Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
1891 | serge | 1 | /* |
2 | * Copyright © 2000 SuSE, Inc. |
||
3 | * Copyright © 2007 Red Hat, Inc. |
||
4 | * |
||
5 | * Permission to use, copy, modify, distribute, and sell this software and its |
||
6 | * documentation for any purpose is hereby granted without fee, provided that |
||
7 | * the above copyright notice appear in all copies and that both that |
||
8 | * copyright notice and this permission notice appear in supporting |
||
9 | * documentation, and that the name of SuSE not be used in advertising or |
||
10 | * publicity pertaining to distribution of the software without specific, |
||
11 | * written prior permission. SuSE makes no representations about the |
||
12 | * suitability of this software for any purpose. It is provided "as is" |
||
13 | * without express or implied warranty. |
||
14 | * |
||
15 | * SuSE DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL |
||
16 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL SuSE |
||
17 | * BE LIABLE FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES |
||
18 | * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION |
||
19 | * OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN |
||
20 | * CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. |
||
21 | */ |
||
22 | #ifdef HAVE_CONFIG_H |
||
23 | #include |
||
24 | #endif |
||
25 | |||
26 | #include |
||
27 | |||
28 | #if defined(USE_ARM_SIMD) && defined(_MSC_VER) |
||
29 | /* Needed for EXCEPTION_ILLEGAL_INSTRUCTION */ |
||
30 | #include |
||
31 | #endif |
||
32 | |||
33 | #include "pixman-private.h" |
||
34 | |||
35 | #ifdef USE_VMX |
||
36 | |||
37 | /* The CPU detection code needs to be in a file not compiled with |
||
38 | * "-maltivec -mabi=altivec", as gcc would try to save vector register |
||
39 | * across function calls causing SIGILL on cpus without Altivec/vmx. |
||
40 | */ |
||
41 | static pixman_bool_t initialized = FALSE; |
||
42 | static volatile pixman_bool_t have_vmx = TRUE; |
||
43 | |||
44 | #ifdef __APPLE__ |
||
45 | #include |
||
46 | |||
47 | static pixman_bool_t |
||
48 | pixman_have_vmx (void) |
||
49 | { |
||
50 | if (!initialized) |
||
51 | { |
||
52 | size_t length = sizeof(have_vmx); |
||
53 | int error = |
||
54 | sysctlbyname ("hw.optional.altivec", &have_vmx, &length, NULL, 0); |
||
55 | |||
56 | if (error) |
||
57 | have_vmx = FALSE; |
||
58 | |||
59 | initialized = TRUE; |
||
60 | } |
||
61 | return have_vmx; |
||
62 | } |
||
63 | |||
64 | #elif defined (__OpenBSD__) |
||
65 | #include |
||
66 | #include |
||
67 | #include |
||
68 | |||
69 | static pixman_bool_t |
||
70 | pixman_have_vmx (void) |
||
71 | { |
||
72 | if (!initialized) |
||
73 | { |
||
74 | int mib[2] = { CTL_MACHDEP, CPU_ALTIVEC }; |
||
75 | size_t length = sizeof(have_vmx); |
||
76 | int error = |
||
77 | sysctl (mib, 2, &have_vmx, &length, NULL, 0); |
||
78 | |||
