1/**************************************************************************
2 *
3 * Copyright 2008 Dennis Smit
4 * All Rights Reserved.
5 *
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:
12 *
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
15 * Software.
16 *
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.
24 *
25 **************************************************************************/
26
27/**
28 * @file
29 * CPU feature detection.
30 *
31 * @author Dennis Smit
32 * @author Based on the work of Eric Anholt <anholt@FreeBSD.org>
33 */
34
35#include "pipe/p_config.h"
36
37#include "u_debug.h"
38#include "u_cpu_detect.h"
39
40#if defined(PIPE_ARCH_PPC)
41#if defined(PIPE_OS_APPLE)
42#include <sys/sysctl.h>
43#else
44#include <signal.h>
45#include <setjmp.h>
46#endif
47#endif
48
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>
53#endif
54
55#if defined(PIPE_OS_FREEBSD)
56#include <sys/types.h>
57#include <sys/sysctl.h>
58#endif
59
60#if defined(PIPE_OS_LINUX)
61#include <signal.h>
62#endif
63
64#ifdef PIPE_OS_UNIX
65#include <unistd.h>
66#endif
67
68#if defined(PIPE_OS_WINDOWS)
69#include <windows.h>
70#if defined(MSVC)
71#include <intrin.h>
72#endif
73#endif
74
75
76#ifdef DEBUG
77DEBUG_GET_ONCE_BOOL_OPTION(dump_cpu, "GALLIUM_DUMP_CPU", FALSE)
78#endif
79
80
81struct util_cpu_caps util_cpu_caps;
82
83#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
84static int has_cpuid(void);
85#endif
86
87
88#if defined(PIPE_ARCH_PPC) && !defined(PIPE_OS_APPLE)
89static jmp_buf  __lv_powerpc_jmpbuf;
90static volatile sig_atomic_t __lv_powerpc_canjump = 0;
91
92static void
93sigill_handler(int sig)
94{
95   if (!__lv_powerpc_canjump) {
96      signal (sig, SIG_DFL);
97      raise (sig);
98   }
99
100   __lv_powerpc_canjump = 0;
101   longjmp(__lv_powerpc_jmpbuf, 1);
102}
103#endif
104
105#if defined(PIPE_ARCH_PPC)
106static void
107check_os_altivec_support(void)
108{
109#if defined(PIPE_OS_APPLE)
110   int sels[2] = {CTL_HW, HW_VECTORUNIT};
111   int has_vu = 0;
112   int len = sizeof (has_vu);
113   int err;
114
115   err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
116
117   if (err == 0) {
118      if (has_vu != 0) {
119         util_cpu_caps.has_altivec = 1;
120      }
121   }
122#else /* !PIPE_OS_APPLE */
123   /* not on Apple/Darwin, do it the brute-force way */
124   /* this is borrowed from the libmpeg2 library */
125   signal(SIGILL, sigill_handler);
126   if (setjmp(__lv_powerpc_jmpbuf)) {
127      signal(SIGILL, SIG_DFL);
128   } else {
129      __lv_powerpc_canjump = 1;
130
131      __asm __volatile
132         ("mtspr 256, %0\n\t"
133          "vand %%v0, %%v0, %%v0"
134          :
135          : "r" (-1));
136
137      signal(SIGILL, SIG_DFL);
138      util_cpu_caps.has_altivec = 1;
139   }
140#endif /* !PIPE_OS_APPLE */
141}
142#endif /* PIPE_ARCH_PPC */
143
144
145#if defined(PIPE_ARCH_X86) || defined (PIPE_ARCH_X86_64)
146static int has_cpuid(void)
147{
148#if defined(PIPE_ARCH_X86)
149#if defined(PIPE_OS_GCC)
150   int a, c;
151
152   __asm __volatile
153      ("pushf\n"
154       "popl %0\n"
155       "movl %0, %1\n"
156       "xorl $0x200000, %0\n"
157       "push %0\n"
158       "popf\n"
159       "pushf\n"
160       "popl %0\n"
161       : "=a" (a), "=c" (c)
162       :
163       : "cc");
164
165   return a != c;
166#else
167   /* FIXME */
168   return 1;
169#endif
170#elif defined(PIPE_ARCH_X86_64)
171   return 1;
172#else
173   return 0;
174#endif
175}
176
177
178/**
179 * @sa cpuid.h included in gcc-4.3 onwards.
180 * @sa http://msdn.microsoft.com/en-us/library/hskdteyh.aspx
181 */
182static INLINE void
183cpuid(uint32_t ax, uint32_t *p)
184{
185#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86)
186   __asm __volatile (
187     "xchgl %%ebx, %1\n\t"
188     "cpuid\n\t"
189     "xchgl %%ebx, %1"
190     : "=a" (p[0]),
191       "=S" (p[1]),
192       "=c" (p[2]),
193       "=d" (p[3])
194     : "0" (ax)
195   );
196#elif defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86_64)
197   __asm __volatile (
198     "cpuid\n\t"
199     : "=a" (p[0]),
200       "=b" (p[1]),
201       "=c" (p[2]),
202       "=d" (p[3])
203     : "0" (ax)
204   );
205#elif defined(PIPE_CC_MSVC)
206   __cpuid(p, ax);
207#else
208   p[0] = 0;
209   p[1] = 0;
210   p[2] = 0;
211   p[3] = 0;
212#endif
213}
214#endif /* X86 or X86_64 */
215
216void
217util_cpu_detect(void)
218{
219   static boolean util_cpu_detect_initialized = FALSE;
220
221   if(util_cpu_detect_initialized)
222      return;
223
224   memset(&util_cpu_caps, 0, sizeof util_cpu_caps);
225
226   /* Count the number of CPUs in system */
227#if defined(PIPE_OS_WINDOWS)
228   {
229      SYSTEM_INFO system_info;
230      GetSystemInfo(&system_info);
231      util_cpu_caps.nr_cpus = system_info.dwNumberOfProcessors;
232   }
233#elif defined(PIPE_OS_UNIX) && defined(_SC_NPROCESSORS_ONLN)
234   util_cpu_caps.nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
235   if (util_cpu_caps.nr_cpus == -1)
236      util_cpu_caps.nr_cpus = 1;
237#elif defined(PIPE_OS_BSD)
238   {
239      int mib[2], ncpu;
240      int len;
241
242      mib[0] = CTL_HW;
243      mib[1] = HW_NCPU;
244
245      len = sizeof (ncpu);
246      sysctl(mib, 2, &ncpu, &len, NULL, 0);
247      util_cpu_caps.nr_cpus = ncpu;
248   }
249#else
250   util_cpu_caps.nr_cpus = 1;
251#endif
252
253#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
254   if (has_cpuid()) {
255      uint32_t regs[4];
256      uint32_t regs2[4];
257
258      util_cpu_caps.cacheline = 32;
259
260      /* Get max cpuid level */
261      cpuid(0x00000000, regs);
262
263      if (regs[0] >= 0x00000001) {
264         unsigned int cacheline;
265
266         cpuid (0x00000001, regs2);
267
268         util_cpu_caps.x86_cpu_type = (regs2[0] >> 8) & 0xf;
269         if (util_cpu_caps.x86_cpu_type == 0xf)
270             util_cpu_caps.x86_cpu_type = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
271
272         /* general feature flags */
273         util_cpu_caps.has_tsc    = (regs2[3] >>  8) & 1; /* 0x0000010 */
274         util_cpu_caps.has_mmx    = (regs2[3] >> 23) & 1; /* 0x0800000 */
275         util_cpu_caps.has_sse    = (regs2[3] >> 25) & 1; /* 0x2000000 */
276         util_cpu_caps.has_sse2   = (regs2[3] >> 26) & 1; /* 0x4000000 */
277         util_cpu_caps.has_sse3   = (regs2[2] >>  0) & 1; /* 0x0000001 */
