2 Simple DirectMedia Layer
3 Copyright (C) 1997-2019 Sam Lantinga <slouken@libsdl.org>
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.
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:
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.
25 * CPU feature detection for SDL.
28 #ifndef SDL_cpuinfo_h_
29 #define SDL_cpuinfo_h_
31 #include "SDL_stdinc.h"
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))
37 /* Many of the intrinsics SDL uses are not implemented by clang with Visual Studio */
57 #endif /* __clang__ */
58 #elif defined(__MINGW64_VERSION_MAJOR)
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)
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). */
72 # include <arm_neon.h>
73 # define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
75 # if defined (_M_ARM64)
77 # include <arm_neon.h>
78 # define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
82 #if defined(__3dNOW__) && !defined(SDL_DISABLE_MM3DNOW_H)
85 #if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
86 #include <immintrin.h>
88 #if defined(__MMX__) && !defined(SDL_DISABLE_MMINTRIN_H)
91 #if defined(__SSE__) && !defined(SDL_DISABLE_XMMINTRIN_H)
92 #include <xmmintrin.h>
94 #if defined(__SSE2__) && !defined(SDL_DISABLE_EMMINTRIN_H)
95 #include <emmintrin.h>
97 #if defined(__SSE3__) && !defined(SDL_DISABLE_PMMINTRIN_H)
98 #include <pmmintrin.h>
100 #endif /* HAVE_IMMINTRIN_H */
101 #endif /* compiler version */
103 #include "begin_code.h"
104 /* Set up for C function definitions, even when using C++ */
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.
114 #define SDL_CACHELINE_SIZE 128
117 * This function returns the number of CPU cores available.
119 extern DECLSPEC int SDLCALL SDL_GetCPUCount(void);
122 * This function returns the L1 cache line size of the CPU
124 * This is useful for determining multi-threaded structure padding
125 * or SIMD prefetch sizes.
127 extern DECLSPEC int SDLCALL SDL_GetCPUCacheLineSize(void);
130 * This function returns true if the CPU has the RDTSC instruction.
132 extern DECLSPEC SDL_bool SDLCALL SDL_HasRDTSC(void);
135 * This function returns true if the CPU has AltiVec features.
137 extern DECLSPEC SDL_bool SDLCALL SDL_HasAltiVec(void);
140 * This function returns true if the CPU has MMX features.
142 extern DECLSPEC SDL_bool SDLCALL SDL_HasMMX(void);
145 * This function returns true if the CPU has 3DNow! features.
147 extern DECLSPEC SDL_bool SDLCALL SDL_Has3DNow(void);
150 * This function returns true if the CPU has SSE features.
152 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE(void);
155 * This function returns true if the CPU has SSE2 features.
157 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE2(void);
160 * This function returns true if the CPU has SSE3 features.
162 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE3(void);
165 * This function returns true if the CPU has SSE4.1 features.
167 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE41(void);
170 * This function returns true if the CPU has SSE4.2 features.
172 extern DECLSPEC SDL_bool SDLCALL SDL_HasSSE42(void);
175 * This function returns true if the CPU has AVX features.
177 extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX(void);
180 * This function returns true if the CPU has AVX2 features.
182 extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX2(void);
185 * This function returns true if the CPU has AVX-512F (foundation) features.
187 extern DECLSPEC SDL_bool SDLCALL SDL_HasAVX512F(void);
190 * This function returns true if the CPU has NEON (ARM SIMD) features.
192 extern DECLSPEC SDL_bool SDLCALL SDL_HasNEON(void);
195 * This function returns the amount of RAM configured in the system, in MB.
197 extern DECLSPEC int SDLCALL SDL_GetSystemRAM(void);
200 * \brief Report the alignment this system needs for SIMD allocations.
202 * This will return the minimum number of bytes to which a pointer must be
203 * aligned to be compatible with SIMD instructions on the current machine.
204 * For example, if the machine supports SSE only, it will return 16, but if
205 * it supports AVX-512F, it'll return 64 (etc). This only reports values for
206 * instruction sets SDL knows about, so if your SDL build doesn't have
207 * SDL_HasAVX512F(), then it might return 16 for the SSE support it sees and
208 * not 64 for the AVX-512 instructions that exist but SDL doesn't know about.
211 extern DECLSPEC size_t SDLCALL SDL_SIMDGetAlignment(void);
214 * \brief Allocate memory in a SIMD-friendly way.
216 * This will allocate a block of memory that is suitable for use with SIMD
217 * instructions. Specifically, it will be properly aligned and padded for
218 * the system's supported vector instructions.
220 * The memory returned will be padded such that it is safe to read or write
221 * an incomplete vector at the end of the memory block. This can be useful
222 * so you don't have to drop back to a scalar fallback at the end of your
223 * SIMD processing loop to deal with the final elements without overflowing
224 * the allocated buffer.
226 * You must free this memory with SDL_FreeSIMD(), not free() or SDL_free()
229 * Note that SDL will only deal with SIMD instruction sets it is aware of;
230 * for example, SDL 2.0.8 knows that SSE wants 16-byte vectors
231 * (SDL_HasSSE()), and AVX2 wants 32 bytes (SDL_HasAVX2()), but doesn't
232 * know that AVX-512 wants 64. To be clear: if you can't decide to use an
233 * instruction set with an SDL_Has*() function, don't use that instruction
234 * set with memory allocated through here.
236 * SDL_AllocSIMD(0) will return a non-NULL pointer, assuming the system isn't
239 * \param len The length, in bytes, of the block to allocated. The actual
240 * allocated block might be larger due to padding, etc.
241 * \return Pointer to newly-allocated block, NULL if out of memory.
243 * \sa SDL_SIMDAlignment
246 extern DECLSPEC void * SDLCALL SDL_SIMDAlloc(const size_t len);
249 * \brief Deallocate memory obtained from SDL_SIMDAlloc
251 * It is not valid to use this function on a pointer from anything but
252 * SDL_SIMDAlloc(). It can't be used on pointers from malloc, realloc,
253 * SDL_malloc, memalign, new[], etc.
255 * However, SDL_SIMDFree(NULL) is a legal no-op.
259 extern DECLSPEC void SDLCALL SDL_SIMDFree(void *ptr);
261 /* vi: set ts=4 sw=4 expandtab: */
262 /* Ends C function definitions when using C++ */
266 #include "close_code.h"
268 #endif /* SDL_cpuinfo_h_ */
270 /* vi: set ts=4 sw=4 expandtab: */