… | |
… | |
40 | |
40 | |
41 | #ifndef ECB_H |
41 | #ifndef ECB_H |
42 | #define ECB_H |
42 | #define ECB_H |
43 | |
43 | |
44 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
45 | #define ECB_VERSION 0x00010004 |
45 | #define ECB_VERSION 0x00010006 |
46 | |
46 | |
47 | #ifdef _WIN32 |
47 | #ifdef _WIN32 |
48 | typedef signed char int8_t; |
48 | typedef signed char int8_t; |
49 | typedef unsigned char uint8_t; |
49 | typedef unsigned char uint8_t; |
50 | typedef signed short int16_t; |
50 | typedef signed short int16_t; |
… | |
… | |
67 | typedef uint32_t uintptr_t; |
67 | typedef uint32_t uintptr_t; |
68 | typedef int32_t intptr_t; |
68 | typedef int32_t intptr_t; |
69 | #endif |
69 | #endif |
70 | #else |
70 | #else |
71 | #include <inttypes.h> |
71 | #include <inttypes.h> |
72 | #if UINTMAX_MAX > 0xffffffffU |
72 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
73 | #define ECB_PTRSIZE 8 |
73 | #define ECB_PTRSIZE 8 |
74 | #else |
74 | #else |
75 | #define ECB_PTRSIZE 4 |
75 | #define ECB_PTRSIZE 4 |
76 | #endif |
76 | #endif |
77 | #endif |
77 | #endif |
78 | |
78 | |
|
|
79 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
|
|
80 | #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
|
|
81 | |
79 | /* work around x32 idiocy by defining proper macros */ |
82 | /* work around x32 idiocy by defining proper macros */ |
80 | #if __amd64 || __x86_64 || _M_AMD64 || _M_X64 |
83 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
81 | #if _ILP32 |
84 | #if _ILP32 |
82 | #define ECB_AMD64_X32 1 |
85 | #define ECB_AMD64_X32 1 |
83 | #else |
86 | #else |
84 | #define ECB_AMD64 1 |
87 | #define ECB_AMD64 1 |
85 | #endif |
88 | #endif |
… | |
… | |
112 | #define ECB_CLANG_EXTENSION(x) 0 |
115 | #define ECB_CLANG_EXTENSION(x) 0 |
113 | #endif |
116 | #endif |
114 | |
117 | |
115 | #define ECB_CPP (__cplusplus+0) |
118 | #define ECB_CPP (__cplusplus+0) |
116 | #define ECB_CPP11 (__cplusplus >= 201103L) |
119 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
120 | #define ECB_CPP14 (__cplusplus >= 201402L) |
|
|
121 | #define ECB_CPP17 (__cplusplus >= 201703L) |
117 | |
122 | |
118 | #if ECB_CPP |
123 | #if ECB_CPP |
119 | #define ECB_C 0 |
124 | #define ECB_C 0 |
120 | #define ECB_STDC_VERSION 0 |
125 | #define ECB_STDC_VERSION 0 |
121 | #else |
126 | #else |
… | |
… | |
123 | #define ECB_STDC_VERSION __STDC_VERSION__ |
128 | #define ECB_STDC_VERSION __STDC_VERSION__ |
124 | #endif |
129 | #endif |
125 | |
130 | |
126 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
131 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
127 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
132 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
133 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
128 | |
134 | |
129 | #if ECB_CPP |
135 | #if ECB_CPP |
130 | #define ECB_EXTERN_C extern "C" |
136 | #define ECB_EXTERN_C extern "C" |
131 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
137 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
132 | #define ECB_EXTERN_C_END } |
138 | #define ECB_EXTERN_C_END } |
… | |
… | |
147 | |
153 | |
148 | #if ECB_NO_SMP |
154 | #if ECB_NO_SMP |
149 | #define ECB_MEMORY_FENCE do { } while (0) |
155 | #define ECB_MEMORY_FENCE do { } while (0) |
150 | #endif |
156 | #endif |
151 | |
157 | |
|
|
158 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ |
|
|
159 | #if __xlC__ && ECB_CPP |
|
|
160 | #include <builtins.h> |
|
|
161 | #endif |
|
|
162 | |
|
|
163 | #if 1400 <= _MSC_VER |
|
|
164 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
|
|
165 | #endif |
|
|
166 | |
152 | #ifndef ECB_MEMORY_FENCE |
167 | #ifndef ECB_MEMORY_FENCE |
153 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
168 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
169 | #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") |
154 | #if __i386 || __i386__ |
170 | #if __i386 || __i386__ |
155 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
156 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
172 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
157 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
173 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
158 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
174 | #elif ECB_GCC_AMD64 |
159 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
175 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
160 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
