|
1
|
1 /*
|
|
|
2 Simple DirectMedia Layer
|
|
|
3 Copyright (C) 1997-2025 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 /* WIKI CATEGORY: Intrinsics */
|
|
|
23
|
|
|
24 /**
|
|
|
25 * # CategoryIntrinsics
|
|
|
26 *
|
|
|
27 * SDL does some preprocessor gymnastics to determine if any CPU-specific
|
|
|
28 * compiler intrinsics are available, as this is not necessarily an easy thing
|
|
|
29 * to calculate, and sometimes depends on quirks of a system, versions of
|
|
|
30 * build tools, and other external forces.
|
|
|
31 *
|
|
|
32 * Apps including SDL's headers will be able to check consistent preprocessor
|
|
|
33 * definitions to decide if it's safe to use compiler intrinsics for a
|
|
|
34 * specific CPU architecture. This check only tells you that the compiler is
|
|
|
35 * capable of using those intrinsics; at runtime, you should still check if
|
|
|
36 * they are available on the current system with the
|
|
|
37 * [CPU info functions](https://wiki.libsdl.org/SDL3/CategoryCPUInfo)
|
|
|
38 * , such as SDL_HasSSE() or SDL_HasNEON(). Otherwise, the process might crash
|
|
|
39 * for using an unsupported CPU instruction.
|
|
|
40 *
|
|
|
41 * SDL only sets preprocessor defines for CPU intrinsics if they are
|
|
|
42 * supported, so apps should check with `#ifdef` and not `#if`.
|
|
|
43 *
|
|
|
44 * SDL will also include the appropriate instruction-set-specific support
|
|
|
45 * headers, so if SDL decides to define SDL_SSE2_INTRINSICS, it will also
|
|
|
46 * `#include <emmintrin.h>` as well.
|
|
|
47 */
|
|
|
48
|
|
|
49 #ifndef SDL_intrin_h_
|
|
|
50 #define SDL_intrin_h_
|
|
|
51
|
|
|
52 #include <SDL3/SDL_stdinc.h>
|
|
|
53
|
|
|
54 #ifdef SDL_WIKI_DOCUMENTATION_SECTION
|
|
|
55
|
|
|
56 /**
|
|
|
57 * Defined if (and only if) the compiler supports Loongarch LSX intrinsics.
|
|
|
58 *
|
|
|
59 * If this macro is defined, SDL will have already included `<lsxintrin.h>`
|
|
|
60 *
|
|
|
61 * \since This macro is available since SDL 3.2.0.
|
|
|
62 *
|
|
|
63 * \sa SDL_LASX_INTRINSICS
|
|
|
64 */
|
|
|
65 #define SDL_LSX_INTRINSICS 1
|
|
|
66
|
|
|
67 /**
|
|
|
68 * Defined if (and only if) the compiler supports Loongarch LSX intrinsics.
|
|
|
69 *
|
|
|
70 * If this macro is defined, SDL will have already included `<lasxintrin.h>`
|
|
|
71 *
|
|
|
72 * \since This macro is available since SDL 3.2.0.
|
|
|
73 *
|
|
|
74 * \sa SDL_LASX_INTRINSICS
|
|
|
75 */
|
|
|
76 #define SDL_LASX_INTRINSICS 1
|
|
|
77
|
|
|
78 /**
|
|
|
79 * Defined if (and only if) the compiler supports ARM NEON intrinsics.
|
|
|
80 *
|
|
|
81 * If this macro is defined, SDL will have already included `<armintr.h>`
|
|
|
82 * `<arm_neon.h>`, `<arm64intr.h>`, and `<arm64_neon.h>`, as appropriate.
|
|
|
83 *
|
|
|
84 * \since This macro is available since SDL 3.2.0.
|
|
|
85 */
|
|
|
86 #define SDL_NEON_INTRINSICS 1
|
|
|
87
|
|
|
88 /**
|
|
|
89 * Defined if (and only if) the compiler supports PowerPC Altivec intrinsics.
|
|
|
90 *
|
|
|
91 * If this macro is defined, SDL will have already included `<altivec.h>`
|
|
|
92 *
|
|
|
93 * \since This macro is available since SDL 3.2.0.
|
|
|
94 */
|
|
|
95 #define SDL_ALTIVEC_INTRINSICS 1
|
|
|
96
|
|
|
97 /**
|
|
|
98 * Defined if (and only if) the compiler supports Intel MMX intrinsics.
|
|
|
99 *
|
|
|
100 * If this macro is defined, SDL will have already included `<mmintrin.h>`
|
|
|
101 *
|
|
|
102 * \since This macro is available since SDL 3.2.0.
|
|
|
103 *
|
|
|
104 * \sa SDL_SSE_INTRINSICS
|
|
|
105 */
|
|
|
106 #define SDL_MMX_INTRINSICS 1
|
|
|
107
|
|
|
108 /**
|
|
|
109 * Defined if (and only if) the compiler supports Intel SSE intrinsics.
|
|
|
110 *
|
|
|
111 * If this macro is defined, SDL will have already included `<xmmintrin.h>`
|
|
|
112 *
|
|
|
113 * \since This macro is available since SDL 3.2.0.
|
|
|
114 *
|
|
|
115 * \sa SDL_SSE2_INTRINSICS
|
|
|
116 * \sa SDL_SSE3_INTRINSICS
|
|
|
117 * \sa SDL_SSE4_1_INTRINSICS
|
|
|
118 * \sa SDL_SSE4_2_INTRINSICS
|
|
|
119 */
|
|
|
120 #define SDL_SSE_INTRINSICS 1
|
|
|
121
|
|
|
122 /**
|
|
|
123 * Defined if (and only if) the compiler supports Intel SSE2 intrinsics.
|
|
|
124 *
|
|
|
125 * If this macro is defined, SDL will have already included `<emmintrin.h>`
|
|
|
126 *
|
|
|
127 * \since This macro is available since SDL 3.2.0.
|
|
|
128 *
|
|
|
129 * \sa SDL_SSE_INTRINSICS
|
|
|
130 * \sa SDL_SSE3_INTRINSICS
|
|
|
131 * \sa SDL_SSE4_1_INTRINSICS
|
|
|
132 * \sa SDL_SSE4_2_INTRINSICS
|
|
|
133 */
|
|
|
134 #define SDL_SSE2_INTRINSICS 1
|
|
|
135
|
|
|
136 /**
|
|
|
137 * Defined if (and only if) the compiler supports Intel SSE3 intrinsics.
|
|
|
138 *
|
|
|
139 * If this macro is defined, SDL will have already included `<pmmintrin.h>`
|
|
|
140 *
|
|
|
141 * \since This macro is available since SDL 3.2.0.
|
|
|
142 *
|
|
|
143 * \sa SDL_SSE_INTRINSICS
|
|
|
144 * \sa SDL_SSE2_INTRINSICS
|
|
|
145 * \sa SDL_SSE4_1_INTRINSICS
|
|
|
146 * \sa SDL_SSE4_2_INTRINSICS
|
|
|
147 */
|
|
|
148 #define SDL_SSE3_INTRINSICS 1
|
|
|
149
|
|
|
150 /**
|
|
|
151 * Defined if (and only if) the compiler supports Intel SSE4.1 intrinsics.
|
|
|
152 *
|
|
|
153 * If this macro is defined, SDL will have already included `<smmintrin.h>`
|
|
|
154 *
|
|
|
155 * \since This macro is available since SDL 3.2.0.
|
|
|
156 *
|
|
|
157 * \sa SDL_SSE_INTRINSICS
|
|
|
158 * \sa SDL_SSE2_INTRINSICS
|
|
|
159 * \sa SDL_SSE3_INTRINSICS
|
|
|
160 * \sa SDL_SSE4_2_INTRINSICS
|
|
|
161 */
|
|
|
162 #define SDL_SSE4_1_INTRINSICS 1
|
|
|
163
|
|
|
164 /**
|
|
|
165 * Defined if (and only if) the compiler supports Intel SSE4.2 intrinsics.
|
|
|
166 *
|
|
|
167 * If this macro is defined, SDL will have already included `<nmmintrin.h>`
|
|
|
168 *
|
|
|
169 * \since This macro is available since SDL 3.2.0.
|
|
|
170 *
|
|
|
171 * \sa SDL_SSE_INTRINSICS
|
|
|
172 * \sa SDL_SSE2_INTRINSICS
|
|
|
173 * \sa SDL_SSE3_INTRINSICS
|
|
|
174 * \sa SDL_SSE4_1_INTRINSICS
|
|
|
175 */
|
|
|
176 #define SDL_SSE4_2_INTRINSICS 1
|
|
|
177
|
|
|
178 /**
|
|
|
179 * Defined if (and only if) the compiler supports Intel AVX intrinsics.
|
|
|
180 *
|
|
|
181 * If this macro is defined, SDL will have already included `<immintrin.h>`
|
|
|
182 *
|
|
|
183 * \since This macro is available since SDL 3.2.0.
|
|
|
184 *
|
|
|
185 * \sa SDL_AVX2_INTRINSICS
|
|
|
186 * \sa SDL_AVX512F_INTRINSICS
|
|
|
187 */
|
|
|
188 #define SDL_AVX_INTRINSICS 1
|
|
|
189
|
|
|
190 /**
|
|
|
191 * Defined if (and only if) the compiler supports Intel AVX2 intrinsics.
|
|
|
192 *
|
|
|
193 * If this macro is defined, SDL will have already included `<immintrin.h>`
|
|
|
194 *
|
|
|
195 * \since This macro is available since SDL 3.2.0.
|
|
|
196 *
|
|
|
197 * \sa SDL_AVX_INTRINSICS
|
|
|
198 * \sa SDL_AVX512F_INTRINSICS
|
|
|
199 */
|
|
|
200 #define SDL_AVX2_INTRINSICS 1
|
|
|
201
|
|
|
202 /**
|
|
|
203 * Defined if (and only if) the compiler supports Intel AVX-512F intrinsics.
|
|
|
204 *
|
|
|
205 * AVX-512F is also sometimes referred to as "AVX-512 Foundation."
|
|
|
206 *
|
|
|
207 * If this macro is defined, SDL will have already included `<immintrin.h>`
|
|
|
208 *
|
|
|
209 * \since This macro is available since SDL 3.2.0.
|
|
|
210 *
|
|
|
211 * \sa SDL_AVX_INTRINSICS
|
|
|
212 * \sa SDL_AVX2_INTRINSICS
|
|
|
213 */
|
|
|
214 #define SDL_AVX512F_INTRINSICS 1
|
|
|
215 #endif
|
|
|
216
|
|
|
217 /* Need to do this here because intrin.h has C++ code in it */
|
|
|
218 /* Visual Studio 2005 has a bug where intrin.h conflicts with winnt.h */
|
|
|
219 #if defined(_MSC_VER) && (_MSC_VER >= 1500) && (defined(_M_IX86) || defined(_M_X64))
|
|
|
220 #ifdef __clang__
|
|
|
221 /* As of Clang 11, '_m_prefetchw' is conflicting with the winnt.h's version,
|
|
|
222 so we define the needed '_m_prefetch' here as a pseudo-header, until the issue is fixed. */
|
|
|
223 #ifndef __PRFCHWINTRIN_H
|
|
|
224 #define __PRFCHWINTRIN_H
|
|
|
225 static __inline__ void __attribute__((__always_inline__, __nodebug__))
|
|
|
226 _m_prefetch(void *__P)
|
|
|
227 {
|
|
|
228 __builtin_prefetch (__P, 0, 3 /* _MM_HINT_T0 */);
|
|
|
229 }
|
|
|
230 #endif /* __PRFCHWINTRIN_H */
|
|
|
231 #endif /* __clang__ */
|
|
|
232 #include <intrin.h>
|
|
|
233
|
|
|
234 #elif defined(__MINGW64_VERSION_MAJOR)
|
|
|
235 #include <intrin.h>
|
|
|
236 #if defined(__ARM_NEON) && !defined(SDL_DISABLE_NEON)
|
|
|
237 # define SDL_NEON_INTRINSICS 1
|
|
|
238 # include <arm_neon.h>
|
|
|
239 #endif
|
|
|
240
|
|
|
241 #else
|
|
|
242 /* altivec.h redefining bool causes a number of problems, see bugs 3993 and 4392, so you need to explicitly define SDL_ENABLE_ALTIVEC to have it included. */
|
|
|
243 #if defined(__ALTIVEC__) && defined(SDL_ENABLE_ALTIVEC)
|
|
|
244 #define SDL_ALTIVEC_INTRINSICS 1
|
|
|
245 #include <altivec.h>
|
|
|
246 #endif
|
|
|
247 #ifndef SDL_DISABLE_NEON
|
|
|
248 # ifdef __ARM_NEON
|
|
|
249 # define SDL_NEON_INTRINSICS 1
|
|
|
250 # include <arm_neon.h>
|
|
|
251 # elif defined(SDL_PLATFORM_WINDOWS)
|
|
|
252 /* Visual Studio doesn't define __ARM_ARCH, but _M_ARM (if set, always 7), and _M_ARM64 (if set, always 1). */
|
|
|
253 # ifdef _M_ARM
|
|
|
254 # define SDL_NEON_INTRINSICS 1
|
|
|
255 # include <armintr.h>
|
|
|
256 # include <arm_neon.h>
|
|
|
257 # define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
|
|
|
258 # endif
|
|
|
259 # if defined (_M_ARM64)
|
|
|
260 # define SDL_NEON_INTRINSICS 1
|
|
|
261 # include <arm64intr.h>
|
|
|
262 # include <arm64_neon.h>
|
|
|
263 # define __ARM_NEON 1 /* Set __ARM_NEON so that it can be used elsewhere, at compile time */
|
|
|
264 # define __ARM_ARCH 8
|
|
|
265 # endif
|
|
|
266 # endif
|
|
|
267 #endif
|
|
|
268 #endif /* compiler version */
|
|
|
269
|
|
|
270 #ifdef SDL_WIKI_DOCUMENTATION_SECTION
|
|
|
271
|
|
|
272 /**
|
|
|
273 * A macro to decide if the compiler supports `__attribute__((target))`.
