps.hasMMX,
348 gCpuCaps.hasMMX2,
349 gCpuCaps.hasSSE,
350 gCpuCaps.hasSSE2,
351 gCpuCaps.has3DNow,
352 gCpuCaps.has3DNowExt);
353 #endif
355 /* FIXME: Does SSE2 need more OS support, too? */
356 if (caps->hasSSE)
357 check_os_katmai_support();
358 if (!caps->hasSSE)
359 caps->hasSSE2 = 0;
360 // caps->has3DNow=1;
361 // caps->hasMMX2 = 0;
362 // caps->hasMMX = 0;
364 #if !CONFIG_RUNTIME_CPUDETECT
365 #if !HAVE_MMX
366 if(caps->hasMMX) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"MMX supported but disabled\n");
367 caps->hasMMX=0;
368 #endif
369 #if !HAVE_MMX2
370 if(caps->hasMMX2) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"MMX2 supported but disabled\n");
371 caps->hasMMX2=0;
372 #endif
373 #if !HAVE_SSE
374 if(caps->hasSSE) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"SSE supported but disabled\n");
375 caps->hasSSE=0;
376 #endif
377 #if !HAVE_SSE2
378 if(caps->hasSSE2) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"SSE2 supported but disabled\n");
379 caps->hasSSE2=0;
380 #endif
381 #if !HAVE_AMD3DNOW
382 if(caps->has3DNow) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"3DNow supported but disabled\n");
383 caps->has3DNow=0;
384 #endif
385 #if !HAVE_AMD3DNOWEXT
386 if(caps->has3DNowExt) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"3DNowExt supported but disabled\n");
387 caps->has3DNowExt=0;
388 #endif
389 #endif // CONFIG_RUNTIME_CPUDETECT
392 char *GetCpuFriendlyName(unsigned int regs[], unsigned int regs2[]){
393 char vendor[13];
394 char *retname;
395 int i;
397 if (NULL==(retname=malloc(256))) {
398 mp_msg(MSGT_CPUDETECT,MSGL_FATAL,"Error: GetCpuFriendlyName() not enough memory\n");
399 exit(1);
401 retname[0] = '\0';
403 sprintf(vendor,"%.4s%.4s%.4s",(char*)(regs+1),(char*)(regs+3),(char*)(regs+2));
405 do_cpuid(0x80000000,regs);
406 if (regs[0] >= 0x80000004)
408 // CPU has built-in namestring
409 for (i = 0x80000002; i <= 0x80000004; i++)
411 do_cpuid(i, regs);
412 strncat(retname, (char*)regs, 16);
415 return retname;
418 #else /* ARCH_X86 */
420 #ifdef __APPLE__
421 #include <sys/sysctl.h>
422 #elif defined(__AMIGAOS4__)
423 /* nothing */
424 #else
425 #include <signal.h>
426 #include <setjmp.h>
428 static sigjmp_buf jmpbuf;
429 static volatile sig_atomic_t canjump = 0;
431 static void sigill_handler (int sig)
433 if (!canjump) {
434 signal (sig, SIG_DFL);
435 raise (sig);
438 canjump = 0;
439 siglongjmp (jmpbuf, 1);
441 #endif /* __APPLE__ */
443 void GetCpuCaps( CpuCaps *caps)
445 caps->cpuType=0;
446 caps->cpuModel=0;
447 caps->cpuStepping=0;
448 caps->hasMMX=0;
449 caps->hasMMX2=0;
450 caps->has3DNow=0;
451 caps->has3DNowExt=0;
452 caps->hasSSE=0;
453 caps->hasSSE2=0;
454 caps->hasSSE3=0;
455 caps->hasSSSE3=0;
456 caps->hasSSE4a=0;
457 caps->isX86=0;
458 caps->hasAltiVec = 0;
459 #if HAVE_ALTIVEC
460 #ifdef __APPLE__
462 rip-off from ffmpeg altivec detection code.
463 this code also appears on Apple's AltiVec pages.
466 int sels[2] = {CTL_HW, HW_VECTORUNIT};
467 int has_vu = 0;
468 size_t len = sizeof(has_vu);
469 int err;
471 err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
473 if (err == 0)
474 if (has_vu != 0)
475 caps->hasAltiVec = 1;
477 #elif defined(__AMIGAOS4__)
478 ULONG result = 0;
480 GetCPUInfoTags(GCIT_VectorUnit, &result, TAG_DONE);
481 if (result == VECTORTYPE_ALTIVEC)
482 caps->hasAltiVec = 1;
483 #else
484 /* no Darwin, do it the brute-force way */
485 /* this is borrowed from the libmpeg2 library */
487 signal (SIGILL, sigill_handler);
488 if (sigsetjmp (jmpbuf, 1)) {
489 signal (SIGILL, SIG_DFL);
490 } else {
491 canjump = 1;
493 __asm__ volatile ("mtspr 256, %0\n\t"
494 "vand %%v0, %%v0, %%v0"
496 : "r" (-1));
498 signal (SIGILL, SIG_DFL);
499 caps->hasAltiVec = 1;
502 #endif /* __APPLE__ */
503 mp_msg(MSGT_CPUDETECT,MSGL_V,"AltiVec %sfound\n", (caps->hasAltiVec ? "" : "not "));
504 #endif /* HAVE_ALTIVEC */
506 if (ARCH_IA64)
507 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Intel Itanium\n");
509 if (ARCH_SPARC)
510 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Sun Sparc\n");
512 if (ARCH_ARM)
513 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: ARM\n");
515 if (ARCH_PPC)
516 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: PowerPC\n");
518 if (ARCH_ALPHA)
519 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Digital Alpha\n");
521 if (ARCH_MIPS)
522 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: MIPS\n");
524 if (ARCH_PA_RISC)
525 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: Hewlett-Packard PA-RISC\n");
527 if (ARCH_S390)
528 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: IBM S/390\n");
530 if (ARCH_S390X)
531 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: IBM S/390X\n");
533 if (ARCH_VAX)
534 mp_msg(MSGT_CPUDETECT,MSGL_V, "CPU: Digital VAX\n" );
536 if (ARCH_XTENSA)
537 mp_msg(MSGT_CPUDETECT,MSGL_V, "CPU: Tensilica Xtensa\n" );
539 #endif /* !ARCH_X86 */