nio4r 0.2.0 → 2.7.5

data/ext/libev/ev.c

@@ -1,7 +1,7 @@
 /*
  * libev event processing core, watcher management
  *
- * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libev@schmorp.de>
+ * Copyright (c) 2007-2019 Marc Alexander Lehmann <libev@schmorp.de>
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without modifica-
@@ -37,6 +37,15 @@
  * either the BSD or the GPL.
  */
 
+/* ########## NIO4R PATCHERY HO! ########## */
+#include "ruby.h"
+#include "ruby/thread.h"
+
+#ifdef __APPLE__
+#include <AvailabilityMacros.h>
+#endif
+/* ######################################## */
+
 /* this big block deduces configuration from config.h */
 #ifndef EV_STANDALONE
 # ifdef EV_CONFIG_H
@@ -45,6 +54,12 @@
 #  include "config.h"
 # endif
 
+# if HAVE_FLOOR
+#  ifndef EV_USE_FLOOR
+#   define EV_USE_FLOOR 1
+#  endif
+# endif
+
 # if HAVE_CLOCK_SYSCALL
 #  ifndef EV_USE_CLOCK_SYSCALL
 #   define EV_USE_CLOCK_SYSCALL 1
@@ -55,7 +70,7 @@
 #    define EV_USE_MONOTONIC 1
 #   endif
 #  endif
-# elif !defined(EV_USE_CLOCK_SYSCALL)
+# elif !defined EV_USE_CLOCK_SYSCALL
 #  define EV_USE_CLOCK_SYSCALL 0
 # endif
 
@@ -111,6 +126,24 @@
 #  define EV_USE_EPOLL 0
 # endif
 
+# if HAVE_LINUX_AIO_ABI_H
+#  ifndef EV_USE_LINUXAIO
+#   define EV_USE_LINUXAIO 0 /* was: EV_FEATURE_BACKENDS, always off by default */
+#  endif
+# else
+#  undef EV_USE_LINUXAIO
+#  define EV_USE_LINUXAIO 0
+# endif
+
+# if HAVE_LINUX_FS_H && HAVE_SYS_TIMERFD_H && HAVE_KERNEL_RWF_T
+#  ifndef EV_USE_IOURING
+#   define EV_USE_IOURING EV_FEATURE_BACKENDS
+#  endif
+# else
+#  undef EV_USE_IOURING
+#  define EV_USE_IOURING 0
+# endif
+
 # if HAVE_KQUEUE && HAVE_SYS_EVENT_H
 #  ifndef EV_USE_KQUEUE
 #   define EV_USE_KQUEUE EV_FEATURE_BACKENDS
@@ -156,9 +189,27 @@
 #  define EV_USE_EVENTFD 0
 # endif
 
+# if HAVE_SYS_TIMERFD_H
+#  ifndef EV_USE_TIMERFD
+#   define EV_USE_TIMERFD EV_FEATURE_OS
+#  endif
+# else
+#  undef EV_USE_TIMERFD
+#  define EV_USE_TIMERFD 0
+# endif
+
 #endif
 
-#include <math.h>
+/* OS X, in its infinite idiocy, actually HARDCODES
+ * a limit of 1024 into their select. Where people have brains,
+ * OS X engineers apparently have a vacuum. Or maybe they were
+ * ordered to have a vacuum, or they do anything for money.
+ * This might help. Or not.
+ * Note that this must be defined early, as other include files
+ * will rely on this define as well.
+ */
+#define _DARWIN_UNLIMITED_SELECT 1
+
 #include <stdlib.h>
 #include <string.h>
 #include <fcntl.h>
@@ -180,7 +231,16 @@
 # include "ev.h"
 #endif
 
-EV_CPP(extern "C" {)
+#if EV_NO_THREADS
+# undef EV_NO_SMP
+# define EV_NO_SMP 1
+# undef ECB_NO_THREADS
+# define ECB_NO_THREADS 1
+#endif
+#if EV_NO_SMP
+# undef EV_NO_SMP
+# define ECB_NO_SMP 1
+#endif
 
 #ifndef _WIN32
 # include <sys/time.h>
@@ -189,6 +249,7 @@ EV_CPP(extern "C" {)
 #else
 # include <io.h>
 # define WIN32_LEAN_AND_MEAN
+# include <winsock2.h>
 # include <windows.h>
 # ifndef EV_SELECT_IS_WINSOCKET
 #  define EV_SELECT_IS_WINSOCKET 1
@@ -196,54 +257,56 @@ EV_CPP(extern "C" {)
 # undef EV_AVOID_STDIO
 #endif
 
-/* OS X, in its infinite idiocy, actually HARDCODES
- * a limit of 1024 into their select. Where people have brains,
- * OS X engineers apparently have a vacuum. Or maybe they were
- * ordered to have a vacuum, or they do anything for money.
- * This might help. Or not.
- */
-#define _DARWIN_UNLIMITED_SELECT 1
-
 /* this block tries to deduce configuration from header-defined symbols and defaults */
 
 /* try to deduce the maximum number of signals on this platform */
-#if defined (EV_NSIG)
+#if defined EV_NSIG
 /* use what's provided */
-#elif defined (NSIG)
+#elif defined NSIG
 # define EV_NSIG (NSIG)
-#elif defined(_NSIG)
+#elif defined _NSIG
 # define EV_NSIG (_NSIG)
-#elif defined (SIGMAX)
+#elif defined SIGMAX
 # define EV_NSIG (SIGMAX+1)
-#elif defined (SIG_MAX)
+#elif defined SIG_MAX
 # define EV_NSIG (SIG_MAX+1)
-#elif defined (_SIG_MAX)
+#elif defined _SIG_MAX
 # define EV_NSIG (_SIG_MAX+1)
-#elif defined (MAXSIG)
+#elif defined MAXSIG
 # define EV_NSIG (MAXSIG+1)
-#elif defined (MAX_SIG)
+#elif defined MAX_SIG
 # define EV_NSIG (MAX_SIG+1)
-#elif defined (SIGARRAYSIZE)
+#elif defined SIGARRAYSIZE
 # define EV_NSIG (SIGARRAYSIZE) /* Assume ary[SIGARRAYSIZE] */
-#elif defined (_sys_nsig)
+#elif defined _sys_nsig
 # define EV_NSIG (_sys_nsig) /* Solaris 2.5 */
 #else
-# error "unable to find value for NSIG, please report"
-/* to make it compile regardless, just remove the above line, */
-/* but consider reporting it, too! :) */
-# define EV_NSIG 65
+# define EV_NSIG (8 * sizeof (sigset_t) + 1)
+#endif
+
+#ifndef EV_USE_FLOOR
+# define EV_USE_FLOOR 0
 #endif
 
 #ifndef EV_USE_CLOCK_SYSCALL
-# if __linux && __GLIBC__ >= 2
+# if __linux && __GLIBC__ == 2 && __GLIBC_MINOR__ < 17
 #  define EV_USE_CLOCK_SYSCALL EV_FEATURE_OS
 # else
 #  define EV_USE_CLOCK_SYSCALL 0
 # endif
 #endif
 
+#if !(_POSIX_TIMERS > 0)
+# ifndef EV_USE_MONOTONIC
+#  define EV_USE_MONOTONIC 0
+# endif
+# ifndef EV_USE_REALTIME
+#  define EV_USE_REALTIME 0
+# endif
+#endif
+
 #ifndef EV_USE_MONOTONIC
-# if defined (_POSIX_MONOTONIC_CLOCK) && _POSIX_MONOTONIC_CLOCK >= 0
+# if defined _POSIX_MONOTONIC_CLOCK && _POSIX_MONOTONIC_CLOCK >= 0
 #  define EV_USE_MONOTONIC EV_FEATURE_OS
 # else
 #  define EV_USE_MONOTONIC 0
@@ -290,6 +353,14 @@ EV_CPP(extern "C" {)
 # define EV_USE_PORT 0
 #endif
 
+#ifndef EV_USE_LINUXAIO
+#  define EV_USE_LINUXAIO 0
+#endif
+
+#ifndef EV_USE_IOURING
+#  define EV_USE_IOURING 0
+#endif
+
 #ifndef EV_USE_INOTIFY
 # if __linux && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 4))
 #  define EV_USE_INOTIFY EV_FEATURE_OS
@@ -322,6 +393,14 @@ EV_CPP(extern "C" {)
 # endif
 #endif
 
+#ifndef EV_USE_TIMERFD
+# if __linux && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 8))
+#  define EV_USE_TIMERFD EV_FEATURE_OS
+# else
+#  define EV_USE_TIMERFD 0
+# endif
+#endif
+
 #if 0 /* debugging */
 # define EV_VERIFY 3
 # define EV_USE_4HEAP 1
@@ -340,14 +419,31 @@ EV_CPP(extern "C" {)
 # define EV_HEAP_CACHE_AT EV_FEATURE_DATA
 #endif
 
+#ifdef __ANDROID__
+/* supposedly, android doesn't typedef fd_mask */
+# undef EV_USE_SELECT
+# define EV_USE_SELECT 0
+/* supposedly, we need to include syscall.h, not sys/syscall.h, so just disable */
+# undef EV_USE_CLOCK_SYSCALL
+# define EV_USE_CLOCK_SYSCALL 0
+#endif
+
+/* aix's poll.h seems to cause lots of trouble */
+#ifdef _AIX
+/* AIX has a completely broken poll.h header */
+# undef EV_USE_POLL
+# define EV_USE_POLL 0
+#endif
+
 /* on linux, we can use a (slow) syscall to avoid a dependency on pthread, */
 /* which makes programs even slower. might work on other unices, too. */
 #if EV_USE_CLOCK_SYSCALL
-# include <syscall.h>
+# include <sys/syscall.h>
 # ifdef SYS_clock_gettime
 #  define clock_gettime(id, ts) syscall (SYS_clock_gettime, (id), (ts))
 #  undef EV_USE_MONOTONIC
 #  define EV_USE_MONOTONIC 1
+#  define EV_NEED_SYSCALL 1
 # else
 #  undef EV_USE_CLOCK_SYSCALL
 #  define EV_USE_CLOCK_SYSCALL 0
@@ -356,12 +452,6 @@ EV_CPP(extern "C" {)
 
 /* this block fixes any misconfiguration where we know we run into trouble otherwise */
 
-#ifdef _AIX
-/* AIX has a completely broken poll.h header */
-# undef EV_USE_POLL
-# define EV_USE_POLL 0
-#endif
-
 #ifndef CLOCK_MONOTONIC
 # undef EV_USE_MONOTONIC
 # define EV_USE_MONOTONIC 0
@@ -379,11 +469,36 @@ EV_CPP(extern "C" {)
 
 #if !EV_USE_NANOSLEEP
 /* hp-ux has it in sys/time.h, which we unconditionally include above */
-# if !defined(_WIN32) && !defined(__hpux)
+# if !defined _WIN32 && !defined __hpux
 #  include <sys/select.h>
 # endif
 #endif
 
+#if EV_USE_LINUXAIO
+# include <sys/syscall.h>
+# if SYS_io_getevents && EV_USE_EPOLL /* linuxaio backend requires epoll backend */
+#  define EV_NEED_SYSCALL 1
+# else
+#  undef EV_USE_LINUXAIO
+#  define EV_USE_LINUXAIO 0
+# endif
+#endif
+
+#if EV_USE_IOURING
+# include <sys/syscall.h>
+# if !SYS_io_uring_setup && __linux && !__alpha
+#  define SYS_io_uring_setup     425
+#  define SYS_io_uring_enter     426
+#  define SYS_io_uring_wregister 427
+# endif
+# if SYS_io_uring_setup && EV_USE_EPOLL /* iouring backend requires epoll backend */
+#  define EV_NEED_SYSCALL 1
+# else
+#  undef EV_USE_IOURING
+#  define EV_USE_IOURING 0
+# endif
+#endif
+
 #if EV_USE_INOTIFY
 # include <sys/statfs.h>
 # include <sys/inotify.h>
@@ -394,12 +509,8 @@ EV_CPP(extern "C" {)
 # endif
 #endif
 
-#if EV_SELECT_IS_WINSOCKET
-# include <winsock.h>
-#endif
-
 #if EV_USE_EVENTFD
-/* our minimum requirement is glibc 2.7 which has the stub, but not the header */
+/* our minimum requirement is glibc 2.7 which has the stub, but not the full header */
 # include <stdint.h>
 # ifndef EFD_NONBLOCK
 #  define EFD_NONBLOCK O_NONBLOCK
@@ -415,7 +526,7 @@ EV_CPP(extern "C") int (eventfd) (unsigned int initval, int flags);
 #endif
 
 #if EV_USE_SIGNALFD
-/* our minimum requirement is glibc 2.7 which has the stub, but not the header */
+/* our minimum requirement is glibc 2.7 which has the stub, but not the full header */
 # include <stdint.h>
 # ifndef SFD_NONBLOCK
 #  define SFD_NONBLOCK O_NONBLOCK
@@ -427,7 +538,7 @@ EV_CPP(extern "C") int (eventfd) (unsigned int initval, int flags);
 #   define SFD_CLOEXEC 02000000
 #  endif
 # endif
-EV_CPP (extern "C") int signalfd (int fd, const sigset_t *mask, int flags);
+EV_CPP (extern "C") int (signalfd) (int fd, const sigset_t *mask, int flags);
 
 struct signalfd_siginfo
 {
@@ -436,7 +547,17 @@ struct signalfd_siginfo
 };
 #endif
 
-/**/
+/* for timerfd, libev core requires TFD_TIMER_CANCEL_ON_SET &c */
+#if EV_USE_TIMERFD
+# include <sys/timerfd.h>
+/* timerfd is only used for periodics */
+# if !(defined (TFD_TIMER_CANCEL_ON_SET) && defined (TFD_CLOEXEC) && defined (TFD_NONBLOCK)) || !EV_PERIODIC_ENABLE
+#  undef EV_USE_TIMERFD
+#  define EV_USE_TIMERFD 0
+# endif
+#endif
+
+/*****************************************************************************/
 
 #if EV_VERIFY >= 3
 # define EV_FREQUENT_CHECK ev_verify (EV_A)
@@ -445,42 +566,1274 @@ struct signalfd_siginfo
 #endif
 
 /*
- * This is used to avoid floating point rounding problems.
- * It is added to ev_rt_now when scheduling periodics
- * to ensure progress, time-wise, even when rounding
- * errors are against us.
+ * This is used to work around floating point rounding problems.
  * This value is good at least till the year 4000.
- * Better solutions welcome.
  */
-#define TIME_EPSILON  0.0001220703125 /* 1/8192 */
+#define MIN_INTERVAL  0.0001220703125 /* 1/2**13, good till 4000 */
+/*#define MIN_INTERVAL  0.00000095367431640625 /* 1/2**20, good till 2200 */
+
+#define MIN_TIMEJUMP   1. /* minimum timejump that gets detected (if monotonic clock available) */
+#define MAX_BLOCKTIME  59.743 /* never wait longer than this time (to detect time jumps) */
+#define MAX_BLOCKTIME2 1500001.07 /* same, but when timerfd is used to detect jumps, also safe delay to not overflow */
+
+/* find a portable timestamp that is "always" in the future but fits into time_t.
+ * this is quite hard, and we are mostly guessing - we handle 32 bit signed/unsigned time_t,
+ * and sizes larger than 32 bit, and maybe the unlikely floating point time_t */
+#define EV_TSTAMP_HUGE \
+  (sizeof (time_t) >= 8     ? 10000000000000.  \
+   : 0 < (time_t)4294967295 ?     4294967295.  \
+   :                              2147483647.) \
+
+#ifndef EV_TS_CONST
+# define EV_TS_CONST(nv) nv
+# define EV_TS_TO_MSEC(a) a * 1e3 + 0.9999
+# define EV_TS_FROM_USEC(us) us * 1e-6
+# define EV_TV_SET(tv,t) do { tv.tv_sec = (long)t; tv.tv_usec = (long)((t - tv.tv_sec) * 1e6); } while (0)
+# define EV_TS_SET(ts,t) do { ts.tv_sec = (long)t; ts.tv_nsec = (long)((t - ts.tv_sec) * 1e9); } while (0)
+# define EV_TV_GET(tv) ((tv).tv_sec + (tv).tv_usec * 1e-6)
+# define EV_TS_GET(ts) ((ts).tv_sec + (ts).tv_nsec * 1e-9)
+#endif
+
+/* the following is ecb.h embedded into libev - use update_ev_c to update from an external copy */
+/* ECB.H BEGIN */
+/*
+ * libecb - http://software.schmorp.de/pkg/libecb
+ *
+ * Copyright (©) 2009-2015,2018-2020 Marc Alexander Lehmann <libecb@schmorp.de>
+ * Copyright (©) 2011 Emanuele Giaquinta
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modifica-
+ * tion, are permitted provided that the following conditions are met:
+ *
+ *   1.  Redistributions of source code must retain the above copyright notice,
+ *       this list of conditions and the following disclaimer.
+ *
+ *   2.  Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
+ * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
+ * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
+ * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Alternatively, the contents of this file may be used under the terms of
+ * the GNU General Public License ("GPL") version 2 or any later version,
+ * in which case the provisions of the GPL are applicable instead of
+ * the above. If you wish to allow the use of your version of this file
+ * only under the terms of the GPL and not to allow others to use your
+ * version of this file under the BSD license, indicate your decision
+ * by deleting the provisions above and replace them with the notice
+ * and other provisions required by the GPL. If you do not delete the
+ * provisions above, a recipient may use your version of this file under
+ * either the BSD or the GPL.
+ */
 
-#define MIN_TIMEJUMP  1. /* minimum timejump that gets detected (if monotonic clock available) */
-#define MAX_BLOCKTIME 59.743 /* never wait longer than this time (to detect time jumps) */
+#ifndef ECB_H
+#define ECB_H
+
+/* 16 bits major, 16 bits minor */
+#define ECB_VERSION 0x00010008
+
+#include <string.h> /* 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;
+  #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
+    #define ECB_PTRSIZE 4
+    typedef uint32_t uintptr_t;
+    typedef  int32_t  intptr_t;
+  #endif
+#else
+  #include <inttypes.h>
+  #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 ECB_GCC_AMD64 || ECB_MSVC_AMD64
+  #if _ILP32
+    #define ECB_AMD64_X32 1
+  #else
+    #define ECB_AMD64 1
+  #endif
+#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.
+ * or so.
+ * we try to detect these and simply assume they are not gcc - if they have
+ * an issue with that they should have done it right in the first place.
+ */
+#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
+  #define ECB_GCC_VERSION(major,minor) 0
+#else
+  #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
+#endif
 
-#define EV_TV_SET(tv,t) do { tv.tv_sec = (long)t; tv.tv_usec = (long)((t - tv.tv_sec) * 1e6); } while (0)
-#define EV_TS_SET(ts,t) do { ts.tv_sec = (long)t; ts.tv_nsec = (long)((t - ts.tv_sec) * 1e9); } while (0)
+#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
 
-#if __GNUC__ >= 4
-# define expect(expr,value)         __builtin_expect ((expr),(value))
-# define noinline                   __attribute__ ((noinline))
+#if __clang__ && defined __has_builtin
+  #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
 #else
-# define expect(expr,value)         (expr)
-# define noinline
-# if __STDC_VERSION__ < 199901L && __GNUC__ < 2
-#  define inline
-# endif
+  #define ECB_CLANG_BUILTIN(x) 0
+#endif
+
+#if __clang__ && defined __has_extension
+  #define ECB_CLANG_EXTENSION(x) __has_extension (x)
+#else
+  #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_C            1
+  #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 }
+#else
+  #define ECB_EXTERN_C extern
+  #define ECB_EXTERN_C_BEG
+  #define ECB_EXTERN_C_END
+#endif
+
+/*****************************************************************************/
+
+/* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
+/* ECB_NO_SMP     - ecb might be used in multiple threads, but only on a single cpu */
+
+#if ECB_NO_THREADS
+  #define ECB_NO_SMP 1
+#endif
+
+#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__ (""                        : : : "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__ (""         : : : "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_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_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__) && !(__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")
+    #elif defined __mips__
+      /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
+      /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
+    #elif defined __alpha__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("mb"       : : : "memory")
+    #elif defined __hppa__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ (""         : : : "memory")
+      #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
+    #elif defined __ia64__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("mf"       : : : "memory")
+    #elif defined __m68k__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ (""         : : : "memory")
+    #elif defined __m88k__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
+    #elif defined __sh__
+      #define ECB_MEMORY_FENCE         __asm__ __volatile__ (""         : : : "memory")
+    #endif
+  #endif
+#endif
+
+#ifndef ECB_MEMORY_FENCE
+  #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... */
+    #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
+    #define ECB_MEMORY_FENCE         _ReadWriteBarrier (); MemoryBarrier()
+    #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
+    #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
+  #elif _MSC_VER >= 1400 /* VC++ 2005 */
+    #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
+    #define ECB_MEMORY_FENCE         _ReadWriteBarrier ()
+    #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
+    #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
+  #elif defined _WIN32
+    #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_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
+#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>
+    #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 you get undefined symbol references to pthread_mutex_lock,
+     * or failure to find pthread.h, then you should implement
+     * the ECB_MEMORY_FENCE operations for your cpu/compiler
+     * OR provide pthread.h and link against the posix thread library
+     * of your system.
+     */
+    #include <pthread.h>
+    #define ECB_NEEDS_PTHREADS 1
+    #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
+
+    static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
+    #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
+  #endif
+#endif
+
+#if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
+  #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
+#endif
+
+#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
+  #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_inline static __inline__
+#elif ECB_C99
+  #define ecb_inline static inline
+#else
+  #define ecb_inline static
+#endif
+
+#if ECB_GCC_VERSION(3,3)
+  #define ecb_restrict __restrict__
+#elif ECB_C99
+  #define ecb_restrict restrict
+#else
+  #define ecb_restrict
+#endif
+
+typedef int ecb_bool;
+
+#define ECB_CONCAT_(a, b) a ## b
+#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
+
+#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
+  #define ecb_attribute(attrlist)        __attribute__ (attrlist)
+#else
+  #define ecb_attribute(attrlist)
+#endif
+
+#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
+  #define ecb_is_constant(expr)          __builtin_constant_p (expr)
+#else
+  /* possible C11 impl for integral types
+  typedef struct ecb_is_constant_struct ecb_is_constant_struct;
+  #define ecb_is_constant(expr)          _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
+
+  #define ecb_is_constant(expr)          0
+#endif
+
+#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
+  #define ecb_expect(expr,value)         __builtin_expect ((expr),(value))
+#else
+  #define ecb_expect(expr,value)         (expr)
+#endif
+
+#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
+  #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
+#else
+  #define ecb_prefetch(addr,rw,locality)
+#endif
+
+/* no emulation for ecb_decltype */
+#if ECB_CPP11
+  // older implementations might have problems with decltype(x)::type, work around it
+  template<class T> struct ecb_decltype_t { typedef T type; };
+  #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
+#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
+  #define ecb_decltype(x) __typeof__ (x)
+#endif
+
+#if _MSC_VER >= 1300
+  #define ecb_deprecated __declspec (deprecated)
+#else
+  #define ecb_deprecated ecb_attribute ((__deprecated__))
+#endif
+
+#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__))
+
+#if ECB_C11 || __IBMC_NORETURN
+  /* 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
+
+#if ECB_GCC_VERSION(4,3)
+  #define ecb_artificial ecb_attribute ((__artificial__))
+  #define ecb_hot        ecb_attribute ((__hot__))
+  #define ecb_cold       ecb_attribute ((__cold__))
+#else
+  #define ecb_artificial
+  #define ecb_hot
+  #define ecb_cold
+#endif
+
+/* put around conditional expressions if you are very sure that the  */
+/* expression is mostly true or mostly false. note that these return */
+/* booleans, not the expression.                                     */
+#define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
+#define ecb_expect_true(expr)  ecb_expect (!!(expr), 1)
+/* for compatibility to the rest of the world */
+#define ecb_likely(expr)   ecb_expect_true  (expr)
+#define ecb_unlikely(expr) ecb_expect_false (expr)
+
+/* count trailing zero bits and count # of one bits */
+#if ECB_GCC_VERSION(3,4) \
+    || (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_popcount32(x) __builtin_popcount (x)
+  /* no popcountll */
+#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
+    r += !!(x & 0xaaaaaaaa) << 0;
+    r += !!(x & 0xcccccccc) << 1;
+    r += !!(x & 0xf0f0f0f0) << 2;
+    r += !!(x & 0xff00ff00) << 3;
+    r += !!(x & 0xffff0000) << 4;
+#else
+    if (x & 0xaaaaaaaa) r +=  1;
+    if (x & 0xcccccccc) r +=  2;
+    if (x & 0xf0f0f0f0) r +=  4;
+    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 & 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)
+  {
+    x -=  (x >> 1) & 0x55555555;
+    x  = ((x >> 2) & 0x33333333) + (x & 0x33333333);
+    x  = ((x >> 4) + x) & 0x0f0f0f0f;
+    x *= 0x01010101;
+
+    return x >> 24;
+  }
+
+  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_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_ 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;
+}
+
+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);
+  x = ((x >>  4) &     0x0f0f) | ((x &     0x0f0f) <<  4);
+  x = ( x >>  8              ) | ( x               <<  8);
+
+  return 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);
+  x = ((x >>  4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) <<  4);
+  x = ((x >>  8) & 0x00ff00ff) | ((x & 0x00ff00ff) <<  8);
+  x = ( x >> 16              ) | ( x               << 16);
+
+  return x;
+}
+
+/* popcount64 is only available on 64 bit cpus as gcc builtin */
+/* so for this version we are lazy */
+ecb_function_ 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 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_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
 
-#define expect_false(expr) expect ((expr) != 0, 0)
-#define expect_true(expr)  expect ((expr) != 0, 1)
-#define inline_size        static inline
+#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)
+  {
+    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)
+  {
+    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)
+  {
+    return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
+  }
+#endif
+
+#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
+  #define ecb_unreachable() __builtin_unreachable ()
+#else
+  /* this seems to work fine, but gcc always emits a warning for it :/ */
+  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 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 */
+  /* 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 (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
+  {
+    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 () == 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; }
+
+/*****************************************************************************/
+/* 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<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; 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) { 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)))
+#endif
+
+#if ECB_CPP
+  template<typename T>
+  static inline T ecb_div_rd (T val, T div)
+  {
+    return val < 0 ? - ((-val + div - 1) / div) : (val          ) / div;
+  }
+  template<typename T>
+  static inline T ecb_div_ru (T val, T div)
+  {
+    return val < 0 ? - ((-val          ) / div) : (val + div - 1) / div;
+  }
+#else
+  #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val)            ) / (div))
+  #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val)            ) / (div)) : ((val) + (div) - 1) / (div))
+#endif
+
+#if ecb_cplusplus_does_not_suck
+  /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
+  template<typename T, int N>
+  static inline int ecb_array_length (const T (&arr)[N])
+  {
+    return 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__ \
+    || 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 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
+#else
+  #define ECB_STDFP 0
+#endif
+
+#ifndef ECB_NO_LIBM
+
+  #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
+
+  /* only the oldest of old doesn't have this one. solaris. */
+  #ifdef INFINITY
+    #define ECB_INFINITY INFINITY
+  #else
+    #define ECB_INFINITY HUGE_VAL
+  #endif
+
+  #ifdef NAN
+    #define ECB_NAN NAN
+  #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 ((double) (x), (e))
+    #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
+  #endif
+
+  /* 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)
+  {
+    uint32_t r;
+
+    #if ECB_STDFP
+      memcpy (&r, &x, 4);
+    #else
+      /* slow emulation, works for anything but -0 */
+      uint32_t m;
+      int e;
+
+      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 = ecb_frexpf (x, &e) * 0x1000000U;
+
+      r = m & 0x80000000U;
+
+      if (r)
+        m = -m;
+
+      if (e <= -126)
+        {
+          m &= 0xffffffU;
+          m >>= (-125 - e);
+          e = -126;
+        }
+
+      r |= (e + 126) << 23;
+      r |= m & 0x7fffffU;
+    #endif
+
+    return r;
+  }
+
+  /* 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)
+  {
+    float r;
+
+    #if ECB_STDFP
+      memcpy (&r, &x, 4);
+    #else
+      /* emulation, only works for normals and subnormals and +0 */
+      int neg = x >> 31;
+      int e = (x >> 23) & 0xffU;
+
+      x &= 0x7fffffU;
+
+      if (e)
+        x |= 0x800000U;
+      else
+        e = 1;
+
+      /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
+      r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
+
+      r = neg ? -r : r;
+    #endif
+
+    return r;
+  }
+
+  /* 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)
+  {
+    uint64_t r;
+
+    #if ECB_STDFP
+      memcpy (&r, &x, 8);
+    #else
+      /* slow emulation, works for anything but -0 */
+      uint64_t m;
+      int e;
+
+      if (x == 0e0                     ) return 0x0000000000000000U;
+      if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
+      if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
+      if (x != x                       ) return 0X7ff7ffffffffffffU;
+
+      m = frexp (x, &e) * 0x20000000000000U;
+
+      r = m & 0x8000000000000000;;
+
+      if (r)
+        m = -m;
+
+      if (e <= -1022)
+        {
+          m &= 0x1fffffffffffffU;
+          m >>= (-1021 - e);
+          e = -1022;
+        }
+
+      r |= ((uint64_t)(e + 1022)) << 52;
+      r |= m & 0xfffffffffffffU;
+    #endif
+
+    return r;
+  }
+
+  /* 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)
+  {
+    double r;
+
+    #if ECB_STDFP
+      memcpy (&r, &x, 8);
+    #else
+      /* emulation, only works for normals and subnormals and +0 */
+      int neg = x >> 63;
+      int e = (x >> 52) & 0x7ffU;
+
+      x &= 0xfffffffffffffU;
+
+      if (e)
+        x |= 0x10000000000000U;
+      else
+        e = 1;
+
+      /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
+      r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
+
+      r = neg ? -r : r;
+    #endif
+
+    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
+
+/* ECB.H END */
+
+#if ECB_MEMORY_FENCE_NEEDS_PTHREADS
+/* if your architecture doesn't need memory fences, e.g. because it is
+ * single-cpu/core, or if you use libev in a project that doesn't use libev
+ * from multiple threads, then you can define ECB_NO_THREADS when compiling
+ * libev, in which cases the memory fences become nops.
+ * alternatively, you can remove this #error and link against libpthread,
+ * which will then provide the memory fences.
+ */
+# error "memory fences not defined for your architecture, please report"
+#endif
+
+#ifndef ECB_MEMORY_FENCE
+# define ECB_MEMORY_FENCE do { } while (0)
+# define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
+# define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
+#endif
+
+#define inline_size        ecb_inline
 
 #if EV_FEATURE_CODE
-# define inline_speed      static inline
+# define inline_speed      ecb_inline
 #else
-# define inline_speed      static noinline
+# define inline_speed      ecb_noinline static
 #endif
 
+/*****************************************************************************/
+/* raw syscall wrappers */
+
+#if EV_NEED_SYSCALL
+
+#include <sys/syscall.h>
+
+/*
+ * define some syscall wrappers for common architectures
+ * this is mostly for nice looks during debugging, not performance.
+ * our syscalls return < 0, not == -1, on error. which is good
+ * enough for linux aio.
+ * TODO: arm is also common nowadays, maybe even mips and x86
+ * TODO: after implementing this, it suddenly looks like overkill, but its hard to remove...
+ */
+#if __GNUC__ && __linux && ECB_AMD64 && !EV_FEATURE_CODE
+  /* the costly errno access probably kills this for size optimisation */
+
+  #define ev_syscall(nr,narg,arg1,arg2,arg3,arg4,arg5,arg6)            \
+    ({                                                                 \
+        long res;                                                      \
+        register unsigned long r6 __asm__ ("r9" );                     \
+        register unsigned long r5 __asm__ ("r8" );                     \
+        register unsigned long r4 __asm__ ("r10");                     \
+        register unsigned long r3 __asm__ ("rdx");                     \
+        register unsigned long r2 __asm__ ("rsi");                     \
+        register unsigned long r1 __asm__ ("rdi");                     \
+        if (narg >= 6) r6 = (unsigned long)(arg6);                     \
+        if (narg >= 5) r5 = (unsigned long)(arg5);                     \
+        if (narg >= 4) r4 = (unsigned long)(arg4);                     \
+        if (narg >= 3) r3 = (unsigned long)(arg3);                     \
+        if (narg >= 2) r2 = (unsigned long)(arg2);                     \
+        if (narg >= 1) r1 = (unsigned long)(arg1);                     \
+        __asm__ __volatile__ (                                         \
+          "syscall\n\t"                                                \
+          : "=a" (res)                                                 \
+          : "0" (nr), "r" (r1), "r" (r2), "r" (r3), "r" (r4), "r" (r5) \
+          : "cc", "r11", "cx", "memory");                              \
+        errno = -res;                                                  \
+        res;                                                           \
+    })
+
+#endif
+
+#ifdef ev_syscall
+  #define ev_syscall0(nr)                               ev_syscall (nr, 0,    0,    0,    0,    0,    0,   0)
+  #define ev_syscall1(nr,arg1)                          ev_syscall (nr, 1, arg1,    0,    0,    0,    0,   0)
+  #define ev_syscall2(nr,arg1,arg2)                     ev_syscall (nr, 2, arg1, arg2,    0,    0,    0,   0)
+  #define ev_syscall3(nr,arg1,arg2,arg3)                ev_syscall (nr, 3, arg1, arg2, arg3,    0,    0,   0)
+  #define ev_syscall4(nr,arg1,arg2,arg3,arg4)           ev_syscall (nr, 3, arg1, arg2, arg3, arg4,    0,   0)
+  #define ev_syscall5(nr,arg1,arg2,arg3,arg4,arg5)      ev_syscall (nr, 5, arg1, arg2, arg3, arg4, arg5,   0)
+  #define ev_syscall6(nr,arg1,arg2,arg3,arg4,arg5,arg6) ev_syscall (nr, 6, arg1, arg2, arg3, arg4, arg5,arg6)
+#else
+  #define ev_syscall0(nr)                               syscall (nr)
+  #define ev_syscall1(nr,arg1)                          syscall (nr, arg1)
+  #define ev_syscall2(nr,arg1,arg2)                     syscall (nr, arg1, arg2)
+  #define ev_syscall3(nr,arg1,arg2,arg3)                syscall (nr, arg1, arg2, arg3)
+  #define ev_syscall4(nr,arg1,arg2,arg3,arg4)           syscall (nr, arg1, arg2, arg3, arg4)
+  #define ev_syscall5(nr,arg1,arg2,arg3,arg4,arg5)      syscall (nr, arg1, arg2, arg3, arg4, arg5)
+  #define ev_syscall6(nr,arg1,arg2,arg3,arg4,arg5,arg6) syscall (nr, arg1, arg2, arg3, arg4, arg5,arg6)
+#endif
+
+#endif
+
+/*****************************************************************************/
+
 #define NUMPRI (EV_MAXPRI - EV_MINPRI + 1)
 
 #if EV_MINPRI == EV_MAXPRI
@@ -489,8 +1842,7 @@ struct signalfd_siginfo
 # define ABSPRI(w) (((W)w)->priority - EV_MINPRI)
 #endif
 
-#define EMPTY       /* required for microsofts broken pseudo-c compiler */
-#define EMPTY2(a,b) /* used to suppress some warnings */
+#define EMPTY /* required for microsofts broken pseudo-c compiler */
 
 typedef ev_watcher *W;
 typedef ev_watcher_list *WL;
@@ -525,11 +1877,65 @@ static EV_ATOMIC_T have_monotonic; /* did clock_gettime (CLOCK_MONOTONIC) work?
 
 /*****************************************************************************/
 
+#if EV_USE_LINUXAIO
+# include <linux/aio_abi.h> /* probably only needed for aio_context_t */
+#endif
+
+/* define a suitable floor function (only used by periodics atm) */
+
+#if EV_USE_FLOOR
+# include <math.h>
+# define ev_floor(v) floor (v)
+#else
+
+#include <float.h>
+
+/* a floor() replacement function, should be independent of ev_tstamp type */
+ecb_noinline
+static ev_tstamp
+ev_floor (ev_tstamp v)
+{
+  /* the choice of shift factor is not terribly important */
+#if FLT_RADIX != 2 /* assume FLT_RADIX == 10 */
+  const ev_tstamp shift = sizeof (unsigned long) >= 8 ? 10000000000000000000. : 1000000000.;
+#else
+  const ev_tstamp shift = sizeof (unsigned long) >= 8 ? 18446744073709551616. : 4294967296.;
+#endif
+
+  /* special treatment for negative arguments */
+  if (ecb_expect_false (v < 0.))
+    {
+      ev_tstamp f = -ev_floor (-v);
+
+      return f - (f == v ? 0 : 1);
+    }
+
+  /* argument too large for an unsigned long? then reduce it */
+  if (ecb_expect_false (v >= shift))
+    {
+      ev_tstamp f;
+
+      if (v == v - 1.)
+        return v; /* very large numbers are assumed to be integer */
+
+      f = shift * ev_floor (v * (1. / shift));
+      return f + ev_floor (v - f);
+    }
+
+  /* fits into an unsigned long */
+  return (unsigned long)v;
+}
+
+#endif
+
+/*****************************************************************************/
+
 #ifdef __linux
 # include <sys/utsname.h>
 #endif
 
-static unsigned int noinline
+ecb_noinline ecb_cold
+static unsigned int
 ev_linux_version (void)
 {
 #ifdef __linux
@@ -568,22 +1974,25 @@ ev_linux_version (void)
 /*****************************************************************************/
 
 #if EV_AVOID_STDIO
-static void noinline
+ecb_noinline ecb_cold
+static void
 ev_printerr (const char *msg)
 {
   write (STDERR_FILENO, msg, strlen (msg));
 }
 #endif
 
-static void (*syserr_cb)(const char *msg);
+static void (*syserr_cb)(const char *msg) EV_NOEXCEPT;
 
+ecb_cold
 void
-ev_set_syserr_cb (void (*cb)(const char *msg))
+ev_set_syserr_cb (void (*cb)(const char *msg) EV_NOEXCEPT) EV_NOEXCEPT
 {
   syserr_cb = cb;
 }
 
-static void noinline
+ecb_noinline ecb_cold
+static void
 ev_syserr (const char *msg)
 {
   if (!msg)
@@ -606,14 +2015,13 @@ ev_syserr (const char *msg)
 }
 
 static void *
-ev_realloc_emul (void *ptr, long size)
+ev_realloc_emul (void *ptr, size_t size) EV_NOEXCEPT
 {
-#if __GLIBC__
-  return realloc (ptr, size);
-#else
   /* some systems, notably openbsd and darwin, fail to properly
    * implement realloc (x, 0) (as required by both ansi c-89 and
    * the single unix specification, so work around them here.
+   * recently, also (at least) fedora and debian started breaking it,
+   * despite documenting it otherwise.
    */
 
   if (size)
@@ -621,19 +2029,19 @@ ev_realloc_emul (void *ptr, long size)
 
   free (ptr);
   return 0;
-#endif
 }
 
-static void *(*alloc)(void *ptr, long size) = ev_realloc_emul;
+static void *(*alloc)(void *ptr, size_t size) EV_NOEXCEPT = ev_realloc_emul;
 
+ecb_cold
 void
-ev_set_allocator (void *(*cb)(void *ptr, long size))
+ev_set_allocator (void *(*cb)(void *ptr, size_t size) EV_NOEXCEPT) EV_NOEXCEPT
 {
   alloc = cb;
 }
 
 inline_speed void *
-ev_realloc (void *ptr, long size)
+ev_realloc (void *ptr, size_t size)
 {
   ptr = alloc (ptr, size);
 
@@ -664,8 +2072,8 @@ typedef struct
   WL head;
   unsigned char events; /* the events watched for */
   unsigned char reify;  /* flag set when this ANFD needs reification (EV_ANFD_REIFY, EV__IOFDSET) */
-  unsigned char emask;  /* the epoll backend stores the actual kernel mask in here */
-  unsigned char unused;
+  unsigned char emask;  /* some backends store the actual kernel mask in here */
+  unsigned char eflags; /* flags field for use by backends */
 #if EV_USE_EPOLL
   unsigned int egen;    /* generation counter to counter epoll bugs */
 #endif
@@ -725,11 +2133,11 @@ typedef struct
   #include "ev_wrap.h"
 
   static struct ev_loop default_loop_struct;
-  struct ev_loop *ev_default_loop_ptr;
+  EV_API_DECL struct ev_loop *ev_default_loop_ptr = 0; /* needs to be initialised to make it a definition despite extern */
 
 #else
 
-  ev_tstamp ev_rt_now;
+  EV_API_DECL ev_tstamp ev_rt_now = EV_TS_CONST (0.); /* needs to be initialised to make it a definition despite extern */
   #define VAR(name,decl) static decl;
     #include "ev_vars.h"
   #undef VAR
@@ -739,8 +2147,8 @@ typedef struct
 #endif
 
 #if EV_FEATURE_API
-# define EV_RELEASE_CB if (expect_false (release_cb)) release_cb (EV_A)
-# define EV_ACQUIRE_CB if (expect_false (acquire_cb)) acquire_cb (EV_A)
+# define EV_RELEASE_CB if (ecb_expect_false (release_cb)) release_cb (EV_A)
+# define EV_ACQUIRE_CB if (ecb_expect_false (acquire_cb)) acquire_cb (EV_A)
 # define EV_INVOKE_PENDING invoke_cb (EV_A)
 #else
 # define EV_RELEASE_CB (void)0
@@ -754,20 +2162,22 @@ typedef struct
 
 #ifndef EV_HAVE_EV_TIME
 ev_tstamp
-ev_time (void)
+ev_time (void) EV_NOEXCEPT
 {
 #if EV_USE_REALTIME
-  if (expect_true (have_realtime))
+  if (ecb_expect_true (have_realtime))
     {
       struct timespec ts;
       clock_gettime (CLOCK_REALTIME, &ts);
-      return ts.tv_sec + ts.tv_nsec * 1e-9;
+      return EV_TS_GET (ts);
     }
 #endif
 
-  struct timeval tv;
-  gettimeofday (&tv, 0);
-  return tv.tv_sec + tv.tv_usec * 1e-6;
+  {
+    struct timeval tv;
+    gettimeofday (&tv, 0);
+    return EV_TV_GET (tv);
+  }
 }
 #endif
 
@@ -775,11 +2185,11 @@ inline_size ev_tstamp
 get_clock (void)
 {
 #if EV_USE_MONOTONIC
-  if (expect_true (have_monotonic))
+  if (ecb_expect_true (have_monotonic))
     {
       struct timespec ts;
       clock_gettime (CLOCK_MONOTONIC, &ts);
-      return ts.tv_sec + ts.tv_nsec * 1e-9;
+      return EV_TS_GET (ts);
     }
 #endif
 
@@ -788,24 +2198,26 @@ get_clock (void)
 
 #if EV_MULTIPLICITY
 ev_tstamp
-ev_now (EV_P)
+ev_now (EV_P) EV_NOEXCEPT
 {
   return ev_rt_now;
 }
 #endif
 
 void
-ev_sleep (ev_tstamp delay)
+ev_sleep (ev_tstamp delay) EV_NOEXCEPT
 {
-  if (delay > 0.)
+  if (delay > EV_TS_CONST (0.))
     {
 #if EV_USE_NANOSLEEP
       struct timespec ts;
 
       EV_TS_SET (ts, delay);
       nanosleep (&ts, 0);
-#elif defined(_WIN32)
-      Sleep ((unsigned long)(delay * 1e3));
+#elif defined _WIN32
+      /* maybe this should round up, as ms is very low resolution */
+      /* compared to select (µs) or nanosleep (ns) */
+      Sleep ((unsigned long)(EV_TS_TO_MSEC (delay)));
 #else
       struct timeval tv;
 
@@ -818,14 +2230,6 @@ ev_sleep (ev_tstamp delay)
     }
 }
 
-inline_speed int
-ev_timeout_to_ms (ev_tstamp timeout)
-{
-  int ms = timeout * 1000. + .999999;
-
-  return expect_true (ms) ? ms : timeout < 1e-6 ? 0 : 1;
-}
-
 /*****************************************************************************/
 
 #define MALLOC_ROUND 4096 /* prefer to allocate in chunks of this size, must be 2**n and >> 4 longs */
@@ -841,7 +2245,7 @@ array_nextsize (int elem, int cur, int cnt)
     ncur <<= 1;
   while (cnt > ncur);
 
-  /* if size is large, round to MALLOC_ROUND - 4 * longs to accomodate malloc overhead */
+  /* if size is large, round to MALLOC_ROUND - 4 * longs to accommodate malloc overhead */
   if (elem * ncur > MALLOC_ROUND - sizeof (void *) * 4)
     {
       ncur *= elem;
@@ -853,23 +2257,26 @@ array_nextsize (int elem, int cur, int cnt)
   return ncur;
 }
 
-static noinline void *
+ecb_noinline ecb_cold
+static void *
 array_realloc (int elem, void *base, int *cur, int cnt)
 {
   *cur = array_nextsize (elem, *cur, cnt);
   return ev_realloc (base, elem * *cur);
 }
 
-#define array_init_zero(base,count)	\
-  memset ((void *)(base), 0, sizeof (*(base)) * (count))
+#define array_needsize_noinit(base,offset,count)
+
+#define array_needsize_zerofill(base,offset,count)	\
+  memset ((void *)(base + offset), 0, sizeof (*(base)) * (count))
 
 #define array_needsize(type,base,cur,cnt,init)			\
-  if (expect_false ((cnt) > (cur)))				\
+  if (ecb_expect_false ((cnt) > (cur)))				\
     {								\
-      int ocur_ = (cur);					\
+      ecb_unused int ocur_ = (cur);				\
       (base) = (type *)array_realloc				\
          (sizeof (type), (base), &(cur), (cnt));		\
-      init ((base) + (ocur_), (cur) - ocur_);			\
+      init ((base), ocur_, ((cur) - ocur_));			\
     }
 
 #if 0
@@ -888,32 +2295,36 @@ array_realloc (int elem, void *base, int *cur, int cnt)
 /*****************************************************************************/
 
 /* dummy callback for pending events */
-static void noinline
+ecb_noinline
+static void
 pendingcb (EV_P_ ev_prepare *w, int revents)
 {
 }
 
-void noinline
-ev_feed_event (EV_P_ void *w, int revents)
+ecb_noinline
+void
+ev_feed_event (EV_P_ void *w, int revents) EV_NOEXCEPT
 {
   W w_ = (W)w;
   int pri = ABSPRI (w_);
 
-  if (expect_false (w_->pending))
+  if (ecb_expect_false (w_->pending))
     pendings [pri][w_->pending - 1].events |= revents;
   else
     {
       w_->pending = ++pendingcnt [pri];
-      array_needsize (ANPENDING, pendings [pri], pendingmax [pri], w_->pending, EMPTY2);
+      array_needsize (ANPENDING, pendings [pri], pendingmax [pri], w_->pending, array_needsize_noinit);
       pendings [pri][w_->pending - 1].w      = w_;
       pendings [pri][w_->pending - 1].events = revents;
     }
+
+  pendingpri = NUMPRI - 1;
 }
 
 inline_speed void
 feed_reverse (EV_P_ W w)
 {
-  array_needsize (W, rfeeds, rfeedmax, rfeedcnt + 1, EMPTY2);
+  array_needsize (W, rfeeds, rfeedmax, rfeedcnt + 1, array_needsize_noinit);
   rfeeds [rfeedcnt++] = w;
 }
 
@@ -958,12 +2369,12 @@ fd_event (EV_P_ int fd, int revents)
 {
   ANFD *anfd = anfds + fd;
 
-  if (expect_true (!anfd->reify))
+  if (ecb_expect_true (!anfd->reify))
     fd_event_nocheck (EV_A_ fd, revents);
 }
 
 void
-ev_feed_fd_event (EV_P_ int fd, int revents)
+ev_feed_fd_event (EV_P_ int fd, int revents) EV_NOEXCEPT
 {
   if (fd >= 0 && fd < anfdmax)
     fd_event_nocheck (EV_A_ fd, revents);
@@ -976,13 +2387,25 @@ fd_reify (EV_P)
 {
   int i;
 
+  /* most backends do not modify the fdchanges list in backend_modfiy.
+   * except io_uring, which has fixed-size buffers which might force us
+   * to handle events in backend_modify, causing fdchanges to be amended,
+   * which could result in an endless loop.
+   * to avoid this, we do not dynamically handle fds that were added
+   * during fd_reify. that means that for those backends, fdchangecnt
+   * might be non-zero during poll, which must cause them to not block.
+   * to not put too much of a burden on other backends, this detail
+   * needs to be handled in the backend.
+   */
+  int changecnt = fdchangecnt;
+
 #if EV_SELECT_IS_WINSOCKET || EV_USE_IOCP
-  for (i = 0; i < fdchangecnt; ++i)
+  for (i = 0; i < changecnt; ++i)
     {
       int fd = fdchanges [i];
       ANFD *anfd = anfds + fd;
 
-      if (anfd->reify & EV__IOFDSET)
+      if (anfd->reify & EV__IOFDSET && anfd->head)
         {
           SOCKET handle = EV_FD_TO_WIN32_HANDLE (fd);
 
@@ -1001,7 +2424,7 @@ fd_reify (EV_P)
     }
 #endif
 
-  for (i = 0; i < fdchangecnt; ++i)
+  for (i = 0; i < changecnt; ++i)
     {
       int fd = fdchanges [i];
       ANFD *anfd = anfds + fd;
@@ -1010,9 +2433,9 @@ fd_reify (EV_P)
       unsigned char o_events = anfd->events;
       unsigned char o_reify  = anfd->reify;
 
-      anfd->reify  = 0;
+      anfd->reify = 0;
 
-      /*if (expect_true (o_reify & EV_ANFD_REIFY)) probably a deoptimisation */
+      /*if (ecb_expect_true (o_reify & EV_ANFD_REIFY)) probably a deoptimisation */
         {
           anfd->events = 0;
 
@@ -1027,26 +2450,34 @@ fd_reify (EV_P)
         backend_modify (EV_A_ fd, o_events, anfd->events);
     }
 
-  fdchangecnt = 0;
+  /* normally, fdchangecnt hasn't changed. if it has, then new fds have been added.
+   * this is a rare case (see beginning comment in this function), so we copy them to the
+   * front and hope the backend handles this case.
+   */
+  if (ecb_expect_false (fdchangecnt != changecnt))
+    memmove (fdchanges, fdchanges + changecnt, (fdchangecnt - changecnt) * sizeof (*fdchanges));
+
+  fdchangecnt -= changecnt;
 }
 
 /* something about the given fd changed */
-inline_size void
+inline_size
+void
 fd_change (EV_P_ int fd, int flags)
 {
   unsigned char reify = anfds [fd].reify;
-  anfds [fd].reify |= flags;
+  anfds [fd].reify = reify | flags;
 
-  if (expect_true (!reify))
+  if (ecb_expect_true (!reify))
     {
       ++fdchangecnt;
-      array_needsize (int, fdchanges, fdchangemax, fdchangecnt, EMPTY2);
+      array_needsize (int, fdchanges, fdchangemax, fdchangecnt, array_needsize_noinit);
       fdchanges [fdchangecnt - 1] = fd;
     }
 }
 
 /* the given fd is invalid/unusable, so make sure it doesn't hurt us anymore */
-inline_speed void
+inline_speed ecb_cold void
 fd_kill (EV_P_ int fd)
 {
   ev_io *w;
@@ -1059,7 +2490,7 @@ fd_kill (EV_P_ int fd)
 }
 
 /* check whether the given fd is actually valid, for error recovery */
-inline_size int
+inline_size ecb_cold int
 fd_valid (int fd)
 {
 #ifdef _WIN32
@@ -1070,7 +2501,8 @@ fd_valid (int fd)
 }
 
 /* called on EBADF to verify fds */
-static void noinline
+ecb_noinline ecb_cold
+static void
 fd_ebadf (EV_P)
 {
   int fd;
@@ -1082,7 +2514,8 @@ fd_ebadf (EV_P)
 }
 
 /* called on ENOMEM in select/poll to kill some fds and retry */
-static void noinline
+ecb_noinline ecb_cold
+static void
 fd_enomem (EV_P)
 {
   int fd;
@@ -1096,7 +2529,8 @@ fd_enomem (EV_P)
 }
 
 /* usually called after fork if backend needs to re-arm all fds from scratch */
-static void noinline
+ecb_noinline
+static void
 fd_rearm_all (EV_P)
 {
   int fd;
@@ -1159,19 +2593,19 @@ downheap (ANHE *heap, int N, int k)
       ANHE *pos = heap + DHEAP * (k - HEAP0) + HEAP0 + 1;
 
       /* find minimum child */
-      if (expect_true (pos + DHEAP - 1 < E))
+      if (ecb_expect_true (pos + DHEAP - 1 < E))
         {
           /* fast path */                               (minpos = pos + 0), (minat = ANHE_at (*minpos));
-          if (               ANHE_at (pos [1]) < minat) (minpos = pos + 1), (minat = ANHE_at (*minpos));
-          if (               ANHE_at (pos [2]) < minat) (minpos = pos + 2), (minat = ANHE_at (*minpos));
-          if (               ANHE_at (pos [3]) < minat) (minpos = pos + 3), (minat = ANHE_at (*minpos));
+          if (               minat > ANHE_at (pos [1])) (minpos = pos + 1), (minat = ANHE_at (*minpos));
+          if (               minat > ANHE_at (pos [2])) (minpos = pos + 2), (minat = ANHE_at (*minpos));
+          if (               minat > ANHE_at (pos [3])) (minpos = pos + 3), (minat = ANHE_at (*minpos));
         }
       else if (pos < E)
         {
           /* slow path */                               (minpos = pos + 0), (minat = ANHE_at (*minpos));
-          if (pos + 1 < E && ANHE_at (pos [1]) < minat) (minpos = pos + 1), (minat = ANHE_at (*minpos));
-          if (pos + 2 < E && ANHE_at (pos [2]) < minat) (minpos = pos + 2), (minat = ANHE_at (*minpos));
-          if (pos + 3 < E && ANHE_at (pos [3]) < minat) (minpos = pos + 3), (minat = ANHE_at (*minpos));
+          if (pos + 1 < E && minat > ANHE_at (pos [1])) (minpos = pos + 1), (minat = ANHE_at (*minpos));
+          if (pos + 2 < E && minat > ANHE_at (pos [2])) (minpos = pos + 2), (minat = ANHE_at (*minpos));
+          if (pos + 3 < E && minat > ANHE_at (pos [3])) (minpos = pos + 3), (minat = ANHE_at (*minpos));
         }
       else
         break;
@@ -1189,7 +2623,7 @@ downheap (ANHE *heap, int N, int k)
   ev_active (ANHE_w (he)) = k;
 }
 
-#else /* 4HEAP */
+#else /* not 4HEAP */
 
 #define HEAP0 1
 #define HPARENT(k) ((k) >> 1)
@@ -1271,7 +2705,7 @@ reheap (ANHE *heap, int N)
 
 /*****************************************************************************/
 
-/* associate signal watchers to a signal signal */
+/* associate signal watchers to a signal */
 typedef struct
 {
   EV_ATOMIC_T pending;
@@ -1287,62 +2721,95 @@ static ANSIG signals [EV_NSIG - 1];
 
 #if EV_SIGNAL_ENABLE || EV_ASYNC_ENABLE
 
-static void noinline
+ecb_noinline ecb_cold
+static void
 evpipe_init (EV_P)
 {
   if (!ev_is_active (&pipe_w))
     {
+      int fds [2];
+
 # if EV_USE_EVENTFD
-      evfd = eventfd (0, EFD_NONBLOCK | EFD_CLOEXEC);
-      if (evfd < 0 && errno == EINVAL)
-        evfd = eventfd (0, 0);
+      fds [0] = -1;
+      fds [1] = eventfd (0, EFD_NONBLOCK | EFD_CLOEXEC);
+      if (fds [1] < 0 && errno == EINVAL)
+        fds [1] = eventfd (0, 0);
 
-      if (evfd >= 0)
+      if (fds [1] < 0)
+# endif
         {
-          evpipe [0] = -1;
-          fd_intern (evfd); /* doing it twice doesn't hurt */
-          ev_io_set (&pipe_w, evfd, EV_READ);
+          while (pipe (fds))
+            ev_syserr ("(libev) error creating signal/async pipe");
+
+          fd_intern (fds [0]);
         }
+
+      evpipe [0] = fds [0];
+
+      if (evpipe [1] < 0)
+        evpipe [1] = fds [1]; /* first call, set write fd */
       else
-# endif
         {
-          while (pipe (evpipe))
-            ev_syserr ("(libev) error creating signal/async pipe");
+          /* on subsequent calls, do not change evpipe [1] */
+          /* so that evpipe_write can always rely on its value. */
+          /* this branch does not do anything sensible on windows, */
+          /* so must not be executed on windows */
 
-          fd_intern (evpipe [0]);
-          fd_intern (evpipe [1]);
-          ev_io_set (&pipe_w, evpipe [0], EV_READ);
+          dup2 (fds [1], evpipe [1]);
+          close (fds [1]);
         }
 
+      fd_intern (evpipe [1]);
+
+      ev_io_set (&pipe_w, evpipe [0] < 0 ? evpipe [1] : evpipe [0], EV_READ);
       ev_io_start (EV_A_ &pipe_w);
       ev_unref (EV_A); /* watcher should not keep loop alive */
     }
 }
 
-inline_size void
+inline_speed void
 evpipe_write (EV_P_ EV_ATOMIC_T *flag)
 {
-  if (!*flag)
+  ECB_MEMORY_FENCE; /* push out the write before this function was called, acquire flag */
+
+  if (ecb_expect_true (*flag))
+    return;
+
+  *flag = 1;
+  ECB_MEMORY_FENCE_RELEASE; /* make sure flag is visible before the wakeup */
+
+  pipe_write_skipped = 1;
+
+  ECB_MEMORY_FENCE; /* make sure pipe_write_skipped is visible before we check pipe_write_wanted */
+
+  if (pipe_write_wanted)
     {
-      int old_errno = errno; /* save errno because write might clobber it */
-      char dummy;
+      int old_errno;
+
+      pipe_write_skipped = 0;
+      ECB_MEMORY_FENCE_RELEASE;
 
-      *flag = 1;
+      old_errno = errno; /* save errno because write will clobber it */
 
 #if EV_USE_EVENTFD
-      if (evfd >= 0)
+      if (evpipe [0] < 0)
         {
           uint64_t counter = 1;
-          write (evfd, &counter, sizeof (uint64_t));
+          write (evpipe [1], &counter, sizeof (uint64_t));
         }
       else
 #endif
-        /* win32 people keep sending patches that change this write() to send() */
-        /* and then run away. but send() is wrong, it wants a socket handle on win32 */
-        /* so when you think this write should be a send instead, please find out */
-        /* where your send() is from - it's definitely not the microsoft send, and */
-        /* tell me. thank you. */
-        write (evpipe [1], &dummy, 1);
+        {
+#ifdef _WIN32
+          WSABUF buf;
+          DWORD sent;
+          buf.buf = (char *)&buf;
+          buf.len = 1;
+          WSASend (EV_FD_TO_WIN32_HANDLE (evpipe [1]), &buf, 1, &sent, 0, 0, 0);
+#else
+          write (evpipe [1], &(evpipe [1]), 1);
+#endif
+        }
 
       errno = old_errno;
     }
@@ -1355,27 +2822,44 @@ pipecb (EV_P_ ev_io *iow, int revents)
 {
   int i;
 
-#if EV_USE_EVENTFD
-  if (evfd >= 0)
+  if (revents & EV_READ)
     {
-      uint64_t counter;
-      read (evfd, &counter, sizeof (uint64_t));
-    }
-  else
+#if EV_USE_EVENTFD
+      if (evpipe [0] < 0)
+        {
+          uint64_t counter;
+          read (evpipe [1], &counter, sizeof (uint64_t));
+        }
+      else
 #endif
-    {
-      char dummy;
-      /* see discussion in evpipe_write when you think this read should be recv in win32 */
-      read (evpipe [0], &dummy, 1);
+        {
+          char dummy[4];
+#ifdef _WIN32
+          WSABUF buf;
+          DWORD recvd;
+          DWORD flags = 0;
+          buf.buf = dummy;
+          buf.len = sizeof (dummy);
+          WSARecv (EV_FD_TO_WIN32_HANDLE (evpipe [0]), &buf, 1, &recvd, &flags, 0, 0);
+#else
+          read (evpipe [0], &dummy, sizeof (dummy));
+#endif
+        }
     }
 
+  pipe_write_skipped = 0;
+
+  ECB_MEMORY_FENCE; /* push out skipped, acquire flags */
+
 #if EV_SIGNAL_ENABLE
   if (sig_pending)
     {
       sig_pending = 0;
 
+      ECB_MEMORY_FENCE;
+
       for (i = EV_NSIG - 1; i--; )
-        if (expect_false (signals [i].pending))
+        if (ecb_expect_false (signals [i].pending))
           ev_feed_signal_event (EV_A_ i + 1);
     }
 #endif
@@ -1385,10 +2869,13 @@ pipecb (EV_P_ ev_io *iow, int revents)
     {
       async_pending = 0;
 
+      ECB_MEMORY_FENCE;
+
       for (i = asynccnt; i--; )
         if (asyncs [i]->sent)
           {
             asyncs [i]->sent = 0;
+            ECB_MEMORY_FENCE_RELEASE;
             ev_feed_event (EV_A_ asyncs [i], EV_ASYNC);
           }
     }
@@ -1398,10 +2885,12 @@ pipecb (EV_P_ ev_io *iow, int revents)
 /*****************************************************************************/
 
 void
-ev_feed_signal (int signum)
+ev_feed_signal (int signum) EV_NOEXCEPT
 {
 #if EV_MULTIPLICITY
-  EV_P = signals [signum - 1].loop;
+  EV_P;
+  ECB_MEMORY_FENCE_ACQUIRE;
+  EV_A = signals [signum - 1].loop;
 
   if (!EV_A)
     return;
@@ -1421,12 +2910,13 @@ ev_sighandler (int signum)
   ev_feed_signal (signum);
 }
 
-void noinline
-ev_feed_signal_event (EV_P_ int signum)
+ecb_noinline
+void
+ev_feed_signal_event (EV_P_ int signum) EV_NOEXCEPT
 {
   WL w;
 
-  if (expect_false (signum <= 0 || signum > EV_NSIG))
+  if (ecb_expect_false (signum <= 0 || signum >= EV_NSIG))
     return;
 
   --signum;
@@ -1435,11 +2925,12 @@ ev_feed_signal_event (EV_P_ int signum)
   /* it is permissible to try to feed a signal to the wrong loop */
   /* or, likely more useful, feeding a signal nobody is waiting for */
 
-  if (expect_false (signals [signum].loop != EV_A))
+  if (ecb_expect_false (signals [signum].loop != EV_A))
     return;
 #endif
 
   signals [signum].pending = 0;
+  ECB_MEMORY_FENCE_RELEASE;
 
   for (w = signals [signum].head; w; w = w->next)
     ev_feed_event (EV_A_ (W)w, EV_SIGNAL);
@@ -1528,6 +3019,57 @@ childcb (EV_P_ ev_signal *sw, int revents)
 
 /*****************************************************************************/
 
+#if EV_USE_TIMERFD
+
+static void periodics_reschedule (EV_P);
+
+static void
+timerfdcb (EV_P_ ev_io *iow, int revents)
+{
+  struct itimerspec its = { 0 };
+
+  its.it_value.tv_sec = ev_rt_now + (int)MAX_BLOCKTIME2;
+  timerfd_settime (timerfd, TFD_TIMER_ABSTIME | TFD_TIMER_CANCEL_ON_SET, &its, 0);
+
+  ev_rt_now = ev_time ();
+  /* periodics_reschedule only needs ev_rt_now */
+  /* but maybe in the future we want the full treatment. */
+  /*
+  now_floor = EV_TS_CONST (0.);
+  time_update (EV_A_ EV_TSTAMP_HUGE);
+  */
+#if EV_PERIODIC_ENABLE
+  periodics_reschedule (EV_A);
+#endif
+}
+
+ecb_noinline ecb_cold
+static void
+evtimerfd_init (EV_P)
+{
+  if (!ev_is_active (&timerfd_w))
+    {
+      timerfd = timerfd_create (CLOCK_REALTIME, TFD_NONBLOCK | TFD_CLOEXEC);
+
+      if (timerfd >= 0)
+        {
+          fd_intern (timerfd); /* just to be sure */
+
+          ev_io_init (&timerfd_w, timerfdcb, timerfd, EV_READ);
+          ev_set_priority (&timerfd_w, EV_MINPRI);
+          ev_io_start (EV_A_ &timerfd_w);
+          ev_unref (EV_A); /* watcher should not keep loop alive */
+
+          /* (re-) arm timer */
+          timerfdcb (EV_A_ 0, 0);
+        }
+    }
+}
+
+#endif
+
+/*****************************************************************************/
+
 #if EV_USE_IOCP
 # include "ev_iocp.c"
 #endif
@@ -1540,6 +3082,12 @@ childcb (EV_P_ ev_signal *sw, int revents)
 #if EV_USE_EPOLL
 # include "ev_epoll.c"
 #endif
+#if EV_USE_LINUXAIO
+# include "ev_linuxaio.c"
+#endif
+#if EV_USE_IOURING
+# include "ev_iouring.c"
+#endif
 #if EV_USE_POLL
 # include "ev_poll.c"
 #endif
@@ -1547,20 +3095,20 @@ childcb (EV_P_ ev_signal *sw, int revents)
 # include "ev_select.c"
 #endif
 
-int
-ev_version_major (void)
+ecb_cold int
+ev_version_major (void) EV_NOEXCEPT
 {
   return EV_VERSION_MAJOR;
 }
 
-int
-ev_version_minor (void)
+ecb_cold int
+ev_version_minor (void) EV_NOEXCEPT
 {
   return EV_VERSION_MINOR;
 }
 
 /* return true if we are running with elevated privileges and should ignore env variables */
-int inline_size
+inline_size ecb_cold int
 enable_secure (void)
 {
 #ifdef _WIN32
@@ -1571,103 +3119,123 @@ enable_secure (void)
 #endif
 }
 
+ecb_cold
 unsigned int
-ev_supported_backends (void)
+ev_supported_backends (void) EV_NOEXCEPT
 {
   unsigned int flags = 0;
 
-  if (EV_USE_PORT  ) flags |= EVBACKEND_PORT;
-  if (EV_USE_KQUEUE) flags |= EVBACKEND_KQUEUE;
-  if (EV_USE_EPOLL ) flags |= EVBACKEND_EPOLL;
-  if (EV_USE_POLL  ) flags |= EVBACKEND_POLL;
-  if (EV_USE_SELECT) flags |= EVBACKEND_SELECT;
+  if (EV_USE_PORT                                       ) flags |= EVBACKEND_PORT;
+  if (EV_USE_KQUEUE                                     ) flags |= EVBACKEND_KQUEUE;
+  if (EV_USE_EPOLL                                      ) flags |= EVBACKEND_EPOLL;
+  if (EV_USE_LINUXAIO && ev_linux_version () >= 0x041300) flags |= EVBACKEND_LINUXAIO; /* 4.19+ */
+  if (EV_USE_IOURING && ev_linux_version () >= 0x050601 ) flags |= EVBACKEND_IOURING; /* 5.6.1+ */
+  if (EV_USE_POLL                                       ) flags |= EVBACKEND_POLL;
+  if (EV_USE_SELECT                                     ) flags |= EVBACKEND_SELECT;
 
   return flags;
 }
 
+ecb_cold
 unsigned int
-ev_recommended_backends (void)
+ev_recommended_backends (void) EV_NOEXCEPT
 {
   unsigned int flags = ev_supported_backends ();
 
-#ifndef __NetBSD__
-  /* kqueue is borked on everything but netbsd apparently */
-  /* it usually doesn't work correctly on anything but sockets and pipes */
-  flags &= ~EVBACKEND_KQUEUE;
-#endif
-#ifdef __APPLE__
+/* apple has a poor track record but post 10.12.2 it seems to work sufficiently well */
+#if defined(__APPLE__) && (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_X_VERSION_10_14)
   /* only select works correctly on that "unix-certified" platform */
   flags &= ~EVBACKEND_KQUEUE; /* horribly broken, even for sockets */
   flags &= ~EVBACKEND_POLL;   /* poll is based on kqueue from 10.5 onwards */
 #endif
+
+#if !defined(__NetBSD__) && !defined(__APPLE__)
+  /* kqueue is borked on everything but netbsd and osx >= 10.12.2 apparently */
+  /* it usually doesn't work correctly on anything but sockets and pipes */
+  flags &= ~EVBACKEND_KQUEUE;
+#endif
+
 #ifdef __FreeBSD__
   flags &= ~EVBACKEND_POLL;   /* poll return value is unusable (http://forums.freebsd.org/archive/index.php/t-10270.html) */
 #endif
 
+#ifdef __linux__
+  /* NOTE: linuxaio is very experimental, never recommend */
+  flags &= ~EVBACKEND_LINUXAIO;
+
+  /* NOTE: io_uring is super experimental, never recommend */
+  flags &= ~EVBACKEND_IOURING;
+#endif
+
   return flags;
 }
 
+ecb_cold
 unsigned int
-ev_embeddable_backends (void)
+ev_embeddable_backends (void) EV_NOEXCEPT
 {
-  int flags = EVBACKEND_EPOLL | EVBACKEND_KQUEUE | EVBACKEND_PORT;
+  int flags = EVBACKEND_EPOLL | EVBACKEND_KQUEUE | EVBACKEND_PORT | EVBACKEND_IOURING;
 
   /* epoll embeddability broken on all linux versions up to at least 2.6.23 */
   if (ev_linux_version () < 0x020620) /* disable it on linux < 2.6.32 */
     flags &= ~EVBACKEND_EPOLL;
 
+  /* EVBACKEND_LINUXAIO is theoretically embeddable, but suffers from a performance overhead */
+
   return flags;
 }
 
 unsigned int
-ev_backend (EV_P)
+ev_backend (EV_P) EV_NOEXCEPT
 {
   return backend;
 }
 
 #if EV_FEATURE_API
 unsigned int
-ev_iteration (EV_P)
+ev_iteration (EV_P) EV_NOEXCEPT
 {
   return loop_count;
 }
 
 unsigned int
-ev_depth (EV_P)
+ev_depth (EV_P) EV_NOEXCEPT
 {
   return loop_depth;
 }
 
 void
-ev_set_io_collect_interval (EV_P_ ev_tstamp interval)
+ev_set_io_collect_interval (EV_P_ ev_tstamp interval) EV_NOEXCEPT
 {
   io_blocktime = interval;
 }
 
 void
-ev_set_timeout_collect_interval (EV_P_ ev_tstamp interval)
+ev_set_timeout_collect_interval (EV_P_ ev_tstamp interval) EV_NOEXCEPT
 {
   timeout_blocktime = interval;
 }
 
 void
-ev_set_userdata (EV_P_ void *data)
+ev_set_userdata (EV_P_ void *data) EV_NOEXCEPT
 {
   userdata = data;
 }
 
 void *
-ev_userdata (EV_P)
+ev_userdata (EV_P) EV_NOEXCEPT
 {
   return userdata;
 }
 
-void ev_set_invoke_pending_cb (EV_P_ void (*invoke_pending_cb)(EV_P))
+void
+ev_set_invoke_pending_cb (EV_P_ ev_loop_callback invoke_pending_cb) EV_NOEXCEPT
 {
   invoke_cb = invoke_pending_cb;
 }
 
-void ev_set_loop_release_cb (EV_P_ void (*release)(EV_P), void (*acquire)(EV_P))
+void
+ev_set_loop_release_cb (EV_P_ void (*release)(EV_P) EV_NOEXCEPT, void (*acquire)(EV_P) EV_NOEXCEPT) EV_NOEXCEPT
 {
   release_cb = release;
   acquire_cb = acquire;
@@ -1675,8 +3243,9 @@ void ev_set_loop_release_cb (EV_P_ void (*release)(EV_P), void (*acquire)(EV_P))
 #endif
 
 /* initialise a loop structure, must be zero-initialised */
-static void noinline
-loop_init (EV_P_ unsigned int flags)
+ecb_noinline ecb_cold
+static void
+loop_init (EV_P_ unsigned int flags) EV_NOEXCEPT
 {
   if (!backend)
     {
@@ -1713,49 +3282,62 @@ loop_init (EV_P_ unsigned int flags)
           && getenv ("LIBEV_FLAGS"))
         flags = atoi (getenv ("LIBEV_FLAGS"));
 
-      ev_rt_now         = ev_time ();
-      mn_now            = get_clock ();
-      now_floor         = mn_now;
-      rtmn_diff         = ev_rt_now - mn_now;
+      ev_rt_now          = ev_time ();
+      mn_now             = get_clock ();
+      now_floor          = mn_now;
+      rtmn_diff          = ev_rt_now - mn_now;
 #if EV_FEATURE_API
-      invoke_cb         = ev_invoke_pending;
+      invoke_cb          = ev_invoke_pending;
 #endif
 
-      io_blocktime      = 0.;
-      timeout_blocktime = 0.;
-      backend           = 0;
-      backend_fd        = -1;
-      sig_pending       = 0;
+      io_blocktime       = 0.;
+      timeout_blocktime  = 0.;
+      backend            = 0;
+      backend_fd         = -1;
+      sig_pending        = 0;
 #if EV_ASYNC_ENABLE
-      async_pending     = 0;
+      async_pending      = 0;
 #endif
+      pipe_write_skipped = 0;
+      pipe_write_wanted  = 0;
+      evpipe [0]         = -1;
+      evpipe [1]         = -1;
 #if EV_USE_INOTIFY
-      fs_fd             = flags & EVFLAG_NOINOTIFY ? -1 : -2;
+      fs_fd              = flags & EVFLAG_NOINOTIFY ? -1 : -2;
 #endif
 #if EV_USE_SIGNALFD
-      sigfd             = flags & EVFLAG_SIGNALFD  ? -2 : -1;
+      sigfd              = flags & EVFLAG_SIGNALFD  ? -2 : -1;
+#endif
+#if EV_USE_TIMERFD
+      timerfd            = flags & EVFLAG_NOTIMERFD ? -1 : -2;
 #endif
 
       if (!(flags & EVBACKEND_MASK))
         flags |= ev_recommended_backends ();
 
 #if EV_USE_IOCP
-      if (!backend && (flags & EVBACKEND_IOCP  )) backend = iocp_init   (EV_A_ flags);
+      if (!backend && (flags & EVBACKEND_IOCP    )) backend = iocp_init      (EV_A_ flags);
 #endif
 #if EV_USE_PORT
-      if (!backend && (flags & EVBACKEND_PORT  )) backend = port_init   (EV_A_ flags);
+      if (!backend && (flags & EVBACKEND_PORT    )) backend = port_init      (EV_A_ flags);
 #endif
 #if EV_USE_KQUEUE
-      if (!backend && (flags & EVBACKEND_KQUEUE)) backend = kqueue_init (EV_A_ flags);
+      if (!backend && (flags & EVBACKEND_KQUEUE  )) backend = kqueue_init    (EV_A_ flags);
+#endif
+#if EV_USE_IOURING
+      if (!backend && (flags & EVBACKEND_IOURING )) backend = iouring_init   (EV_A_ flags);
+#endif
+#if EV_USE_LINUXAIO
+      if (!backend && (flags & EVBACKEND_LINUXAIO)) backend = linuxaio_init  (EV_A_ flags);
 #endif
 #if EV_USE_EPOLL
-      if (!backend && (flags & EVBACKEND_EPOLL )) backend = epoll_init  (EV_A_ flags);
+      if (!backend && (flags & EVBACKEND_EPOLL   )) backend = epoll_init     (EV_A_ flags);
 #endif
 #if EV_USE_POLL
-      if (!backend && (flags & EVBACKEND_POLL  )) backend = poll_init   (EV_A_ flags);
+      if (!backend && (flags & EVBACKEND_POLL    )) backend = poll_init      (EV_A_ flags);
 #endif
 #if EV_USE_SELECT
-      if (!backend && (flags & EVBACKEND_SELECT)) backend = select_init (EV_A_ flags);
+      if (!backend && (flags & EVBACKEND_SELECT  )) backend = select_init    (EV_A_ flags);
 #endif
 
       ev_prepare_init (&pending_w, pendingcb);
@@ -1768,6 +3350,7 @@ loop_init (EV_P_ unsigned int flags)
 }
 
 /* free up a loop structure */
+ecb_cold
 void
 ev_loop_destroy (EV_P)
 {
@@ -1781,7 +3364,7 @@ ev_loop_destroy (EV_P)
 
 #if EV_CLEANUP_ENABLE
   /* queue cleanup watchers (and execute them) */
-  if (expect_false (cleanupcnt))
+  if (ecb_expect_false (cleanupcnt))
     {
       queue_events (EV_A_ (W *)cleanups, cleanupcnt, EV_CLEANUP);
       EV_INVOKE_PENDING;
@@ -1789,7 +3372,7 @@ ev_loop_destroy (EV_P)
 #endif
 
 #if EV_CHILD_ENABLE
-  if (ev_is_active (&childev))
+  if (ev_is_default_loop (EV_A) && ev_is_active (&childev))
     {
       ev_ref (EV_A); /* child watcher */
       ev_signal_stop (EV_A_ &childev);
@@ -1801,16 +3384,8 @@ ev_loop_destroy (EV_P)
       /*ev_ref (EV_A);*/
       /*ev_io_stop (EV_A_ &pipe_w);*/
 
-#if EV_USE_EVENTFD
-      if (evfd >= 0)
-        close (evfd);
-#endif
-
-      if (evpipe [0] >= 0)
-        {
-          EV_WIN32_CLOSE_FD (evpipe [0]);
-          EV_WIN32_CLOSE_FD (evpipe [1]);
-        }
+      if (evpipe [0] >= 0) EV_WIN32_CLOSE_FD (evpipe [0]);
+      if (evpipe [1] >= 0) EV_WIN32_CLOSE_FD (evpipe [1]);
     }
 
 #if EV_USE_SIGNALFD
@@ -1818,6 +3393,11 @@ ev_loop_destroy (EV_P)
     close (sigfd);
 #endif
 
+#if EV_USE_TIMERFD
+  if (ev_is_active (&timerfd_w))
+    close (timerfd);
+#endif
+
 #if EV_USE_INOTIFY
   if (fs_fd >= 0)
     close (fs_fd);
@@ -1827,22 +3407,28 @@ ev_loop_destroy (EV_P)
     close (backend_fd);
 
 #if EV_USE_IOCP
-  if (backend == EVBACKEND_IOCP  ) iocp_destroy   (EV_A);
+  if (backend == EVBACKEND_IOCP    ) iocp_destroy     (EV_A);
 #endif
 #if EV_USE_PORT
-  if (backend == EVBACKEND_PORT  ) port_destroy   (EV_A);
+  if (backend == EVBACKEND_PORT    ) port_destroy     (EV_A);
 #endif
 #if EV_USE_KQUEUE
-  if (backend == EVBACKEND_KQUEUE) kqueue_destroy (EV_A);
+  if (backend == EVBACKEND_KQUEUE  ) kqueue_destroy   (EV_A);
+#endif
+#if EV_USE_IOURING
+  if (backend == EVBACKEND_IOURING ) iouring_destroy  (EV_A);
+#endif
+#if EV_USE_LINUXAIO
+  if (backend == EVBACKEND_LINUXAIO) linuxaio_destroy (EV_A);
 #endif
 #if EV_USE_EPOLL
-  if (backend == EVBACKEND_EPOLL ) epoll_destroy  (EV_A);
+  if (backend == EVBACKEND_EPOLL   ) epoll_destroy    (EV_A);
 #endif
 #if EV_USE_POLL
-  if (backend == EVBACKEND_POLL  ) poll_destroy   (EV_A);
+  if (backend == EVBACKEND_POLL    ) poll_destroy     (EV_A);
 #endif
 #if EV_USE_SELECT
-  if (backend == EVBACKEND_SELECT) select_destroy (EV_A);
+  if (backend == EVBACKEND_SELECT  ) select_destroy   (EV_A);
 #endif
 
   for (i = NUMPRI; i--; )
@@ -1894,46 +3480,61 @@ inline_size void
 loop_fork (EV_P)
 {
 #if EV_USE_PORT
-  if (backend == EVBACKEND_PORT  ) port_fork   (EV_A);
+  if (backend == EVBACKEND_PORT    ) port_fork     (EV_A);
 #endif
 #if EV_USE_KQUEUE
-  if (backend == EVBACKEND_KQUEUE) kqueue_fork (EV_A);
+  if (backend == EVBACKEND_KQUEUE  ) kqueue_fork   (EV_A);
+#endif
+#if EV_USE_IOURING
+  if (backend == EVBACKEND_IOURING ) iouring_fork  (EV_A);
+#endif
+#if EV_USE_LINUXAIO
+  if (backend == EVBACKEND_LINUXAIO) linuxaio_fork (EV_A);
 #endif
 #if EV_USE_EPOLL
-  if (backend == EVBACKEND_EPOLL ) epoll_fork  (EV_A);
+  if (backend == EVBACKEND_EPOLL   ) epoll_fork    (EV_A);
 #endif
 #if EV_USE_INOTIFY
   infy_fork (EV_A);
 #endif
 
-  if (ev_is_active (&pipe_w))
+  if (postfork != 2)
     {
-      /* this "locks" the handlers against writing to the pipe */
-      /* while we modify the fd vars */
-      sig_pending   = 1;
-#if EV_ASYNC_ENABLE
-      async_pending = 1;
-#endif
+      #if EV_USE_SIGNALFD
+        /* surprisingly, nothing needs to be done for signalfd, accoridng to docs, it does the right thing on fork */
+      #endif
 
-      ev_ref (EV_A);
-      ev_io_stop (EV_A_ &pipe_w);
+      #if EV_USE_TIMERFD
+        if (ev_is_active (&timerfd_w))
+          {
+            ev_ref (EV_A);
+            ev_io_stop (EV_A_ &timerfd_w);
 
-#if EV_USE_EVENTFD
-      if (evfd >= 0)
-        close (evfd);
-#endif
+            close (timerfd);
+            timerfd = -2;
 
-      if (evpipe [0] >= 0)
-        {
-          EV_WIN32_CLOSE_FD (evpipe [0]);
-          EV_WIN32_CLOSE_FD (evpipe [1]);
-        }
+            evtimerfd_init (EV_A);
+            /* reschedule periodics, in case we missed something */
+            ev_feed_event (EV_A_ &timerfd_w, EV_CUSTOM);
+          }
+      #endif
 
-#if EV_SIGNAL_ENABLE || EV_ASYNC_ENABLE
-      evpipe_init (EV_A);
-      /* now iterate over everything, in case we missed something */
-      pipecb (EV_A_ &pipe_w, EV_READ);
-#endif
+      #if EV_SIGNAL_ENABLE || EV_ASYNC_ENABLE
+        if (ev_is_active (&pipe_w))
+          {
+            /* pipe_write_wanted must be false now, so modifying fd vars should be safe */
+
+            ev_ref (EV_A);
+            ev_io_stop (EV_A_ &pipe_w);
+
+            if (evpipe [0] >= 0)
+              EV_WIN32_CLOSE_FD (evpipe [0]);
+
+            evpipe_init (EV_A);
+            /* iterate over everything, in case we missed something before */
+            ev_feed_event (EV_A_ &pipe_w, EV_CUSTOM);
+          }
+      #endif
     }
 
   postfork = 0;
@@ -1941,8 +3542,9 @@ loop_fork (EV_P)
 
 #if EV_MULTIPLICITY
 
+ecb_cold
 struct ev_loop *
-ev_loop_new (unsigned int flags)
+ev_loop_new (unsigned int flags) EV_NOEXCEPT
 {
   EV_P = (struct ev_loop *)ev_malloc (sizeof (struct ev_loop));
 
@@ -1959,7 +3561,8 @@ ev_loop_new (unsigned int flags)
 #endif /* multiplicity */
 
 #if EV_VERIFY
-static void noinline
+ecb_noinline ecb_cold
+static void
 verify_watcher (EV_P_ W w)
 {
   assert (("libev: watcher has invalid priority", ABSPRI (w) >= 0 && ABSPRI (w) < NUMPRI));
@@ -1968,7 +3571,8 @@ verify_watcher (EV_P_ W w)
     assert (("libev: pending watcher not on pending queue", pendings [ABSPRI (w)][w->pending - 1].w == w));
 }
 
-static void noinline
+ecb_noinline ecb_cold
+static void
 verify_heap (EV_P_ ANHE *heap, int N)
 {
   int i;
@@ -1983,7 +3587,8 @@ verify_heap (EV_P_ ANHE *heap, int N)
     }
 }
 
-static void noinline
+ecb_noinline ecb_cold
+static void
 array_verify (EV_P_ W *ws, int cnt)
 {
   while (cnt--)
@@ -1995,12 +3600,12 @@ array_verify (EV_P_ W *ws, int cnt)
 #endif
 
 #if EV_FEATURE_API
-void
-ev_verify (EV_P)
+void ecb_cold
+ev_verify (EV_P) EV_NOEXCEPT
 {
 #if EV_VERIFY
   int i;
-  WL w;
+  WL w, w2;
 
   assert (activecnt >= -1);
 
@@ -2010,12 +3615,23 @@ ev_verify (EV_P)
 
   assert (anfdmax >= 0);
   for (i = 0; i < anfdmax; ++i)
-    for (w = anfds [i].head; w; w = w->next)
-      {
-        verify_watcher (EV_A_ (W)w);
-        assert (("libev: inactive fd watcher on anfd list", ev_active (w) == 1));
-        assert (("libev: fd mismatch between watcher and anfd", ((ev_io *)w)->fd == i));
-      }
+    {
+      int j = 0;
+
+      for (w = w2 = anfds [i].head; w; w = w->next)
+        {
+          verify_watcher (EV_A_ (W)w);
+
+          if (j++ & 1)
+            {
+              assert (("libev: io watcher list contains a loop", w != w2));
+              w2 = w2->next;
+            }
+
+          assert (("libev: inactive fd watcher on anfd list", ev_active (w) == 1));
+          assert (("libev: fd mismatch between watcher and anfd", ((ev_io *)w)->fd == i));
+        }
+    }
 
   assert (timermax >= timercnt);
   verify_heap (EV_A_ timers, timercnt);
@@ -2071,11 +3687,12 @@ ev_verify (EV_P)
 #endif
 
 #if EV_MULTIPLICITY
+ecb_cold
 struct ev_loop *
 #else
 int
 #endif
-ev_default_loop (unsigned int flags)
+ev_default_loop (unsigned int flags) EV_NOEXCEPT
 {
   if (!ev_default_loop_ptr)
     {
@@ -2104,9 +3721,9 @@ ev_default_loop (unsigned int flags)
 }
 
 void
-ev_loop_fork (EV_P)
+ev_loop_fork (EV_P) EV_NOEXCEPT
 {
-  postfork = 1; /* must be in line with ev_default_fork */
+  postfork = 1;
 }
 
 /*****************************************************************************/
@@ -2118,7 +3735,7 @@ ev_invoke (EV_P_ void *w, int revents)
 }
 
 unsigned int
-ev_pending_count (EV_P)
+ev_pending_count (EV_P) EV_NOEXCEPT
 {
   int pri;
   unsigned int count = 0;
@@ -2129,20 +3746,27 @@ ev_pending_count (EV_P)
   return count;
 }
 
-void noinline
+ecb_noinline
+void
 ev_invoke_pending (EV_P)
 {
-  int pri;
+  pendingpri = NUMPRI;
 
-  for (pri = NUMPRI; pri--; )
-    while (pendingcnt [pri])
-      {
-        ANPENDING *p = pendings [pri] + --pendingcnt [pri];
+  do
+    {
+      --pendingpri;
 
-        p->w->pending = 0;
-        EV_CB_INVOKE (p->w, p->events);
-        EV_FREQUENT_CHECK;
-      }
+      /* pendingpri possibly gets modified in the inner loop */
+      while (pendingcnt [pendingpri])
+        {
+          ANPENDING *p = pendings [pendingpri] + --pendingcnt [pendingpri];
+
+          p->w->pending = 0;
+          EV_CB_INVOKE (p->w, p->events);
+          EV_FREQUENT_CHECK;
+        }
+    }
+  while (pendingpri);
 }
 
 #if EV_IDLE_ENABLE
@@ -2151,7 +3775,7 @@ ev_invoke_pending (EV_P)
 inline_size void
 idle_reify (EV_P)
 {
-  if (expect_false (idleall))
+  if (ecb_expect_false (idleall))
     {
       int pri;
 
@@ -2191,7 +3815,7 @@ timers_reify (EV_P)
               if (ev_at (w) < mn_now)
                 ev_at (w) = mn_now;
 
-              assert (("libev: negative ev_timer repeat value found while processing timers", w->repeat > 0.));
+              assert (("libev: negative ev_timer repeat value found while processing timers", w->repeat > EV_TS_CONST (0.)));
 
               ANHE_at_cache (timers [HEAP0]);
               downheap (timers, timercnt, HEAP0);
@@ -2210,12 +3834,29 @@ timers_reify (EV_P)
 
 #if EV_PERIODIC_ENABLE
 
-inline_speed void
+ecb_noinline
+static void
 periodic_recalc (EV_P_ ev_periodic *w)
 {
-  /* TODO: use slow but potentially more correct incremental algo, */
-  /* also do not rely on ceil */
-  ev_at (w) = w->offset + ceil ((ev_rt_now - w->offset) / w->interval) * w->interval;
+  ev_tstamp interval = w->interval > MIN_INTERVAL ? w->interval : MIN_INTERVAL;
+  ev_tstamp at = w->offset + interval * ev_floor ((ev_rt_now - w->offset) / interval);
+
+  /* the above almost always errs on the low side */
+  while (at <= ev_rt_now)
+    {
+      ev_tstamp nat = at + w->interval;
+
+      /* when resolution fails us, we use ev_rt_now */
+      if (ecb_expect_false (nat == at))
+        {
+          at = ev_rt_now;
+          break;
+        }
+
+      at = nat;
+    }
+
+  ev_at (w) = at;
 }
 
 /* make periodics pending */
@@ -2226,8 +3867,6 @@ periodics_reify (EV_P)
 
   while (periodiccnt && ANHE_at (periodics [HEAP0]) < ev_rt_now)
     {
-      int feed_count = 0;
-
       do
         {
           ev_periodic *w = (ev_periodic *)ANHE_w (periodics [HEAP0]);
@@ -2247,20 +3886,6 @@ periodics_reify (EV_P)
           else if (w->interval)
             {
               periodic_recalc (EV_A_ w);
-
-              /* if next trigger time is not sufficiently in the future, put it there */
-              /* this might happen because of floating point inexactness */
-              if (ev_at (w) - ev_rt_now < TIME_EPSILON)
-                {
-                  ev_at (w) += w->interval;
-
-                  /* if interval is unreasonably low we might still have a time in the past */
-                  /* so correct this. this will make the periodic very inexact, but the user */
-                  /* has effectively asked to get triggered more often than possible */
-                  if (ev_at (w) < ev_rt_now)
-                    ev_at (w) = ev_rt_now;
-                }
-
               ANHE_at_cache (periodics [HEAP0]);
               downheap (periodics, periodiccnt, HEAP0);
             }
@@ -2278,7 +3903,8 @@ periodics_reify (EV_P)
 
 /* simply recalculate all periodics */
 /* TODO: maybe ensure that at least one event happens when jumping forward? */
-static void noinline
+ecb_noinline ecb_cold
+static void
 periodics_reschedule (EV_P)
 {
   int i;
@@ -2301,7 +3927,8 @@ periodics_reschedule (EV_P)
 #endif
 
 /* adjust all timers by a given offset */
-static void noinline
+ecb_noinline ecb_cold
+static void
 timers_reschedule (EV_P_ ev_tstamp adjust)
 {
   int i;
@@ -2320,7 +3947,7 @@ inline_speed void
 time_update (EV_P_ ev_tstamp max_block)
 {
 #if EV_USE_MONOTONIC
-  if (expect_true (have_monotonic))
+  if (ecb_expect_true (have_monotonic))
     {
       int i;
       ev_tstamp odiff = rtmn_diff;
@@ -2329,7 +3956,7 @@ time_update (EV_P_ ev_tstamp max_block)
 
       /* only fetch the realtime clock every 0.5*MIN_TIMEJUMP seconds */
       /* interpolate in the meantime */
-      if (expect_true (mn_now - now_floor < MIN_TIMEJUMP * .5))
+      if (ecb_expect_true (mn_now - now_floor < EV_TS_CONST (MIN_TIMEJUMP * .5)))
         {
           ev_rt_now = rtmn_diff + mn_now;
           return;
@@ -2348,9 +3975,12 @@ time_update (EV_P_ ev_tstamp max_block)
        */
       for (i = 4; --i; )
         {
+          ev_tstamp diff;
           rtmn_diff = ev_rt_now - mn_now;
 
-          if (expect_true (fabs (odiff - rtmn_diff) < MIN_TIMEJUMP))
+          diff = odiff - rtmn_diff;
+
+          if (ecb_expect_true ((diff < EV_TS_CONST (0.) ? -diff : diff) < EV_TS_CONST (MIN_TIMEJUMP)))
             return; /* all is well */
 
           ev_rt_now = ev_time ();
@@ -2369,7 +3999,7 @@ time_update (EV_P_ ev_tstamp max_block)
     {
       ev_rt_now = ev_time ();
 
-      if (expect_false (mn_now > ev_rt_now || ev_rt_now > mn_now + max_block + MIN_TIMEJUMP))
+      if (ecb_expect_false (mn_now > ev_rt_now || ev_rt_now > mn_now + max_block + EV_TS_CONST (MIN_TIMEJUMP)))
         {
           /* adjust timers. this is easy, as the offset is the same for all of them */
           timers_reschedule (EV_A_ ev_rt_now - mn_now);
@@ -2382,9 +4012,30 @@ time_update (EV_P_ ev_tstamp max_block)
     }
 }
 
-void
+/* ########## NIO4R PATCHERY HO! ########## */
+struct ev_poll_args {
+  struct ev_loop *loop;
+  ev_tstamp waittime;
+};
+
+static
+void * ev_backend_poll(void *ptr)
+{
+  struct ev_poll_args *args = (struct ev_poll_args *)ptr;
+  struct ev_loop *loop = args->loop;
+  backend_poll (EV_A_ args->waittime);
+
+  return NULL;
+}
+/* ######################################## */
+
+int
 ev_run (EV_P_ int flags)
 {
+/* ########## NIO4R PATCHERY HO! ########## */
+  struct ev_poll_args poll_args;
+/* ######################################## */
+
 #if EV_FEATURE_API
   ++loop_depth;
 #endif
@@ -2402,8 +4053,8 @@ ev_run (EV_P_ int flags)
 #endif
 
 #ifndef _WIN32
-      if (expect_false (curpid)) /* penalise the forking check even more */
-        if (expect_false (getpid () != curpid))
+      if (ecb_expect_false (curpid)) /* penalise the forking check even more */
+        if (ecb_expect_false (getpid () != curpid))
           {
             curpid = getpid ();
             postfork = 1;
@@ -2412,7 +4063,7 @@ ev_run (EV_P_ int flags)
 
 #if EV_FORK_ENABLE
       /* we might have forked, so queue fork handlers */
-      if (expect_false (postfork))
+      if (ecb_expect_false (postfork))
         if (forkcnt)
           {
             queue_events (EV_A_ (W *)forks, forkcnt, EV_FORK);
@@ -2422,18 +4073,18 @@ ev_run (EV_P_ int flags)
 
 #if EV_PREPARE_ENABLE
       /* queue prepare watchers (and execute them) */
-      if (expect_false (preparecnt))
+      if (ecb_expect_false (preparecnt))
         {
           queue_events (EV_A_ (W *)prepares, preparecnt, EV_PREPARE);
           EV_INVOKE_PENDING;
         }
 #endif
 
-      if (expect_false (loop_done))
+      if (ecb_expect_false (loop_done))
         break;
 
       /* we might have forked, so reify kernel state if necessary */
-      if (expect_false (postfork))
+      if (ecb_expect_false (postfork))
         loop_fork (EV_A);
 
       /* update fd-related kernel structures */
@@ -2448,39 +4099,66 @@ ev_run (EV_P_ int flags)
         ev_tstamp prev_mn_now = mn_now;
 
         /* update time to cancel out callback processing overhead */
-        time_update (EV_A_ 1e100);
+        time_update (EV_A_ EV_TS_CONST (EV_TSTAMP_HUGE));
+
+        /* from now on, we want a pipe-wake-up */
+        pipe_write_wanted = 1;
 
-        if (expect_true (!(flags & EVRUN_NOWAIT || idleall || !activecnt)))
+        ECB_MEMORY_FENCE; /* make sure pipe_write_wanted is visible before we check for potential skips */
+
+        if (ecb_expect_true (!(flags & EVRUN_NOWAIT || idleall || !activecnt || pipe_write_skipped)))
           {
-            waittime = MAX_BLOCKTIME;
+            waittime = EV_TS_CONST (MAX_BLOCKTIME);
+
+#if EV_USE_TIMERFD
+            /* sleep a lot longer when we can reliably detect timejumps */
+            if (ecb_expect_true (timerfd >= 0))
+              waittime = EV_TS_CONST (MAX_BLOCKTIME2);
+#endif
+#if !EV_PERIODIC_ENABLE
+            /* without periodics but with monotonic clock there is no need */
+            /* for any time jump detection, so sleep longer */
+            if (ecb_expect_true (have_monotonic))
+              waittime = EV_TS_CONST (MAX_BLOCKTIME2);
+#endif
 
             if (timercnt)
               {
-                ev_tstamp to = ANHE_at (timers [HEAP0]) - mn_now + backend_fudge;
+                ev_tstamp to = ANHE_at (timers [HEAP0]) - mn_now;
                 if (waittime > to) waittime = to;
               }
 
 #if EV_PERIODIC_ENABLE
             if (periodiccnt)
               {
-                ev_tstamp to = ANHE_at (periodics [HEAP0]) - ev_rt_now + backend_fudge;
+                ev_tstamp to = ANHE_at (periodics [HEAP0]) - ev_rt_now;
                 if (waittime > to) waittime = to;
               }
 #endif
 
             /* don't let timeouts decrease the waittime below timeout_blocktime */
-            if (expect_false (waittime < timeout_blocktime))
+            if (ecb_expect_false (waittime < timeout_blocktime))
               waittime = timeout_blocktime;
 
+            /* now there are two more special cases left, either we have
+             * already-expired timers, so we should not sleep, or we have timers
+             * that expire very soon, in which case we need to wait for a minimum
+             * amount of time for some event loop backends.
+             */
+            if (ecb_expect_false (waittime < backend_mintime))
+              waittime = waittime <= EV_TS_CONST (0.)
+                 ? EV_TS_CONST (0.)
+                 : backend_mintime;
+
             /* extra check because io_blocktime is commonly 0 */
-            if (expect_false (io_blocktime))
+            if (ecb_expect_false (io_blocktime))
               {
                 sleeptime = io_blocktime - (mn_now - prev_mn_now);
 
-                if (sleeptime > waittime - backend_fudge)
-                  sleeptime = waittime - backend_fudge;
+                if (sleeptime > waittime - backend_mintime)
+                  sleeptime = waittime - backend_mintime;
 
-                if (expect_true (sleeptime > 0.))
+                if (ecb_expect_true (sleeptime > EV_TS_CONST (0.)))
                   {
                     ev_sleep (sleeptime);
                     waittime -= sleeptime;
@@ -2492,9 +4170,65 @@ ev_run (EV_P_ int flags)
         ++loop_count;
 #endif
         assert ((loop_done = EVBREAK_RECURSE, 1)); /* assert for side effect */
-        backend_poll (EV_A_ waittime);
+
+/*
+########################## NIO4R PATCHERY HO! ##########################
+
+According to the grandwizards of Ruby, locking and unlocking of the global
+interpreter lock are apparently too powerful a concept for a mere mortal to
+wield (although redefining what + and - do to numbers is totally cool).
+And so it came to pass that the only acceptable way to release the global
+interpreter lock is through a convoluted callback system that thakes a
+function pointer. While the grandwizard of libev foresaw this sort of scenario,
+he too attempted to place an API with callbacks on it, one that runs before
+the system call, and one that runs immediately after.
+
+And so it came to pass that trying to wrap everything up in callbacks created
+two incompatible APIs, Ruby's which releases the global interpreter lock and
+reacquires it when the callback returns, and libev's, which wants two
+callbacks, one which runs before the polling operation starts, and one which
+runs after it finishes.
+
+These two systems are incompatible as they both want to use callbacks to
+solve the same problem, however libev wants to use before/after callbacks,
+and Ruby wants to use an "around" callback. This presents a significant
+problem as these two patterns of callbacks are diametrical opposites of each
+other and thus cannot be composed.
+
+And thus we are left with no choice but to patch the internals of libev in
+order to release a mutex at just the precise moment.
+
+This is a great example of a situation where granular locking and unlocking
+of the GVL is practically required. The goal is to get as close to the
+system call as possible, and to keep the GVL unlocked for the shortest
+amount of time possible.
+
+Perhaps Ruby could benefit from such an API, e.g:
+
+rb_thread_unsafe_dangerous_crazy_blocking_region_begin(...);
+rb_thread_unsafe_dangerous_crazy_blocking_region_end(...);
+
+#######################################################################
+*/
+
+        poll_args.loop = loop;
+        poll_args.waittime = waittime;
+        rb_thread_call_without_gvl(ev_backend_poll, (void *)&poll_args, RUBY_UBF_IO, 0);
+/*
+############################# END PATCHERY ############################
+*/
+
         assert ((loop_done = EVBREAK_CANCEL, 1)); /* assert for side effect */
 
+        pipe_write_wanted = 0; /* just an optimisation, no fence needed */
+
+        ECB_MEMORY_FENCE_ACQUIRE;
+        if (pipe_write_skipped)
+          {
+            assert (("libev: pipe_w not active, but pipe not written", ev_is_active (&pipe_w)));
+            ev_feed_event (EV_A_ &pipe_w, EV_CUSTOM);
+          }
+
         /* update ev_rt_now, do magic */
         time_update (EV_A_ waittime + sleeptime);
       }
@@ -2512,13 +4246,13 @@ ev_run (EV_P_ int flags)
 
 #if EV_CHECK_ENABLE
       /* queue check watchers, to be executed first */
-      if (expect_false (checkcnt))
+      if (ecb_expect_false (checkcnt))
         queue_events (EV_A_ (W *)checks, checkcnt, EV_CHECK);
 #endif
 
       EV_INVOKE_PENDING;
     }
-  while (expect_true (
+  while (ecb_expect_true (
     activecnt
     && !loop_done
     && !(flags & (EVRUN_ONCE | EVRUN_NOWAIT))
@@ -2530,40 +4264,42 @@ ev_run (EV_P_ int flags)
 #if EV_FEATURE_API
   --loop_depth;
 #endif
+
+  return activecnt;
 }
 
 void
-ev_break (EV_P_ int how)
+ev_break (EV_P_ int how) EV_NOEXCEPT
 {
   loop_done = how;
 }
 
 void
-ev_ref (EV_P)
+ev_ref (EV_P) EV_NOEXCEPT
 {
   ++activecnt;
 }
 
 void
-ev_unref (EV_P)
+ev_unref (EV_P) EV_NOEXCEPT
 {
   --activecnt;
 }
 
 void
-ev_now_update (EV_P)
+ev_now_update (EV_P) EV_NOEXCEPT
 {
-  time_update (EV_A_ 1e100);
+  time_update (EV_A_ EV_TSTAMP_HUGE);
 }
 
 void
-ev_suspend (EV_P)
+ev_suspend (EV_P) EV_NOEXCEPT
 {
   ev_now_update (EV_A);
 }
 
 void
-ev_resume (EV_P)
+ev_resume (EV_P) EV_NOEXCEPT
 {
   ev_tstamp mn_prev = mn_now;
 
@@ -2590,7 +4326,7 @@ wlist_del (WL *head, WL elem)
 {
   while (*head)
     {
-      if (expect_true (*head == elem))
+      if (ecb_expect_true (*head == elem))
         {
           *head = elem->next;
           break;
@@ -2612,12 +4348,12 @@ clear_pending (EV_P_ W w)
 }
 
 int
-ev_clear_pending (EV_P_ void *w)
+ev_clear_pending (EV_P_ void *w) EV_NOEXCEPT
 {
   W w_ = (W)w;
   int pending = w_->pending;
 
-  if (expect_true (pending))
+  if (ecb_expect_true (pending))
     {
       ANPENDING *p = pendings [ABSPRI (w_)] + pending - 1;
       p->w = (W)&pending_w;
@@ -2654,38 +4390,49 @@ ev_stop (EV_P_ W w)
 
 /*****************************************************************************/
 
-void noinline
-ev_io_start (EV_P_ ev_io *w)
+ecb_noinline
+void
+ev_io_start (EV_P_ ev_io *w) EV_NOEXCEPT
 {
   int fd = w->fd;
 
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   assert (("libev: ev_io_start called with negative fd", fd >= 0));
   assert (("libev: ev_io_start called with illegal event mask", !(w->events & ~(EV__IOFDSET | EV_READ | EV_WRITE))));
 
+#if EV_VERIFY >= 2
+  assert (("libev: ev_io_start called on watcher with invalid fd", fd_valid (fd)));
+#endif
   EV_FREQUENT_CHECK;
 
   ev_start (EV_A_ (W)w, 1);
-  array_needsize (ANFD, anfds, anfdmax, fd + 1, array_init_zero);
+  array_needsize (ANFD, anfds, anfdmax, fd + 1, array_needsize_zerofill);
   wlist_add (&anfds[fd].head, (WL)w);
 
+  /* common bug, apparently */
+  assert (("libev: ev_io_start called with corrupted watcher", ((WL)w)->next != (WL)w));
+
   fd_change (EV_A_ fd, w->events & EV__IOFDSET | EV_ANFD_REIFY);
   w->events &= ~EV__IOFDSET;
 
   EV_FREQUENT_CHECK;
 }
 
-void noinline
-ev_io_stop (EV_P_ ev_io *w)
+ecb_noinline
+void
+ev_io_stop (EV_P_ ev_io *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   assert (("libev: ev_io_stop called with illegal fd (must stay constant after start!)", w->fd >= 0 && w->fd < anfdmax));
 
+#if EV_VERIFY >= 2
+  assert (("libev: ev_io_stop called on watcher with invalid fd", fd_valid (w->fd)));
+#endif
   EV_FREQUENT_CHECK;
 
   wlist_del (&anfds[w->fd].head, (WL)w);
@@ -2696,10 +4443,11 @@ ev_io_stop (EV_P_ ev_io *w)
   EV_FREQUENT_CHECK;
 }
 
-void noinline
-ev_timer_start (EV_P_ ev_timer *w)
+ecb_noinline
+void
+ev_timer_start (EV_P_ ev_timer *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   ev_at (w) += mn_now;
@@ -2710,7 +4458,7 @@ ev_timer_start (EV_P_ ev_timer *w)
 
   ++timercnt;
   ev_start (EV_A_ (W)w, timercnt + HEAP0 - 1);
-  array_needsize (ANHE, timers, timermax, ev_active (w) + 1, EMPTY2);
+  array_needsize (ANHE, timers, timermax, ev_active (w) + 1, array_needsize_noinit);
   ANHE_w (timers [ev_active (w)]) = (WT)w;
   ANHE_at_cache (timers [ev_active (w)]);
   upheap (timers, ev_active (w));
@@ -2720,11 +4468,12 @@ ev_timer_start (EV_P_ ev_timer *w)
   /*assert (("libev: internal timer heap corruption", timers [ev_active (w)] == (WT)w));*/
 }
 
-void noinline
-ev_timer_stop (EV_P_ ev_timer *w)
+ecb_noinline
+void
+ev_timer_stop (EV_P_ ev_timer *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -2736,7 +4485,7 @@ ev_timer_stop (EV_P_ ev_timer *w)
 
     --timercnt;
 
-    if (expect_true (active < timercnt + HEAP0))
+    if (ecb_expect_true (active < timercnt + HEAP0))
       {
         timers [active] = timers [timercnt + HEAP0];
         adjustheap (timers, timercnt, active);
@@ -2750,11 +4499,14 @@ ev_timer_stop (EV_P_ ev_timer *w)
   EV_FREQUENT_CHECK;
 }
 
-void noinline
-ev_timer_again (EV_P_ ev_timer *w)
+ecb_noinline
+void
+ev_timer_again (EV_P_ ev_timer *w) EV_NOEXCEPT
 {
   EV_FREQUENT_CHECK;
 
+  clear_pending (EV_A_ (W)w);
+
   if (ev_is_active (w))
     {
       if (w->repeat)
@@ -2776,18 +4528,24 @@ ev_timer_again (EV_P_ ev_timer *w)
 }
 
 ev_tstamp
-ev_timer_remaining (EV_P_ ev_timer *w)
+ev_timer_remaining (EV_P_ ev_timer *w) EV_NOEXCEPT
 {
-  return ev_at (w) - (ev_is_active (w) ? mn_now : 0.);
+  return ev_at (w) - (ev_is_active (w) ? mn_now : EV_TS_CONST (0.));
 }
 
 #if EV_PERIODIC_ENABLE
-void noinline
-ev_periodic_start (EV_P_ ev_periodic *w)
+ecb_noinline
+void
+ev_periodic_start (EV_P_ ev_periodic *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
+#if EV_USE_TIMERFD
+  if (timerfd == -2)
+    evtimerfd_init (EV_A);
+#endif
+
   if (w->reschedule_cb)
     ev_at (w) = w->reschedule_cb (w, ev_rt_now);
   else if (w->interval)
@@ -2802,7 +4560,7 @@ ev_periodic_start (EV_P_ ev_periodic *w)
 
   ++periodiccnt;
   ev_start (EV_A_ (W)w, periodiccnt + HEAP0 - 1);
-  array_needsize (ANHE, periodics, periodicmax, ev_active (w) + 1, EMPTY2);
+  array_needsize (ANHE, periodics, periodicmax, ev_active (w) + 1, array_needsize_noinit);
   ANHE_w (periodics [ev_active (w)]) = (WT)w;
   ANHE_at_cache (periodics [ev_active (w)]);
   upheap (periodics, ev_active (w));
@@ -2812,11 +4570,12 @@ ev_periodic_start (EV_P_ ev_periodic *w)
   /*assert (("libev: internal periodic heap corruption", ANHE_w (periodics [ev_active (w)]) == (WT)w));*/
 }
 
-void noinline
-ev_periodic_stop (EV_P_ ev_periodic *w)
+ecb_noinline
+void
+ev_periodic_stop (EV_P_ ev_periodic *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -2828,7 +4587,7 @@ ev_periodic_stop (EV_P_ ev_periodic *w)
 
     --periodiccnt;
 
-    if (expect_true (active < periodiccnt + HEAP0))
+    if (ecb_expect_true (active < periodiccnt + HEAP0))
       {
         periodics [active] = periodics [periodiccnt + HEAP0];
         adjustheap (periodics, periodiccnt, active);
@@ -2840,8 +4599,9 @@ ev_periodic_stop (EV_P_ ev_periodic *w)
   EV_FREQUENT_CHECK;
 }
 
-void noinline
-ev_periodic_again (EV_P_ ev_periodic *w)
+ecb_noinline
+void
+ev_periodic_again (EV_P_ ev_periodic *w) EV_NOEXCEPT
 {
   /* TODO: use adjustheap and recalculation */
   ev_periodic_stop (EV_A_ w);
@@ -2855,10 +4615,11 @@ ev_periodic_again (EV_P_ ev_periodic *w)
 
 #if EV_SIGNAL_ENABLE
 
-void noinline
-ev_signal_start (EV_P_ ev_signal *w)
+ecb_noinline
+void
+ev_signal_start (EV_P_ ev_signal *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   assert (("libev: ev_signal_start called with illegal signal number", w->signum > 0 && w->signum < EV_NSIG));
@@ -2868,6 +4629,7 @@ ev_signal_start (EV_P_ ev_signal *w)
            !signals [w->signum - 1].loop || signals [w->signum - 1].loop == loop));
 
   signals [w->signum - 1].loop = EV_A;
+  ECB_MEMORY_FENCE_RELEASE;
 #endif
 
   EV_FREQUENT_CHECK;
@@ -2936,11 +4698,12 @@ ev_signal_start (EV_P_ ev_signal *w)
   EV_FREQUENT_CHECK;
 }
 
-void noinline
-ev_signal_stop (EV_P_ ev_signal *w)
+ecb_noinline
+void
+ev_signal_stop (EV_P_ ev_signal *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -2978,12 +4741,12 @@ ev_signal_stop (EV_P_ ev_signal *w)
 #if EV_CHILD_ENABLE
 
 void
-ev_child_start (EV_P_ ev_child *w)
+ev_child_start (EV_P_ ev_child *w) EV_NOEXCEPT
 {
 #if EV_MULTIPLICITY
   assert (("libev: child watchers are only supported in the default loop", loop == ev_default_loop_ptr));
 #endif
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -2995,10 +4758,10 @@ ev_child_start (EV_P_ ev_child *w)
 }
 
 void
-ev_child_stop (EV_P_ ev_child *w)
+ev_child_stop (EV_P_ ev_child *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3022,17 +4785,21 @@ ev_child_stop (EV_P_ ev_child *w)
 #define NFS_STAT_INTERVAL 30.1074891 /* for filesystems potentially failing inotify */
 #define MIN_STAT_INTERVAL  0.1074891
 
-static void noinline stat_timer_cb (EV_P_ ev_timer *w_, int revents);
+ecb_noinline static void stat_timer_cb (EV_P_ ev_timer *w_, int revents);
 
 #if EV_USE_INOTIFY
 
 /* the * 2 is to allow for alignment padding, which for some reason is >> 8 */
 # define EV_INOTIFY_BUFSIZE (sizeof (struct inotify_event) * 2 + NAME_MAX)
 
-static void noinline
+ecb_noinline
+static void
 infy_add (EV_P_ ev_stat *w)
 {
-  w->wd = inotify_add_watch (fs_fd, w->path, IN_ATTRIB | IN_DELETE_SELF | IN_MOVE_SELF | IN_MODIFY | IN_DONT_FOLLOW | IN_MASK_ADD);
+  w->wd = inotify_add_watch (fs_fd, w->path,
+                             IN_ATTRIB | IN_DELETE_SELF | IN_MOVE_SELF | IN_MODIFY
+                             | IN_CREATE | IN_DELETE | IN_MOVED_FROM | IN_MOVED_TO
+                             | IN_DONT_FOLLOW | IN_MASK_ADD);
 
   if (w->wd >= 0)
     {
@@ -3046,10 +4813,16 @@ infy_add (EV_P_ ev_stat *w)
         w->timer.repeat = w->interval ? w->interval : DEF_STAT_INTERVAL;
       else if (!statfs (w->path, &sfs)
                && (sfs.f_type == 0x1373 /* devfs */
+                   || sfs.f_type == 0x4006 /* fat */
+                   || sfs.f_type == 0x4d44 /* msdos */
                    || sfs.f_type == 0xEF53 /* ext2/3 */
+                   || sfs.f_type == 0x72b6 /* jffs2 */
+                   || sfs.f_type == 0x858458f6 /* ramfs */
+                   || sfs.f_type == 0x5346544e /* ntfs */
                    || sfs.f_type == 0x3153464a /* jfs */
+                   || sfs.f_type == 0x9123683e /* btrfs */
                    || sfs.f_type == 0x52654973 /* reiser3 */
-                   || sfs.f_type == 0x01021994 /* tempfs */
+                   || sfs.f_type == 0x01021994 /* tmpfs */
                    || sfs.f_type == 0x58465342 /* xfs */))
         w->timer.repeat = 0.; /* filesystem is local, kernel new enough */
       else
@@ -3094,7 +4867,8 @@ infy_add (EV_P_ ev_stat *w)
   if (ev_is_active (&w->timer)) ev_unref (EV_A);
 }
 
-static void noinline
+ecb_noinline
+static void
 infy_del (EV_P_ ev_stat *w)
 {
   int slot;
@@ -3111,7 +4885,8 @@ infy_del (EV_P_ ev_stat *w)
   inotify_rm_watch (fs_fd, wd);
 }
 
-static void noinline
+ecb_noinline
+static void
 infy_wd (EV_P_ int slot, int wd, struct inotify_event *ev)
 {
   if (slot < 0)
@@ -3157,7 +4932,8 @@ infy_cb (EV_P_ ev_io *w, int revents)
     }
 }
 
-inline_size void
+inline_size ecb_cold
+void
 ev_check_2625 (EV_P)
 {
   /* kernels < 2.6.25 are borked
@@ -3172,7 +4948,7 @@ ev_check_2625 (EV_P)
 inline_size int
 infy_newfd (void)
 {
-#if defined (IN_CLOEXEC) && defined (IN_NONBLOCK)
+#if defined IN_CLOEXEC && defined IN_NONBLOCK
   int fd = inotify_init1 (IN_CLOEXEC | IN_NONBLOCK);
   if (fd >= 0)
     return fd;
@@ -3257,7 +5033,7 @@ infy_fork (EV_P)
 #endif
 
 void
-ev_stat_stat (EV_P_ ev_stat *w)
+ev_stat_stat (EV_P_ ev_stat *w) EV_NOEXCEPT
 {
   if (lstat (w->path, &w->attr) < 0)
     w->attr.st_nlink = 0;
@@ -3265,7 +5041,8 @@ ev_stat_stat (EV_P_ ev_stat *w)
     w->attr.st_nlink = 1;
 }
 
-static void noinline
+ecb_noinline
+static void
 stat_timer_cb (EV_P_ ev_timer *w_, int revents)
 {
   ev_stat *w = (ev_stat *)(((char *)w_) - offsetof (ev_stat, timer));
@@ -3306,9 +5083,9 @@ stat_timer_cb (EV_P_ ev_timer *w_, int revents)
 }
 
 void
-ev_stat_start (EV_P_ ev_stat *w)
+ev_stat_start (EV_P_ ev_stat *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   ev_stat_stat (EV_A_ w);
@@ -3337,10 +5114,10 @@ ev_stat_start (EV_P_ ev_stat *w)
 }
 
 void
-ev_stat_stop (EV_P_ ev_stat *w)
+ev_stat_stop (EV_P_ ev_stat *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3363,9 +5140,9 @@ ev_stat_stop (EV_P_ ev_stat *w)
 
 #if EV_IDLE_ENABLE
 void
-ev_idle_start (EV_P_ ev_idle *w)
+ev_idle_start (EV_P_ ev_idle *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   pri_adjust (EV_A_ (W)w);
@@ -3378,7 +5155,7 @@ ev_idle_start (EV_P_ ev_idle *w)
     ++idleall;
     ev_start (EV_A_ (W)w, active);
 
-    array_needsize (ev_idle *, idles [ABSPRI (w)], idlemax [ABSPRI (w)], active, EMPTY2);
+    array_needsize (ev_idle *, idles [ABSPRI (w)], idlemax [ABSPRI (w)], active, array_needsize_noinit);
     idles [ABSPRI (w)][active - 1] = w;
   }
 
@@ -3386,10 +5163,10 @@ ev_idle_start (EV_P_ ev_idle *w)
 }
 
 void
-ev_idle_stop (EV_P_ ev_idle *w)
+ev_idle_stop (EV_P_ ev_idle *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3410,25 +5187,25 @@ ev_idle_stop (EV_P_ ev_idle *w)
 
 #if EV_PREPARE_ENABLE
 void
-ev_prepare_start (EV_P_ ev_prepare *w)
+ev_prepare_start (EV_P_ ev_prepare *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
 
   ev_start (EV_A_ (W)w, ++preparecnt);
-  array_needsize (ev_prepare *, prepares, preparemax, preparecnt, EMPTY2);
+  array_needsize (ev_prepare *, prepares, preparemax, preparecnt, array_needsize_noinit);
   prepares [preparecnt - 1] = w;
 
   EV_FREQUENT_CHECK;
 }
 
 void
-ev_prepare_stop (EV_P_ ev_prepare *w)
+ev_prepare_stop (EV_P_ ev_prepare *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3448,25 +5225,25 @@ ev_prepare_stop (EV_P_ ev_prepare *w)
 
 #if EV_CHECK_ENABLE
 void
-ev_check_start (EV_P_ ev_check *w)
+ev_check_start (EV_P_ ev_check *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
 
   ev_start (EV_A_ (W)w, ++checkcnt);
-  array_needsize (ev_check *, checks, checkmax, checkcnt, EMPTY2);
+  array_needsize (ev_check *, checks, checkmax, checkcnt, array_needsize_noinit);
   checks [checkcnt - 1] = w;
 
   EV_FREQUENT_CHECK;
 }
 
 void
-ev_check_stop (EV_P_ ev_check *w)
+ev_check_stop (EV_P_ ev_check *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3485,8 +5262,9 @@ ev_check_stop (EV_P_ ev_check *w)
 #endif
 
 #if EV_EMBED_ENABLE
-void noinline
-ev_embed_sweep (EV_P_ ev_embed *w)
+ecb_noinline
+void
+ev_embed_sweep (EV_P_ ev_embed *w) EV_NOEXCEPT
 {
   ev_run (w->other, EVRUN_NOWAIT);
 }
@@ -3518,6 +5296,7 @@ embed_prepare_cb (EV_P_ ev_prepare *prepare, int revents)
   }
 }
 
+#if EV_FORK_ENABLE
 static void
 embed_fork_cb (EV_P_ ev_fork *fork_w, int revents)
 {
@@ -3534,6 +5313,7 @@ embed_fork_cb (EV_P_ ev_fork *fork_w, int revents)
 
   ev_embed_start (EV_A_ w);
 }
+#endif
 
 #if 0
 static void
@@ -3544,9 +5324,9 @@ embed_idle_cb (EV_P_ ev_idle *idle, int revents)
 #endif
 
 void
-ev_embed_start (EV_P_ ev_embed *w)
+ev_embed_start (EV_P_ ev_embed *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   {
@@ -3564,8 +5344,10 @@ ev_embed_start (EV_P_ ev_embed *w)
   ev_set_priority (&w->prepare, EV_MINPRI);
   ev_prepare_start (EV_A_ &w->prepare);
 
+#if EV_FORK_ENABLE
   ev_fork_init (&w->fork, embed_fork_cb);
   ev_fork_start (EV_A_ &w->fork);
+#endif
 
   /*ev_idle_init (&w->idle, e,bed_idle_cb);*/
 
@@ -3575,17 +5357,19 @@ ev_embed_start (EV_P_ ev_embed *w)
 }
 
 void
-ev_embed_stop (EV_P_ ev_embed *w)
+ev_embed_stop (EV_P_ ev_embed *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
 
   ev_io_stop      (EV_A_ &w->io);
   ev_prepare_stop (EV_A_ &w->prepare);
+#if EV_FORK_ENABLE
   ev_fork_stop    (EV_A_ &w->fork);
+#endif
 
   ev_stop (EV_A_ (W)w);
 
@@ -3595,25 +5379,25 @@ ev_embed_stop (EV_P_ ev_embed *w)
 
 #if EV_FORK_ENABLE
 void
-ev_fork_start (EV_P_ ev_fork *w)
+ev_fork_start (EV_P_ ev_fork *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
 
   ev_start (EV_A_ (W)w, ++forkcnt);
-  array_needsize (ev_fork *, forks, forkmax, forkcnt, EMPTY2);
+  array_needsize (ev_fork *, forks, forkmax, forkcnt, array_needsize_noinit);
   forks [forkcnt - 1] = w;
 
   EV_FREQUENT_CHECK;
 }
 
 void
-ev_fork_stop (EV_P_ ev_fork *w)
+ev_fork_stop (EV_P_ ev_fork *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3633,15 +5417,15 @@ ev_fork_stop (EV_P_ ev_fork *w)
 
 #if EV_CLEANUP_ENABLE
 void
-ev_cleanup_start (EV_P_ ev_cleanup *w)
+ev_cleanup_start (EV_P_ ev_cleanup *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
 
   ev_start (EV_A_ (W)w, ++cleanupcnt);
-  array_needsize (ev_cleanup *, cleanups, cleanupmax, cleanupcnt, EMPTY2);
+  array_needsize (ev_cleanup *, cleanups, cleanupmax, cleanupcnt, array_needsize_noinit);
   cleanups [cleanupcnt - 1] = w;
 
   /* cleanup watchers should never keep a refcount on the loop */
@@ -3650,10 +5434,10 @@ ev_cleanup_start (EV_P_ ev_cleanup *w)
 }
 
 void
-ev_cleanup_stop (EV_P_ ev_cleanup *w)
+ev_cleanup_stop (EV_P_ ev_cleanup *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3674,9 +5458,9 @@ ev_cleanup_stop (EV_P_ ev_cleanup *w)
 
 #if EV_ASYNC_ENABLE
 void
-ev_async_start (EV_P_ ev_async *w)
+ev_async_start (EV_P_ ev_async *w) EV_NOEXCEPT
 {
-  if (expect_false (ev_is_active (w)))
+  if (ecb_expect_false (ev_is_active (w)))
     return;
 
   w->sent = 0;
@@ -3686,17 +5470,17 @@ ev_async_start (EV_P_ ev_async *w)
   EV_FREQUENT_CHECK;
 
   ev_start (EV_A_ (W)w, ++asynccnt);
-  array_needsize (ev_async *, asyncs, asyncmax, asynccnt, EMPTY2);
+  array_needsize (ev_async *, asyncs, asyncmax, asynccnt, array_needsize_noinit);
   asyncs [asynccnt - 1] = w;
 
   EV_FREQUENT_CHECK;
 }
 
 void
-ev_async_stop (EV_P_ ev_async *w)
+ev_async_stop (EV_P_ ev_async *w) EV_NOEXCEPT
 {
   clear_pending (EV_A_ (W)w);
-  if (expect_false (!ev_is_active (w)))
+  if (ecb_expect_false (!ev_is_active (w)))
     return;
 
   EV_FREQUENT_CHECK;
@@ -3714,7 +5498,7 @@ ev_async_stop (EV_P_ ev_async *w)
 }
 
 void
-ev_async_send (EV_P_ ev_async *w)
+ev_async_send (EV_P_ ev_async *w) EV_NOEXCEPT
 {
   w->sent = 1;
   evpipe_write (EV_A_ &async_pending);
@@ -3761,16 +5545,10 @@ once_cb_to (EV_P_ ev_timer *w, int revents)
 }
 
 void
-ev_once (EV_P_ int fd, int events, ev_tstamp timeout, void (*cb)(int revents, void *arg), void *arg)
+ev_once (EV_P_ int fd, int events, ev_tstamp timeout, void (*cb)(int revents, void *arg), void *arg) EV_NOEXCEPT
 {
   struct ev_once *once = (struct ev_once *)ev_malloc (sizeof (struct ev_once));
 
-  if (expect_false (!once))
-    {
-      cb (EV_ERROR | EV_READ | EV_WRITE | EV_TIMER, arg);
-      return;
-    }
-
   once->cb  = cb;
   once->arg = arg;
 
@@ -3792,8 +5570,9 @@ ev_once (EV_P_ int fd, int events, ev_tstamp timeout, void (*cb)(int revents, vo
 /*****************************************************************************/
 
 #if EV_WALK_ENABLE
+ecb_cold
 void
-ev_walk (EV_P_ int types, void (*cb)(EV_P_ int type, void *w))
+ev_walk (EV_P_ int types, void (*cb)(EV_P_ int type, void *w)) EV_NOEXCEPT
 {
   int i, j;
   ev_watcher_list *wl, *wn;
@@ -3846,7 +5625,7 @@ ev_walk (EV_P_ int types, void (*cb)(EV_P_ int type, void *w))
 
 #if EV_IDLE_ENABLE
   if (types & EV_IDLE)
-    for (j = NUMPRI; i--; )
+    for (j = NUMPRI; j--; )
       for (i = idlecnt [j]; i--; )
         cb (EV_A_ EV_IDLE, idles [j][i]);
 #endif
@@ -3908,6 +5687,3 @@ ev_walk (EV_P_ int types, void (*cb)(EV_P_ int type, void *w))
 #if EV_MULTIPLICITY
   #include "ev_wrap.h"
 #endif
-
-EV_CPP(})
-