Go to most recent revision | Details | Last modification | View Log | RSS feed
Rev | Author | Line No. | Line |
---|---|---|---|
1901 | serge | 1 | /* |
2 | * Mesa 3-D graphics library |
||
3 | * Version: 6.5.1 |
||
4 | * |
||
5 | * Copyright (C) 1999-2006 Brian Paul All Rights Reserved. |
||
6 | * |
||
7 | * Permission is hereby granted, free of charge, to any person obtaining a |
||
8 | * copy of this software and associated documentation files (the "Software"), |
||
9 | * to deal in the Software without restriction, including without limitation |
||
10 | * the rights to use, copy, modify, merge, publish, distribute, sublicense, |
||
11 | * and/or sell copies of the Software, and to permit persons to whom the |
||
12 | * Software is furnished to do so, subject to the following conditions: |
||
13 | * |
||
14 | * The above copyright notice and this permission notice shall be included |
||
15 | * in all copies or substantial portions of the Software. |
||
16 | * |
||
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS |
||
18 | * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
||
19 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL |
||
20 | * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN |
||
21 | * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN |
||
22 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. |
||
23 | */ |
||
24 | |||
25 | /** |
||
26 | * \file common_x86.c |
||
27 | * |
||
28 | * Check CPU capabilities & initialize optimized funtions for this particular |
||
29 | * processor. |
||
30 | * |
||
31 | * Changed by Andre Werthmann for using the new SSE functions. |
||
32 | * |
||
33 | * \author Holger Waechtler |
||
34 | * \author Andre Werthmann |
||
35 | */ |
||
36 | |||
37 | /* XXX these includes should probably go into imports.h or glheader.h */ |
||
38 | #if defined(USE_SSE_ASM) && defined(__linux__) |
||
39 | #include |
||
40 | #endif |
||
41 | #if defined(USE_SSE_ASM) && defined(__FreeBSD__) |
||
42 | #include |
||
43 | #include |
||
44 | #endif |
||
45 | #if defined(USE_SSE_ASM) && defined(__OpenBSD__) |
||
46 | #include |
||
47 | #include |
||
48 | #include |
||
49 | #endif |
||
50 | |||
51 | #include "main/imports.h" |
||
52 | #include "common_x86_asm.h" |
||
53 | |||
54 | #undef WIN32 |
||
55 | #undef _WIN32 |
||
56 | |||
57 | /** Bitmask of X86_FEATURE_x bits */ |
||
58 | int _mesa_x86_cpu_features = 0x0; |
||
59 | |||
60 | static int detection_debug = GL_FALSE; |
||
61 | |||
62 | /* No reason for this to be public. |
||
63 | */ |
||
64 | extern GLuint _ASMAPI _mesa_x86_has_cpuid(void); |
||
65 | extern void _ASMAPI _mesa_x86_cpuid(GLuint op, GLuint *reg_eax, GLuint *reg_ebx, GLuint *reg_ecx, GLuint *reg_edx); |
||
66 | extern GLuint _ASMAPI _mesa_x86_cpuid_eax(GLuint op); |
||
67 | extern GLuint _ASMAPI _mesa_x86_cpuid_ebx(GLuint op); |
||
68 | extern GLuint _ASMAPI _mesa_x86_cpuid_ecx(GLuint op); |
||
69 | extern GLuint _ASMAPI _mesa_x86_cpuid_edx(GLuint op); |
||
70 | |||
71 | |||
72 | #if defined(USE_SSE_ASM) |
||
73 | /* |
||