176 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
161 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
177 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
162 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
178 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
163 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
179 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
180 | #elif defined __ARM_ARCH_2__ \ |
|
|
181 | || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ |
|
|
182 | || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ |
|
|
183 | || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ |
|
|
184 | || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ |
|
|
185 | || defined __ARM_ARCH_5TEJ__ |
|
|
186 | /* should not need any, unless running old code on newer cpu - arm doesn't support that */ |
164 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
187 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
165 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
188 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
189 | || defined __ARM_ARCH_6T2__ |
166 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
190 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
167 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
191 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
168 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
192 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
169 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
193 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
170 | #elif __aarch64__ |
194 | #elif __aarch64__ |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
195 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
172 | #elif (__sparc || __sparc__) && !__sparcv8 |
196 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
173 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
197 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
174 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
198 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
175 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
199 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
176 | #elif defined __s390__ || defined __s390x__ |
200 | #elif defined __s390__ || defined __s390x__ |
177 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
201 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
… | |
… | |
200 | #if ECB_GCC_VERSION(4,7) |
224 | #if ECB_GCC_VERSION(4,7) |
201 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
225 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
202 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
226 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
203 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
227 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
204 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
228 | #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
229 | #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) |
205 | |
230 | |
206 | #elif ECB_CLANG_EXTENSION(c_atomic) |
231 | #elif ECB_CLANG_EXTENSION(c_atomic) |
207 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
232 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
208 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
233 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
209 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
234 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
210 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
235 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
|
|
236 | #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) |
211 | |
237 | |
212 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
238 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
213 | #define ECB_MEMORY_FENCE __sync_synchronize () |
239 | #define ECB_MEMORY_FENCE __sync_synchronize () |
214 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
240 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
215 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
241 | /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ |
… | |
… | |
225 | #elif defined _WIN32 |
251 | #elif defined _WIN32 |
226 | #include <WinNT.h> |
252 | #include <WinNT.h> |
227 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
253 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
228 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
