1 /**************************************************************************
3 * Copyright 2008 Dennis Smit
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:
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
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.
25 **************************************************************************/
29 * CPU feature detection.
32 * @author Based on the work of Eric Anholt <anholt@FreeBSD.org>
35 #include "pipe/p_config.h"
38 #include "u_cpu_detect.h"
40 #if defined(PIPE_ARCH_PPC)
41 #if defined(PIPE_OS_APPLE)
42 #include <sys/sysctl.h>
49 #if defined(PIPE_OS_NETBSD) || defined(PIPE_OS_OPENBSD)
50 #include <sys/param.h>
51 #include <sys/sysctl.h>
52 #include <machine/cpu.h>
55 #if defined(PIPE_OS_FREEBSD)
56 #include <sys/types.h>
57 #include <sys/sysctl.h>
60 #if defined(PIPE_OS_LINUX)
68 #if defined(PIPE_OS_WINDOWS)
76 DEBUG_GET_ONCE_BOOL_OPTION(dump_cpu
, "GALLIUM_DUMP_CPU", FALSE
)
79 struct util_cpu_caps util_cpu_caps
;
81 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
82 static int has_cpuid(void);
86 #if defined(PIPE_ARCH_X86)
88 /* The sigill handlers */
89 #if defined(PIPE_OS_LINUX) /*&& defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)*/
91 sigill_handler_sse(int signal
, struct sigcontext sc
)
93 /* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
94 * instructions are 3 bytes long. We must increment the instruction
95 * pointer manually to avoid repeated execution of the offending
98 * If the SIGILL is caused by a divide-by-zero when unmasked
99 * exceptions aren't supported, the SIMD FPU status and control
100 * word will be restored at the end of the test, so we don't need
101 * to worry about doing it here. Besides, we may not be able to...
105 util_cpu_caps
.has_sse
=0;
109 sigfpe_handler_sse(int signal
, struct sigcontext sc
)
111 if (sc
.fpstate
->magic
!= 0xffff) {
112 /* Our signal context has the extended FPU state, so reset the
113 * divide-by-zero exception mask and clear the divide-by-zero
116 sc
.fpstate
->mxcsr
|= 0x00000200;
117 sc
.fpstate
->mxcsr
&= 0xfffffffb;
119 /* If we ever get here, we're completely hosed.
123 #endif /* PIPE_OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
125 #if defined(PIPE_OS_WINDOWS)
127 win32_sig_handler_sse(EXCEPTION_POINTERS
* ep
)
129 if(ep
->ExceptionRecord
->ExceptionCode
==EXCEPTION_ILLEGAL_INSTRUCTION
){
130 ep
->ContextRecord
->Eip
+=3;
131 util_cpu_caps
.has_sse
=0;
132 return EXCEPTION_CONTINUE_EXECUTION
;
134 return EXCEPTION_CONTINUE_SEARCH
;
136 #endif /* PIPE_OS_WINDOWS */
138 #endif /* PIPE_ARCH_X86 */
141 #if defined(PIPE_ARCH_PPC) && !defined(PIPE_OS_APPLE)
142 static jmp_buf __lv_powerpc_jmpbuf
;
143 static volatile sig_atomic_t __lv_powerpc_canjump
= 0;
146 sigill_handler(int sig
)
148 if (!__lv_powerpc_canjump
) {
149 signal (sig
, SIG_DFL
);
153 __lv_powerpc_canjump
= 0;
154 longjmp(__lv_powerpc_jmpbuf
, 1);
158 #if defined(PIPE_ARCH_PPC)
160 check_os_altivec_support(void)
162 #if defined(PIPE_OS_APPLE)
163 int sels
[2] = {CTL_HW
, HW_VECTORUNIT
};
165 int len
= sizeof (has_vu
);
168 err
= sysctl(sels
, 2, &has_vu
, &len
, NULL
, 0);
172 util_cpu_caps
.has_altivec
= 1;
175 #else /* !PIPE_OS_APPLE */
176 /* not on Apple/Darwin, do it the brute-force way */
177 /* this is borrowed from the libmpeg2 library */
178 signal(SIGILL
, sigill_handler
);
179 if (setjmp(__lv_powerpc_jmpbuf
)) {
180 signal(SIGILL
, SIG_DFL
);
182 __lv_powerpc_canjump
= 1;
186 "vand %%v0, %%v0, %%v0"
190 signal(SIGILL
, SIG_DFL
);
191 util_cpu_caps
.has_altivec
= 1;
193 #endif /* !PIPE_OS_APPLE */
195 #endif /* PIPE_ARCH_PPC */
198 #if defined(PIPE_ARCH_X86) || defined (PIPE_ARCH_X86_64)
199 static int has_cpuid(void)
201 #if defined(PIPE_ARCH_X86)
202 #if defined(PIPE_OS_GCC)
209 "xorl $0x200000, %0\n"
223 #elif defined(PIPE_ARCH_X86_64)
232 * @sa cpuid.h included in gcc-4.3 onwards.
233 * @sa http://msdn.microsoft.com/en-us/library/hskdteyh.aspx
236 cpuid(uint32_t ax
, uint32_t *p
)
238 #if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
240 "xchgl %%ebx, %1\n\t"
249 #elif defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86_64)
258 #elif defined(PIPE_CC_MSVC)
267 #endif /* X86 or X86_64 */
270 util_cpu_detect(void)
272 static boolean util_cpu_detect_initialized
= FALSE
;
274 if(util_cpu_detect_initialized
)
277 memset(&util_cpu_caps
, 0, sizeof util_cpu_caps
);
279 /* Count the number of CPUs in system */
280 #if defined(PIPE_OS_WINDOWS)
282 SYSTEM_INFO system_info
;
283 GetSystemInfo(&system_info
);
284 util_cpu_caps
.nr_cpus
= system_info
.dwNumberOfProcessors
;
286 #elif defined(PIPE_OS_UNIX) && defined(_SC_NPROCESSORS_ONLN)
287 util_cpu_caps
.nr_cpus
= sysconf(_SC_NPROCESSORS_ONLN
);
288 if (util_cpu_caps
.nr_cpus
== -1)
289 util_cpu_caps
.nr_cpus
= 1;
290 #elif defined(PIPE_OS_BSD)
299 sysctl(mib
, 2, &ncpu
, &len
, NULL
, 0);
300 util_cpu_caps
.nr_cpus
= ncpu
;
303 util_cpu_caps
.nr_cpus
= 1;
306 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
311 util_cpu_caps
.cacheline
= 32;
313 /* Get max cpuid level */
314 cpuid(0x00000000, regs
);
316 if (regs
[0] >= 0x00000001) {
317 unsigned int cacheline
;
319 cpuid (0x00000001, regs2
);
321 util_cpu_caps
.x86_cpu_type
= (regs2
[0] >> 8) & 0xf;
322 if (util_cpu_caps
.x86_cpu_type
== 0xf)
323 util_cpu_caps
.x86_cpu_type
= 8 + ((regs2
[0] >> 20) & 255); /* use extended family (P4, IA64) */
325 /* general feature flags */
326 util_cpu_caps
.has_tsc
= (regs2
[3] & (1 << 8 )) >> 8; /* 0x0000010 */
327 util_cpu_caps
.has_mmx
= (regs2
[3] & (1 << 23 )) >> 23; /* 0x0800000 */
328 util_cpu_caps
.has_sse
= (regs2
[3] & (1 << 25 )) >> 25; /* 0x2000000 */
329 util_cpu_caps
.has_sse2
= (regs2
[3] & (1 << 26 )) >> 26; /* 0x4000000 */
330 util_cpu_caps
.has_sse3
= (regs2
[2] & (1)); /* 0x0000001 */
331 util_cpu_caps
.has_ssse3
= (regs2
[2] & (1 << 9 )) >> 9; /* 0x0000020 */
332 util_cpu_caps
.has_sse4_1
= (regs2
[2] & (1 << 19)) >> 19;
333 util_cpu_caps
.has_mmx2
= util_cpu_caps
.has_sse
; /* SSE cpus supports mmxext too */
335 cacheline
= ((regs2
[1] >> 8) & 0xFF) * 8;
337 util_cpu_caps
.cacheline
= cacheline
;
340 cpuid(0x80000000, regs
);
342 if (regs
[0] >= 0x80000001) {
344 cpuid(0x80000001, regs2
);
346 util_cpu_caps
.has_mmx
|= (regs2
[3] & (1 << 23 )) >> 23; /* 0x0800000 */
347 util_cpu_caps
.has_mmx2
|= (regs2
[3] & (1 << 22 )) >> 22; /* 0x400000 */
348 util_cpu_caps
.has_3dnow
= (regs2
[3] & (1 << 31 )) >> 31; /* 0x80000000 */
349 util_cpu_caps
.has_3dnow_ext
= (regs2
[3] & (1 << 30 )) >> 30;
352 if (regs
[0] >= 0x80000006) {
353 cpuid(0x80000006, regs2
);
354 util_cpu_caps
.cacheline
= regs2
[2] & 0xFF;
357 if (!util_cpu_caps
.has_sse
) {
358 util_cpu_caps
.has_sse2
= 0;
359 util_cpu_caps
.has_sse3
= 0;
360 util_cpu_caps
.has_ssse3
= 0;
361 util_cpu_caps
.has_sse4_1
= 0;
364 #endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
366 #if defined(PIPE_ARCH_PPC)
367 check_os_altivec_support();
368 #endif /* PIPE_ARCH_PPC */
371 if (debug_get_option_dump_cpu()) {
372 debug_printf("util_cpu_caps.nr_cpus = %u\n", util_cpu_caps
.nr_cpus
);
374 debug_printf("util_cpu_caps.x86_cpu_type = %u\n", util_cpu_caps
.x86_cpu_type
);
375 debug_printf("util_cpu_caps.cacheline = %u\n", util_cpu_caps
.cacheline
);
377 debug_printf("util_cpu_caps.has_tsc = %u\n", util_cpu_caps
.has_tsc
);
378 debug_printf("util_cpu_caps.has_mmx = %u\n", util_cpu_caps
.has_mmx
);
379 debug_printf("util_cpu_caps.has_mmx2 = %u\n", util_cpu_caps
.has_mmx2
);
380 debug_printf("util_cpu_caps.has_sse = %u\n", util_cpu_caps
.has_sse
);
381 debug_printf("util_cpu_caps.has_sse2 = %u\n", util_cpu_caps
.has_sse2
);
382 debug_printf("util_cpu_caps.has_sse3 = %u\n", util_cpu_caps
.has_sse3
);
383 debug_printf("util_cpu_caps.has_ssse3 = %u\n", util_cpu_caps
.has_ssse3
);
384 debug_printf("util_cpu_caps.has_sse4_1 = %u\n", util_cpu_caps
.has_sse4_1
);
385 debug_printf("util_cpu_caps.has_3dnow = %u\n", util_cpu_caps
.has_3dnow
);
386 debug_printf("util_cpu_caps.has_3dnow_ext = %u\n", util_cpu_caps
.has_3dnow_ext
);
387 debug_printf("util_cpu_caps.has_altivec = %u\n", util_cpu_caps
.has_altivec
);
391 util_cpu_detect_initialized
= TRUE
;