|
|
|
274 *
|
|
|
275 * Even though this is defined in SDL's public headers, it is generally not
|
|
|
276 * used directly by apps. Apps should probably just use SDL_TARGETING
|
|
|
277 * directly, instead.
|
|
|
278 *
|
|
|
279 * \since This macro is available since SDL 3.2.0.
|
|
|
280 *
|
|
|
281 * \sa SDL_TARGETING
|
|
|
282 */
|
|
|
283 #define SDL_HAS_TARGET_ATTRIBS
|
|
|
284 #elif defined(__loongarch64) && defined(__GNUC__) && (__GNUC__ >= 15)
|
|
|
285 /* LoongArch requires GCC 15+ for target attribute support */
|
|
|
286 # define SDL_HAS_TARGET_ATTRIBS
|
|
|
287 #elif defined(__clang__) && defined(__has_attribute)
|
|
|
288 # if __has_attribute(target)
|
|
|
289 # define SDL_HAS_TARGET_ATTRIBS
|
|
|
290 # endif
|
|
|
291 #elif defined(__GNUC__) && !defined(__loongarch64) && (__GNUC__ + (__GNUC_MINOR__ >= 9) > 4) /* gcc >= 4.9 */
|
|
|
292 # define SDL_HAS_TARGET_ATTRIBS
|
|
|
293 #elif defined(__ICC) && __ICC >= 1600
|
|
|
294 # define SDL_HAS_TARGET_ATTRIBS
|
|
|
295 #endif
|
|
|
296
|
|
|
297
|
|
|
298 #ifdef SDL_WIKI_DOCUMENTATION_SECTION
|
|
|
299
|
|
|
300 /**
|
|
|
301 * A macro to tag a function as targeting a specific CPU architecture.
|
|
|
302 *
|
|
|
303 * This is a hint to the compiler that a function should be built with support
|
|
|
304 * for a CPU instruction set that might be different than the rest of the
|
|
|
305 * program.
|
|
|
306 *
|
|
|
307 * The particulars of this are explained in the GCC documentation:
|
|
|
308 *
|
|
|
309 * https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-target-function-attribute
|
|
|
310 *
|
|
|
311 * An example of using this feature is to turn on SSE2 support for a specific
|
|
|
312 * function, even if the rest of the source code is not compiled to use SSE2
|
|
|
313 * code:
|
|
|
314 *
|
|
|
315 * ```c
|
|
|
316 * #ifdef SDL_SSE2_INTRINSICS
|
|
|
317 * static void SDL_TARGETING("sse2") DoSomethingWithSSE2(char *x) {
|
|
|
318 * ...use SSE2 intrinsic functions, etc...
|
|
|
319 * }
|
|
|
320 * #endif
|
|
|
321 *
|
|
|
322 * // later...
|
|
|
323 * #ifdef SDL_SSE2_INTRINSICS
|
|
|
324 * if (SDL_HasSSE2()) {
|
|
|
325 * DoSomethingWithSSE2(str);
|
|
|
326 * }
|
|
|
327 * #endif
|
|
|
328 * ```
|
|
|
329 *
|
|
|
330 * The application is, on a whole, built without SSE2 instructions, so it will
|
|
|
331 * run on Intel machines that don't support SSE2. But then at runtime, it
|
|
|
332 * checks if the system supports the instructions, and then calls into a
|
|
|
333 * function that uses SSE2 opcodes. The ifdefs make sure that this code isn't
|
|
|
334 * used on platforms that don't have SSE2 at all.
|
|
|
335 *
|
|
|
336 * On compilers without target support, this is defined to nothing.
|
|
|
337 *
|
|
|
338 * This symbol is used by SDL internally, but apps and other libraries are
|
|
|
339 * welcome to use it for their own interfaces as well.
|
|
|
340 *
|
|
|
341 * \since This macro is available since SDL 3.2.0.
