--- CBOR-XS/ecb.h 2015/04/27 20:21:53 1.11 +++ CBOR-XS/ecb.h 2020/11/29 21:31:58 1.16 @@ -1,7 +1,7 @@ /* * libecb - http://software.schmorp.de/pkg/libecb * - * Copyright (©) 2009-2015 Marc Alexander Lehmann + * Copyright (©) 2009-2015,2018-2020 Marc Alexander Lehmann * Copyright (©) 2011 Emanuele Giaquinta * All rights reserved. * @@ -42,15 +42,23 @@ #define ECB_H /* 16 bits major, 16 bits minor */ -#define ECB_VERSION 0x00010004 +#define ECB_VERSION 0x00010008 -#ifdef _WIN32 +#include /* for memcpy */ + +#if defined (_WIN32) && !defined (__MINGW32__) typedef signed char int8_t; typedef unsigned char uint8_t; + typedef signed char int_fast8_t; + typedef unsigned char uint_fast8_t; typedef signed short int16_t; typedef unsigned short uint16_t; + typedef signed int int_fast16_t; + typedef unsigned int uint_fast16_t; typedef signed int int32_t; typedef unsigned int uint32_t; + typedef signed int int_fast32_t; + typedef unsigned int uint_fast32_t; #if __GNUC__ typedef signed long long int64_t; typedef unsigned long long uint64_t; @@ -58,6 +66,8 @@ typedef signed __int64 int64_t; typedef unsigned __int64 uint64_t; #endif + typedef int64_t int_fast64_t; + typedef uint64_t uint_fast64_t; #ifdef _WIN64 #define ECB_PTRSIZE 8 typedef uint64_t uintptr_t; @@ -69,15 +79,26 @@ #endif #else #include - #if UINTMAX_MAX > 0xffffffffU + #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU #define ECB_PTRSIZE 8 #else #define ECB_PTRSIZE 4 #endif #endif +#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) +#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) + +#ifndef ECB_OPTIMIZE_SIZE + #if __OPTIMIZE_SIZE__ + #define ECB_OPTIMIZE_SIZE 1 + #else + #define ECB_OPTIMIZE_SIZE 0 + #endif +#endif + /* work around x32 idiocy by defining proper macros */ -#if __amd64 || __x86_64 || _M_AMD64 || _M_X64 +#if ECB_GCC_AMD64 || ECB_MSVC_AMD64 #if _ILP32 #define ECB_AMD64_X32 1 #else @@ -114,6 +135,8 @@ #define ECB_CPP (__cplusplus+0) #define ECB_CPP11 (__cplusplus >= 201103L) +#define ECB_CPP14 (__cplusplus >= 201402L) +#define ECB_CPP17 (__cplusplus >= 201703L) #if ECB_CPP #define ECB_C 0 @@ -125,6 +148,7 @@ #define ECB_C99 (ECB_STDC_VERSION >= 199901L) #define ECB_C11 (ECB_STDC_VERSION >= 201112L) +#define ECB_C17 (ECB_STDC_VERSION >= 201710L) #if ECB_CPP #define ECB_EXTERN_C extern "C" @@ -149,27 +173,45 @@ #define ECB_MEMORY_FENCE do { } while (0) #endif +/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ +#if __xlC__ && ECB_CPP + #include +#endif + +#if 1400 <= _MSC_VER + #include /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ +#endif + #ifndef ECB_MEMORY_FENCE #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 + #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory") #if __i386 || __i386__ #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") #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_RELEASE __asm__ __volatile__ ("" : : : "memory") + #elif ECB_GCC_AMD64 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") - #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") + #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") + #elif defined __ARM_ARCH_2__ \ + || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ + || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ + || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \ + || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \ + || defined __ARM_ARCH_5TEJ__ + /* should not need any, unless running old code on newer cpu - arm doesn't support that */ #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ - || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ + || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ + || defined __ARM_ARCH_6T2__ #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__ + || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") #elif __aarch64__ #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") - #elif (__sparc || __sparc__) && !__sparcv8 + #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) #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") @@ -202,12 +244,14 @@ #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) + #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) #elif ECB_CLANG_EXTENSION(c_atomic) /* see comment below (stdatomic.h) about the C11 memory model. */ #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) + #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ #define ECB_MEMORY_FENCE __sync_synchronize () @@ -227,9 +271,10 @@ #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 () + #define ECB_MEMORY_FENCE __machine_rw_barrier () + #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () + #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier () + #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier () #elif __xlC__ #define ECB_MEMORY_FENCE __sync () #endif @@ -240,15 +285,9 @@ /* 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 */ - /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ - /* for all __atomic_thread_fence's except seq_cst */ #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) + #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire) + #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release) #endif #endif @@ -278,6 +317,10 @@ #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE #endif +#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE + #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */ +#endif + /*****************************************************************************/ #if ECB_CPP @@ -351,7 +394,7 @@ #define ecb_deprecated ecb_attribute ((__deprecated__)) #endif -#if __MSC_VER >= 1500 +#if _MSC_VER >= 1500 #define ecb_deprecated_message(msg) __declspec (deprecated (msg)) #elif ECB_GCC_VERSION(4,5) #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg)) @@ -369,13 +412,13 @@ #define ecb_const ecb_attribute ((__const__)) #define ecb_pure ecb_attribute ((__pure__)) -/* TODO http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */ #if ECB_C11 || __IBMC_NORETURN - /* http://pic.dhe.ibm.com/infocenter/compbg/v121v141/topic/com.ibm.xlcpp121.bg.doc/language_ref/noreturn.html */ + /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */ #define ecb_noreturn _Noreturn #elif ECB_CPP11 #define ecb_noreturn [[noreturn]] #elif _MSC_VER >= 1200 + /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */ #define ecb_noreturn __declspec (noreturn) #else #define ecb_noreturn ecb_attribute ((__noreturn__)) @@ -417,6 +460,11 @@ ecb_function_ ecb_const int ecb_ctz32 (uint32_t x) { +#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) + unsigned long r; + _BitScanForward (&r, x); + return (int)r; +#else int r = 0; x &= ~x + 1; /* this isolates the lowest bit */ @@ -436,14 +484,21 @@ #endif return r; +#endif } ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); ecb_function_ ecb_const int ecb_ctz64 (uint64_t x) { - int shift = x & 0xffffffffU ? 0 : 32; +#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) + unsigned long r; + _BitScanForward64 (&r, x); + return (int)r; +#else + int shift = x & 0xffffffff ? 0 : 32; return ecb_ctz32 (x >> shift) + shift; +#endif } ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); @@ -461,6 +516,11 @@ ecb_function_ ecb_const int ecb_ld32 (uint32_t x); ecb_function_ ecb_const int ecb_ld32 (uint32_t x) { +#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) + unsigned long r; + _BitScanReverse (&r, x); + return (int)r; +#else int r = 0; if (x >> 16) { x >>= 16; r += 16; } @@ -470,16 +530,23 @@ if (x >> 1) { r += 1; } return r; +#endif } ecb_function_ ecb_const int ecb_ld64 (uint64_t x); ecb_function_ ecb_const int ecb_ld64 (uint64_t x) { +#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) + unsigned long r; + _BitScanReverse64 (&r, x); + return (int)r; +#else int r = 0; if (x >> 32) { x >>= 32; r += 32; } return r + ecb_ld32 (x); +#endif } #endif @@ -545,10 +612,57 @@ ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } +#if ECB_CPP + +inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } +inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } +inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); } +inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); } + +inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); } +inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); } +inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); } +inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); } + +inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); } +inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); } +inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); } +inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); } + +inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); } +inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); } +inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); } +inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); } + +inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); } +inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); } +inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); } + +inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); } +inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); } +inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); } +inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); } + +inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); } +inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); } +inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); } +inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); } + +#endif + #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) + #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) + #define ecb_bswap16(x) __builtin_bswap16 (x) + #else #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) + #endif #define ecb_bswap32(x) __builtin_bswap32 (x) #define ecb_bswap64(x) __builtin_bswap64 (x) +#elif _MSC_VER + #include + #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x))) + #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x))) + #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x))) #else ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x); ecb_function_ ecb_const uint16_t @@ -583,8 +697,8 @@ /* try to tell the compiler that some condition is definitely true */ #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 -ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); -ecb_inline ecb_const unsigned char +ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); +ecb_inline ecb_const uint32_t ecb_byteorder_helper (void) { /* the union code still generates code under pressure in gcc, */ @@ -593,28 +707,104 @@ /* the reason why we have this horrible preprocessor mess */ /* is to avoid it in all cases, at least on common architectures */ /* or when using a recent enough gcc version (>= 4.6) */ -#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__ - return 0x11; +#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ + || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) + #define ECB_LITTLE_ENDIAN 1 + return 0x44332211; +#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ + || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) + #define ECB_BIG_ENDIAN 1 + return 0x11223344; #else union { - uint32_t i; - uint8_t c; - } u = { 0x11223344 }; - return u.c; + uint8_t c[4]; + uint32_t u; + } u = { 0x11, 0x22, 0x33, 0x44 }; + return u.u; #endif } ecb_inline ecb_const ecb_bool ecb_big_endian (void); -ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } +ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } ecb_inline ecb_const ecb_bool ecb_little_endian (void); -ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } +ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } + +/*****************************************************************************/ +/* unaligned load/store */ + +ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } +ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } +ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } + +ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } +ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } +ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } + +ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; } +ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; } +ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; } + +ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); } +ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); } +ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); } + +ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); } +ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); } +ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); } + +ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } +ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } +ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } + +ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } +ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } +ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } + +ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); } +ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); } +ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } + +ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } +ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } +ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } + +ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } +ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } +ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } + +#if ECB_CPP + +inline uint8_t ecb_bswap (uint8_t v) { return v; } +inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); } +inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); } +inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); } + +template inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } +template inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } +template inline T ecb_peek (const void *ptr) { return *(const T *)ptr; } +template inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek (ptr)); } +template inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek (ptr)); } +template inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; } +template inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u (ptr)); } +template inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u (ptr)); } + +template inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } +template inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } +template inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; } +template inline void ecb_poke_be (void *ptr, T v) { return ecb_poke (ptr, ecb_host_to_be (v)); } +template inline void ecb_poke_le (void *ptr, T v) { return ecb_poke (ptr, ecb_host_to_le (v)); } +template inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } +template inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u (ptr, ecb_host_to_be (v)); } +template inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u (ptr, ecb_host_to_le (v)); } + +#endif + +/*****************************************************************************/ +/* division */ #if ECB_GCC_VERSION(3,0) || ECB_C99 + /* C99 tightened the definition of %, so we can use a more efficient version */ #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) #else #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) @@ -636,6 +826,9 @@ #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) #endif +/*****************************************************************************/ +/* array length */ + #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) */ template @@ -647,6 +840,105 @@ #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) #endif +/*****************************************************************************/ +/* IEEE 754-2008 half float conversions */ + +ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); +ecb_function_ ecb_const uint32_t +ecb_binary16_to_binary32 (uint32_t x) +{ + unsigned int s = (x & 0x8000) << (31 - 15); + int e = (x >> 10) & 0x001f; + unsigned int m = x & 0x03ff; + + if (ecb_expect_false (e == 31)) + /* infinity or NaN */ + e = 255 - (127 - 15); + else if (ecb_expect_false (!e)) + { + if (ecb_expect_true (!m)) + /* zero, handled by code below by forcing e to 0 */ + e = 0 - (127 - 15); + else + { + /* subnormal, renormalise */ + unsigned int s = 10 - ecb_ld32 (m); + + m = (m << s) & 0x3ff; /* mask implicit bit */ + e -= s - 1; + } + } + + /* e and m now are normalised, or zero, (or inf or nan) */ + e += 127 - 15; + + return s | (e << 23) | (m << (23 - 10)); +} + +ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); +ecb_function_ ecb_const uint16_t +ecb_binary32_to_binary16 (uint32_t x) +{ + unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ + unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ + unsigned int m = x & 0x007fffff; + + x &= 0x7fffffff; + + /* if it's within range of binary16 normals, use fast path */ + if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff)) + { + /* mantissa round-to-even */ + m += 0x00000fff + ((m >> (23 - 10)) & 1); + + /* handle overflow */ + if (ecb_expect_false (m >= 0x00800000)) + { + m >>= 1; + e += 1; + } + + return s | (e << 10) | (m >> (23 - 10)); + } + + /* handle large numbers and infinity */ + if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) + return s | 0x7c00; + + /* handle zero, subnormals and small numbers */ + if (ecb_expect_true (x < 0x38800000)) + { + /* zero */ + if (ecb_expect_true (!x)) + return s; + + /* handle subnormals */ + + /* too small, will be zero */ + if (e < (14 - 24)) /* might not be sharp, but is good enough */ + return s; + + m |= 0x00800000; /* make implicit bit explicit */ + + /* very tricky - we need to round to the nearest e (+10) bit value */ + { + unsigned int bits = 14 - e; + unsigned int half = (1 << (bits - 1)) - 1; + unsigned int even = (m >> bits) & 1; + + /* if this overflows, we will end up with a normalised number */ + m = (m + half + even) >> bits; + } + + return s | m; + } + + /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ + m >>= 13; + + return s | 0x7c00 | m | !m; +} + /*******************************************************************************/ /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ @@ -654,7 +946,7 @@ /* the only noteworthy exception is ancient armle, which uses order 43218765 */ #if 0 \ || __i386 || __i386__ \ - || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ + || ECB_GCC_AMD64 \ || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ || defined __s390__ || defined __s390x__ \ || defined __mips__ \ @@ -664,11 +956,10 @@ || defined __m68k__ \ || defined __m88k__ \ || defined __sh__ \ - || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \ + || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ || defined __aarch64__ #define ECB_STDFP 1 - #include /* for memcpy */ #else #define ECB_STDFP 0 #endif @@ -692,27 +983,12 @@ #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L #define ecb_ldexpf(x,e) ldexpf ((x), (e)) + #define ecb_frexpf(x,e) frexpf ((x), (e)) #else - #define ecb_ldexpf(x,e) (float) ldexp ((x), (e)) + #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e)) + #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e)) #endif - /* converts an ieee half/binary16 to a float */ - ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); - ecb_function_ ecb_const float - ecb_binary16_to_float (uint16_t x) - { - int e = (x >> 10) & 0x1f; - int m = x & 0x3ff; - float r; - - if (!e ) r = ecb_ldexpf (m , -24); - else if (e != 31) r = ecb_ldexpf (m + 0x400, e - 25); - else if (m ) r = ECB_NAN; - else r = ECB_INFINITY; - - return x & 0x8000 ? -r : r; - } - /* convert a float to ieee single/binary32 */ ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x); ecb_function_ ecb_const uint32_t @@ -732,7 +1008,7 @@ if (x < -3.40282346638528860e+38f) return 0xff800000U; if (x != x ) return 0x7fbfffffU; - m = frexpf (x, &e) * 0x1000000U; + m = ecb_frexpf (x, &e) * 0x1000000U; r = m & 0x80000000U; @@ -853,6 +1129,22 @@ return r; } + /* convert a float to ieee half/binary16 */ + ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x); + ecb_function_ ecb_const uint16_t + ecb_float_to_binary16 (float x) + { + return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x)); + } + + /* convert an ieee half/binary16 to float */ + ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x); + ecb_function_ ecb_const float + ecb_binary16_to_float (uint16_t x) + { + return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); + } + #endif #endif