79 | if (error != 0) |
||
80 | have_vmx = FALSE; |
||
81 | |||
82 | initialized = TRUE; |
||
83 | } |
||
84 | return have_vmx; |
||
85 | } |
||
86 | |||
87 | #elif defined (__linux__) |
||
88 | #include |
||
89 | #include |
||
90 | #include |
||
91 | #include |
||
92 | #include |
||
93 | #include |
||
94 | #include |
||
95 | |||
96 | static pixman_bool_t |
||
97 | pixman_have_vmx (void) |
||
98 | { |
||
99 | if (!initialized) |
||
100 | { |
||
101 | char fname[64]; |
||
102 | unsigned long buf[64]; |
||
103 | ssize_t count = 0; |
||
104 | pid_t pid; |
||
105 | int fd, i; |
||
106 | |||
107 | pid = getpid (); |
||
108 | snprintf (fname, sizeof(fname) - 1, "/proc/%d/auxv", pid); |
||
109 | |||
110 | fd = open (fname, O_RDONLY); |
||
111 | if (fd >= 0) |
||
112 | { |
||
113 | for (i = 0; i <= (count / sizeof(unsigned long)); i += 2) |
||
114 | { |
||
115 | /* Read more if buf is empty... */ |
||
116 | if (i == (count / sizeof(unsigned long))) |
||
117 | { |
||
118 | count = read (fd, buf, sizeof(buf)); |
||
119 | if (count <= 0) |
||
120 | break; |
||
121 | i = 0; |
||
122 | } |
||
123 | |||
124 | if (buf[i] == AT_HWCAP) |
||
125 | { |
||
126 | have_vmx = !!(buf[i + 1] & PPC_FEATURE_HAS_ALTIVEC); |
||
127 | initialized = TRUE; |
||
128 | break; |
||
129 | } |
||
130 | else if (buf[i] == AT_NULL) |
||
131 | { |
||
132 | break; |
||
133 | } |
||
134 | } |
||
135 | close (fd); |
||
136 | } |
||
137 | } |
||
138 | if (!initialized) |
||
139 | { |
||
140 | /* Something went wrong. Assume 'no' rather than playing |
||
141 | fragile tricks with catching SIGILL. */ |
||
142 | have_vmx = FALSE; |
||
143 | initialized = TRUE; |
||
144 | } |
||
145 | |||
146 | return have_vmx; |
||
147 | } |
||
148 | |||
149 | #else /* !__APPLE__ && !__OpenBSD__ && !__linux__ */ |
||
150 | #include |
||
151 | #include |
||
152 | |||
153 | static jmp_buf jump_env; |
||
154 | |||
155 | static void |
||
156 | vmx_test (int sig, |
||
157 | siginfo_t *si, |
||
158 | void * unused) |
||
159 | { |
||
160 | longjmp (jump_env, 1); |
||
161 | } |
||
162 | |||
163 | static pixman_bool_t |
||
164 | pixman_have_vmx (void) |
||
165 | { |
||
166 | struct sigaction sa, osa; |
||
167 | int jmp_result; |
||
168 | |||
169 | if (!initialized) |
||
170 | { |
||
171 | sa.sa_flags = SA_SIGINFO; |
||
172 | sigemptyset (&sa.sa_mask); |
||
173 | sa.sa_sigaction = vmx_test; |
||
174 | sigaction (SIGILL, &sa, &osa); |
||
175 | jmp_result = setjmp (jump_env); |
||
176 | if (jmp_result == 0) |
||
177 | { |
||
178 | asm volatile ( "vor 0, 0, 0" ); |
||
179 | } |
||
180 | sigaction (SIGILL, &osa, NULL); |
||
181 | have_vmx = (jmp_result == 0); |
||
182 | initialized = TRUE; |
||
183 | } |
||
184 | return have_vmx; |
||
185 | } |
||
186 | |||
187 | #endif /* __APPLE__ */ |
||
188 | #endif /* USE_VMX */ |
||
189 | |||
190 | #if defined(USE_ARM_SIMD) || defined(USE_ARM_NEON) |
||
191 | |||
192 | #if defined(_MSC_VER) |