254 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
229 | #include <mbarrier.h> |
255 | #include <mbarrier.h> |
230 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
256 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
231 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
257 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
232 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
258 | #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () |
|
|
259 | #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () |
233 | #elif __xlC__ |
260 | #elif __xlC__ |
234 | #define ECB_MEMORY_FENCE __sync () |
261 | #define ECB_MEMORY_FENCE __sync () |
235 | #endif |
262 | #endif |
236 | #endif |
263 | #endif |
237 | |
264 | |
238 | #ifndef ECB_MEMORY_FENCE |
265 | #ifndef ECB_MEMORY_FENCE |
239 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
266 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
240 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
267 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
241 | /* not just C11 atomics and atomic accesses */ |
268 | /* not just C11 atomics and atomic accesses */ |
242 | #include <stdatomic.h> |
269 | #include <stdatomic.h> |
243 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
244 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
245 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
246 | /* for most usages, or gcc and clang have a bug */ |
|
|
247 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
248 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
249 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
250 | /* for all __atomic_thread_fence's except seq_cst */ |
|
|
251 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
270 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
271 | #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) |
|
|
272 | #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) |
252 | #endif |
273 | #endif |
253 | #endif |
274 | #endif |
254 | |
275 | |
255 | #ifndef ECB_MEMORY_FENCE |
276 | #ifndef ECB_MEMORY_FENCE |
256 | #if !ECB_AVOID_PTHREADS |
277 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
276 | |
297 | |
277 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
298 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
278 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
299 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
279 | #endif |
300 | #endif |
280 | |
301 | |
|
|
302 | #if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE |
|
|
303 | #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ |
|
|
304 | #endif |
|
|
305 | |
281 | /*****************************************************************************/ |
306 | /*****************************************************************************/ |
282 | |
307 | |
283 | #if ECB_CPP |
308 | #if ECB_CPP |
284 | #define ecb_inline static inline |
309 | #define ecb_inline static inline |
285 | #elif ECB_GCC_VERSION(2,5) |
310 | #elif ECB_GCC_VERSION(2,5) |
… | |
… | |
349 | #define ecb_deprecated __declspec (deprecated) |
374 | #define ecb_deprecated __declspec (deprecated) |
350 | #else |
375 | #else |
351 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
376 | #define ecb_deprecated ecb_attribute ((__deprecated__)) |
352 | #endif |
377 | #endif |
353 | |
378 | |
354 | #if __MSC_VER >= 1500 |
379 | #if _MSC_VER >= 1500 |
355 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
380 | #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) |
356 | #elif ECB_GCC_VERSION(4,5) |
381 | #elif ECB_GCC_VERSION(4,5) |
357 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
382 | #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) |
358 | #else |
383 | #else |
359 | #define ecb_deprecated_message(msg) ecb_deprecated |
384 | #define ecb_deprecated_message(msg) ecb_deprecated |
… | |
… | |
368 | #define ecb_unused ecb_attribute ((__unused__)) |
393 | #define ecb_unused ecb_attribute ((__unused__)) |
369 | #define ecb_const ecb_attribute ((__const__)) |
394 | #define ecb_const ecb_attribute ((__const__)) |
370 | #define ecb_pure ecb_attribute ((__pure__)) |
395 | #define ecb_pure ecb_attribute ((__pure__)) |
371 | |
396 | |
372 | #if ECB_C11 || __IBMC_NORETURN |
397 | #if ECB_C11 || __IBMC_NORETURN |
373 | /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */ |
398 | /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ |
374 | #define ecb_noreturn _Noreturn |
