libfaad2: change lrintf availability logic
[mplayer.git] / cpudetect.c
blobd394bdcdab6deb44a5099fb984251b67ad85034b
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 void
93 do_cpuid(unsigned int ax, unsigned int *p)
95 // code from libavcodec:
96 __asm__ volatile
97 ("mov %%"REG_b", %%"REG_S"\n\t"
98 "cpuid\n\t"
99 "xchg %%"REG_b", %%"REG_S
100 : "=a" (p[0]), "=S" (p[1]),
101 "=c" (p[2]), "=d" (p[3])
102 : "0" (ax));
105 void GetCpuCaps( CpuCaps *caps)
107 unsigned int regs[4];
108 unsigned int regs2[4];
110 memset(caps, 0, sizeof(*caps));
111 caps->isX86=1;
112 caps->cl_size=32; /* default */
113 if (!has_cpuid()) {
114 mp_msg(MSGT_CPUDETECT,MSGL_WARN,"CPUID not supported!??? (maybe an old 486?)\n");
115 return;
117 do_cpuid(0x00000000, regs); // get _max_ cpuid level and vendor name
118 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU vendor name: %.4s%.4s%.4s max cpuid level: %d\n",
119 (char*) (regs+1),(char*) (regs+3),(char*) (regs+2), regs[0]);
120 if (regs[0]>=0x00000001)
122 char *tmpstr, *ptmpstr;
123 unsigned cl_size;
125 do_cpuid(0x00000001, regs2);
127 caps->cpuType=(regs2[0] >> 8)&0xf;
128 caps->cpuModel=(regs2[0] >> 4)&0xf;
130 // see AMD64 Architecture Programmer's Manual, Volume 3: General-purpose and
131 // System Instructions, Table 3-2: Effective family computation, page 120.
132 if(caps->cpuType==0xf){
133 // use extended family (P4, IA64, K8)
134 caps->cpuType=0xf+((regs2[0]>>20)&255);
136 if(caps->cpuType==0xf || caps->cpuType==6)
137 caps->cpuModel |= ((regs2[0]>>16)&0xf) << 4;
139 caps->cpuStepping=regs2[0] & 0xf;
141 // general feature flags:
142 caps->hasTSC = (regs2[3] & (1 << 8 )) >> 8; // 0x0000010
143 caps->hasMMX = (regs2[3] & (1 << 23 )) >> 23; // 0x0800000
144 caps->hasSSE = (regs2[3] & (1 << 25 )) >> 25; // 0x2000000
145 caps->hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; // 0x4000000
146 caps->hasSSE3 = (regs2[2] & 1); // 0x0000001
147 caps->hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9; // 0x0000200
148 caps->hasMMX2 = caps->hasSSE; // SSE cpus supports mmxext too
149 cl_size = ((regs2[1] >> 8) & 0xFF)*8;
150 if(cl_size) caps->cl_size = cl_size;
152 ptmpstr=tmpstr=GetCpuFriendlyName(regs, regs2);
153 while(*ptmpstr == ' ') // strip leading spaces
154 ptmpstr++;
155 mp_msg(MSGT_CPUDETECT,MSGL_V,"CPU: %s ", ptmpstr);
156 free(tmpstr);
157 mp_msg(MSGT_CPUDETECT,MSGL_V,"(Family: %d, Model: %d, Stepping: %d)\n",
158 caps->cpuType, caps->cpuModel, caps->cpuStepping);
161 do_cpuid(0x80000000, regs);
162 if (regs[0]>=0x80000001) {
163 mp_msg(MSGT_CPUDETECT,MSGL_V,"extended cpuid-level: %d\n",regs[0]&0x7FFFFFFF);
164 do_cpuid(0x80000001, regs2);
165 caps->hasMMX |= (regs2[3] & (1 << 23 )) >> 23; // 0x0800000
166 caps->hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; // 0x400000
167 caps->has3DNow = (regs2[3] & (1 << 31 )) >> 31; //0x80000000
168 caps->has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
169 caps->hasSSE4a = (regs2[2] & (1 << 6 )) >> 6; // 0x0000040
171 if(regs[0]>=0x80000006)
173 do_cpuid(0x80000006, regs2);
174 mp_msg(MSGT_CPUDETECT,MSGL_V,"extended cache-info: %d\n",regs2[2]&0x7FFFFFFF);
175 caps->cl_size = regs2[2] & 0xFF;
177 mp_msg(MSGT_CPUDETECT,MSGL_V,"Detected cache-line size is %u bytes\n",caps->cl_size);
178 #if 0
179 mp_msg(MSGT_CPUDETECT,MSGL_INFO,"cpudetect: MMX=%d MMX2=%d SSE=%d SSE2=%d 3DNow=%d 3DNowExt=%d\n",
180 gCpuCaps.hasMMX,
181 gCpuCaps.hasMMX2,
182 gCpuCaps.hasSSE,
183 gCpuCaps.hasSSE2,
184 gCpuCaps.has3DNow,
185 gCpuCaps.has3DNowExt);
186 #endif
188 /* FIXME: Does SSE2 need more OS support, too? */
189 if (caps->hasSSE)
190 check_os_katmai_support();
191 if (!caps->hasSSE)
