| 1 | /* |
| 2 | SDL - Simple DirectMedia Layer |
| 3 | Copyright (C) 1997-2009 Sam Lantinga |
| 4 | |
| 5 | This library is free software; you can redistribute it and/or |
| 6 | modify it under the terms of the GNU Lesser General Public |
| 7 | License as published by the Free Software Foundation; either |
| 8 | version 2.1 of the License, or (at your option) any later version. |
| 9 | |
| 10 | This library is distributed in the hope that it will be useful, |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | Lesser General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU Lesser General Public |
| 16 | License along with this library; if not, write to the Free Software |
| 17 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 18 | |
| 19 | Sam Lantinga |
| 20 | slouken@libsdl.org |
| 21 | */ |
| 22 | #include "SDL_config.h" |
| 23 | |
| 24 | /* CPU feature detection for SDL */ |
| 25 | |
| 26 | #include "SDL.h" |
| 27 | #include "SDL_cpuinfo.h" |
| 28 | |
| 29 | #if defined(__MACOSX__) && defined(__ppc__) |
| 30 | #include <sys/sysctl.h> /* For AltiVec check */ |
| 31 | #elif SDL_ALTIVEC_BLITTERS && HAVE_SETJMP |
| 32 | #include <signal.h> |
| 33 | #include <setjmp.h> |
| 34 | #endif |
| 35 | |
| 36 | #define CPU_HAS_RDTSC 0x00000001 |
| 37 | #define CPU_HAS_MMX 0x00000002 |
| 38 | #define CPU_HAS_MMXEXT 0x00000004 |
| 39 | #define CPU_HAS_3DNOW 0x00000010 |
| 40 | #define CPU_HAS_3DNOWEXT 0x00000020 |
| 41 | #define CPU_HAS_SSE 0x00000040 |
| 42 | #define CPU_HAS_SSE2 0x00000080 |
| 43 | #define CPU_HAS_ALTIVEC 0x00000100 |
| 44 | |
| 45 | #if SDL_ALTIVEC_BLITTERS && HAVE_SETJMP && !__MACOSX__ |
| 46 | /* This is the brute force way of detecting instruction sets... |
| 47 | the idea is borrowed from the libmpeg2 library - thanks! |
| 48 | */ |
| 49 | static jmp_buf jmpbuf; |
| 50 | static void illegal_instruction(int sig) |
| 51 | { |
| 52 | longjmp(jmpbuf, 1); |
| 53 | } |
| 54 | #endif /* HAVE_SETJMP */ |
| 55 | |
| 56 | static __inline__ int CPU_haveCPUID(void) |
| 57 | { |
| 58 | int has_CPUID = 0; |
| 59 | #if defined(__GNUC__) && defined(i386) |
| 60 | __asm__ ( |
| 61 | " pushfl # Get original EFLAGS \n" |
| 62 | " popl %%eax \n" |
| 63 | " movl %%eax,%%ecx \n" |
| 64 | " xorl $0x200000,%%eax # Flip ID bit in EFLAGS \n" |
| 65 | " pushl %%eax # Save new EFLAGS value on stack \n" |
| 66 | " popfl # Replace current EFLAGS value \n" |
| 67 | " pushfl # Get new EFLAGS \n" |
| 68 | " popl %%eax # Store new EFLAGS in EAX \n" |
| 69 | " xorl %%ecx,%%eax # Can not toggle ID bit, \n" |
| 70 | " jz 1f # Processor=80486 \n" |
| 71 | " movl $1,%0 # We have CPUID support \n" |
| 72 | "1: \n" |
| 73 | : "=m" (has_CPUID) |
| 74 | : |
| 75 | : "%eax", "%ecx" |
| 76 | ); |
| 77 | #elif defined(__GNUC__) && defined(__x86_64__) |
| 78 | /* Technically, if this is being compiled under __x86_64__ then it has |
| 79 | CPUid by definition. But it's nice to be able to prove it. :) */ |
| 80 | __asm__ ( |
| 81 | " pushfq # Get original EFLAGS \n" |
| 82 | " popq %%rax \n" |
| 83 | " movq %%rax,%%rcx \n" |
