--- libecb/ecb.h 2011/08/24 23:28:47 1.67 +++ libecb/ecb.h 2012/06/07 21:28:36 1.99 @@ -1,7 +1,7 @@ /* * libecb - http://software.schmorp.de/pkg/libecb * - * Copyright (©) 2009-2011 Marc Alexander Lehmann + * Copyright (©) 2009-2012 Marc Alexander Lehmann * Copyright (©) 2011 Emanuele Giaquinta * All rights reserved. * @@ -30,6 +30,9 @@ #ifndef ECB_H #define ECB_H +/* 16 bits major, 16 bits minor */ +#define ECB_VERSION 0x00010001 + #ifdef _WIN32 typedef signed char int8_t; typedef unsigned char uint8_t; @@ -44,8 +47,23 @@ typedef signed __int64 int64_t; typedef unsigned __int64 uint64_t; #endif + #ifdef _WIN64 + #define ECB_PTRSIZE 8 + typedef uint64_t uintptr_t; + typedef int64_t intptr_t; + #else + #define ECB_PTRSIZE 4 + typedef uint32_t uintptr_t; + typedef int32_t intptr_t; + #endif + typedef intptr_t ptrdiff_t; #else #include + #if UINTMAX_MAX > 0xffffffffU + #define ECB_PTRSIZE 8 + #else + #define ECB_PTRSIZE 4 + #endif #endif /* many compilers define _GNUC_ to some versions but then only implement @@ -56,57 +74,108 @@ * an issue with that they should have done it right in the first place. */ #ifndef ECB_GCC_VERSION - #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) + #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ #define ECB_GCC_VERSION(major,minor) 0 #else #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) #endif #endif +#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ +#define ECB_C99 (__STDC_VERSION__ >= 199901L) +#define ECB_C11 (__STDC_VERSION__ >= 201112L) +#define ECB_CPP (__cplusplus+0) +#define ECB_CPP11 (__cplusplus >= 201103L) + /*****************************************************************************/ /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ -#if ECB_NO_THREADS || ECB_NO_SMP +#if ECB_NO_THREADS + #define ECB_NO_SMP 1 +#endif + +#if ECB_NO_SMP #define ECB_MEMORY_FENCE do { } while (0) #endif #ifndef ECB_MEMORY_FENCE - #if ECB_GCC_VERSION(2,5) - #if __i386__ + #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 + #if __i386 || __i386__ #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") - #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE /* non-lock xchg might be enough */ - #define ECB_MEMORY_FENCE_RELEASE do { } while (0) /* unlikely to change in future cpus */ - #elif __amd64 - #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") - #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("lfence" : : : "memory") - #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("sfence") /* play safe - not needed in any current cpu */ + #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") + #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") + #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") + #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") + #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ - #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") - #elif defined(__ARM_ARCH_6__ ) || defined(__ARM_ARCH_6J__ ) \ - || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6ZK__) - #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") - #elif defined(__ARM_ARCH_7__ ) || defined(__ARM_ARCH_7A__ ) \ - || defined(__ARM_ARCH_7M__) || defined(__ARM_ARCH_7R__ ) - #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") + #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ + || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") + #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ + || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") + #elif __sparc || __sparc__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") + #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") + #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") + #elif defined __s390__ || defined __s390x__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") + #elif defined __mips__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") + #elif defined __alpha__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") + #elif defined __hppa__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") + #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") + #elif defined __ia64__ + #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") #endif #endif #endif #ifndef ECB_MEMORY_FENCE - #if ECB_GCC_VERSION(4,4) || defined(__INTEL_COMPILER) + #if ECB_GCC_VERSION(4,7) + /* see comment below (stdatomic.h) about the C11 memory model. */ + #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) + #elif defined __clang && __has_feature (cxx_atomic) + /* see comment below (stdatomic.h) about the C11 memory model. */ + #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) + #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ #define ECB_MEMORY_FENCE __sync_synchronize () - /*#define ECB_MEMORY_FENCE_ACQUIRE ({ char dummy = 0; __sync_lock_test_and_set (&dummy, 1); }) */ - /*#define ECB_MEMORY_FENCE_RELEASE ({ char dummy = 1; __sync_lock_release (&dummy ); }) */ #elif _MSC_VER >= 1400 /* VC++ 2005 */ #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) #define ECB_MEMORY_FENCE _ReadWriteBarrier () #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () - #elif defined(_WIN32) + #elif defined _WIN32 #include #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ + #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 + #include + #define ECB_MEMORY_FENCE __machine_rw_barrier () + #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () + #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () + #elif __xlC__ + #define ECB_MEMORY_FENCE __sync () + #endif +#endif + +#ifndef ECB_MEMORY_FENCE + #if ECB_C11 && !defined __STDC_NO_ATOMICS__ + /* we assume