[ViewVC] Diff of: cvs/libecb/ecb.h

Comparing libecb/ecb.h (file contents):
Revision 1.157 by root, Fri Feb 20 17:17:26 2015 UTC vs.
Revision 1.181 by root, Mon Jan 20 13:16:01 2020 UTC

 #ifndef ECB_H
 #define ECB_H
 /* 16 bits major, 16 bits minor */
-#define ECB_VERSION 0x00010004
+#define ECB_VERSION 0x00010008
 #ifdef _WIN32
   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;
   #else /* _MSC_VER || __BORLANDC__ */
     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;
     typedef  int64_t  intptr_t;
   #else
     typedef uint32_t uintptr_t;
     typedef  int32_t  intptr_t;
   #endif
 #else
   #include <inttypes.h>
-  #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
     #define ECB_AMD64 1
   #endif
   #define ECB_CLANG_EXTENSION(x) 0
 #endif
 #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
   #define ECB_STDC_VERSION 0
 #else
   #define ECB_STDC_VERSION __STDC_VERSION__
 #endif
 #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"
   #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
   #define ECB_EXTERN_C_END }
 #if ECB_NO_SMP
   #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 <builtins.h>
+#endif
+#if 1400 <= _MSC_VER
+  #include <intrin.h> /* 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__ ("")
+      #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ (""                        : : : "memory")
-    #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__ ("")
+      #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")
     #elif defined __s390__ || defined __s390x__
       #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("bcr 15,0" : : : "memory")
   #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)
     #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 ()
   #elif _MSC_VER >= 1500 /* VC++ 2008 */
     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
   #elif defined _WIN32
     #include <WinNT.h>
     #define ECB_MEMORY_FENCE         MemoryBarrier () /* actually just xchg on x86... scary */
   #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
     #include <mbarrier.h>
-    #define ECB_MEMORY_FENCE         __machine_rw_barrier ()
+    #define ECB_MEMORY_FENCE         __machine_rw_barrier  ()
-    #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier  ()
+    #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
-    #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier  ()
+    #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
+    #define ECB_MEMORY_FENCE_RELAXED __compiler_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 <stdatomic.h>
-    /* 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
 #ifndef ECB_MEMORY_FENCE
   #if !ECB_AVOID_PTHREADS
 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
   #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
   #define ecb_inline static inline
 #elif ECB_GCC_VERSION(2,5)
   #define ecb_deprecated __declspec (deprecated)
 #else
   #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))
 #else
   #define ecb_deprecated_message(msg) ecb_deprecated
 #define ecb_unused     ecb_attribute ((__unused__))
 #define ecb_const      ecb_attribute ((__const__))
 #define ecb_pure       ecb_attribute ((__pure__))
 #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 */
 #else
   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 */
 #if ECB_branchless_on_i386
     if (x & 0xff00ff00) r +=  8;
     if (x & 0xffff0000) r += 16;
 #endif
     return r;
+#endif
   }
   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;
+    _BitScanForward64 (&r, x);
+    return (int)r;
+#else
-    int shift = x & 0xffffffffU ? 0 : 32;
+    int shift = x & 0xffffffff ? 0 : 32;
     return ecb_ctz32 (x >> shift) + shift;
+#endif
   }
   ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
   ecb_function_ ecb_const int
   ecb_popcount32 (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; }
     if (x >>  8) { x >>=  8; r +=  8; }
     if (x >>  4) { x >>=  4; r +=  4; }
     if (x >>  2) { x >>=  2; r +=  2; }
     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
 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_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 <stdlib.h>
+  #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
   ecb_bswap16 (uint16_t x)
   {
 #endif
 /* 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 uint32_t ecb_byteorder_helper (void);
-ecb_inline ecb_const unsigned char
+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 */
   /* 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 */
   /* or when using a recent enough gcc version (>= 4.6) */
-#if ((__i386 || __i386__) && !__VOS__) || _M_X86 || __amd64 || __amd64__ || _M_X64
-  return 0x44;
-#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+#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 0x44;
+  return 0x44332211;
-#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
+      || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
+  #define ECB_BIG_ENDIAN 1
-  return 0x11;
+  return 0x11223344;
 #else
   union
   {
+    uint8_t c[4];
-    uint32_t i;
+    uint32_t u;
-    uint8_t c;
-  } u = { 0x11223344 };
+  } u = { 0x11, 0x22, 0x33, 0x44 };
-  return u.c;
+  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)); }
+#ifdef 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<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
+template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian    () ? ecb_bswap (v) : v; }
+template<typename T> inline T ecb_peek       (const void *ptr) { return *(const T *)ptr; }
+template<typename T> inline T ecb_peek_be    (const void *ptr) { return ecb_be_to_host (ecb_peek  <T> (ptr)); }
+template<typename T> inline T ecb_peek_le    (const void *ptr) { return ecb_le_to_host (ecb_peek  <T> (ptr)); }
+template<typename T> inline T ecb_peek_u     (const void *ptr) { T v; std::memcpy (&v, ptr, sizeof (v)); return v; }
+template<typename T> inline T ecb_peek_be_u  (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
+template<typename T> inline T ecb_peek_le_u  (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
+template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
+template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian    () ? ecb_bswap (v) : v; }
+template<typename T> inline void ecb_poke      (void *ptr, T v) { *(T *)ptr = v; }
+template<typename T> inline void ecb_poke_be   (void *ptr, T v) { return ecb_poke  <T> (ptr, ecb_host_to_be (v)); }
+template<typename T> inline void ecb_poke_le   (void *ptr, T v) { return ecb_poke  <T> (ptr, ecb_host_to_le (v)); }
+template<typename T> inline void ecb_poke_u    (void *ptr, T v) { std::memcpy (ptr, &v, sizeof (v)); }
+template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
+template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
+#endif
+/*****************************************************************************/
 #if ECB_GCC_VERSION(3,0) || ECB_C99
   #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)))
   }
 #else
   #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 */
 /* basically, everything uses "ieee pure-endian" floating point numbers */
 /* 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__ \
     || defined __alpha__ \
     || defined __hppa__ \
     || defined __ia64__ \
     || 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 <string.h> /* for memcpy */
 #else
     #define ECB_NAN ECB_INFINITY
   #endif
   #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
   ecb_float_to_binary32 (float x)
       if (x == 0e0f                    ) return 0x00000000U;
       if (x > +3.40282346638528860e+38f) return 0x7f800000U;
       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;
       if (r)
         m = -m;
     #endif
     return r;
   }
-#endif
+  /* 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));
+  }
-#endif
+  /* 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

Diff Legend

-–
+Removed lines
-+
+Added lines
-<
+Changed lines
->
+Changed lines

Comparing libecb/ecb.h (file contents): Revision 1.157 by root, Fri Feb 20 17:17:26 2015 UTC vs. Revision 1.181 by root, Mon Jan 20 13:16:01 2020 UTC

Diff Legend

Comparing libecb/ecb.h (file contents):
Revision 1.157 by root, Fri Feb 20 17:17:26 2015 UTC vs.
Revision 1.181 by root, Mon Jan 20 13:16:01 2020 UTC