278         util_cpu_caps.has_ssse3  = (regs2[2] >>  9) & 1; /* 0x0000020 */
279         util_cpu_caps.has_sse4_1 = (regs2[2] >> 19) & 1;
280         util_cpu_caps.has_sse4_2 = (regs2[2] >> 20) & 1;
281         util_cpu_caps.has_avx    = (regs2[2] >> 28) & 1;
282         util_cpu_caps.has_mmx2   = util_cpu_caps.has_sse; /* SSE cpus supports mmxext too */
283
284         cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
285         if (cacheline > 0)
286            util_cpu_caps.cacheline = cacheline;
287      }
288
289      cpuid(0x80000000, regs);
290
291      if (regs[0] >= 0x80000001) {
292
293         cpuid(0x80000001, regs2);
294
295         util_cpu_caps.has_mmx  |= (regs2[3] >> 23) & 1;
296         util_cpu_caps.has_mmx2 |= (regs2[3] >> 22) & 1;
297         util_cpu_caps.has_3dnow = (regs2[3] >> 31) & 1;
298         util_cpu_caps.has_3dnow_ext = (regs2[3] >> 30) & 1;
299      }
300
301      if (regs[0] >= 0x80000006) {
302         cpuid(0x80000006, regs2);
303         util_cpu_caps.cacheline = regs2[2] & 0xFF;
304      }
305
306      if (!util_cpu_caps.has_sse) {
307         util_cpu_caps.has_sse2 = 0;
308         util_cpu_caps.has_sse3 = 0;
309         util_cpu_caps.has_ssse3 = 0;
310         util_cpu_caps.has_sse4_1 = 0;
311      }
312   }
313#endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
314
315#if defined(PIPE_ARCH_PPC)
316   check_os_altivec_support();
317#endif /* PIPE_ARCH_PPC */
318
319#ifdef DEBUG
320   if (debug_get_option_dump_cpu()) {
321      debug_printf("util_cpu_caps.nr_cpus = %u\n", util_cpu_caps.nr_cpus);
322
323      debug_printf("util_cpu_caps.x86_cpu_type = %u\n", util_cpu_caps.x86_cpu_type);
324      debug_printf("util_cpu_caps.cacheline = %u\n", util_cpu_caps.cacheline);
325
326      debug_printf("util_cpu_caps.has_tsc = %u\n", util_cpu_caps.has_tsc);
327      debug_printf("util_cpu_caps.has_mmx = %u\n", util_cpu_caps.has_mmx);
328      debug_printf("util_cpu_caps.has_mmx2 = %u\n", util_cpu_caps.has_mmx2);
329      debug_printf("util_cpu_caps.has_sse = %u\n", util_cpu_caps.has_sse);
330      debug_printf("util_cpu_caps.has_sse2 = %u\n", util_cpu_caps.has_sse2);
331      debug_printf("util_cpu_caps.has_sse3 = %u\n", util_cpu_caps.has_sse3);
332      debug_printf("util_cpu_caps.has_ssse3 = %u\n", util_cpu_caps.has_ssse3);
333      debug_printf("util_cpu_caps.has_sse4_1 = %u\n", util_cpu_caps.has_sse4_1);
334      debug_printf("util_cpu_caps.has_sse4_2 = %u\n", util_cpu_caps.has_sse4_2);
335      debug_printf("util_cpu_caps.has_avx = %u\n", util_cpu_caps.has_avx);
336      debug_printf("util_cpu_caps.has_3dnow = %u\n", util_cpu_caps.has_3dnow);
337      debug_printf("util_cpu_caps.has_3dnow_ext = %u\n", util_cpu_caps.has_3dnow_ext);
338      debug_printf("util_cpu_caps.has_altivec = %u\n", util_cpu_caps.has_altivec);
339   }
340#endif
341
342   util_cpu_detect_initialized = TRUE;
343}
344