that these memory fences work on all variables/all memory accesses, */ + /* not just C11 atomics and atomic accesses */ + #include + /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ + /* any fence other than seq_cst, which isn't very efficient for us. */ + /* Why that is, we don't know - either the C11 memory model is quite useless */ + /* for most usages, or gcc and clang have a bug */ + /* I *currently* lean towards the latter, and inefficiently implement */ + /* all three of ecb's fences as a seq_cst fence */ + #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) #endif #endif @@ -128,18 +197,16 @@ #endif #endif -#if !defined(ECB_MEMORY_FENCE_ACQUIRE) && defined(ECB_MEMORY_FENCE) +#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE #endif -#if !defined(ECB_MEMORY_FENCE_RELEASE) && defined(ECB_MEMORY_FENCE) +#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE #endif /*****************************************************************************/ -#define ECB_C99 (__STDC_VERSION__ >= 199901L) - #if __cplusplus #define ecb_inline static inline #elif ECB_GCC_VERSION(2,5) @@ -187,11 +254,16 @@ #endif #define ecb_noinline ecb_attribute ((__noinline__)) -#define ecb_noreturn ecb_attribute ((__noreturn__)) #define ecb_unused ecb_attribute ((__unused__)) #define ecb_const ecb_attribute ((__const__)) #define ecb_pure ecb_attribute ((__pure__)) +#if ECB_C11 + #define ecb_noreturn _Noreturn +#else + #define ecb_noreturn ecb_attribute ((__noreturn__)) +#endif + #if ECB_GCC_VERSION(4,3) #define ecb_artificial ecb_attribute ((__artificial__)) #define ecb_hot ecb_attribute ((__hot__)) @@ -291,6 +363,41 @@ } #endif +ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; +ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } +ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; +ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } + +ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; +ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) +{ + return ( (x * 0x0802U & 0x22110U) + | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; +} + +ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; +ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) +{ + x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); + x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); + x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); + x = ( x >> 8 ) | ( x << 8); + + return x; +} + +ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; +ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) +{ + x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); + x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); + x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); + x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); + x = ( x >> 16 ) | ( x << 16); + + return x; +} + /* popcount64 is only available on 64 bit cpus as gcc builtin */ /* so for this version we are lazy */ ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; @@ -349,25 +456,43 @@ #define ecb_unreachable() __builtin_unreachable () #else /* this seems to work fine, but gcc always emits a warning for it :/ */ - ecb_function_ void ecb_unreachable (void) ecb_noreturn; - ecb_function_ void ecb_unreachable (void) { } + ecb_inline void ecb_unreachable (void) ecb_noreturn; + ecb_inline void ecb_unreachable (void) { } #endif /* try to tell the compiler that some condition is definitely true */ #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) -ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; -ecb_function_ unsigned char +ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; +ecb_inline unsigned char ecb_byteorder_helper (void) { - const uint32_t u = 0x11223344; - return *(unsigned char *)&u; + /* the union code still generates code under pressure in gcc, */ + /* but less than using pointers, and always seem to */ + /* successfully return a constant. */ + /* the reason why we have this horrible preprocessor mess */ + /* is to avoid it in all cases, at least on common architectures */ + /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ +#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 + return 0x44; +#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ + return 0x44; +#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ + retrurn 0x11; +#else + union + { + uint32_t i; + uint8_t c; + } u = { 0x11223344 }; + return u.c; +#endif } -ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; -ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } -ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; -ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } +ecb_inline ecb_bool ecb_big_endian (void) ecb_const; +ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } +ecb_inline ecb_bool ecb_little_endian (void) ecb_const; +ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } #if ECB_GCC_VERSION(3,0) || ECB_C99 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) @@ -375,8 +500,21 @@ #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) #endif -#define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) -#define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) +#if __cplusplus + template + static inline T ecb_div_rd (T val, T div) + { + return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; + } + template + static inline T ecb_div_ru (T val, T div) + { + return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; + } +#else + #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) + #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) +#endif #if ecb_cplusplus_does_not_suck /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */