1 /*
2   Simple DirectMedia Layer
3   Copyright (C) 1997-2020 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 /* Many of the intrinsics SDL uses are not implemented by clang with Visual Studio */
38 #undef __MMX__
39 #undef __SSE__
40 #undef __SSE2__
41 #else
42 #include <intrin.h>
43 #ifndef _WIN64
44 #ifndef __MMX__
45 #define __MMX__
46 #endif
47 #ifndef __3dNOW__
48 #define __3dNOW__
49 #endif
50 #endif
51 #ifndef __SSE__
52 #define __SSE__
53 #endif
54 #ifndef __SSE2__
55 #define __SSE2__
56 #endif
57 #endif /* __clang__ */
58 #elif defined(__MINGW64_VERSION_MAJOR)
59 #include <intrin.h>
60 #else
61 /* 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. */
62 #if defined(HAVE_ALTIVEC_H) && defined(__ALTIVEC__) && !defined(__APPLE_ALTIVEC__) && defined(SDL_ENABLE_ALTIVEC_H)
63 #include <altivec.h>
64 #endif
65 #if !defined(SDL_DISABLE_ARM_NEON_H)
66 #  if defined(__ARM_NEON)
67 #    include <arm_neon.h>
68 #  elif defined(__WINDOWS__) || defined(__WINRT__)
69 /* Visual Studio doesn't define __ARM_ARCH, but _M_ARM (if set, always 7), and _M_ARM64 (if set, always 1). */
70 #    if defined(_M_ARM)
71 #      include <armintr.h>
72 #      include <arm_neon.h>
73 #      define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
74 #    endif
75 #    if defined (_M_ARM64)
76 #      include <arm64intr.h>
77 #      include <arm64_neon.h>
78 #      define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
79 #    endif
80 #  endif
81 #endif
82 #if defined(__3dNOW__) && !defined(SDL_DISABLE_MM3DNOW_H)
83 #include <mm3dnow.h>
84 #endif
85 #if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
86 #include <immintrin.h>
87 #else
88 #if defined(__MMX__) && !defined(SDL_DISABLE_MMINTRIN_H)
89 #include <mmintrin.h>
90 #endif
91 #if defined(__SSE__) && !defined(SDL_DISABLE_XMMINTRIN_H)
92 #include <xmmintrin.h>
93 #endif
94 #if defined(__SSE2__) && !defined(SDL_DISABLE_EMMINTRIN_H)
95 #include <emmintrin.h>
96 #endif
97 #if defined(__SSE3__) && !defined(SDL_DISABLE_PMMINTRIN_H)
98 #include <pmmintrin.h>
99 #endif
100 #endif /* HAVE_IMMINTRIN_H */
101 #endif /* compiler version */
102 
103 #include "begin_code.h"
104 /* Set up for C function definitions, even when using C++ */
105 #ifdef __cplusplus
106 extern "C" {
107 #endif
108 
109 /* This is a guess for the cacheline size used for padding.
110  * Most x86 processors have a 64 byte cache line.
111  * The 64-bit PowerPC processors have a 128 byte cache line.
112  * We'll use the larger value to be generally safe.
113  */
114 #define SDL_CACHELINE_SIZE  128
115 
116 /**
117  *  This function returns the number of CPU cores available.
118  */
119 extern DECLSPEC int SDLCALL SDL_GetCPUCount(void);
120 
121 /**
122  *  This function returns the L1 cache line size of the CPU
123  *
124  *  This is useful for determining multi-threaded structure padding
125  *  or SIMD prefetch sizes.
126  */
127 extern DECLSPEC int SDLCALL SDL_GetCPUCacheLineSize(void);
128 
129 /**
130  *  This function returns true if the CPU has the RDTSC instruction.
131  */
132 extern DECLSPEC SDL_bool SDLCALL SDL_HasRDTSC(void);
133 
134 /**
135  *  This function returns true if the CPU has AltiVec features.
136  */
137 extern DECLSPEC SDL_bool SDLCALL SDL_HasAltiVec(void);
138 
139 /**
140  *  This function returns true if the CPU has MMX features.
141  */
142 extern DECLSPEC SDL_bool SDLCALL SDL_HasMMX(void);
143 
144 /**
145  *  This function returns true if the CPU has 3DNow! features.
146  */
147 extern DECLSPEC SDL_bool SDLCALL SDL_Has3DNow(void);
148 
149 /**
150  *  This function returns true if the CPU has SSE features.
151  */
152 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE(void);
153 
154 /**
155  *  This function returns true if the CPU has SSE2 features.
156  */
157 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE2(void);
158 
159 /**
160  *  This function returns true if the CPU has SSE3 features.
161  */
162 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE3(void);
163 
164 /**
165  *  This function returns true if the CPU has SSE4.1 features.
166  */
167 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE41(void);
168 
169 /**
170  *  This function returns true if the CPU has SSE4.2 features.
171  */
172 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE42(void);
173 
174 /**
175  *  This function returns true if the CPU has AVX features.
176  */
177 extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX(void);
178 
179 /**
180  *  This function returns true if the CPU has AVX2 features.
181  */
182 extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX2(void);
183 
184 /**
185  *  This function returns true if the CPU has AVX-512F (foundation) features.