| 84 | " xorl $0x200000,%%eax # Flip ID bit in EFLAGS \n" |
| 85 | " pushq %%rax # Save new EFLAGS value on stack \n" |
| 86 | " popfq # Replace current EFLAGS value \n" |
| 87 | " pushfq # Get new EFLAGS \n" |
| 88 | " popq %%rax # Store new EFLAGS in EAX \n" |
| 89 | " xorl %%ecx,%%eax # Can not toggle ID bit, \n" |
| 90 | " jz 1f # Processor=80486 \n" |
| 91 | " movl $1,%0 # We have CPUID support \n" |
| 92 | "1: \n" |
| 93 | : "=m" (has_CPUID) |
| 94 | : |
| 95 | : "%rax", "%rcx" |
| 96 | ); |
| 97 | #elif (defined(_MSC_VER) && defined(_M_IX86)) || defined(__WATCOMC__) |
| 98 | __asm { |
| 99 | pushfd ; Get original EFLAGS |
| 100 | pop eax |
| 101 | mov ecx, eax |
| 102 | xor eax, 200000h ; Flip ID bit in EFLAGS |
| 103 | push eax ; Save new EFLAGS value on stack |
| 104 | popfd ; Replace current EFLAGS value |
| 105 | pushfd ; Get new EFLAGS |
| 106 | pop eax ; Store new EFLAGS in EAX |
| 107 | xor eax, ecx ; Can not toggle ID bit, |
| 108 | jz done ; Processor=80486 |
| 109 | mov has_CPUID,1 ; We have CPUID support |
| 110 | done: |
| 111 | } |
| 112 | #elif defined(__sun) && defined(__i386) |
| 113 | __asm ( |
| 114 | " pushfl \n" |
| 115 | " popl %eax \n" |
| 116 | " movl %eax,%ecx \n" |
| 117 | " xorl $0x200000,%eax \n" |
| 118 | " pushl %eax \n" |
| 119 | " popfl \n" |
| 120 | " pushfl \n" |
| 121 | " popl %eax \n" |
| 122 | " xorl %ecx,%eax \n" |
| 123 | " jz 1f \n" |
| 124 | " movl $1,-8(%ebp) \n" |
| 125 | "1: \n" |
| 126 | ); |
| 127 | #elif defined(__sun) && defined(__amd64) |
| 128 | __asm ( |
| 129 | " pushfq \n" |
| 130 | " popq %rax \n" |
| 131 | " movq %rax,%rcx \n" |
| 132 | " xorl $0x200000,%eax \n" |
| 133 | " pushq %rax \n" |
| 134 | " popfq \n" |
| 135 | " pushfq \n" |
| 136 | " popq %rax \n" |
| 137 | " xorl %ecx,%eax \n" |
| 138 | " jz 1f \n" |
| 139 | " movl $1,-8(%rbp) \n" |
| 140 | "1: \n" |
| 141 | ); |
| 142 | #endif |
| 143 | return has_CPUID; |
| 144 | } |
| 145 | |
| 146 | static __inline__ int CPU_getCPUIDFeatures(void) |
| 147 | { |
| 148 | int features = 0; |
| 149 | #if defined(__GNUC__) && defined(i386) |
| 150 | __asm__ ( |
| 151 | " movl %%ebx,%%edi\n" |
| 152 | " xorl %%eax,%%eax # Set up for CPUID instruction \n" |
| 153 | " cpuid # Get and save vendor ID \n" |
| 154 | " cmpl $1,%%eax # Make sure 1 is valid input for CPUID\n" |
| 155 | " jl 1f # We dont have the CPUID instruction\n" |
| 156 | " xorl %%eax,%%eax \n" |
| 157 | " incl %%eax \n" |
| 158 | " cpuid # Get family/model/stepping/features\n" |
| 159 | " movl %%edx,%0 \n" |
| 160 | "1: \n" |
| 161 | " movl %%edi,%%ebx\n" |
| 162 | : "=m" (features) |
| 163 | : |
| 164 | : "%eax", "%ecx", "%edx", "%edi" |
| 165 | ); |
| 166 | #elif defined(__GNUC__) && defined(__x86_64__) |
| 167 | __asm__ ( |
| 168 | " movq %%rbx,%%rdi\n" |
| 169 | " xorl %%eax,%%eax # Set up for CPUID instruction \n" |
| 170 | " cpuid # Get and save vendor ID \n" |
| 171 | " cmpl $1,%%eax # Make sure 1 is valid input for CPUID\n" |
| 172 | " jl 1f # We dont have the CPUID instruction\n" |
| 173 | " xorl %%eax,%%eax \n" |