||
193 | |||
194 | #if defined(USE_ARM_SIMD) |
||
195 | extern int pixman_msvc_try_arm_simd_op (); |
||
196 | |||
197 | pixman_bool_t |
||
198 | pixman_have_arm_simd (void) |
||
199 | { |
||
200 | static pixman_bool_t initialized = FALSE; |
||
201 | static pixman_bool_t have_arm_simd = FALSE; |
||
202 | |||
203 | if (!initialized) |
||
204 | { |
||
205 | __try { |
||
206 | pixman_msvc_try_arm_simd_op (); |
||
207 | have_arm_simd = TRUE; |
||
208 | } __except (GetExceptionCode () == EXCEPTION_ILLEGAL_INSTRUCTION) { |
||
209 | have_arm_simd = FALSE; |
||
210 | } |
||
211 | initialized = TRUE; |
||
212 | } |
||
213 | |||
214 | return have_arm_simd; |
||
215 | } |
||
216 | |||
217 | #endif /* USE_ARM_SIMD */ |
||
218 | |||
219 | #if defined(USE_ARM_NEON) |
||
220 | extern int pixman_msvc_try_arm_neon_op (); |
||
221 | |||
222 | pixman_bool_t |
||
223 | pixman_have_arm_neon (void) |
||
224 | { |
||
225 | static pixman_bool_t initialized = FALSE; |
||
226 | static pixman_bool_t have_arm_neon = FALSE; |
||
227 | |||
228 | if (!initialized) |
||
229 | { |
||
230 | __try |
||
231 | { |
||
232 | pixman_msvc_try_arm_neon_op (); |
||
233 | have_arm_neon = TRUE; |
||
234 | } |
||
235 | __except (GetExceptionCode () == EXCEPTION_ILLEGAL_INSTRUCTION) |
||
236 | { |
||
237 | have_arm_neon = FALSE; |
||
238 | } |
||
239 | initialized = TRUE; |
||
240 | } |
||
241 | |||
242 | return have_arm_neon; |
||
243 | } |
||
244 | |||
245 | #endif /* USE_ARM_NEON */ |
||
246 | |||
247 | #else /* linux ELF */ |
||
248 | |||
249 | #include |
||
250 | #include |
||
251 | #include |
||
252 | #include |
||
253 | #include |
||
254 | #include |
||
255 | #include |
||
256 | #include |
||
257 | |||
258 | static pixman_bool_t arm_has_v7 = FALSE; |
||
259 | static pixman_bool_t arm_has_v6 = FALSE; |
||
260 | static pixman_bool_t arm_has_vfp = FALSE; |
||
261 | static pixman_bool_t arm_has_neon = FALSE; |
||
262 | static pixman_bool_t arm_has_iwmmxt = FALSE; |
||
263 | static pixman_bool_t arm_tests_initialized = FALSE; |
||
264 | |||
265 | static void |
||
266 | pixman_arm_read_auxv () |
||
267 | { |
||
268 | int fd; |
||
269 | Elf32_auxv_t aux; |
||
270 | |||
271 | fd = open ("/proc/self/auxv", O_RDONLY); |
||
272 | if (fd >= 0) |
||
273 | { |
||
274 | while (read (fd, &aux, sizeof(Elf32_auxv_t)) == sizeof(Elf32_auxv_t)) |
||
275 | { |
||
276 | if (aux.a_type == AT_HWCAP) |
||
277 | { |
||
278 | uint32_t hwcap = aux.a_un.a_val; |
||
279 | /* hardcode these values to avoid depending on specific |
||
280 | * versions of the hwcap header, e.g. HWCAP_NEON |
||
281 | */ |
||
282 | arm_has_vfp = (hwcap & 64) != 0; |
||
283 | arm_has_iwmmxt = (hwcap & 512) != 0; |
||
284 | /* this flag is only present on kernel 2.6.29 */ |
||
285 | arm_has_neon = (hwcap & 4096) != 0; |
||
286 | } |
||
287 | else if (aux.a_type == AT_PLATFORM) |
||
288 | { |
||
289 | const char *plat = (const char*) aux.a_un.a_val; |
