1/*
2 Simple DirectMedia Layer
3 Copyright (C) 1997-2024 Sam Lantinga <slouken@libsdl.org>
4
5 This software is provided 'as-is', without any express or implied
6 warranty. In no event will the authors be held liable for any damages
7 arising from the use of this software.
8
9 Permission is granted to anyone to use this software for any purpose,
10 including commercial applications, and to alter it and redistribute it
11 freely, subject to the following restrictions:
12
13 1. The origin of this software must not be misrepresented; you must not
14 claim that you wrote the original software. If you use this software
15 in a product, an acknowledgment in the product documentation would be
16 appreciated but is not required.
17 2. Altered source versions must be plainly marked as such, and must not be
18 misrepresented as being the original software.
19 3. This notice may not be removed or altered from any source distribution.
20*/
21
22/**
23 * \file SDL_cpuinfo.h
24 *
25 * CPU feature detection for SDL.
26 */
27
28#ifndef SDL_cpuinfo_h_
29#define SDL_cpuinfo_h_
30
31#include "SDL_stdinc.h"
32
33/* Need to do this here because intrin.h has C++ code in it */
34/* Visual Studio 2005 has a bug where intrin.h conflicts with winnt.h */
35#if defined(_MSC_VER) && (_MSC_VER >= 1500) && (defined(_M_IX86) || defined(_M_X64))
36#ifdef __clang__
37/* As of Clang 11, '_m_prefetchw' is conflicting with the winnt.h's version,
38 so we define the needed '_m_prefetch' here as a pseudo-header, until the issue is fixed. */
39
40#ifndef __PRFCHWINTRIN_H
41#define __PRFCHWINTRIN_H
42
43static __inline__ void __attribute__((__always_inline__, __nodebug__))
44_m_prefetch(void *__P)
45{
46 __builtin_prefetch (__P, 0, 3 /* _MM_HINT_T0 */);
47}
48
49#endif /* __PRFCHWINTRIN_H */
50#endif /* __clang__ */
51#include <intrin.h>
52#ifndef _WIN64
53#ifndef __MMX__
54#define __MMX__
55#endif
56#ifndef __3dNOW__
57#define __3dNOW__
58#endif
59#endif
60#ifndef __SSE__
61#define __SSE__
62#endif
63#ifndef __SSE2__
64#define __SSE2__
65#endif
66#ifndef __SSE3__
67#define __SSE3__
68#endif
69#elif defined(__MINGW64_VERSION_MAJOR)
70#include <intrin.h>
71#if !defined(SDL_DISABLE_ARM_NEON_H) && defined(__ARM_NEON)
72# include <arm_neon.h>
73#endif
74#else
75/* altivec.h redefining bool causes a number of problems, see bugs 3993 and 4392, so you need to explicitly define SDL_ENABLE_ALTIVEC_H to have it included. */
76#if defined(HAVE_ALTIVEC_H) && defined(__ALTIVEC__) && !defined(__APPLE_ALTIVEC__) && defined(SDL_ENABLE_ALTIVEC_H)
77#include <altivec.h>
78#endif
79#if !defined(SDL_DISABLE_ARM_NEON_H)
80# if defined(__ARM_NEON)
81# include <arm_neon.h>
82# elif defined(__WINDOWS__) || defined(__WINRT__) || defined(__GDK__)
83/* Visual Studio doesn't define __ARM_ARCH, but _M_ARM (if set, always 7), and _M_ARM64 (if set, always 1). */
84# if defined(_M_ARM)
85# include <armintr.h>
86# include <arm_neon.h>
87# define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
88# endif
89# if defined (_M_ARM64)
90# include <arm64intr.h>
91# include <arm64_neon.h>
92# define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
93# define __ARM_ARCH 8
94# endif
95# endif
96#endif
97#endif /* compiler version */
98
99#if defined(__3dNOW__) && !defined(SDL_DISABLE_MM3DNOW_H)
100#include <mm3dnow.h>
101#endif
102#if defined(__loongarch_sx) && !defined(SDL_DISABLE_LSX_H)
103#include <lsxintrin.h>
104#define __LSX__
105#endif
106#if defined(__loongarch_asx) && !defined(SDL_DISABLE_LASX_H)
107#include <lasxintrin.h>
108#define __LASX__
109#endif
110#if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
111#include <immintrin.h>
112#else
113#if defined(__MMX__) && !defined(SDL_DISABLE_MMINTRIN_H)
114#include <mmintrin.h>
115#endif
116#if defined(__SSE__) && !defined(SDL_DISABLE_XMMINTRIN_H)
117#include <xmmintrin.h>
118#endif
119#if defined(__SSE2__) && !defined(SDL_DISABLE_EMMINTRIN_H)
120#include <emmintrin.h>
121#endif
122#if defined(__SSE3__) && !defined(SDL_DISABLE_PMMINTRIN_H)
123#include <pmmintrin.h>
124#endif
125#endif /* HAVE_IMMINTRIN_H */
126
127#include "begin_code.h"
128/* Set up for C function definitions, even when using C++ */
129#ifdef __cplusplus
130extern "C" {
131#endif
132
133/* This is a guess for the cacheline size used for padding.