399 | #define ecb_noreturn _Noreturn |
375 | #elif ECB_CPP11 |
400 | #elif ECB_CPP11 |
376 | #define ecb_noreturn [[noreturn]] |
401 | #define ecb_noreturn [[noreturn]] |
377 | #elif _MSC_VER >= 1200 |
402 | #elif _MSC_VER >= 1200 |
378 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
403 | /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ |
… | |
… | |
415 | #else |
440 | #else |
416 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
441 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
417 | ecb_function_ ecb_const int |
442 | ecb_function_ ecb_const int |
418 | ecb_ctz32 (uint32_t x) |
443 | ecb_ctz32 (uint32_t x) |
419 | { |
444 | { |
|
|
445 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
446 | unsigned long r; |
|
|
447 | _BitScanForward (&r, x); |
|
|
448 | return (int)r; |
|
|
449 | #else |
420 | int r = 0; |
450 | int r = 0; |
421 | |
451 | |
422 | x &= ~x + 1; /* this isolates the lowest bit */ |
452 | x &= ~x + 1; /* this isolates the lowest bit */ |
423 | |
453 | |
424 | #if ECB_branchless_on_i386 |
454 | #if ECB_branchless_on_i386 |
… | |
… | |
434 | if (x & 0xff00ff00) r += 8; |
464 | if (x & 0xff00ff00) r += 8; |
435 | if (x & 0xffff0000) r += 16; |
465 | if (x & 0xffff0000) r += 16; |
436 | #endif |
466 | #endif |
437 | |
467 | |
438 | return r; |
468 | return r; |
|
|
469 | #endif |
439 | } |
470 | } |
440 | |
471 | |
441 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
472 | ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); |
442 | ecb_function_ ecb_const int |
473 | ecb_function_ ecb_const int |
443 | ecb_ctz64 (uint64_t x) |
474 | ecb_ctz64 (uint64_t x) |
444 | { |
475 | { |
|
|
476 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
477 | unsigned long r; |
|
|
478 | _BitScanForward64 (&r, x); |
|
|
479 | return (int)r; |
|
|
480 | #else |
445 | int shift = x & 0xffffffffU ? 0 : 32; |
481 | int shift = x & 0xffffffff ? 0 : 32; |
446 | return ecb_ctz32 (x >> shift) + shift; |
482 | return ecb_ctz32 (x >> shift) + shift; |
|
|
483 | #endif |
447 | } |
484 | } |
448 | |
485 | |
449 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
486 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
450 | ecb_function_ ecb_const int |
487 | ecb_function_ ecb_const int |
451 | ecb_popcount32 (uint32_t x) |
488 | ecb_popcount32 (uint32_t x) |
… | |
… | |
459 | } |
496 | } |
460 | |
497 | |
461 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
498 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x); |
462 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
499 | ecb_function_ ecb_const int ecb_ld32 (uint32_t x) |
463 | { |
500 | { |
|
|
501 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
502 | unsigned long r; |
|
|
503 | _BitScanReverse (&r, x); |
|
|
504 | return (int)r; |
|
|
505 | #else |
464 | int r = 0; |
506 | int r = 0; |
465 | |
507 | |
466 | if (x >> 16) { x >>= 16; r += 16; } |
508 | if (x >> 16) { x >>= 16; r += 16; } |
467 | if (x >> 8) { x >>= 8; r += 8; } |
509 | if (x >> 8) { x >>= 8; r += 8; } |
468 | if (x >> 4) { x >>= 4; r += 4; } |
510 | if (x >> 4) { x >>= 4; r += 4; } |
469 | if (x >> 2) { x >>= 2; r += 2; } |
511 | if (x >> 2) { x >>= 2; r += 2; } |
470 | if (x >> 1) { r += 1; } |
512 | if (x >> 1) { r += 1; } |
471 | |
513 | |
472 | return r; |
514 | return r; |
|
|
515 | #endif |
473 | } |
516 | } |
474 | |
517 | |
475 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
518 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x); |
476 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
519 | ecb_function_ ecb_const int ecb_ld64 (uint64_t x) |
477 | { |
520 | { |
|
|
521 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
522 | unsigned long r; |
|
|
523 | _BitScanReverse64 (&r, x); |
|
|
524 | return (int)r; |
|
|
525 | #else |
478 | int r = 0; |
526 | int r = 0; |
479 | |
527 | |
480 | if (x >> 32) { x >>= 32; r += 32; } |
528 | if (x >> 32) { x >>= 32; r += 32; } |
481 | |
529 | |
482 | return r + ecb_ld32 (x); |
530 | return r + ecb_ld32 (x); |
|
|
531 | #endif |
483 | } |
532 | } |
484 | #endif |
533 | #endif |
485 | |
534 | |
486 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
535 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); |
487 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
536 | ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
… | |
… | |
544 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
593 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