||
290 | if (strncmp (plat, "v7l", 3) == 0) |
||
291 | { |
||
292 | arm_has_v7 = TRUE; |
||
293 | arm_has_v6 = TRUE; |
||
294 | } |
||
295 | else if (strncmp (plat, "v6l", 3) == 0) |
||
296 | { |
||
297 | arm_has_v6 = TRUE; |
||
298 | } |
||
299 | } |
||
300 | } |
||
301 | close (fd); |
||
302 | } |
||
303 | |||
304 | arm_tests_initialized = TRUE; |
||
305 | } |
||
306 | |||
307 | #if defined(USE_ARM_SIMD) |
||
308 | pixman_bool_t |
||
309 | pixman_have_arm_simd (void) |
||
310 | { |
||
311 | if (!arm_tests_initialized) |
||
312 | pixman_arm_read_auxv (); |
||
313 | |||
314 | return arm_has_v6; |
||
315 | } |
||
316 | |||
317 | #endif /* USE_ARM_SIMD */ |
||
318 | |||
319 | #if defined(USE_ARM_NEON) |
||
320 | pixman_bool_t |
||
321 | pixman_have_arm_neon (void) |
||
322 | { |
||
323 | if (!arm_tests_initialized) |
||
324 | pixman_arm_read_auxv (); |
||
325 | |||
326 | return arm_has_neon; |
||
327 | } |
||
328 | |||
329 | #endif /* USE_ARM_NEON */ |
||
330 | |||
331 | #endif /* linux */ |
||
332 | |||
333 | #endif /* USE_ARM_SIMD || USE_ARM_NEON */ |
||
334 | |||
335 | #if defined(USE_MMX) || defined(USE_SSE2) |
||
336 | /* The CPU detection code needs to be in a file not compiled with |
||
337 | * "-mmmx -msse", as gcc would generate CMOV instructions otherwise |
||
338 | * that would lead to SIGILL instructions on old CPUs that don't have |
||
339 | * it. |
||
340 | */ |
||
341 | #if !defined(__amd64__) && !defined(__x86_64__) && !defined(_M_AMD64) |
||
342 | |||
343 | #ifdef HAVE_GETISAX |
||
344 | #include |
||
345 | #endif |
||
346 | |||
347 | typedef enum |
||
348 | { |
||
349 | NO_FEATURES = 0, |
||
350 | MMX = 0x1, |
||
351 | MMX_EXTENSIONS = 0x2, |
||
352 | SSE = 0x6, |
||
353 | SSE2 = 0x8, |
||
354 | CMOV = 0x10 |
||
355 | } cpu_features_t; |
||
356 | |||
357 | |||
358 | static unsigned int |
||
359 | detect_cpu_features (void) |
||
360 | { |
||
361 | unsigned int features = 0; |
||
362 | unsigned int result = 0; |
||
363 | |||
364 | #ifdef HAVE_GETISAX |
||
365 | if (getisax (&result, 1)) |
||
366 | { |
||
367 | if (result & AV_386_CMOV) |
||
368 | features |= CMOV; |
||
369 | if (result & AV_386_MMX) |
||
370 | features |= MMX; |
||
371 | if (result & AV_386_AMD_MMX) |
||
372 | features |= MMX_EXTENSIONS; |
||
373 | if (result & AV_386_SSE) |
||
374 | features |= SSE; |
||
375 | if (result & AV_386_SSE2) |
||
376 | features |= SSE2; |
||
377 | } |
||
378 | #else |
||
379 | char vendor[13]; |
||
380 | #ifdef _MSC_VER |
||
381 | int vendor0 = 0, vendor1, vendor2; |
||
382 | #endif |
||
383 | vendor[0] = 0; |
||
384 | vendor[12] = 0; |
||
385 | |||
386 | #ifdef __GNUC__ |
||
387 | /* see p. 118 of amd64 instruction set manual Vol3 */ |
||
388 | /* We need to be careful about the handling of %ebx and |
||
389 | * %esp here. We can't declare either one as clobbered |
||
390 | * since they are special registers (%ebx is the "PIC |