134 * Most x86 processors have a 64 byte cache line.
135 * The 64-bit PowerPC processors have a 128 byte cache line.
136 * We'll use the larger value to be generally safe.
137 */
138#define SDL_CACHELINE_SIZE 128
139
140/**
141 * Get the number of CPU cores available.
142 *
143 * \returns the total number of logical CPU cores. On CPUs that include
144 * technologies such as hyperthreading, the number of logical cores
145 * may be more than the number of physical cores.
146 *
147 * \since This function is available since SDL 2.0.0.
148 */
149extern DECLSPEC int SDLCALL SDL_GetCPUCount(void);
150
151/**
152 * Determine the L1 cache line size of the CPU.
153 *
154 * This is useful for determining multi-threaded structure padding or SIMD
155 * prefetch sizes.
156 *
157 * \returns the L1 cache line size of the CPU, in bytes.
158 *
159 * \since This function is available since SDL 2.0.0.
160 */
161extern DECLSPEC int SDLCALL SDL_GetCPUCacheLineSize(void);
162
163/**
164 * Determine whether the CPU has the RDTSC instruction.
165 *
166 * This always returns false on CPUs that aren't using Intel instruction sets.
167 *
168 * \returns SDL_TRUE if the CPU has the RDTSC instruction or SDL_FALSE if not.
169 *
170 * \since This function is available since SDL 2.0.0.
171 *
172 * \sa SDL_Has3DNow
173 * \sa SDL_HasAltiVec
174 * \sa SDL_HasAVX
175 * \sa SDL_HasAVX2
176 * \sa SDL_HasMMX
177 * \sa SDL_HasSSE
178 * \sa SDL_HasSSE2
179 * \sa SDL_HasSSE3
180 * \sa SDL_HasSSE41
181 * \sa SDL_HasSSE42
182 */
183extern DECLSPEC SDL_bool SDLCALL SDL_HasRDTSC(void);
184
185/**
186 * Determine whether the CPU has AltiVec features.
187 *
188 * This always returns false on CPUs that aren't using PowerPC instruction
189 * sets.
190 *
191 * \returns SDL_TRUE if the CPU has AltiVec features or SDL_FALSE if not.
192 *
193 * \since This function is available since SDL 2.0.0.
194 *
195 * \sa SDL_Has3DNow
196 * \sa SDL_HasAVX
197 * \sa SDL_HasAVX2
198 * \sa SDL_HasMMX
199 * \sa SDL_HasRDTSC
200 * \sa SDL_HasSSE
201 * \sa SDL_HasSSE2
202 * \sa SDL_HasSSE3
203 * \sa SDL_HasSSE41
204 * \sa SDL_HasSSE42
205 */
206extern DECLSPEC SDL_bool SDLCALL SDL_HasAltiVec(void);
207
208/**
209 * Determine whether the CPU has MMX features.
210 *
211 * This always returns false on CPUs that aren't using Intel instruction sets.
212 *
213 * \returns SDL_TRUE if the CPU has MMX features or SDL_FALSE if not.
214 *
215 * \since This function is available since SDL 2.0.0.
216 *
217 * \sa SDL_Has3DNow
218 * \sa SDL_HasAltiVec
219 * \sa SDL_HasAVX
220 * \sa SDL_HasAVX2
221 * \sa SDL_HasRDTSC
222 * \sa SDL_HasSSE
223 * \sa SDL_HasSSE2
224 * \sa SDL_HasSSE3
225 * \sa SDL_HasSSE41
226 * \sa SDL_HasSSE42
227 */
228extern DECLSPEC SDL_bool SDLCALL SDL_HasMMX(void);
229
230/**
231 * Determine whether the CPU has 3DNow! features.
232 *
233 * This always returns false on CPUs that aren't using AMD instruction sets.
234 *
235 * \returns SDL_TRUE if the CPU has 3DNow! features or SDL_FALSE if not.
236 *
237 * \since This function is available since SDL 2.0.0.
238 *
239 * \sa SDL_HasAltiVec
240 * \sa SDL_HasAVX
241 * \sa SDL_HasAVX2
242 * \sa SDL_HasMMX
243 * \sa SDL_HasRDTSC
244 * \sa SDL_HasSSE
245 * \sa SDL_HasSSE2
246 * \sa SDL_HasSSE3
247 * \sa SDL_HasSSE41
248 * \sa SDL_HasSSE42
249 */
250extern DECLSPEC SDL_bool SDLCALL SDL_Has3DNow(void);
251
252/**
253 * Determine whether the CPU has SSE features.
254 *
255 * This always returns false on CPUs that aren't using Intel instruction sets.
256 *
257 * \returns SDL_TRUE if the CPU has SSE features or SDL_FALSE if not.
258 *
259 * \since This function is available since SDL 2.0.0.
260 *
261 * \sa SDL_Has3DNow
262 * \sa SDL_HasAltiVec
263 * \sa SDL_HasAVX
264 * \sa SDL_HasAVX2
265 * \sa SDL_HasMMX
266 * \sa SDL_HasRDTSC
267 * \sa SDL_HasSSE2
268 * \sa SDL_HasSSE3
269 * \sa SDL_HasSSE41
270 * \sa SDL_HasSSE42
271 */
272extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE(void);
273
274/**
275 * Determine whether the CPU has SSE2 features.
276 *
277 * This always returns false on CPUs that aren't using Intel instruction sets.
278 *
279 * \returns SDL_TRUE if the CPU has SSE2 features or SDL_FALSE if not.
280 *
281 * \since This function is available since SDL 2.0.0.
282 *
283 * \sa SDL_Has3DNow
284 * \sa SDL_HasAltiVec
285 * \sa SDL_HasAVX
286 * \sa SDL_HasAVX2
287 * \sa SDL_HasMMX
288 * \sa SDL_HasRDTSC
289 * \sa SDL_HasSSE
290 * \sa SDL_HasSSE3
291 * \sa SDL_HasSSE41
292 * \sa SDL_HasSSE42
293 */
294extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE2(void);
295
296/**
297 * Determine whether the CPU has SSE3 features.
298 *
299 * This always returns false on CPUs that aren't using Intel instruction sets.
300 *
301 * \returns SDL_TRUE if the CPU has SSE3 features or SDL_FALSE if not.
302 *
303 * \since This function is available since SDL 2.0.0.
304 *
305 * \sa SDL_Has3DNow
306 * \sa SDL_HasAltiVec
307 * \sa SDL_HasAVX
308 * \sa SDL_HasAVX2
309 * \sa SDL_HasMMX
310 * \sa SDL_HasRDTSC
311 * \sa SDL_HasSSE
312 * \sa SDL_HasSSE2
313 * \sa SDL_HasSSE41
314 * \sa SDL_HasSSE42
315 */
316extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE3(void);
317
318/**
319 * Determine whether the CPU has SSE4.1 features.
320 *
321 * This always returns false on CPUs that aren't using Intel instruction sets.
322 *
323 * \returns SDL_TRUE if the CPU has SSE4.1 features or SDL_FALSE if not.
324 *
325 * \since This function is available since SDL 2.0.0.
326 *
327 * \sa SDL_Has3DNow
328 * \sa SDL_HasAltiVec
329 * \sa SDL_HasAVX
330 * \sa SDL_HasAVX2
331 * \sa SDL_HasMMX
332 * \sa SDL_HasRDTSC
333 * \sa SDL_HasSSE
334 * \sa SDL_HasSSE2
335 * \sa SDL_HasSSE3
336 * \sa SDL_HasSSE42
337 */
338extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE41(void);
339
340/**
341 * Determine whether the CPU has SSE4.2 features.
342 *
343 * This always returns false on CPUs that aren't using Intel instruction sets.
344 *
345 * \returns SDL_TRUE if the CPU has SSE4.2 features or SDL_FALSE if not.
346 *
347 * \since This function is available since SDL 2.0.0.
348 *
349 * \sa SDL_Has3DNow
350 * \sa SDL_HasAltiVec
351 * \sa SDL_HasAVX
352 * \sa SDL_HasAVX2
353 * \sa SDL_HasMMX
354 * \sa SDL_HasRDTSC
355 * \sa SDL_HasSSE
356 * \sa SDL_HasSSE2
357 * \sa SDL_HasSSE3
358 * \sa SDL_HasSSE41
359 */
360extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE42(void);
361
362/**
363 * Determine whether the CPU has AVX features.
364 *
365 * This always returns false on CPUs that aren't using Intel instruction sets.
366 *
367 * \returns SDL_TRUE if the CPU has AVX features or SDL_FALSE if not.
368 *
369 * \since This function is available since SDL 2.0.2.
370 *
371 * \sa SDL_Has3DNow
372 * \sa SDL_HasAltiVec
373 * \sa SDL_HasAVX2
374 * \sa SDL_HasMMX
375 * \sa SDL_HasRDTSC
376 * \sa SDL_HasSSE
377 * \sa SDL_HasSSE2
378 * \sa SDL_HasSSE3
379 * \sa SDL_HasSSE41
380 * \sa SDL_HasSSE42
381 */
382extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX(void);
383
384/**
385 * Determine whether the CPU has AVX2 features.
386 *
387 * This always returns false on CPUs that aren't using Intel instruction sets.
388 *
389 * \returns SDL_TRUE if the CPU has AVX2 features or SDL_FALSE if not.
390 *
391 * \since This function is available since SDL 2.0.4.
392 *
393 * \sa SDL_Has3DNow
394 * \sa SDL_HasAltiVec
395 * \sa SDL_HasAVX
396 * \sa SDL_HasMMX
397 * \sa SDL_HasRDTSC
398 * \sa SDL_HasSSE
399 * \sa SDL_HasSSE2
400 * \sa SDL_HasSSE3
401 * \sa SDL_HasSSE41
402 * \sa SDL_HasSSE42
403 */
404extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX2(void);
405
406/**
407 * Determine whether the CPU has AVX-512F (foundation) features.
408 *
409 * This always returns false on CPUs that aren't using Intel instruction sets.
410 *
411 * \returns SDL_TRUE if the CPU has AVX-512F features or SDL_FALSE if not.
412 *
413 * \since This function is available since SDL 2.0.9.
414 *
415 * \sa SDL_HasAVX
416 */
417extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX512F(void);
418
419/**
420 * Determine whether the CPU has ARM SIMD (ARMv6) features.
421 *
422 * This is different from ARM NEON, which is a different instruction set.
423 *
424 * This always returns false on CPUs that aren't using ARM instruction sets.
425 *
426 * \returns SDL_TRUE if the CPU has ARM SIMD features or SDL_FALSE if not.
427 *
428 * \since This function is available since SDL 2.0.12.
429 *
430 * \sa SDL_HasNEON
431 */
432extern DECLSPEC SDL_bool SDLCALL SDL_HasARMSIMD(void);
433
434/**
435 * Determine whether the CPU has NEON (ARM SIMD) features.
436 *
437 * This always returns false on CPUs that aren't using ARM instruction sets.
438 *
439 * \returns SDL_TRUE if the CPU has ARM NEON features or SDL_FALSE if not.
440 *
441 * \since This function is available since SDL 2.0.6.
442 */
443extern DECLSPEC SDL_bool SDLCALL SDL_HasNEON(void);
444
445/**
446 * Determine whether the CPU has LSX (LOONGARCH SIMD) features.
447 *
448 * This always returns false on CPUs that aren't using LOONGARCH instruction
449 * sets.
450 *
451 * \returns SDL_TRUE if the CPU has LOONGARCH LSX features or SDL_FALSE if
452 * not.
453 *
454 * \since This function is available since SDL 2.24.0.
455 */
456extern DECLSPEC SDL_bool SDLCALL SDL_HasLSX(void);
457
458/**
459 * Determine whether the CPU has LASX (LOONGARCH SIMD) features.
460 *
461 * This always returns false on CPUs that aren't using LOONGARCH instruction
462 * sets.
463 *
464 * \returns SDL_TRUE if the CPU has LOONGARCH LASX features or SDL_FALSE if
465 * not.
466 *
467 * \since This function is available since SDL 2.24.0.
468 */
469extern DECLSPEC SDL_bool SDLCALL SDL_HasLASX(void);
470
471/**
472 * Get the amount of RAM configured in the system.
473 *
474 * \returns the amount of RAM configured in the system in MiB.
475 *
476 * \since This function is available since SDL 2.0.1.
477 */
478extern DECLSPEC int SDLCALL SDL_GetSystemRAM(void);
479
480/**
481 * Report the alignment this system needs for SIMD allocations.
482 *
483 * This will return the minimum number of bytes to which a pointer must be
484 * aligned to be compatible with SIMD instructions on the current machine. For
485 * example, if the machine supports SSE only, it will return 16, but if it
486 * supports AVX-512F, it'll return 64 (etc). This only reports values for
487 * instruction sets SDL knows about, so if your SDL build doesn't have
488 * SDL_HasAVX512F(), then it might return 16 for the SSE support it sees and
489 * not 64 for the AVX-512 instructions that exist but SDL doesn't know about.