192 caps->hasSSE2 = 0;
193 // caps->has3DNow=1;
194 // caps->hasMMX2 = 0;
195 // caps->hasMMX = 0;
197 #if !CONFIG_RUNTIME_CPUDETECT
198 #if !HAVE_MMX
199 if(caps->hasMMX) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"MMX supported but disabled\n");
200 caps->hasMMX=0;
201 #endif
202 #if !HAVE_MMX2
203 if(caps->hasMMX2) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"MMX2 supported but disabled\n");
204 caps->hasMMX2=0;
205 #endif
206 #if !HAVE_SSE
207 if(caps->hasSSE) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"SSE supported but disabled\n");
208 caps->hasSSE=0;
209 #endif
210 #if !HAVE_SSE2
211 if(caps->hasSSE2) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"SSE2 supported but disabled\n");
212 caps->hasSSE2=0;
213 #endif
214 #if !HAVE_AMD3DNOW
215 if(caps->has3DNow) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"3DNow supported but disabled\n");
216 caps->has3DNow=0;
217 #endif
218 #if !HAVE_AMD3DNOWEXT
219 if(caps->has3DNowExt) mp_msg(MSGT_CPUDETECT,MSGL_WARN,"3DNowExt supported but disabled\n");
220 caps->has3DNowExt=0;
221 #endif
222 #endif // CONFIG_RUNTIME_CPUDETECT
225 char *GetCpuFriendlyName(unsigned int regs[], unsigned int regs2[]){
226 char vendor[13];
227 char *retname;
228 int i;
230 if (NULL==(retname=malloc(256))) {
231 mp_msg(MSGT_CPUDETECT,MSGL_FATAL,"Error: GetCpuFriendlyName() not enough memory\n");
232 exit(1);
234 retname[0] = '\0';
236 sprintf(vendor,"%.4s%.4s%.4s",(char*)(regs+1),(char*)(regs+3),(char*)(regs+2));
238 do_cpuid(0x80000000,regs);
239 if (regs[0] >= 0x80000004)
241 // CPU has built-in namestring
242 for (i = 0x80000002; i <= 0x80000004; i++)
244 do_cpuid(i, regs);
245 strncat(retname, (char*)regs, 16);
248 return retname;
251 #if defined(__linux__) && defined(_POSIX_SOURCE) && !ARCH_X86_64
252 static void sigill_handler_sse( int signal, struct sigcontext sc )
254 mp_msg(MSGT_CPUDETECT,MSGL_V, "SIGILL, " );
256 /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
257 * instructions are 3 bytes long. We must increment the instruction
258 * pointer manually to avoid repeated execution of the offending
259 * instruction.
261 * If the SIGILL is caused by a divide-by-zero when unmasked
262 * exceptions aren't supported, the SIMD FPU status and control
263 * word will be restored at the end of the test, so we don't need
264 * to worry about doing it here. Besides, we may not be able to...
266 sc.eip += 3;
268 gCpuCaps.hasSSE=0;
270 #endif /* __linux__ && _POSIX_SOURCE */
272 #if (defined(__MINGW32__) || defined(__CYGWIN__)) && !ARCH_X86_64
273 LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
275 if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
276 mp_msg(MSGT_CPUDETECT,MSGL_V, "SIGILL, " );
277 ep->ContextRecord->Eip +=3;
278 gCpuCaps.hasSSE=0;
279 return EXCEPTION_CONTINUE_EXECUTION;
281 return EXCEPTION_CONTINUE_SEARCH;
283 #endif /* defined(__MINGW32__) || defined(__CYGWIN__) */
285 #ifdef __OS2__
286 ULONG _System os2_sig_handler_sse(PEXCEPTIONREPORTRECORD p1,
287 PEXCEPTIONREGISTRATIONRECORD p2,
288 PCONTEXTRECORD p3,
289 PVOID p4)
291 if(p1->ExceptionNum == XCPT_ILLEGAL_INSTRUCTION){
292 mp_msg(MSGT_CPUDETECT, MSGL_V, "SIGILL, ");
294 p3->ctx_RegEip += 3;
295 gCpuCaps.hasSSE = 0;
297 return XCPT_CONTINUE_EXECUTION;
299 return XCPT_CONTINUE_SEARCH;
301 #endif
303 /* If we're running on a processor that can do SSE, let's see if we
304 * are allowed to or not. This will catch 2.4.0 or later kernels that
305 * haven't been configured for a Pentium III but are running on one,
306 * and RedHat patched 2.2 kernels that have broken exception handling
307 * support for user space apps that do SSE.