||
391 | * register" holding an offset to global data, %esp the |
||
392 | * stack pointer), so we need to make sure they have their |
||
393 | * original values when we access the output operands. |
||
394 | */ |
||
395 | __asm__ ( |
||
396 | "pushf\n" |
||
397 | "pop %%eax\n" |
||
398 | "mov %%eax, %%ecx\n" |
||
399 | "xor $0x00200000, %%eax\n" |
||
400 | "push %%eax\n" |
||
401 | "popf\n" |
||
402 | "pushf\n" |
||
403 | "pop %%eax\n" |
||
404 | "mov $0x0, %%edx\n" |
||
405 | "xor %%ecx, %%eax\n" |
||
406 | "jz 1f\n" |
||
407 | |||
408 | "mov $0x00000000, %%eax\n" |
||
409 | "push %%ebx\n" |
||
410 | "cpuid\n" |
||
411 | "mov %%ebx, %%eax\n" |
||
412 | "pop %%ebx\n" |
||
413 | "mov %%eax, %1\n" |
||
414 | "mov %%edx, %2\n" |
||
415 | "mov %%ecx, %3\n" |
||
416 | "mov $0x00000001, %%eax\n" |
||
417 | "push %%ebx\n" |
||
418 | "cpuid\n" |
||
419 | "pop %%ebx\n" |
||
420 | "1:\n" |
||
421 | "mov %%edx, %0\n" |
||
422 | : "=r" (result), |
||
423 | "=m" (vendor[0]), |
||
424 | "=m" (vendor[4]), |
||
425 | "=m" (vendor[8]) |
||
426 | : |
||
427 | : "%eax", "%ecx", "%edx" |
||
428 | ); |
||
429 | |||
430 | #elif defined (_MSC_VER) |
||
431 | |||
432 | _asm { |
||
433 | pushfd |
||
434 | pop eax |
||
435 | mov ecx, eax |
||
436 | xor eax, 00200000h |
||
437 | push eax |
||
438 | popfd |
||
439 | pushfd |
||
440 | pop eax |
||
441 | mov edx, 0 |
||
442 | xor eax, ecx |
||
443 | jz nocpuid |
||
444 | |||
445 | mov eax, 0 |
||
446 | push ebx |
||
447 | cpuid |
||
448 | mov eax, ebx |
||
449 | pop ebx |
||
450 | mov vendor0, eax |
||
451 | mov vendor1, edx |
||
452 | mov vendor2, ecx |
||
453 | mov eax, 1 |
||
454 | push ebx |
||
455 | cpuid |
||
456 | pop ebx |
||
457 | nocpuid: |
||
458 | mov result, edx |
||
459 | } |
||
460 | memmove (vendor + 0, &vendor0, 4); |
||
461 | memmove (vendor + 4, &vendor1, 4); |
||
462 | memmove (vendor + 8, &vendor2, 4); |
||
463 | |||
464 | #else |
||
465 | # error unsupported compiler |
||
466 | #endif |
||
467 | |||
468 | features = 0; |
||
469 | if (result) |
||
470 | { |
||
471 | /* result now contains the standard feature bits */ |
||
472 | if (result & (1 << 15)) |
||
473 | features |= CMOV; |
||
474 | if (result & (1 << 23)) |
||
475 | features |= MMX; |
||
476 | if (result & (1 << 25)) |
||
477 | features |= SSE; |
||
478 | if (result & (1 << 26)) |
||
479 | features |= SSE2; |
||
480 | if ((features & MMX) && !(features & SSE) && |
||
481 | (strcmp (vendor, "AuthenticAMD") == 0 || |
||
482 | strcmp (vendor, "Geode by NSC") == 0)) |
||
483 | { |
||
484 | /* check for AMD MMX extensions */ |
||
485 | #ifdef __GNUC__ |
||
486 | __asm__ ( |
||
487 | " push %%ebx\n" |
||
488 | " mov $0x80000000, %%eax\n" |
||
489 | " cpuid\n" |
||
490 | " xor %%edx, %%edx\n" |
||
491 | " cmp $0x1, %%eax\n" |
||
492 | " jge 2f\n" |
||
493 | " mov $0x80000001, %%eax\n" |
||
494 | " cpuid\n" |
||
495 | "2:\n" |
||
496 | " pop %%ebx\n" |