74 | * We must verify that the Streaming SIMD Extensions are truly supported |
||
75 | * on this processor before we go ahead and hook out the optimized code. |
||
76 | * |
||
77 | * However, I have been told by Alan Cox that all 2.4 (and later) Linux |
||
78 | * kernels provide full SSE support on all processors that expose SSE via |
||
79 | * the CPUID mechanism. |
||
80 | */ |
||
81 | |||
82 | /* These are assembly functions: */ |
||
83 | extern void _mesa_test_os_sse_support( void ); |
||
84 | extern void _mesa_test_os_sse_exception_support( void ); |
||
85 | |||
86 | |||
87 | #if defined(WIN32) |
||
88 | #ifndef STATUS_FLOAT_MULTIPLE_TRAPS |
||
89 | # define STATUS_FLOAT_MULTIPLE_TRAPS (0xC00002B5L) |
||
90 | #endif |
||
91 | static LONG WINAPI ExceptionFilter(LPEXCEPTION_POINTERS exp) |
||
92 | { |
||
93 | PEXCEPTION_RECORD rec = exp->ExceptionRecord; |
||
94 | PCONTEXT ctx = exp->ContextRecord; |
||
95 | |||
96 | if ( rec->ExceptionCode == EXCEPTION_ILLEGAL_INSTRUCTION ) { |
||
97 | _mesa_debug(NULL, "EXCEPTION_ILLEGAL_INSTRUCTION\n" ); |
||
98 | _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM); |
||
99 | } else if ( rec->ExceptionCode == STATUS_FLOAT_MULTIPLE_TRAPS ) { |
||
100 | _mesa_debug(NULL, "STATUS_FLOAT_MULTIPLE_TRAPS\n"); |
||
101 | /* Windows seems to clear the exception flag itself, we just have to increment Eip */ |
||
102 | } else { |
||
103 | _mesa_debug(NULL, "UNEXPECTED EXCEPTION (0x%08x), terminating!\n" ); |
||
104 | return EXCEPTION_EXECUTE_HANDLER; |
||
105 | } |
||
106 | |||
107 | if ( (ctx->ContextFlags & CONTEXT_CONTROL) != CONTEXT_CONTROL ) { |
||
108 | _mesa_debug(NULL, "Context does not contain control registers, terminating!\n"); |
||
109 | return EXCEPTION_EXECUTE_HANDLER; |
||
110 | } |
||
111 | ctx->Eip += 3; |
||
112 | |||
113 | return EXCEPTION_CONTINUE_EXECUTION; |
||
114 | } |
||
115 | #endif /* WIN32 */ |
||
116 | |||
117 | |||
118 | /** |
||
119 | * Check if SSE is supported. |
||
120 | * If not, turn off the X86_FEATURE_XMM flag in _mesa_x86_cpu_features. |
||
121 | */ |
||
122 | void _mesa_check_os_sse_support( void ) |
||
123 | { |
||
124 | #if defined(__FreeBSD__) |
||
125 | { |
||
126 | int ret, enabled; |
||
127 | unsigned int len; |
||
128 | len = sizeof(enabled); |
||
129 | ret = sysctlbyname("hw.instruction_sse", &enabled, &len, NULL, 0); |
||
130 | if (ret || !enabled) |
||
131 | _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM); |
||
132 | } |
||
133 | #elif defined (__NetBSD__) |
||
134 | { |
||
135 | int ret, enabled; |
||
136 | size_t len = sizeof(enabled); |
||
137 | ret = sysctlbyname("machdep.sse", &enabled, &len, (void *)NULL, 0); |
||
138 | if (ret || !enabled) |
||
139 | _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM); |
||
140 | } |
||
141 | #elif defined(__OpenBSD__) |
||
142 | { |
||
143 | int mib[2]; |
||
144 | int ret, enabled; |
||
145 | size_t len = sizeof(enabled); |
||
146 | |||
147 | mib[0] = CTL_MACHDEP; |
||
148 | mib[1] = CPU_SSE; |
||
149 | |||
150 | ret = sysctl(mib, 2, &enabled, &len, NULL, 0); |
||
151 | if (ret || !enabled) |