| 174 | " incl %%eax \n" |
| 175 | " cpuid # Get family/model/stepping/features\n" |
| 176 | " movl %%edx,%0 \n" |
| 177 | "1: \n" |
| 178 | " movq %%rdi,%%rbx\n" |
| 179 | : "=m" (features) |
| 180 | : |
| 181 | : "%rax", "%rbx", "%rcx", "%rdx", "%rdi" |
| 182 | ); |
| 183 | #elif (defined(_MSC_VER) && defined(_M_IX86)) || defined(__WATCOMC__) |
| 184 | __asm { |
| 185 | xor eax, eax ; Set up for CPUID instruction |
| 186 | cpuid ; Get and save vendor ID |
| 187 | cmp eax, 1 ; Make sure 1 is valid input for CPUID |
| 188 | jl done ; We dont have the CPUID instruction |
| 189 | xor eax, eax |
| 190 | inc eax |
| 191 | cpuid ; Get family/model/stepping/features |
| 192 | mov features, edx |
| 193 | done: |
| 194 | } |
| 195 | #elif defined(__sun) && (defined(__i386) || defined(__amd64)) |
| 196 | __asm( |
| 197 | " movl %ebx,%edi\n" |
| 198 | " xorl %eax,%eax \n" |
| 199 | " cpuid \n" |
| 200 | " cmpl $1,%eax \n" |
| 201 | " jl 1f \n" |
| 202 | " xorl %eax,%eax \n" |
| 203 | " incl %eax \n" |
| 204 | " cpuid \n" |
| 205 | #ifdef __i386 |
| 206 | " movl %edx,-8(%ebp) \n" |
| 207 | #else |
| 208 | " movl %edx,-8(%rbp) \n" |
| 209 | #endif |
| 210 | "1: \n" |
| 211 | " movl %edi,%ebx\n" ); |
| 212 | #endif |
| 213 | return features; |
| 214 | } |
| 215 | |
| 216 | static __inline__ int CPU_getCPUIDFeaturesExt(void) |
| 217 | { |
| 218 | int features = 0; |
| 219 | #if defined(__GNUC__) && defined(i386) |
| 220 | __asm__ ( |
| 221 | " movl %%ebx,%%edi\n" |
| 222 | " movl $0x80000000,%%eax # Query for extended functions \n" |
| 223 | " cpuid # Get extended function limit \n" |
| 224 | " cmpl $0x80000001,%%eax \n" |
| 225 | " jl 1f # Nope, we dont have function 800000001h\n" |
| 226 | " movl $0x80000001,%%eax # Setup extended function 800000001h\n" |
| 227 | " cpuid # and get the information \n" |
| 228 | " movl %%edx,%0 \n" |
| 229 | "1: \n" |
| 230 | " movl %%edi,%%ebx\n" |
| 231 | : "=m" (features) |
| 232 | : |
| 233 | : "%eax", "%ecx", "%edx", "%edi" |
| 234 | ); |
| 235 | #elif defined(__GNUC__) && defined (__x86_64__) |
| 236 | __asm__ ( |
| 237 | " movq %%rbx,%%rdi\n" |
| 238 | " movl $0x80000000,%%eax # Query for extended functions \n" |
| 239 | " cpuid # Get extended function limit \n" |
| 240 | " cmpl $0x80000001,%%eax \n" |
| 241 | " jl 1f # Nope, we dont have function 800000001h\n" |
| 242 | " movl $0x80000001,%%eax # Setup extended function 800000001h\n" |
| 243 | " cpuid # and get the information \n" |
| 244 | " movl %%edx,%0 \n" |
| 245 | "1: \n" |
| 246 | " movq %%rdi,%%rbx\n" |
| 247 | : "=m" (features) |
| 248 | : |
| 249 | : "%rax", "%rbx", "%rcx", "%rdx", "%rdi" |
| 250 | ); |
| 251 | #elif (defined(_MSC_VER) && defined(_M_IX86)) || defined(__WATCOMC__) |
| 252 | __asm { |
| 253 | mov eax,80000000h ; Query for extended functions |
| 254 | cpuid ; Get extended function limit |
| 255 | cmp eax,80000001h |
| 256 | jl done ; Nope, we dont have function 800000001h |
| 257 | mov eax,80000001h ; Setup extended function 800000001h |
| 258 | cpuid ; and get the information |