545 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
594 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
546 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
595 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
547 | |
596 | |
548 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
597 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
|
|
598 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
|
|
599 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
|
|
600 | #else |
549 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
601 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
602 | #endif |
550 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
603 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
551 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
604 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
605 | #elif _MSC_VER |
|
|
606 | #include <stdlib.h> |
|
|
607 | #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) |
|
|
608 | #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) |
|
|
609 | #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) |
552 | #else |
610 | #else |
553 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
611 | ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); |
554 | ecb_function_ ecb_const uint16_t |
612 | ecb_function_ ecb_const uint16_t |
555 | ecb_bswap16 (uint16_t x) |
613 | ecb_bswap16 (uint16_t x) |
556 | { |
614 | { |
… | |
… | |
581 | #endif |
639 | #endif |
582 | |
640 | |
583 | /* try to tell the compiler that some condition is definitely true */ |
641 | /* try to tell the compiler that some condition is definitely true */ |
584 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
642 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
585 | |
643 | |
586 | ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); |
644 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
587 | ecb_inline ecb_const unsigned char |
645 | ecb_inline ecb_const uint32_t |
588 | ecb_byteorder_helper (void) |
646 | ecb_byteorder_helper (void) |
589 | { |
647 | { |
590 | /* the union code still generates code under pressure in gcc, */ |
648 | /* the union code still generates code under pressure in gcc, */ |
591 | /* but less than using pointers, and always seems to */ |
649 | /* but less than using pointers, and always seems to */ |
592 | /* successfully return a constant. */ |
650 | /* successfully return a constant. */ |
593 | /* the reason why we have this horrible preprocessor mess */ |
651 | /* the reason why we have this horrible preprocessor mess */ |
594 | /* is to avoid it in all cases, at least on common architectures */ |
652 | /* is to avoid it in all cases, at least on common architectures */ |
595 | /* or when using a recent enough gcc version (>= 4.6) */ |
653 | /* or when using a recent enough gcc version (>= 4.6) */ |
596 | #if ((__i386 || __i386__) && !__VOS__) || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
597 | return 0x44; |
|
|
598 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
654 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
655 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
656 | #define ECB_LITTLE_ENDIAN 1 |
599 | return 0x44; |
657 | return 0x44332211; |
600 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
658 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
659 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
660 | #define ECB_BIG_ENDIAN 1 |
601 | return 0x11; |
661 | return 0x11223344; |
602 | #else |
662 | #else |
603 | union |
663 | union |
604 | { |
664 | { |
|
|
665 | uint8_t c[4]; |
605 | uint32_t i; |
666 | uint32_t u; |
606 | uint8_t c; |
|
|
607 | } u = { 0x11223344 }; |
667 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
608 | return u.c; |
668 | return u.u; |
609 | #endif |
669 | #endif |
610 | } |
670 | } |
611 | |
671 | |
612 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
672 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
613 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
673 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
614 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
674 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
615 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
675 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
616 | |
676 | |
617 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