||
152 | _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM); |
||
153 | } |
||
154 | #elif defined(WIN32) |
||
155 | LPTOP_LEVEL_EXCEPTION_FILTER oldFilter; |
||
156 | |||
157 | /* Install our ExceptionFilter */ |
||
158 | oldFilter = SetUnhandledExceptionFilter( ExceptionFilter ); |
||
159 | |||
160 | if ( cpu_has_xmm ) { |
||
161 | _mesa_debug(NULL, "Testing OS support for SSE...\n"); |
||
162 | |||
163 | _mesa_test_os_sse_support(); |
||
164 | |||
165 | if ( cpu_has_xmm ) { |
||
166 | _mesa_debug(NULL, "Yes.\n"); |
||
167 | } else { |
||
168 | _mesa_debug(NULL, "No!\n"); |
||
169 | } |
||
170 | } |
||
171 | |||
172 | if ( cpu_has_xmm ) { |
||
173 | _mesa_debug(NULL, "Testing OS support for SSE unmasked exceptions...\n"); |
||
174 | |||
175 | _mesa_test_os_sse_exception_support(); |
||
176 | |||
177 | if ( cpu_has_xmm ) { |
||
178 | _mesa_debug(NULL, "Yes.\n"); |
||
179 | } else { |
||
180 | _mesa_debug(NULL, "No!\n"); |
||
181 | } |
||
182 | } |
||
183 | |||
184 | /* Restore previous exception filter */ |
||
185 | SetUnhandledExceptionFilter( oldFilter ); |
||
186 | |||
187 | if ( cpu_has_xmm ) { |
||
188 | _mesa_debug(NULL, "Tests of OS support for SSE passed.\n"); |
||
189 | } else { |
||
190 | _mesa_debug(NULL, "Tests of OS support for SSE failed!\n"); |
||
191 | } |
||
192 | #else |
||
193 | /* Do nothing on other platforms for now. |
||
194 | */ |
||
195 | if (detection_debug) |
||
196 | _mesa_debug(NULL, "Not testing OS support for SSE, leaving enabled.\n"); |
||
197 | #endif /* __FreeBSD__ */ |
||
198 | } |
||
199 | |||
200 | #endif /* USE_SSE_ASM */ |
||
201 | |||
202 | |||
203 | /** |
||
204 | * Initialize the _mesa_x86_cpu_features bitfield. |
||
205 | * This is a no-op if called more than once. |
||
206 | */ |
||
207 | void |
||
208 | _mesa_get_x86_features(void) |
||
209 | { |
||
210 | static int called = 0; |
||
211 | |||
212 | if (called) |
||
213 | return; |
||
214 | |||
215 | called = 1; |
||
216 | |||
217 | #ifdef USE_X86_ASM |
||
218 | _mesa_x86_cpu_features = 0x0; |
||
219 | |||
220 | if (_mesa_getenv( "MESA_NO_ASM")) { |
||
221 | return; |
||
222 | } |
||
223 | |||
224 | if (!_mesa_x86_has_cpuid()) { |
||
225 | _mesa_debug(NULL, "CPUID not detected\n"); |
||
226 | } |
||
227 | else { |
||
228 | GLuint cpu_features; |
||
229 | GLuint cpu_ext_features; |
||
230 | GLuint cpu_ext_info; |
||
231 | char cpu_vendor[13]; |
||
232 | GLuint result; |
||
233 | |||
234 | /* get vendor name */ |
||
235 | _mesa_x86_cpuid(0, &result, (GLuint *)(cpu_vendor + 0), (GLuint *)(cpu_vendor + 8), (GLuint *)(cpu_vendor + 4)); |
||
236 | cpu_vendor[12] = '\0'; |
||
237 | |||
238 | if (detection_debug) |
||
239 | _mesa_debug(NULL, "CPU vendor: %s\n", cpu_vendor); |
||
240 | |||
241 | /* get cpu features */ |
||
242 | cpu_features = _mesa_x86_cpuid_edx(1); |
||
243 | |||
244 | if (cpu_features & X86_CPU_FPU) |
||
245 | _mesa_x86_cpu_features |= X86_FEATURE_FPU; |
||
246 | if (cpu_features & X86_CPU_CMOV) |
||
247 | _mesa_x86_cpu_features |= X86_FEATURE_CMOV; |
||
248 | |||
249 | #ifdef USE_MMX_ASM |
||
