--- libecb/ecb.h 2019/06/24 22:22:59 1.178 +++ libecb/ecb.h 2022/03/25 15:36:36 1.214 @@ -1,7 +1,7 @@ /* * libecb - http://software.schmorp.de/pkg/libecb * - * Copyright (©) 2009-2015 Marc Alexander Lehmann + * Copyright (©) 2009-2015,2018-2021 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 0x00010006 +#define ECB_VERSION 0x0001000c -#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; @@ -79,6 +89,14 @@ #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 ECB_GCC_AMD64 || ECB_MSVC_AMD64 #if _ILP32 @@ -88,6 +106,12 @@ #endif #endif +#if ECB_PTRSIZE >= 8 || ECB_AMD64_X32 + #define ECB_64BIT_NATIVE 1 +#else + #define ECB_64BIT_NATIVE 0 +#endif + /* many compilers define _GNUC_ to some versions but then only implement * what their idiot authors think are the "more important" extensions, * causing enormous grief in return for some better fake benchmark numbers. @@ -226,6 +250,7 @@ #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) + #undef ECB_MEMORY_FENCE_RELAXED #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED) #elif ECB_CLANG_EXTENSION(c_atomic) @@ -233,6 +258,7 @@ #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) + #undef ECB_MEMORY_FENCE_RELAXED #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED) #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ @@ -331,6 +357,10 @@ #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) +/* This marks larger functions that do not neccessarily need to be inlined */ +/* The idea is to possibly compile the header twice, */ +/* once exposing only the declarations, another time to define external functions */ +/* TODO: possibly static would be best for these at the moment? */ #define ecb_function_ ecb_inline #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) @@ -430,34 +460,53 @@ || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ && ECB_CLANG_BUILTIN(__builtin_popcount)) - /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ - #define ecb_ld32(x) (__builtin_clz (x) ^ 31) - #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) - #define ecb_ctz32(x) __builtin_ctz (x) - #define ecb_ctz64(x) __builtin_ctzll (x) + #define ecb_ctz32(x) __builtin_ctz (x) + #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x)) + #define ecb_clz32(x) __builtin_clz (x) + #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x)) + #define ecb_ld32(x) (ecb_clz32 (x) ^ 31) + #define ecb_ld64(x) (ecb_clz64 (x) ^ 63) #define ecb_popcount32(x) __builtin_popcount (x) - /* no popcountll */ + /* ecb_popcount64 is more difficult, see below */ #else ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); - ecb_function_ ecb_const int - ecb_ctz32 (uint32_t x) + 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; + int r; x &= ~x + 1; /* this isolates the lowest bit */ -#if ECB_branchless_on_i386 + #if 1 + /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */ + /* This happens to return 32 for x == 0, but the API does not support this */ + + /* -0 marks unused entries */ + static unsigned char table[64] = + { + 32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14, + 10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15, + 31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26, + 30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0 + }; + + /* magic constant results in 33 unique values in the upper 6 bits */ + x *= 0x0450fbafU; /* == 17 * 65 * 65535 */ + + r = table [x >> 26]; + #elif 0 /* branchless on i386, typically */ + r = 0; r += !!(x & 0xaaaaaaaa) << 0; r += !!(x & 0xcccccccc) << 1; r += !!(x & 0xf0f0f0f0) << 2; r += !!(x & 0xff00ff00) << 3; r += !!(x & 0xffff0000) << 4; -#else + #else /* branchless on modern compilers, typically */ + r = 0; if (x & 0xaaaaaaaa) r += 1; if (x & 0xcccccccc) r += 2; if (x & 0xf0f0f0f0) r += 4; @@ -470,8 +519,7 @@ } ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); - ecb_function_ ecb_const int - ecb_ctz64 (uint64_t x) + ecb_function_ ecb_const int ecb_ctz64 (uint64_t x) { #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) unsigned long r; @@ -483,6 +531,55 @@ #endif } + ecb_function_ ecb_const int ecb_clz32 (uint32_t x); + ecb_function_ ecb_const int ecb_clz32 (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 + + /* Robert Harley's algorithm from comp.arch 1996-12-07 */ + /* This happens to return 32 for x == 0, but the API does not support this */ + + /* -0 marks unused table elements */ + static unsigned char table[64] = + { + 32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0, + -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0, + 17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18, + 5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0 + }; + + /* propagate leftmost 1 bit to the right */ + x |= x >> 1; + x |= x >> 2; + x |= x >> 4; + x |= x >> 8; + x |= x >> 16; + + /* magic constant results in 33 unique values in the upper 6 bits */ + x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */ + + return table [x >> 26]; +#endif + } + + ecb_function_ ecb_const int ecb_clz64 (uint64_t x); + ecb_function_ ecb_const int ecb_clz64 (uint64_t x) + { +#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) + unsigned long r; + _BitScanReverse64 (&r, x); + return (int)r; +#else + uint32_t l = x >> 32; + int shift = l ? 