||
497 | " mov %%edx, %0\n" |
||
498 | : "=r" (result) |
||
499 | : |
||
500 | : "%eax", "%ecx", "%edx" |
||
501 | ); |
||
502 | #elif defined _MSC_VER |
||
503 | _asm { |
||
504 | push ebx |
||
505 | mov eax, 80000000h |
||
506 | cpuid |
||
507 | xor edx, edx |
||
508 | cmp eax, 1 |
||
509 | jge notamd |
||
510 | mov eax, 80000001h |
||
511 | cpuid |
||
512 | notamd: |
||
513 | pop ebx |
||
514 | mov result, edx |
||
515 | } |
||
516 | #endif |
||
517 | if (result & (1 << 22)) |
||
518 | features |= MMX_EXTENSIONS; |
||
519 | } |
||
520 | } |
||
521 | #endif /* HAVE_GETISAX */ |
||
522 | |||
523 | return features; |
||
524 | } |
||
525 | |||
526 | static pixman_bool_t |
||
527 | pixman_have_mmx (void) |
||
528 | { |
||
529 | static pixman_bool_t initialized = FALSE; |
||
530 | static pixman_bool_t mmx_present; |
||
531 | |||
532 | if (!initialized) |
||
533 | { |
||
534 | unsigned int features = detect_cpu_features (); |
||
535 | mmx_present = (features & (MMX | MMX_EXTENSIONS)) == (MMX | MMX_EXTENSIONS); |
||
536 | initialized = TRUE; |
||
537 | } |
||
538 | |||
539 | return mmx_present; |
||
540 | } |
||
541 | |||
542 | #ifdef USE_SSE2 |
||
543 | static pixman_bool_t |
||
544 | pixman_have_sse2 (void) |
||
545 | { |
||
546 | static pixman_bool_t initialized = FALSE; |
||
547 | static pixman_bool_t sse2_present; |
||
548 | |||
549 | if (!initialized) |
||
550 | { |
||
551 | unsigned int features = detect_cpu_features (); |
||
552 | sse2_present = (features & (MMX | MMX_EXTENSIONS | SSE | SSE2)) == (MMX | MMX_EXTENSIONS | SSE | SSE2); |
||
553 | initialized = TRUE; |
||
554 | } |
||
555 | |||
556 | return sse2_present; |
||
557 | } |
||
558 | |||
559 | #endif |
||
560 | |||
561 | #else /* __amd64__ */ |
||
562 | #ifdef USE_MMX |
||
563 | #define pixman_have_mmx() TRUE |
||
564 | #endif |
||
565 | #ifdef USE_SSE2 |
||
566 | #define pixman_have_sse2() TRUE |
||
567 | #endif |
||
568 | #endif /* __amd64__ */ |
||
569 | #endif |
||
570 | |||
571 | pixman_implementation_t * |
||
572 | _pixman_choose_implementation (void) |
||
573 | { |
||
574 | #ifdef USE_SSE2 |
||
575 | if (pixman_have_sse2 ()) |
||
576 | return _pixman_implementation_create_sse2 (); |
||
577 | #endif |
||
578 | #ifdef USE_MMX |
||
579 | if (pixman_have_mmx ()) |
||
580 | return _pixman_implementation_create_mmx (); |
||
581 | #endif |
||
582 | |||
583 | #ifdef USE_ARM_NEON |
||
584 | if (pixman_have_arm_neon ()) |
||
585 | return _pixman_implementation_create_arm_neon (); |
||
586 | #endif |
||
587 | #ifdef USE_ARM_SIMD |
||
588 | if (pixman_have_arm_simd ()) |
||
589 | return _pixman_implementation_create_arm_simd (); |
||
590 | #endif |
||
591 | #ifdef USE_VMX |
||
592 | if (pixman_have_vmx ()) |
||
593 | return _pixman_implementation_create_vmx (); |
||
594 | #endif |
||
595 | |||
596 | return _pixman_implementation_create_fast_path (); |
||
597 | }><>><>><>><>><>=>=> |
||
598 |