250 | if (cpu_features & X86_CPU_MMX) |
||
251 | _mesa_x86_cpu_features |= X86_FEATURE_MMX; |
||
252 | #endif |
||
253 | |||
254 | #ifdef USE_SSE_ASM |
||
255 | if (cpu_features & X86_CPU_XMM) |
||
256 | _mesa_x86_cpu_features |= X86_FEATURE_XMM; |
||
257 | if (cpu_features & X86_CPU_XMM2) |
||
258 | _mesa_x86_cpu_features |= X86_FEATURE_XMM2; |
||
259 | #endif |
||
260 | |||
261 | /* query extended cpu features */ |
||
262 | if ((cpu_ext_info = _mesa_x86_cpuid_eax(0x80000000)) > 0x80000000) { |
||
263 | if (cpu_ext_info >= 0x80000001) { |
||
264 | |||
265 | cpu_ext_features = _mesa_x86_cpuid_edx(0x80000001); |
||
266 | |||
267 | if (cpu_features & X86_CPU_MMX) { |
||
268 | |||
269 | #ifdef USE_3DNOW_ASM |
||
270 | if (cpu_ext_features & X86_CPUEXT_3DNOW) |
||
271 | _mesa_x86_cpu_features |= X86_FEATURE_3DNOW; |
||
272 | if (cpu_ext_features & X86_CPUEXT_3DNOW_EXT) |
||
273 | _mesa_x86_cpu_features |= X86_FEATURE_3DNOWEXT; |
||
274 | #endif |
||
275 | |||
276 | #ifdef USE_MMX_ASM |
||
277 | if (cpu_ext_features & X86_CPUEXT_MMX_EXT) |
||
278 | _mesa_x86_cpu_features |= X86_FEATURE_MMXEXT; |
||
279 | #endif |
||
280 | } |
||
281 | } |
||
282 | |||
283 | /* query cpu name */ |
||
284 | if (cpu_ext_info >= 0x80000002) { |
||
285 | GLuint ofs; |
||
286 | char cpu_name[49]; |
||
287 | for (ofs = 0; ofs < 3; ofs++) |
||
288 | _mesa_x86_cpuid(0x80000002+ofs, (GLuint *)(cpu_name + (16*ofs)+0), (GLuint *)(cpu_name + (16*ofs)+4), (GLuint *)(cpu_name + (16*ofs)+8), (GLuint *)(cpu_name + (16*ofs)+12)); |
||
289 | cpu_name[48] = '\0'; /* the name should be NULL terminated, but just to be sure */ |
||
290 | |||
291 | if (detection_debug) |
||
292 | _mesa_debug(NULL, "CPU name: %s\n", cpu_name); |
||
293 | } |
||
294 | } |
||
295 | |||
296 | } |
||
297 | |||
298 | #ifdef USE_MMX_ASM |
||
299 | if ( cpu_has_mmx ) { |
||
300 | if ( _mesa_getenv( "MESA_NO_MMX" ) == 0 ) { |
||
301 | if (detection_debug) |
||
302 | _mesa_debug(NULL, "MMX cpu detected.\n"); |
||
303 | } else { |
||
304 | _mesa_x86_cpu_features &= ~(X86_FEATURE_MMX); |
||
305 | } |
||
306 | } |
||
307 | #endif |
||
308 | |||
309 | #ifdef USE_3DNOW_ASM |
||
310 | if ( cpu_has_3dnow ) { |
||
311 | if ( _mesa_getenv( "MESA_NO_3DNOW" ) == 0 ) { |
||
312 | if (detection_debug) |
||
313 | _mesa_debug(NULL, "3DNow! cpu detected.\n"); |
||
314 | } else { |
||
315 | _mesa_x86_cpu_features &= ~(X86_FEATURE_3DNOW); |
||
316 | } |
||
317 | } |
||
318 | #endif |
||
319 | |||
320 | #ifdef USE_SSE_ASM |
||
321 | if ( cpu_has_xmm ) { |
||
322 | if ( _mesa_getenv( "MESA_NO_SSE" ) == 0 ) { |
||
323 | if (detection_debug) |
||
324 | _mesa_debug(NULL, "SSE cpu detected.\n"); |
||
325 | if ( _mesa_getenv( "MESA_FORCE_SSE" ) == 0 ) { |
||
326 | _mesa_check_os_sse_support(); |
||
327 | } |
||
328 | } else { |
||
329 | _mesa_debug(NULL, "SSE cpu detected, but switched off by user.\n"); |
||
330 | _mesa_x86_cpu_features &= ~(X86_FEATURE_XMM); |
||
331 | } |
||
332 | } |
||
333 | #endif |
||
334 | |||
335 | #endif /* USE_X86_ASM */ |
||
336 | |||
337 | (void) detection_debug; |
||
338 | }> |