310 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__)
311 #define SSE_SYSCTL_NAME "hw.instruction_sse"
312 #elif defined(__APPLE__)
313 #define SSE_SYSCTL_NAME "hw.optional.sse"
314 #endif
316 static void check_os_katmai_support( void )
318 #if ARCH_X86_64
319 gCpuCaps.hasSSE=1;
320 gCpuCaps.hasSSE2=1;
321 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) || defined(__DragonFly__) || defined(__APPLE__)
322 int has_sse=0, ret;
323 size_t len=sizeof(has_sse);
325 ret = sysctlbyname(SSE_SYSCTL_NAME, &has_sse, &len, NULL, 0);
326 if (ret || !has_sse)
327 gCpuCaps.hasSSE=0;
329 #elif defined(__NetBSD__) || defined (__OpenBSD__)
330 #if __NetBSD_Version__ >= 105250000 || (defined __OpenBSD__)
331 int has_sse, has_sse2, ret, mib[2];
332 size_t varlen;
334 mib[0] = CTL_MACHDEP;
335 mib[1] = CPU_SSE;
336 varlen = sizeof(has_sse);
338 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
339 ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
340 gCpuCaps.hasSSE = ret >= 0 && has_sse;
341 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
343 mib[1] = CPU_SSE2;
344 varlen = sizeof(has_sse2);
345 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE2... " );
346 ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
347 gCpuCaps.hasSSE2 = ret >= 0 && has_sse2;
348 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE2 ? "yes.\n" : "no!\n" );
349 #else
350 gCpuCaps.hasSSE = 0;
351 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "No OS support for SSE, disabling to be safe.\n" );
352 #endif
353 #elif defined(__MINGW32__) || defined(__CYGWIN__)
354 LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
355 if ( gCpuCaps.hasSSE ) {
356 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
357 exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
358 __asm__ volatile ("xorps %xmm0, %xmm0");
359 SetUnhandledExceptionFilter(exc_fil);
360 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
362 #elif defined(__OS2__)
363 EXCEPTIONREGISTRATIONRECORD RegRec = { 0, &os2_sig_handler_sse };
364 if ( gCpuCaps.hasSSE ) {
365 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
366 DosSetExceptionHandler( &RegRec );
367 __asm__ volatile ("xorps %xmm0, %xmm0");
368 DosUnsetExceptionHandler( &RegRec );
369 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
371 #elif defined(__linux__)
372 #if defined(_POSIX_SOURCE)
373 struct sigaction saved_sigill;
375 /* Save the original signal handlers.
377 sigaction( SIGILL, NULL, &saved_sigill );
379 signal( SIGILL, (void (*)(int))sigill_handler_sse );
381 /* Emulate test for OSFXSR in CR4. The OS will set this bit if it
382 * supports the extended FPU save and restore required for SSE. If
383 * we execute an SSE instruction on a PIII and get a SIGILL, the OS
384 * doesn't support Streaming SIMD Exceptions, even if the processor
385 * does.
387 if ( gCpuCaps.hasSSE ) {
388 mp_msg(MSGT_CPUDETECT,MSGL_V, "Testing OS support for SSE... " );
390 // __asm__ volatile ("xorps %%xmm0, %%xmm0");
391 __asm__ volatile ("xorps %xmm0, %xmm0");
393 mp_msg(MSGT_CPUDETECT,MSGL_V, gCpuCaps.hasSSE ? "yes.\n" : "no!\n" );
396 /* Restore the original signal handlers.
398 sigaction( SIGILL, &saved_sigill, NULL );
400 /* If we've gotten to here and the XMM CPUID bit is still set, we're
401 * safe to go ahead and hook out the SSE code throughout Mesa.
403 mp_msg(MSGT_CPUDETECT,MSGL_V, "Tests of OS support for SSE %s\n", gCpuCaps.hasSSE ? "passed." : "failed!" );
404 #else
405 /* We can't use POSIX signal handling to test the availability of
406 * SSE, so we disable it by default.
408 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "Cannot test OS support for SSE, disabling to be safe.\n" );
409 gCpuCaps.hasSSE=0;
410 #endif /* _POSIX_SOURCE */
411 #else
412 /* Do nothing on other platforms for now.
414 mp_msg(MSGT_CPUDETECT,MSGL_WARN, "Cannot test OS support for SSE, leaving disabled.\n" );
415 gCpuCaps.hasSSE=0;
416 #endif /* __linux__ */
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 */