| 259 | mov features,edx |
| 260 | done: |
| 261 | } |
| 262 | #elif defined(__sun) && ( defined(__i386) || defined(__amd64) ) |
| 263 | __asm ( |
| 264 | " movl %ebx,%edi\n" |
| 265 | " movl $0x80000000,%eax \n" |
| 266 | " cpuid \n" |
| 267 | " cmpl $0x80000001,%eax \n" |
| 268 | " jl 1f \n" |
| 269 | " movl $0x80000001,%eax \n" |
| 270 | " cpuid \n" |
| 271 | #ifdef __i386 |
| 272 | " movl %edx,-8(%ebp) \n" |
| 273 | #else |
| 274 | " movl %edx,-8(%rbp) \n" |
| 275 | #endif |
| 276 | "1: \n" |
| 277 | " movl %edi,%ebx\n" |
| 278 | ); |
| 279 | #endif |
| 280 | return features; |
| 281 | } |
| 282 | |
| 283 | static __inline__ int CPU_haveRDTSC(void) |
| 284 | { |
| 285 | if ( CPU_haveCPUID() ) { |
| 286 | return (CPU_getCPUIDFeatures() & 0x00000010); |
| 287 | } |
| 288 | return 0; |
| 289 | } |
| 290 | |
| 291 | static __inline__ int CPU_haveMMX(void) |
| 292 | { |
| 293 | if ( CPU_haveCPUID() ) { |
| 294 | return (CPU_getCPUIDFeatures() & 0x00800000); |
| 295 | } |
| 296 | return 0; |
| 297 | } |
| 298 | |
| 299 | static __inline__ int CPU_haveMMXExt(void) |
| 300 | { |
| 301 | if ( CPU_haveCPUID() ) { |
| 302 | return (CPU_getCPUIDFeaturesExt() & 0x00400000); |
| 303 | } |
| 304 | return 0; |
| 305 | } |
| 306 | |
| 307 | static __inline__ int CPU_have3DNow(void) |
| 308 | { |
| 309 | if ( CPU_haveCPUID() ) { |
| 310 | return (CPU_getCPUIDFeaturesExt() & 0x80000000); |
| 311 | } |
| 312 | return 0; |
| 313 | } |
| 314 | |
| 315 | static __inline__ int CPU_have3DNowExt(void) |
| 316 | { |
| 317 | if ( CPU_haveCPUID() ) { |
| 318 | return (CPU_getCPUIDFeaturesExt() & 0x40000000); |
| 319 | } |
| 320 | return 0; |
| 321 | } |
| 322 | |
| 323 | static __inline__ int CPU_haveSSE(void) |
| 324 | { |
| 325 | if ( CPU_haveCPUID() ) { |
| 326 | return (CPU_getCPUIDFeatures() & 0x02000000); |
| 327 | } |
| 328 | return 0; |
| 329 | } |
| 330 | |
| 331 | static __inline__ int CPU_haveSSE2(void) |
| 332 | { |
| 333 | if ( CPU_haveCPUID() ) { |
| 334 | return (CPU_getCPUIDFeatures() & 0x04000000); |
| 335 | } |
| 336 | return 0; |
| 337 | } |
| 338 | |
| 339 | static __inline__ int CPU_haveAltiVec(void) |
| 340 | { |
| 341 | volatile int altivec = 0; |
| 342 | #if defined(__MACOSX__) && defined(__ppc__) |
| 343 | int selectors[2] = { CTL_HW, HW_VECTORUNIT }; |
| 344 | int hasVectorUnit = 0; |
| 345 | size_t length = sizeof(hasVectorUnit); |
| 346 | int error = sysctl(selectors, 2, &hasVectorUnit, &length, NULL, 0); |
| 347 | if( 0 == error ) |
| 348 | altivec = (hasVectorUnit != 0); |
| 349 | #elif SDL_ALTIVEC_BLITTERS && HAVE_SETJMP |
| 350 | void (*handler)(int sig); |
| 351 | handler = signal(SIGILL, illegal_instruction); |
| 352 | if ( setjmp(jmpbuf) == 0 ) { |
| 353 | asm volatile ("mtspr 256, %0\n\t" |
| 354 | "vand %%v0, %%v0, %%v0" |
| 355 | : |
| 356 | : "r" (-1)); |
| 357 | altivec = 1; |
| 358 | } |
| 359 | signal(SIGILL, handler); |
| 360 | #endif |
| 361 | return altivec; |
| 362 | } |
| 363 | |
| 364 | static Uint32 SDL_CPUFeatures = 0xFFFFFFFF; |
| 365 | |
| 366 | static Uint32 SDL_GetCPUFeatures(void) |