677 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
618 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
678 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
619 | #else |
679 | #else |
620 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
680 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
… | |
… | |
645 | } |
705 | } |
646 | #else |
706 | #else |
647 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
707 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
648 | #endif |
708 | #endif |
649 | |
709 | |
|
|
710 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
|
|
711 | ecb_function_ ecb_const uint32_t |
|
|
712 | ecb_binary16_to_binary32 (uint32_t x) |
|
|
713 | { |
|
|
714 | unsigned int s = (x & 0x8000) << (31 - 15); |
|
|
715 | int e = (x >> 10) & 0x001f; |
|
|
716 | unsigned int m = x & 0x03ff; |
|
|
717 | |
|
|
718 | if (ecb_expect_false (e == 31)) |
|
|
719 | /* infinity or NaN */ |
|
|
720 | e = 255 - (127 - 15); |
|
|
721 | else if (ecb_expect_false (!e)) |
|
|
722 | { |
|
|
723 | if (ecb_expect_true (!m)) |
|
|
724 | /* zero, handled by code below by forcing e to 0 */ |
|
|
725 | e = 0 - (127 - 15); |
|
|
726 | else |
|
|
727 | { |
|
|
728 | /* subnormal, renormalise */ |
|
|
729 | unsigned int s = 10 - ecb_ld32 (m); |
|
|
730 | |
|
|
731 | m = (m << s) & 0x3ff; /* mask implicit bit */ |
|
|
732 | e -= s - 1; |
|
|
733 | } |
|
|
734 | } |
|
|
735 | |
|
|
736 | /* e and m now are normalised, or zero, (or inf or nan) */ |
|
|
737 | e += 127 - 15; |
|
|
738 | |
|
|
739 | return s | (e << 23) | (m << (23 - 10)); |
|
|
740 | } |
|
|
741 | |
|
|
742 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
|
|
743 | ecb_function_ ecb_const uint16_t |
|
|
744 | ecb_binary32_to_binary16 (uint32_t x) |
|
|
745 | { |
|
|
746 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
|
|
747 | unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
|
|
748 | unsigned int m = x & 0x007fffff; |
|
|
749 | |
|
|
750 | x &= 0x7fffffff; |
|
|
751 | |
|
|
752 | /* if it's within range of binary16 normals, use fast path */ |
|
|
753 | if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) |
|
|
754 | { |
|
|
755 | /* mantissa round-to-even */ |
|
|
756 | m += 0x00000fff + ((m >> (23 - 10)) & 1); |
|
|
757 | |
|
|
758 | /* handle overflow */ |
|
|
759 | if (ecb_expect_false (m >= 0x00800000)) |
|
|
760 | { |
|
|
761 | m >>= 1; |
|
|
762 | e += 1; |
|
|
763 | } |
|
|
764 | |
|
|
765 | return s | (e << 10) | (m >> (23 - 10)); |
|
|
766 | } |
|
|
767 | |
|
|
768 | /* handle large numbers and infinity */ |
|
|
769 | if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) |
|
|
770 | return s | 0x7c00; |
|
|
771 | |
|
|
772 | /* handle zero, subnormals and small numbers */ |
|
|
773 | if (ecb_expect_true (x < 0x38800000)) |
|
|
774 | { |
|
|
775 | /* zero */ |
|
|
776 | if (ecb_expect_true (!x)) |
|
|
777 | return s; |
|
|
778 | |
|
|
779 | /* handle subnormals */ |
|
|
780 | |
|
|
781 | /* too small, will be zero */ |
|
|
782 | if (e < (14 - 24)) /* might not be sharp, but is good enough */ |
|
|
783 | return s; |
|
|
784 | |
|
|
785 | m |= 0x00800000; /* make implicit bit explicit */ |
|
|
786 | |
|
|
787 | /* very tricky - we need to round to the nearest e (+10) bit value */ |
|
|
788 | { |
|
|
789 | unsigned int bits = 14 - e; |
|
|
790 | unsigned int half = (1 << (bits - 1)) - 1; |
|
|
791 | unsigned int even = (m >> bits) & 1; |
|
|
792 | |
|
|
793 | /* if this overflows, we will end up with a normalised number */ |
|
|
794 | m = (m + half + even) >> bits; |
|
|
795 | } |
|
|
796 | |
|
|
797 | return s | m; |
|
|
798 | } |
|
|
799 | |
|
|
800 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
|
|
801 | m >>= 13; |
|
|
802 | |
|
|
803 | return s | 0x7c00 | m | !m; |
|
|
804 | } |
|
|
805 | |
650 | /*******************************************************************************/ |
806 | /*******************************************************************************/ |
651 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
807 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
652 | |
808 | |
653 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
809 | /* basically, everything uses "ieee pure-endian" floating point numbers */ |
654 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
810 | /* the only noteworthy exception is ancient armle, which uses order 43218765 */ |
655 | #if 0 \ |
811 | #if 0 \ |
656 | || __i386 || __i386__ \ |
812 | || __i386 || __i386__ \ |