186  */
187 extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX512F(void);
188 
189 /**
190  *  This function returns true if the CPU has ARM SIMD (ARMv6) features.
191  */
192 extern DECLSPEC SDL_bool SDLCALL SDL_HasARMSIMD(void);
193 
194 /**
195  *  This function returns true if the CPU has NEON (ARM SIMD) features.
196  */
197 extern DECLSPEC SDL_bool SDLCALL SDL_HasNEON(void);
198 
199 /**
200  *  This function returns the amount of RAM configured in the system, in MB.
201  */
202 extern DECLSPEC int SDLCALL SDL_GetSystemRAM(void);
203 
204 /**
205  * \brief Report the alignment this system needs for SIMD allocations.
206  *
207  * This will return the minimum number of bytes to which a pointer must be
208  *  aligned to be compatible with SIMD instructions on the current machine.
209  *  For example, if the machine supports SSE only, it will return 16, but if
210  *  it supports AVX-512F, it'll return 64 (etc). This only reports values for
211  *  instruction sets SDL knows about, so if your SDL build doesn't have
212  *  SDL_HasAVX512F(), then it might return 16 for the SSE support it sees and
213  *  not 64 for the AVX-512 instructions that exist but SDL doesn't know about.
214  *  Plan accordingly.
215  */
216 extern DECLSPEC size_t SDLCALL SDL_SIMDGetAlignment(void);
217 
218 /**
219  * \brief Allocate memory in a SIMD-friendly way.
220  *
221  * This will allocate a block of memory that is suitable for use with SIMD
222  *  instructions. Specifically, it will be properly aligned and padded for
223  *  the system's supported vector instructions.
224  *
225  * The memory returned will be padded such that it is safe to read or write
226  *  an incomplete vector at the end of the memory block. This can be useful
227  *  so you don't have to drop back to a scalar fallback at the end of your
228  *  SIMD processing loop to deal with the final elements without overflowing
229  *  the allocated buffer.
230  *
231  * You must free this memory with SDL_FreeSIMD(), not free() or SDL_free()
232  *  or delete[], etc.
233  *
234  * Note that SDL will only deal with SIMD instruction sets it is aware of;
235  *  for example, SDL 2.0.8 knows that SSE wants 16-byte vectors
236  *  (SDL_HasSSE()), and AVX2 wants 32 bytes (SDL_HasAVX2()), but doesn't
237  *  know that AVX-512 wants 64. To be clear: if you can't decide to use an
238  *  instruction set with an SDL_Has*() function, don't use that instruction
239  *  set with memory allocated through here.
240  *
241  * SDL_AllocSIMD(0) will return a non-NULL pointer, assuming the system isn't
242  *  out of memory.
243  *
244  *  \param len The length, in bytes, of the block to allocated. The actual
245  *             allocated block might be larger due to padding, etc.
246  * \return Pointer to newly-allocated block, NULL if out of memory.
247  *
248  * \sa SDL_SIMDAlignment
249  * \sa SDL_SIMDRealloc
250  * \sa SDL_SIMDFree
251  */
252 extern DECLSPEC void * SDLCALL SDL_SIMDAlloc(const size_t len);
253 
254 /**
255  * \brief Reallocate memory obtained from SDL_SIMDAlloc
256  *
257  * It is not valid to use this function on a pointer from anything but
258  *  SDL_SIMDAlloc(). It can't be used on pointers from malloc, realloc,
259  *  SDL_malloc, memalign, new[], etc.
260  *
261  *  \param mem The pointer obtained from SDL_SIMDAlloc. This function also
262  *             accepts NULL, at which point this function is the same as
263  *             calling SDL_realloc with a NULL pointer.
264  *  \param len The length, in bytes, of the block to allocated. The actual
265  *             allocated block might be larger due to padding, etc. Passing 0
266  *             will return a non-NULL pointer, assuming the system isn't out of
267  *             memory.
268  * \return Pointer to newly-reallocated block, NULL if out of memory.
269  *
270  * \sa SDL_SIMDAlignment
271  * \sa SDL_SIMDAlloc
272  * \sa SDL_SIMDFree
273  */
274 extern DECLSPEC void * SDLCALL SDL_SIMDRealloc(void *mem, size_t len);
275 
276 /**
277  * \brief Deallocate memory obtained from SDL_SIMDAlloc
278  *
279  * It is not valid to use this function on a pointer from anything but
280  *  SDL_SIMDAlloc(). It can't be used on pointers from malloc, realloc,
281  *  SDL_malloc, memalign, new[], etc.
282  *
283  * However, SDL_SIMDFree(NULL) is a legal no-op.
284  *
285  * \sa SDL_SIMDAlloc
286  * \sa SDL_SIMDRealloc
287  */
288 extern DECLSPEC void SDLCALL SDL_SIMDFree(void *ptr);
289 
290 /* vi: set ts=4 sw=4 expandtab: */
291 /* Ends C function definitions when using C++ */
292 #ifdef __cplusplus
293 }
294 #endif
295 #include "close_code.h"
296 
297 #endif /* SDL_cpuinfo_h_ */
298 
299 /* vi: set ts=4 sw=4 expandtab: */
300