--- cvsroot/CBOR-XS/ecb.h 2014/10/25 06:36:34 1.9 +++ cvsroot/CBOR-XS/ecb.h 2016/02/08 04:11:11 1.13 @@ -1,7 +1,7 @@ /* * libecb - http://software.schmorp.de/pkg/libecb * - * Copyright (©) 2009-2014 Marc Alexander Lehmann + * Copyright (©) 2009-2015 Marc Alexander Lehmann * Copyright (©) 2011 Emanuele Giaquinta * All rights reserved. * @@ -42,7 +42,7 @@ #define ECB_H /* 16 bits major, 16 bits minor */ -#define ECB_VERSION 0x00010004 +#define ECB_VERSION 0x00010005 #ifdef _WIN32 typedef signed char int8_t; @@ -69,15 +69,18 @@ #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) + /* 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 @@ -149,27 +152,44 @@ #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 #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__ + #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__ ("") #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") @@ -304,6 +324,7 @@ #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) #define ECB_STRINGIFY_(a) # a #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) +#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) #define ecb_function_ ecb_inline @@ -350,15 +371,32 @@ #define ecb_deprecated ecb_attribute ((__deprecated__)) #endif -#define ecb_noinline ecb_attribute ((__noinline__)) +#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)) +#else + #define ecb_deprecated_message(msg) ecb_deprecated +#endif + +#if _MSC_VER >= 1400 + #define ecb_noinline __declspec (noinline) +#else + #define ecb_noinline ecb_attribute ((__noinline__)) +#endif + #define ecb_unused ecb_attribute ((__unused__)) #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__)) #endif @@ -395,10 +433,15 @@ #define ecb_popcount32(x) __builtin_popcount (x) /* no popcountll */ #else - ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; - ecb_function_ int + 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; x &= ~x + 1; /* this isolates the lowest bit */ @@ -418,18 +461,25 @@ #endif return r; +#endif } - ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; - ecb_function_ int + 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_ int ecb_popcount32 (uint32_t x) ecb_const; - ecb_function_ int + ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); + ecb_function_ ecb_const int ecb_popcount32 (uint32_t x) { x -= (x >> 1) & 0x55555555; @@ -440,9 +490,14 @@ return x >> 24; } - ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; - ecb_function_ int ecb_ld32 (uint32_t x) + 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; } @@ -452,33 +507,40 @@ if (x >> 1) { r += 1; } return r; +#endif } - ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; - ecb_function_ int ecb_ld64 (uint64_t x) + 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 -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_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); +ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } +ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x); +ecb_function_ ecb_const 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) +ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x); +ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x) { return ( (x * 0x0802U & 0x22110U) - | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; + | (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) +ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x); +ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x) { x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); @@ -488,8 +550,8 @@ return x; } -ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; -ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) +ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x); +ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x) { x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); @@ -502,52 +564,61 @@ /* 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; -ecb_function_ int +ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); +ecb_function_ ecb_const int ecb_popcount64 (uint64_t x) { return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); } -ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; -ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; -ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; -ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; -ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; -ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; -ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; -ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; - -ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } -ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } -ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } -ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } -ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } -ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } -ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } -ecb_inline 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); +ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); +ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); +ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count); +ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); +ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); +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); } #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_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; - ecb_function_ uint16_t + 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_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; - ecb_function_ uint32_t + 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_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; - ecb_function_ uint64_t + 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); @@ -558,15 +629,15 @@ #define ecb_unreachable() __builtin_unreachable () #else /* this seems to work fine, but gcc always emits a warning for it :/ */ - ecb_inline void ecb_unreachable (void) ecb_noreturn; - ecb_inline void ecb_unreachable (void) { } + ecb_inline ecb_noreturn void ecb_unreachable (void); + ecb_inline ecb_noreturn void ecb_unreachable (void) { } #endif /* try to tell the compiler that some condition is definitely true */ #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 -ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; -ecb_inline 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, */ @@ -575,26 +646,28 @@ /* 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_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; } +ecb_inline ecb_const ecb_bool ecb_big_endian (void); +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 () == 0x44332211; } #if ECB_GCC_VERSION(3,0) || ECB_C99 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) @@ -629,6 +702,102 @@ #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) #endif +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 */ @@ -636,7 +805,7 @@ /* 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__ \ @@ -646,7 +815,7 @@ || 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 @@ -673,31 +842,16 @@ #endif #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L - #define ecb_ldexpf(x,e) ldexpf (x, e) + #define ecb_ldexpf(x,e) ldexpf ((x), (e)) + #define ecb_frexpf(x,e) frexpf ((x), (e)) #else - #define ecb_ldexpf(x,e) (float) ldexp ((double)(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_ float ecb_binary16_to_float (uint16_t x) ecb_const; - ecb_function_ 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_ uint32_t ecb_float_to_binary32 (float x) ecb_const; - ecb_function_ uint32_t + 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; @@ -714,7 +868,7 @@ 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; @@ -736,8 +890,8 @@ } /* converts an ieee single/binary32 to a float */ - ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; - ecb_function_ 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) { float r; @@ -766,8 +920,8 @@ } /* convert a double to ieee double/binary64 */ - ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; - ecb_function_ uint64_t + 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; @@ -806,8 +960,8 @@ } /* converts an ieee double/binary64 to a double */ - ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; - ecb_function_ 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) { double r; @@ -835,6 +989,22 @@ 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