0 : 32; + return ecb_clz32 (l ? l : x) + shift; +#endif + } + ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); ecb_function_ ecb_const int ecb_popcount32 (uint32_t x) @@ -567,13 +664,16 @@ return x; } -/* popcount64 is only available on 64 bit cpus as gcc builtin */ -/* so for this version we are lazy */ ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); -ecb_function_ ecb_const int -ecb_popcount64 (uint64_t x) +ecb_function_ ecb_const int ecb_popcount64 (uint64_t x) { + /* popcount64 is only available on 64 bit cpus as gcc builtin. */ + /* also, gcc/clang make this surprisingly difficult to use */ +#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl)) + return __builtin_popcountl (x); +#else return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); +#endif } ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); @@ -585,14 +685,52 @@ ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); -ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } -ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } -ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } -ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } -ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } -ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } -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); } +ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); } +ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); } +ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); } +ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); } +ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); } +ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); } +ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); } +ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); } + +#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) @@ -609,22 +747,19 @@ #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 - ecb_bswap16 (uint16_t x) + ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x) { return ecb_rotl16 (x, 8); } ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x); - ecb_function_ ecb_const uint32_t - ecb_bswap32 (uint32_t x) + ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x) { return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); } ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x); - ecb_function_ ecb_const uint64_t - ecb_bswap64 (uint64_t x) + ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x) { return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); } @@ -642,8 +777,7 @@ #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); -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, */ /* but less than using pointers, and always seems to */ @@ -674,7 +808,374 @@ ecb_inline ecb_const ecb_bool ecb_little_endian (void); 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 + +/*****************************************************************************/ +/* pointer/integer hashing */ + +/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */ +ecb_function_ uint32_t ecb_mix32 (uint32_t v); +ecb_function_ uint32_t ecb_mix32 (uint32_t v) +{ + v ^= v >> 16; v *= 0x7feb352dU; + v ^= v >> 15; v *= 0x846ca68bU; + v ^= v >> 16; + return v; +} + +ecb_function_ uint32_t ecb_unmix32 (uint32_t v); +ecb_function_ uint32_t ecb_unmix32 (uint32_t v) +{ + v ^= v >> 16 ; v *= 0x43021123U; + v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U; + v ^= v >> 16 ; + return v; +} + +/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */ +ecb_function_ uint64_t ecb_mix64 (uint64_t v); +ecb_function_ uint64_t ecb_mix64 (uint64_t v) +{ + v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U; + v ^= v >> 27; v *= 0x94d049bb133111ebU; + v ^= v >> 31; + return v; +} + +ecb_function_ uint64_t ecb_unmix64 (uint64_t v); +ecb_function_ uint64_t ecb_unmix64 (uint64_t v) +{ + v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U; + v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U; + v ^= v >> 30 ^ v >> 60; + return v; +} + +ecb_function_ uintptr_t ecb_ptrmix (void *p); +ecb_function_ uintptr_t ecb_ptrmix (void *p) +{ + #if ECB_PTRSIZE <= 4 + return ecb_mix32 ((uint32_t)p); + #else + return ecb_mix64 ((uint64_t)p); + #endif +} + +ecb_function_ void *ecb_ptrunmix (uintptr_t v); +ecb_function_ void *ecb_ptrunmix (uintptr_t v) +{ + #if ECB_PTRSIZE <= 4 + return (void *)ecb_unmix32 (v); + #else + return (void *)ecb_unmix64 (v); + #endif +} + +#if ECB_CPP + +template +inline uintptr_t ecb_ptrmix (T *p) +{ + return ecb_ptrmix (static_cast(p)); +} + +template +inline T *ecb_ptrunmix (uintptr_t v) +{ + return static_cast(ecb_ptrunmix (v)); +} + +#endif + +/*****************************************************************************/ +/* gray code */ + +ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); } +ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); } +ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); } +ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); } + +ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g); +ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g) +{ + g ^= g >> 1; + g ^= g >> 2; + g ^= g >> 4; + + return g; +} + +ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g); +ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g) +{ + g ^= g >> 1; + g ^= g >> 2; + g ^= g >> 4; + g ^= g >> 8; + + return g; +} + +ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g); +ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g) +{ + g ^= g >> 1; + g ^= g >> 2; + g ^= g >> 4; + g ^= g >> 8; + g ^= g >> 16; + + return g; +} + +ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g); +ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g) +{ + g ^= g >> 1; + g ^= g >> 2; + g ^= g >> 4; + g ^= g >> 8; + g ^= g >> 16; + g ^= g >> 32; + + return g; +} + +#if ECB_CPP + +ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); } +ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); } +ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); } +ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); } + +ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); } +ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); } +ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); } +ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); } + +#endif + +/*****************************************************************************/ +/* 2d hilbert curves */ + +/* algorithm from the book Hacker's Delight, modified to not */ +/* run into undefined behaviour for n==16 */ +static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s); +static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s) +{ + uint32_t comp, swap, cs, t, sr; + + /* pad s on the left (unused) bits with 01 (no change groups) */ + s |= 0x55555555U << n << n; + /* "s shift right" */ + sr = (s >> 1) & 0x55555555U; + /* compute complement and swap info in two-bit groups */ + cs = ((s & 0x55555555U) + sr) ^ 0x55555555U; + + /* parallel prefix xor op to propagate both complement + * and swap info together from left to right (there is + * no step "cs ^= cs >> 1", so in effect it computes + * two independent parallel prefix operations on two + * interleaved sets of sixteen bits). + */ + cs ^= cs >> 2; + cs ^= cs >> 4; + cs ^= cs >> 8; + cs ^= cs >> 16; + + /* separate swap and complement bits */ + swap = cs & 0x55555555U; + comp = (cs >> 1) & 0x55555555U; + + /* calculate coordinates in odd and even bit positions */ + t = (s & swap) ^ comp; + s = s ^ sr ^ t ^ (t << 1); + + /* unpad/clear out any junk on the left */ + s = s & ((1 << n << n) - 1); + + /* Now "unshuffle" to separate the x and y bits. */ + t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1); + t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2); + t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4); + t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8); + + /* now s contains two 16-bit coordinates */ + return s; +} + +/* 64 bit, a straightforward extension to the 32 bit case */ +static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s); +static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s) +{ + uint64_t comp, swap, cs, t, sr; + + /* pad s on the left (unused) bits with 01 (no change groups) */ + s |= 0x5555555555555555U << n << n; + /* "s shift right" */ + sr = (s >> 1) & 0x5555555555555555U; + /* compute complement and swap info in two-bit groups */ + cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U; + + /* parallel prefix xor op to propagate both complement + * and swap info together from left to right (there is + * no step "cs ^= cs >> 1", so in effect it computes + * two independent parallel prefix operations on two + * interleaved sets of thirty-two bits). + */ + cs ^= cs >> 2; + cs ^= cs >> 4; + cs ^= cs >> 8; + cs ^= cs >> 16; + cs ^= cs >> 32; + + /* separate swap and complement bits */ + swap = cs & 0x5555555555555555U; + comp = (cs >> 1) & 0x5555555555555555U; + + /* calculate coordinates in odd and even bit positions */ + t = (s & swap) ^ comp; + s = s ^ sr ^ t ^ (t << 1); + + /* unpad/clear out any junk on the left */ + s = s & ((1 << n << n) - 1); + + /* Now "unshuffle" to separate the x and y bits. */ + t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1); + t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2); + t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4); + t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8); + t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16); + + /* now s contains two 32-bit coordinates */ + return s; +} + +/* algorithm from the book Hacker's Delight, but a similar algorithm*/ +/* is given in https://doi.org/10.1002/spe.4380160103 */ +/* this has been slightly improved over the original version */ +ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy); +ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy) +{ + uint32_t row; + uint32_t state = 0; + uint32_t s = 0; + + do + { + --n; + + row = 4 * state + | (2 & (xy >> n >> 15)) + | (1 & (xy >> n )); + + /* these funky constants are lookup tables for two-bit values */ + s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; + state = (0x8fe65831U >> 2 * row) & 3; + } + while (n > 0); + + return s; +} + +/* 64 bit, essentially the same as 32 bit */ +ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy); +ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy) +{ + uint32_t row; + uint32_t state = 0; + uint64_t s = 0; + + do + { + --n; + + row = 4 * state + | (2 & (xy >> n >> 31)) + | (1 & (xy >> n )); + + /* these funky constants are lookup tables for two-bit values */ + s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; + state = (0x8fe65831U >> 2 * row) & 3; + } + while (n > 0); + + return s; +} + +/*****************************************************************************/ +/* 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))) @@ -696,6 +1197,9 @@ #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 @@ -707,9 +1211,11 @@ #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) +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; @@ -740,11 +1246,10 @@ } 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) +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 */ + int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ unsigned int m = x & 0x007fffff; x &= 0x7fffffff; @@ -804,6 +1309,228 @@ } /*******************************************************************************/ +/* fast integer to ascii */ + +/* + * This code is pretty complicated because it is general. The idea behind it, + * however, is pretty simple: first, the number is multiplied with a scaling + * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point + * number with the first digit in the upper bits. + * Then this digit is converted to text and masked out. The resulting number + * is then multiplied by 10, by multiplying the fixed point representation + * by 5 and shifting the (binary) decimal point one to the right, so a 4.28 + * format becomes 5.27, 6.26 and so on. + * The rest involves only advancing the pointer if we already generated a + * non-zero digit, so leading zeroes are overwritten. + */ + +/* simply return a mask with "bits" bits set */ +#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) + +/* oputput a single digit. maskvalue is 10**digitidx */ +#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ + if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ + { \ + char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ + *ptr = digit + '0'; /* output it */ \ + nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ + ptr += nz; /* output digit only if non-zero digit seen */ \ + x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ + } + +/* convert integer to fixed point format and multiply out digits, highest first */ +/* requires magic constants: max. digits and number of bits after the decimal point */ +#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ +ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ +{ \ + char nz = lz; /* non-zero digit seen? */ \ + /* convert to x.bits fixed-point */ \ + type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \ + /* output up to 10 digits */ \ + ecb_i2a_digit (type,bits,digitmask, 1, 0); \ + ecb_i2a_digit (type,bits,digitmask, 10, 1); \ + ecb_i2a_digit (type,bits,digitmask, 100, 2); \ + ecb_i2a_digit (type,bits,digitmask, 1000, 3); \ + ecb_i2a_digit (type,bits,digitmask, 10000, 4); \ + ecb_i2a_digit (type,bits,digitmask, 100000, 5); \ + ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \ + ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \ + ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ + ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ + return ptr; \ +} + +/* predefined versions of the above, for various digits */ +/* ecb_i2a_xN = almost N digits, limit defined by macro */ +/* ecb_i2a_N = up to N digits, leading zeroes suppressed */ +/* ecb_i2a_0N = exactly N digits, including leading zeroes */ + +/* non-leading-zero versions, limited range */ +#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */ +#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */ +ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) +ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) + +/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */ +ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) +ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) +ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) +ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) +ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) +ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) +ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) +ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) + +/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */ +ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) +ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) +ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) +ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) +ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) +ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1) +ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1) +ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1) + +#define ECB_I2A_I32_DIGITS 11 +#define ECB_I2A_U32_DIGITS 10 +#define ECB_I2A_I64_DIGITS 20 +#define ECB_I2A_U64_DIGITS 21 +#define ECB_I2A_MAX_DIGITS 21 + +ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u); +ecb_function_ char * ecb_i2a_u32 (char *ptr, uint32_t u) +{ + #if ECB_64BIT_NATIVE + if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) + ptr = ecb_i2a_x10 (ptr, u); + else /* x10 almost, but not fully, covers 32 bit */ + { + uint32_t u1 = u % 1000000000; + uint32_t u2 = u / 1000000000; + + *ptr++ = u2 + '0'; + ptr = ecb_i2a_09 (ptr, u1); + } + #else + if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) + ecb_i2a_x5 (ptr, u); + else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000)) + { + uint32_t u1 = u % 10000; + uint32_t u2 = u / 10000; + + ptr = ecb_i2a_x5 (ptr, u2); + ptr = ecb_i2a_04 (ptr, u1); + } + else + { + uint32_t u1 = u % 10000; + uint32_t ua = u / 10000; + uint32_t u2 = ua % 10000; + uint32_t u3 = ua / 10000; + + ptr = ecb_i2a_2 (ptr, u3); + ptr = ecb_i2a_04 (ptr, u2); + ptr = ecb_i2a_04 (ptr, u1); + } + #endif + + return ptr; +} + +ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v); +ecb_function_ char * ecb_i2a_i32 (char *ptr, int32_t v) +{ + *ptr = '-'; ptr += v < 0; + uint32_t u = v < 0 ? -(uint32_t)v : v; + + #if ECB_64BIT_NATIVE + ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */ + #else + ptr = ecb_i2a_u32 (ptr, u); + #endif + + return ptr; +} + +ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u); +ecb_function_ char * ecb_i2a_u64 (char *ptr, uint64_t u) +{ + #if ECB_64BIT_NATIVE + if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) + ptr = ecb_i2a_x10 (ptr, u); + else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) + { + uint64_t u1 = u % 1000000000; + uint64_t u2 = u / 1000000000; + + ptr = ecb_i2a_x10 (ptr, u2); + ptr = ecb_i2a_09 (ptr, u1); + } + else + { + uint64_t u1 = u % 1000000000; + uint64_t ua = u / 1000000000; + uint64_t u2 = ua % 1000000000; + uint64_t u3 = ua / 1000000000; + + ptr = ecb_i2a_2 (ptr, u3); + ptr = ecb_i2a_09 (ptr, u2); + ptr = ecb_i2a_09 (ptr, u1); + } + #else + if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) + ptr = ecb_i2a_x5 (ptr, u); + else + { + uint64_t u1 = u % 10000; + uint64_t u2 = u / 10000; + + ptr = ecb_i2a_u64 (ptr, u2); + ptr = ecb_i2a_04 (ptr, u1); + } + #endif + + return ptr; +} + +ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v); +ecb_function_ char * ecb_i2a_i64 (char *ptr, int64_t v) +{ + *ptr = '-'; ptr += v < 0; + uint64_t u = v < 0 ? -(uint64_t)v : v; + + #if ECB_64BIT_NATIVE + if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) + ptr = ecb_i2a_x10 (ptr, u); + else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) + { + uint64_t u1 = u % 1000000000; + uint64_t u2 = u / 1000000000; + + ptr = ecb_i2a_x10 (ptr, u2); + ptr = ecb_i2a_09 (ptr, u1); + } + else + { + uint64_t u1 = u % 1000000000; + uint64_t ua = u / 1000000000; + uint64_t u2 = ua % 1000000000; + uint64_t u3 = ua / 1000000000; + + /* 2**31 is 19 digits, so the top is exactly one digit */ + *ptr++ = u3 + '0'; + ptr = ecb_i2a_09 (ptr, u2); + ptr = ecb_i2a_09 (ptr, u1); + } + #else + ptr = ecb_i2a_u64 (ptr, u); + #endif + + return ptr; +} + +/*******************************************************************************/ /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ /* basically, everything uses "ieee pure-endian" floating point numbers */ @@ -824,7 +1551,6 @@ || (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 @@ -856,8 +1582,7 @@ /* 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 - ecb_float_to_binary32 (float x) + ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x) { uint32_t r; @@ -896,8 +1621,7 @@ /* converts an ieee single/binary32 to a float */ ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x); - ecb_function_ ecb_const float - ecb_binary32_to_float (uint32_t x) + ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x) { float r; @@ -926,8 +1650,7 @@ /* convert a double to ieee double/binary64 */ ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x); - ecb_function_ ecb_const uint64_t - ecb_double_to_binary64 (double x) + ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x) { uint64_t r; @@ -966,8 +1689,7 @@ /* converts an ieee double/binary64 to a double */ ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x); - ecb_function_ ecb_const double - ecb_binary64_to_double (uint64_t x) + ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x) { double r; @@ -996,16 +1718,14 @@ /* 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) + 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) + ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x) { return ecb_binary32_to_float (ecb_binary16_to_binary32 (x)); }