657 | || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ |
813 | || ECB_GCC_AMD64 \ |
658 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
814 | || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ |
659 | || defined __s390__ || defined __s390x__ \ |
815 | || defined __s390__ || defined __s390x__ \ |
660 | || defined __mips__ \ |
816 | || defined __mips__ \ |
661 | || defined __alpha__ \ |
817 | || defined __alpha__ \ |
662 | || defined __hppa__ \ |
818 | || defined __hppa__ \ |
663 | || defined __ia64__ \ |
819 | || defined __ia64__ \ |
664 | || defined __m68k__ \ |
820 | || defined __m68k__ \ |
665 | || defined __m88k__ \ |
821 | || defined __m88k__ \ |
666 | || defined __sh__ \ |
822 | || defined __sh__ \ |
667 | || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ |
823 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
668 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
824 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
669 | || defined __aarch64__ |
825 | || defined __aarch64__ |
670 | #define ECB_STDFP 1 |
826 | #define ECB_STDFP 1 |
671 | #include <string.h> /* for memcpy */ |
827 | #include <string.h> /* for memcpy */ |
672 | #else |
828 | #else |
… | |
… | |
690 | #define ECB_NAN ECB_INFINITY |
846 | #define ECB_NAN ECB_INFINITY |
691 | #endif |
847 | #endif |
692 | |
848 | |
693 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
849 | #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L |
694 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
850 | #define ecb_ldexpf(x,e) ldexpf ((x), (e)) |
|
|
851 | #define ecb_frexpf(x,e) frexpf ((x), (e)) |
695 | #else |
852 | #else |
696 | #define ecb_ldexpf(x,e) (float) ldexp ((x), (e)) |
853 | #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) |
|
|
854 | #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) |
697 | #endif |
855 | #endif |
698 | |
|
|
699 | /* converts an ieee half/binary16 to a float */ |
|
|
700 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
701 | ecb_function_ ecb_const float |
|
|
702 | ecb_binary16_to_float (uint16_t x) |
|
|
703 | { |
|
|
704 | int e = (x >> 10) & 0x1f; |
|
|
705 | int m = x & 0x3ff; |
|
|
706 | float r; |
|
|
707 | |
|
|
708 | if (!e ) r = ecb_ldexpf (m , -24); |
|
|
709 | else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25); |
|
|
710 | else if (m ) r = ECB_NAN; |
|
|
711 | else r = ECB_INFINITY; |
|
|
712 | |
|
|
713 | return x & 0x8000 ? -r : r; |
|
|
714 | } |
|
|
715 | |
856 | |
716 | /* convert a float to ieee single/binary32 */ |
857 | /* convert a float to ieee single/binary32 */ |
717 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
858 | ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); |
718 | ecb_function_ ecb_const uint32_t |
859 | ecb_function_ ecb_const uint32_t |
719 | ecb_float_to_binary32 (float x) |
860 | ecb_float_to_binary32 (float x) |
… | |
… | |
730 | if (x == 0e0f ) return 0x00000000U; |
871 | if (x == 0e0f ) return 0x00000000U; |
731 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
872 | if (x > +3.40282346638528860e+38f) return 0x7f800000U; |
732 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
873 | if (x < -3.40282346638528860e+38f) return 0xff800000U; |
733 | if (x != x ) return 0x7fbfffffU; |
874 | if (x != x ) return 0x7fbfffffU; |
734 | |
875 | |
735 | m = frexpf (x, &e) * 0x1000000U; |
876 | m = ecb_frexpf (x, &e) * 0x1000000U; |
736 | |
877 | |
737 | r = m & 0x80000000U; |
878 | r = m & 0x80000000U; |
738 | |
879 | |
739 | if (r) |
880 | if (r) |
740 | m = -m; |
881 | m = -m; |
… | |
… | |
851 | #endif |
992 | #endif |
852 | |
993 | |
853 | return r; |
994 | return r; |
854 | } |
995 | } |
855 | |
996 | |
856 | #endif |
997 | /* convert a float to ieee half/binary16 */ |
|
|
998 | ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); |
|
|
999 | ecb_function_ ecb_const uint16_t |
|
|
1000 | ecb_float_to_binary16 (float x) |
|
|
1001 | { |
|
|
1002 | return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); |
|
|
1003 | } |
857 | |
1004 | |
858 | #endif |
1005 | /* convert an ieee half/binary16 to float */ |
|
|
1006 | ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); |
|
|
1007 | ecb_function_ ecb_const float |
|
|
1008 | ecb_binary16_to_float (uint16_t x) |
|
|
1009 | { |
|
|
1010 | return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); |
|
|
1011 | } |
859 | |
1012 | |
|
|
1013 | #endif |
|
|
1014 | |
|
|
1015 | #endif |
|
|
1016 | |