490 * Plan accordingly.
491 *
492 * \returns the alignment in bytes needed for available, known SIMD
493 * instructions.
494 *
495 * \since This function is available since SDL 2.0.10.
496 */
497extern DECLSPEC size_t SDLCALL SDL_SIMDGetAlignment(void);
498
499/**
500 * Allocate memory in a SIMD-friendly way.
501 *
502 * This will allocate a block of memory that is suitable for use with SIMD
503 * instructions. Specifically, it will be properly aligned and padded for the
504 * system's supported vector instructions.
505 *
506 * The memory returned will be padded such that it is safe to read or write an
507 * incomplete vector at the end of the memory block. This can be useful so you
508 * don't have to drop back to a scalar fallback at the end of your SIMD
509 * processing loop to deal with the final elements without overflowing the
510 * allocated buffer.
511 *
512 * You must free this memory with SDL_FreeSIMD(), not free() or SDL_free() or
513 * delete[], etc.
514 *
515 * Note that SDL will only deal with SIMD instruction sets it is aware of; for
516 * example, SDL 2.0.8 knows that SSE wants 16-byte vectors (SDL_HasSSE()), and
517 * AVX2 wants 32 bytes (SDL_HasAVX2()), but doesn't know that AVX-512 wants
518 * 64. To be clear: if you can't decide to use an instruction set with an
519 * SDL_Has*() function, don't use that instruction set with memory allocated
520 * through here.
521 *
522 * SDL_AllocSIMD(0) will return a non-NULL pointer, assuming the system isn't
523 * out of memory, but you are not allowed to dereference it (because you only
524 * own zero bytes of that buffer).
525 *
526 * \param len The length, in bytes, of the block to allocate. The actual
527 * allocated block might be larger due to padding, etc.
528 * \returns a pointer to the newly-allocated block, NULL if out of memory.
529 *
530 * \since This function is available since SDL 2.0.10.
531 *
532 * \sa SDL_SIMDGetAlignment
533 * \sa SDL_SIMDRealloc
534 * \sa SDL_SIMDFree
535 */
536extern DECLSPEC void * SDLCALL SDL_SIMDAlloc(const size_t len);
537
538/**
539 * Reallocate memory obtained from SDL_SIMDAlloc
540 *
541 * It is not valid to use this function on a pointer from anything but
542 * SDL_SIMDAlloc(). It can't be used on pointers from malloc, realloc,
543 * SDL_malloc, memalign, new[], etc.
544 *
545 * \param mem The pointer obtained from SDL_SIMDAlloc. This function also
546 * accepts NULL, at which point this function is the same as
547 * calling SDL_SIMDAlloc with a NULL pointer.
548 * \param len The length, in bytes, of the block to allocated. The actual
549 * allocated block might be larger due to padding, etc. Passing 0
550 * will return a non-NULL pointer, assuming the system isn't out of
551 * memory.
552 * \returns a pointer to the newly-reallocated block, NULL if out of memory.
553 *
554 * \since This function is available since SDL 2.0.14.
555 *
556 * \sa SDL_SIMDGetAlignment
557 * \sa SDL_SIMDAlloc
558 * \sa SDL_SIMDFree
559 */
560extern DECLSPEC void * SDLCALL SDL_SIMDRealloc(void *mem, const size_t len);
561
562/**
563 * Deallocate memory obtained from SDL_SIMDAlloc
564 *
565 * It is not valid to use this function on a pointer from anything but
566 * SDL_SIMDAlloc() or SDL_SIMDRealloc(). It can't be used on pointers from
567 * malloc, realloc, SDL_malloc, memalign, new[], etc.
568 *
569 * However, SDL_SIMDFree(NULL) is a legal no-op.
570 *
571 * The memory pointed to by `ptr` is no longer valid for access upon return,
572 * and may be returned to the system or reused by a future allocation. The
573 * pointer passed to this function is no longer safe to dereference once this
574 * function returns, and should be discarded.
575 *
576 * \param ptr The pointer, returned from SDL_SIMDAlloc or SDL_SIMDRealloc, to
577 * deallocate. NULL is a legal no-op.
578 *
579 * \since This function is available since SDL 2.0.10.
580 *
581 * \sa SDL_SIMDAlloc
582 * \sa SDL_SIMDRealloc
583 */
584extern DECLSPEC void SDLCALL SDL_SIMDFree(void *ptr);
585
586/* Ends C function definitions when using C++ */
587#ifdef __cplusplus
588}
589#endif
590#include "close_code.h"
591
592#endif /* SDL_cpuinfo_h_ */
593
594/* vi: set ts=4 sw=4 expandtab: */
595