| 367 | { |
| 368 | if ( SDL_CPUFeatures == 0xFFFFFFFF ) { |
| 369 | SDL_CPUFeatures = 0; |
| 370 | if ( CPU_haveRDTSC() ) { |
| 371 | SDL_CPUFeatures |= CPU_HAS_RDTSC; |
| 372 | } |
| 373 | if ( CPU_haveMMX() ) { |
| 374 | SDL_CPUFeatures |= CPU_HAS_MMX; |
| 375 | } |
| 376 | if ( CPU_haveMMXExt() ) { |
| 377 | SDL_CPUFeatures |= CPU_HAS_MMXEXT; |
| 378 | } |
| 379 | if ( CPU_have3DNow() ) { |
| 380 | SDL_CPUFeatures |= CPU_HAS_3DNOW; |
| 381 | } |
| 382 | if ( CPU_have3DNowExt() ) { |
| 383 | SDL_CPUFeatures |= CPU_HAS_3DNOWEXT; |
| 384 | } |
| 385 | if ( CPU_haveSSE() ) { |
| 386 | SDL_CPUFeatures |= CPU_HAS_SSE; |
| 387 | } |
| 388 | if ( CPU_haveSSE2() ) { |
| 389 | SDL_CPUFeatures |= CPU_HAS_SSE2; |
| 390 | } |
| 391 | if ( CPU_haveAltiVec() ) { |
| 392 | SDL_CPUFeatures |= CPU_HAS_ALTIVEC; |
| 393 | } |
| 394 | } |
| 395 | return SDL_CPUFeatures; |
| 396 | } |
| 397 | |
| 398 | SDL_bool SDL_HasRDTSC(void) |
| 399 | { |
| 400 | if ( SDL_GetCPUFeatures() & CPU_HAS_RDTSC ) { |
| 401 | return SDL_TRUE; |
| 402 | } |
| 403 | return SDL_FALSE; |
| 404 | } |
| 405 | |
| 406 | SDL_bool SDL_HasMMX(void) |
| 407 | { |
| 408 | if ( SDL_GetCPUFeatures() & CPU_HAS_MMX ) { |
| 409 | return SDL_TRUE; |
| 410 | } |
| 411 | return SDL_FALSE; |
| 412 | } |
| 413 | |
| 414 | SDL_bool SDL_HasMMXExt(void) |
| 415 | { |
| 416 | if ( SDL_GetCPUFeatures() & CPU_HAS_MMXEXT ) { |
| 417 | return SDL_TRUE; |
| 418 | } |
| 419 | return SDL_FALSE; |
| 420 | } |
| 421 | |
| 422 | SDL_bool SDL_Has3DNow(void) |
| 423 | { |
| 424 | if ( SDL_GetCPUFeatures() & CPU_HAS_3DNOW ) { |
| 425 | return SDL_TRUE; |
| 426 | } |
| 427 | return SDL_FALSE; |
| 428 | } |
| 429 | |
| 430 | SDL_bool SDL_Has3DNowExt(void) |
| 431 | { |
| 432 | if ( SDL_GetCPUFeatures() & CPU_HAS_3DNOWEXT ) { |
| 433 | return SDL_TRUE; |
| 434 | } |
| 435 | return SDL_FALSE; |
| 436 | } |
| 437 | |
| 438 | SDL_bool SDL_HasSSE(void) |
| 439 | { |
| 440 | if ( SDL_GetCPUFeatures() & CPU_HAS_SSE ) { |
| 441 | return SDL_TRUE; |
| 442 | } |
| 443 | return SDL_FALSE; |
| 444 | } |
| 445 | |
| 446 | SDL_bool SDL_HasSSE2(void) |
| 447 | { |
| 448 | if ( SDL_GetCPUFeatures() & CPU_HAS_SSE2 ) { |
| 449 | return SDL_TRUE; |
| 450 | } |
| 451 | return SDL_FALSE; |
| 452 | } |
| 453 | |
| 454 | SDL_bool SDL_HasAltiVec(void) |
| 455 | { |
| 456 | if ( SDL_GetCPUFeatures() & CPU_HAS_ALTIVEC ) { |
| 457 | return SDL_TRUE; |
| 458 | } |
| 459 | return SDL_FALSE; |
| 460 | } |
| 461 | |
| 462 | #ifdef TEST_MAIN |
| 463 | |
| 464 | #include <stdio.h> |
| 465 | |
| 466 | int main() |
| 467 | { |
| 468 | printf("RDTSC: %d\n", SDL_HasRDTSC()); |
| 469 | printf("MMX: %d\n", SDL_HasMMX()); |
| 470 | printf("MMXExt: %d\n", SDL_HasMMXExt()); |
| 471 | printf("3DNow: %d\n", SDL_Has3DNow()); |
| 472 | printf("3DNowExt: %d\n", SDL_Has3DNowExt()); |
| 473 | printf("SSE: %d\n", SDL_HasSSE()); |
| 474 | printf("SSE2: %d\n", SDL_HasSSE2()); |
| 475 | printf("AltiVec: %d\n", SDL_HasAltiVec()); |
| 476 | return 0; |
| 477 | } |
| 478 | |
| 479 | #endif /* TEST_MAIN */ |