|
|
|
342 */
|
|
|
343 #define SDL_TARGETING(x) __attribute__((target(x)))
|
|
|
344
|
|
|
345 #elif defined(SDL_HAS_TARGET_ATTRIBS)
|
|
|
346 # define SDL_TARGETING(x) __attribute__((target(x)))
|
|
|
347 #else
|
|
|
348 # define SDL_TARGETING(x)
|
|
|
349 #endif
|
|
|
350
|
|
|
351 #ifdef __loongarch64
|
|
|
352 # ifndef SDL_DISABLE_LSX
|
|
|
353 # define SDL_LSX_INTRINSICS 1
|
|
|
354 # include <lsxintrin.h>
|
|
|
355 # endif
|
|
|
356 # ifndef SDL_DISABLE_LASX
|
|
|
357 # define SDL_LASX_INTRINSICS 1
|
|
|
358 # include <lasxintrin.h>
|
|
|
359 # endif
|
|
|
360 #endif
|
|
|
361
|
|
|
362 #if defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86)
|
|
|
363 # if ((defined(_MSC_VER) && !defined(_M_X64)) || defined(__MMX__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(SDL_DISABLE_MMX)
|
|
|
364 # define SDL_MMX_INTRINSICS 1
|
|
|
365 # include <mmintrin.h>
|
|
|
366 # endif
|
|
|
367 # if (defined(_MSC_VER) || defined(__SSE__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(SDL_DISABLE_SSE)
|
|
|
368 # define SDL_SSE_INTRINSICS 1
|
|
|
369 # include <xmmintrin.h>
|
|
|
370 # endif
|
|
|
371 # if (defined(_MSC_VER) || defined(__SSE2__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(SDL_DISABLE_SSE2)
|
|
|
372 # define SDL_SSE2_INTRINSICS 1
|
|
|
373 # include <emmintrin.h>
|
|
|
374 # endif
|
|
|
375 # if (defined(_MSC_VER) || defined(__SSE3__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(SDL_DISABLE_SSE3)
|
|
|
376 # define SDL_SSE3_INTRINSICS 1
|
|
|
377 # include <pmmintrin.h>
|
|
|
378 # endif
|
|
|
379 # if (defined(_MSC_VER) || defined(__SSE4_1__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(SDL_DISABLE_SSE4_1)
|
|
|
380 # define SDL_SSE4_1_INTRINSICS 1
|
|
|
381 # include <smmintrin.h>
|
|
|
382 # endif
|
|
|
383 # if (defined(_MSC_VER) || defined(__SSE4_2__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(SDL_DISABLE_SSE4_2)
|
|
|
384 # define SDL_SSE4_2_INTRINSICS 1
|
|
|
385 # include <nmmintrin.h>
|
|
|
386 # endif
|
|
|
387 # if defined(__clang__) && (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX__) && !defined(SDL_DISABLE_AVX)
|
|
|
388 # define SDL_DISABLE_AVX /* see https://reviews.llvm.org/D20291 and https://reviews.llvm.org/D79194 */
|
|
|
389 # endif
|
|
|
390 # if (defined(_MSC_VER) || defined(__AVX__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(_M_ARM64EC) && !defined(SDL_DISABLE_AVX)
|
|
|
391 # define SDL_AVX_INTRINSICS 1
|
|
|
392 # include <immintrin.h>
|
|
|
393 # endif
|
|
|
394 # if defined(__clang__) && (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX2__) && !defined(SDL_DISABLE_AVX2)
|
|
|
395 # define SDL_DISABLE_AVX2 /* see https://reviews.llvm.org/D20291 and https://reviews.llvm.org/D79194 */
|
|
|
396 # endif
|
|
|
397 # if (defined(_MSC_VER) || defined(__AVX2__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(_M_ARM64EC) && !defined(SDL_DISABLE_AVX2)
|
|
|
398 # define SDL_AVX2_INTRINSICS 1
|
|
|
399 # include <immintrin.h>
|
|
|
400 # endif
|
|
|
401 # if defined(__clang__) && (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX512F__) && !defined(SDL_DISABLE_AVX512F)
|
|
|
402 # define SDL_DISABLE_AVX512F /* see https://reviews.llvm.org/D20291 and https://reviews.llvm.org/D79194 */
|
|
|
403 # endif
|
|
|
404 # if (defined(_MSC_VER) || defined(__AVX512F__) || defined(SDL_HAS_TARGET_ATTRIBS)) && !defined(_M_ARM64EC) && !defined(SDL_DISABLE_AVX512F)
|
|
|
405 # define SDL_AVX512F_INTRINSICS 1
|
|
|
406 # include <immintrin.h>
|
|
|
407 # endif
|
|
|
408 #endif /* defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || defined(_M_IX86) */
|
|
|
409
|
|
|
410 #endif /* SDL_intrin_h_ */
|
