1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
6 | * All rights reserved. |
6 | * All rights reserved. |
7 | * |
7 | * |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
… | |
… | |
54 | * or so. |
54 | * or so. |
55 | * we try to detect these and simply assume they are not gcc - if they have |
55 | * we try to detect these and simply assume they are not gcc - if they have |
56 | * an issue with that they should have done it right in the first place. |
56 | * an issue with that they should have done it right in the first place. |
57 | */ |
57 | */ |
58 | #ifndef ECB_GCC_VERSION |
58 | #ifndef ECB_GCC_VERSION |
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
59 | #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ |
60 | #define ECB_GCC_VERSION(major,minor) 0 |
60 | #define ECB_GCC_VERSION(major,minor) 0 |
61 | #else |
61 | #else |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
63 | #endif |
63 | #endif |
64 | #endif |
64 | #endif |
… | |
… | |
66 | /*****************************************************************************/ |
66 | /*****************************************************************************/ |
67 | |
67 | |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
68 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
69 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
69 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
70 | |
70 | |
|
|
71 | #if ECB_NO_THREADS |
|
|
72 | # define ECB_NO_SMP 1 |
|
|
73 | #endif |
|
|
74 | |
71 | #if ECB_NO_THREADS || ECB_NO_SMP |
75 | #if ECB_NO_THREADS || ECB_NO_SMP |
72 | #define ECB_MEMORY_FENCE do { } while (0) |
76 | #define ECB_MEMORY_FENCE do { } while (0) |
73 | #endif |
77 | #endif |
74 | |
78 | |
75 | #ifndef ECB_MEMORY_FENCE |
79 | #ifndef ECB_MEMORY_FENCE |
76 | #if ECB_GCC_VERSION(2,5) || defined(__INTEL_COMPILER) || defined(__clang__) |
80 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
77 | #if __i386__ |
81 | #if __i386 || __i386__ |
78 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
79 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
83 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ |
80 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
84 | #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ |
81 | #elif __amd64 |
85 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
82 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
83 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
87 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") |
84 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
88 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ |
85 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
89 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
86 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
90 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
87 | #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ |
91 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
88 | || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) |
92 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
89 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
93 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
90 | #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ |
94 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
91 | || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) |
95 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
92 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
96 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
97 | #elif __sparc || __sparc__ |
|
|
98 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad | " : : : "memory") |
|
|
99 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
100 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
101 | #elif defined __s390__ || defined __s390x__ |
|
|
102 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
103 | #elif defined __mips__ |
|
|
104 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
105 | #elif defined __alpha__ |
|
|
106 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
93 | #endif |
107 | #endif |
94 | #endif |
108 | #endif |
95 | #endif |
109 | #endif |
96 | |
110 | |
97 | #ifndef ECB_MEMORY_FENCE |
111 | #ifndef ECB_MEMORY_FENCE |
98 | #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) || defined(__clang__) |
112 | #if ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
99 | #define ECB_MEMORY_FENCE __sync_synchronize () |
113 | #define ECB_MEMORY_FENCE __sync_synchronize () |
100 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
114 | /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ |
101 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
115 | /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ |
102 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
116 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
103 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
117 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
104 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
118 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
105 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
119 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
106 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
120 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
107 | #elif defined(_WIN32) |
121 | #elif defined _WIN32 |
108 | #include <WinNT.h> |
122 | #include <WinNT.h> |
109 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
123 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
124 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
125 | #include <mbarrier.h> |
|
|
126 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
127 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
128 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
129 | #elif __xlC__ |
|
|
130 | #define ECB_MEMORY_FENCE __sync () |
110 | #endif |
131 | #endif |
111 | #endif |
132 | #endif |
112 | |
133 | |
113 | #ifndef ECB_MEMORY_FENCE |
134 | #ifndef ECB_MEMORY_FENCE |
114 | #if !ECB_AVOID_PTHREADS |
135 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
126 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
147 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
127 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
148 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
128 | #endif |
149 | #endif |
129 | #endif |
150 | #endif |
130 | |
151 | |
131 | #if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) |
152 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
132 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
153 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
133 | #endif |
154 | #endif |
134 | |
155 | |
135 | #if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) |
156 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
136 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
157 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
137 | #endif |
158 | #endif |
138 | |
159 | |
139 | /*****************************************************************************/ |
160 | /*****************************************************************************/ |
140 | |
161 | |
… | |
… | |
289 | |
310 | |
290 | return r + ecb_ld32 (x); |
311 | return r + ecb_ld32 (x); |
291 | } |
312 | } |
292 | #endif |
313 | #endif |
293 | |
314 | |
|
|
315 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
316 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
317 | { |
|
|
318 | return ( (x * 0x0802U & 0x22110U) |
|
|
319 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
320 | } |
|
|
321 | |
|
|
322 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
323 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
324 | { |
|
|
325 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
326 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
327 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
328 | x = ( x >> 8 ) | ( x << 8); |
|
|
329 | |
|
|
330 | return x; |
|
|
331 | } |
|
|
332 | |
|
|
333 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
334 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
335 | { |
|
|
336 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
337 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
338 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
339 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
340 | x = ( x >> 16 ) | ( x << 16); |
|
|
341 | |
|
|
342 | return x; |
|
|
343 | } |
|
|
344 | |
294 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
345 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
295 | /* so for this version we are lazy */ |
346 | /* so for this version we are lazy */ |
296 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
347 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
297 | ecb_function_ int |
348 | ecb_function_ int |
298 | ecb_popcount64 (uint64_t x) |
349 | ecb_popcount64 (uint64_t x) |
… | |
… | |
347 | |
398 | |
348 | #if ECB_GCC_VERSION(4,5) |
399 | #if ECB_GCC_VERSION(4,5) |
349 | #define ecb_unreachable() __builtin_unreachable () |
400 | #define ecb_unreachable() __builtin_unreachable () |
350 | #else |
401 | #else |
351 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
402 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
352 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
403 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
353 | ecb_function_ void ecb_unreachable (void) { } |
404 | ecb_inline void ecb_unreachable (void) { } |
354 | #endif |
405 | #endif |
355 | |
406 | |
356 | /* try to tell the compiler that some condition is definitely true */ |
407 | /* try to tell the compiler that some condition is definitely true */ |
357 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
408 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
358 | |
409 | |
359 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
410 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
360 | ecb_function_ unsigned char |
411 | ecb_inline unsigned char |
361 | ecb_byteorder_helper (void) |
412 | ecb_byteorder_helper (void) |
362 | { |
413 | { |
363 | const uint32_t u = 0x11223344; |
414 | const uint32_t u = 0x11223344; |
364 | return *(unsigned char *)&u; |
415 | return *(unsigned char *)&u; |
365 | } |
416 | } |
366 | |
417 | |
367 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
418 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
368 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
419 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
369 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
420 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
370 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
421 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
371 | |
422 | |
372 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
423 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
373 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
424 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
374 | #else |
425 | #else |
375 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
426 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |