diff --git a/.gitignore b/.gitignore
index 052acb3c..a4f2d6f5 100644
--- a/.gitignore
+++ b/.gitignore
@@ -3,6 +3,9 @@
 /bin/
 /lib/
 
+# Cross-compilation output (tools/crossBuild.sh: musl + Windows artifacts).
+/build/
+
 # Vendored-engine build output: mruby's Rake build and libssh2's CMake build emit whole
 # trees (generated .c/.h, the mrbc tool, CMake cache) under their own build/ dirs.
 /vendor/mruby/build/
diff --git a/LICENSE.md b/LICENSE.md
index 7330ac72..760b54a8 100644
--- a/LICENSE.md
+++ b/LICENSE.md
@@ -114,3 +114,14 @@ All of the following are linked into the default build and into `bin/calog`.
 > permissive binary can build without the MySQL backend -- drop
 > `-DCALOG_WITH_MYSQL` and the `$(MYSQLARCH)` archive from the `bin/calog` link in
 > the Makefile -- and keep SQLite + PostgreSQL.
+
+### Platform support (Windows builds only)
+
+Linked only into **Windows** builds; not present in the Linux or macOS builds (those
+use the native pthreads).
+
+- **winpthreads** (from mingw-w64) -- MIT License. Copyright (c) 2011 mingw-w64
+  project; parts derived from the Pthreads-win32 (POSIX Threads for Windows) library.
+  A POSIX threads implementation over the Win32 API, so calog's pthread code compiles
+  on Windows. Vendored from source (see `vendor/winpthreads/NOTICE.calog`); license in
+  `vendor/winpthreads/COPYING`.
diff --git a/Makefile b/Makefile
index 75c170af..6b3bb27e 100644
--- a/Makefile
+++ b/Makefile
@@ -50,21 +50,44 @@ LUAINC    = -I$(LUADIR)/src
 # is Unix-only anyway -- pthreads, sanitizers, setarch -- so the Windows branch just
 # keeps the Lua define correct, it does not by itself make a Windows build work.)
 ifeq ($(OS),Windows_NT)
+  PLATFORM    = windows
   LUAPLAT     =
   LUAPLATLIBS =
+  DLLIB       =
+  SOCKETLIBS  = -lws2_32 -lwinmm
 else
   UNAMES := $(shell uname -s)
   ifeq ($(UNAMES),Linux)
+    PLATFORM    = linux
     LUAPLAT     = -DLUA_USE_LINUX
     LUAPLATLIBS = -ldl
+    DLLIB       = -ldl
+    SOCKETLIBS  =
   else ifeq ($(UNAMES),Darwin)
+    PLATFORM    = darwin
     LUAPLAT     = -DLUA_USE_MACOSX
     LUAPLATLIBS =
+    DLLIB       =
+    SOCKETLIBS  =
   else
+    PLATFORM    = posix
     LUAPLAT     = -DLUA_USE_POSIX
     LUAPLATLIBS =
+    DLLIB       = -ldl
+    SOCKETLIBS  =
   endif
 endif
+
+# Extra link flags for the `release` target (empty for the default dev build): the C++ runtime
+# and libgcc linked statically so the CLI's only shared-library dependency is the system libc.
+# CXXLIB is how the explicit C++ runtime is pulled in; release wraps it in -Bstatic/-Bdynamic so
+# the STATIC libstdc++.a is used (a driver -static-libstdc++ does not override an explicit -lstdc++).
+RELEASELD =
+CXXLIB    = -lstdc++
+# macOS ships libc++ (LLVM), not libstdc++; the Squirrel C++ engine links against it there.
+ifeq ($(PLATFORM),darwin)
+  CXXLIB  = -lc++
+endif
 LUAFLAGS  = -std=c99 -w -g -O2 $(LUAPLAT)
 LUASRC    = $(filter-out $(LUADIR)/src/lua.c $(LUADIR)/src/luac.c, $(wildcard $(LUADIR)/src/*.c))
 LUAOBJ    = $(patsubst $(LUADIR)/src/%.c,obj/%.o,$(LUASRC))
@@ -108,7 +131,7 @@ S7DIR     = vendor/s7
 S7INC     = -I$(S7DIR)
 S7FLAGS   = -std=c99 -w -g -O1 -D_GNU_SOURCE
 S7OBJ     = obj/s7.o
-S7LIBS    = -ldl -lm
+S7LIBS    = $(DLLIB) -lm
 
 # --- vendored Wren (C, amalgamated single file wren.c + wren.h; regenerate via
 # util/generate_amalgamation.py from upstream). Links -lm. ---
@@ -144,7 +167,7 @@ $(MRUBYLIB): $(MRUBYCONF)
 TCLDIR   = vendor/tcl
 TCLINC   = -I$(TCLDIR)/generic -I$(TCLDIR)/unix
 TCLLIB   = $(TCLDIR)/build/libtcl9.0.a
-TCLLIBS  = -ldl -lz -lpthread -lm
+TCLLIBS  = $(DLLIB) -lz -lpthread -lm
 
 $(TCLLIB):
 	mkdir -p $(TCLDIR)/build && cd $(TCLDIR)/build && ../unix/configure --disable-shared >/dev/null && $(MAKE) libtcl9.0.a
@@ -161,9 +184,28 @@ SQLITELIBS  = -lm -lpthread
 ENETDIR   = vendor/enet
 ENETINC   = -I$(ENETDIR)/include
 ENETFLAGS = -std=c11 -w -g -O2 -D_GNU_SOURCE -DHAS_FCNTL=1 -DHAS_POLL=1 -DHAS_GETADDRINFO=1 -DHAS_GETNAMEINFO=1 -DHAS_GETHOSTBYNAME_R=1 -DHAS_GETHOSTBYADDR_R=1 -DHAS_INET_PTON=1 -DHAS_INET_NTOP=1 -DHAS_MSGHDR_FLAGS=1 -DHAS_SOCKLEN_T=1
-ENETNAMES = callbacks compress host list packet peer protocol unix
+# ENet's platform backend: unix.c on Linux/macOS, win32.c on Windows (which also needs -lwinmm).
+ENETBACKEND = unix
+ifeq ($(PLATFORM),windows)
+  ENETBACKEND = win32
+endif
+ENETNAMES = callbacks compress host list packet peer protocol $(ENETBACKEND)
 ENETOBJ   = $(foreach n,$(ENETNAMES),obj/enet_$(n).o)
 
+# --- vendored winpthreads (WINDOWS ONLY): POSIX threads over the Win32 API, so calog's pthread
+# code (actor model, timer, pubsub, ...) builds on Windows. Built from source like every
+# dependency; unused on Linux/macOS, which have native pthreads. A Windows build sets CC to a
+# mingw-w64 compiler (or uses tools/crossBuild.sh with zig). See vendor/winpthreads/NOTICE.calog. ---
+WPTHDIR   = vendor/winpthreads
+WPTHINC   = -I$(WPTHDIR)/include -DWINPTHREAD_STATIC=1
+WPTHLIB   = lib/libwinpthreads.a
+WPTHSRC   = $(wildcard $(WPTHDIR)/src/*.c)
+WPTHOBJ   = $(patsubst $(WPTHDIR)/src/%.c,obj/wpth_%.o,$(WPTHSRC))
+obj/wpth_%.o: $(WPTHDIR)/src/%.c | obj
+	$(CC) -O2 -w -I$(WPTHDIR)/include -I$(WPTHDIR)/src -DWINPTHREAD_STATIC=1 -DIN_WINPTHREAD=1 -c -o $@ $<
+$(WPTHLIB): $(WPTHOBJ) | lib
+	$(AR) rcs $@ $(WPTHOBJ)
+
 BINS = bin/testBroker bin/testLua bin/testMyBasic bin/testPolyglot bin/testActor \
        bin/testEngineLua bin/testEngineMyBasic bin/testSquirrel bin/testEngineSquirrel bin/testJs bin/testEngineJs \
        bin/testEngineBerry bin/testEngineS7 bin/testEngineWren bin/testEngineMruby bin/testEngineTcl bin/testLoad bin/testDb bin/testNet bin/testTask bin/testExport bin/testJson bin/testTime bin/testFs bin/testCrypto bin/testKv bin/testTimer bin/testPubsub bin/testHttp bin/testHttps bin/embed bin/calog
@@ -398,10 +440,10 @@ bin/testPolyglot: obj/testPolyglot.o lib/libcalog.a lib/liblua.a lib/libmybasic.
 	$(CC) $(LDFLAGS) -o $@ $^ $(LUALIBS)
 
 bin/testSquirrel: obj/testSquirrel.o lib/libcalog.a lib/libsquirrel.a | bin
-	$(CC) $(LDFLAGS) -o $@ $^ -lstdc++ -lm
+	$(CC) $(LDFLAGS) -o $@ $^ $(CXXLIB) -lm
 
 bin/testEngineSquirrel: obj/testEngineSquirrel.o lib/libcalog.a lib/libsquirrel.a | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ -lstdc++ -lm
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(CXXLIB) -lm
 
 bin/testJs: obj/testJs.o lib/libcalog.a lib/libquickjs.a | bin
 	$(CC) $(LDFLAGS) -o $@ $^ -lm
@@ -441,7 +483,19 @@ bin/calog: obj/calogMain.o \
            obj/calogCrypto.o obj/calogDbFull.o obj/calogExport.o obj/calogFs.o obj/calogHttp.o obj/calogJson.o obj/calogKv.o obj/calogNet.o obj/calogPubsub.o obj/calogSsh.o obj/calogTask.o obj/calogTime.o obj/calogTimer.o obj/calogHandle.o \
            lib/libcalog.a lib/liblua.a lib/libquickjs.a lib/libsquirrel.a lib/libmybasic.a lib/libberry.a lib/libs7.a lib/libwren.a $(MRUBYLIB) $(TCLLIB) \
            lib/libsqlite3.a lib/libenet.a $(LIBSSH2LIB) $(DBARCH) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $(filter-out $(DBARCH),$^) -Wl,--start-group $(PGARCHIVES) -Wl,--end-group $(MYSQLARCH) $(SSLARCH) -lstdc++ -ldl -lm -lpthread
+	$(CC) $(LDFLAGS) $(RELEASELD) -pthread -o $@ $(filter-out $(DBARCH),$^) -Wl,--start-group $(PGARCHIVES) -Wl,--end-group $(MYSQLARCH) $(SSLARCH) $(CXXLIB) $(DLLIB) -lm -lpthread $(SOCKETLIBS)
+
+# ---- release build (fewest system dependencies while keeping DNS) --------------------
+# The full CLI with sanitizers off and the C++/gcc runtimes linked statically, so `ldd` collapses
+# to just the system libc (glibc stays dynamic, so getaddrinfo/DNS keeps working -- see PORTING.md).
+# A `clean` first keeps build modes from mixing; the slow vendored archives (OpenSSL/Tcl/mruby/
+# libssh2) live under vendor/ and survive it, so this is a moderate rebuild, not a full one.
+release:
+	$(MAKE) clean
+	$(MAKE) SAN= RELEASELD='-static-libgcc' CXXLIB='-Wl,-Bstatic -lstdc++ -Wl,-Bdynamic' bin/calog
+	@echo "== release bin/calog shared-library dependencies (want just libc/libm + loader) =="
+	@ldd bin/calog 2>&1 | sed 's/^/  /'
+	@echo "== note: run 'make clean && make' to return to the sanitized dev build =="
 
 # ---- fully-static build --------------------------------------------------------------
 # A self-contained calog executable with NO shared-library dependencies at runtime. The
@@ -479,15 +533,15 @@ obj/testLoad.o: tests/testLoad.c src/calog.h | obj
 	$(CC) $(COREFLAGS) -Isrc -pthread $(ENGINEDEFS) -c -o $@ $<
 
 bin/testLoad: obj/testLoad.o lib/libcalog.a lib/liblua.a lib/libquickjs.a lib/libsquirrel.a lib/libmybasic.a lib/libberry.a lib/libs7.a lib/libwren.a $(MRUBYLIB) $(TCLLIB) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -lstdc++ -lm $(MRUBYLIBS) $(TCLLIBS)
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(CXXLIB) -lm $(MRUBYLIBS) $(TCLLIBS)
 
 # task library test: the task lib names every engine, so (like testLoad) it links them all.
 bin/testTask: obj/testTask.o $(TASKADP) obj/calogHandle.o lib/libcalog.a lib/liblua.a lib/libquickjs.a lib/libsquirrel.a lib/libmybasic.a lib/libberry.a lib/libs7.a lib/libwren.a $(MRUBYLIB) $(TCLLIB) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -lstdc++ -lm $(MRUBYLIBS) $(TCLLIBS)
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(CXXLIB) -lm $(MRUBYLIBS) $(TCLLIBS)
 
 # export library test: publish from Lua, call by bare name from Lua + via callExport from JS.
 bin/testExport: obj/testExport.o obj/calogExport.o lib/libcalog.a lib/liblua.a lib/libquickjs.a lib/libsquirrel.a lib/libs7.a lib/libmybasic.a | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -lstdc++ -ldl -lm
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(CXXLIB) $(DLLIB) -lm
 
 bin/testJson: obj/testJson.o obj/calogJson.o lib/libcalog.a lib/liblua.a | bin
 	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS)
@@ -499,7 +553,7 @@ bin/testFs: obj/testFs.o obj/calogFs.o lib/libcalog.a lib/liblua.a | bin
 	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS)
 
 bin/testCrypto: obj/testCrypto.o obj/calogCrypto.o lib/libcalog.a lib/liblua.a $(SSLARCH) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -ldl
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(DLLIB)
 
 bin/testKv: obj/testKv.o obj/calogKv.o lib/libcalog.a lib/liblua.a | bin
 	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS)
@@ -511,14 +565,14 @@ bin/testPubsub: obj/testPubsub.o obj/calogPubsub.o lib/libcalog.a lib/liblua.a |
 	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS)
 
 bin/testHttp: obj/testHttp.o obj/calogHttp.o lib/libcalog.a lib/liblua.a $(SSLARCH) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -ldl
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(DLLIB)
 
 bin/testHttps: obj/testHttps.o obj/calogHttp.o lib/libcalog.a lib/liblua.a $(SSLARCH) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -ldl
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(DLLIB)
 
 # SSH/SFTP over vendored libssh2 (before OpenSSL in the link line: libssh2 depends on it).
 bin/testSsh: obj/testSsh.o obj/calogSsh.o obj/calogHandle.o lib/libcalog.a lib/liblua.a $(LIBSSH2LIB) $(SSLARCH) | bin
-	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) -ldl
+	$(CC) $(LDFLAGS) -pthread -o $@ $^ $(LUALIBS) $(DLLIB)
 
 # The SSH test spawns a throwaway unprivileged sshd (needs system sshd + ssh-keygen), so it is
 # NOT part of `make test`. Run it explicitly.
@@ -559,7 +613,7 @@ tsansq: $(SQSRC) | bin obj
 	$(CC) $(TSANFLAGS) $(INC) $(SQDEF) $(SQINC) -o bin/testEngineSquirrelTsan \
 		tests/testEngineSquirrel.c src/squirrel/squirrelEngine.c src/squirrel/squirrelAdapter.c \
 		$(TSANCORE) \
-		$(patsubst $(SQDIR)/squirrel/%.cpp,obj/%.tsan.o,$(SQSRC)) -lstdc++ -lm
+		$(patsubst $(SQDIR)/squirrel/%.cpp,obj/%.tsan.o,$(SQSRC)) $(CXXLIB) -lm
 	setarch -R ./bin/testEngineSquirrelTsan
 	rm -f $(patsubst $(SQDIR)/squirrel/%.cpp,obj/%.tsan.o,$(SQSRC))
 
@@ -651,4 +705,4 @@ clean:
 -include $(wildcard obj/*.d)
 -include $(wildcard obj/rel/*.d)
 
-.PHONY: all test tsan tsansq tsanjs tsanmb tsanberry tsans7 tsanwren tsanmruby tsantcl tsanlibs clean
+.PHONY: all test tsan tsansq tsanjs tsanmb tsanberry tsans7 tsanwren tsanmruby tsantcl tsanlibs release clean
diff --git a/PORTING.md b/PORTING.md
new file mode 100644
index 00000000..28bb71cd
--- /dev/null
+++ b/PORTING.md
@@ -0,0 +1,149 @@
+# Porting calog
+
+calog targets **Linux, macOS, and Windows**. Its core -- values, broker, the actor/threading
+model, and every engine adapter -- is portable C11. The only operating-system-specific surface is
+small and centralized:
+
+- **Threads**: POSIX `pthreads`. Native on Linux and macOS; on Windows via mingw-w64's `winpthreads`.
+- **Sockets + DNS**: BSD sockets + `getaddrinfo`, used only by the net/ssh/http libraries. BSD
+  sockets on Linux/macOS, Winsock on Windows. All of this is behind `src/calogPlatform.h`.
+
+There is no `epoll`/`eventfd`/`kqueue`, no `fork`/`exec`, and no `dlopen` in calog's own code (the
+actor model uses blocking sockets on per-context threads), so a port is a thin shim, not a rewrite.
+
+## The platform abstraction: `src/calogPlatform.h`
+
+The net/ssh/http libraries use these instead of raw socket calls, so the same source compiles on all
+three platforms:
+
+| Abstraction | POSIX (Linux/macOS) | Windows |
+|---|---|---|
+| `CalogSocketT` | `int` | `SOCKET` |
+| `CALOG_INVALID_SOCKET` | `-1` | `INVALID_SOCKET` |
+| `calogSockClose(fd)` | `close` | `closesocket` |
+| `calogSockErrno()` / `calogSockErrStr()` | `errno` / `strerror` | `WSAGetLastError` / `FormatMessage` |
+| `calogSockSetNonblock(fd, on)` | `fcntl(O_NONBLOCK)` | `ioctlsocket(FIONBIO)` |
+| `calogSockInProgress(err)` | `EINPROGRESS` | `WSAEWOULDBLOCK`/`WSAEINPROGRESS` |
+| `calogPoll(...)` | `poll` | `WSAPoll` |
+| `calogPlatformNetInit()` / `Shutdown()` | no-op | `WSAStartup` / `WSACleanup` |
+| `CALOG_MSG_NOSIGNAL` | `MSG_NOSIGNAL` | `0` (Windows has no `SIGPIPE`) |
+
+A socket `fd` is unsigned on Windows, so the code compares against `CALOG_INVALID_SOCKET` (never
+`fd < 0`). The abstraction is behavior-identical on POSIX (it maps to the same names), so the Linux
+and macOS builds are unchanged by it.
+
+## Build matrix
+
+| Target | Command | Runtime dependencies | DNS |
+|---|---|---|---|
+| Linux dev | `make` | glibc, libstdc++, libgcc, ASan/UBSan (dev only) | yes |
+| Linux release | `make release` | **glibc only** (libstdc++/libgcc linked static) | yes |
+| Linux static | `make static` on Alpine/musl | **none** (fully static) | yes (musl resolver) |
+| Linux static | `make static` on glibc | none, but see caveat | **no** (glibc NSS needs dlopen) |
+| macOS | `make` (Darwin auto-detected) | libSystem (always present) | yes |
+| Windows | mingw-w64 build | core OS DLLs (`kernel32`, `ws2_32`) | yes |
+
+The Makefile auto-detects the platform (`PLATFORM` = `linux`/`darwin`/`windows`/`posix`) and adjusts
+the link line: `DLLIB` is `-ldl` on Linux (empty on macOS, where `dlopen` is in libSystem) and
+`SOCKETLIBS` is `-lws2_32` on Windows.
+
+## Linux
+
+- **`make`** -- the sanitized development build (ASan + UBSan + strict warnings).
+- **`make release`** -- the distribution build: sanitizers off, `-static-libstdc++ -static-libgcc`,
+  glibc dynamic. `ldd` collapses to just `libc`/`libm` + the loader. Because glibc stays dynamic,
+  `getaddrinfo` can still `dlopen` its NSS modules, so **DNS works**. Runs on any mainstream
+  glibc distro with nothing to install.
+- **`make static` on Alpine (musl)** -- a fully static binary with **zero** runtime dependencies and
+  **working DNS** (musl has a built-in resolver, no NSS/dlopen). This is the minimal-dependency
+  artifact. Because calog builds every dependency from source, and Alpine's default toolchain is
+  musl, no per-dependency changes are needed -- build it in an Alpine container.
+- **`make static` on glibc** -- fully static, but name resolution and `dlopen`-based Lua C modules
+  do not work (glibc NSS requires runtime `dlopen`). Connect by IP, or use the musl route above.
+
+## macOS (Darwin)
+
+The C is already POSIX-compatible (pthreads, BSD sockets, `getaddrinfo` are all native), so the port
+is build configuration, not code -- confirmed by `tools/crossBuild.sh`, which links valid Mach-O
+executables for **both x86_64 (Intel) and arm64 (Apple Silicon)** using zig's bundled `libSystem`
+stub (calog only needs `libSystem` -- libc/pthreads/sockets -- so no Apple SDK/frameworks are
+required for the core). Apple does not support statically linking libc, but `libSystem` is part of
+the OS and always present, so "static everything except libSystem" is the natural minimal-deps build.
+Each vendored dependency builds its own way and needs a macOS configuration:
+
+- OpenSSL: `./Configure darwin64-<arch>-cc`
+- Tcl: its `configure` detects Darwin
+- mruby / Lua / SQLite / ENet / QuickJS / Squirrel / Berry / s7 / Wren: compile as-is with clang
+- libssh2 / MariaDB: CMake with the macOS toolchain
+
+Link against dynamic `libSystem` (the default); `-ldl` is dropped automatically (`DLLIB` is empty on
+Darwin) and the C++ runtime is `libc++` (`CXXLIB = -lc++`). SIGPIPE is ignored at CLI startup, since
+macOS `send()` has no `MSG_NOSIGNAL`. ENet's `unix.c` backend is correct on macOS (BSD sockets).
+
+## Windows (mingw-w64)
+
+calog's pthread code compiles unchanged on Windows against **winpthreads** (POSIX threads over the
+Win32 API), which is **vendored** at `vendor/winpthreads/` and built from source like every other
+dependency (see its `NOTICE.calog`). `src/calogPlatform.h` handles the Winsock differences. Building
+uses a **mingw-w64** toolchain (not MSVC) for its POSIX-ish environment; `tools/crossBuild.sh` cross-
+compiles from Linux with a single zig toolchain and is the tested path. Remaining work is build
+configuration:
+
+- Build each vendored dependency for mingw (OpenSSL `mingw64`, Tcl's `win/` makefile or configure,
+  libssh2/MariaDB via CMake with the mingw toolchain, mruby with a mingw build_config).
+- ENet selects its `win32.c` backend automatically (`ENETBACKEND`); the link adds `-lws2_32 -lwinmm`
+  (handled by `SOCKETLIBS`).
+- Static-link the runtime (`-static -static-libgcc -static-libstdc++`) so the binary depends only on
+  core Windows DLLs, which are always present. DNS works through Winsock's `getaddrinfo`.
+
+Remaining code review for Windows: `calogFs.c` and `calogHttp.c` use POSIX headers (`dirent.h`,
+`sys/stat.h`, `sys/time.h`) that mingw-w64 mostly provides, but path handling (`/` vs `\`) and a few
+calls may need adjustment; `calogSsh.c` already guards `sys/select.h` (Winsock supplies
+`fd_set`/`select`), and `httpSetTimeouts` already branches on the Winsock `SO_RCVTIMEO` DWORD form.
+
+## Cross-compiling from Linux with zig (`tools/crossBuild.sh`)
+
+A single [zig](https://ziglang.org/download/) toolchain cross-compiles C and C++ to musl-static and
+Windows without installing per-target toolchains, so the ports can be exercised on a plain Linux box:
+
+```
+ZIG=/path/to/zig ./tools/crossBuild.sh
+```
+
+It builds a representative set -- core + one engine of each kind (Lua, QuickJS, Squirrel/C++,
+my-basic) + the socket library (`calogNet`) -- because the crypto/db/http/ssh libraries pull in
+OpenSSL/MariaDB/PostgreSQL/libssh2, each needing its own per-target build. What it does:
+
+- **musl**: fully static Linux binaries. These **run on the build host**, so the script builds AND
+  runs them -- verifying the core, all four engine kinds, C++, and the socket abstraction on musl.
+- **Windows**: builds `vendor/winpthreads` into `libwinpthreads.a`, then links real `.exe`s
+  (including the Winsock + ENet-win32 socket path). Verified as valid PE executables; running them
+  needs wine.
+
+Verified results (this repo): musl -- Lua, QuickJS, my-basic, Squirrel, and `testNet` all build and
+**pass** (fully static); Windows -- `testEngineLua.exe` and `testNet.exe` build against vendored
+winpthreads. Since the Windows binaries are the *same source* that passes on musl, the build is strong
+evidence; full run-verification needs a Windows host or wine.
+
+## Status
+
+- **Platform abstraction** (`calogPlatform.h`) and the net/ssh/http conversion: **done**, verified
+  behavior-identical on Linux (`make test` 30/30, socket + HTTP loopback smoke, ASan- and TSan-clean).
+- **Linux release** target: **done and verified** -- `ldd bin/calog` shows only `libc`/`libm` + loader.
+- **Linux musl-static**: **run-verified** via `tools/crossBuild.sh` (zig) -- core + Lua + QuickJS +
+  my-basic + Squirrel + `testNet` all build fully static and pass on the host. Also works via
+  `make static` in an Alpine container (musl default toolchain).
+- **Windows**: **build-verified** via `tools/crossBuild.sh` -- `testEngineLua.exe` and `testNet.exe`
+  (core + engine + Winsock socket abstraction) link against vendored `winpthreads` into real PE
+  executables. Run-verification needs a Windows host or wine; the binaries are the same source that
+  passes on musl.
+- **Build system**: OS-aware (`PLATFORM`, `DLLIB`, `CXXLIB`, `SOCKETLIBS`, `ENETBACKEND`, vendored
+  `winpthreads`); the Linux values are identical to before, so the dev build is unchanged.
+- **macOS**: **build-verified** via `tools/crossBuild.sh` -- core + Lua + `testNet` link into valid
+  Mach-O executables for both **x86_64 and arm64** (zig's libSystem stub; no Apple SDK needed for
+  calog's libc surface). Run-verification needs a Mac. The C code and Makefile hooks are all in place
+  (`libc++`, no `-ldl`, SIGPIPE, native pthreads + BSD sockets).
+- The heavy vendored libraries (OpenSSL, Tcl, mruby, libssh2, MariaDB, PostgreSQL) still need a
+  per-target build to reach a *full-featured* calog on macOS/Windows; the cross-build proves the core
+  broker + engines + socket layer, which is where all the portability risk lived. No code rewrite is
+  expected -- the OS surface is only threads + sockets + DNS, all abstracted.
diff --git a/README.md b/README.md
index 6838918a..db6de50e 100644
--- a/README.md
+++ b/README.md
@@ -39,6 +39,16 @@ Swap `&calogJsEngine` for `&calogLuaEngine`, `&calogSquirrelEngine`,
 - **Callbacks both ways.** A script can hand you a function value to keep and call later
   (routed back to the engine that owns it); and you can hand a native to a script as a
   callable value with `calogFnFromNative`.
+- **Scripts export to each other, across languages.** A script publishes a function by name
+  with `calogExport`; any *other* script -- in any language -- calls it with `calogCall` (or
+  by bare name on the hook engines). A Lua helper is callable from Ruby, a Tcl proc from
+  JavaScript. Functions cross the language boundary, not just the host boundary, so you can
+  write each part of a system in whatever language fits it best and still share code.
+- **Hot reload -- patch a running system.** A live context is a *persistent* interpreter:
+  feed it new code at any time (`taskEval`, or `calogContextEval` from the host) to redefine a
+  function, fix a bug, or add a feature while the process keeps running and keeps its state --
+  no restart. Re-export the fixed function and every caller, in every language, picks up the
+  new version on the next call.
 - **Many runtimes.** Independent `CalogT` runtimes coexist in one process; one host
   thread can drive several.
 - **Load by filename.** `calogContextLoad(calog, "config")` finds `config.lua` /
@@ -47,6 +57,11 @@ Swap `&calogJsEngine` for `&calogLuaEngine`, `&calogSquirrelEngine`,
   packages). The core is warning-clean on gcc **and** clang (`-Wall -Wextra -Werror
   -Wconversion -Wsign-conversion`) and runs clean under AddressSanitizer, UBSan, and
   ThreadSanitizer.
+- **Portable, few dependencies.** Linux, macOS, and Windows (mingw-w64). The OS surface is
+  just threads + sockets + DNS, behind one abstraction (`src/calogPlatform.h`); everything
+  else is portable C11. On Linux, `make release` yields a binary whose only shared-library
+  dependency is the system libc, and `make static` on Alpine a fully static one. See
+  [PORTING.md](PORTING.md).
 
 ## Engines
 
diff --git a/libs/calogHttp.c b/libs/calogHttp.c
index be486a3f..dc852931 100644
--- a/libs/calogHttp.c
+++ b/libs/calogHttp.c
@@ -11,21 +11,15 @@
 
 #include "calogHttp.h"
 #include "calogInternal.h"
+#include "calogPlatform.h"
 
-#include <errno.h>
-#include <fcntl.h>
 #include <limits.h>
-#include <netdb.h>
-#include <netinet/in.h>
-#include <poll.h>
 #include <signal.h>
 #include <stdint.h>
 #include <stdlib.h>
 #include <string.h>
 #include <strings.h>
-#include <sys/socket.h>
 #include <sys/time.h>
-#include <unistd.h>
 
 #include <openssl/ssl.h>
 #include <openssl/x509v3.h>
@@ -69,10 +63,10 @@ typedef struct HttpUrlT {
 // requests certificate + hostname authentication for an https connection (default; an https
 // request can opt out with insecure=true).
 typedef struct HttpConnT {
-    int      fd;
-    SSL     *ssl;
-    SSL_CTX *ctx;
-    bool     verify;
+    CalogSocketT fd;
+    SSL         *ssl;
+    SSL_CTX     *ctx;
+    bool         verify;
 } HttpConnT;
 
 static CalogValueT *httpAggField(CalogAggT *agg, const char *name);
@@ -83,7 +77,7 @@ static int32_t httpBufPutBytes(HttpBufT *buf, const char *bytes, size_t length);
 static int32_t httpBufPutStr(HttpBufT *buf, const char *s);
 static int32_t httpBuildResult(const char *raw, size_t rawLen, CalogValueT *result);
 static void    httpConnClose(HttpConnT *conn);
-static bool    httpConnectTimeout(int fd, const struct sockaddr *addr, socklen_t addrLen, int timeoutSec);
+static bool    httpConnectTimeout(CalogSocketT fd, const struct sockaddr *addr, socklen_t addrLen, int timeoutSec);
 static int32_t httpConnOpen(HttpConnT *conn, const HttpUrlT *url, CalogValueT *result);
 static ssize_t httpConnRead(HttpConnT *conn, char *buf, size_t length);
 static int32_t httpConnWriteAll(HttpConnT *conn, const char *bytes, size_t length);
@@ -112,7 +106,7 @@ static bool    httpResolveLocation(const HttpUrlT *base, const char *loc, char *
 static const CalogValueT *httpResultHeader(const CalogValueT *result, const char *name);
 static int32_t httpResultStatus(const CalogValueT *result);
 static bool    httpSameOrigin(const HttpUrlT *a, const HttpUrlT *b);
-static void    httpSetTimeouts(int fd, int timeoutSec);
+static void    httpSetTimeouts(CalogSocketT fd, int timeoutSec);
 static int32_t httpTlsHandshake(HttpConnT *conn, const HttpUrlT *url, CalogValueT *result);
 static int     httpWriteAuthority(char *out, size_t cap, const char *scheme, const HttpUrlT *base);
 
@@ -419,46 +413,46 @@ static void httpConnClose(HttpConnT *conn) {
         conn->ctx = NULL;
     }
     // SSL_set_fd installs a BIO_NOCLOSE socket BIO, so SSL_free never closes the fd -- we own it.
-    if (conn->fd >= 0) {
-        close(conn->fd);
-        conn->fd = -1;
+    if (conn->fd != CALOG_INVALID_SOCKET) {
+        calogSockClose(conn->fd);
+        conn->fd = CALOG_INVALID_SOCKET;
     }
 }
 
 
-static bool httpConnectTimeout(int fd, const struct sockaddr *addr, socklen_t addrLen, int timeoutSec) {
-    int           flags;
+static bool httpConnectTimeout(CalogSocketT fd, const struct sockaddr *addr, socklen_t addrLen, int timeoutSec) {
     int           rc;
     int           err;
     socklen_t     errLen;
     struct pollfd pfd;
 
-    flags = fcntl(fd, F_GETFL, 0);
-    if (flags < 0 || fcntl(fd, F_SETFL, flags | O_NONBLOCK) < 0) {
+    // A fresh socket starts blocking; drive a bounded connect by flipping it non-blocking, polling
+    // for writability, then restoring blocking mode (portable across BSD sockets and Winsock).
+    if (calogSockSetNonblock(fd, 1) != 0) {
         return connect(fd, addr, addrLen) == 0;   // best effort: no bounded-connect fallback
     }
     rc = connect(fd, addr, addrLen);
     if (rc == 0) {
-        fcntl(fd, F_SETFL, flags);
+        calogSockSetNonblock(fd, 0);
         return true;
     }
-    if (errno != EINPROGRESS) {
-        fcntl(fd, F_SETFL, flags);
+    if (!calogSockInProgress(calogSockErrno())) {
+        calogSockSetNonblock(fd, 0);
         return false;
     }
     pfd.fd     = fd;
     pfd.events = POLLOUT;
-    rc = poll(&pfd, 1, timeoutSec * 1000);
+    rc = calogPoll(&pfd, 1, timeoutSec * 1000);
     if (rc <= 0) {
-        fcntl(fd, F_SETFL, flags);
+        calogSockSetNonblock(fd, 0);
         return false;   // timeout or poll error
     }
     errLen = sizeof(err);
-    if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errLen) < 0 || err != 0) {
-        fcntl(fd, F_SETFL, flags);
+    if (getsockopt(fd, SOL_SOCKET, SO_ERROR, (char *)&err, &errLen) < 0 || err != 0) {
+        calogSockSetNonblock(fd, 0);
         return false;
     }
-    fcntl(fd, F_SETFL, flags);
+    calogSockSetNonblock(fd, 0);
     return true;
 }
 
@@ -467,10 +461,10 @@ static int32_t httpConnOpen(HttpConnT *conn, const HttpUrlT *url, CalogValueT *r
     struct addrinfo  hints;
     struct addrinfo *res;
     struct addrinfo *rp;
-    int              fd;
+    CalogSocketT     fd;
     int              rc;
 
-    conn->fd  = -1;
+    conn->fd  = CALOG_INVALID_SOCKET;
     conn->ssl = NULL;
     conn->ctx = NULL;
     memset(&hints, 0, sizeof(hints));
@@ -480,21 +474,21 @@ static int32_t httpConnOpen(HttpConnT *conn, const HttpUrlT *url, CalogValueT *r
     if (rc != 0) {
         return calogFail(result, calogErrArgE, gai_strerror(rc));
     }
-    fd = -1;
+    fd = CALOG_INVALID_SOCKET;
     for (rp = res; rp != NULL; rp = rp->ai_next) {
         fd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
-        if (fd < 0) {
+        if (fd == CALOG_INVALID_SOCKET) {
             continue;
         }
         httpSetTimeouts(fd, HTTP_TIMEOUT_SEC);
-        if (httpConnectTimeout(fd, rp->ai_addr, rp->ai_addrlen, HTTP_TIMEOUT_SEC)) {
+        if (httpConnectTimeout(fd, rp->ai_addr, (socklen_t)rp->ai_addrlen, HTTP_TIMEOUT_SEC)) {
             break;
         }
-        close(fd);
-        fd = -1;
+        calogSockClose(fd);
+        fd = CALOG_INVALID_SOCKET;
     }
     freeaddrinfo(res);
-    if (fd < 0) {
+    if (fd == CALOG_INVALID_SOCKET) {
         return calogFail(result, calogErrArgE, "http: could not connect to host");
     }
     conn->fd = fd;
@@ -532,7 +526,7 @@ static ssize_t httpConnRead(HttpConnT *conn, char *buf, size_t length) {
     for (;;) {
         ssize_t r;
         r = recv(conn->fd, buf, length, 0);
-        if (r < 0 && errno == EINTR) {
+        if (r < 0 && calogSockErrno() == EINTR) {
             continue;
         }
         return r;
@@ -556,9 +550,9 @@ static int32_t httpConnWriteAll(HttpConnT *conn, const char *bytes, size_t lengt
             sent += (size_t)w;
         } else {
             ssize_t w;
-            w = send(conn->fd, bytes + sent, length - sent, MSG_NOSIGNAL);
+            w = send(conn->fd, bytes + sent, length - sent, CALOG_MSG_NOSIGNAL);
             if (w < 0) {
-                if (errno == EINTR) {
+                if (calogSockErrno() == EINTR) {
                     continue;
                 }
                 return calogErrArgE;
@@ -1451,15 +1445,22 @@ static bool httpSameOrigin(const HttpUrlT *a, const HttpUrlT *b) {
 }
 
 
-static void httpSetTimeouts(int fd, int timeoutSec) {
+static void httpSetTimeouts(CalogSocketT fd, int timeoutSec) {
+    // Best effort: if the kernel rejects these (unusual), the connection just falls back to no
+    // read/write deadline rather than failing the request outright. Winsock's SO_RCVTIMEO takes a
+    // DWORD of milliseconds; BSD sockets take a struct timeval.
+#if defined(_WIN32)
+    DWORD ms;
+    ms = (DWORD)timeoutSec * 1000u;
+    (void)setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, (const char *)&ms, sizeof(ms));
+    (void)setsockopt(fd, SOL_SOCKET, SO_SNDTIMEO, (const char *)&ms, sizeof(ms));
+#else
     struct timeval tv;
-
     tv.tv_sec  = timeoutSec;
     tv.tv_usec = 0;
-    // Best effort: if the kernel rejects these (unusual), the connection just falls back to no
-    // read/write deadline rather than failing the request outright.
-    (void)setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv));
-    (void)setsockopt(fd, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv));
+    (void)setsockopt(fd, SOL_SOCKET, SO_RCVTIMEO, (const char *)&tv, sizeof(tv));
+    (void)setsockopt(fd, SOL_SOCKET, SO_SNDTIMEO, (const char *)&tv, sizeof(tv));
+#endif
 }
 
 
diff --git a/libs/calogNet.c b/libs/calogNet.c
index 9c1a0ce3..7de42668 100644
--- a/libs/calogNet.c
+++ b/libs/calogNet.c
@@ -8,18 +8,13 @@
 
 #include "calogHandle.h"
 #include "calogInternal.h"
+#include "calogPlatform.h"
 
-#include <arpa/inet.h>
-#include <errno.h>
-#include <netdb.h>
-#include <netinet/in.h>
 #include <pthread.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/socket.h>
-#include <unistd.h>
 
 #include <enet/enet.h>
 
@@ -38,7 +33,7 @@
 #define NET_PORT_MAX 65535
 
 typedef struct NetSocketT {
-    int fd;
+    CalogSocketT fd;
 } NetSocketT;
 
 // Process-wide network library state shared by every runtime that registers the natives.
@@ -63,12 +58,12 @@ static int32_t enetHost(CalogValueT *args, int32_t argCount, CalogValueT *result
 static int32_t enetSend(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData);
 static int32_t enetService(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData);
 static void    netCloser(uint32_t type, void *resource);
-static int32_t netOpenBound(uint16_t port, int socktype, bool doListen, CalogValueT *result, int *fdOut);
+static int32_t netOpenBound(uint16_t port, int socktype, bool doListen, CalogValueT *result, CalogSocketT *fdOut);
 static bool    netPortOk(int64_t port);
 static int     netResolve(const char *host, uint16_t port, int socktype, bool passive, struct addrinfo **out);
 static int32_t netSocketClose(NetLibT *lib, int64_t handleId, uint32_t type1, uint32_t type2, CalogValueT *result, const char *message);
 static void    netSocketFree(NetSocketT *sock);
-static int32_t netStore(NetLibT *lib, int fd, uint32_t type, CalogValueT *result);
+static int32_t netStore(NetLibT *lib, CalogSocketT fd, uint32_t type, CalogValueT *result);
 static int32_t tcpAccept(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData);
 static int32_t tcpClose(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData);
 static int32_t tcpConnect(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData);
@@ -110,20 +105,28 @@ int32_t calogNetRegister(CalogT *calog) {
     pthread_mutex_lock(&gNetLibMutex);
     if (gNetLib == NULL) {
         NetLibT *newLib;
+        // Windows requires WSAStartup before any socket use (no-op on POSIX).
+        if (calogPlatformNetInit() != 0) {
+            pthread_mutex_unlock(&gNetLibMutex);
+            return calogErrOomE;
+        }
         newLib = (NetLibT *)calloc(1, sizeof(*newLib));
         if (newLib == NULL) {
+            calogPlatformNetShutdown();
             pthread_mutex_unlock(&gNetLibMutex);
             return calogErrOomE;
         }
         newLib->handles = calogHandleTableCreate();
         if (newLib->handles == NULL) {
             free(newLib);
+            calogPlatformNetShutdown();
             pthread_mutex_unlock(&gNetLibMutex);
             return calogErrOomE;
         }
         if (enet_initialize() != 0) {
             calogHandleTableDestroy(newLib->handles, NULL);
             free(newLib);
+            calogPlatformNetShutdown();
             pthread_mutex_unlock(&gNetLibMutex);
             return calogErrOomE;
         }
@@ -159,6 +162,7 @@ void calogNetShutdown(void) {
     if (gNetLib->refCount <= 0) {
         calogHandleTableDestroy(gNetLib->handles, netCloser);
         enet_deinitialize();
+        calogPlatformNetShutdown();
         free(gNetLib);
         gNetLib = NULL;
     }
@@ -187,40 +191,40 @@ static void netCloser(uint32_t type, void *resource) {
 
 // Create a socket bound to the given local port (0 = ephemeral), optionally listening.
 // Returns calogOkE with *fdOut set, or an error with result populated.
-static int32_t netOpenBound(uint16_t port, int socktype, bool doListen, CalogValueT *result, int *fdOut) {
+static int32_t netOpenBound(uint16_t port, int socktype, bool doListen, CalogValueT *result, CalogSocketT *fdOut) {
     struct addrinfo *res;
     struct addrinfo *rp;
-    int              fd;
+    CalogSocketT     fd;
     int              rc;
     int              yes;
 
-    *fdOut = -1;
+    *fdOut = CALOG_INVALID_SOCKET;
     yes    = 1;
     rc     = netResolve(NULL, port, socktype, true, &res);
     if (rc != 0) {
         return calogFail(result, calogErrArgE, gai_strerror(rc));
     }
-    fd = -1;
+    fd = CALOG_INVALID_SOCKET;
     for (rp = res; rp != NULL; rp = rp->ai_next) {
         fd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
-        if (fd < 0) {
+        if (fd == CALOG_INVALID_SOCKET) {
             continue;
         }
-        setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes));
-        if (bind(fd, rp->ai_addr, rp->ai_addrlen) == 0) {
+        setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, (const char *)&yes, sizeof(yes));
+        if (bind(fd, rp->ai_addr, (socklen_t)rp->ai_addrlen) == 0) {
             break;
         }
-        close(fd);
-        fd = -1;
+        calogSockClose(fd);
+        fd = CALOG_INVALID_SOCKET;
     }
     freeaddrinfo(res);
-    if (fd < 0) {
+    if (fd == CALOG_INVALID_SOCKET) {
         return calogFail(result, calogErrArgE, "could not bind the requested port");
     }
     if (doListen && listen(fd, SOMAXCONN) != 0) {
         int32_t status;
-        status = calogFail(result, calogErrArgE, strerror(errno));
-        close(fd);
+        status = calogFail(result, calogErrArgE, calogSockErrStr());
+        calogSockClose(fd);
         return status;
     }
     *fdOut = fd;
@@ -270,26 +274,26 @@ static int32_t netSocketClose(NetLibT *lib, int64_t handleId, uint32_t type1, ui
 // Close the fd and release the NetSocketT. Shared by netSocketClose and the handle-table
 // teardown path (netCloser).
 static void netSocketFree(NetSocketT *sock) {
-    close(sock->fd);
+    calogSockClose(sock->fd);
     free(sock);
 }
 
 
 // Wrap an open fd in a handle-table entry, transferring ownership. On failure the fd is
 // closed. Sets result to the new integer handle on success.
-static int32_t netStore(NetLibT *lib, int fd, uint32_t type, CalogValueT *result) {
+static int32_t netStore(NetLibT *lib, CalogSocketT fd, uint32_t type, CalogValueT *result) {
     NetSocketT *sock;
     int64_t     handle;
 
     sock = (NetSocketT *)malloc(sizeof(*sock));
     if (sock == NULL) {
-        close(fd);
+        calogSockClose(fd);
         return calogFail(result, calogErrOomE, "out of memory");
     }
     sock->fd = fd;
     handle   = calogHandleAdd(lib->handles, type, sock);
     if (handle == 0) {
-        close(fd);
+        calogSockClose(fd);
         free(sock);
         return calogFail(result, calogErrOomE, "out of memory");
     }
@@ -299,9 +303,9 @@ static int32_t netStore(NetLibT *lib, int fd, uint32_t type, CalogValueT *result
 
 
 static int32_t tcpAccept(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData) {
-    NetLibT    *lib;
-    NetSocketT *listener;
-    int         fd;
+    NetLibT     *lib;
+    NetSocketT  *listener;
+    CalogSocketT fd;
 
     lib = (NetLibT *)userData;
     calogValueNil(result);
@@ -313,8 +317,8 @@ static int32_t tcpAccept(CalogValueT *args, int32_t argCount, CalogValueT *resul
         return calogFail(result, calogErrArgE, "tcpAccept: invalid listener handle");
     }
     fd = accept(listener->fd, NULL, NULL);
-    if (fd < 0) {
-        return calogFail(result, calogErrArgE, strerror(errno));
+    if (fd == CALOG_INVALID_SOCKET) {
+        return calogFail(result, calogErrArgE, calogSockErrStr());
     }
     return netStore(lib, fd, NET_TYPE_TCP, result);
 }
@@ -336,7 +340,7 @@ static int32_t tcpConnect(CalogValueT *args, int32_t argCount, CalogValueT *resu
     NetLibT         *lib;
     struct addrinfo *res;
     struct addrinfo *rp;
-    int              fd;
+    CalogSocketT     fd;
     int              rc;
 
     lib = (NetLibT *)userData;
@@ -351,20 +355,20 @@ static int32_t tcpConnect(CalogValueT *args, int32_t argCount, CalogValueT *resu
     if (rc != 0) {
         return calogFail(result, calogErrArgE, gai_strerror(rc));
     }
-    fd = -1;
+    fd = CALOG_INVALID_SOCKET;
     for (rp = res; rp != NULL; rp = rp->ai_next) {
         fd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
-        if (fd < 0) {
+        if (fd == CALOG_INVALID_SOCKET) {
             continue;
         }
-        if (connect(fd, rp->ai_addr, rp->ai_addrlen) == 0) {
+        if (connect(fd, rp->ai_addr, (socklen_t)rp->ai_addrlen) == 0) {
             break;
         }
-        close(fd);
-        fd = -1;
+        calogSockClose(fd);
+        fd = CALOG_INVALID_SOCKET;
     }
     freeaddrinfo(res);
-    if (fd < 0) {
+    if (fd == CALOG_INVALID_SOCKET) {
         return calogFail(result, calogErrArgE, "tcpConnect: could not connect");
     }
     return netStore(lib, fd, NET_TYPE_TCP, result);
@@ -372,9 +376,9 @@ static int32_t tcpConnect(CalogValueT *args, int32_t argCount, CalogValueT *resu
 
 
 static int32_t tcpListen(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData) {
-    NetLibT *lib;
-    int      fd;
-    int32_t  status;
+    NetLibT     *lib;
+    CalogSocketT fd;
+    int32_t      status;
 
     lib = (NetLibT *)userData;
     calogValueNil(result);
@@ -417,7 +421,7 @@ static int32_t tcpRecv(CalogValueT *args, int32_t argCount, CalogValueT *result,
     }
     received = recv(sock->fd, buffer, (size_t)args[1].as.i, 0);
     if (received < 0) {
-        status = calogFail(result, calogErrArgE, strerror(errno));
+        status = calogFail(result, calogErrArgE, calogSockErrStr());
         free(buffer);
         return status;
     }
@@ -449,9 +453,9 @@ static int32_t tcpSend(CalogValueT *args, int32_t argCount, CalogValueT *result,
     total = 0;
     while (total < args[1].as.s.length) {
         ssize_t sent;
-        sent = send(sock->fd, args[1].as.s.bytes + total, (size_t)(args[1].as.s.length - total), MSG_NOSIGNAL);
+        sent = send(sock->fd, args[1].as.s.bytes + total, (size_t)(args[1].as.s.length - total), CALOG_MSG_NOSIGNAL);
         if (sent < 0) {
-            return calogFail(result, calogErrArgE, strerror(errno));
+            return calogFail(result, calogErrArgE, calogSockErrStr());
         }
         total += sent;
     }
@@ -473,9 +477,9 @@ static int32_t udpClose(CalogValueT *args, int32_t argCount, CalogValueT *result
 
 
 static int32_t udpOpen(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData) {
-    NetLibT *lib;
-    int      fd;
-    int32_t  status;
+    NetLibT     *lib;
+    CalogSocketT fd;
+    int32_t      status;
 
     lib = (NetLibT *)userData;
     calogValueNil(result);
@@ -523,7 +527,7 @@ static int32_t udpRecvFrom(CalogValueT *args, int32_t argCount, CalogValueT *res
     fromLength = sizeof(from);
     received   = recvfrom(sock->fd, buffer, (size_t)args[1].as.i, 0, (struct sockaddr *)&from, &fromLength);
     if (received < 0) {
-        status = calogFail(result, calogErrArgE, strerror(errno));
+        status = calogFail(result, calogErrArgE, calogSockErrStr());
         free(buffer);
         return status;
     }
@@ -578,7 +582,7 @@ static int32_t udpSendTo(CalogValueT *args, int32_t argCount, CalogValueT *resul
     sent = sendto(sock->fd, args[3].as.s.bytes, (size_t)args[3].as.s.length, 0, res->ai_addr, res->ai_addrlen);
     freeaddrinfo(res);
     if (sent < 0) {
-        return calogFail(result, calogErrArgE, strerror(errno));
+        return calogFail(result, calogErrArgE, calogSockErrStr());
     }
     calogValueInt(result, (int64_t)sent);
     return calogOkE;
diff --git a/libs/calogSsh.c b/libs/calogSsh.c
index 49d7a7e8..70e44af5 100644
--- a/libs/calogSsh.c
+++ b/libs/calogSsh.c
@@ -10,17 +10,17 @@
 
 #include "calogHandle.h"
 #include "calogInternal.h"
+#include "calogPlatform.h"
 
-#include <netdb.h>
 #include <pthread.h>
 #include <stdbool.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
-#include <sys/select.h>
-#include <sys/socket.h>
-#include <unistd.h>
+#ifndef _WIN32
+#include <sys/select.h>   // fd_set/select come from winsock2.h on Windows (via calogPlatform.h)
+#endif
 
 #include <libssh2.h>
 #include <libssh2_sftp.h>
@@ -46,7 +46,7 @@
 // SFTP subsystem. The handle is a plain process-wide entry with no owning context: any context
 // it is passed to may operate on it (see the threading notes in calogSsh.h).
 typedef struct SshConnT {
-    int             sock;
+    CalogSocketT    sock;
     LIBSSH2_SESSION *session;
     LIBSSH2_SFTP    *sftp;
 } SshConnT;
@@ -88,26 +88,33 @@ static int32_t sshConnect(CalogValueT *args, int32_t argCount, CalogValueT *resu
 static int32_t sshDrainStep(SshReadFnT readFn, void *handle, int streamId, char **bytes, int64_t *cap, int64_t *length, SshStepE *step);
 static int32_t sshEnsureSftp(SshConnT *conn, CalogValueT *result);
 static int32_t sshExec(CalogValueT *args, int32_t argCount, CalogValueT *result, void *userData);
-static int32_t sshExecDrain(int sock, LIBSSH2_SESSION *session, LIBSSH2_CHANNEL *channel, char **outBytes, int64_t *outLength, char **errBytes, int64_t *errLength);
+static int32_t sshExecDrain(CalogSocketT sock, LIBSSH2_SESSION *session, LIBSSH2_CHANNEL *channel, char **outBytes, int64_t *outLength, char **errBytes, int64_t *errLength);
 static ssize_t sshReadChannel(void *handle, int streamId, char *buffer, size_t length);
 static ssize_t sshReadSftpFile(void *handle, int streamId, char *buffer, size_t length);
 static int32_t sshResolve(SshLibT *lib, int64_t handle, CalogValueT *result, SshConnT **out);
-static void    sshSessionDestroy(LIBSSH2_SESSION *session, int sock, const char *reason);
-static void    sshWaitSocket(int sock, LIBSSH2_SESSION *session);
+static void    sshSessionDestroy(LIBSSH2_SESSION *session, CalogSocketT sock, const char *reason);
+static void    sshWaitSocket(CalogSocketT sock, LIBSSH2_SESSION *session);
 
 
 int32_t calogSshRegister(CalogT *calog) {
     pthread_mutex_lock(&gSshLibMutex);
     if (gSshLib == NULL) {
         SshLibT *lib;
+        // Windows needs WSAStartup before any socket use (no-op on POSIX).
+        if (calogPlatformNetInit() != 0) {
+            pthread_mutex_unlock(&gSshLibMutex);
+            return calogErrUnsupportedE;
+        }
         lib = (SshLibT *)calloc(1, sizeof(*lib));
         if (lib == NULL) {
+            calogPlatformNetShutdown();
             pthread_mutex_unlock(&gSshLibMutex);
             return calogErrOomE;
         }
         lib->handles = calogHandleTableCreate();
         if (lib->handles == NULL) {
             free(lib);
+            calogPlatformNetShutdown();
             pthread_mutex_unlock(&gSshLibMutex);
             return calogErrOomE;
         }
@@ -115,6 +122,7 @@ int32_t calogSshRegister(CalogT *calog) {
         if (libssh2_init(0) != 0) {
             calogHandleTableDestroy(lib->handles, NULL);
             free(lib);
+            calogPlatformNetShutdown();
             pthread_mutex_unlock(&gSshLibMutex);
             return calogErrUnsupportedE;
         }
@@ -147,6 +155,7 @@ void calogSshShutdown(void) {
     if (gSshLib->refCount <= 0) {
         calogHandleTableDestroy(gSshLib->handles, sshCloser);
         libssh2_exit();
+        calogPlatformNetShutdown();
         free(gSshLib);
         gSshLib = NULL;
     }
@@ -546,7 +555,7 @@ static int32_t sshConnect(CalogValueT *args, int32_t argCount, CalogValueT *resu
     char             portBuffer[8];
     int64_t          port;
     int64_t          handle;
-    int              fd;
+    CalogSocketT     fd;
     int              rc;
 
     lib = (SshLibT *)userData;
@@ -572,31 +581,31 @@ static int32_t sshConnect(CalogValueT *args, int32_t argCount, CalogValueT *resu
     if (rc != 0) {
         return calogFail(result, calogErrArgE, gai_strerror(rc));
     }
-    fd = -1;
+    fd = CALOG_INVALID_SOCKET;
     for (rp = res; rp != NULL; rp = rp->ai_next) {
         fd = socket(rp->ai_family, rp->ai_socktype, rp->ai_protocol);
-        if (fd < 0) {
+        if (fd == CALOG_INVALID_SOCKET) {
             continue;
         }
-        if (connect(fd, rp->ai_addr, rp->ai_addrlen) == 0) {
+        if (connect(fd, rp->ai_addr, (socklen_t)rp->ai_addrlen) == 0) {
             break;
         }
-        close(fd);
-        fd = -1;
+        calogSockClose(fd);
+        fd = CALOG_INVALID_SOCKET;
     }
     freeaddrinfo(res);
-    if (fd < 0) {
+    if (fd == CALOG_INVALID_SOCKET) {
         return calogFail(result, calogErrArgE, "sshConnect: could not connect");
     }
     session = libssh2_session_init();
     if (session == NULL) {
-        close(fd);
+        calogSockClose(fd);
         return calogFail(result, calogErrOomE, "sshConnect: could not create a session");
     }
     libssh2_session_set_blocking(session, 1);
     if (libssh2_session_handshake(session, fd) != 0) {
         libssh2_session_free(session);
-        close(fd);
+        calogSockClose(fd);
         return calogFail(result, calogErrArgE, "sshConnect: SSH handshake failed");
     }
     conn = (SshConnT *)calloc(1, sizeof(*conn));
@@ -766,7 +775,7 @@ static int32_t sshExec(CalogValueT *args, int32_t argCount, CalogValueT *result,
 // duration (restored before every return) so a stream with nothing ready right now cannot
 // block progress on the other; sshWaitSocket parks the calling thread only when neither
 // stream can make progress right now. On failure both output buffers are freed here.
-static int32_t sshExecDrain(int sock, LIBSSH2_SESSION *session, LIBSSH2_CHANNEL *channel, char **outBytes, int64_t *outLength, char **errBytes, int64_t *errLength) {
+static int32_t sshExecDrain(CalogSocketT sock, LIBSSH2_SESSION *session, LIBSSH2_CHANNEL *channel, char **outBytes, int64_t *outLength, char **errBytes, int64_t *errLength) {
     char    *outBuf;
     char    *errBuf;
     int64_t  outCap;
@@ -862,17 +871,17 @@ static int32_t sshResolve(SshLibT *lib, int64_t handle, CalogValueT *result, Ssh
 // Tear down a raw session and its socket (disconnect, free the session, close the fd). Used
 // by sshConnDestroy and by sshConnect's post-handshake failure paths, before a SshConnT
 // exists to own the session.
-static void sshSessionDestroy(LIBSSH2_SESSION *session, int sock, const char *reason) {
+static void sshSessionDestroy(LIBSSH2_SESSION *session, CalogSocketT sock, const char *reason) {
     libssh2_session_disconnect(session, reason);
     libssh2_session_free(session);
-    close(sock);
+    calogSockClose(sock);
 }
 
 
 // Block the calling thread until the socket is ready in whatever direction libssh2 is
 // currently waiting on. Used only by sshExecDrain's non-blocking interleave; every other
 // native in this file runs its session in blocking mode and never calls this.
-static void sshWaitSocket(int sock, LIBSSH2_SESSION *session) {
+static void sshWaitSocket(CalogSocketT sock, LIBSSH2_SESSION *session) {
     fd_set  fdSet;
     fd_set *readSet;
     fd_set *writeSet;
diff --git a/src/calogMain.c b/src/calogMain.c
index 73d7af82..2eb52f68 100644
--- a/src/calogMain.c
+++ b/src/calogMain.c
@@ -470,6 +470,11 @@ int main(int argc, char **argv) {
 
     signal(SIGINT, onSignal);
     signal(SIGTERM, onSignal);
+#ifdef SIGPIPE
+    // Ignore SIGPIPE so a peer resetting a socket cannot terminate the process. Linux avoids the
+    // signal via MSG_NOSIGNAL on send(), but macOS lacks that flag, so this is the portable guard.
+    signal(SIGPIPE, SIG_IGN);
+#endif
 
     // Launch every script fire-and-forget on its own context, recording the live ones so a
     // later error can retire exactly that context. calogContextEval keeps its own copy of the
diff --git a/src/calogPlatform.h b/src/calogPlatform.h
new file mode 100644
index 00000000..4ef2e47b
--- /dev/null
+++ b/src/calogPlatform.h
@@ -0,0 +1,100 @@
+// calogPlatform.h -- OS abstraction for calog's small platform surface.
+//
+// calog's core (values, broker, actor, engines) is portable C11. The only OS-specific surface is
+// the socket layer used by the net/ssh/http libraries: BSD sockets on Linux and macOS, Winsock on
+// Windows. pthreads and getaddrinfo are available on all three (on Windows via mingw-w64's
+// winpthreads + Winsock), so only the socket type, close/error accessors, non-blocking toggle, and
+// a one-time subsystem init differ. This header papers over exactly those.
+//
+// POSIX (Linux/macOS): everything maps to the standard names; init/shutdown are no-ops.
+// Windows: maps to Winsock (closesocket / WSAGetLastError / ioctlsocket) and WSAStartup/WSACleanup.
+
+#ifndef CALOG_PLATFORM_H
+#define CALOG_PLATFORM_H
+
+#if defined(_WIN32)
+
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN
+#endif
+#ifndef _WIN32_WINNT
+#define _WIN32_WINNT 0x0601   // Windows 7+: exposes WSAPoll and struct pollfd
+#endif
+#include <winsock2.h>
+#include <ws2tcpip.h>
+#include <windows.h>
+#include <stdio.h>
+
+// poll on a small fd set (WSAPoll is API-compatible with poll for our single-socket use).
+#define calogPoll WSAPoll
+
+typedef SOCKET CalogSocketT;
+#define CALOG_INVALID_SOCKET INVALID_SOCKET
+#define CALOG_SOCKET_ERROR   SOCKET_ERROR
+
+static inline int  calogSockClose(CalogSocketT s)          { return closesocket(s); }
+static inline int  calogSockErrno(void)                    { return WSAGetLastError(); }
+static inline int  calogSockWouldBlock(int err)            { return err == WSAEWOULDBLOCK; }
+static inline int  calogSockInProgress(int err)            { return err == WSAEWOULDBLOCK || err == WSAEINPROGRESS; }
+static inline int  calogSockSetNonblock(CalogSocketT s, int on) { u_long mode = (u_long)(on != 0); return ioctlsocket(s, FIONBIO, &mode); }
+static inline int  calogPlatformNetInit(void)              { WSADATA data; return WSAStartup(MAKEWORD(2, 2), &data); }
+static inline void calogPlatformNetShutdown(void)          { WSACleanup(); }
+
+// Best-effort message for the last socket error (Winsock has no errno; strerror does not map it).
+// Uses a thread-local buffer so concurrent context threads do not clobber each other.
+static inline const char *calogSockErrStr(void) {
+    static __declspec(thread) char buffer[256];
+    int   err;
+    DWORD written;
+    err     = WSAGetLastError();
+    written = FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, NULL, (DWORD)err, 0, buffer, (DWORD)sizeof(buffer), NULL);
+    if (written == 0) {
+        snprintf(buffer, sizeof(buffer), "winsock error %d", err);
+    }
+    return buffer;
+}
+
+#else
+
+#include <sys/socket.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+#include <netdb.h>
+#include <unistd.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <string.h>
+#include <poll.h>
+
+#define calogPoll poll
+
+typedef int CalogSocketT;
+#define CALOG_INVALID_SOCKET (-1)
+#define CALOG_SOCKET_ERROR   (-1)
+
+static inline int  calogSockClose(CalogSocketT s)          { return close(s); }
+static inline int  calogSockErrno(void)                    { return errno; }
+static inline int  calogSockWouldBlock(int err)            { return err == EWOULDBLOCK || err == EAGAIN; }
+static inline int  calogSockInProgress(int err)            { return err == EINPROGRESS; }
+static inline int  calogSockSetNonblock(CalogSocketT s, int on) {
+    int flags = fcntl(s, F_GETFL, 0);
+    if (flags < 0) {
+        return -1;
+    }
+    return fcntl(s, F_SETFL, on ? (flags | O_NONBLOCK) : (flags & ~O_NONBLOCK));
+}
+static inline const char *calogSockErrStr(void)            { return strerror(errno); }
+static inline int  calogPlatformNetInit(void)              { return 0; }
+static inline void calogPlatformNetShutdown(void)          { (void)0; }
+
+#endif
+
+// SIGPIPE suppression on send(): Linux uses the MSG_NOSIGNAL flag; macOS/Windows lack it (macOS
+// uses the SO_NOSIGPIPE socket option, Windows has no SIGPIPE). Falls back to 0 where absent.
+#ifdef MSG_NOSIGNAL
+#define CALOG_MSG_NOSIGNAL MSG_NOSIGNAL
+#else
+#define CALOG_MSG_NOSIGNAL 0
+#endif
+
+#endif
diff --git a/tools/crossBuild.sh b/tools/crossBuild.sh
new file mode 100755
index 00000000..c7bba977
--- /dev/null
+++ b/tools/crossBuild.sh
@@ -0,0 +1,109 @@
+#!/usr/bin/env bash
+#
+# crossBuild.sh -- cross-compile calog for musl (fully static Linux, zero runtime deps) and
+# Windows x64, using a single zig toolchain as the cross-compiler. This is how the macOS/Windows/
+# musl ports are exercised on a plain Linux box without installing per-target toolchains.
+#
+# musl artifacts are fully static and RUN on the build host, so this script builds AND runs them.
+# Windows artifacts (.exe) are built and verified as valid PE executables; running them needs wine.
+#
+# Requirements: zig (https://ziglang.org/download/). Point ZIG at the binary, or put it on PATH:
+#   ZIG=/path/to/zig ./tools/crossBuild.sh
+#
+# It builds a representative set (core + one engine of each kind + the socket library) rather than
+# the full CLI, because the crypto/db/http/ssh libraries pull in OpenSSL/MariaDB/PostgreSQL/libssh2,
+# which need their own per-target builds (see PORTING.md).
+
+set -u
+cd "$(dirname "$0")/.."
+ROOT=$(pwd)
+ZIG=${ZIG:-zig}
+OUT=${OUT:-build/cross}
+mkdir -p "$OUT"
+
+if ! command -v "$ZIG" >/dev/null 2>&1; then
+    echo "error: zig not found. Set ZIG=/path/to/zig (see https://ziglang.org/download/)." >&2
+    exit 1
+fi
+echo "zig: $($ZIG version)"
+
+CORE="src/broker.c src/value.c src/context.c"
+LUASRC=$(ls vendor/lua/src/*.c | grep -vE '/(lua|luac)\.c$' | tr '\n' ' ')
+ENET_UNIX=$(ls vendor/enet/*.c | grep -vE '/win32\.c$' | tr '\n' ' ')
+ENET_WIN=$(ls vendor/enet/*.c | grep -vE '/unix\.c$' | tr '\n' ' ')
+SQSRC=$(ls vendor/squirrel-src/squirrel/*.cpp | tr '\n' ' ')
+MBSRC=$(ls vendor/ourbasic/*.c | tr '\n' ' ')
+ENETDEF="-D_GNU_SOURCE -DHAS_FCNTL=1 -DHAS_POLL=1 -DHAS_GETADDRINFO=1 -DHAS_GETNAMEINFO=1 -DHAS_GETHOSTBYNAME_R=1 -DHAS_GETHOSTBYADDR_R=1 -DHAS_INET_PTON=1 -DHAS_INET_NTOP=1 -DHAS_MSGHDR_FLAGS=1 -DHAS_SOCKLEN_T=1"
+
+pass=0; fail=0
+mrun() {  # name, then compiler args -- build static musl, then RUN it
+    local name=$1; shift
+    if "$@" -o "$OUT/$name-musl" 2>"$OUT/$name-musl.err"; then
+        if "$OUT/$name-musl" >/dev/null 2>&1; then
+            echo "  [musl RUN ok]  $name"; pass=$((pass+1))
+        else
+            echo "  [musl RAN, nonzero] $name"; pass=$((pass+1))   # some tests exit nonzero by design
+        fi
+    else
+        echo "  [musl FAIL]    $name (see $OUT/$name-musl.err)"; fail=$((fail+1))
+    fi
+}
+wbuild() {  # name, then compiler args -- build a Windows .exe, verify it is a PE
+    local name=$1; shift
+    if "$@" -o "$OUT/$name.exe" 2>"$OUT/$name.exe.err"; then
+        if file "$OUT/$name.exe" | grep -q 'PE32+ executable'; then
+            echo "  [win  BUILT]   $name.exe"; pass=$((pass+1))
+        else
+            echo "  [win  ??]      $name.exe (not a PE?)"; fail=$((fail+1))
+        fi
+    else
+        echo "  [win  FAIL]    $name.exe (see $OUT/$name.exe.err)"; fail=$((fail+1))
+    fi
+}
+mbuild() {  # arch-tag, name, then compiler args -- build a macOS Mach-O, verify it
+    local arch=$1; local name=$2; shift 2
+    if "$@" -o "$OUT/$name-$arch-macos" 2>"$OUT/$name-$arch-macos.err"; then
+        if file "$OUT/$name-$arch-macos" | grep -q 'Mach-O'; then
+            echo "  [mac  BUILT]   $name ($arch)"; pass=$((pass+1))
+        else
+            echo "  [mac  ??]      $name ($arch, not Mach-O?)"; fail=$((fail+1))
+        fi
+    else
+        echo "  [mac  FAIL]    $name ($arch, see $OUT/$name-$arch-macos.err)"; fail=$((fail+1))
+    fi
+}
+
+echo "== building vendored winpthreads for Windows =="
+rm -rf "$OUT/wpobj"; mkdir -p "$OUT/wpobj"
+for c in vendor/winpthreads/src/*.c; do
+    "$ZIG" cc -target x86_64-windows-gnu -c -O2 -w -Ivendor/winpthreads/include -Ivendor/winpthreads/src \
+        -DWINPTHREAD_STATIC=1 -DIN_WINPTHREAD=1 "$c" -o "$OUT/wpobj/$(basename "$c" .c).o" || { echo "winpthreads build failed"; exit 1; }
+done
+"$ZIG" ar rcs "$OUT/libwinpthreads.a" "$OUT"/wpobj/*.o
+WPA="$OUT/libwinpthreads.a"
+WPINC="-Ivendor/winpthreads/include -DWINPTHREAD_STATIC=1"
+
+echo "== musl (fully static Linux, built AND run) =="
+mrun testEngineLua   "$ZIG" cc -target x86_64-linux-musl -static -O2 -pthread -w -Isrc -Isrc/lua -Ilibs -Ivendor/lua/src -DLUA_USE_POSIX $CORE src/lua/luaEngine.c src/lua/luaAdapter.c tests/testEngineLua.c $LUASRC -lm
+mrun testEngineJs    "$ZIG" cc -target x86_64-linux-musl -static -O2 -pthread -w -Isrc -Isrc/js -Ilibs -Ivendor/quickjs -D_GNU_SOURCE $CORE src/js/jsEngine.c src/js/jsAdapter.c tests/testEngineJs.c vendor/quickjs/quickjs.c vendor/quickjs/libregexp.c vendor/quickjs/libunicode.c vendor/quickjs/dtoa.c -lm
+mrun testEngineMyBasic "$ZIG" cc -target x86_64-linux-musl -static -O2 -pthread -w -Isrc -Isrc/mybasic -Ilibs -Ivendor/ourbasic -DMB_DOUBLE_FLOAT $CORE src/mybasic/mybasicEngine.c src/mybasic/mybasicAdapter.c tests/testEngineMyBasic.c $MBSRC -lm
+mrun testNet         "$ZIG" cc -target x86_64-linux-musl -static -O2 -pthread -w -Isrc -Isrc/lua -Ilibs -Ivendor/lua/src -Ivendor/enet/include -DLUA_USE_POSIX $ENETDEF $CORE src/lua/luaEngine.c src/lua/luaAdapter.c libs/calogNet.c libs/calogHandle.c tests/testNet.c $LUASRC $ENET_UNIX -lm
+# Squirrel is C++ (vendored VM); the calog .c files stay C (-x c), the VM is C++ (-x c++).
+if "$ZIG" c++ -target x86_64-linux-musl -static -O2 -pthread -w -Isrc -Isrc/squirrel -Ilibs -Ivendor/squirrel-src/include -Ivendor/squirrel-src/squirrel -D_SQ64 -DSQUSEDOUBLE \
+    -x c $CORE src/squirrel/squirrelEngine.c src/squirrel/squirrelAdapter.c tests/testEngineSquirrel.c -x c++ $SQSRC -o "$OUT/testEngineSquirrel-musl" 2>"$OUT/sq-musl.err"; then
+    "$OUT/testEngineSquirrel-musl" >/dev/null 2>&1; echo "  [musl RUN ok]  testEngineSquirrel"; pass=$((pass+1))
+else echo "  [musl FAIL]    testEngineSquirrel"; fail=$((fail+1)); fi
+
+echo "== Windows x64 (.exe built against vendored winpthreads; run needs wine) =="
+wbuild testEngineLua "$ZIG" cc -target x86_64-windows-gnu -O2 -w -Isrc -Isrc/lua -Ilibs -Ivendor/lua/src $WPINC $CORE src/lua/luaEngine.c src/lua/luaAdapter.c tests/testEngineLua.c $LUASRC "$WPA"
+wbuild testNet       "$ZIG" cc -target x86_64-windows-gnu -O2 -w -Isrc -Isrc/lua -Ilibs -Ivendor/lua/src -Ivendor/enet/include $WPINC $CORE src/lua/luaEngine.c src/lua/luaAdapter.c libs/calogNet.c libs/calogHandle.c tests/testNet.c $LUASRC $ENET_WIN "$WPA" -lws2_32 -lwinmm
+
+echo "== macOS (Mach-O built via zig's libSystem stub -- Intel + Apple Silicon; run needs a Mac) =="
+for arch in x86_64 aarch64; do
+    mbuild "$arch" testEngineLua "$ZIG" cc -target $arch-macos -O2 -w -Isrc -Isrc/lua -Ilibs -Ivendor/lua/src -DLUA_USE_MACOSX $CORE src/lua/luaEngine.c src/lua/luaAdapter.c tests/testEngineLua.c $LUASRC
+    mbuild "$arch" testNet       "$ZIG" cc -target $arch-macos -O2 -w -Isrc -Isrc/lua -Ilibs -Ivendor/lua/src -Ivendor/enet/include -DLUA_USE_MACOSX $ENETDEF $CORE src/lua/luaEngine.c src/lua/luaAdapter.c libs/calogNet.c libs/calogHandle.c tests/testNet.c $LUASRC $ENET_UNIX
+done
+
+echo
+echo "== cross-build: $pass ok, $fail failed  (artifacts in $OUT/) =="
+[ "$fail" -eq 0 ]
diff --git a/vendor/winpthreads/COPYING b/vendor/winpthreads/COPYING
new file mode 100644
index 00000000..35077012
--- /dev/null
+++ b/vendor/winpthreads/COPYING
@@ -0,0 +1,57 @@
+Copyright (c) 2011 mingw-w64 project
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+the rights to use, copy, modify, merge, publish, distribute, sublicense,
+and/or sell copies of the Software, and to permit persons to whom the
+Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+DEALINGS IN THE SOFTWARE.
+
+
+/*
+ * Parts of this library are derived by:
+ *
+ * Posix Threads library for Microsoft Windows
+ *
+ * Use at own risk, there is no implied warranty to this code.
+ * It uses undocumented features of Microsoft Windows that can change
+ * at any time in the future.
+ *
+ * (C) 2010 Lockless Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification,
+ * are permitted provided that the following conditions are met:
+ *
+ *
+ *  * Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *  * 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.
+ *  * Neither the name of Lockless Inc. nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AN
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, 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 OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
diff --git a/vendor/winpthreads/NOTICE.calog b/vendor/winpthreads/NOTICE.calog
new file mode 100644
index 00000000..41cc0629
--- /dev/null
+++ b/vendor/winpthreads/NOTICE.calog
@@ -0,0 +1,26 @@
+winpthreads (vendored into calog)
+=================================
+
+This is the winpthreads library from mingw-w64 (mingw-w64-libraries/winpthreads),
+a POSIX threads implementation over the Win32 API. calog uses pthreads throughout
+(the actor model, the timer, pubsub, kv, ...); on Windows those pthread calls resolve
+against this library, so the same threading code compiles on Linux, macOS, and Windows.
+
+Vendored from the mingw-w64 source tree. Only the buildable library is kept:
+
+  src/       the implementation (.c/.h)
+  include/   the public headers (pthread.h, sched.h, semaphore.h, ...)
+  COPYING    the license (permissive, MIT/BSD-style; see the file)
+  README     upstream readme
+
+Removed from the upstream subtree: tests/, build-aux/, m4/, and the autoconf machinery
+(Makefile.am, configure.ac). calog builds the sources directly (no autoconf); the one
+generated header the sources expect, src/config.h, is a minimal hand-written stub
+(winpthreads only uses it for optional feature probes, whose portable fallbacks are used).
+
+Build (static, for a Windows target), as done by the Makefile and tools/crossBuild.sh:
+
+  cc -c -O2 -Iinclude -Isrc -DWINPTHREAD_STATIC=1 -DIN_WINPTHREAD=1 src/*.c
+  ar rcs libwinpthreads.a *.o
+
+This library is only needed for Windows builds; Linux and macOS use their native pthreads.
diff --git a/vendor/winpthreads/README b/vendor/winpthreads/README
new file mode 100644
index 00000000..44c71510
--- /dev/null
+++ b/vendor/winpthreads/README
@@ -0,0 +1,18 @@
+# The Winpthreads Library
+
+This library provides POSIX threading APIs for mingw-w64.
+
+## Linking Against Winpthreads
+
+For maximum compatibility, winpthreads headers expose APIs without the
+`dllimport` attribute by default.
+
+If your program is linked against the DLL, you may define the macro
+`WINPTHREADS_USE_DLLIMPORT` to add `dllimport` attributes to all APIs,
+which makes function calls to them a bit more efficient.
+
+Be careful when you build static libraries; you probably do not want to define
+`WINPTHREADS_USE_DLLIMPORT` unconditionally, to ensure that your code can link
+against either shared or static winpthreads. Note that winpthreads does not
+export any variables, which means that `dllimport` is not requred, with both
+MSVC and mingw-w64 toolchains.
diff --git a/vendor/winpthreads/include/pthread.h b/vendor/winpthreads/include/pthread.h
new file mode 100644
index 00000000..3ab5db74
--- /dev/null
+++ b/vendor/winpthreads/include/pthread.h
@@ -0,0 +1,691 @@
+/*
+   Copyright (c) 2011-2016 mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+/*
+ * Parts of this library are derived by:
+ *
+ * Posix Threads library for Microsoft Windows
+ *
+ * Use at own risk, there is no implied warranty to this code.
+ * It uses undocumented features of Microsoft Windows that can change
+ * at any time in the future.
+ *
+ * (C) 2010 Lockless Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification,
+ * are permitted provided that the following conditions are met:
+ *
+ *
+ *  * Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *  * 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.
+ *  * Neither the name of Lockless Inc. nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AN
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, 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 OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#ifndef WIN_PTHREADS_H
+#define WIN_PTHREADS_H
+
+#include <stddef.h>
+#include <errno.h>
+#include <sys/types.h>
+
+#include <process.h>
+#include <limits.h>
+#include <signal.h>
+#include <time.h>
+
+#include "pthread_compat.h"
+#include "sched.h"
+
+#ifdef _MSC_VER
+#include "pthread_time.h"
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define __WINPTHREADS_VERSION_MAJOR 0
+#define __WINPTHREADS_VERSION_MINOR 5
+#define __WINPTHREADS_VERSION_PATCHLEVEL 0
+
+/* MSB 8-bit major version, 8-bit minor version, 16-bit patch level.  */
+#define __WINPTHREADS_VERSION 0x00050000
+
+/* Compatibility stuff: */
+#define RWLS_PER_THREAD						8
+
+/* pthread specific defines.  */
+
+#define PTHREAD_CANCEL_DISABLE 0
+#define PTHREAD_CANCEL_ENABLE 0x01
+
+#define PTHREAD_CANCEL_DEFERRED 0
+#define PTHREAD_CANCEL_ASYNCHRONOUS 0x02
+
+#define PTHREAD_CREATE_JOINABLE 0
+#define PTHREAD_CREATE_DETACHED 0x04
+
+#define PTHREAD_EXPLICIT_SCHED 0
+#define PTHREAD_INHERIT_SCHED 0x08
+
+#define PTHREAD_SCOPE_PROCESS 0
+#define PTHREAD_SCOPE_SYSTEM 0x10
+
+#define PTHREAD_DEFAULT_ATTR (PTHREAD_CANCEL_ENABLE)
+
+#define PTHREAD_CANCELED ((void *) (intptr_t) 0xDEADBEEF)
+
+#define _PTHREAD_NULL_THREAD ((pthread_t) 0)
+
+#define PTHREAD_ONCE_INIT 0
+
+#define PTHREAD_DESTRUCTOR_ITERATIONS 256
+#define PTHREAD_KEYS_MAX (1<<20)
+
+#define PTHREAD_MUTEX_NORMAL 0
+#define PTHREAD_MUTEX_ERRORCHECK 1
+#define PTHREAD_MUTEX_RECURSIVE 2
+#define PTHREAD_MUTEX_DEFAULT PTHREAD_MUTEX_NORMAL
+
+#define PTHREAD_MUTEX_SHARED 1
+#define PTHREAD_MUTEX_PRIVATE 0
+
+#define PTHREAD_PRIO_NONE 0
+#define PTHREAD_PRIO_INHERIT 8
+#define PTHREAD_PRIO_PROTECT 16
+#define PTHREAD_PRIO_MULT 32
+#define PTHREAD_PROCESS_SHARED 1
+#define PTHREAD_PROCESS_PRIVATE 0
+
+#define PTHREAD_MUTEX_FAST_NP		PTHREAD_MUTEX_NORMAL
+#define PTHREAD_MUTEX_TIMED_NP		PTHREAD_MUTEX_FAST_NP
+#define PTHREAD_MUTEX_ADAPTIVE_NP	PTHREAD_MUTEX_FAST_NP
+#define PTHREAD_MUTEX_ERRORCHECK_NP	PTHREAD_MUTEX_ERRORCHECK
+#define PTHREAD_MUTEX_RECURSIVE_NP	PTHREAD_MUTEX_RECURSIVE
+
+WINPTHREAD_API void * pthread_timechange_handler_np(void * dummy);
+WINPTHREAD_API int    pthread_delay32_np (const struct _timespec32 *interval);
+WINPTHREAD_API int    pthread_delay64_np (const struct _timespec64 *interval);
+WINPTHREAD_THREAD_DECL int    pthread_delay_np (const struct timespec *interval)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return pthread_delay32_np ((const struct _timespec32 *) interval);
+#else
+  return pthread_delay64_np ((const struct _timespec64 *) interval);
+#endif
+}
+WINPTHREAD_API int    pthread_num_processors_np(void);
+WINPTHREAD_API int    pthread_set_num_processors_np(int n);
+
+#define PTHREAD_BARRIER_SERIAL_THREAD 1
+
+/* maximum number of times a read lock may be obtained */
+#define	MAX_READ_LOCKS		(INT_MAX - 1)
+
+/* No fork() in windows - so ignore this */
+#define pthread_atfork(F1,F2,F3) 0
+
+/* unsupported stuff: */
+#define pthread_mutex_getprioceiling(M, P) ENOTSUP
+#define pthread_mutex_setprioceiling(M, P) ENOTSUP
+#define pthread_getcpuclockid(T, C) ENOTSUP
+#define pthread_attr_getguardsize(A, S) ENOTSUP
+#define pthread_attr_setguardsize(A, S) ENOTSUP
+
+typedef long pthread_once_t;
+typedef unsigned pthread_mutexattr_t;
+typedef unsigned pthread_key_t;
+typedef void *pthread_barrierattr_t;
+typedef int pthread_condattr_t;
+typedef int pthread_rwlockattr_t;
+typedef uintptr_t pthread_t;
+
+typedef struct _pthread_cleanup _pthread_cleanup;
+struct _pthread_cleanup
+{
+    void (*func)(void *);
+    void *arg;
+    _pthread_cleanup *next;
+};
+
+/* Using MemoryBarrier() requires including Windows headers. User code
+ * may want to use pthread_cleanup_push without including Windows headers
+ * first, thus prefer GCC specific intrinsics where possible. */
+#ifdef __GNUC__
+#define __pthread_MemoryBarrier() __sync_synchronize()
+#else
+#define __pthread_MemoryBarrier() MemoryBarrier()
+#endif
+
+#define pthread_cleanup_push(F, A)                                      \
+    do {                                                                \
+        const _pthread_cleanup _pthread_cup =                           \
+            { (F), (A), *pthread_getclean() };                          \
+        __pthread_MemoryBarrier();                                      \
+        *pthread_getclean() = (_pthread_cleanup *) &_pthread_cup;       \
+        __pthread_MemoryBarrier();                                      \
+        do {                                                            \
+            do {} while (0)
+
+/* Note that if async cancelling is used, then there is a race here */
+#define pthread_cleanup_pop(E)                                          \
+        } while (0);                                                    \
+        *pthread_getclean() = _pthread_cup.next;                        \
+        if ((E)) _pthread_cup.func((pthread_once_t *)_pthread_cup.arg); \
+    } while (0)
+
+typedef struct pthread_attr_t pthread_attr_t;
+struct pthread_attr_t
+{
+    unsigned p_state;
+    void *stack;
+    size_t s_size;
+    struct sched_param param;
+};
+
+WINPTHREAD_API int pthread_attr_setschedparam(pthread_attr_t *attr, const struct sched_param *param);
+WINPTHREAD_API int pthread_attr_getschedparam(const pthread_attr_t *attr, struct sched_param *param);
+WINPTHREAD_API int pthread_getschedparam(pthread_t thread, int *pol, struct sched_param *param);
+WINPTHREAD_API int pthread_setschedparam(pthread_t thread, int pol, const struct sched_param *param);
+WINPTHREAD_API int pthread_attr_setschedpolicy (pthread_attr_t *attr, int pol);
+WINPTHREAD_API int pthread_attr_getschedpolicy (const pthread_attr_t *attr, int *pol);
+
+/* synchronization objects */
+typedef intptr_t pthread_spinlock_t;
+typedef intptr_t pthread_mutex_t;
+typedef intptr_t pthread_cond_t;
+typedef intptr_t pthread_rwlock_t;
+typedef void	*pthread_barrier_t;
+
+#define PTHREAD_MUTEX_NORMAL 0
+#define PTHREAD_MUTEX_ERRORCHECK 1
+#define PTHREAD_MUTEX_RECURSIVE 2
+
+#define GENERIC_INITIALIZER				-1
+#define GENERIC_ERRORCHECK_INITIALIZER			-2
+#define GENERIC_RECURSIVE_INITIALIZER			-3
+#define GENERIC_NORMAL_INITIALIZER			-1
+#define PTHREAD_MUTEX_INITIALIZER			(pthread_mutex_t)GENERIC_INITIALIZER
+#define PTHREAD_RECURSIVE_MUTEX_INITIALIZER		(pthread_mutex_t)GENERIC_RECURSIVE_INITIALIZER
+#define PTHREAD_ERRORCHECK_MUTEX_INITIALIZER		(pthread_mutex_t)GENERIC_ERRORCHECK_INITIALIZER
+#define PTHREAD_NORMAL_MUTEX_INITIALIZER		(pthread_mutex_t)GENERIC_NORMAL_INITIALIZER
+#define PTHREAD_DEFAULT_MUTEX_INITIALIZER		PTHREAD_NORMAL_MUTEX_INITIALIZER
+#define PTHREAD_COND_INITIALIZER			(pthread_cond_t)GENERIC_INITIALIZER
+#define PTHREAD_RWLOCK_INITIALIZER			(pthread_rwlock_t)GENERIC_INITIALIZER
+#define PTHREAD_SPINLOCK_INITIALIZER			(pthread_spinlock_t)GENERIC_INITIALIZER
+
+WINPTHREAD_API extern void (**_pthread_key_dest)(void *);
+WINPTHREAD_API int         pthread_key_create(pthread_key_t *key, void (* dest)(void *));
+WINPTHREAD_API int         pthread_key_delete(pthread_key_t key);
+WINPTHREAD_API void *      pthread_getspecific(pthread_key_t key);
+WINPTHREAD_API int         pthread_setspecific(pthread_key_t key, const void *value);
+
+WINPTHREAD_API pthread_t pthread_self(void);
+WINPTHREAD_API int       pthread_once(pthread_once_t *o, void (*func)(void));
+WINPTHREAD_API void      pthread_testcancel(void);
+WINPTHREAD_API int       pthread_equal(pthread_t t1, pthread_t t2);
+WINPTHREAD_API void      pthread_tls_init(void);
+WINPTHREAD_API void      _pthread_cleanup_dest(pthread_t t);
+WINPTHREAD_API int       pthread_get_concurrency(int *val);
+WINPTHREAD_API int       pthread_set_concurrency(int val);
+WINPTHREAD_API void      pthread_exit(void *res);
+WINPTHREAD_API void      _pthread_invoke_cancel(void);
+WINPTHREAD_API int       pthread_cancel(pthread_t t);
+WINPTHREAD_API int       pthread_kill(pthread_t t, int sig);
+WINPTHREAD_API unsigned  _pthread_get_state(const pthread_attr_t *attr, unsigned flag);
+WINPTHREAD_API int       _pthread_set_state(pthread_attr_t *attr, unsigned flag, unsigned val);
+WINPTHREAD_API int       pthread_setcancelstate(int state, int *oldstate);
+WINPTHREAD_API int       pthread_setcanceltype(int type, int *oldtype);
+WINPTHREAD_API unsigned  __stdcall pthread_create_wrapper(void *args);
+WINPTHREAD_API int       pthread_create(pthread_t *th, const pthread_attr_t *attr, void *(* func)(void *), void *arg);
+WINPTHREAD_API int       pthread_join(pthread_t t, void **res);
+WINPTHREAD_API int       pthread_detach(pthread_t t);
+WINPTHREAD_API int       pthread_setname_np(pthread_t thread, const char *name);
+WINPTHREAD_API int       pthread_getname_np(pthread_t thread, char *name, size_t len);
+
+WINPTHREAD_API int pthread_rwlock_init(pthread_rwlock_t *rwlock_, const pthread_rwlockattr_t *attr);
+WINPTHREAD_API int pthread_rwlock_wrlock(pthread_rwlock_t *l);
+WINPTHREAD_API int pthread_rwlock_timedwrlock32(pthread_rwlock_t *rwlock, const struct _timespec32 *ts);
+WINPTHREAD_API int pthread_rwlock_timedwrlock64(pthread_rwlock_t *rwlock, const struct _timespec64 *ts);
+WINPTHREAD_RWLOCK_DECL int pthread_rwlock_timedwrlock(pthread_rwlock_t *rwlock, const struct timespec *ts)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return pthread_rwlock_timedwrlock32 (rwlock, (const struct _timespec32 *) ts);
+#else
+  return pthread_rwlock_timedwrlock64 (rwlock, (const struct _timespec64 *) ts);
+#endif
+}
+WINPTHREAD_API int pthread_rwlock_rdlock(pthread_rwlock_t *l);
+WINPTHREAD_API int pthread_rwlock_timedrdlock32(pthread_rwlock_t *l, const struct _timespec32 *ts);
+WINPTHREAD_API int pthread_rwlock_timedrdlock64(pthread_rwlock_t *l, const struct _timespec64 *ts);
+WINPTHREAD_RWLOCK_DECL int pthread_rwlock_timedrdlock(pthread_rwlock_t *l, const struct timespec *ts)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return pthread_rwlock_timedrdlock32 (l, (const struct _timespec32 *) ts);
+#else
+  return pthread_rwlock_timedrdlock64 (l, (const struct _timespec64 *) ts);
+#endif
+}
+WINPTHREAD_API int pthread_rwlock_unlock(pthread_rwlock_t *l);
+WINPTHREAD_API int pthread_rwlock_tryrdlock(pthread_rwlock_t *l);
+WINPTHREAD_API int pthread_rwlock_trywrlock(pthread_rwlock_t *l);
+WINPTHREAD_API int pthread_rwlock_destroy (pthread_rwlock_t *l);
+
+WINPTHREAD_API int pthread_cond_init(pthread_cond_t *cv, const pthread_condattr_t *a);
+WINPTHREAD_API int pthread_cond_destroy(pthread_cond_t *cv);
+WINPTHREAD_API int pthread_cond_signal (pthread_cond_t *cv);
+WINPTHREAD_API int pthread_cond_broadcast (pthread_cond_t *cv);
+WINPTHREAD_API int pthread_cond_wait (pthread_cond_t *cv, pthread_mutex_t *external_mutex);
+WINPTHREAD_API int pthread_cond_timedwait32(pthread_cond_t *cv, pthread_mutex_t *external_mutex, const struct _timespec32 *t);
+WINPTHREAD_API int pthread_cond_timedwait64(pthread_cond_t *cv, pthread_mutex_t *external_mutex, const struct _timespec64 *t);
+WINPTHREAD_COND_DECL int pthread_cond_timedwait(pthread_cond_t *cv, pthread_mutex_t *external_mutex, const struct timespec *t)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return pthread_cond_timedwait32 (cv, external_mutex, (const struct _timespec32 *) t);
+#else
+  return pthread_cond_timedwait64 (cv, external_mutex, (const struct _timespec64 *) t);
+#endif
+}
+WINPTHREAD_API int pthread_cond_timedwait32_relative_np(pthread_cond_t *cv, pthread_mutex_t *external_mutex, const struct _timespec32 *t);
+WINPTHREAD_API int pthread_cond_timedwait64_relative_np(pthread_cond_t *cv, pthread_mutex_t *external_mutex, const struct _timespec64 *t);
+WINPTHREAD_COND_DECL int pthread_cond_timedwait_relative_np(pthread_cond_t *cv, pthread_mutex_t *external_mutex, const struct timespec *t)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return pthread_cond_timedwait32_relative_np (cv, external_mutex, (const struct _timespec32 *) t);
+#else
+  return pthread_cond_timedwait64_relative_np (cv, external_mutex, (const struct _timespec64 *) t);
+#endif
+}
+
+WINPTHREAD_API int pthread_mutex_lock(pthread_mutex_t *m);
+WINPTHREAD_API int pthread_mutex_timedlock32(pthread_mutex_t *m, const struct _timespec32 *ts);
+WINPTHREAD_API int pthread_mutex_timedlock64(pthread_mutex_t *m, const struct _timespec64 *ts);
+WINPTHREAD_MUTEX_DECL int pthread_mutex_timedlock(pthread_mutex_t *m, const struct timespec *ts)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return pthread_mutex_timedlock32 (m, (const struct _timespec32 *) ts);
+#else
+  return pthread_mutex_timedlock64 (m, (const struct _timespec64 *) ts);
+#endif
+}
+WINPTHREAD_API int pthread_mutex_unlock(pthread_mutex_t *m);
+WINPTHREAD_API int pthread_mutex_trylock(pthread_mutex_t *m);
+WINPTHREAD_API int pthread_mutex_init(pthread_mutex_t *m, const pthread_mutexattr_t *a);
+WINPTHREAD_API int pthread_mutex_destroy(pthread_mutex_t *m);
+
+WINPTHREAD_API int pthread_barrier_destroy(pthread_barrier_t *b);
+WINPTHREAD_API int pthread_barrier_init(pthread_barrier_t *b, const void *attr, unsigned int count);
+WINPTHREAD_API int pthread_barrier_wait(pthread_barrier_t *b);
+
+WINPTHREAD_API int pthread_spin_init(pthread_spinlock_t *l, int pshared);
+WINPTHREAD_API int pthread_spin_destroy(pthread_spinlock_t *l);
+/* No-fair spinlock due to lack of knowledge of thread number.  */
+WINPTHREAD_API int pthread_spin_lock(pthread_spinlock_t *l);
+WINPTHREAD_API int pthread_spin_trylock(pthread_spinlock_t *l);
+WINPTHREAD_API int pthread_spin_unlock(pthread_spinlock_t *l);
+
+WINPTHREAD_API int pthread_attr_init(pthread_attr_t *attr);
+WINPTHREAD_API int pthread_attr_destroy(pthread_attr_t *attr);
+WINPTHREAD_API int pthread_attr_setdetachstate(pthread_attr_t *a, int flag);
+WINPTHREAD_API int pthread_attr_getdetachstate(const pthread_attr_t *a, int *flag);
+WINPTHREAD_API int pthread_attr_setinheritsched(pthread_attr_t *a, int flag);
+WINPTHREAD_API int pthread_attr_getinheritsched(const pthread_attr_t *a, int *flag);
+WINPTHREAD_API int pthread_attr_setscope(pthread_attr_t *a, int flag);
+WINPTHREAD_API int pthread_attr_getscope(const pthread_attr_t *a, int *flag);
+WINPTHREAD_API int pthread_attr_getstack(const pthread_attr_t *attr, void **stack, size_t *size);
+WINPTHREAD_API int pthread_attr_setstack(pthread_attr_t *attr, void *stack, size_t size);
+WINPTHREAD_API int pthread_attr_getstackaddr(const pthread_attr_t *attr, void **stack);
+WINPTHREAD_API int pthread_attr_setstackaddr(pthread_attr_t *attr, void *stack);
+WINPTHREAD_API int pthread_attr_getstacksize(const pthread_attr_t *attr, size_t *size);
+WINPTHREAD_API int pthread_attr_setstacksize(pthread_attr_t *attr, size_t size);
+
+WINPTHREAD_API int pthread_mutexattr_init(pthread_mutexattr_t *a);
+WINPTHREAD_API int pthread_mutexattr_destroy(pthread_mutexattr_t *a);
+WINPTHREAD_API int pthread_mutexattr_gettype(const pthread_mutexattr_t *a, int *type);
+WINPTHREAD_API int pthread_mutexattr_settype(pthread_mutexattr_t *a, int type);
+WINPTHREAD_API int pthread_mutexattr_getpshared(const pthread_mutexattr_t *a, int *type);
+WINPTHREAD_API int pthread_mutexattr_setpshared(pthread_mutexattr_t * a, int type);
+WINPTHREAD_API int pthread_mutexattr_getprotocol(const pthread_mutexattr_t *a, int *type);
+WINPTHREAD_API int pthread_mutexattr_setprotocol(pthread_mutexattr_t *a, int type);
+WINPTHREAD_API int pthread_mutexattr_getprioceiling(const pthread_mutexattr_t *a, int * prio);
+WINPTHREAD_API int pthread_mutexattr_setprioceiling(pthread_mutexattr_t *a, int prio);
+WINPTHREAD_API int pthread_getconcurrency(void);
+WINPTHREAD_API int pthread_setconcurrency(int new_level);
+
+WINPTHREAD_API int pthread_condattr_destroy(pthread_condattr_t *a);
+WINPTHREAD_API int pthread_condattr_init(pthread_condattr_t *a);
+WINPTHREAD_API int pthread_condattr_getpshared(const pthread_condattr_t *a, int *s);
+WINPTHREAD_API int pthread_condattr_setpshared(pthread_condattr_t *a, int s);
+
+WINPTHREAD_API int pthread_condattr_getclock (const pthread_condattr_t *attr,
+       clockid_t *clock_id);
+WINPTHREAD_API int pthread_condattr_setclock(pthread_condattr_t *attr,
+       clockid_t clock_id);
+
+WINPTHREAD_API int pthread_barrierattr_init(void **attr);
+WINPTHREAD_API int pthread_barrierattr_destroy(void **attr);
+WINPTHREAD_API int pthread_barrierattr_setpshared(void **attr, int s);
+WINPTHREAD_API int pthread_barrierattr_getpshared(void **attr, int *s);
+
+/* Private extensions for analysis and internal use.  */
+WINPTHREAD_API struct _pthread_cleanup ** pthread_getclean (void);
+WINPTHREAD_API void * pthread_gethandle (pthread_t t);
+WINPTHREAD_API void * pthread_getevent (void);
+
+WINPTHREAD_API int _pthread_tryjoin (pthread_t t, void **res);
+WINPTHREAD_API int pthread_rwlockattr_destroy(pthread_rwlockattr_t *a);
+WINPTHREAD_API int pthread_rwlockattr_getpshared(pthread_rwlockattr_t *a, int *s);
+WINPTHREAD_API int pthread_rwlockattr_init(pthread_rwlockattr_t *a);
+WINPTHREAD_API int pthread_rwlockattr_setpshared(pthread_rwlockattr_t *a, int s);
+
+#ifndef SIG_BLOCK
+#define SIG_BLOCK 0
+#endif
+#ifndef SIG_UNBLOCK
+#define SIG_UNBLOCK 1
+#endif
+#ifndef SIG_SETMASK
+#define SIG_SETMASK 2
+#endif
+
+#include <pthread_unistd.h>
+
+#undef _POSIX_THREAD_DESTRUCTOR_ITERATIONS
+#define _POSIX_THREAD_DESTRUCTOR_ITERATIONS     PTHREAD_DESTRUCTOR_ITERATIONS
+
+#undef _POSIX_THREAD_KEYS_MAX
+#define _POSIX_THREAD_KEYS_MAX                  PTHREAD_KEYS_MAX
+
+#undef PTHREAD_THREADS_MAX
+#define PTHREAD_THREADS_MAX                     2019
+
+#undef _POSIX_SEM_NSEMS_MAX
+#define _POSIX_SEM_NSEMS_MAX                    256
+
+#undef SEM_NSEMS_MAX
+#define SEM_NSEMS_MAX                           1024
+
+/* Wrap cancellation points.  */
+#if defined(__WINPTHREAD_ENABLE_WRAP_API) \
+    || defined(__WINPTRHEAD_ENABLE_WRAP_API) /* historical typo */
+#define accept(...) (pthread_testcancel(), accept(__VA_ARGS__))
+#define aio_suspend(...) (pthread_testcancel(), aio_suspend(__VA_ARGS__))
+#define clock_nanosleep(...) (pthread_testcancel(), clock_nanosleep(__VA_ARGS__))
+#define close(...) (pthread_testcancel(), close(__VA_ARGS__))
+#define connect(...) (pthread_testcancel(), connect(__VA_ARGS__))
+#define creat(...) (pthread_testcancel(), creat(__VA_ARGS__))
+#define fcntl(...) (pthread_testcancel(), fcntl(__VA_ARGS__))
+#define fdatasync(...) (pthread_testcancel(), fdatasync(__VA_ARGS__))
+#define fsync(...) (pthread_testcancel(), fsync(__VA_ARGS__))
+#define getmsg(...) (pthread_testcancel(), getmsg(__VA_ARGS__))
+#define getpmsg(...) (pthread_testcancel(), getpmsg(__VA_ARGS__))
+#define lockf(...) (pthread_testcancel(), lockf(__VA_ARGS__))
+#define mg_receive(...) (pthread_testcancel(), mg_receive(__VA_ARGS__))
+#define mg_send(...) (pthread_testcancel(), mg_send(__VA_ARGS__))
+#define mg_timedreceive(...) (pthread_testcancel(), mg_timedreceive(__VA_ARGS__))
+#define mg_timessend(...) (pthread_testcancel(), mg_timedsend(__VA_ARGS__))
+#define msgrcv(...) (pthread_testcancel(), msgrecv(__VA_ARGS__))
+#define msgsnd(...) (pthread_testcancel(), msgsnd(__VA_ARGS__))
+#define msync(...) (pthread_testcancel(), msync(__VA_ARGS__))
+#define nanosleep(...) (pthread_testcancel(), nanosleep(__VA_ARGS__))
+#define open(...) (pthread_testcancel(), open(__VA_ARGS__))
+#define pause(...) (pthread_testcancel(), pause(__VA_ARGS__))
+#define poll(...) (pthread_testcancel(), poll(__VA_ARGS__))
+#define pread(...) (pthread_testcancel(), pread(__VA_ARGS__))
+#define pselect(...) (pthread_testcancel(), pselect(__VA_ARGS__))
+#define putmsg(...) (pthread_testcancel(), putmsg(__VA_ARGS__))
+#define putpmsg(...) (pthread_testcancel(), putpmsg(__VA_ARGS__))
+#define pwrite(...) (pthread_testcancel(), pwrite(__VA_ARGS__))
+#define read(...) (pthread_testcancel(), read(__VA_ARGS__))
+#define readv(...) (pthread_testcancel(), readv(__VA_ARGS__))
+#define recv(...) (pthread_testcancel(), recv(__VA_ARGS__))
+#define recvfrom(...) (pthread_testcancel(), recvfrom(__VA_ARGS__))
+#define recvmsg(...) (pthread_testcancel(), recvmsg(__VA_ARGS__))
+#define select(...) (pthread_testcancel(), select(__VA_ARGS__))
+#define sem_timedwait(...) (pthread_testcancel(), sem_timedwait(__VA_ARGS__))
+#define sem_wait(...) (pthread_testcancel(), sem_wait(__VA_ARGS__))
+#define send(...) (pthread_testcancel(), send(__VA_ARGS__))
+#define sendmsg(...) (pthread_testcancel(), sendmsg(__VA_ARGS__))
+#define sendto(...) (pthread_testcancel(), sendto(__VA_ARGS__))
+#define sigpause(...) (pthread_testcancel(), sigpause(__VA_ARGS__))
+#define sigsuspend(...) (pthread_testcancel(), sigsuspend(__VA_ARGS__))
+#define sigwait(...) (pthread_testcancel(), sigwait(__VA_ARGS__))
+#define sigwaitinfo(...) (pthread_testcancel(), sigwaitinfo(__VA_ARGS__))
+#define sleep(...) (pthread_testcancel(), sleep(__VA_ARGS__))
+//#define Sleep(...) (pthread_testcancel(), Sleep(__VA_ARGS__))
+#define system(...) (pthread_testcancel(), system(__VA_ARGS__))
+#define access(...) (pthread_testcancel(), access(__VA_ARGS__))
+#define asctime(...) (pthread_testcancel(), asctime(__VA_ARGS__))
+#define catclose(...) (pthread_testcancel(), catclose(__VA_ARGS__))
+#define catgets(...) (pthread_testcancel(), catgets(__VA_ARGS__))
+#define catopen(...) (pthread_testcancel(), catopen(__VA_ARGS__))
+#define closedir(...) (pthread_testcancel(), closedir(__VA_ARGS__))
+#define closelog(...) (pthread_testcancel(), closelog(__VA_ARGS__))
+#define ctermid(...) (pthread_testcancel(), ctermid(__VA_ARGS__))
+#define ctime(...) (pthread_testcancel(), ctime(__VA_ARGS__))
+#define dbm_close(...) (pthread_testcancel(), dbm_close(__VA_ARGS__))
+#define dbm_delete(...) (pthread_testcancel(), dbm_delete(__VA_ARGS__))
+#define dbm_fetch(...) (pthread_testcancel(), dbm_fetch(__VA_ARGS__))
+#define dbm_nextkey(...) (pthread_testcancel(), dbm_nextkey(__VA_ARGS__))
+#define dbm_open(...) (pthread_testcancel(), dbm_open(__VA_ARGS__))
+#define dbm_store(...) (pthread_testcancel(), dbm_store(__VA_ARGS__))
+#define dlclose(...) (pthread_testcancel(), dlclose(__VA_ARGS__))
+#define dlopen(...) (pthread_testcancel(), dlopen(__VA_ARGS__))
+#define endgrent(...) (pthread_testcancel(), endgrent(__VA_ARGS__))
+#define endhostent(...) (pthread_testcancel(), endhostent(__VA_ARGS__))
+#define endnetent(...) (pthread_testcancel(), endnetent(__VA_ARGS__))
+#define endprotoent(...) (pthread_testcancel(), endprotoend(__VA_ARGS__))
+#define endpwent(...) (pthread_testcancel(), endpwent(__VA_ARGS__))
+#define endservent(...) (pthread_testcancel(), endservent(__VA_ARGS__))
+#define endutxent(...) (pthread_testcancel(), endutxent(__VA_ARGS__))
+#define fclose(...) (pthread_testcancel(), fclose(__VA_ARGS__))
+#define fflush(...) (pthread_testcancel(), fflush(__VA_ARGS__))
+#define fgetc(...) (pthread_testcancel(), fgetc(__VA_ARGS__))
+#define fgetpos(...) (pthread_testcancel(), fgetpos(__VA_ARGS__))
+#define fgets(...) (pthread_testcancel(), fgets(__VA_ARGS__))
+#define fgetwc(...) (pthread_testcancel(), fgetwc(__VA_ARGS__))
+#define fgetws(...) (pthread_testcancel(), fgetws(__VA_ARGS__))
+#define fmtmsg(...) (pthread_testcancel(), fmtmsg(__VA_ARGS__))
+#define fopen(...) (pthread_testcancel(), fopen(__VA_ARGS__))
+#define fpathconf(...) (pthread_testcancel(), fpathconf(__VA_ARGS__))
+#define fprintf(...) (pthread_testcancel(), fprintf(__VA_ARGS__))
+#define fputc(...) (pthread_testcancel(), fputc(__VA_ARGS__))
+#define fputs(...) (pthread_testcancel(), fputs(__VA_ARGS__))
+#define fputwc(...) (pthread_testcancel(), fputwc(__VA_ARGS__))
+#define fputws(...) (pthread_testcancel(), fputws(__VA_ARGS__))
+#define fread(...) (pthread_testcancel(), fread(__VA_ARGS__))
+#define freopen(...) (pthread_testcancel(), freopen(__VA_ARGS__))
+#define fscanf(...) (pthread_testcancel(), fscanf(__VA_ARGS__))
+#define fseek(...) (pthread_testcancel(), fseek(__VA_ARGS__))
+#define fseeko(...) (pthread_testcancel(), fseeko(__VA_ARGS__))
+#define fsetpos(...) (pthread_testcancel(), fsetpos(__VA_ARGS__))
+#define fstat(...) (pthread_testcancel(), fstat(__VA_ARGS__))
+#define ftell(...) (pthread_testcancel(), ftell(__VA_ARGS__))
+#define ftello(...) (pthread_testcancel(), ftello(__VA_ARGS__))
+#define ftw(...) (pthread_testcancel(), ftw(__VA_ARGS__))
+#define fwprintf(...) (pthread_testcancel(), fwprintf(__VA_ARGS__))
+#define fwrite(...) (pthread_testcancel(), fwrite(__VA_ARGS__))
+#define fwscanf(...) (pthread_testcancel(), fwscanf(__VA_ARGS__))
+#define getaddrinfo(...) (pthread_testcancel(), getaddrinfo(__VA_ARGS__))
+#define getc(...) (pthread_testcancel(), getc(__VA_ARGS__))
+#define getc_unlocked(...) (pthread_testcancel(), getc_unlocked(__VA_ARGS__))
+#define getchar(...) (pthread_testcancel(), getchar(__VA_ARGS__))
+#define getchar_unlocked(...) (pthread_testcancel(), getchar_unlocked(__VA_ARGS__))
+#define getcwd(...) (pthread_testcancel(), getcwd(__VA_ARGS__))
+#define getdate(...) (pthread_testcancel(), getdate(__VA_ARGS__))
+#define getgrent(...) (pthread_testcancel(), getgrent(__VA_ARGS__))
+#define getgrgid(...) (pthread_testcancel(), getgrgid(__VA_ARGS__))
+#define getgrgid_r(...) (pthread_testcancel(), getgrgid_r(__VA_ARGS__))
+#define gergrnam(...) (pthread_testcancel(), getgrnam(__VA_ARGS__))
+#define getgrnam_r(...) (pthread_testcancel(), getgrnam_r(__VA_ARGS__))
+#define gethostbyaddr(...) (pthread_testcancel(), gethostbyaddr(__VA_ARGS__))
+#define gethostbyname(...) (pthread_testcancel(), gethostbyname(__VA_ARGS__))
+#define gethostent(...) (pthread_testcancel(), gethostent(__VA_ARGS__))
+#define gethostid(...) (pthread_testcancel(), gethostid(__VA_ARGS__))
+#define gethostname(...) (pthread_testcancel(), gethostname(__VA_ARGS__))
+#define getlogin(...) (pthread_testcancel(), getlogin(__VA_ARGS__))
+#define getlogin_r(...) (pthread_testcancel(), getlogin_r(__VA_ARGS__))
+#define getnameinfo(...) (pthread_testcancel(), getnameinfo(__VA_ARGS__))
+#define getnetbyaddr(...) (pthread_testcancel(), getnetbyaddr(__VA_ARGS__))
+#define getnetbyname(...) (pthread_testcancel(), getnetbyname(__VA_ARGS__))
+#define getnetent(...) (pthread_testcancel(), getnetent(__VA_ARGS__))
+#define getopt(...) (pthread_testcancel(), getopt(__VA_ARGS__))
+#define getprotobyname(...) (pthread_testcancel(), getprotobyname(__VA_ARGS__))
+#define getprotobynumber(...) (pthread_testcancel(), getprotobynumber(__VA_ARGS__))
+#define getprotoent(...) (pthread_testcancel(), getprotoent(__VA_ARGS__))
+#define getpwent(...) (pthread_testcancel(), getpwent(__VA_ARGS__))
+#define getpwnam(...) (pthread_testcancel(), getpwnam(__VA_ARGS__))
+#define getpwnam_r(...) (pthread_testcancel(), getpwnam_r(__VA_ARGS__))
+#define getpwuid(...) (pthread_testcancel(), getpwuid(__VA_ARGS__))
+#define getpwuid_r(...) (pthread_testcancel(), getpwuid_r(__VA_ARGS__))
+#define gets(...) (pthread_testcancel(), gets(__VA_ARGS__))
+#define getservbyname(...) (pthread_testcancel(), getservbyname(__VA_ARGS__))
+#define getservbyport(...) (pthread_testcancel(), getservbyport(__VA_ARGS__))
+#define getservent(...) (pthread_testcancel(), getservent(__VA_ARGS__))
+#define getutxent(...) (pthread_testcancel(), getutxent(__VA_ARGS__))
+#define getutxid(...) (pthread_testcancel(), getutxid(__VA_ARGS__))
+#define getutxline(...) (pthread_testcancel(), getutxline(__VA_ARGS__))
+#undef getwc
+#define getwc(...) (pthread_testcancel(), getwc(__VA_ARGS__))
+#undef getwchar
+#define getwchar(...) (pthread_testcancel(), getwchar(__VA_ARGS__))
+#define getwd(...) (pthread_testcancel(), getwd(__VA_ARGS__))
+#define glob(...) (pthread_testcancel(), glob(__VA_ARGS__))
+#define iconv_close(...) (pthread_testcancel(), iconv_close(__VA_ARGS__))
+#define iconv_open(...) (pthread_testcancel(), iconv_open(__VA_ARGS__))
+#define ioctl(...) (pthread_testcancel(), ioctl(__VA_ARGS__))
+#define link(...) (pthread_testcancel(), link(__VA_ARGS__))
+#define localtime(...) (pthread_testcancel(), localtime(__VA_ARGS__))
+#define lseek(...) (pthread_testcancel(), lseek(__VA_ARGS__))
+#define lstat(...) (pthread_testcancel(), lstat(__VA_ARGS__))
+#define mkstemp(...) (pthread_testcancel(), mkstemp(__VA_ARGS__))
+#define nftw(...) (pthread_testcancel(), nftw(__VA_ARGS__))
+#define opendir(...) (pthread_testcancel(), opendir(__VA_ARGS__))
+#define openlog(...) (pthread_testcancel(), openlog(__VA_ARGS__))
+#define pathconf(...) (pthread_testcancel(), pathconf(__VA_ARGS__))
+#define pclose(...) (pthread_testcancel(), pclose(__VA_ARGS__))
+#define perror(...) (pthread_testcancel(), perror(__VA_ARGS__))
+#define popen(...) (pthread_testcancel(), popen(__VA_ARGS__))
+#define posix_fadvise(...) (pthread_testcancel(), posix_fadvise(__VA_ARGS__))
+#define posix_fallocate(...) (pthread_testcancel(), posix_fallocate(__VA_ARGS__))
+#define posix_madvise(...) (pthread_testcancel(), posix_madvise(__VA_ARGS__))
+#define posix_openpt(...) (pthread_testcancel(), posix_openpt(__VA_ARGS__))
+#define posix_spawn(...) (pthread_testcancel(), posix_spawn(__VA_ARGS__))
+#define posix_spawnp(...) (pthread_testcancel(), posix_spawnp(__VA_ARGS__))
+#define posix_trace_clear(...) (pthread_testcancel(), posix_trace_clear(__VA_ARGS__))
+#define posix_trace_close(...) (pthread_testcancel(), posix_trace_close(__VA_ARGS__))
+#define posix_trace_create(...) (pthread_testcancel(), posix_trace_create(__VA_ARGS__))
+#define posix_trace_create_withlog(...) (pthread_testcancel(), posix_trace_create_withlog(__VA_ARGS__))
+#define posix_trace_eventtypelist_getne(...) (pthread_testcancel(), posix_trace_eventtypelist_getne(__VA_ARGS__))
+#define posix_trace_eventtypelist_rewin(...) (pthread_testcancel(), posix_trace_eventtypelist_rewin(__VA_ARGS__))
+#define posix_trace_flush(...) (pthread_testcancel(), posix_trace_flush(__VA_ARGS__))
+#define posix_trace_get_attr(...) (pthread_testcancel(), posix_trace_get_attr(__VA_ARGS__))
+#define posix_trace_get_filter(...) (pthread_testcancel(), posix_trace_get_filter(__VA_ARGS__))
+#define posix_trace_get_status(...) (pthread_testcancel(), posix_trace_get_status(__VA_ARGS__))
+#define posix_trace_getnext_event(...) (pthread_testcancel(), posix_trace_getnext_event(__VA_ARGS__))
+#define posix_trace_open(...) (pthread_testcancel(), posix_trace_open(__VA_ARGS__))
+#define posix_trace_rewind(...) (pthread_testcancel(), posix_trace_rewind(__VA_ARGS__))
+#define posix_trace_setfilter(...) (pthread_testcancel(), posix_trace_setfilter(__VA_ARGS__))
+#define posix_trace_shutdown(...) (pthread_testcancel(), posix_trace_shutdown(__VA_ARGS__))
+#define posix_trace_timedgetnext_event(...) (pthread_testcancel(), posix_trace_timedgetnext_event(__VA_ARGS__))
+#define posix_typed_mem_open(...) (pthread_testcancel(), posix_typed_mem_open(__VA_ARGS__))
+#define printf(...) (pthread_testcancel(), printf(__VA_ARGS__))
+#define putc(...) (pthread_testcancel(), putc(__VA_ARGS__))
+#define putc_unlocked(...) (pthread_testcancel(), putc_unlocked(__VA_ARGS__))
+#define putchar(...) (pthread_testcancel(), putchar(__VA_ARGS__))
+#define putchar_unlocked(...) (pthread_testcancel(), putchar_unlocked(__VA_ARGS__))
+#define puts(...) (pthread_testcancel(), puts(__VA_ARGS__))
+#define pututxline(...) (pthread_testcancel(), pututxline(__VA_ARGS__))
+#undef putwc
+#define putwc(...) (pthread_testcancel(), putwc(__VA_ARGS__))
+#undef putwchar
+#define putwchar(...) (pthread_testcancel(), putwchar(__VA_ARGS__))
+#define readdir(...) (pthread_testcancel(), readdir(__VA_ARSG__))
+#define readdir_r(...) (pthread_testcancel(), readdir_r(__VA_ARGS__))
+#define remove(...) (pthread_testcancel(), remove(__VA_ARGS__))
+#define rename(...) (pthread_testcancel(), rename(__VA_ARGS__))
+#define rewind(...) (pthread_testcancel(), rewind(__VA_ARGS__))
+#define rewinddir(...) (pthread_testcancel(), rewinddir(__VA_ARGS__))
+#define scanf(...) (pthread_testcancel(), scanf(__VA_ARGS__))
+#define seekdir(...) (pthread_testcancel(), seekdir(__VA_ARGS__))
+#define semop(...) (pthread_testcancel(), semop(__VA_ARGS__))
+#define setgrent(...) (pthread_testcancel(), setgrent(__VA_ARGS__))
+#define sethostent(...) (pthread_testcancel(), sethostemt(__VA_ARGS__))
+#define setnetent(...) (pthread_testcancel(), setnetent(__VA_ARGS__))
+#define setprotoent(...) (pthread_testcancel(), setprotoent(__VA_ARGS__))
+#define setpwent(...) (pthread_testcancel(), setpwent(__VA_ARGS__))
+#define setservent(...) (pthread_testcancel(), setservent(__VA_ARGS__))
+#define setutxent(...) (pthread_testcancel(), setutxent(__VA_ARGS__))
+#define stat(...) (pthread_testcancel(), stat(__VA_ARGS__))
+#define strerror(...) (pthread_testcancel(), strerror(__VA_ARGS__))
+#define strerror_r(...) (pthread_testcancel(), strerror_r(__VA_ARGS__))
+#define strftime(...) (pthread_testcancel(), strftime(__VA_ARGS__))
+#define symlink(...) (pthread_testcancel(), symlink(__VA_ARGS__))
+#define sync(...) (pthread_testcancel(), sync(__VA_ARGS__))
+#define syslog(...) (pthread_testcancel(), syslog(__VA_ARGS__))
+#define tmpfile(...) (pthread_testcancel(), tmpfile(__VA_ARGS__))
+#define tmpnam(...) (pthread_testcancel(), tmpnam(__VA_ARGS__))
+#define ttyname(...) (pthread_testcancel(), ttyname(__VA_ARGS__))
+#define ttyname_r(...) (pthread_testcancel(), ttyname_r(__VA_ARGS__))
+#define tzset(...) (pthread_testcancel(), tzset(__VA_ARGS__))
+#define ungetc(...) (pthread_testcancel(), ungetc(__VA_ARGS__))
+#define ungetwc(...) (pthread_testcancel(), ungetwc(__VA_ARGS__))
+#define unlink(...) (pthread_testcancel(), unlink(__VA_ARGS__))
+#define vfprintf(...) (pthread_testcancel(), vfprintf(__VA_ARGS__))
+#define vfwprintf(...) (pthread_testcancel(), vfwprintf(__VA_ARGS__))
+#define vprintf(...) (pthread_testcancel(), vprintf(__VA_ARGS__))
+#define vwprintf(...) (pthread_testcancel(), vwprintf(__VA_ARGS__))
+#define wcsftime(...) (pthread_testcancel(), wcsftime(__VA_ARGS__))
+#define wordexp(...) (pthread_testcancel(), wordexp(__VA_ARGS__))
+#define wprintf(...) (pthread_testcancel(), wprintf(__VA_ARGS__))
+#define wscanf(...) (pthread_testcancel(), wscanf(__VA_ARGS__))
+#endif
+
+/* We deal here with a gcc issue for posix threading on Windows.
+   We would need to change here gcc's gthr-posix.h header, but this
+   got rejected.  So we deal it within this header.  */
+#ifdef _GTHREAD_USE_MUTEX_INIT_FUNC
+#undef _GTHREAD_USE_MUTEX_INIT_FUNC
+#endif
+#define _GTHREAD_USE_MUTEX_INIT_FUNC 1
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* WIN_PTHREADS_H */
diff --git a/vendor/winpthreads/include/pthread_compat.h b/vendor/winpthreads/include/pthread_compat.h
new file mode 100644
index 00000000..81600559
--- /dev/null
+++ b/vendor/winpthreads/include/pthread_compat.h
@@ -0,0 +1,204 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+/*
+ * Parts of this library are derived by:
+ *
+ * Posix Threads library for Microsoft Windows
+ *
+ * Use at own risk, there is no implied warranty to this code.
+ * It uses undocumented features of Microsoft Windows that can change
+ * at any time in the future.
+ *
+ * (C) 2010 Lockless Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification,
+ * are permitted provided that the following conditions are met:
+ *
+ *
+ *  * Redistributions of source code must retain the above copyright notice,
+ *    this list of conditions and the following disclaimer.
+ *  * 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.
+ *  * Neither the name of Lockless Inc. nor the names of its contributors may be
+ *    used to endorse or promote products derived from this software without
+ *    specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AN
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
+ * IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
+ * INDIRECT, INCIDENTAL, SPECIAL, 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 OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef WIN_PTHREADS_PTHREAD_COMPAT_H
+#define WIN_PTHREADS_PTHREAD_COMPAT_H
+
+#if defined(__cplusplus) && __cplusplus >= 201103L
+#define WINPTHREADS_STATIC_ASSERT(expr, msg) static_assert ((expr), msg)
+#elif defined(__STDC_VERSION__) && __STDC_VERSION__ >= 202311L
+#define WINPTHREADS_STATIC_ASSERT(expr, msg) static_assert ((expr), msg)
+#elif defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L
+#define WINPTHREADS_STATIC_ASSERT(expr, msg) _Static_assert ((expr), msg)
+#else
+#define WINPTHREADS_STATIC_ASSERT(expr, msg) extern int winpthreads_static_assert[((expr) ? 1 : -1)]
+#endif
+
+#if defined(_USE_32BIT_TIME_T)
+#define WINPTHREADS_TIME_BITS 32
+#else
+#define WINPTHREADS_TIME_BITS 64
+#endif
+
+#ifndef WINPTHREAD_API
+  /**
+   * Allow `WINPTHREAD_STATIC` to override `WINPTHREADS_USE_DLLIMPORT`.
+   */
+# if defined(WINPTHREADS_USE_DLLIMPORT) && !defined(WINPTHREAD_STATIC)
+#  define WINPTHREAD_API  __declspec(dllimport)
+# else
+#  define WINPTHREAD_API
+# endif
+#endif
+
+#ifndef __clockid_t_defined
+typedef int clockid_t;
+#define __clockid_t_defined 1
+#endif  /* __clockid_t_defined */
+
+#ifndef _MODE_T_
+#define	_MODE_T_
+typedef unsigned short mode_t;
+#endif
+
+/* Error-codes.  */
+#ifndef ETIMEDOUT
+#define ETIMEDOUT	138
+#endif
+#ifndef ENOTSUP
+#define ENOTSUP		129
+#endif
+#ifndef EWOULDBLOCK
+#define EWOULDBLOCK	140
+#endif
+
+#ifdef __GNUC__
+
+#define WINPTHREADS_INLINE __inline__
+#define WINPTHREADS_ALWAYS_INLINE __inline__ __attribute__((__always_inline__))
+#define WINPTHREADS_ATTRIBUTE(X) __attribute__(X)
+#define WINPTHREADS_SECTION(X) __section__(X)
+
+#elif _MSC_VER
+
+/**
+ * If package which includes this header file is using autoconf and calls
+ * AC_TYPE_PID_T macro, the check for pid_t will fail and it will define pid_t
+ * as a macro in config.h - this eventually results in compilation error.
+ *
+ * Luckily, it defines it to the same base type, so we can simply undefine it.
+ */
+#ifdef _WIN64
+#ifdef pid_t
+WINPTHREADS_STATIC_ASSERT (sizeof (pid_t) == sizeof(__int64), "pid_t is defined as a macro with mismatching base type");
+#undef pid_t
+#endif
+typedef __int64 pid_t;
+#else
+#ifdef pid_t
+WINPTHREADS_STATIC_ASSERT (sizeof (pid_t) == sizeof(int), "pid_t is defined as a macro with mismatching base type");
+#undef pid_t
+#endif
+typedef int     pid_t;
+#endif
+
+#define WINPTHREADS_INLINE __inline
+#define WINPTHREADS_ALWAYS_INLINE __inline __forceinline
+#define WINPTHREADS_ATTRIBUTE(X) __declspec X
+#define WINPTHREADS_SECTION(X) allocate(X)
+
+#endif
+
+#ifndef WINPTHREAD_CLOCK_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_CLOCK_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_CLOCK_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#ifndef WINPTHREAD_COND_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_COND_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_COND_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#ifndef WINPTHREAD_MUTEX_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_MUTEX_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_MUTEX_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#ifndef WINPTHREAD_NANOSLEEP_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_NANOSLEEP_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_NANOSLEEP_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#ifndef WINPTHREAD_RWLOCK_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_RWLOCK_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_RWLOCK_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#ifndef WINPTHREAD_SEM_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_SEM_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_SEM_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#ifndef WINPTHREAD_THREAD_DECL
+# ifdef __cplusplus
+#  define WINPTHREAD_THREAD_DECL WINPTHREADS_ALWAYS_INLINE
+# else
+#  define WINPTHREAD_THREAD_DECL static WINPTHREADS_ALWAYS_INLINE
+# endif
+#endif
+
+#endif
diff --git a/vendor/winpthreads/include/pthread_signal.h b/vendor/winpthreads/include/pthread_signal.h
new file mode 100644
index 00000000..42a50ae0
--- /dev/null
+++ b/vendor/winpthreads/include/pthread_signal.h
@@ -0,0 +1,29 @@
+/*
+   Copyright (c) 2013-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_SIGNAL_H
+#define WIN_PTHREADS_SIGNAL_H
+
+/* Windows has rudimentary signals support.  */
+#define pthread_sigmask(H, S1, S2) 0
+
+#endif /* WIN_PTHREADS_SIGNAL_H */
diff --git a/vendor/winpthreads/include/pthread_time.h b/vendor/winpthreads/include/pthread_time.h
new file mode 100644
index 00000000..27f7566f
--- /dev/null
+++ b/vendor/winpthreads/include/pthread_time.h
@@ -0,0 +1,136 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_TIME_H
+#define WIN_PTHREADS_TIME_H
+
+#include <sys/timeb.h>
+#include "pthread_compat.h"
+
+/* Posix timers are supported */
+#ifndef _POSIX_TIMERS
+#define _POSIX_TIMERS           200809L
+#endif
+
+/* Monotonic clocks are available.  */
+#ifndef _POSIX_MONOTONIC_CLOCK
+#define _POSIX_MONOTONIC_CLOCK  200809L
+#endif
+
+/* CPU-time clocks are available.  */
+#ifndef _POSIX_CPUTIME
+#define _POSIX_CPUTIME          200809L
+#endif
+
+/* Clock support in threads are available.  */
+#ifndef _POSIX_THREAD_CPUTIME
+#define _POSIX_THREAD_CPUTIME   200809L
+#endif
+
+#ifndef TIMER_ABSTIME
+#define TIMER_ABSTIME   1
+#endif
+
+#ifndef CLOCK_REALTIME
+#define CLOCK_REALTIME              0
+#endif
+
+#ifndef CLOCK_MONOTONIC
+#define CLOCK_MONOTONIC             1
+#endif
+
+#ifndef CLOCK_PROCESS_CPUTIME_ID
+#define CLOCK_PROCESS_CPUTIME_ID    2
+#endif
+
+#ifndef CLOCK_THREAD_CPUTIME_ID
+#define CLOCK_THREAD_CPUTIME_ID     3
+#endif
+
+#ifndef CLOCK_REALTIME_COARSE
+#define CLOCK_REALTIME_COARSE       4
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+WINPTHREAD_API int __cdecl nanosleep32(const struct _timespec32 *request, struct _timespec32 *remain);
+WINPTHREAD_API int __cdecl nanosleep64(const struct _timespec64 *request, struct _timespec64 *remain);
+WINPTHREAD_NANOSLEEP_DECL int __cdecl nanosleep(const struct timespec *request, struct timespec *remain)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return nanosleep32 ((struct _timespec32 *)request, (struct _timespec32 *)remain);
+#else
+  return nanosleep64 ((struct _timespec64 *)request, (struct _timespec64 *)remain);
+#endif
+}
+
+WINPTHREAD_API int __cdecl clock_nanosleep32(clockid_t clock_id, int flags, const struct _timespec32 *request, struct _timespec32 *remain);
+WINPTHREAD_API int __cdecl clock_nanosleep64(clockid_t clock_id, int flags, const struct _timespec64 *request, struct _timespec64 *remain);
+WINPTHREAD_CLOCK_DECL int __cdecl clock_nanosleep(clockid_t clock_id, int flags, const struct timespec *request, struct timespec *remain)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return clock_nanosleep32 (clock_id, flags, (struct _timespec32 *)request, (struct _timespec32 *)remain);
+#else
+  return clock_nanosleep64 (clock_id, flags, (struct _timespec64 *)request, (struct _timespec64 *)remain);
+#endif
+}
+
+WINPTHREAD_API int __cdecl clock_getres32(clockid_t clock_id, struct _timespec32 *res);
+WINPTHREAD_API int __cdecl clock_getres64(clockid_t clock_id, struct _timespec64 *res);
+WINPTHREAD_CLOCK_DECL int __cdecl clock_getres(clockid_t clock_id, struct timespec *res)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return clock_getres32 (clock_id, (struct _timespec32 *)res);
+#else
+  return clock_getres64 (clock_id, (struct _timespec64 *)res);
+#endif
+}
+
+WINPTHREAD_API int __cdecl clock_gettime32(clockid_t clock_id, struct _timespec32 *tp);
+WINPTHREAD_API int __cdecl clock_gettime64(clockid_t clock_id, struct _timespec64 *tp);
+WINPTHREAD_CLOCK_DECL int __cdecl clock_gettime(clockid_t clock_id, struct timespec *tp)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return clock_gettime32 (clock_id, (struct _timespec32 *)tp);
+#else
+  return clock_gettime64 (clock_id, (struct _timespec64 *)tp);
+#endif
+}
+
+WINPTHREAD_API int __cdecl clock_settime32(clockid_t clock_id, const struct _timespec32 *tp);
+WINPTHREAD_API int __cdecl clock_settime64(clockid_t clock_id, const struct _timespec64 *tp);
+WINPTHREAD_CLOCK_DECL int __cdecl clock_settime(clockid_t clock_id, const struct timespec *tp)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return clock_settime32 (clock_id, (struct _timespec32 *)tp);
+#else
+  return clock_settime64 (clock_id, (struct _timespec64 *)tp);
+#endif
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* WIN_PTHREADS_TIME_H */
diff --git a/vendor/winpthreads/include/pthread_unistd.h b/vendor/winpthreads/include/pthread_unistd.h
new file mode 100644
index 00000000..64693534
--- /dev/null
+++ b/vendor/winpthreads/include/pthread_unistd.h
@@ -0,0 +1,192 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_UNISTD_H
+#define WIN_PTHREADS_UNISTD_H
+
+/* Set defines described by the POSIX Threads Extension (1003.1c-1995) */
+/* _SC_THREADS
+  Basic support for POSIX threads is available. The functions
+
+  pthread_atfork(),
+  pthread_attr_destroy(),
+  pthread_attr_getdetachstate(),
+  pthread_attr_getschedparam(),
+  pthread_attr_init(),
+  pthread_attr_setdetachstate(),
+  pthread_attr_setschedparam(),
+  pthread_cancel(),
+  pthread_cleanup_push(),
+  pthread_cleanup_pop(),
+  pthread_cond_broadcast(),
+  pthread_cond_destroy(),
+  pthread_cond_init(),
+  pthread_cond_signal(),
+  pthread_cond_timedwait(),
+  pthread_cond_wait(),
+  pthread_condattr_destroy(),
+  pthread_condattr_init(),
+  pthread_create(),
+  pthread_detach(),
+  pthread_equal(),
+  pthread_exit(),
+  pthread_getspecific(),
+  pthread_join(,
+  pthread_key_create(),
+  pthread_key_delete(),
+  pthread_mutex_destroy(),
+  pthread_mutex_init(),
+  pthread_mutex_lock(),
+  pthread_mutex_trylock(),
+  pthread_mutex_unlock(),
+  pthread_mutexattr_destroy(),
+  pthread_mutexattr_init(),
+  pthread_once(),
+  pthread_rwlock_destroy(),
+  pthread_rwlock_init(),
+  pthread_rwlock_rdlock(),
+  pthread_rwlock_tryrdlock(),
+  pthread_rwlock_trywrlock(),
+  pthread_rwlock_unlock(),
+  pthread_rwlock_wrlock(),
+  pthread_rwlockattr_destroy(),
+  pthread_rwlockattr_init(),
+  pthread_self(),
+  pthread_setcancelstate(),
+  pthread_setcanceltype(),
+  pthread_setspecific(),
+  pthread_testcancel()
+
+  are present. */
+#undef _POSIX_THREADS
+#define _POSIX_THREADS 200112L
+
+/* _SC_READER_WRITER_LOCKS
+  This option implies the _POSIX_THREADS option. Conversely, under
+  POSIX 1003.1-2001 the _POSIX_THREADS option implies this option.
+
+  The functions
+  pthread_rwlock_destroy(),
+  pthread_rwlock_init(),
+  pthread_rwlock_rdlock(),
+  pthread_rwlock_tryrdlock(),
+  pthread_rwlock_trywrlock(),
+  pthread_rwlock_unlock(),
+  pthread_rwlock_wrlock(),
+  pthread_rwlockattr_destroy(),
+  pthread_rwlockattr_init()
+
+  are present.
+*/
+#undef _POSIX_READER_WRITER_LOCKS
+#define _POSIX_READER_WRITER_LOCKS 200112L
+
+/* _SC_SPIN_LOCKS
+  This option implies the _POSIX_THREADS and _POSIX_THREAD_SAFE_FUNCTIONS
+  options. The functions
+
+  pthread_spin_destroy(),
+  pthread_spin_init(),
+  pthread_spin_lock(),
+  pthread_spin_trylock(),
+  pthread_spin_unlock()
+
+  are present. */
+#undef _POSIX_SPIN_LOCKS
+#define _POSIX_SPIN_LOCKS 200112L
+
+/* _SC_BARRIERS
+  This option implies the _POSIX_THREADS and _POSIX_THREAD_SAFE_FUNCTIONS
+  options. The functions
+
+  pthread_barrier_destroy(),
+  pthread_barrier_init(),
+  pthread_barrier_wait(),
+  pthread_barrierattr_destroy(),
+  pthread_barrierattr_init()
+
+  are present.
+*/
+#undef _POSIX_BARRIERS
+#define _POSIX_BARRIERS 200112L
+
+/* _SC_TIMEOUTS
+  The functions
+
+  mq_timedreceive(), - not supported
+  mq_timedsend(), - not supported
+  posix_trace_timedgetnext_event(), - not supported
+  pthread_mutex_timedlock(),
+  pthread_rwlock_timedrdlock(),
+  pthread_rwlock_timedwrlock(),
+  sem_timedwait(),
+
+  are present. */
+#undef _POSIX_TIMEOUTS
+#define _POSIX_TIMEOUTS 200112L
+
+/* _SC_TIMERS - not supported
+  The functions
+
+  clock_getres(),
+  clock_gettime(),
+  clock_settime(),
+  nanosleep(),
+  timer_create(),
+  timer_delete(),
+  timer_gettime(),
+  timer_getoverrun(),
+  timer_settime()
+
+  are present.  */
+/* #undef _POSIX_TIMERS */
+
+/* _SC_CLOCK_SELECTION
+   This option implies the _POSIX_TIMERS option. The functions
+
+   pthread_condattr_getclock(),
+   pthread_condattr_setclock(),
+   clock_nanosleep()
+
+   are present.
+*/
+#undef _POSIX_CLOCK_SELECTION
+#define _POSIX_CLOCK_SELECTION 200112
+
+/* _SC_SEMAPHORES
+  The include file <semaphore.h> is present. The functions
+
+  sem_close(),
+  sem_destroy(),
+  sem_getvalue(),
+  sem_init(),
+  sem_open(),
+  sem_post(),
+  sem_trywait(),
+  sem_unlink(),
+  sem_wait()
+
+  are present. */
+#undef _POSIX_SEMAPHORES
+#define _POSIX_SEMAPHORES 200112
+
+#endif /* WIN_PTHREADS_UNISTD_H */
diff --git a/vendor/winpthreads/include/sched.h b/vendor/winpthreads/include/sched.h
new file mode 100644
index 00000000..abbf5cf0
--- /dev/null
+++ b/vendor/winpthreads/include/sched.h
@@ -0,0 +1,65 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+#ifndef WIN_PTHREADS_SCHED_H
+#define WIN_PTHREADS_SCHED_H
+
+#include <stddef.h>
+#include <errno.h>
+#include <sys/types.h>
+
+#include <process.h>
+#include <limits.h>
+#include <signal.h>
+
+#include "pthread_compat.h"
+
+/* Some POSIX realtime extensions, mostly stubbed */
+#define SCHED_OTHER     0
+#define SCHED_FIFO      1
+#define SCHED_RR        2
+#define SCHED_MIN       SCHED_OTHER
+#define SCHED_MAX       SCHED_RR
+
+struct sched_param {
+  int sched_priority;
+};
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+WINPTHREAD_API int sched_yield(void);
+WINPTHREAD_API int sched_get_priority_min(int pol);
+WINPTHREAD_API int sched_get_priority_max(int pol);
+WINPTHREAD_API int sched_getscheduler(pid_t pid);
+WINPTHREAD_API int sched_setscheduler(pid_t pid, int pol, const struct sched_param *param);
+
+#ifdef __cplusplus
+}
+#endif
+
+#ifndef sched_rr_get_interval
+#define sched_rr_get_interval(_p, _i) \
+  ( errno = ENOTSUP, (int) -1 )
+#endif
+
+#endif /* WIN_PTHREADS_SCHED_H */
diff --git a/vendor/winpthreads/include/semaphore.h b/vendor/winpthreads/include/semaphore.h
new file mode 100644
index 00000000..9a6c211a
--- /dev/null
+++ b/vendor/winpthreads/include/semaphore.h
@@ -0,0 +1,75 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_SEMAPHORE_H
+#define WIN_PTHREADS_SEMAPHORE_H
+
+#include <sys/timeb.h>
+#include "pthread_compat.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define SEM_VALUE_MAX   INT_MAX
+
+typedef void		*sem_t;
+
+#define SEM_FAILED 		NULL
+
+WINPTHREAD_API int sem_init(sem_t * sem, int pshared, unsigned int value);
+
+WINPTHREAD_API int sem_destroy(sem_t *sem);
+
+WINPTHREAD_API int sem_trywait(sem_t *sem);
+
+WINPTHREAD_API int sem_wait(sem_t *sem);
+
+WINPTHREAD_API int sem_timedwait32(sem_t * sem, const struct _timespec32 *t);
+WINPTHREAD_API int sem_timedwait64(sem_t * sem, const struct _timespec64 *t);
+WINPTHREAD_SEM_DECL int sem_timedwait(sem_t * sem, const struct timespec *t)
+{
+#if WINPTHREADS_TIME_BITS == 32
+  return sem_timedwait32 (sem, (const struct _timespec32 *) t);
+#else
+  return sem_timedwait64 (sem, (const struct _timespec64 *) t);
+#endif
+}
+
+WINPTHREAD_API int sem_post(sem_t *sem);
+
+WINPTHREAD_API int sem_post_multiple(sem_t *sem, int count);
+
+/* yes, it returns a semaphore (or SEM_FAILED) */
+WINPTHREAD_API sem_t * sem_open(const char * name, int oflag, mode_t mode, unsigned int value);
+
+WINPTHREAD_API int sem_close(sem_t * sem);
+
+WINPTHREAD_API int sem_unlink(const char * name);
+
+WINPTHREAD_API int sem_getvalue(sem_t * sem, int * sval);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* WIN_PTHREADS_SEMAPHORE_H */
diff --git a/vendor/winpthreads/src/barrier.c b/vendor/winpthreads/src/barrier.c
new file mode 100644
index 00000000..b1be5126
--- /dev/null
+++ b/vendor/winpthreads/src/barrier.c
@@ -0,0 +1,254 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <assert.h>
+#include <malloc.h>
+#include <stdio.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+/* public header files */
+#include "pthread.h"
+#include "semaphore.h"
+/* internal header files */
+#include "barrier.h"
+#include "misc.h"
+
+static pthread_spinlock_t barrier_global = PTHREAD_SPINLOCK_INITIALIZER;
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int
+barrier_unref(volatile pthread_barrier_t *barrier, int res)
+{
+    pthread_spin_lock(&barrier_global);
+    assert((((barrier_t *)*barrier)->valid == LIFE_BARRIER) && (((barrier_t *)*barrier)->busy > 0));
+     ((barrier_t *)*barrier)->busy -= 1;
+    pthread_spin_unlock(&barrier_global);
+    return res;
+}
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int barrier_ref(volatile pthread_barrier_t *barrier)
+{
+    int r = 0;
+    pthread_spin_lock(&barrier_global);
+
+    if (!barrier || !*barrier || ((barrier_t *)*barrier)->valid != LIFE_BARRIER) r = EINVAL;
+    else {
+        ((barrier_t *)*barrier)->busy += 1;
+    }
+
+    pthread_spin_unlock(&barrier_global);
+
+    return r;
+}
+
+static WINPTHREADS_ATTRIBUTE((noinline))  int
+barrier_ref_destroy(volatile pthread_barrier_t *barrier, pthread_barrier_t *bDestroy)
+{
+    int r = 0;
+
+    *bDestroy = NULL;
+    pthread_spin_lock(&barrier_global);
+
+    if (!barrier || !*barrier || ((barrier_t *)*barrier)->valid != LIFE_BARRIER) r = EINVAL;
+    else {
+        barrier_t *b_ = (barrier_t *)*barrier;
+        if (b_->busy) r = EBUSY;
+        else {
+            *bDestroy = *barrier;
+            *barrier = NULL;
+        }
+    }
+
+    pthread_spin_unlock(&barrier_global);
+    return r;
+}
+
+static WINPTHREADS_ATTRIBUTE((noinline)) void
+barrier_ref_set (volatile pthread_barrier_t *barrier, void *v)
+{
+  pthread_spin_lock(&barrier_global);
+  *barrier = v;
+  pthread_spin_unlock(&barrier_global);
+}
+
+int pthread_barrier_destroy(pthread_barrier_t *b_)
+{
+    pthread_barrier_t bDestroy;
+    barrier_t *b;
+    int r;
+
+    while ((r = barrier_ref_destroy(b_,&bDestroy)) == EBUSY)
+      Sleep(0);
+
+    if (r)
+      return r;
+
+    b = (barrier_t *)bDestroy;
+
+    pthread_mutex_lock(&b->m);
+
+    if (sem_destroy(&b->sems[0]) != 0)
+    {
+        /* Could this happen? */
+        *b_ = bDestroy;
+        pthread_mutex_unlock (&b->m);
+        return EBUSY;
+    }
+    if (sem_destroy(&b->sems[1]) != 0)
+    {
+      sem_init (&b->sems[0], b->share, 0);
+      *b_ = bDestroy;
+      pthread_mutex_unlock (&b->m);
+      return -1;
+    }
+    pthread_mutex_unlock(&b->m);
+    if(pthread_mutex_destroy(&b->m) != 0) {
+     sem_init (&b->sems[0], b->share, 0);
+     sem_init (&b->sems[1], b->share, 0);
+     *b_ = bDestroy;
+     return -1;
+    }
+    b->valid = DEAD_BARRIER;
+    free(bDestroy);
+    return 0;
+
+}
+
+int
+pthread_barrier_init (pthread_barrier_t *b_, const void *attr,
+		      unsigned int count)
+{
+    barrier_t *b;
+
+    if (!count || !b_)
+      return EINVAL;
+
+    if ((b = (pthread_barrier_t)calloc(1,sizeof(*b))) == NULL)
+       return ENOMEM;
+    if (!attr || *((int **)attr) == NULL)
+      b->share = PTHREAD_PROCESS_PRIVATE;
+    else
+      memcpy (&b->share, *((void **) attr), sizeof (int));
+    b->total = count;
+    b->count = count;
+    b->valid = LIFE_BARRIER;
+    b->sel = 0;
+
+    if (pthread_mutex_init(&b->m, NULL) != 0)
+    {
+      free (b);
+      return ENOMEM;
+    }
+
+    if (sem_init(&b->sems[0], b->share, 0) != 0)
+    {
+       pthread_mutex_destroy(&b->m);
+       free (b);
+       return ENOMEM;
+    }
+    if (sem_init(&b->sems[1], b->share, 0) != 0)
+    {
+       pthread_mutex_destroy(&b->m);
+       sem_destroy(&b->sems[0]);
+       free (b);
+       return ENOMEM;
+    }
+    barrier_ref_set (b_,b);
+
+    return 0;
+}
+
+int pthread_barrier_wait(pthread_barrier_t *b_)
+{
+  long sel;
+  int r, e, rslt;
+  barrier_t *b;
+
+  r = barrier_ref(b_);
+  if(r) return r;
+
+  b = (barrier_t *)*b_;
+
+  if ((r = pthread_mutex_lock(&b->m)) != 0) return  barrier_unref(b_,EINVAL);
+  sel = b->sel;
+  InterlockedDecrement((long*)&b->total);
+  if (b->total == 0)
+  {
+    b->total = b->count;
+    b->sel = (sel != 0 ? 0 : 1);
+    e = 1;
+    rslt = PTHREAD_BARRIER_SERIAL_THREAD;
+    r = (b->count > 1 ? sem_post_multiple (&b->sems[sel], b->count - 1) : 0);
+  }
+  else { e = 0; rslt= 0; }
+  pthread_mutex_unlock(&b->m);
+  if (!e)
+    r = sem_wait(&b->sems[sel]);
+
+  if (!r) r = rslt;
+  return barrier_unref(b_,r);
+}
+
+int pthread_barrierattr_init(void **attr)
+{
+  int *p;
+
+  if ((p = (int *) calloc (1, sizeof (int))) == NULL)
+    return ENOMEM;
+
+  *p = PTHREAD_PROCESS_PRIVATE;
+  *attr = p;
+
+  return 0;
+}
+
+int pthread_barrierattr_destroy(void **attr)
+{
+  void *p;
+  if (!attr || (p = *attr) == NULL)
+    return EINVAL;
+  *attr = NULL;
+  free (p);
+  return 0;
+}
+
+int pthread_barrierattr_setpshared(void **attr, int s)
+{
+  if (!attr || *attr == NULL
+      || (s != PTHREAD_PROCESS_SHARED && s != PTHREAD_PROCESS_PRIVATE))
+    return EINVAL;
+  memcpy (*attr, &s, sizeof (int));
+  return 0;
+}
+
+int pthread_barrierattr_getpshared(void **attr, int *s)
+{
+  if (!attr || !s || *attr == NULL)
+    return EINVAL;
+  memcpy (s, *attr, sizeof (int));
+  return 0;
+}
diff --git a/vendor/winpthreads/src/barrier.h b/vendor/winpthreads/src/barrier.h
new file mode 100644
index 00000000..1d1e83da
--- /dev/null
+++ b/vendor/winpthreads/src/barrier.h
@@ -0,0 +1,50 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_BARRIER_H
+#define WIN_PTHREADS_BARRIER_H
+
+#define LIFE_BARRIER 0xBAB1FEED
+#define DEAD_BARRIER 0xDEADB00F
+
+#define _PTHREAD_BARRIER_FLAG (1<<30)
+
+#define CHECK_BARRIER(b)                                                \
+    do {                                                                \
+        if (!(b) || ( ((barrier_t *)(*b))->valid != (unsigned int)LIFE_BARRIER ) ) \
+            return EINVAL;                                              \
+    } while (0)
+
+typedef struct barrier_t barrier_t;
+struct barrier_t
+{
+    int valid;
+    int busy;
+    int count;
+    int total;
+    int share;
+    long sel;
+    pthread_mutex_t m;
+    sem_t sems[2];
+};
+
+#endif
diff --git a/vendor/winpthreads/src/clock.c b/vendor/winpthreads/src/clock.c
new file mode 100644
index 00000000..922c5de5
--- /dev/null
+++ b/vendor/winpthreads/src/clock.c
@@ -0,0 +1,354 @@
+/**
+ * This file has no copyright assigned and is placed in the Public Domain.
+ * This file is part of the w64 mingw-runtime package.
+ * No warranty is given; refer to the file DISCLAIMER.PD within this package.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <assert.h>
+#include <errno.h>
+#include <stdint.h>
+#include <time.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define WINPTHREAD_CLOCK_DECL WINPTHREAD_API
+
+/* public header files */
+#include "pthread_time.h"
+/* internal header files */
+#include "misc.h"
+
+#define POW10_7                 10000000
+#define POW10_9                 1000000000
+
+/* Number of 100ns-seconds between the beginning of the Windows epoch
+ * (Jan. 1, 1601) and the Unix epoch (Jan. 1, 1970)
+ */
+#define DELTA_EPOCH_IN_100NS    INT64_C(116444736000000000)
+
+static WINPTHREADS_INLINE int lc_set_errno(int result)
+{
+    if (result != 0) {
+        errno = result;
+        return -1;
+    }
+    return 0;
+}
+
+/**
+ * Get the resolution of the specified clock clock_id and
+ * stores it in the struct timespec pointed to by res.
+ * @param  clock_id The clock_id argument is the identifier of the particular
+ *         clock on which to act. The following clocks are supported:
+ * <pre>
+ *     CLOCK_REALTIME  System-wide real-time clock. Setting this clock
+ *                 requires appropriate privileges.
+ *     CLOCK_MONOTONIC Clock that cannot be set and represents monotonic
+ *                 time since some unspecified starting point.
+ *     CLOCK_PROCESS_CPUTIME_ID High-resolution per-process timer from the CPU.
+ *     CLOCK_THREAD_CPUTIME_ID  Thread-specific CPU-time clock.
+ * </pre>
+ * @param  res The pointer to a timespec structure to receive the time
+ *         resolution.
+ * @return If the function succeeds, the return value is 0.
+ *         If the function fails, the return value is -1,
+ *         with errno set to indicate the error.
+ */
+static int __clock_getres(clockid_t clock_id, struct _timespec64 *res)
+{
+    clockid_t id = clock_id;
+
+    if (id == CLOCK_REALTIME && _pthread_get_system_time_best_as_file_time == GetSystemTimeAsFileTime)
+        id = CLOCK_REALTIME_COARSE; /* GetSystemTimePreciseAsFileTime() not available */
+
+    switch(id) {
+    case CLOCK_REALTIME:
+    case CLOCK_MONOTONIC:
+        {
+            LARGE_INTEGER pf;
+
+            if (QueryPerformanceFrequency(&pf) == 0)
+                return lc_set_errno(EINVAL);
+
+            res->tv_sec = 0;
+            res->tv_nsec = (int) ((POW10_9 + (pf.QuadPart >> 1)) / pf.QuadPart);
+            if (res->tv_nsec < 1)
+                res->tv_nsec = 1;
+
+            return 0;
+        }
+
+    case CLOCK_REALTIME_COARSE:
+    case CLOCK_PROCESS_CPUTIME_ID:
+    case CLOCK_THREAD_CPUTIME_ID:
+        {
+            DWORD   timeAdjustment, timeIncrement;
+            BOOL    isTimeAdjustmentDisabled;
+
+            (void) GetSystemTimeAdjustment(&timeAdjustment, &timeIncrement, &isTimeAdjustmentDisabled);
+            res->tv_sec = 0;
+            res->tv_nsec = timeIncrement * 100;
+
+            return 0;
+        }
+    default:
+        break;
+    }
+
+    return lc_set_errno(EINVAL);
+}
+
+/**
+ * Get the time of the specified clock clock_id and stores it in the struct
+ * timespec pointed to by tp.
+ * @param  clock_id The clock_id argument is the identifier of the particular
+ *         clock on which to act. The following clocks are supported:
+ * <pre>
+ *     CLOCK_REALTIME  System-wide real-time clock. Setting this clock
+ *                 requires appropriate privileges.
+ *     CLOCK_MONOTONIC Clock that cannot be set and represents monotonic
+ *                 time since some unspecified starting point.
+ *     CLOCK_PROCESS_CPUTIME_ID High-resolution per-process timer from the CPU.
+ *     CLOCK_THREAD_CPUTIME_ID  Thread-specific CPU-time clock.
+ * </pre>
+ * @param  tp The pointer to a timespec structure to receive the time.
+ * @return If the function succeeds, the return value is 0.
+ *         If the function fails, the return value is -1,
+ *         with errno set to indicate the error.
+ */
+static int __clock_gettime(clockid_t clock_id, struct _timespec64 *tp)
+{
+    unsigned __int64 t;
+    LARGE_INTEGER pf, pc;
+    union {
+        unsigned __int64 u64;
+        FILETIME ft;
+    }  ct, et, kt, ut;
+
+    switch(clock_id) {
+    case CLOCK_REALTIME:
+        {
+            _pthread_get_system_time_best_as_file_time(&ct.ft);
+            t = ct.u64 - DELTA_EPOCH_IN_100NS;
+            tp->tv_sec = t / POW10_7;
+            tp->tv_nsec = ((int) (t % POW10_7)) * 100;
+
+            return 0;
+        }
+
+    case CLOCK_REALTIME_COARSE:
+        {
+            GetSystemTimeAsFileTime(&ct.ft);
+            t = ct.u64 - DELTA_EPOCH_IN_100NS;
+            tp->tv_sec = t / POW10_7;
+            tp->tv_nsec = ((int) (t % POW10_7)) * 100;
+
+            return 0;
+        }
+
+    case CLOCK_MONOTONIC:
+        {
+            if (QueryPerformanceFrequency(&pf) == 0)
+                return lc_set_errno(EINVAL);
+
+            if (QueryPerformanceCounter(&pc) == 0)
+                return lc_set_errno(EINVAL);
+
+            tp->tv_sec = pc.QuadPart / pf.QuadPart;
+            tp->tv_nsec = (int) (((pc.QuadPart % pf.QuadPart) * POW10_9 + (pf.QuadPart >> 1)) / pf.QuadPart);
+            if (tp->tv_nsec >= POW10_9) {
+                tp->tv_sec ++;
+                tp->tv_nsec -= POW10_9;
+            }
+
+            return 0;
+        }
+
+    case CLOCK_PROCESS_CPUTIME_ID:
+        {
+        if(0 == GetProcessTimes(GetCurrentProcess(), &ct.ft, &et.ft, &kt.ft, &ut.ft))
+            return lc_set_errno(EINVAL);
+        t = kt.u64 + ut.u64;
+        tp->tv_sec = t / POW10_7;
+        tp->tv_nsec = ((int) (t % POW10_7)) * 100;
+
+        return 0;
+        }
+
+    case CLOCK_THREAD_CPUTIME_ID:
+        {
+            if(0 == GetThreadTimes(GetCurrentThread(), &ct.ft, &et.ft, &kt.ft, &ut.ft))
+                return lc_set_errno(EINVAL);
+            t = kt.u64 + ut.u64;
+            tp->tv_sec = t / POW10_7;
+            tp->tv_nsec = ((int) (t % POW10_7)) * 100;
+
+            return 0;
+        }
+
+    default:
+        break;
+    }
+
+    return lc_set_errno(EINVAL);
+}
+
+/**
+ * Sleep for the specified time.
+ * @param  clock_id This argument should always be CLOCK_REALTIME (0).
+ * @param  flags 0 for relative sleep interval, others for absolute waking up.
+ * @param  request The desired sleep interval or absolute waking up time.
+ * @param  remain The remain amount of time to sleep.
+ *         The current implemention just ignore it.
+ * @return If the function succeeds, the return value is 0.
+ *         If the function fails, the return value is -1,
+ *         with errno set to indicate the error.
+ */
+static int __clock_nanosleep(clockid_t clock_id, int flags,
+                           const struct _timespec64 *request,
+                           struct _timespec64 *remain)
+{
+    struct _timespec64 tp;
+
+    if (clock_id != CLOCK_REALTIME)
+        return lc_set_errno(EINVAL);
+
+    if (flags == 0)
+        return nanosleep64(request, remain);
+
+    /* TIMER_ABSTIME = 1 */
+    __clock_gettime(CLOCK_REALTIME, &tp);
+
+    tp.tv_sec = request->tv_sec - tp.tv_sec;
+    tp.tv_nsec = request->tv_nsec - tp.tv_nsec;
+    if (tp.tv_nsec < 0) {
+        tp.tv_nsec += POW10_9;
+        tp.tv_sec --;
+    }
+
+    return nanosleep64(&tp, remain);
+}
+
+/**
+ * Set the time of the specified clock clock_id.
+ * @param  clock_id This argument should always be CLOCK_REALTIME (0).
+ * @param  tp The requested time.
+ * @return If the function succeeds, the return value is 0.
+ *         If the function fails, the return value is -1,
+ *         with errno set to indicate the error.
+ */
+static int __clock_settime(clockid_t clock_id, const struct _timespec64 *tp)
+{
+    SYSTEMTIME st;
+
+    union {
+        unsigned __int64 u64;
+        FILETIME ft;
+    }  t;
+
+    if (clock_id != CLOCK_REALTIME)
+        return lc_set_errno(EINVAL);
+
+    t.u64 = tp->tv_sec * (__int64) POW10_7 + tp->tv_nsec / 100 + DELTA_EPOCH_IN_100NS;
+    if (FileTimeToSystemTime(&t.ft, &st) == 0)
+        return lc_set_errno(EINVAL);
+
+    if (SetSystemTime(&st) == 0)
+        return lc_set_errno(EPERM);
+
+    return 0;
+}
+
+/**
+ * Versions to use with 64-bit time_t (struct _timespec64)
+ */
+
+int clock_getres64 (clockid_t clock_id, struct _timespec64 *tp)
+{
+    return __clock_getres (clock_id, tp);
+}
+
+int clock_gettime64 (clockid_t clock_id, struct _timespec64 *tp)
+{
+    return __clock_gettime (clock_id, tp);
+}
+
+int clock_settime64 (clockid_t clock_id, const struct _timespec64 *tp)
+{
+    return __clock_settime (clock_id, tp);
+}
+
+int clock_nanosleep64 (clockid_t clock_id, int flags,
+                const struct _timespec64 *request, struct _timespec64 *remain)
+{
+    return __clock_nanosleep (clock_id, flags, request, remain);
+}
+
+/**
+ * Versions to use with 32-bit time_t (struct _timespec32)
+ */
+
+int clock_getres32 (clockid_t clock_id, struct _timespec32 *tp)
+{
+    struct _timespec64 tp64 = {0};
+
+    if (__clock_getres (clock_id, &tp64) == -1)
+        return -1;
+
+    tp->tv_sec = (__time32_t) tp64.tv_sec;
+    tp->tv_nsec = tp64.tv_nsec;
+
+    return 0;
+}
+
+int clock_gettime32 (clockid_t clock_id, struct _timespec32 *tp)
+{
+    struct _timespec64 tp64 = {0};
+
+    if (__clock_gettime (clock_id, &tp64) == -1)
+        return -1;
+
+    if (tp64.tv_sec > INT_MAX)
+    {
+        _set_errno (EOVERFLOW);
+        return -1;
+    }
+
+    tp->tv_sec = (__time32_t) tp64.tv_sec;
+    tp->tv_nsec = tp64.tv_nsec;
+
+    return 0;
+}
+
+int clock_settime32 (clockid_t clock_id, const struct _timespec32 *tp)
+{
+    struct _timespec64 tp64 = {.tv_sec = tp->tv_sec, .tv_nsec = tp->tv_nsec};
+    return __clock_settime (clock_id, &tp64);
+}
+
+int clock_nanosleep32 (clockid_t clock_id, int flags,
+                const struct _timespec32 *request, struct _timespec32 *remain)
+{
+    struct _timespec64 request64 = {
+        .tv_sec = request->tv_sec,
+        .tv_nsec = request->tv_nsec
+    };
+    struct _timespec64 remain64 = {0};
+
+    if (__clock_nanosleep (clock_id, flags, &request64, &remain64) == -1)
+        return -1;
+
+    assert (remain64.tv_sec <= INT_MAX);
+
+    if (remain != NULL) {
+        remain->tv_sec = (__time32_t)remain64.tv_sec;
+        remain->tv_nsec = remain64.tv_nsec;
+    }
+
+    return 0;
+}
diff --git a/vendor/winpthreads/src/cond.c b/vendor/winpthreads/src/cond.c
new file mode 100644
index 00000000..a77cd3ef
--- /dev/null
+++ b/vendor/winpthreads/src/cond.c
@@ -0,0 +1,717 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+/*
+ * Posix Condition Variables for Microsoft Windows.
+ * 22-9-2010 Partly based on the ACE framework implementation.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <malloc.h>
+#include <stdio.h>
+#include <time.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define WINPTHREAD_COND_DECL WINPTHREAD_API
+
+/* public header files */
+#include "pthread.h"
+#include "pthread_time.h"
+/* internal header files */
+#include "cond.h"
+#include "misc.h"
+#include "thread.h"
+
+#include "pthread_compat.h"
+
+int __pthread_shallcancel (void);
+
+static int do_sema_b_wait (HANDLE sema, int nointerrupt, DWORD timeout,CRITICAL_SECTION *cs, LONG *val);
+static int do_sema_b_release(HANDLE sema, LONG count,CRITICAL_SECTION *cs, LONG *val);
+static void cleanup_wait(void *arg);
+
+typedef struct sCondWaitHelper {
+    cond_t *c;
+    pthread_mutex_t *external_mutex;
+    int *r;
+} sCondWaitHelper;
+
+int do_sema_b_wait_intern (HANDLE sema, int nointerrupt, DWORD timeout);
+
+static pthread_spinlock_t cond_locked = PTHREAD_SPINLOCK_INITIALIZER;
+
+static int
+cond_static_init (pthread_cond_t *c)
+{
+  int r = 0;
+
+  pthread_spin_lock (&cond_locked);
+  if (c == NULL)
+    r = EINVAL;
+  else if (*c == PTHREAD_COND_INITIALIZER)
+    r = pthread_cond_init (c, NULL);
+  else
+    /* We assume someone was faster ... */
+    r = 0;
+  pthread_spin_unlock (&cond_locked);
+  return r;
+}
+
+int
+pthread_condattr_destroy (pthread_condattr_t *a)
+{
+  if (!a)
+    return EINVAL;
+   *a = 0;
+   return 0;
+}
+
+int
+pthread_condattr_init (pthread_condattr_t *a)
+{
+  if (!a)
+    return EINVAL;
+  *a = 0;
+  return 0;
+}
+
+int
+pthread_condattr_getpshared (const pthread_condattr_t *a, int *s)
+{
+  if (!a || !s)
+    return EINVAL;
+  *s = *a;
+  return 0;
+}
+
+int
+pthread_condattr_getclock (const pthread_condattr_t *a, clockid_t *clock_id)
+{
+  if (!a || !clock_id)
+    return EINVAL;
+  *clock_id = 0;
+  return 0;
+}
+
+int
+pthread_condattr_setclock(pthread_condattr_t *a, clockid_t clock_id)
+{
+  if (!a || clock_id != 0)
+    return EINVAL;
+  return 0;
+}
+
+int
+pthread_condattr_setpshared (pthread_condattr_t *a, int s)
+{
+  if (!a || (s != PTHREAD_PROCESS_SHARED && s != PTHREAD_PROCESS_PRIVATE))
+    return EINVAL;
+  if (s == PTHREAD_PROCESS_SHARED)
+    {
+       *a = PTHREAD_PROCESS_PRIVATE;
+       return ENOSYS;
+    }
+  *a = s;
+  return 0;
+}
+
+int
+pthread_cond_init (pthread_cond_t *c, const pthread_condattr_t *a)
+{
+  cond_t *_c;
+  int r = 0;
+
+  if (!c)
+    return EINVAL;
+  if (a && *a == PTHREAD_PROCESS_SHARED)
+    return ENOSYS;
+
+  if ((_c = calloc(1, sizeof(*_c))) == NULL)
+    return ENOMEM;
+
+  _c->valid  = DEAD_COND;
+  _c->busy = 0;
+  _c->waiters_count_ = 0;
+  _c->waiters_count_gone_ = 0;
+  _c->waiters_count_unblock_ = 0;
+
+  _c->sema_q = CreateSemaphore (NULL,       /* no security */
+      0,          /* initially 0 */
+      0x7fffffff, /* max count */
+      NULL);      /* unnamed  */
+  _c->sema_b =  CreateSemaphore (NULL,       /* no security */
+      0,          /* initially 0 */
+      0x7fffffff, /* max count */
+      NULL);
+  if (_c->sema_q == NULL || _c->sema_b == NULL) {
+      if (_c->sema_q != NULL)
+	CloseHandle (_c->sema_q);
+      if (_c->sema_b != NULL)
+	CloseHandle (_c->sema_b);
+      free (_c);
+      r = EAGAIN;
+  } else {
+      InitializeCriticalSection(&_c->waiters_count_lock_);
+      InitializeCriticalSection(&_c->waiters_b_lock_);
+      InitializeCriticalSection(&_c->waiters_q_lock_);
+      _c->value_q = 0;
+      _c->value_b = 1;
+  }
+  if (!r)
+    {
+      _c->valid = LIFE_COND;
+      *c = (pthread_cond_t)_c;
+    }
+  else
+    *c = (pthread_cond_t)NULL;
+  return r;
+}
+
+int
+pthread_cond_destroy (pthread_cond_t *c)
+{
+  cond_t *_c;
+  int r;
+  if (!c || !*c)
+    return EINVAL;
+  if (*c == PTHREAD_COND_INITIALIZER)
+    {
+      pthread_spin_lock (&cond_locked);
+      if (*c == PTHREAD_COND_INITIALIZER)
+      {
+	*c = (pthread_cond_t)NULL;
+	r = 0;
+      }
+      else
+	r = EBUSY;
+      pthread_spin_unlock (&cond_locked);
+      return r;
+    }
+  _c = (cond_t *) *c;
+  r = do_sema_b_wait(_c->sema_b, 0, INFINITE,&_c->waiters_b_lock_,&_c->value_b);
+  if (r != 0)
+    return r;
+  if (!TryEnterCriticalSection (&_c->waiters_count_lock_))
+    {
+       do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+       return EBUSY;
+    }
+  if (_c->waiters_count_ > _c->waiters_count_gone_)
+    {
+      r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+      if (!r) r = EBUSY;
+      LeaveCriticalSection(&_c->waiters_count_lock_);
+      return r;
+    }
+  *c = (pthread_cond_t)NULL;
+  do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+
+  if (!CloseHandle (_c->sema_q) && !r)
+    r = EINVAL;
+  if (!CloseHandle (_c->sema_b) && !r)
+    r = EINVAL;
+  LeaveCriticalSection (&_c->waiters_count_lock_);
+  DeleteCriticalSection(&_c->waiters_count_lock_);
+  DeleteCriticalSection(&_c->waiters_b_lock_);
+  DeleteCriticalSection(&_c->waiters_q_lock_);
+  _c->valid  = DEAD_COND;
+  free(_c);
+  return 0;
+}
+
+int
+pthread_cond_signal (pthread_cond_t *c)
+{
+  cond_t *_c;
+  int r;
+
+  if (!c || !*c)
+    return EINVAL;
+  _c = (cond_t *)*c;
+  if (_c == (cond_t *)PTHREAD_COND_INITIALIZER)
+    return 0;
+  else if (_c->valid != (unsigned int)LIFE_COND)
+    return EINVAL;
+
+  EnterCriticalSection (&_c->waiters_count_lock_);
+  /* If there aren't any waiters, then this is a no-op.   */
+  if (_c->waiters_count_unblock_ != 0)
+    {
+      if (_c->waiters_count_ == 0)
+      {
+	LeaveCriticalSection (&_c->waiters_count_lock_);
+	/* pthread_testcancel(); */
+	return 0;
+      }
+      _c->waiters_count_ -= 1;
+      _c->waiters_count_unblock_ += 1;
+    }
+  else if (_c->waiters_count_ > _c->waiters_count_gone_)
+    {
+      r = do_sema_b_wait (_c->sema_b, 1, INFINITE,&_c->waiters_b_lock_,&_c->value_b);
+      if (r != 0)
+      {
+	LeaveCriticalSection (&_c->waiters_count_lock_);
+	/* pthread_testcancel(); */
+	return r;
+      }
+      if (_c->waiters_count_gone_ != 0)
+      {
+	_c->waiters_count_ -= _c->waiters_count_gone_;
+	_c->waiters_count_gone_ = 0;
+      }
+      _c->waiters_count_ -= 1;
+      _c->waiters_count_unblock_ = 1;
+    }
+  else
+    {
+      LeaveCriticalSection (&_c->waiters_count_lock_);
+      /* pthread_testcancel(); */
+      return 0;
+    }
+  LeaveCriticalSection (&_c->waiters_count_lock_);
+  r = do_sema_b_release(_c->sema_q, 1,&_c->waiters_q_lock_,&_c->value_q);
+  /* pthread_testcancel(); */
+  return r;
+}
+
+int
+pthread_cond_broadcast (pthread_cond_t *c)
+{
+  cond_t *_c;
+  int r;
+  int relCnt = 0;
+
+  if (!c || !*c)
+    return EINVAL;
+  _c = (cond_t *)*c;
+  if (_c == (cond_t*)PTHREAD_COND_INITIALIZER)
+    return 0;
+  else if (_c->valid != (unsigned int)LIFE_COND)
+    return EINVAL;
+
+  EnterCriticalSection (&_c->waiters_count_lock_);
+  /* If there aren't any waiters, then this is a no-op.   */
+  if (_c->waiters_count_unblock_ != 0)
+    {
+      if (_c->waiters_count_ == 0)
+      {
+	LeaveCriticalSection (&_c->waiters_count_lock_);
+	/* pthread_testcancel(); */
+	return 0;
+      }
+      relCnt = _c->waiters_count_;
+      _c->waiters_count_ = 0;
+      _c->waiters_count_unblock_ += relCnt;
+    }
+  else if (_c->waiters_count_ > _c->waiters_count_gone_)
+    {
+      r = do_sema_b_wait (_c->sema_b, 1, INFINITE,&_c->waiters_b_lock_,&_c->value_b);
+      if (r != 0)
+      {
+	LeaveCriticalSection (&_c->waiters_count_lock_);
+	/* pthread_testcancel(); */
+	return r;
+      }
+      if (_c->waiters_count_gone_ != 0)
+      {
+	_c->waiters_count_ -= _c->waiters_count_gone_;
+	_c->waiters_count_gone_ = 0;
+      }
+      relCnt = _c->waiters_count_;
+      _c->waiters_count_ = 0;
+      _c->waiters_count_unblock_ = relCnt;
+    }
+  else
+    {
+      LeaveCriticalSection (&_c->waiters_count_lock_);
+      /* pthread_testcancel(); */
+      return 0;
+    }
+  LeaveCriticalSection (&_c->waiters_count_lock_);
+  r = do_sema_b_release(_c->sema_q, relCnt,&_c->waiters_q_lock_,&_c->value_q);
+  /* pthread_testcancel(); */
+  return r;
+}
+
+int
+pthread_cond_wait (pthread_cond_t *c, pthread_mutex_t *external_mutex)
+{
+  sCondWaitHelper ch;
+  cond_t *_c;
+  int r;
+
+  /* pthread_testcancel(); */
+
+  if (!c || *c == (pthread_cond_t)NULL)
+    return EINVAL;
+  _c = (cond_t *)*c;
+  if (*c == PTHREAD_COND_INITIALIZER)
+  {
+    r = cond_static_init(c);
+    if (r != 0 && r != EBUSY)
+      return r;
+    _c = (cond_t *) *c;
+  } else if (_c->valid != (unsigned int)LIFE_COND)
+    return EINVAL;
+
+tryagain:
+  r = do_sema_b_wait (_c->sema_b, 0, INFINITE,&_c->waiters_b_lock_,&_c->value_b);
+  if (r != 0)
+    return r;
+
+  if (!TryEnterCriticalSection (&_c->waiters_count_lock_))
+  {
+    r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+    if (r != 0)
+      return r;
+    sched_yield();
+    goto tryagain;
+  }
+
+  _c->waiters_count_++;
+  LeaveCriticalSection(&_c->waiters_count_lock_);
+  r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+  if (r != 0)
+    return r;
+
+  ch.c = _c;
+  ch.r = &r;
+  ch.external_mutex = external_mutex;
+
+  pthread_cleanup_push(cleanup_wait, (void *) &ch);
+  r = pthread_mutex_unlock(external_mutex);
+  if (!r)
+    r = do_sema_b_wait (_c->sema_q, 0, INFINITE,&_c->waiters_q_lock_,&_c->value_q);
+
+  pthread_cleanup_pop(1);
+  return r;
+}
+
+static int
+pthread_cond_timedwait_impl (pthread_cond_t *c, pthread_mutex_t *external_mutex, const struct _timespec64 *t, int rel)
+{
+  sCondWaitHelper ch;
+  DWORD dwr;
+  int r;
+  cond_t *_c;
+
+  /* pthread_testcancel(); */
+
+  if (!c || !*c)
+    return EINVAL;
+  _c = (cond_t *)*c;
+  if (_c == (cond_t *)PTHREAD_COND_INITIALIZER)
+  {
+    r = cond_static_init(c);
+    if (r && r != EBUSY)
+      return r;
+    _c = (cond_t *) *c;
+  } else if ((_c)->valid != (unsigned int)LIFE_COND)
+    return EINVAL;
+
+  if (rel == 0)
+  {
+    dwr = dwMilliSecs(_pthread_rel_time_in_ms(t));
+  }
+  else
+  {
+    dwr = dwMilliSecs(_pthread_time_in_ms_from_timespec(t));
+  }
+
+tryagain:
+  r = do_sema_b_wait (_c->sema_b, 0, INFINITE,&_c->waiters_b_lock_,&_c->value_b);
+  if (r != 0)
+    return r;
+
+  if (!TryEnterCriticalSection (&_c->waiters_count_lock_))
+  {
+    r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+    if (r != 0)
+      return r;
+    sched_yield();
+    goto tryagain;
+  }
+
+  _c->waiters_count_++;
+  LeaveCriticalSection(&_c->waiters_count_lock_);
+  r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+  if (r != 0)
+    return r;
+
+  ch.c = _c;
+  ch.r = &r;
+  ch.external_mutex = external_mutex;
+  {
+    pthread_cleanup_push(cleanup_wait, (void *) &ch);
+
+    r = pthread_mutex_unlock(external_mutex);
+    if (!r)
+      r = do_sema_b_wait (_c->sema_q, 0, dwr,&_c->waiters_q_lock_,&_c->value_q);
+
+    pthread_cleanup_pop(1);
+  }
+  return r;
+}
+
+int
+pthread_cond_timedwait64(pthread_cond_t *c, pthread_mutex_t *m, const struct _timespec64 *t)
+{
+  return pthread_cond_timedwait_impl(c, m, t, 0);
+}
+
+int
+pthread_cond_timedwait32(pthread_cond_t *c, pthread_mutex_t *m, const struct _timespec32 *t)
+{
+  struct _timespec64 t64 = {.tv_sec = t->tv_sec, .tv_nsec = t->tv_nsec};
+  return pthread_cond_timedwait_impl(c, m, &t64, 0);
+}
+
+int
+pthread_cond_timedwait64_relative_np(pthread_cond_t *c, pthread_mutex_t *m, const struct _timespec64 *t)
+{
+  return pthread_cond_timedwait_impl(c, m, t, 1);
+}
+
+int
+pthread_cond_timedwait32_relative_np(pthread_cond_t *c, pthread_mutex_t *m, const struct _timespec32 *t)
+{
+  struct _timespec64 t64 = {.tv_sec = t->tv_sec, .tv_nsec = t->tv_nsec};
+  return pthread_cond_timedwait_impl(c, m, &t64, 1);
+}
+
+static void
+cleanup_wait (void *arg)
+{
+  int n, r;
+  sCondWaitHelper *ch = (sCondWaitHelper *) arg;
+  cond_t *_c;
+
+  _c = ch->c;
+  EnterCriticalSection (&_c->waiters_count_lock_);
+  n = _c->waiters_count_unblock_;
+  if (n != 0)
+    _c->waiters_count_unblock_ -= 1;
+  else if ((INT_MAX/2) - 1 == _c->waiters_count_gone_)
+  {
+    _c->waiters_count_gone_ += 1;
+    r = do_sema_b_wait (_c->sema_b, 1, INFINITE,&_c->waiters_b_lock_,&_c->value_b);
+    if (r != 0)
+    {
+      LeaveCriticalSection(&_c->waiters_count_lock_);
+      ch->r[0] = r;
+      return;
+    }
+    _c->waiters_count_ -= _c->waiters_count_gone_;
+    r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+    if (r != 0)
+    {
+      LeaveCriticalSection(&_c->waiters_count_lock_);
+      ch->r[0] = r;
+      return;
+    }
+    _c->waiters_count_gone_ = 0;
+  }
+  else
+    _c->waiters_count_gone_ += 1;
+  LeaveCriticalSection (&_c->waiters_count_lock_);
+
+  if (n == 1)
+  {
+    r = do_sema_b_release (_c->sema_b, 1,&_c->waiters_b_lock_,&_c->value_b);
+    if (r != 0)
+    {
+      ch->r[0] = r;
+      return;
+    }
+  }
+  r = pthread_mutex_lock(ch->external_mutex);
+  if (r != 0)
+    ch->r[0] = r;
+}
+
+static int
+do_sema_b_wait (HANDLE sema, int nointerrupt, DWORD timeout,CRITICAL_SECTION *cs, LONG *val)
+{
+  int r;
+  LONG v;
+  EnterCriticalSection(cs);
+  InterlockedDecrement(val);
+  v = val[0];
+  LeaveCriticalSection(cs);
+  if (v >= 0)
+    return 0;
+  r = do_sema_b_wait_intern (sema, nointerrupt, timeout);
+  EnterCriticalSection(cs);
+  if (r != 0)
+    InterlockedIncrement(val);
+  LeaveCriticalSection(cs);
+  return r;
+}
+
+int
+do_sema_b_wait_intern (HANDLE sema, int nointerrupt, DWORD timeout)
+{
+  HANDLE arr[2];
+  DWORD maxH = 1;
+  int r = 0;
+  DWORD res, dt;
+  if (nointerrupt == 1)
+  {
+    res = _pthread_wait_for_single_object(sema, timeout);
+    switch (res) {
+    case WAIT_TIMEOUT:
+	r = ETIMEDOUT;
+	break;
+    case WAIT_ABANDONED:
+	r = EPERM;
+	break;
+    case WAIT_OBJECT_0:
+	break;
+    default:
+	/*We can only return EINVAL though it might not be posix compliant  */
+	r = EINVAL;
+    }
+    if (r != 0 && r != EINVAL && WaitForSingleObject(sema, 0) == WAIT_OBJECT_0)
+      r = 0;
+    return r;
+  }
+  arr[0] = sema;
+  arr[1] = (HANDLE) pthread_getevent ();
+  if (arr[1] != NULL) maxH += 1;
+  if (maxH == 2)
+  {
+redo:
+      res = _pthread_wait_for_multiple_objects(maxH, arr, 0, timeout);
+      switch (res) {
+      case WAIT_TIMEOUT:
+	  r = ETIMEDOUT;
+	  break;
+      case (WAIT_OBJECT_0 + 1):
+          ResetEvent(arr[1]);
+          if (nointerrupt != 2)
+	    {
+            pthread_testcancel();
+            return EINVAL;
+	    }
+	  pthread_testcancel ();
+	  goto redo;
+      case WAIT_ABANDONED:
+	  r = EPERM;
+	  break;
+      case WAIT_OBJECT_0:
+          r = 0;
+	  break;
+      default:
+	  /*We can only return EINVAL though it might not be posix compliant  */
+	  r = EINVAL;
+      }
+      if (r != 0 && r != EINVAL && WaitForSingleObject(arr[0], 0) == WAIT_OBJECT_0)
+	r = 0;
+      if (r != 0 && nointerrupt != 2 && __pthread_shallcancel ())
+	return EINVAL;
+      return r;
+  }
+  if (timeout == INFINITE)
+  {
+    do {
+      res = _pthread_wait_for_single_object(sema, 40);
+      switch (res) {
+      case WAIT_TIMEOUT:
+	  r = ETIMEDOUT;
+	  break;
+      case WAIT_ABANDONED:
+	  r = EPERM;
+	  break;
+      case WAIT_OBJECT_0:
+          r = 0;
+	  break;
+      default:
+	  /*We can only return EINVAL though it might not be posix compliant  */
+	  r = EINVAL;
+      }
+      if (r != 0 && __pthread_shallcancel ())
+      {
+	if (nointerrupt != 2)
+	  pthread_testcancel();
+	return EINVAL;
+      }
+    } while (r == ETIMEDOUT);
+    if (r != 0 && r != EINVAL && WaitForSingleObject(sema, 0) == WAIT_OBJECT_0)
+      r = 0;
+    return r;
+  }
+  dt = 20;
+  do {
+    if (dt > timeout) dt = timeout;
+    res = _pthread_wait_for_single_object(sema, dt);
+    switch (res) {
+    case WAIT_TIMEOUT:
+	r = ETIMEDOUT;
+	break;
+    case WAIT_ABANDONED:
+	r = EPERM;
+	break;
+    case WAIT_OBJECT_0:
+	r = 0;
+	break;
+    default:
+	/*We can only return EINVAL though it might not be posix compliant  */
+	r = EINVAL;
+    }
+    timeout -= dt;
+    if (timeout != 0 && r != 0 && __pthread_shallcancel ())
+      return EINVAL;
+  } while (r == ETIMEDOUT && timeout != 0);
+  if (r != 0 && r == ETIMEDOUT && WaitForSingleObject(sema, 0) == WAIT_OBJECT_0)
+    r = 0;
+  if (r != 0 && nointerrupt != 2)
+    pthread_testcancel();
+  return r;
+}
+
+static int
+do_sema_b_release(HANDLE sema, LONG count,CRITICAL_SECTION *cs, LONG *val)
+{
+  int wc;
+  EnterCriticalSection(cs);
+  if (((long long) val[0] + (long long) count) > (long long) 0x7fffffffLL)
+  {
+    LeaveCriticalSection(cs);
+    return ERANGE;
+  }
+  wc = -val[0];
+  InterlockedExchangeAdd(val, count);
+  if (wc <= 0 || ReleaseSemaphore(sema, (wc < count ? wc : count), NULL))
+  {
+    LeaveCriticalSection(cs);
+    return 0;
+  }
+  InterlockedExchangeAdd(val, -count);
+  LeaveCriticalSection(cs);
+  return EINVAL;
+}
diff --git a/vendor/winpthreads/src/cond.h b/vendor/winpthreads/src/cond.h
new file mode 100644
index 00000000..fcf7db35
--- /dev/null
+++ b/vendor/winpthreads/src/cond.h
@@ -0,0 +1,57 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_COND_H
+#define WIN_PTHREADS_COND_H
+
+#define CHECK_COND(c)                                                   \
+    do {                                                                \
+        if (!(c) || !*c || (*c == PTHREAD_COND_INITIALIZER)             \
+            || ( ((cond_t *)(*c))->valid != (unsigned int)LIFE_COND ) ) \
+            return EINVAL;                                              \
+    } while (0)
+
+#define LIFE_COND 0xC0BAB1FD
+#define DEAD_COND 0xC0DEADBF
+
+#define STATIC_COND_INITIALIZER(x)		((pthread_cond_t)(x) == ((pthread_cond_t)PTHREAD_COND_INITIALIZER))
+
+typedef struct cond_t cond_t;
+struct cond_t
+{
+    unsigned int valid;
+    int busy;
+    LONG waiters_count_; /* Number of waiting threads.  */
+    LONG waiters_count_unblock_; /* Number of waiting threads whitch can be unblocked.  */
+    LONG waiters_count_gone_; /* Number of waiters which are gone.  */
+    CRITICAL_SECTION waiters_count_lock_; /* Serialize access to <waiters_count_>.  */
+    CRITICAL_SECTION waiters_q_lock_; /* Serialize access to sema_q.  */
+    LONG value_q;
+    CRITICAL_SECTION waiters_b_lock_; /* Serialize access to sema_b.  */
+    LONG value_b;
+    HANDLE sema_q; /* Semaphore used to queue up threads waiting for the condition to
+                 become signaled.  */
+    HANDLE sema_b; /* Semaphore used to queue up threads waiting for the condition which
+                 became signaled.  */
+};
+
+#endif
diff --git a/vendor/winpthreads/src/config.h b/vendor/winpthreads/src/config.h
new file mode 100644
index 00000000..2a0269a3
--- /dev/null
+++ b/vendor/winpthreads/src/config.h
@@ -0,0 +1,5 @@
+/* Minimal hand-written config.h for building vendored winpthreads with a modern toolchain.
+   winpthreads only uses it for optional feature probes; the portable fallbacks are correct. */
+#ifndef WINPTHREADS_CONFIG_H
+#define WINPTHREADS_CONFIG_H
+#endif
diff --git a/vendor/winpthreads/src/libgcc/dll_dependency.S b/vendor/winpthreads/src/libgcc/dll_dependency.S
new file mode 100644
index 00000000..21e1c890
--- /dev/null
+++ b/vendor/winpthreads/src/libgcc/dll_dependency.S
@@ -0,0 +1,90 @@
+/* Implementation for gcc's internal stack-allocation routines.  */
+#if defined(__i386__) || (defined(__x86_64__) && !defined(__arm64ec__))
+.global ___chkstk
+.global	__alloca
+
+.global ___chkstk_ms
+___chkstk_ms:
+#ifdef _WIN64
+	pushq	%rax
+	pushq	%rcx
+	cmpq	$0x1000, %rax
+	leaq	24(%rsp), %rcx
+	jb	.Lchkstk_ms_end
+.Lchkstk_ms_loop:
+	subq	$0x1000, %rcx
+	subq	$0x1000, %rax
+	orq	$0x0, (%rcx)
+	cmpq	$0x1000, %rax
+	ja	.Lchkstk_ms_loop
+.Lchkstk_ms_end:
+	subq	%rax, %rcx
+	orq	$0x0, (%rcx)
+	popq	%rcx
+	popq	%rax
+	ret
+#else
+	pushl	%eax
+	pushl	%ecx
+	cmpl	$0x1000, %eax
+	leal	12(%esp), %ecx
+	jb	chkstk_ms_end
+chkstk_ms_loop:
+	subl	$0x1000, %ecx
+	subl	$0x1000, %eax
+	orl	$0x0, (%ecx)
+	cmpl	$0x1000, %eax
+	ja	chkstk_ms_loop
+chkstk_ms_end:
+	subl	%eax, %ecx
+	orl	$0x0, (%ecx)
+	popl	%ecx
+	popl	%eax
+	ret
+#endif
+
+#ifdef _WIN64
+__alloca:
+	movq	%rcx, %rax
+.align	4
+___chkstk:
+	popq	%r11
+	movq	%rsp, %r10
+	cmpq	$0x1000, %rax
+	jb	.Lchkstk_end
+.Lchkstk_loop:
+	subq	$0x1000, %r10
+	subq	$0x1000, %rax
+	orl	$0x0, (%r10)
+	cmpq	$0x1000, %rax
+	ja	.Lchkstk_loop
+.Lchkstk_end:
+	subq	%rax, %r10
+	movq	%rsp, %rax
+	orl	$0x0, (%r10)
+	movq	%r10, %rsp
+	pushq	%r11
+	ret
+#else
+___chkstk:
+__alloca:
+	pushl	%ecx
+	leal	8(%esp), %ecx
+	cmpl	$0x1000, %eax		/* > 4k ?*/
+	jb	chkstk_end
+chkstk_loop:
+	subl	$0x1000, %ecx
+	subl	$0x1000, %eax
+	orl	$0x0, (%ecx)
+	cmpl	$0x1000, %eax
+	ja	chkstk_loop
+chkstk_end:
+	subl	%eax, %ecx	   
+	orl	$0x0, (%ecx)
+	movl	%esp, %eax
+	movl	%ecx, %esp
+	movl	(%eax), %ecx
+	pushl	4(%eax)
+	ret
+#endif
+#endif
diff --git a/vendor/winpthreads/src/libgcc/dll_math.c b/vendor/winpthreads/src/libgcc/dll_math.c
new file mode 100644
index 00000000..8b9ffaf0
--- /dev/null
+++ b/vendor/winpthreads/src/libgcc/dll_math.c
@@ -0,0 +1,574 @@
+/*-
+ * Copyright (c) 1992, 1993
+ *	The Regents of the University of California.  All rights reserved.
+ *
+ * This software was developed by the Computer Systems Engineering group
+ * at Lawrence Berkeley Laboratory under DARPA contract BG 91-66 and
+ * contributed to Berkeley.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, 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.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 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 OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#ifndef _LIBKERN_QUAD_H_
+#define	_LIBKERN_QUAD_H_
+
+/*
+ * Quad arithmetic.
+ *
+ * This library makes the following assumptions:
+ *
+ *  - The type long long (aka quad_t) exists.
+ *
+ *  - A quad variable is exactly twice as long as `long'.
+ *
+ *  - The machine's arithmetic is two's complement.
+ *
+ * This library can provide 128-bit arithmetic on a machine with 128-bit
+ * quads and 64-bit longs, for instance, or 96-bit arithmetic on machines
+ * with 48-bit longs.
+ */
+/*
+#include <sys/cdefs.h>
+#include <sys/types.h>
+#include <sys/limits.h>
+#include <sys/syslimits.h>
+*/
+
+#include <limits.h>
+typedef long long quad_t;
+typedef unsigned long long u_quad_t;
+typedef unsigned long u_long;
+#ifndef CHAR_BIT
+#define CHAR_BIT __CHAR_BIT__
+#endif
+
+/*
+ * Define the order of 32-bit words in 64-bit words.
+ * For little endian only.
+ */
+#define _QUAD_HIGHWORD 1
+#define _QUAD_LOWWORD 0
+
+/*
+ * Depending on the desired operation, we view a `long long' (aka quad_t) in
+ * one or more of the following formats.
+ */
+union uu {
+	quad_t	q;		/* as a (signed) quad */
+	quad_t	uq;		/* as an unsigned quad */
+	long	sl[2];		/* as two signed longs */
+	u_long	ul[2];		/* as two unsigned longs */
+};
+
+/*
+ * Define high and low longwords.
+ */
+#define	H		_QUAD_HIGHWORD
+#define	L		_QUAD_LOWWORD
+
+/*
+ * Total number of bits in a quad_t and in the pieces that make it up.
+ * These are used for shifting, and also below for halfword extraction
+ * and assembly.
+ */
+#define	QUAD_BITS	(sizeof(quad_t) * CHAR_BIT)
+#define	LONG_BITS	(sizeof(long) * CHAR_BIT)
+#define	HALF_BITS	(sizeof(long) * CHAR_BIT / 2)
+
+/*
+ * Extract high and low shortwords from longword, and move low shortword of
+ * longword to upper half of long, i.e., produce the upper longword of
+ * ((quad_t)(x) << (number_of_bits_in_long/2)).  (`x' must actually be u_long.)
+ *
+ * These are used in the multiply code, to split a longword into upper
+ * and lower halves, and to reassemble a product as a quad_t, shifted left
+ * (sizeof(long)*CHAR_BIT/2).
+ */
+#define	HHALF(x)	((x) >> HALF_BITS)
+#define	LHALF(x)	((x) & ((1 << HALF_BITS) - 1))
+#define	LHUP(x)		((x) << HALF_BITS)
+
+typedef unsigned int	qshift_t;
+
+quad_t		__ashldi3(quad_t, qshift_t);
+quad_t		__ashrdi3(quad_t, qshift_t);
+int		__cmpdi2(quad_t a, quad_t b);
+quad_t		__divdi3(quad_t a, quad_t b);
+quad_t		__lshrdi3(quad_t, qshift_t);
+quad_t		__moddi3(quad_t a, quad_t b);
+u_quad_t	__qdivrem(u_quad_t u, u_quad_t v, u_quad_t *rem);
+u_quad_t	__udivdi3(u_quad_t a, u_quad_t b);
+u_quad_t	__umoddi3(u_quad_t a, u_quad_t b);
+int		__ucmpdi2(u_quad_t a, u_quad_t b);
+quad_t	__divmoddi4(quad_t a, quad_t b, quad_t *rem);
+u_quad_t	__udivmoddi4(u_quad_t a, u_quad_t b, u_quad_t *rem);
+
+#endif /* !_LIBKERN_QUAD_H_ */
+
+#if defined (_X86_) && !defined (__x86_64__)
+/*
+ * Shift a (signed) quad value left (arithmetic shift left).
+ * This is the same as logical shift left!
+ */
+quad_t
+__ashldi3(quad_t a, qshift_t shift)
+{
+	union uu aa;
+
+	aa.q = a;
+	if (shift >= LONG_BITS) {
+		aa.ul[H] = shift >= QUAD_BITS ? 0 :
+		    aa.ul[L] << (shift - LONG_BITS);
+		aa.ul[L] = 0;
+	} else if (shift > 0) {
+		aa.ul[H] = (aa.ul[H] << shift) |
+		    (aa.ul[L] >> (LONG_BITS - shift));
+		aa.ul[L] <<= shift;
+	}
+	return (aa.q);
+}
+
+/*
+ * Shift a (signed) quad value right (arithmetic shift right).
+ */
+quad_t
+__ashrdi3(quad_t a, qshift_t shift)
+{
+	union uu aa;
+
+	aa.q = a;
+	if (shift >= LONG_BITS) {
+		long s;
+
+		/*
+		 * Smear bits rightward using the machine's right-shift
+		 * method, whether that is sign extension or zero fill,
+		 * to get the `sign word' s.  Note that shifting by
+		 * LONG_BITS is undefined, so we shift (LONG_BITS-1),
+		 * then 1 more, to get our answer.
+		 */
+		s = (aa.sl[H] >> (LONG_BITS - 1)) >> 1;
+		aa.ul[L] = shift >= QUAD_BITS ? s :
+		    aa.sl[H] >> (shift - LONG_BITS);
+		aa.ul[H] = s;
+	} else if (shift > 0) {
+		aa.ul[L] = (aa.ul[L] >> shift) |
+		    (aa.ul[H] << (LONG_BITS - shift));
+		aa.sl[H] >>= shift;
+	}
+	return (aa.q);
+}
+
+/*
+ * Return 0, 1, or 2 as a <, =, > b respectively.
+ * Both a and b are considered signed---which means only the high word is
+ * signed.
+ */
+int
+__cmpdi2(quad_t a, quad_t b)
+{
+	union uu aa, bb;
+
+	aa.q = a;
+	bb.q = b;
+	return (aa.sl[H] < bb.sl[H] ? 0 : aa.sl[H] > bb.sl[H] ? 2 :
+	    aa.ul[L] < bb.ul[L] ? 0 : aa.ul[L] > bb.ul[L] ? 2 : 1);
+}
+
+/*
+ * Divide two signed quads.
+ * ??? if -1/2 should produce -1 on this machine, this code is wrong
+ */
+quad_t
+__divdi3(quad_t a, quad_t b)
+{
+	u_quad_t ua, ub, uq;
+	int neg;
+
+	if (a < 0)
+		ua = -(u_quad_t)a, neg = 1;
+	else
+		ua = a, neg = 0;
+	if (b < 0)
+		ub = -(u_quad_t)b, neg ^= 1;
+	else
+		ub = b;
+	uq = __qdivrem(ua, ub, (u_quad_t *)0);
+	return (neg ? -uq : uq);
+}
+
+/*
+ * Shift an (unsigned) quad value right (logical shift right).
+ */
+quad_t
+__lshrdi3(quad_t a, qshift_t shift)
+{
+	union uu aa;
+
+	aa.q = a;
+	if (shift >= LONG_BITS) {
+		aa.ul[L] = shift >= QUAD_BITS ? 0 :
+		    aa.ul[H] >> (shift - LONG_BITS);
+		aa.ul[H] = 0;
+	} else if (shift > 0) {
+		aa.ul[L] = (aa.ul[L] >> shift) |
+		    (aa.ul[H] << (LONG_BITS - shift));
+		aa.ul[H] >>= shift;
+	}
+	return (aa.q);
+}
+
+/*
+ * Return remainder after dividing two signed quads.
+ *
+ * XXX
+ * If -1/2 should produce -1 on this machine, this code is wrong.
+ */
+quad_t
+__moddi3(quad_t a, quad_t b)
+{
+	u_quad_t ua, ub, ur;
+	int neg;
+
+	if (a < 0)
+		ua = -(u_quad_t)a, neg = 1;
+	else
+		ua = a, neg = 0;
+	if (b < 0)
+		ub = -(u_quad_t)b;
+	else
+		ub = b;
+	(void)__qdivrem(ua, ub, &ur);
+	return (neg ? -ur : ur);
+}
+
+
+/*
+ * Multiprecision divide.  This algorithm is from Knuth vol. 2 (2nd ed),
+ * section 4.3.1, pp. 257--259.
+ */
+
+#define	B	(1 << HALF_BITS)	/* digit base */
+
+/* Combine two `digits' to make a single two-digit number. */
+#define	COMBINE(a, b) (((u_long)(a) << HALF_BITS) | (b))
+
+/* select a type for digits in base B: use unsigned short if they fit */
+#if ULONG_MAX == 0xffffffff && USHRT_MAX >= 0xffff
+typedef unsigned short digit;
+#else
+typedef u_long digit;
+#endif
+
+/*
+ * Shift p[0]..p[len] left `sh' bits, ignoring any bits that
+ * `fall out' the left (there never will be any such anyway).
+ * We may assume len >= 0.  NOTE THAT THIS WRITES len+1 DIGITS.
+ */
+static void
+__shl(register digit *p, register int len, register int sh)
+{
+	register int i;
+
+	for (i = 0; i < len; i++)
+		p[i] = LHALF(p[i] << sh) | (p[i + 1] >> (HALF_BITS - sh));
+	p[i] = LHALF(p[i] << sh);
+}
+
+/*
+ * __qdivrem(u, v, rem) returns u/v and, optionally, sets *rem to u%v.
+ *
+ * We do this in base 2-sup-HALF_BITS, so that all intermediate products
+ * fit within u_long.  As a consequence, the maximum length dividend and
+ * divisor are 4 `digits' in this base (they are shorter if they have
+ * leading zeros).
+ */
+u_quad_t
+__qdivrem(u_quad_t uq, u_quad_t vq, u_quad_t *arq)
+{
+	union uu tmp;
+	digit *u, *v, *q;
+	register digit v1, v2;
+	u_long qhat, rhat, t;
+	int m, n, d, j, i;
+	digit uspace[5], vspace[5], qspace[5];
+
+	/*
+	 * Take care of special cases: divide by zero, and u < v.
+	 */
+	if (vq == 0) {
+		/* divide by zero. */
+		static volatile const unsigned int zero = 0;
+
+		tmp.ul[H] = tmp.ul[L] = 1 / zero;
+		if (arq)
+			*arq = uq;
+		return (tmp.q);
+	}
+	if (uq < vq) {
+		if (arq)
+			*arq = uq;
+		return (0);
+	}
+	u = &uspace[0];
+	v = &vspace[0];
+	q = &qspace[0];
+
+	/*
+	 * Break dividend and divisor into digits in base B, then
+	 * count leading zeros to determine m and n.  When done, we
+	 * will have:
+	 *	u = (u[1]u[2]...u[m+n]) sub B
+	 *	v = (v[1]v[2]...v[n]) sub B
+	 *	v[1] != 0
+	 *	1 < n <= 4 (if n = 1, we use a different division algorithm)
+	 *	m >= 0 (otherwise u < v, which we already checked)
+	 *	m + n = 4
+	 * and thus
+	 *	m = 4 - n <= 2
+	 */
+	tmp.uq = uq;
+	u[0] = 0;
+	u[1] = HHALF(tmp.ul[H]);
+	u[2] = LHALF(tmp.ul[H]);
+	u[3] = HHALF(tmp.ul[L]);
+	u[4] = LHALF(tmp.ul[L]);
+	tmp.uq = vq;
+	v[1] = HHALF(tmp.ul[H]);
+	v[2] = LHALF(tmp.ul[H]);
+	v[3] = HHALF(tmp.ul[L]);
+	v[4] = LHALF(tmp.ul[L]);
+	for (n = 4; v[1] == 0; v++) {
+		if (--n == 1) {
+			u_long rbj;	/* r*B+u[j] (not root boy jim) */
+			digit q1, q2, q3, q4;
+
+			/*
+			 * Change of plan, per exercise 16.
+			 *	r = 0;
+			 *	for j = 1..4:
+			 *		q[j] = floor((r*B + u[j]) / v),
+			 *		r = (r*B + u[j]) % v;
+			 * We unroll this completely here.
+			 */
+			t = v[2];	/* nonzero, by definition */
+			q1 = u[1] / t;
+			rbj = COMBINE(u[1] % t, u[2]);
+			q2 = rbj / t;
+			rbj = COMBINE(rbj % t, u[3]);
+			q3 = rbj / t;
+			rbj = COMBINE(rbj % t, u[4]);
+			q4 = rbj / t;
+			if (arq)
+				*arq = rbj % t;
+			tmp.ul[H] = COMBINE(q1, q2);
+			tmp.ul[L] = COMBINE(q3, q4);
+			return (tmp.q);
+		}
+	}
+
+	/*
+	 * By adjusting q once we determine m, we can guarantee that
+	 * there is a complete four-digit quotient at &qspace[1] when
+	 * we finally stop.
+	 */
+	for (m = 4 - n; u[1] == 0; u++)
+		m--;
+	for (i = 4 - m; --i >= 0;)
+		q[i] = 0;
+	q += 4 - m;
+
+	/*
+	 * Here we run Program D, translated from MIX to C and acquiring
+	 * a few minor changes.
+	 *
+	 * D1: choose multiplier 1 << d to ensure v[1] >= B/2.
+	 */
+	d = 0;
+	for (t = v[1]; t < B / 2; t <<= 1)
+		d++;
+	if (d > 0) {
+		__shl(&u[0], m + n, d);		/* u <<= d */
+		__shl(&v[1], n - 1, d);		/* v <<= d */
+	}
+	/*
+	 * D2: j = 0.
+	 */
+	j = 0;
+	v1 = v[1];	/* for D3 -- note that v[1..n] are constant */
+	v2 = v[2];	/* for D3 */
+	do {
+		register digit uj0, uj1, uj2;
+
+		/*
+		 * D3: Calculate qhat (\^q, in TeX notation).
+		 * Let qhat = min((u[j]*B + u[j+1])/v[1], B-1), and
+		 * let rhat = (u[j]*B + u[j+1]) mod v[1].
+		 * While rhat < B and v[2]*qhat > rhat*B+u[j+2],
+		 * decrement qhat and increase rhat correspondingly.
+		 * Note that if rhat >= B, v[2]*qhat < rhat*B.
+		 */
+		uj0 = u[j + 0];	/* for D3 only -- note that u[j+...] change */
+		uj1 = u[j + 1];	/* for D3 only */
+		uj2 = u[j + 2];	/* for D3 only */
+		if (uj0 == v1) {
+			qhat = B;
+			rhat = uj1;
+			goto qhat_too_big;
+		} else {
+			u_long nn = COMBINE(uj0, uj1);
+			qhat = nn / v1;
+			rhat = nn % v1;
+		}
+		while (v2 * qhat > COMBINE(rhat, uj2)) {
+	qhat_too_big:
+			qhat--;
+			if ((rhat += v1) >= B)
+				break;
+		}
+		/*
+		 * D4: Multiply and subtract.
+		 * The variable `t' holds any borrows across the loop.
+		 * We split this up so that we do not require v[0] = 0,
+		 * and to eliminate a final special case.
+		 */
+		for (t = 0, i = n; i > 0; i--) {
+			t = u[i + j] - v[i] * qhat - t;
+			u[i + j] = LHALF(t);
+			t = (B - HHALF(t)) & (B - 1);
+		}
+		t = u[j] - t;
+		u[j] = LHALF(t);
+		/*
+		 * D5: test remainder.
+		 * There is a borrow if and only if HHALF(t) is nonzero;
+		 * in that (rare) case, qhat was too large (by exactly 1).
+		 * Fix it by adding v[1..n] to u[j..j+n].
+		 */
+		if (HHALF(t)) {
+			qhat--;
+			for (t = 0, i = n; i > 0; i--) { /* D6: add back. */
+				t += u[i + j] + v[i];
+				u[i + j] = LHALF(t);
+				t = HHALF(t);
+			}
+			u[j] = LHALF(u[j] + t);
+		}
+		q[j] = qhat;
+	} while (++j <= m);		/* D7: loop on j. */
+
+	/*
+	 * If caller wants the remainder, we have to calculate it as
+	 * u[m..m+n] >> d (this is at most n digits and thus fits in
+	 * u[m+1..m+n], but we may need more source digits).
+	 */
+	if (arq) {
+		if (d) {
+			for (i = m + n; i > m; --i)
+				u[i] = (u[i] >> d) |
+				    LHALF(u[i - 1] << (HALF_BITS - d));
+			u[i] = 0;
+		}
+		tmp.ul[H] = COMBINE(uspace[1], uspace[2]);
+		tmp.ul[L] = COMBINE(uspace[3], uspace[4]);
+		*arq = tmp.q;
+	}
+
+	tmp.ul[H] = COMBINE(qspace[1], qspace[2]);
+	tmp.ul[L] = COMBINE(qspace[3], qspace[4]);
+	return (tmp.q);
+}
+
+/*
+ * Return 0, 1, or 2 as a <, =, > b respectively.
+ * Neither a nor b are considered signed.
+ */
+int
+__ucmpdi2(u_quad_t a, u_quad_t b)
+{
+	union uu aa, bb;
+
+	aa.uq = a;
+	bb.uq = b;
+	return (aa.ul[H] < bb.ul[H] ? 0 : aa.ul[H] > bb.ul[H] ? 2 :
+	    aa.ul[L] < bb.ul[L] ? 0 : aa.ul[L] > bb.ul[L] ? 2 : 1);
+}
+
+/*
+ * Divide two unsigned quads.
+ */
+u_quad_t
+__udivdi3(u_quad_t a, u_quad_t b)
+{
+
+	return (__qdivrem(a, b, (u_quad_t *)0));
+}
+
+/*
+ * Return remainder after dividing two unsigned quads.
+ */
+u_quad_t
+__umoddi3(u_quad_t a, u_quad_t b)
+{
+	u_quad_t r;
+
+	(void)__qdivrem(a, b, &r);
+	return (r);
+}
+
+/*
+ * Divide two signed quads.
+ * This function is new in GCC 7.
+ */
+quad_t
+__divmoddi4(quad_t a, quad_t b, quad_t *rem)
+{
+	u_quad_t ua, ub, uq, ur;
+	int negq, negr;
+
+	if (a < 0)
+		ua = -(u_quad_t)a, negq = 1, negr = 1;
+	else
+		ua = a, negq = 0, negr = 0;
+	if (b < 0)
+		ub = -(u_quad_t)b, negq ^= 1;
+	else
+		ub = b;
+	uq = __qdivrem(ua, ub, &ur);
+	if (rem)
+		*rem = (negr ? -ur : ur);
+	return (negq ? -uq : uq);
+}
+
+u_quad_t
+__udivmoddi4(u_quad_t a, u_quad_t b, u_quad_t *rem)
+{
+  return __qdivrem(a, b, rem);
+}
+
+#else
+static int __attribute__((unused)) dummy;
+#endif /*deined (_X86_) && !defined (__x86_64__)*/
+
diff --git a/vendor/winpthreads/src/misc.c b/vendor/winpthreads/src/misc.c
new file mode 100644
index 00000000..b2d5cebc
--- /dev/null
+++ b/vendor/winpthreads/src/misc.c
@@ -0,0 +1,234 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+/* public header files */
+#include "pthread.h"
+/* internal header files */
+#include "misc.h"
+
+void (WINAPI *_pthread_get_system_time_best_as_file_time) (LPFILETIME) = NULL;
+static ULONGLONG (WINAPI *_pthread_get_tick_count_64) (VOID);
+HRESULT (WINAPI *_pthread_set_thread_description) (HANDLE, PCWSTR) = NULL;
+BOOL (WINAPI *_pthread_get_handle_information) (HANDLE, LPDWORD) = NULL;
+
+#if defined(__GNUC__) || defined(__clang__)
+#if __GNUC__ >= 9 && !defined(__clang__)
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wprio-ctor-dtor"
+#endif
+__attribute__((constructor(0)))
+#endif
+static void winpthreads_init(void)
+{
+    HMODULE mod = GetModuleHandleA("kernel32.dll");
+    if (mod)
+    {
+        _pthread_get_handle_information =
+            (BOOL (WINAPI *)(HANDLE, LPDWORD))(void*) GetProcAddress(mod, "GetHandleInformation");
+
+        _pthread_get_tick_count_64 =
+            (ULONGLONG (WINAPI *)(VOID))(void*) GetProcAddress(mod, "GetTickCount64");
+
+        _pthread_set_thread_description =
+            (HRESULT (WINAPI *)(HANDLE, PCWSTR))(void*) GetProcAddress(mod, "SetThreadDescription");
+
+        /* <1us precision on Windows 10 */
+        _pthread_get_system_time_best_as_file_time =
+            (void (WINAPI *)(LPFILETIME))(void*) GetProcAddress(mod, "GetSystemTimePreciseAsFileTime");
+    }
+
+    if (!_pthread_get_system_time_best_as_file_time)
+        /* >15ms precision on Windows 10 */
+        _pthread_get_system_time_best_as_file_time = GetSystemTimeAsFileTime;
+
+    /* Although SetThreadDescription lives in kernel32.dll, on Windows Server 2016,
+     * Windows 10 LTSB 2016 and Windows 10 version 1607, it was only available in
+     * kernelbase.dll. So, load it from there for maximum coverage.
+     */
+    if (!_pthread_set_thread_description)
+    {
+        mod = GetModuleHandleA("kernelbase.dll");
+        if (mod)
+        {
+            _pthread_set_thread_description =
+                (HRESULT (WINAPI *)(HANDLE, PCWSTR))(void*) GetProcAddress(mod, "SetThreadDescription");
+        }
+    }
+}
+#if defined(__GNUC__) && __GNUC__ >= 9 && !defined(__clang__)
+#pragma GCC diagnostic pop
+#endif
+
+#if defined(_MSC_VER) && !defined(__clang__)
+/* Force a reference to __xc_t to prevent whole program optimization
+ * from discarding the variable. */
+
+/* On x86, symbols are prefixed with an underscore. */
+# if defined(_M_IX86)
+#   pragma comment(linker, "/include:___xc_t")
+# else
+#   pragma comment(linker, "/include:__xc_t")
+# endif
+
+#pragma section(".CRT$XCT", long, read)
+__declspec(allocate(".CRT$XCT"))
+extern const _PVFV __xc_t;
+const _PVFV __xc_t = winpthreads_init;
+#endif
+
+unsigned long long _pthread_time_in_ms(void)
+{
+    FILETIME ft;
+
+    GetSystemTimeAsFileTime(&ft);
+    return (((unsigned long long)ft.dwHighDateTime << 32) + ft.dwLowDateTime
+            - 0x19DB1DED53E8000ULL) / 10000ULL;
+}
+
+unsigned long long _pthread_time_in_ms_from_timespec(const struct _timespec64 *ts)
+{
+    unsigned long long t = (unsigned long long) ts->tv_sec * 1000LL;
+    /* The +999999 is here to ensure that the division always rounds up */
+    t += (unsigned long long) (ts->tv_nsec + 999999) / 1000000;
+
+    return t;
+}
+
+unsigned long long _pthread_rel_time_in_ms(const struct _timespec64 *ts)
+{
+    unsigned long long t1 = _pthread_time_in_ms_from_timespec(ts);
+    unsigned long long t2 = _pthread_time_in_ms();
+
+    /* Prevent underflow */
+    if (t1 < t2) return 0;
+    return t1 - t2;
+}
+
+static unsigned long long
+_pthread_get_tick_count (long long *frequency)
+{
+  if (_pthread_get_tick_count_64 != NULL)
+    return _pthread_get_tick_count_64 ();
+
+  LARGE_INTEGER freq, timestamp;
+
+  if (*frequency == 0)
+  {
+    if (QueryPerformanceFrequency (&freq))
+      *frequency = freq.QuadPart;
+    else
+      *frequency = -1;
+  }
+
+  if (*frequency > 0 && QueryPerformanceCounter (&timestamp))
+    return timestamp.QuadPart / (*frequency / 1000);
+
+  /* Fallback */
+  return GetTickCount ();
+}
+
+/* A wrapper around WaitForSingleObject() that ensures that
+ * the wait function does not time out before the time
+ * actually runs out. This is needed because WaitForSingleObject()
+ * might have poor accuracy, returning earlier than expected.
+ * On the other hand, returning a bit *later* than expected
+ * is acceptable in a preemptive multitasking environment.
+ */
+unsigned long
+_pthread_wait_for_single_object (void *handle, unsigned long timeout)
+{
+  DWORD result;
+  unsigned long long start_time, end_time;
+  unsigned long wait_time;
+  long long frequency = 0;
+
+  if (timeout == INFINITE || timeout == 0)
+    return WaitForSingleObject ((HANDLE) handle, (DWORD) timeout);
+
+  start_time = _pthread_get_tick_count (&frequency);
+  end_time = start_time + timeout;
+  wait_time = timeout;
+
+  do
+  {
+    unsigned long long current_time;
+
+    result = WaitForSingleObject ((HANDLE) handle, (DWORD) wait_time);
+    if (result != WAIT_TIMEOUT)
+      break;
+
+    current_time = _pthread_get_tick_count (&frequency);
+    if (current_time >= end_time)
+      break;
+
+    wait_time = (DWORD) (end_time - current_time);
+  } while (TRUE);
+
+  return result;
+}
+
+/* A wrapper around WaitForMultipleObjects() that ensures that
+ * the wait function does not time out before the time
+ * actually runs out. This is needed because WaitForMultipleObjects()
+ * might have poor accuracy, returning earlier than expected.
+ * On the other hand, returning a bit *later* than expected
+ * is acceptable in a preemptive multitasking environment.
+ */
+unsigned long
+_pthread_wait_for_multiple_objects (unsigned long count, void **handles, unsigned int all, unsigned long timeout)
+{
+  DWORD result;
+  unsigned long long start_time, end_time;
+  unsigned long wait_time;
+  long long frequency = 0;
+
+  if (timeout == INFINITE || timeout == 0)
+    return WaitForMultipleObjects ((DWORD) count, (HANDLE *) handles, all, (DWORD) timeout);
+
+  start_time = _pthread_get_tick_count (&frequency);
+  end_time = start_time + timeout;
+  wait_time = timeout;
+
+  do
+  {
+    unsigned long long current_time;
+
+    result = WaitForMultipleObjects ((DWORD) count, (HANDLE *) handles, all, (DWORD) wait_time);
+    if (result != WAIT_TIMEOUT)
+      break;
+
+    current_time = _pthread_get_tick_count (&frequency);
+    if (current_time >= end_time)
+      break;
+
+    wait_time = (DWORD) (end_time - current_time);
+  } while (TRUE);
+
+  return result;
+}
diff --git a/vendor/winpthreads/src/misc.h b/vendor/winpthreads/src/misc.h
new file mode 100644
index 00000000..dc11eaff
--- /dev/null
+++ b/vendor/winpthreads/src/misc.h
@@ -0,0 +1,112 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_MISC_H
+#define WIN_PTHREADS_MISC_H
+
+#include <limits.h>
+/* public header files */
+#include "pthread_compat.h"
+
+#define PTR2INT(x)	((int)(uintptr_t)(x))
+
+#if SIZE_MAX>UINT_MAX
+typedef long long LONGBAG;
+#else
+typedef long LONGBAG;
+#endif
+
+extern BOOL (WINAPI *_pthread_get_handle_information) (HANDLE, LPDWORD);
+
+/* For gcc and clang define DUMMY_WRITABLE_DWORD as C99 compound literal.
+ * For other pre-C99 compilers declare DUMMY_WRITABLE_DWORD as static variable.
+ */
+#if defined(__GNUC__) || defined(__clang__)
+#define DUMMY_WRITABLE_DWORD (DWORD){0}
+#else
+static DWORD DUMMY_WRITABLE_DWORD;
+#endif
+
+#define TEST_HANDLE(h)                                                  \
+  (((h) != NULL && (h) != INVALID_HANDLE_VALUE) && (                    \
+     _pthread_get_handle_information == NULL ||                         \
+     _pthread_get_handle_information((h), &DUMMY_WRITABLE_DWORD) ||     \
+     GetLastError() == ERROR_CALL_NOT_IMPLEMENTED                       \
+  ))
+
+#define CHECK_HANDLE2(h, e)                                             \
+  do {                                                                  \
+    if (!TEST_HANDLE(h))                                                \
+      return e;                                                         \
+  } while (0)
+
+#define CHECK_HANDLE(h) CHECK_HANDLE2(h, EINVAL)
+
+#define CHECK_PTR(p) do { if (!(p)) return EINVAL; } while (0)
+
+#define UPD_RESULT(x,r) do { int _r = (x); (r) = (r) ? (r) : _r; } while (0)
+
+#define CHECK_THREAD(t)                         \
+  do {                                          \
+    CHECK_PTR(t);                               \
+    CHECK_HANDLE((t)->h);                       \
+  } while (0)
+
+#define CHECK_OBJECT(o, e)                                              \
+  do {                                                                  \
+    if (!(o)) return e;                                                 \
+    CHECK_HANDLE2((o)->h, e);                                           \
+  } while (0)
+
+#define VALID(x)    if (!(p)) return EINVAL;
+
+/* ms can be 64 bit, solve wrap-around issues: */
+static WINPTHREADS_INLINE unsigned long dwMilliSecs(unsigned long long ms)
+{
+  if (ms >= 0xffffffffULL) return 0xfffffffful;
+  return (unsigned long) ms;
+}
+
+unsigned long long _pthread_time_in_ms(void);
+unsigned long long _pthread_time_in_ms_from_timespec(const struct _timespec64 *ts);
+unsigned long long _pthread_rel_time_in_ms(const struct _timespec64 *ts);
+unsigned long _pthread_wait_for_single_object (void *handle, unsigned long timeout);
+unsigned long _pthread_wait_for_multiple_objects (unsigned long count, void **handles, unsigned int all, unsigned long timeout);
+
+extern void (WINAPI *_pthread_get_system_time_best_as_file_time) (LPFILETIME);
+extern HRESULT (WINAPI *_pthread_set_thread_description) (HANDLE, PCWSTR);
+
+#if defined(__GNUC__) || defined(__clang__)
+#define likely(cond) __builtin_expect((cond) != 0, 1)
+#define unlikely(cond) __builtin_expect((cond) != 0, 0)
+#else
+#define likely(cond) (cond)
+#define unlikely(cond) (cond)
+#endif
+
+#if defined(__GNUC__) || defined(__clang__)
+#define UNREACHABLE() __builtin_unreachable()
+#elif defined(_MSC_VER)
+#define UNREACHABLE() __assume(0)
+#endif
+
+#endif
diff --git a/vendor/winpthreads/src/mutex.c b/vendor/winpthreads/src/mutex.c
new file mode 100644
index 00000000..ae478586
--- /dev/null
+++ b/vendor/winpthreads/src/mutex.c
@@ -0,0 +1,403 @@
+/*
+   Copyright (c) 2011, 2014 mingw-w64 project
+   Copyright (c) 2015 Intel Corporation
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <malloc.h>
+#include <stdio.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define WINPTHREAD_MUTEX_DECL WINPTHREAD_API
+
+/* public header files */
+#include "pthread.h"
+/* internal header files */
+#include "misc.h"
+
+typedef enum {
+  Unlocked,        /* Not locked. */
+  Locked,          /* Locked but without waiters. */
+  Waiting,         /* Locked, may have waiters. */
+} mutex_state_t;
+
+typedef enum {
+  Normal,
+  Errorcheck,
+  Recursive,
+} mutex_type_t;
+
+/* The heap-allocated part of a mutex. */
+typedef struct {
+  mutex_state_t state;
+  mutex_type_t type;
+  HANDLE event;       /* Auto-reset event, or NULL if not yet allocated. */
+  unsigned rec_lock;  /* For recursive mutexes, the number of times the
+                         mutex has been locked in excess by the same thread. */
+  volatile DWORD owner;  /* For recursive and error-checking mutexes, the
+                            ID of the owning thread if the mutex is locked. */
+} mutex_impl_t;
+
+/* Whether a mutex is still a static initializer (not a pointer to
+   a mutex_impl_t). */
+static BOOL
+is_static_initializer(pthread_mutex_t m)
+{
+  /* Treat 0 as a static initializer as well (for normal mutexes),
+     to tolerate sloppy code in libgomp. (We should rather fix that code!) */
+  intptr_t v = (intptr_t)m;
+  return v >= -3 && v <= 0;
+/* Should be simple:
+  return (uintptr_t)m >= (uintptr_t)-3; */
+}
+
+/* Create and return the implementation part of a mutex from a static
+   initialiser. Return NULL on out-of-memory error. */
+static WINPTHREADS_ATTRIBUTE((noinline)) mutex_impl_t *
+mutex_impl_init(pthread_mutex_t *m, mutex_impl_t *mi)
+{
+  mutex_impl_t *new_mi = malloc(sizeof(mutex_impl_t));
+  if (new_mi == NULL)
+    return NULL;
+  new_mi->state = Unlocked;
+  new_mi->type = (mi == (void *)PTHREAD_RECURSIVE_MUTEX_INITIALIZER ? Recursive
+                  : mi == (void *)PTHREAD_ERRORCHECK_MUTEX_INITIALIZER ? Errorcheck
+                  : Normal);
+  new_mi->event = NULL;
+  new_mi->rec_lock = 0;
+  new_mi->owner = (DWORD)-1;
+  if (InterlockedCompareExchangePointer((PVOID volatile *)m, new_mi, mi) == mi) {
+    return new_mi;
+  } else {
+    /* Someone created the struct before us. */
+    free(new_mi);
+    return (mutex_impl_t *)*m;
+  }
+}
+
+/* Return the implementation part of a mutex, creating it if necessary.
+   Return NULL on out-of-memory error. */
+static WINPTHREADS_INLINE mutex_impl_t *
+mutex_impl(pthread_mutex_t *m)
+{
+  mutex_impl_t *mi = (mutex_impl_t *)*m;
+  if (is_static_initializer((pthread_mutex_t)mi)) {
+    return mutex_impl_init(m, mi);
+  } else {
+    /* mi cannot be null here; avoid a test in the fast path. */
+    if (mi == NULL)
+      UNREACHABLE();
+    return mi;
+  }
+}
+
+/* Lock a mutex. Give up after 'timeout' ms (with ETIMEDOUT),
+   or never if timeout=INFINITE. */
+static WINPTHREADS_INLINE int
+pthread_mutex_lock_intern (pthread_mutex_t *m, DWORD timeout)
+{
+  mutex_impl_t *mi = mutex_impl(m);
+  if (mi == NULL)
+    return ENOMEM;
+  mutex_state_t old_state = InterlockedExchange((long *)&mi->state, Locked);
+  if (unlikely(old_state != Unlocked)) {
+    /* The mutex is already locked. */
+
+    if (mi->type != Normal) {
+      /* Recursive or Errorcheck */
+      if (mi->owner == GetCurrentThreadId()) {
+        /* FIXME: A recursive mutex should not need two atomic ops when locking
+           recursively.  We could rewrite by doing compare-and-swap instead of
+           test-and-set the first time, but it would lead to more code
+           duplication and add a conditional branch to the critical path. */
+        InterlockedCompareExchange((long *)&mi->state, old_state, Locked);
+        if (mi->type == Recursive) {
+          mi->rec_lock++;
+          return 0;
+        } else {
+          /* type == Errorcheck */
+          return EDEADLK;
+        }
+      }
+    }
+
+    /* Make sure there is an event object on which to wait. */
+    if (mi->event == NULL) {
+      /* Make an auto-reset event object. */
+      HANDLE ev = CreateEvent(NULL, FALSE, FALSE, NULL);
+      if (ev == NULL) {
+        switch (GetLastError()) {
+        case ERROR_ACCESS_DENIED:
+          return EPERM;
+        default:
+          return ENOMEM;    /* Probably accurate enough. */
+        }
+      }
+      if (InterlockedCompareExchangePointer(&mi->event, ev, NULL) != NULL) {
+        /* Someone created the event before us. */
+        CloseHandle(ev);
+      }
+    }
+
+    /* At this point, mi->event is non-NULL. */
+
+    while (InterlockedExchange((long *)&mi->state, Waiting) != Unlocked) {
+      /* For timed locking attempts, it is possible (although unlikely)
+         that we are woken up but someone else grabs the lock before us,
+         and we have to go back to sleep again. In that case, the total
+         wait may be longer than expected. */
+
+      unsigned r = _pthread_wait_for_single_object(mi->event, timeout);
+      switch (r) {
+      case WAIT_TIMEOUT:
+        return ETIMEDOUT;
+      case WAIT_OBJECT_0:
+        break;
+      default:
+        return EINVAL;
+      }
+    }
+  }
+
+  if (mi->type != Normal)
+    mi->owner = GetCurrentThreadId();
+
+  return 0;
+}
+
+int
+pthread_mutex_lock (pthread_mutex_t *m)
+{
+  return pthread_mutex_lock_intern (m, INFINITE);
+}
+
+/* Internal version which always uses `struct _timespec64`. */
+static int __pthread_mutex_timedlock(pthread_mutex_t *m, const struct _timespec64 *ts)
+{
+  unsigned long long patience;
+  if (ts != NULL) {
+    unsigned long long end = _pthread_time_in_ms_from_timespec(ts);
+    unsigned long long now = _pthread_time_in_ms();
+    patience = end > now ? end - now : 0;
+    if (patience > 0xffffffff)
+      patience = INFINITE;
+  } else {
+    patience = INFINITE;
+  }
+  return pthread_mutex_lock_intern(m, patience);
+}
+
+int pthread_mutex_timedlock64(pthread_mutex_t *m, const struct _timespec64 *ts)
+{
+  return __pthread_mutex_timedlock (m, ts);
+}
+
+int pthread_mutex_timedlock32(pthread_mutex_t *m, const struct _timespec32 *ts)
+{
+  struct _timespec64 ts64 = {.tv_sec = ts->tv_sec, .tv_nsec = ts->tv_nsec};
+  return __pthread_mutex_timedlock (m, &ts64);
+}
+
+int pthread_mutex_unlock(pthread_mutex_t *m)
+{
+  /* Here m might an initialiser of an error-checking or recursive mutex, in
+     which case the behaviour is well-defined, so we can't skip this check. */
+  mutex_impl_t *mi = mutex_impl(m);
+  if (mi == NULL)
+    return ENOMEM;
+
+  if (unlikely(mi->type != Normal)) {
+    if (mi->state == Unlocked)
+      return EPERM;
+    if (mi->owner != GetCurrentThreadId())
+      return EPERM;
+    if (mi->rec_lock > 0) {
+      mi->rec_lock--;
+      return 0;
+    }
+    mi->owner = (DWORD)-1;
+  }
+  if (unlikely(InterlockedExchange((long *)&mi->state, Unlocked) == Waiting)) {
+    if (!SetEvent(mi->event))
+      return EPERM;
+  }
+  return 0;
+}
+
+int pthread_mutex_trylock(pthread_mutex_t *m)
+{
+  mutex_impl_t *mi = mutex_impl(m);
+  if (mi == NULL)
+    return ENOMEM;
+
+  if (InterlockedCompareExchange((long *)&mi->state, Locked, Unlocked) == Unlocked) {
+    if (mi->type != Normal)
+      mi->owner = GetCurrentThreadId();
+    return 0;
+  } else {
+    if (mi->type == Recursive && mi->owner == GetCurrentThreadId()) {
+      mi->rec_lock++;
+      return 0;
+    }
+    return EBUSY;
+  }
+}
+
+int
+pthread_mutex_init (pthread_mutex_t *m, const pthread_mutexattr_t *a)
+{
+  pthread_mutex_t init = PTHREAD_MUTEX_INITIALIZER;
+  if (a != NULL) {
+    int pshared;
+    if (pthread_mutexattr_getpshared(a, &pshared) == 0
+        && pshared == PTHREAD_PROCESS_SHARED)
+      return ENOSYS;
+
+    int type;
+    if (pthread_mutexattr_gettype(a, &type) == 0) {
+      switch (type) {
+      case PTHREAD_MUTEX_ERRORCHECK:
+        init = PTHREAD_ERRORCHECK_MUTEX_INITIALIZER;
+        break;
+      case PTHREAD_MUTEX_RECURSIVE:
+        init = PTHREAD_RECURSIVE_MUTEX_INITIALIZER;
+        break;
+      default:
+        init = PTHREAD_MUTEX_INITIALIZER;
+        break;
+      }
+    }
+  }
+  *m = init;
+  return 0;
+}
+
+int pthread_mutex_destroy (pthread_mutex_t *m)
+{
+  mutex_impl_t *mi = (mutex_impl_t *)*m;
+  if (!is_static_initializer((pthread_mutex_t)mi)) {
+    if (mi->event != NULL)
+      CloseHandle(mi->event);
+    free(mi);
+    /* Sabotage attempts to re-use the mutex before initialising it again. */
+    *m = (pthread_mutex_t)NULL;
+  }
+
+  return 0;
+}
+
+int pthread_mutexattr_init(pthread_mutexattr_t *a)
+{
+  *a = PTHREAD_MUTEX_NORMAL | (PTHREAD_PROCESS_PRIVATE << 3);
+  return 0;
+}
+
+int pthread_mutexattr_destroy(pthread_mutexattr_t *a)
+{
+  if (!a)
+    return EINVAL;
+
+  return 0;
+}
+
+int pthread_mutexattr_gettype(const pthread_mutexattr_t *a, int *type)
+{
+  if (!a || !type)
+    return EINVAL;
+
+  *type = *a & 3;
+
+  return 0;
+}
+
+int pthread_mutexattr_settype(pthread_mutexattr_t *a, int type)
+{
+    if (!a || (type != PTHREAD_MUTEX_NORMAL && type != PTHREAD_MUTEX_RECURSIVE && type != PTHREAD_MUTEX_ERRORCHECK))
+      return EINVAL;
+    *a &= ~3;
+    *a |= type;
+
+    return 0;
+}
+
+int pthread_mutexattr_getpshared(const pthread_mutexattr_t *a, int *type)
+{
+    if (!a || !type)
+      return EINVAL;
+    *type = (*a & 4 ? PTHREAD_PROCESS_SHARED : PTHREAD_PROCESS_PRIVATE);
+
+    return 0;
+}
+
+int pthread_mutexattr_setpshared(pthread_mutexattr_t * a, int type)
+{
+    int r = 0;
+    if (!a || (type != PTHREAD_PROCESS_SHARED
+	&& type != PTHREAD_PROCESS_PRIVATE))
+      return EINVAL;
+    if (type == PTHREAD_PROCESS_SHARED)
+    {
+      type = PTHREAD_PROCESS_PRIVATE;
+      r = ENOSYS;
+    }
+    type = (type == PTHREAD_PROCESS_SHARED ? 4 : 0);
+
+    *a &= ~4;
+    *a |= type;
+
+    return r;
+}
+
+int pthread_mutexattr_getprotocol(const pthread_mutexattr_t *a, int *type)
+{
+    *type = *a & (8 + 16);
+
+    return 0;
+}
+
+int pthread_mutexattr_setprotocol(pthread_mutexattr_t *a, int type)
+{
+    if ((type & (8 + 16)) != 8 + 16) return EINVAL;
+
+    *a &= ~(8 + 16);
+    *a |= type;
+
+    return 0;
+}
+
+int pthread_mutexattr_getprioceiling(const pthread_mutexattr_t *a, int * prio)
+{
+    *prio = *a / PTHREAD_PRIO_MULT;
+    return 0;
+}
+
+int pthread_mutexattr_setprioceiling(pthread_mutexattr_t *a, int prio)
+{
+    *a &= (PTHREAD_PRIO_MULT - 1);
+    *a += prio * PTHREAD_PRIO_MULT;
+
+    return 0;
+}
diff --git a/vendor/winpthreads/src/nanosleep.c b/vendor/winpthreads/src/nanosleep.c
new file mode 100644
index 00000000..a81bf2c8
--- /dev/null
+++ b/vendor/winpthreads/src/nanosleep.c
@@ -0,0 +1,109 @@
+/**
+ * This file has no copyright assigned and is placed in the Public Domain.
+ * This file is part of the w64 mingw-runtime package.
+ * No warranty is given; refer to the file DISCLAIMER.PD within this package.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <assert.h>
+#include <errno.h>
+#include <time.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define WINPTHREAD_NANOSLEEP_DECL WINPTHREAD_API
+
+/* public header files */
+#include "pthread.h"
+#include "pthread_time.h"
+/* internal header files */
+#include "thread.h"
+
+#define POW10_3                 1000
+#define POW10_4                 10000
+#define POW10_6                 1000000
+#define POW10_9                 1000000000
+#define MAX_SLEEP_IN_MS         4294967294UL
+
+/**
+ * Sleep for the specified time.
+ * @param  request The desired amount of time to sleep.
+ * @param  remain The remain amount of time to sleep.
+ * @return If the function succeeds, the return value is 0.
+ *         If the function fails, the return value is -1,
+ *         with errno set to indicate the error.
+ */
+static int __nanosleep(const struct _timespec64 *request, struct _timespec64 *remain)
+{
+    unsigned long ms, rc = 0;
+    unsigned __int64 u64, want, real;
+
+    union {
+        unsigned __int64 ns100;
+        FILETIME ft;
+    }  _start, _end;
+
+    if (request->tv_sec < 0 || request->tv_nsec < 0 || request->tv_nsec >= POW10_9) {
+        errno = EINVAL;
+        return -1;
+    }
+
+    if (remain != NULL) GetSystemTimeAsFileTime(&_start.ft);
+
+    want = u64 = request->tv_sec * POW10_3 + request->tv_nsec / POW10_6;
+    while (u64 > 0 && rc == 0) {
+        if (u64 >= MAX_SLEEP_IN_MS) ms = MAX_SLEEP_IN_MS;
+        else ms = (unsigned long) u64;
+
+        u64 -= ms;
+        rc = _pthread_delay_np_ms(ms);
+    }
+
+    if (rc != 0) { /* WAIT_IO_COMPLETION (192) */
+        if (remain != NULL) {
+            GetSystemTimeAsFileTime(&_end.ft);
+            real = (_end.ns100 - _start.ns100) / POW10_4;
+
+            if (real >= want) u64 = 0;
+            else u64 = want - real;
+
+            remain->tv_sec = u64 / POW10_3;
+            remain->tv_nsec = (long) (u64 % POW10_3) * POW10_6;
+        }
+
+        errno = EINTR;
+        return -1;
+    }
+
+    return 0;
+}
+
+int nanosleep64(const struct _timespec64 *request, struct _timespec64 *remain)
+{
+    return __nanosleep (request, remain);
+}
+
+int nanosleep32(const struct _timespec32 *request, struct _timespec32 *remain)
+{
+    struct _timespec64 request64 = {
+        .tv_sec = request->tv_sec,
+        .tv_nsec = request->tv_nsec
+    };
+    struct _timespec64 remain64 = {0};
+
+    if (__nanosleep (&request64, &remain64) == -1)
+        return -1;
+
+    assert (remain64.tv_sec <= INT_MAX);
+
+    if (remain != NULL) {
+        remain->tv_sec = (__time32_t)remain64.tv_sec;
+        remain->tv_nsec = remain64.tv_nsec;
+    }
+
+    return 0;
+}
diff --git a/vendor/winpthreads/src/rwlock.c b/vendor/winpthreads/src/rwlock.c
new file mode 100644
index 00000000..68ef2a73
--- /dev/null
+++ b/vendor/winpthreads/src/rwlock.c
@@ -0,0 +1,546 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <assert.h>
+#include <malloc.h>
+#include <stdio.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define WINPTHREAD_RWLOCK_DECL WINPTHREAD_API
+
+/* public header files */
+#include "pthread.h"
+/* internal header files */
+#include "misc.h"
+#include "rwlock.h"
+#include "thread.h"
+
+static pthread_spinlock_t rwl_global = PTHREAD_SPINLOCK_INITIALIZER;
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int rwlock_static_init(pthread_rwlock_t *rw);
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int rwl_unref(volatile pthread_rwlock_t *rwl, int res)
+{
+    pthread_spin_lock(&rwl_global);
+    assert((((rwlock_t *)*rwl)->valid == LIFE_RWLOCK) && (((rwlock_t *)*rwl)->busy > 0));
+     ((rwlock_t *)*rwl)->busy--;
+    pthread_spin_unlock(&rwl_global);
+    return res;
+}
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int rwl_ref(pthread_rwlock_t *rwl, int f )
+{
+    int r = 0;
+    if (STATIC_RWL_INITIALIZER(*rwl)) {
+        r = rwlock_static_init(rwl);
+        if (r != 0 && r != EBUSY)
+            return r;
+    }
+    pthread_spin_lock(&rwl_global);
+
+    if (!rwl || !*rwl || ((rwlock_t *)*rwl)->valid != LIFE_RWLOCK) r = EINVAL;
+    else {
+        ((rwlock_t *)*rwl)->busy ++;
+    }
+
+    pthread_spin_unlock(&rwl_global);
+
+    return r;
+}
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int rwl_ref_unlock(pthread_rwlock_t *rwl )
+{
+    int r = 0;
+
+    pthread_spin_lock(&rwl_global);
+
+    if (!rwl || !*rwl || ((rwlock_t *)*rwl)->valid != LIFE_RWLOCK) r = EINVAL;
+    else if (STATIC_RWL_INITIALIZER(*rwl)) r= EPERM;
+    else {
+        ((rwlock_t *)*rwl)->busy ++;
+    }
+
+    pthread_spin_unlock(&rwl_global);
+
+    return r;
+}
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int rwl_ref_destroy(pthread_rwlock_t *rwl, pthread_rwlock_t *rDestroy )
+{
+    int r = 0;
+
+    *rDestroy = (pthread_rwlock_t)NULL;
+    pthread_spin_lock(&rwl_global);
+
+    if (!rwl || !*rwl) r = EINVAL;
+    else {
+        rwlock_t *r_ = (rwlock_t *)*rwl;
+        if (STATIC_RWL_INITIALIZER(*rwl)) *rwl = (pthread_rwlock_t)NULL;
+        else if (r_->valid != LIFE_RWLOCK) r = EINVAL;
+        else if (r_->busy) r = EBUSY;
+        else {
+            *rDestroy = *rwl;
+            *rwl = (pthread_rwlock_t)NULL;
+        }
+    }
+
+    pthread_spin_unlock(&rwl_global);
+    return r;
+}
+
+static int rwlock_gain_both_locks(rwlock_t *rwlock)
+{
+  int ret;
+  ret = pthread_mutex_lock(&rwlock->mex);
+  if (ret != 0)
+    return ret;
+  ret = pthread_mutex_lock(&rwlock->mcomplete);
+  if (ret != 0)
+    pthread_mutex_unlock(&rwlock->mex);
+  return ret;
+}
+
+static int rwlock_free_both_locks(rwlock_t *rwlock, int last_fail)
+{
+  int ret, ret2;
+  ret = pthread_mutex_unlock(&rwlock->mcomplete);
+  ret2 = pthread_mutex_unlock(&rwlock->mex);
+  if (last_fail && ret2 != 0)
+    ret = ret2;
+  else if (!last_fail && !ret)
+    ret = ret2;
+  return ret;
+}
+
+static pthread_spinlock_t cond_locked = PTHREAD_SPINLOCK_INITIALIZER;
+
+static WINPTHREADS_ATTRIBUTE((noinline)) int rwlock_static_init(pthread_rwlock_t *rw)
+{
+  int r;
+  pthread_spin_lock(&cond_locked);
+  if (*rw != PTHREAD_RWLOCK_INITIALIZER)
+  {
+    pthread_spin_unlock(&cond_locked);
+    return EINVAL;
+  }
+  r = pthread_rwlock_init (rw, NULL);
+  pthread_spin_unlock(&cond_locked);
+
+  return r;
+}
+
+int pthread_rwlock_init (pthread_rwlock_t *rwlock_, const pthread_rwlockattr_t *attr)
+{
+    rwlock_t *rwlock;
+    int r;
+
+    if(!rwlock_)
+      return EINVAL;
+    *rwlock_ = (pthread_rwlock_t)NULL;
+    if ((rwlock = calloc(1, sizeof(*rwlock))) == NULL)
+      return ENOMEM;
+    rwlock->valid = DEAD_RWLOCK;
+
+    rwlock->nex_count = rwlock->nsh_count = rwlock->ncomplete = 0;
+    if ((r = pthread_mutex_init (&rwlock->mex, NULL)) != 0)
+    {
+        free(rwlock);
+        return r;
+    }
+    if ((r = pthread_mutex_init (&rwlock->mcomplete, NULL)) != 0)
+    {
+      pthread_mutex_destroy(&rwlock->mex);
+      free(rwlock);
+      return r;
+    }
+    if ((r = pthread_cond_init (&rwlock->ccomplete, NULL)) != 0)
+    {
+      pthread_mutex_destroy(&rwlock->mex);
+      pthread_mutex_destroy (&rwlock->mcomplete);
+      free(rwlock);
+      return r;
+    }
+    rwlock->valid = LIFE_RWLOCK;
+    *rwlock_ = (pthread_rwlock_t)rwlock;
+    return r;
+}
+
+int pthread_rwlock_destroy (pthread_rwlock_t *rwlock_)
+{
+    rwlock_t *rwlock;
+    pthread_rwlock_t rDestroy;
+    int r, r2;
+
+    pthread_spin_lock(&cond_locked);
+    r = rwl_ref_destroy(rwlock_,&rDestroy);
+    pthread_spin_unlock(&cond_locked);
+
+    if(r) return r;
+    if(!rDestroy) return 0; /* destroyed a (still) static initialized rwl */
+
+    rwlock = (rwlock_t *)rDestroy;
+    r = rwlock_gain_both_locks (rwlock);
+    if (r != 0)
+    {
+      *rwlock_ = rDestroy;
+      return r;
+    }
+    if (rwlock->nsh_count > rwlock->ncomplete || rwlock->nex_count > 0)
+    {
+      *rwlock_ = rDestroy;
+      r = rwlock_free_both_locks(rwlock, 1);
+      if (!r)
+        r = EBUSY;
+      return r;
+    }
+    rwlock->valid  = DEAD_RWLOCK;
+    r = rwlock_free_both_locks(rwlock, 0);
+    if (r != 0) { *rwlock_ = rDestroy; return r; }
+
+    r = pthread_cond_destroy(&rwlock->ccomplete);
+    r2 = pthread_mutex_destroy(&rwlock->mex);
+    if (!r) r = r2;
+    r2 = pthread_mutex_destroy(&rwlock->mcomplete);
+    if (!r) r = r2;
+    rwlock->valid  = DEAD_RWLOCK;
+    free((void *)rDestroy);
+    return 0;
+}
+
+int pthread_rwlock_rdlock (pthread_rwlock_t *rwlock_)
+{
+  rwlock_t *rwlock;
+  int ret;
+
+  /* pthread_testcancel(); */
+
+  ret = rwl_ref(rwlock_,0);
+  if(ret != 0) return ret;
+
+  rwlock = (rwlock_t *)*rwlock_;
+
+  ret = pthread_mutex_lock(&rwlock->mex);
+  if (ret != 0) return rwl_unref(rwlock_, ret);
+  InterlockedIncrement((long*)&rwlock->nsh_count);
+  if (rwlock->nsh_count == INT_MAX)
+  {
+    ret = pthread_mutex_lock(&rwlock->mcomplete);
+    if (ret != 0)
+    {
+      pthread_mutex_unlock(&rwlock->mex);
+      return rwl_unref(rwlock_,ret);
+    }
+    rwlock->nsh_count -= rwlock->ncomplete;
+    rwlock->ncomplete = 0;
+    ret = rwlock_free_both_locks(rwlock, 0);
+    return rwl_unref(rwlock_, ret);
+  }
+  ret = pthread_mutex_unlock(&rwlock->mex);
+  return rwl_unref(rwlock_, ret);
+}
+
+/* Internal version which always uses `struct _timespec64`. */
+static int __pthread_rwlock_timedrdlock (pthread_rwlock_t *rwlock_, const struct _timespec64 *ts)
+{
+  rwlock_t *rwlock;
+  int ret;
+
+  /* pthread_testcancel(); */
+
+  ret = rwl_ref(rwlock_,0);
+  if(ret != 0) return ret;
+
+  rwlock = (rwlock_t *)*rwlock_;
+  if ((ret = pthread_mutex_timedlock64 (&rwlock->mex, ts)) != 0)
+      return rwl_unref(rwlock_, ret);
+  InterlockedIncrement(&rwlock->nsh_count);
+  if (rwlock->nsh_count == INT_MAX)
+  {
+    ret = pthread_mutex_timedlock64(&rwlock->mcomplete, ts);
+    if (ret != 0)
+    {
+      if (ret == ETIMEDOUT)
+	InterlockedIncrement(&rwlock->ncomplete);
+      pthread_mutex_unlock(&rwlock->mex);
+      return rwl_unref(rwlock_, ret);
+    }
+    rwlock->nsh_count -= rwlock->ncomplete;
+    rwlock->ncomplete = 0;
+    ret = rwlock_free_both_locks(rwlock, 0);
+    return rwl_unref(rwlock_, ret);
+  }
+  ret = pthread_mutex_unlock(&rwlock->mex);
+  return rwl_unref(rwlock_, ret);
+}
+
+int pthread_rwlock_timedrdlock64(pthread_rwlock_t *l, const struct _timespec64 *ts)
+{
+  return __pthread_rwlock_timedrdlock (l, ts);
+}
+
+int pthread_rwlock_timedrdlock32(pthread_rwlock_t *l, const struct _timespec32 *ts)
+{
+  struct _timespec64 ts64 = {.tv_sec = ts->tv_sec, .tv_nsec = ts->tv_nsec};
+  return __pthread_rwlock_timedrdlock (l, &ts64);
+}
+
+int pthread_rwlock_tryrdlock (pthread_rwlock_t *rwlock_)
+{
+  rwlock_t *rwlock;
+  int ret;
+
+  ret = rwl_ref(rwlock_,RWL_TRY);
+  if(ret != 0) return ret;
+
+  rwlock = (rwlock_t *)*rwlock_;
+  ret = pthread_mutex_trylock(&rwlock->mex);
+  if (ret != 0)
+      return rwl_unref(rwlock_, ret);
+  InterlockedIncrement(&rwlock->nsh_count);
+  if (rwlock->nsh_count == INT_MAX)
+  {
+    ret = pthread_mutex_lock(&rwlock->mcomplete);
+    if (ret != 0)
+    {
+      pthread_mutex_unlock(&rwlock->mex);
+      return rwl_unref(rwlock_, ret);
+    }
+    rwlock->nsh_count -= rwlock->ncomplete;
+    rwlock->ncomplete = 0;
+    ret = rwlock_free_both_locks(rwlock, 0);
+    return rwl_unref(rwlock_, ret);
+  }
+  ret = pthread_mutex_unlock(&rwlock->mex);
+  return rwl_unref(rwlock_,ret);
+}
+
+int pthread_rwlock_trywrlock (pthread_rwlock_t *rwlock_)
+{
+  rwlock_t *rwlock;
+  int ret;
+
+  ret = rwl_ref(rwlock_,RWL_TRY);
+  if(ret != 0) return ret;
+
+  rwlock = (rwlock_t *)*rwlock_;
+  ret = pthread_mutex_trylock (&rwlock->mex);
+  if (ret != 0)
+    return rwl_unref(rwlock_, ret);
+  ret = pthread_mutex_trylock(&rwlock->mcomplete);
+  if (ret != 0)
+  {
+    int r1 = pthread_mutex_unlock(&rwlock->mex);
+    if (r1 != 0)
+      ret = r1;
+    return rwl_unref(rwlock_, ret);
+  }
+  if (rwlock->nex_count != 0)
+    return rwl_unref(rwlock_, EBUSY);
+  if (rwlock->ncomplete > 0)
+  {
+    rwlock->nsh_count -= rwlock->ncomplete;
+    rwlock->ncomplete = 0;
+  }
+  if (rwlock->nsh_count > 0)
+  {
+    ret = rwlock_free_both_locks(rwlock, 0);
+    if (!ret)
+      ret = EBUSY;
+    return rwl_unref(rwlock_, ret);
+  }
+  rwlock->nex_count = 1;
+  return rwl_unref(rwlock_, 0);
+}
+
+int pthread_rwlock_unlock (pthread_rwlock_t *rwlock_)
+{
+  rwlock_t *rwlock;
+  int ret;
+
+  ret = rwl_ref_unlock(rwlock_);
+  if(ret != 0) return ret;
+
+  rwlock = (rwlock_t *)*rwlock_;
+  if (rwlock->nex_count == 0)
+  {
+    ret = pthread_mutex_lock(&rwlock->mcomplete);
+    if (!ret)
+    {
+      int r1;
+      InterlockedIncrement(&rwlock->ncomplete);
+      if (rwlock->ncomplete == 0)
+	ret = pthread_cond_signal(&rwlock->ccomplete);
+      r1 = pthread_mutex_unlock(&rwlock->mcomplete);
+      if (!ret)
+	ret = r1;
+    }
+  }
+  else
+  {
+    InterlockedDecrement(&rwlock->nex_count);
+    ret = rwlock_free_both_locks(rwlock, 0);
+  }
+  return rwl_unref(rwlock_, ret);
+}
+
+static void st_cancelwrite (void *arg)
+{
+    rwlock_t *rwlock = (rwlock_t *)arg;
+
+    rwlock->nsh_count = - rwlock->ncomplete;
+    rwlock->ncomplete = 0;
+    rwlock_free_both_locks(rwlock, 0);
+}
+
+int pthread_rwlock_wrlock (pthread_rwlock_t *rwlock_)
+{
+  rwlock_t *rwlock;
+  int ret;
+
+  /* pthread_testcancel(); */
+  ret = rwl_ref(rwlock_,0);
+  if(ret != 0) return ret;
+
+  rwlock = (rwlock_t *)*rwlock_;
+  ret = rwlock_gain_both_locks(rwlock);
+  if (ret != 0)
+    return rwl_unref(rwlock_,ret);
+
+  if (rwlock->nex_count == 0)
+  {
+    if (rwlock->ncomplete > 0)
+    {
+      rwlock->nsh_count -= rwlock->ncomplete;
+      rwlock->ncomplete = 0;
+    }
+    if (rwlock->nsh_count > 0)
+    {
+      rwlock->ncomplete = -rwlock->nsh_count;
+      pthread_cleanup_push(st_cancelwrite, (void *) rwlock);
+      do {
+	ret = pthread_cond_wait(&rwlock->ccomplete, &rwlock->mcomplete);
+      } while (!ret && rwlock->ncomplete < 0);
+
+      pthread_cleanup_pop(!ret ? 0 : 1);
+      if (!ret)
+	rwlock->nsh_count = 0;
+    }
+  }
+  if(!ret)
+    InterlockedIncrement((long*)&rwlock->nex_count);
+  return rwl_unref(rwlock_,ret);
+}
+
+/* Internal version which always uses `struct _timespec64`. */
+static int __pthread_rwlock_timedwrlock (pthread_rwlock_t *rwlock_, const struct _timespec64 *ts)
+{
+  int ret;
+  rwlock_t *rwlock;
+
+  /* pthread_testcancel(); */
+  if (!rwlock_ || !ts)
+    return EINVAL;
+  if ((ret = rwl_ref(rwlock_,0)) != 0)
+    return ret;
+  rwlock = (rwlock_t *)*rwlock_;
+
+  ret = pthread_mutex_timedlock64(&rwlock->mex, ts);
+  if (ret != 0)
+    return rwl_unref(rwlock_,ret);
+  ret = pthread_mutex_timedlock64(&rwlock->mcomplete, ts);
+  if (ret != 0)
+  {
+    pthread_mutex_unlock(&rwlock->mex);
+    return rwl_unref(rwlock_,ret);
+  }
+  if (rwlock->nex_count == 0)
+  {
+    if (rwlock->ncomplete > 0)
+    {
+      rwlock->nsh_count -= rwlock->ncomplete;
+      rwlock->ncomplete = 0;
+    }
+    if (rwlock->nsh_count > 0)
+    {
+      rwlock->ncomplete = -rwlock->nsh_count;
+      pthread_cleanup_push(st_cancelwrite, (void *) rwlock);
+      do {
+	ret = pthread_cond_timedwait64(&rwlock->ccomplete, &rwlock->mcomplete, ts);
+      } while (rwlock->ncomplete < 0 && !ret);
+      pthread_cleanup_pop(!ret ? 0 : 1);
+
+      if (!ret)
+	rwlock->nsh_count = 0;
+    }
+  }
+  if(!ret)
+    InterlockedIncrement((long*)&rwlock->nex_count);
+  return rwl_unref(rwlock_,ret);
+}
+
+int pthread_rwlock_timedwrlock64(pthread_rwlock_t *rwlock, const struct _timespec64 *ts)
+{
+  return __pthread_rwlock_timedwrlock (rwlock, ts);
+}
+
+int pthread_rwlock_timedwrlock32(pthread_rwlock_t *rwlock, const struct _timespec32 *ts)
+{
+  struct _timespec64 ts64 = {.tv_sec = ts->tv_sec, .tv_nsec = ts->tv_nsec};
+  return __pthread_rwlock_timedwrlock (rwlock, &ts64);
+}
+
+int pthread_rwlockattr_destroy(pthread_rwlockattr_t *a)
+{
+  if (!a)
+    return EINVAL;
+  return 0;
+}
+
+int pthread_rwlockattr_init(pthread_rwlockattr_t *a)
+{
+  if (!a)
+    return EINVAL;
+  *a = PTHREAD_PROCESS_PRIVATE;
+  return 0;
+}
+
+int pthread_rwlockattr_getpshared(pthread_rwlockattr_t *a, int *s)
+{
+  if (!a || !s)
+    return EINVAL;
+  *s = *a;
+  return 0;
+}
+
+int pthread_rwlockattr_setpshared(pthread_rwlockattr_t *a, int s)
+{
+  if (!a || (s != PTHREAD_PROCESS_SHARED && s != PTHREAD_PROCESS_PRIVATE))
+    return EINVAL;
+  *a = s;
+  return 0;
+}
diff --git a/vendor/winpthreads/src/rwlock.h b/vendor/winpthreads/src/rwlock.h
new file mode 100644
index 00000000..8faf12b9
--- /dev/null
+++ b/vendor/winpthreads/src/rwlock.h
@@ -0,0 +1,46 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREADS_RWLOCK_H
+#define WIN_PTHREADS_RWLOCK_H
+
+#define LIFE_RWLOCK 0xBAB1F0ED
+#define DEAD_RWLOCK 0xDEADB0EF
+
+#define STATIC_RWL_INITIALIZER(x)		((pthread_rwlock_t)(x) == ((pthread_rwlock_t)PTHREAD_RWLOCK_INITIALIZER))
+
+typedef struct rwlock_t rwlock_t;
+struct rwlock_t {
+    unsigned int valid;
+    int busy;
+    LONG nex_count; /* Exclusive access counter.  */
+    LONG nsh_count; /* Shared access counter. */
+    LONG ncomplete; /* Shared completed counter. */
+    pthread_mutex_t mex; /* Exclusive access protection.  */
+    pthread_mutex_t mcomplete; /* Shared completed protection. */
+    pthread_cond_t ccomplete; /* Shared access completed queue.  */
+};
+
+#define RWL_SET	0x01
+#define RWL_TRY	0x02
+
+#endif
diff --git a/vendor/winpthreads/src/sched.c b/vendor/winpthreads/src/sched.c
new file mode 100644
index 00000000..0bdcc3ca
--- /dev/null
+++ b/vendor/winpthreads/src/sched.c
@@ -0,0 +1,226 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <stdio.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+/* public header files */
+#include "pthread.h"
+/* internal header files */
+#include "misc.h"
+#include "thread.h"
+
+int sched_get_priority_min(int pol)
+{
+  if (pol < SCHED_MIN || pol > SCHED_MAX) {
+      errno = EINVAL;
+      return -1;
+  }
+
+  return THREAD_PRIORITY_IDLE;
+}
+
+int sched_get_priority_max(int pol)
+{
+  if (pol < SCHED_MIN || pol > SCHED_MAX) {
+      errno = EINVAL;
+      return -1;
+  }
+
+  return THREAD_PRIORITY_TIME_CRITICAL;
+}
+
+int pthread_attr_setschedparam(pthread_attr_t *attr, const struct sched_param *p)
+{
+    int r = 0;
+
+    if (attr == NULL || p == NULL) {
+        return EINVAL;
+    }
+    memcpy(&attr->param, p, sizeof (*p));
+    return r;
+}
+
+int pthread_attr_getschedparam(const pthread_attr_t *attr, struct sched_param *p)
+{
+    int r = 0;
+
+    if (attr == NULL || p == NULL) {
+        return EINVAL;
+    }
+    memcpy(p, &attr->param, sizeof (*p));
+    return r;
+}
+
+int pthread_attr_setschedpolicy (pthread_attr_t *attr, int pol)
+{
+  if (!attr || pol < SCHED_MIN || pol > SCHED_MAX)
+    return EINVAL;
+  if (pol != SCHED_OTHER)
+    return ENOTSUP;
+  return 0;
+}
+
+int pthread_attr_getschedpolicy (const pthread_attr_t *attr, int *pol)
+{
+  if (!attr || !pol)
+    return EINVAL;
+  *pol = SCHED_OTHER;
+  return 0;
+}
+
+static int pthread_check(pthread_t t)
+{
+  struct _pthread_v *pv;
+
+  if (!t)
+    return ESRCH;
+  pv = __pth_gpointer_locked (t);
+  if (pv->ended == 0)
+    return 0;
+  CHECK_OBJECT(pv, ESRCH);
+  return 0;
+}
+
+int pthread_getschedparam(pthread_t t, int *pol, struct sched_param *p)
+{
+    int r;
+    //if (!t)
+    //  t = pthread_self();
+
+    if ((r = pthread_check(t)) != 0)
+    {
+        return r;
+    }
+
+    if (!p || !pol)
+    {
+        return EINVAL;
+    }
+    *pol = __pth_gpointer_locked (t)->sched_pol;
+    p->sched_priority = __pth_gpointer_locked (t)->sched.sched_priority;
+
+    return 0;
+}
+
+int pthread_setschedparam(pthread_t t, int pol,  const struct sched_param *p)
+{
+  struct _pthread_v *pv;
+    int r, pr = 0;
+    //if (!t.p) t = pthread_self();
+
+    if ((r = pthread_check(t)) != 0)
+        return r;
+
+    if (pol < SCHED_MIN || pol > SCHED_MAX || p == NULL)
+        return EINVAL;
+    if (pol != SCHED_OTHER)
+        return ENOTSUP;
+    pr = p->sched_priority;
+    if (pr < sched_get_priority_min(pol) || pr > sched_get_priority_max(pol))
+      return EINVAL;
+
+    /* See msdn: there are actually 7 priorities:
+    THREAD_PRIORITY_IDLE    -      -15
+    THREAD_PRIORITY_LOWEST          -2
+    THREAD_PRIORITY_BELOW_NORMAL    -1
+    THREAD_PRIORITY_NORMAL           0
+    THREAD_PRIORITY_ABOVE_NORMAL     1
+    THREAD_PRIORITY_HIGHEST          2
+    THREAD_PRIORITY_TIME_CRITICAL   15
+    */
+    if (pr <= THREAD_PRIORITY_IDLE) {
+        pr = THREAD_PRIORITY_IDLE;
+    } else if (pr <= THREAD_PRIORITY_LOWEST) {
+        pr = THREAD_PRIORITY_LOWEST;
+    } else if (pr >= THREAD_PRIORITY_TIME_CRITICAL) {
+        pr = THREAD_PRIORITY_TIME_CRITICAL;
+    } else if (pr >= THREAD_PRIORITY_HIGHEST) {
+        pr = THREAD_PRIORITY_HIGHEST;
+    }
+    pv = __pth_gpointer_locked (t);
+    if (SetThreadPriority(pv->h, pr)) {
+        pv->sched_pol = pol;
+        pv->sched.sched_priority = p->sched_priority;
+    } else
+        r = EINVAL;
+    return r;
+}
+
+int sched_getscheduler(pid_t pid)
+{
+  if (pid != 0)
+  {
+      HANDLE h = NULL;
+      int selfPid = (int) GetCurrentProcessId ();
+
+      if (pid != (pid_t) selfPid && (h = OpenProcess (PROCESS_QUERY_INFORMATION, 0, (DWORD) pid)) == NULL)
+      {
+	  errno = (GetLastError () == (0xFF & ERROR_ACCESS_DENIED)) ? EPERM : ESRCH;
+	  return -1;
+      }
+      if (h)
+	  CloseHandle (h);
+  }
+  return SCHED_OTHER;
+}
+
+int sched_setscheduler(pid_t pid, int pol, const struct sched_param *param)
+{
+  if (!param)
+    {
+      errno = EINVAL;
+      return -1;
+    }
+  if (pid != 0)
+  {
+      HANDLE h = NULL;
+      int selfPid = (int) GetCurrentProcessId ();
+
+      if (pid != (pid_t) selfPid && (h = OpenProcess (PROCESS_SET_INFORMATION, 0, (DWORD) pid)) == NULL)
+      {
+	  errno = (GetLastError () == (0xFF & ERROR_ACCESS_DENIED)) ? EPERM : ESRCH;
+	  return -1;
+      }
+      if (h)
+          CloseHandle (h);
+  }
+
+  if (pol != SCHED_OTHER)
+  {
+      errno = ENOSYS;
+      return -1;
+  }
+  return SCHED_OTHER;
+}
+
+int sched_yield(void)
+{
+  Sleep(0);
+  return 0;
+}
diff --git a/vendor/winpthreads/src/sem.c b/vendor/winpthreads/src/sem.c
new file mode 100644
index 00000000..68b33b8b
--- /dev/null
+++ b/vendor/winpthreads/src/sem.c
@@ -0,0 +1,381 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <malloc.h>
+#include <stdio.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+/* public header files */
+#include "pthread.h"
+#include "semaphore.h"
+/* internal header files */
+#include "misc.h"
+#include "sem.h"
+#include "thread.h"
+
+int do_sema_b_wait_intern (HANDLE sema, int nointerrupt, DWORD timeout);
+
+static int
+sem_result (int res)
+{
+  if (res != 0) {
+    errno = res;
+    return -1;
+  }
+  return 0;
+}
+
+int
+sem_init (sem_t *sem, int pshared, unsigned int value)
+{
+  _sem_t *sv;
+
+  if (!sem || value > (unsigned int)SEM_VALUE_MAX)
+    return sem_result (EINVAL);
+  if (pshared != PTHREAD_PROCESS_PRIVATE)
+    return sem_result (EPERM);
+
+  if ((sv = (sem_t) calloc (1,sizeof (*sv))) == NULL)
+    return sem_result (ENOMEM);
+
+  sv->value = value;
+  if (pthread_mutex_init (&sv->vlock, NULL) != 0)
+    {
+      free (sv);
+      return sem_result (ENOSPC);
+    }
+  if ((sv->s = CreateSemaphore (NULL, 0, SEM_VALUE_MAX, NULL)) == NULL)
+    {
+      pthread_mutex_destroy (&sv->vlock);
+      free (sv);
+      return sem_result (ENOSPC);
+    }
+
+  sv->valid = LIFE_SEM;
+  *sem = sv;
+  return 0;
+}
+
+int
+sem_destroy (sem_t *sem)
+{
+  int r;
+  _sem_t *sv = NULL;
+
+  if (!sem || (sv = *sem) == NULL)
+    return sem_result (EINVAL);
+  if ((r = pthread_mutex_lock (&sv->vlock)) != 0)
+    return sem_result (r);
+
+#if 0
+  /* We don't wait for destroying a semaphore ...
+     or?  */
+  if (sv->value < 0)
+    {
+      pthread_mutex_unlock (&sv->vlock);
+      return sem_result (EBUSY);
+    }
+#endif
+
+  if (!CloseHandle (sv->s))
+    {
+      pthread_mutex_unlock (&sv->vlock);
+      return sem_result (EINVAL);
+    }
+  *sem = NULL;
+  sv->value = SEM_VALUE_MAX;
+  pthread_mutex_unlock(&sv->vlock);
+  Sleep (0);
+  while (pthread_mutex_destroy (&sv->vlock) == EBUSY)
+    Sleep (0);
+  sv->valid = DEAD_SEM;
+  free (sv);
+  return 0;
+}
+
+static int
+sem_std_enter (sem_t *sem,_sem_t **svp, int do_test)
+{
+  int r;
+  _sem_t *sv;
+
+  if (do_test)
+    pthread_testcancel ();
+  if (!sem)
+    return sem_result (EINVAL);
+  sv = *sem;
+  if (sv == NULL)
+    return sem_result (EINVAL);
+
+  if ((r = pthread_mutex_lock (&sv->vlock)) != 0)
+    return sem_result (r);
+
+  if (*sem == NULL)
+    {
+      pthread_mutex_unlock(&sv->vlock);
+      return sem_result (EINVAL);
+    }
+  *svp = sv;
+  return 0;
+}
+
+int
+sem_trywait (sem_t *sem)
+{
+  _sem_t *sv;
+
+  if (sem_std_enter (sem, &sv, 0) != 0)
+    return -1;
+  if (sv->value <= 0)
+    {
+      pthread_mutex_unlock (&sv->vlock);
+      return sem_result (EAGAIN);
+    }
+  sv->value--;
+  pthread_mutex_unlock (&sv->vlock);
+
+  return 0;
+}
+
+struct sSemTimedWait
+{
+  sem_t *p;
+  int *ret;
+};
+
+static void
+clean_wait_sem (void *s)
+{
+  struct sSemTimedWait *p = (struct sSemTimedWait *) s;
+  _sem_t *sv = NULL;
+
+  if (sem_std_enter (p->p, &sv, 0) != 0)
+    return;
+
+  if (WaitForSingleObject (sv->s, 0) != WAIT_OBJECT_0)
+    InterlockedIncrement (&sv->value);
+  else if (p->ret)
+    p->ret[0] = 0;
+  pthread_mutex_unlock (&sv->vlock);
+}
+
+int
+sem_wait (sem_t *sem)
+{
+  long cur_v;
+  int ret = 0;
+  _sem_t *sv;
+  HANDLE semh;
+  struct sSemTimedWait arg;
+
+  if (sem_std_enter (sem, &sv, 1) != 0)
+    return -1;
+
+  arg.ret = &ret;
+  arg.p = sem;
+  InterlockedDecrement (&sv->value);
+  cur_v = sv->value;
+  semh = sv->s;
+  pthread_mutex_unlock (&sv->vlock);
+
+  if (cur_v >= 0)
+    return 0;
+  else
+    {
+      pthread_cleanup_push (clean_wait_sem, (void *) &arg);
+      ret = do_sema_b_wait_intern (semh, 2, INFINITE);
+      pthread_cleanup_pop (ret);
+      if (ret == EINVAL)
+        return 0;
+    }
+
+  if (!ret)
+    return 0;
+
+  return sem_result (ret);
+}
+
+/* Internal version which always uses `struct _timespec64`. */
+static int
+__sem_timedwait (sem_t *sem, const struct _timespec64 *t)
+{
+  int cur_v, ret = 0;
+  DWORD dwr;
+  HANDLE semh;
+  _sem_t *sv;
+  struct sSemTimedWait arg;
+
+  if (!t)
+    return sem_wait (sem);
+  dwr = dwMilliSecs(_pthread_rel_time_in_ms (t));
+
+  if (sem_std_enter (sem, &sv, 1) != 0)
+    return -1;
+
+  arg.ret = &ret;
+  arg.p = sem;
+  InterlockedDecrement (&sv->value);
+  cur_v = sv->value;
+  semh = sv->s;
+  pthread_mutex_unlock(&sv->vlock);
+
+  if (cur_v >= 0)
+    return 0;
+  else
+    {
+      pthread_cleanup_push (clean_wait_sem, (void *) &arg);
+      ret = do_sema_b_wait_intern (semh, 2, dwr);
+      pthread_cleanup_pop (ret);
+      if (ret == EINVAL)
+        return 0;
+    }
+
+  if (!ret)
+    return 0;
+  return sem_result (ret);
+}
+
+int sem_timedwait64(sem_t *sem, const struct _timespec64 *t)
+{
+  return __sem_timedwait (sem, t);
+}
+
+int sem_timedwait32(sem_t *sem, const struct _timespec32 *t)
+{
+  struct _timespec64 t64 = {0};
+
+  if (t != NULL)
+  {
+    t64.tv_sec = t->tv_sec;
+    t64.tv_nsec = t->tv_nsec;
+  }
+
+  return __sem_timedwait (sem, t == NULL ? NULL : &t64);
+}
+
+int
+sem_post (sem_t *sem)
+{
+  _sem_t *sv;
+
+  if (sem_std_enter (sem, &sv, 0) != 0)
+    return -1;
+
+  if (sv->value >= SEM_VALUE_MAX)
+    {
+      pthread_mutex_unlock (&sv->vlock);
+      return sem_result (ERANGE);
+    }
+  InterlockedIncrement (&sv->value);
+  if (sv->value > 0 || ReleaseSemaphore (sv->s, 1, NULL))
+    {
+      pthread_mutex_unlock (&sv->vlock);
+      return 0;
+    }
+  InterlockedDecrement (&sv->value);
+  pthread_mutex_unlock (&sv->vlock);
+
+  return sem_result (EINVAL);
+}
+
+int
+sem_post_multiple (sem_t *sem, int count)
+{
+  int waiters_count;
+  _sem_t *sv;
+
+  if (count <= 0)
+    return sem_result (EINVAL);
+  if (sem_std_enter (sem, &sv, 0) != 0)
+    return -1;
+
+  if (sv->value > (SEM_VALUE_MAX - count))
+  {
+    pthread_mutex_unlock (&sv->vlock);
+    return sem_result (ERANGE);
+  }
+  waiters_count = -sv->value;
+  sv->value += count;
+  /*InterlockedExchangeAdd((long*)&sv->value, (long) count);*/
+  if (waiters_count <= 0
+      || ReleaseSemaphore (sv->s,
+			   (waiters_count < count ? waiters_count
+			   			  : count), NULL))
+  {
+    pthread_mutex_unlock(&sv->vlock);
+    return 0;
+  }
+  /*InterlockedExchangeAdd((long*)&sv->value, -((long) count));*/
+  sv->value -= count;
+  pthread_mutex_unlock(&sv->vlock);
+  return sem_result (EINVAL);
+}
+
+sem_t *
+sem_open (const char *name, int oflag, mode_t mode, unsigned int value)
+{
+  sem_result (ENOSYS);
+  return NULL;
+}
+
+int
+sem_close (sem_t *sem)
+{
+  return sem_result (ENOSYS);
+}
+
+int
+sem_unlink (const char *name)
+{
+  return sem_result (ENOSYS);
+}
+
+int
+sem_getvalue (sem_t *sem, int *sval)
+{
+  _sem_t *sv;
+  int r;
+
+  if (!sval)
+    return sem_result (EINVAL);
+
+  if (!sem || (sv = *sem) == NULL)
+    return sem_result (EINVAL);
+
+  if ((r = pthread_mutex_lock (&sv->vlock)) != 0)
+    return sem_result (r);
+  if (*sem == NULL)
+    {
+      pthread_mutex_unlock (&sv->vlock);
+      return sem_result (EINVAL);
+    }
+
+  *sval = (int) sv->value;
+  pthread_mutex_unlock (&sv->vlock);
+  return 0;
+}
diff --git a/vendor/winpthreads/src/sem.h b/vendor/winpthreads/src/sem.h
new file mode 100644
index 00000000..ddfc00a6
--- /dev/null
+++ b/vendor/winpthreads/src/sem.h
@@ -0,0 +1,38 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_SEM
+#define WIN_SEM
+
+#define LIFE_SEM 0xBAB1F00D
+#define DEAD_SEM 0xDEADBEEF
+
+typedef struct _sem_t _sem_t;
+struct _sem_t
+{
+    unsigned int valid;
+    HANDLE s;
+    volatile long value;
+    pthread_mutex_t vlock;
+};
+
+#endif /* WIN_SEM */
diff --git a/vendor/winpthreads/src/spinlock.c b/vendor/winpthreads/src/spinlock.c
new file mode 100644
index 00000000..c6648bc8
--- /dev/null
+++ b/vendor/winpthreads/src/spinlock.c
@@ -0,0 +1,371 @@
+/*
+   Copyright (c) 2026 mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <stdlib.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+/* public header files */
+#include "pthread.h"
+/* internal header files */
+#include "misc.h"
+
+/**
+ * Reference:
+ *
+ * pthread_spin_init(), pthread_spin_destroy():
+ *  <https://pubs.opengroup.org/onlinepubs/9799919799.2024edition/functions/pthread_spin_destroy.html>
+ *
+ * pthread_spin_lock(), pthread_spin_trylock():
+ *  <https://pubs.opengroup.org/onlinepubs/9799919799.2024edition/functions/pthread_spin_lock.html>
+ *
+ * pthread_spin_unlock():
+ *  <https://pubs.opengroup.org/onlinepubs/9799919799.2024edition/functions/pthread_spin_unlock.html>
+ */
+
+/**
+ * Internal structure pointed to by `pthread_spinlock_t` objects.
+ */
+typedef struct {
+  /**
+   * This value is used to indicate that spin lock has no owner.
+   */
+#define THREAD_ID_NO_OWNER ((DWORD)-1)
+  /**
+   * ID of the thread which holds the lock.
+   */
+  DWORD ThreadId;
+  /**
+   * Auto-reset event.
+   *
+   * This event is created in signaled state, which means any thread can
+   * `WaitForSingleObject` on it; only one thread will be released at a time,
+   * after which this event will be automatically put into non-signaled state.
+   *
+   * The released thread owns the lock; it unlocks it by calling `SetEvent` on
+   * this event, which puts this event into signaled state, allowing system
+   * release another thread which waits on it.
+   *
+   * The cycle repeats until this event is destroyed.
+   */
+  HANDLE Event;
+} WinpthreadsSpinlock;
+
+#define WINPTHREADS_SPINLOCK_INITIALIZER ((WinpthreadsSpinlock *) PTHREAD_SPINLOCK_INITIALIZER)
+
+/**
+ * Obtain pointer to `WinpthreadsSpinlock` structure pointed to by `lock`.
+ *
+ * If `lock` points to statically initialzied `pthread_spinlock_t` object,
+ * allocate `WinpthreadsSpinlock` structure and store its address in `*lock`.
+ *
+ * On success, stores pointer to `WinpthreadsSpinlock` structure in `*spinlock`.
+ *
+ * Returns zero on success and an error-code on failure.
+ */
+static WINPTHREADS_INLINE int WinpthreadsSpinlockGet (pthread_spinlock_t *lock, WinpthreadsSpinlock **spinlock)
+{
+  WinpthreadsSpinlock **pSpinlock = (WinpthreadsSpinlock **) lock;
+
+  /**
+   * POSIX:
+   *
+   * If an implementation detects that the value specified by the lock argument
+   * to pthread_spin_*() does not refer to an initialized spin lock object,
+   * it is recommended that the function should fail and report an [EINVAL] error.
+   */
+  if (unlikely (pSpinlock == NULL)) {
+    return EINVAL;
+  }
+
+  if (*pSpinlock == WINPTHREADS_SPINLOCK_INITIALIZER) {
+    /**
+     * We need to avoid race condition when more than one thread calls
+     * `pthread_spin_[try]lock` on statically initialized `pthread_spinlock_t`
+     * at the same time.
+     *
+     * Store newly initialized spinlock in `spinlock`, which points to local
+     * variable supplied by the caller, and then store it in `*lock`.
+     *
+     * If some other thread was faster then us, destroy newly created spinlock
+     * and use spinlock pointed to by `lock`.
+     */
+    int error_code = pthread_spin_init ((pthread_spinlock_t *) spinlock, PTHREAD_PROCESS_PRIVATE);
+
+    if (error_code) {
+      return error_code;
+    }
+
+    WinpthreadsSpinlock *oldLock = (WinpthreadsSpinlock *) InterlockedCompareExchangePointer (
+      (void **) pSpinlock, *spinlock, WINPTHREADS_SPINLOCK_INITIALIZER
+    );
+
+    /**
+     * Some other thread was faster than us.
+     */
+    if (unlikely (oldLock != WINPTHREADS_SPINLOCK_INITIALIZER)) {
+      pthread_spin_destroy ((pthread_spinlock_t *) spinlock);
+      *spinlock = oldLock;
+    }
+  } else {
+    *spinlock = *pSpinlock;
+  }
+
+  if (unlikely (*spinlock == NULL)) {
+    return EINVAL;
+  }
+
+  return 0;
+}
+
+int pthread_spin_init (pthread_spinlock_t *lock, int pshared)
+{
+  WinpthreadsSpinlock **pSpinlock = (WinpthreadsSpinlock **) lock;
+
+  if (unlikely (pSpinlock == NULL)) {
+    return EINVAL;
+  }
+
+  if (unlikely (pshared != PTHREAD_PROCESS_PRIVATE && pshared != PTHREAD_PROCESS_SHARED)) {
+    return EINVAL;
+  }
+
+  if (unlikely (pshared == PTHREAD_PROCESS_SHARED)) {
+    return ENOSYS;
+  }
+
+  WinpthreadsSpinlock *wSpinlock = calloc (1, sizeof (WinpthreadsSpinlock));
+
+  /**
+   * The pthread_spin_init() function shall fail if:
+   *
+   * [ENOMEM]
+   *  Insufficient memory exists to initialize the lock.
+   */
+  if (wSpinlock == NULL) {
+    return ENOMEM;
+  }
+
+  wSpinlock->ThreadId = THREAD_ID_NO_OWNER;
+  wSpinlock->Event    = CreateEventA (NULL, FALSE, TRUE, NULL);
+
+  /**
+   * The pthread_spin_init() function shall fail if:
+   *
+   * [EAGAIN]
+   *  The system lacks the necessary resources to initialize another spin lock.
+   */
+  if (wSpinlock->Event == NULL) {
+    free (wSpinlock);
+    return EAGAIN;
+  }
+
+  *pSpinlock = wSpinlock;
+
+  return 0;
+}
+
+int pthread_spin_destroy (pthread_spinlock_t *lock)
+{
+  WinpthreadsSpinlock **pSpinlock = (WinpthreadsSpinlock **) lock;
+
+  /**
+   * POSIX:
+   *
+   * If an implementation detects that the value specified by the lock argument
+   * to pthread_spin_destroy() does not refer to an initialized spin lock object,
+   * it is recommended that the function should fail and report an [EINVAL] error.
+   */
+  if (unlikely (pSpinlock == NULL)) {
+    return EINVAL;
+  }
+
+  /**
+   * If `lock` points to statically initialized `pthread_spinlock_t` object,
+   * this will immediately invalidate it, minimizing the window for
+   * `pthread_spin_lock` and `pthread_spin_trylock` to attempt using it.
+   */
+  WinpthreadsSpinlock *wSpinlock = InterlockedCompareExchangePointer (
+    (void **) pSpinlock, NULL, WINPTHREADS_SPINLOCK_INITIALIZER
+  );
+
+  if (unlikely (wSpinlock == NULL)) {
+    return EINVAL;
+  }
+
+  if (unlikely (wSpinlock == WINPTHREADS_SPINLOCK_INITIALIZER)) {
+    return 0;
+  }
+
+  switch (_pthread_wait_for_single_object (wSpinlock->Event, 0)) {
+    /**
+     * `wSpinlock->Event` was in signaled state, which means it was unlocked;
+     * we are holding the lock now which prevents other threads from locking it.
+     */
+    case WAIT_OBJECT_0:
+      break;
+    /**
+     * `wSpinlock->Event` was in not-signaled state, which means some thread
+     * holds the lock.
+     *
+     * POSIX:
+     *
+     * If an implementation detects that the value specified by the lock argument
+     * to pthread_spin_destroy() or pthread_spin_init() refers to a locked spin
+     * lock object, or detects that the value specified by the lock argument to
+     * pthread_spin_init() refers to an already initialized spin lock object,
+     * it is recommended that the function should fail and report an [EBUSY] error.
+     */
+    case WAIT_TIMEOUT:
+      return EBUSY;
+    default:
+      return EINVAL;
+  }
+
+  /**
+   * Invalidate `pthread_spinlock_t` object pointed by `lock`.
+   */
+  InterlockedExchangePointer ((void **) pSpinlock, NULL);
+
+  CloseHandle (wSpinlock->Event);
+  free (wSpinlock);
+
+  return 0;
+}
+
+int pthread_spin_lock (pthread_spinlock_t *lock)
+{
+  WinpthreadsSpinlock *wSpinlock = NULL;
+
+  int error_code = WinpthreadsSpinlockGet (lock, &wSpinlock);
+
+  if (error_code) {
+    return error_code;
+  }
+
+  DWORD threadId = GetCurrentThreadId ();
+
+  /**
+   * The pthread_spin_lock() function may fail if:
+   *
+   * [EDEADLK]
+   *  A deadlock condition was detected.
+   */
+  if (unlikely (wSpinlock->ThreadId == threadId)) {
+    return EDEADLK;
+  }
+
+  switch (_pthread_wait_for_single_object (wSpinlock->Event, INFINITE)) {
+    /**
+     * We are holding the lock now and `wSpinlock->Event` was reset to
+     * non-signaled state.
+     */
+    case WAIT_OBJECT_0:
+      break;
+    default:
+      return EINVAL;
+  }
+
+  wSpinlock->ThreadId = threadId;
+
+  return 0;
+}
+
+int pthread_spin_trylock (pthread_spinlock_t *lock)
+{
+  WinpthreadsSpinlock *wSpinlock = NULL;
+
+  int error_code = WinpthreadsSpinlockGet (lock, &wSpinlock);
+
+  if (error_code) {
+    return error_code;
+  }
+
+  /**
+   * Unlike `pthread_spin_lock`, no deadlock can occur even if calling thread
+   * owns the lock.
+   */
+  switch (_pthread_wait_for_single_object (wSpinlock->Event, 0)) {
+    /**
+     * `wSpinlock->Event` was in signaled state, which means it was unlocked;
+     * we are holding the lock now and `wSpinlock->Event` was reset to
+     * non-signaled state.
+     */
+    case WAIT_OBJECT_0:
+      break;
+    /**
+     * `wSpinlock->Event` was in non-signaled state, which means some thread
+     * holds the lock.
+     *
+     * The pthread_spin_trylock() function shall fail if:
+     *
+     * [EBUSY]
+     *  A thread currently holds the lock.
+     */
+    case WAIT_TIMEOUT:
+      return EBUSY;
+    default:
+      return EINVAL;
+  }
+
+  wSpinlock->ThreadId = GetCurrentThreadId ();
+
+  return 0;
+}
+
+int pthread_spin_unlock (pthread_spinlock_t *lock)
+{
+  WinpthreadsSpinlock *wSpinlock = NULL;
+
+  int error_code = WinpthreadsSpinlockGet (lock, &wSpinlock);
+
+  if (error_code) {
+    return error_code;
+  }
+
+  DWORD threadId = GetCurrentThreadId ();
+
+  /**
+   * POSIX:
+   *
+   * If an implementation detects that the value specified by the lock argument
+   * to pthread_spin_unlock() refers to a spin lock object for which the
+   * current thread does not hold the lock, it is recommended that the function
+   * should fail and report an [EPERM] error.
+   */
+  if (unlikely (wSpinlock->ThreadId != threadId)) {
+    return EPERM;
+  }
+
+  wSpinlock->ThreadId = THREAD_ID_NO_OWNER;
+
+  if (!SetEvent (wSpinlock->Event)) {
+    return EINVAL;
+  }
+
+  return 0;
+}
diff --git a/vendor/winpthreads/src/thread.c b/vendor/winpthreads/src/thread.c
new file mode 100644
index 00000000..47637042
--- /dev/null
+++ b/vendor/winpthreads/src/thread.c
@@ -0,0 +1,1910 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#if defined(__arm__) || defined(__aarch64__)
+/* We use setjmp/longjmp through asynchronous function calls via
+ * SetThreadContext below. This makes unwinding from longjmp not
+ * work reliably; therefore use a version of setjmp/longjmp that doesn't
+ * rely on SEH. */
+#define __USE_MINGW_SETJMP_NON_SEH
+#endif
+
+#define __LARGE_MBSTATE_T
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <malloc.h>
+#include <signal.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <wchar.h>
+
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#define WINPTHREAD_THREAD_DECL WINPTHREAD_API
+
+/* public header files */
+#include "pthread.h"
+/* internal header files */
+#include "misc.h"
+#include "thread.h"
+
+static _pthread_v *__pthread_self_lite (void);
+
+void (**_pthread_key_dest)(void *) = NULL;
+
+static volatile long _pthread_cancelling;
+static int _pthread_concur;
+
+/* FIXME Will default to zero as needed */
+static pthread_once_t _pthread_tls_once;
+static DWORD _pthread_tls = 0xffffffff;
+
+static pthread_rwlock_t _pthread_key_lock = PTHREAD_RWLOCK_INITIALIZER;
+static unsigned long _pthread_key_max=0L;
+static unsigned long _pthread_key_sch=0L;
+
+static _pthread_v *pthr_root = NULL, *pthr_last = NULL;
+static pthread_mutex_t mtx_pthr_locked = PTHREAD_RECURSIVE_MUTEX_INITIALIZER;
+
+static __pthread_idlist *idList = NULL;
+static size_t idListCnt = 0;
+static size_t idListMax = 0;
+static pthread_t idListNextId = 0;
+
+#if defined(__SEH__) && (!defined(__clang__) || __clang_major__ >= 7)
+#define SEH_INLINE_ASM
+#ifdef __arm__
+#define ASM_EXCEPT "%%except"
+#else
+#define ASM_EXCEPT "@except"
+#endif
+#endif
+
+#if !defined(_MSC_VER) && (defined(__i386__) || defined(SEH_INLINE_ASM))
+static EXCEPTION_DISPOSITION __cdecl
+SetThreadName_SEH (EXCEPTION_RECORD *ExceptionRecord, PVOID EstablisherFrame, CONTEXT *ContextRecord, PVOID DispatcherContext)
+{
+  /* Do not be confused with VEH handlers and CRT except filters which returns LONG value with UPPER_CASE constants.
+   * SEH handlers like this one return value from EXCEPTION_DISPOSITION enum which has CamelCase constants.
+   * UPPER_CASE EXCEPTION_CONTINUE_SEARCH and CamelCase ExceptionContinueSearch are different constants.
+   */
+  if (!(ExceptionRecord->ExceptionFlags & EXCEPTION_UNWINDING) &&
+      !(ExceptionRecord->ExceptionFlags & EXCEPTION_NONCONTINUABLE) &&
+      ExceptionRecord->ExceptionCode == EXCEPTION_SET_THREAD_NAME)
+    return ExceptionContinueExecution;
+
+  return ExceptionContinueSearch;
+}
+#endif
+
+typedef struct _THREADNAME_INFO
+{
+  DWORD  dwType;	/* must be 0x1000 */
+  LPCSTR szName;	/* pointer to name (in user addr space) */
+  DWORD  dwThreadID;	/* thread ID (-1=caller thread) */
+  DWORD  dwFlags;	/* reserved for future use, must be zero */
+} THREADNAME_INFO;
+
+#if !defined(_MSC_VER) && !defined(__i386__) && defined(SEH_INLINE_ASM)
+WINPTHREADS_ATTRIBUTE((noinline)) /* required for asm .seh_handler directive */
+#endif
+static void
+SetThreadName (DWORD dwThreadID, LPCSTR szThreadName)
+{
+   THREADNAME_INFO info;
+   DWORD infosize;
+
+   info.dwType = 0x1000;
+   info.szName = szThreadName;
+   info.dwThreadID = dwThreadID;
+   info.dwFlags = 0;
+
+   infosize = sizeof (info) / sizeof (ULONG_PTR);
+
+   /* Exception has to be processed otherwise it will crash the process. */
+
+#if defined(_MSC_VER)
+   /* msvc supports __try / __except syntax, so use it */
+   __try
+     {
+       RaiseException (EXCEPTION_SET_THREAD_NAME, 0, infosize, (ULONG_PTR *)&info);
+     }
+   __except (EXCEPTION_EXECUTE_HANDLER)
+     {
+     }
+#else
+   /* gcc does not support __try / __except syntax, so manually register SEH handler */
+#if defined(__i386__)
+   /* On 32-bit x86 is SEH handler registered and unregistered at runtime */
+   EXCEPTION_REGISTRATION_RECORD exception_record = {
+     .Next = (EXCEPTION_REGISTRATION_RECORD *) __readfsdword (0), /* current SEH handler */
+     .Handler = (PEXCEPTION_ROUTINE)(void*) SetThreadName_SEH,
+   };
+   __writefsdword (0, (DWORD) &exception_record); /* register our SEH handler */
+   RaiseException (EXCEPTION_SET_THREAD_NAME, 0, infosize, (ULONG_PTR *) &info);
+   __writefsdword (0, (DWORD) exception_record.Next); /* unregister our SEH handler */
+#elif defined(SEH_INLINE_ASM)
+   /* On other platforms SEH handlers are registered at compile time.
+      Assembler directive .seh_handler statically register SEH handler for
+      the whole current function. It does not matter at which line is this
+      directive called. It always applies for the whole function, so also
+      for code before the directive itself. As this function does not do
+      anything else, we can register our SEH handler for the whole function.
+      This function has to be marked as noinline to ensure that the SEH
+      handler would not be registered for a caller.
+    */
+   asm volatile (".seh_handler %c0, " ASM_EXCEPT :: "i" (SetThreadName_SEH));
+   RaiseException (EXCEPTION_SET_THREAD_NAME, 0, infosize, (ULONG_PTR *) &info);
+#else
+   /* Other compilers / platforms do not provide SEH support */
+#endif
+#endif
+}
+
+/* Search the list idList for an element with identifier ID.  If
+   found, its associated _pthread_v pointer is returned, otherwise
+   NULL.
+   NOTE: This method is not locked.  */
+static struct _pthread_v *
+__pthread_get_pointer (pthread_t id)
+{
+  size_t l, r, p;
+  if (!idListCnt)
+    return NULL;
+  if (idListCnt == 1)
+    return (idList[0].id == id ? idList[0].ptr : NULL);
+  l = 0; r = idListCnt - 1;
+  while (l <= r)
+  {
+    p = (l + r) >> 1;
+    if (idList[p].id == id)
+      return idList[p].ptr;
+    else if (idList[p].id > id)
+      {
+	if (p == l)
+	  return NULL;
+	r = p - 1;
+      }
+    else
+      {
+	l = p + 1;
+      }
+  }
+
+  return NULL;
+}
+
+static void
+__pth_remove_use_for_key (pthread_key_t key)
+{
+  int i;
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  for (i = 0; i < idListCnt; i++)
+    {
+      if (idList[i].ptr != NULL
+          && idList[i].ptr->keyval != NULL
+          && key < idList[i].ptr->keymax)
+        {
+	  idList[i].ptr->keyval[key] = NULL;
+	  idList[i].ptr->keyval_set[key] = 0;
+	}
+    }
+  pthread_mutex_unlock (&mtx_pthr_locked);
+}
+
+/* Search the list idList for an element with identifier ID.  If
+   found, its associated _pthread_v pointer is returned, otherwise
+   NULL.
+   NOTE: This method uses lock mtx_pthr_locked.  */
+struct _pthread_v *
+__pth_gpointer_locked (pthread_t id)
+{
+  struct _pthread_v *ret;
+  if (!id)
+    return NULL;
+  pthread_mutex_lock (&mtx_pthr_locked);
+  ret =  __pthread_get_pointer (id);
+  pthread_mutex_unlock (&mtx_pthr_locked);
+  return ret;
+}
+
+/* Registers in the list idList an element with _pthread_v pointer
+   and creates and unique identifier ID.  If successful created the
+   ID of this element is returned, otherwise on failure zero ID gets
+   returned.
+   NOTE: This method is not locked.  */
+static pthread_t
+__pthread_register_pointer (struct _pthread_v *ptr)
+{
+  __pthread_idlist *e;
+  size_t i;
+
+  if (!ptr)
+    return 0;
+  /* Check if a resize of list is necessary.  */
+  if (idListCnt >= idListMax)
+    {
+      if (!idListCnt)
+        {
+	  e = (__pthread_idlist *) malloc (sizeof (__pthread_idlist) * 16);
+	  if (!e)
+	    return 0;
+	  idListMax = 16;
+	  idList = e;
+	}
+      else
+        {
+	  e = (__pthread_idlist *) realloc (idList, sizeof (__pthread_idlist) * (idListMax + 16));
+	  if (!e)
+	    return 0;
+	  idListMax += 16;
+	  idList = e;
+	}
+    }
+  do
+    {
+      ++idListNextId;
+      /* If two MSB are set we reset to id 1.  We need to check here bits
+         to avoid gcc's no-overflow issue on increment.  Additionally we
+         need to handle different size of pthread_t on 32-bit/64-bit.  */
+      if ((idListNextId & ( ((pthread_t) 1) << ((sizeof (pthread_t) * 8) - 2))) != 0)
+        idListNextId = 1;
+    }
+  while (idListNextId == 0 || __pthread_get_pointer (idListNextId));
+  /* We assume insert at end of list.  */
+  i = idListCnt;
+  if (i != 0)
+    {
+      /* Find position we can actual insert sorted.  */
+      while (i > 0 && idList[i - 1].id > idListNextId)
+        --i;
+      if (i != idListCnt)
+	memmove (&idList[i + 1], &idList[i], sizeof (__pthread_idlist) * (idListCnt - i));
+    }
+  idList[i].id = idListNextId;
+  idList[i].ptr = ptr;
+  ++idListCnt;
+  return idListNextId;
+}
+
+/* Deregisters in the list idList an element with identifier ID and
+   returns its _pthread_v pointer on success.  Otherwise NULL is returned.
+   NOTE: This method is not locked.  */
+static struct _pthread_v *
+__pthread_deregister_pointer (pthread_t id)
+{
+  size_t l, r, p;
+  if (!idListCnt)
+    return NULL;
+  l = 0; r = idListCnt - 1;
+  while (l <= r)
+  {
+    p = (l + r) >> 1;
+    if (idList[p].id == id)
+      {
+	struct _pthread_v *ret = idList[p].ptr;
+	p++;
+	if (p < idListCnt)
+	  memmove (&idList[p - 1], &idList[p], sizeof (__pthread_idlist) * (idListCnt - p));
+	--idListCnt;
+	/* Is this last element in list then free list.  */
+	if (idListCnt == 0)
+	{
+	  free (idList);
+	  idListCnt = idListMax = 0;
+	}
+	return ret;
+      }
+    else if (idList[p].id > id)
+      {
+	if (p == l)
+	  return NULL;
+	r = p - 1;
+      }
+    else
+      {
+	l = p + 1;
+      }
+  }
+  return NULL;
+}
+
+/* Save a _pthread_v element for reuse in pool.  */
+static void
+push_pthread_mem (_pthread_v *sv)
+{
+  if (!sv || sv->next != NULL)
+    return;
+  pthread_mutex_lock (&mtx_pthr_locked);
+  if (sv->x != 0)
+    __pthread_deregister_pointer (sv->x);
+  if (sv->keyval)
+    free (sv->keyval);
+  if (sv->keyval_set)
+    free (sv->keyval_set);
+  if (sv->thread_name)
+    free (sv->thread_name);
+  memset (sv, 0, sizeof(struct _pthread_v));
+  if (pthr_last == NULL)
+    pthr_root = pthr_last = sv;
+  else
+  {
+    pthr_last->next = sv;
+    pthr_last = sv;
+  }
+  pthread_mutex_unlock (&mtx_pthr_locked);
+}
+
+/* Get a _pthread_v element from pool, or allocate it.
+   Note the unique identifier is created for the element here, too.  */
+static _pthread_v *
+pop_pthread_mem (void)
+{
+  _pthread_v *r = NULL;
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  if ((r = pthr_root) == NULL)
+    {
+      if ((r = (_pthread_v *)calloc (1,sizeof(struct _pthread_v))) != NULL)
+	{
+	  r->x = __pthread_register_pointer (r);
+	  if (r->x == 0)
+	    {
+	      free (r);
+	      r = NULL;
+	    }
+	}
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      return r;
+    }
+  r->x = __pthread_register_pointer (r);
+  if (r->x == 0)
+    r = NULL;
+  else
+    {
+      if((pthr_root = r->next) == NULL)
+	pthr_last = NULL;
+
+      r->next = NULL;
+    }
+  pthread_mutex_unlock (&mtx_pthr_locked);
+  return r;
+}
+
+/* Free memory consumed in _pthread_v pointer pool.  */
+static void
+free_pthread_mem (void)
+{
+#if 0
+  _pthread_v *t;
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  t = pthr_root;
+  while (t != NULL)
+  {
+    _pthread_v *sv = t;
+    t = t->next;
+    if (sv->x != 0 && sv->ended == 0 && sv->valid != DEAD_THREAD)
+      {
+	pthread_mutex_unlock (&mtx_pthr_locked);
+	pthread_cancel (t->x);
+	Sleep (0);
+	pthread_mutex_lock (&mtx_pthr_locked);
+	t = pthr_root;
+	continue;
+      }
+    else if (sv->x != 0 && sv->valid != DEAD_THREAD)
+      {
+	pthread_mutex_unlock (&mtx_pthr_locked);
+	Sleep (0);
+	pthread_mutex_lock (&mtx_pthr_locked);
+	continue;
+      }
+    if (sv->x != 0)
+      __pthread_deregister_pointer (sv->x);
+    sv->x = 0;
+    free (sv);
+    pthr_root = t;
+  }
+  pthread_mutex_unlock (&mtx_pthr_locked);
+#endif
+  return;
+}
+
+static void
+replace_spin_keys (pthread_spinlock_t *old, pthread_spinlock_t new)
+{
+  if (old == NULL)
+    return;
+
+  if (EPERM == pthread_spin_destroy (old))
+    {
+#define THREADERR "Error cleaning up spin_keys for thread %lu.\n"
+      char threaderr[sizeof(THREADERR) + 8] = { 0 };
+      snprintf(threaderr, sizeof(threaderr), THREADERR, GetCurrentThreadId());
+#undef THREADERR
+      OutputDebugStringA (threaderr);
+      abort ();
+    }
+
+  *old = new;
+}
+
+/* Hook for TLS-based deregistration/registration of thread.  */
+static void WINAPI
+__dyn_tls_pthread (HANDLE hDllHandle, DWORD dwReason, LPVOID lpreserved)
+{
+  _pthread_v *t = NULL;
+  pthread_spinlock_t new_spin_keys = PTHREAD_SPINLOCK_INITIALIZER;
+
+  if (dwReason == DLL_PROCESS_DETACH)
+    {
+      free_pthread_mem ();
+    }
+  else if (dwReason == DLL_PROCESS_ATTACH)
+    {
+    }
+  else if (dwReason == DLL_THREAD_DETACH)
+    {
+      if (_pthread_tls != 0xffffffff)
+	t = (_pthread_v *)TlsGetValue(_pthread_tls);
+      if (t && t->thread_noposix != 0)
+	{
+	  _pthread_cleanup_dest (t->x);
+	  if (t->h != NULL)
+	    {
+	      CloseHandle (t->h);
+	      if (t->evStart)
+		CloseHandle (t->evStart);
+	      t->evStart = NULL;
+	      t->h = NULL;
+	    }
+	  pthread_mutex_destroy (&t->p_clock);
+	  replace_spin_keys (&t->spin_keys, new_spin_keys);
+	  push_pthread_mem (t);
+	  t = NULL;
+	  TlsSetValue (_pthread_tls, t);
+	}
+      else if (t && t->ended == 0)
+	{
+	  if (t->evStart)
+	    CloseHandle(t->evStart);
+	  t->evStart = NULL;
+	  t->ended = 1;
+	  _pthread_cleanup_dest (t->x);
+	  if ((t->p_state & PTHREAD_CREATE_DETACHED) == PTHREAD_CREATE_DETACHED)
+	    {
+	      t->valid = DEAD_THREAD;
+	      if (t->h != NULL)
+		CloseHandle (t->h);
+	      t->h = NULL;
+	      pthread_mutex_destroy(&t->p_clock);
+	      replace_spin_keys (&t->spin_keys, new_spin_keys);
+	      push_pthread_mem (t);
+	      t = NULL;
+	      TlsSetValue (_pthread_tls, t);
+	      return;
+	    }
+	  pthread_mutex_destroy(&t->p_clock);
+	  replace_spin_keys (&t->spin_keys, new_spin_keys);
+	}
+      else if (t)
+	{
+	  if (t->evStart)
+	    CloseHandle (t->evStart);
+	  t->evStart = NULL;
+	  pthread_mutex_destroy (&t->p_clock);
+	  replace_spin_keys (&t->spin_keys, new_spin_keys);
+	}
+    }
+}
+
+/* TLS-runtime section variable.  */
+
+/* Force a reference to _tls_used to make the linker create the TLS
+ * directory if it's not already there.  (e.g. if __declspec(thread)
+ * is not used).
+ * Force a reference to __xl_f to prevent whole program optimization
+ * from discarding the variable. */
+#if defined(__GNUC__)
+extern const IMAGE_TLS_DIRECTORY _tls_used;
+static __attribute__((used)) const IMAGE_TLS_DIRECTORY *const _include_tls_used = &_tls_used;
+#elif defined(_MSC_VER)
+/* On x86, symbols are prefixed with an underscore. */
+# if defined(_M_IX86)
+#   pragma comment(linker, "/include:__tls_used")
+#   pragma comment(linker, "/include:___xl_f")
+# else
+#   pragma comment(linker, "/include:_tls_used")
+#   pragma comment(linker, "/include:__xl_f")
+# endif
+
+/* .CRT$XLA to .CRT$XLZ is an array of PIMAGE_TLS_CALLBACK
+ * pointers. Pick an arbitrary location for our callback.
+ *
+ * See VC\...\crt\src\vcruntime\tlssup.cpp for reference. */
+
+# pragma section(".CRT$XLF", long, read)
+#endif
+
+#if defined(__GNUC__)
+static __attribute__((used))
+#endif
+WINPTHREADS_ATTRIBUTE((WINPTHREADS_SECTION(".CRT$XLF")))
+const PIMAGE_TLS_CALLBACK __xl_f = __dyn_tls_pthread;
+
+/* Internal collect-once structure.  */
+typedef struct collect_once_t {
+  pthread_once_t *o;
+  pthread_mutex_t m;
+  int count;
+  struct collect_once_t *next;
+} collect_once_t;
+
+static collect_once_t *once_obj = NULL;
+
+static pthread_spinlock_t once_global = PTHREAD_SPINLOCK_INITIALIZER;
+
+static collect_once_t *
+enterOnceObject (pthread_once_t *o)
+{
+  collect_once_t *c, *p = NULL;
+  pthread_spin_lock (&once_global);
+  c = once_obj;
+  while (c != NULL && c->o != o)
+    {
+      c = (p = c)->next;
+    }
+  if (!c)
+    {
+      c = (collect_once_t *) calloc(1,sizeof(collect_once_t));
+      c->o = o;
+      c->count = 1;
+      if (!p)
+        once_obj = c;
+      else
+        p->next = c;
+      pthread_mutex_init(&c->m, NULL);
+    }
+  else
+    c->count += 1;
+  pthread_spin_unlock (&once_global);
+  return c;
+}
+
+static void
+leaveOnceObject (collect_once_t *c)
+{
+  collect_once_t *h, *p = NULL;
+  if (!c)
+    return;
+  pthread_spin_lock (&once_global);
+  h = once_obj;
+  while (h != NULL && c != h)
+    h = (p = h)->next;
+
+  if (h)
+    {
+      c->count -= 1;
+      if (c->count == 0)
+	{
+	  pthread_mutex_destroy(&c->m);
+	  if (!p)
+	    once_obj = c->next;
+	  else
+	    p->next = c->next;
+	  free (c);
+	}
+    }
+  else
+    fprintf(stderr, "%p not found?!?!\n", (void *) c);
+  pthread_spin_unlock (&once_global);
+}
+
+static void
+_pthread_once_cleanup (void *o)
+{
+  collect_once_t *co = (collect_once_t *) o;
+  pthread_mutex_unlock (&co->m);
+  leaveOnceObject (co);
+}
+
+static int
+_pthread_once_raw (pthread_once_t *o, void (*func)(void))
+{
+  collect_once_t *co;
+  long state = *o;
+
+  CHECK_PTR(o);
+  CHECK_PTR(func);
+
+  if (state == 1)
+    return 0;
+  co = enterOnceObject(o);
+  pthread_mutex_lock(&co->m);
+  if (*o == 0)
+    {
+      func();
+      *o = 1;
+    }
+  else if (*o != 1)
+    fprintf (stderr," once %p is %ld\n", (void *) o, (long) *o);
+  pthread_mutex_unlock(&co->m);
+  leaveOnceObject(co);
+
+  /* Done */
+  return 0;
+}
+
+/* Unimplemented.  */
+void *
+pthread_timechange_handler_np(void *dummy)
+{
+  return NULL;
+}
+
+/* Compatibility routine for pthread-win32.  It waits for ellapse of
+   interval and additionally checks for possible thread-cancelation.  */
+static int
+__pthread_delay_np (const struct _timespec64 *interval)
+{
+  DWORD to = (!interval ? 0 : dwMilliSecs (_pthread_time_in_ms_from_timespec (interval)));
+  struct _pthread_v *s = __pthread_self_lite ();
+
+  if (!to)
+    {
+      pthread_testcancel ();
+      Sleep (0);
+      pthread_testcancel ();
+      return 0;
+    }
+  pthread_testcancel ();
+  if (s->evStart)
+    _pthread_wait_for_single_object (s->evStart, to);
+  else
+    Sleep (to);
+  pthread_testcancel ();
+  return 0;
+}
+
+int
+pthread_delay64_np (const struct _timespec64 *interval)
+{
+  return __pthread_delay_np(interval);
+}
+
+int
+pthread_delay32_np (const struct _timespec32 *interval)
+{
+  struct _timespec64 interval64 = {.tv_sec = interval->tv_sec, .tv_nsec = interval->tv_nsec};
+  return __pthread_delay_np(&interval64);
+}
+
+int
+_pthread_delay_np_ms (DWORD to)
+{
+  struct _pthread_v *s = __pthread_self_lite ();
+
+  if (!to)
+    {
+      pthread_testcancel ();
+      Sleep (0);
+      pthread_testcancel ();
+      return 0;
+    }
+  pthread_testcancel ();
+  if (s->evStart)
+    _pthread_wait_for_single_object (s->evStart, to);
+  else
+    Sleep (to);
+  pthread_testcancel ();
+  return 0;
+}
+
+/* Compatibility routine for pthread-win32.  It returns the
+   amount of available CPUs on system.  */
+int
+pthread_num_processors_np(void)
+{
+  int r = 0;
+  DWORD_PTR ProcessAffinityMask, SystemAffinityMask;
+
+  if (GetProcessAffinityMask(GetCurrentProcess(), &ProcessAffinityMask, &SystemAffinityMask))
+    {
+      for(; ProcessAffinityMask != 0; ProcessAffinityMask >>= 1)
+	r += (ProcessAffinityMask & 1) != 0;
+    }
+  /* assume at least 1 */
+  return r ? r : 1;
+}
+
+/* Compatiblity routine for pthread-win32.  Allows to set amount of used
+   CPUs for process.  */
+int
+pthread_set_num_processors_np(int n)
+{
+  DWORD_PTR ProcessAffinityMask, ProcessNewAffinityMask = 0, SystemAffinityMask;
+  int r = 0;
+  /* need at least 1 */
+  n = n ? n : 1;
+  if (GetProcessAffinityMask (GetCurrentProcess (), &ProcessAffinityMask, &SystemAffinityMask))
+    {
+      for (; ProcessAffinityMask != 0; ProcessAffinityMask >>= 1)
+	{
+	  ProcessNewAffinityMask <<= 1;
+	  if ((ProcessAffinityMask & 1) != 0 && r < n)
+	    {
+	      ProcessNewAffinityMask |= 1;
+	      r++;
+	    }
+	}
+      SetProcessAffinityMask (GetCurrentProcess (),ProcessNewAffinityMask);
+    }
+  return r;
+}
+
+int
+pthread_once (pthread_once_t *o, void (*func)(void))
+{
+  collect_once_t *co;
+  long state = *o;
+
+  CHECK_PTR(o);
+  CHECK_PTR(func);
+
+  if (state == 1)
+    return 0;
+  co = enterOnceObject(o);
+  pthread_mutex_lock(&co->m);
+  if (*o == 0)
+    {
+      pthread_cleanup_push(_pthread_once_cleanup, co);
+      func();
+      pthread_cleanup_pop(0);
+      *o = 1;
+    }
+  else if (*o != 1)
+    fprintf (stderr," once %p is %ld\n", (void *) o, (long) *o);
+  pthread_mutex_unlock(&co->m);
+  leaveOnceObject(co);
+
+  return 0;
+}
+
+int
+pthread_key_create (pthread_key_t *key, void (* dest)(void *))
+{
+	unsigned int i;
+	long nmax;
+	void (**d)(void *);
+
+	if (!key)
+		return EINVAL;
+
+	pthread_rwlock_wrlock (&_pthread_key_lock);
+
+	for (i = _pthread_key_sch; i < _pthread_key_max; i++)
+	{
+		if (!_pthread_key_dest[i])
+		{
+			*key = i;
+			if (dest)
+				_pthread_key_dest[i] = dest;
+			else
+				_pthread_key_dest[i] = (void(*)(void *))1;
+			pthread_rwlock_unlock (&_pthread_key_lock);
+			return 0;
+		}
+	}
+
+	for (i = 0; i < _pthread_key_sch; i++)
+	{
+		if (!_pthread_key_dest[i])
+		{
+			*key = i;
+			if (dest)
+				_pthread_key_dest[i] = dest;
+			else
+				_pthread_key_dest[i] = (void(*)(void *))1;
+			pthread_rwlock_unlock (&_pthread_key_lock);
+
+			return 0;
+		}
+	}
+
+	if (_pthread_key_max == PTHREAD_KEYS_MAX)
+	{
+		pthread_rwlock_unlock(&_pthread_key_lock);
+		return ENOMEM;
+	}
+
+	nmax = _pthread_key_max * 2;
+	if (nmax == 0)
+		nmax = _pthread_key_max + 1;
+	if (nmax > PTHREAD_KEYS_MAX)
+		nmax = PTHREAD_KEYS_MAX;
+
+	/* No spare room anywhere */
+	d = (void (__cdecl **)(void *))realloc(_pthread_key_dest, nmax * sizeof(*d));
+	if (!d)
+	{
+		pthread_rwlock_unlock (&_pthread_key_lock);
+		return ENOMEM;
+	}
+
+	/* Clear new region */
+	memset ((void *) &d[_pthread_key_max], 0, (nmax-_pthread_key_max)*sizeof(void *));
+
+	/* Use new region */
+	_pthread_key_dest = d;
+	_pthread_key_sch = _pthread_key_max + 1;
+	*key = _pthread_key_max;
+	_pthread_key_max = nmax;
+
+	if (dest)
+		_pthread_key_dest[*key] = dest;
+	else
+		_pthread_key_dest[*key] = (void(*)(void *))1;
+
+	pthread_rwlock_unlock (&_pthread_key_lock);
+	return 0;
+}
+
+int
+pthread_key_delete (pthread_key_t key)
+{
+  if (key >= _pthread_key_max || !_pthread_key_dest)
+    return EINVAL;
+
+  pthread_rwlock_wrlock (&_pthread_key_lock);
+
+  _pthread_key_dest[key] = NULL;
+
+  /* Start next search from our location */
+  if (_pthread_key_sch > key)
+    _pthread_key_sch = key;
+  /* So now we need to walk the complete list of threads
+     and remove key's reference for it.  */
+  __pth_remove_use_for_key (key);
+
+  pthread_rwlock_unlock (&_pthread_key_lock);
+  return 0;
+}
+
+void *
+pthread_getspecific (pthread_key_t key)
+{
+  DWORD lasterr = GetLastError ();
+  void *r;
+  _pthread_v *t = __pthread_self_lite ();
+  pthread_spin_lock (&t->spin_keys);
+  r = (key >= t->keymax || t->keyval_set[key] == 0 ? NULL : t->keyval[key]);
+  pthread_spin_unlock (&t->spin_keys);
+  SetLastError (lasterr);
+  return r;
+}
+
+int
+pthread_setspecific (pthread_key_t key, const void *value)
+{
+  DWORD lasterr = GetLastError ();
+  _pthread_v *t = __pthread_self_lite ();
+
+  pthread_spin_lock (&t->spin_keys);
+
+  if (key >= t->keymax)
+    {
+      int keymax = (key + 1);
+      void **kv;
+      unsigned char *kv_set;
+
+      kv = (void **) realloc (t->keyval, keymax * sizeof (void *));
+
+      if (!kv)
+        {
+	  pthread_spin_unlock (&t->spin_keys);
+	  return ENOMEM;
+	}
+      kv_set = (unsigned char *) realloc (t->keyval_set, keymax);
+      if (!kv_set)
+        {
+	  pthread_spin_unlock (&t->spin_keys);
+	  return ENOMEM;
+	}
+
+      /* Clear new region */
+      memset (&kv[t->keymax], 0, (keymax - t->keymax)*sizeof(void *));
+      memset (&kv_set[t->keymax], 0, (keymax - t->keymax));
+
+      t->keyval = kv;
+      t->keyval_set = kv_set;
+      t->keymax = keymax;
+    }
+
+  t->keyval[key] = (void *) value;
+  t->keyval_set[key] = 1;
+  pthread_spin_unlock (&t->spin_keys);
+  SetLastError (lasterr);
+
+  return 0;
+}
+
+int
+pthread_equal (pthread_t t1, pthread_t t2)
+{
+  return (t1 == t2);
+}
+
+void
+pthread_tls_init (void)
+{
+  _pthread_tls = TlsAlloc();
+
+  /* Cannot continue if out of indexes */
+  if (_pthread_tls == TLS_OUT_OF_INDEXES)
+    abort();
+}
+
+void
+_pthread_cleanup_dest (pthread_t t)
+{
+	_pthread_v *tv;
+	unsigned int j;
+	int i;
+
+	if (!t)
+		return;
+	tv = __pth_gpointer_locked (t);
+	if (!tv)
+		return;
+
+	for (j = 0; j < PTHREAD_DESTRUCTOR_ITERATIONS; j++)
+	{
+		int flag = 0;
+
+		pthread_spin_lock (&tv->spin_keys);
+		for (i = tv->keymax - 1; i >= 0; i--)
+		{
+			void *val = tv->keyval[i];
+
+			if (tv->keyval_set[i])
+			{
+				pthread_rwlock_rdlock (&_pthread_key_lock);
+				if ((uintptr_t) _pthread_key_dest[i] > 1)
+				{
+					/* Call destructor */
+					tv->keyval[i] = NULL;
+					tv->keyval_set[i] = 0;
+					pthread_spin_unlock (&tv->spin_keys);
+					_pthread_key_dest[i](val);
+					pthread_spin_lock (&tv->spin_keys);
+					flag = 1;
+				}
+				else
+				{
+					tv->keyval[i] = NULL;
+					tv->keyval_set[i] = 0;
+				}
+				pthread_rwlock_unlock(&_pthread_key_lock);
+			}
+		}
+		pthread_spin_unlock (&tv->spin_keys);
+		/* Nothing to do? */
+		if (!flag)
+			return;
+	}
+}
+
+static _pthread_v *
+__pthread_self_lite (void)
+{
+  _pthread_v *t;
+  pthread_spinlock_t new_spin_keys = PTHREAD_SPINLOCK_INITIALIZER;
+
+  _pthread_once_raw (&_pthread_tls_once, pthread_tls_init);
+
+  t = (_pthread_v *) TlsGetValue (_pthread_tls);
+  if (t)
+    return t;
+  /* Main thread? */
+  t = (struct _pthread_v *) pop_pthread_mem ();
+
+  /* If cannot initialize main thread, then the only thing we can do is return null pthread_t */
+  if (!__xl_f || !t)
+    return 0;
+
+  t->p_state = PTHREAD_DEFAULT_ATTR /*| PTHREAD_CREATE_DETACHED*/;
+  t->tid = GetCurrentThreadId();
+  t->evStart = CreateEvent (NULL, 1, 0, NULL);
+  t->p_clock = PTHREAD_MUTEX_INITIALIZER;
+  replace_spin_keys (&t->spin_keys, new_spin_keys);
+  t->sched_pol = SCHED_OTHER;
+  t->h = NULL; //GetCurrentThread();
+  if (!DuplicateHandle(GetCurrentProcess(), GetCurrentThread(), GetCurrentProcess(), &t->h, 0, FALSE, DUPLICATE_SAME_ACCESS))
+    abort ();
+  t->sched.sched_priority = GetThreadPriority(t->h);
+  t->ended = 0;
+  t->thread_noposix = 1;
+
+  /* Save for later */
+  if (!TlsSetValue(_pthread_tls, t))
+    abort ();
+  return t;
+}
+
+pthread_t
+pthread_self (void)
+{
+  _pthread_v *t = __pthread_self_lite ();
+
+  if (!t)
+    return 0;
+  return t->x;
+}
+
+/* Internal helper for getting event handle of thread T.  */
+void *
+pthread_getevent (void)
+{
+  _pthread_v *t = __pthread_self_lite ();
+  return (!t ? NULL : t->evStart);
+}
+
+/* Internal helper for getting thread handle of thread T.  */
+void *
+pthread_gethandle (pthread_t t)
+{
+  struct _pthread_v *tv = __pth_gpointer_locked (t);
+  return (!tv ? NULL : tv->h);
+}
+
+/* Internal helper for getting pointer of clean of current thread.  */
+struct _pthread_cleanup **
+pthread_getclean (void)
+{
+  struct _pthread_v *t = __pthread_self_lite ();
+  if (!t) return NULL;
+  return &t->clean;
+}
+
+int
+pthread_get_concurrency (int *val)
+{
+  *val = _pthread_concur;
+  return 0;
+}
+
+int
+pthread_set_concurrency (int val)
+{
+  _pthread_concur = val;
+  return 0;
+}
+
+void
+pthread_exit (void *res)
+{
+  _pthread_v *t = NULL;
+  unsigned rslt = (unsigned) ((intptr_t) res);
+  struct _pthread_v *id = __pthread_self_lite ();
+
+  id->ret_arg = res;
+
+  _pthread_cleanup_dest (id->x);
+  if (id->thread_noposix == 0)
+    longjmp(id->jb, 1);
+
+  /* Make sure we free ourselves if we are detached */
+  if ((t = (_pthread_v *)TlsGetValue(_pthread_tls)) != NULL)
+    {
+      if (!t->h)
+	{
+	  t->valid = DEAD_THREAD;
+	  if (t->evStart)
+	    CloseHandle (t->evStart);
+	  t->evStart = NULL;
+	  rslt = (unsigned) (size_t) t->ret_arg;
+	  push_pthread_mem(t);
+	  t = NULL;
+	  TlsSetValue (_pthread_tls, t);
+	}
+      else
+	{
+	  rslt = (unsigned) (size_t) t->ret_arg;
+	  t->ended = 1;
+	  if (t->evStart)
+	    CloseHandle (t->evStart);
+	  t->evStart = NULL;
+	  if ((t->p_state & PTHREAD_CREATE_DETACHED) == PTHREAD_CREATE_DETACHED)
+	    {
+	      t->valid = DEAD_THREAD;
+	      CloseHandle (t->h);
+	      t->h = NULL;
+	      push_pthread_mem(t);
+	      t = NULL;
+	      TlsSetValue(_pthread_tls, t);
+	    }
+	}
+    }
+  /* Time to die */
+  _endthreadex(rslt);
+}
+
+void
+_pthread_invoke_cancel (void)
+{
+  _pthread_cleanup *pcup;
+  struct _pthread_v *se = __pthread_self_lite ();
+  se->in_cancel = 1;
+  _pthread_setnobreak (1);
+  InterlockedDecrement(&_pthread_cancelling);
+
+  /* Call cancel queue */
+  for (pcup = se->clean; pcup; pcup = pcup->next)
+    {
+      pcup->func((pthread_once_t *)pcup->arg);
+    }
+
+  _pthread_setnobreak (0);
+  pthread_exit(PTHREAD_CANCELED);
+}
+
+int
+__pthread_shallcancel (void)
+{
+  struct _pthread_v *t;
+  if (!_pthread_cancelling)
+    return 0;
+  t = __pthread_self_lite ();
+  if (t == NULL)
+    return 0;
+  if (t->nobreak <= 0 && t->cancelled && (t->p_state & PTHREAD_CANCEL_ENABLE))
+    return 1;
+  return 0;
+}
+
+void
+_pthread_setnobreak (int v)
+{
+  struct _pthread_v *t = __pthread_self_lite ();
+  if (t == NULL)
+    return;
+  if (v > 0)
+    InterlockedIncrement ((long*)&t->nobreak);
+  else
+    InterlockedDecrement((long*)&t->nobreak);
+}
+
+void
+pthread_testcancel (void)
+{
+  struct _pthread_v *self = __pthread_self_lite ();
+
+  if (!self || self->in_cancel)
+    return;
+  if (!_pthread_cancelling)
+    return;
+  pthread_mutex_lock (&self->p_clock);
+
+  if (self->cancelled && (self->p_state & PTHREAD_CANCEL_ENABLE) && self->nobreak <= 0)
+    {
+      self->in_cancel = 1;
+      self->p_state &= ~PTHREAD_CANCEL_ENABLE;
+      if (self->evStart)
+	ResetEvent (self->evStart);
+      pthread_mutex_unlock (&self->p_clock);
+      _pthread_invoke_cancel ();
+    }
+  pthread_mutex_unlock (&self->p_clock);
+}
+
+int
+pthread_cancel (pthread_t t)
+{
+  struct _pthread_v *tv = __pth_gpointer_locked (t);
+
+  if (tv == NULL)
+    return ESRCH;
+  CHECK_OBJECT(tv, ESRCH);
+  /*if (tv->ended) return ESRCH;*/
+  pthread_mutex_lock(&tv->p_clock);
+  if (pthread_equal(pthread_self(), t))
+    {
+      if(tv->cancelled)
+	{
+	  pthread_mutex_unlock(&tv->p_clock);
+	  return (tv->in_cancel ? ESRCH : 0);
+	}
+      tv->cancelled = 1;
+      InterlockedIncrement(&_pthread_cancelling);
+      if(tv->evStart) SetEvent(tv->evStart);
+      if ((tv->p_state & PTHREAD_CANCEL_ASYNCHRONOUS) != 0 && (tv->p_state & PTHREAD_CANCEL_ENABLE) != 0)
+	{
+	  tv->p_state &= ~PTHREAD_CANCEL_ENABLE;
+	  tv->in_cancel = 1;
+	  pthread_mutex_unlock(&tv->p_clock);
+	  _pthread_invoke_cancel();
+	}
+      else
+	pthread_mutex_unlock(&tv->p_clock);
+      return 0;
+    }
+
+  if ((tv->p_state & PTHREAD_CANCEL_ASYNCHRONOUS) != 0 && (tv->p_state & PTHREAD_CANCEL_ENABLE) != 0)
+    {
+      /* Dangerous asynchronous cancelling */
+      CONTEXT ctxt;
+
+      if(tv->in_cancel)
+	{
+	  pthread_mutex_unlock(&tv->p_clock);
+	  return (tv->in_cancel ? ESRCH : 0);
+	}
+      /* Already done? */
+      if(tv->cancelled || tv->in_cancel)
+	{
+	  /* ??? pthread_mutex_unlock (&tv->p_clock); */
+	  return ESRCH;
+	}
+
+      ctxt.ContextFlags = CONTEXT_CONTROL;
+
+      SuspendThread (tv->h);
+      if (WaitForSingleObject (tv->h, 0) == WAIT_TIMEOUT)
+	{
+	  GetThreadContext(tv->h, &ctxt);
+#ifdef _M_X64
+	  ctxt.Rip = (uintptr_t) _pthread_invoke_cancel;
+#elif defined(_M_IX86)
+	  ctxt.Eip = (uintptr_t) _pthread_invoke_cancel;
+#elif defined(_M_ARM) || defined(_M_ARM64)
+	  ctxt.Pc = (uintptr_t) _pthread_invoke_cancel;
+#else
+#error Unsupported architecture
+#endif
+	  SetThreadContext (tv->h, &ctxt);
+
+	  /* Also try deferred Cancelling */
+	  tv->cancelled = 1;
+	  tv->p_state &= ~PTHREAD_CANCEL_ENABLE;
+	  tv->in_cancel = 1;
+
+	  /* Notify everyone to look */
+	  InterlockedIncrement (&_pthread_cancelling);
+	  if (tv->evStart)
+	    SetEvent (tv->evStart);
+	  pthread_mutex_unlock (&tv->p_clock);
+
+	  ResumeThread (tv->h);
+	}
+    }
+  else
+    {
+      if (tv->cancelled == 0)
+	{
+	  /* Safe deferred Cancelling */
+	  tv->cancelled = 1;
+
+	  /* Notify everyone to look */
+	  InterlockedIncrement (&_pthread_cancelling);
+	  if (tv->evStart)
+	    SetEvent (tv->evStart);
+	}
+      else
+	{
+	  pthread_mutex_unlock (&tv->p_clock);
+	  return (tv->in_cancel ? ESRCH : 0);
+	}
+    }
+  pthread_mutex_unlock (&tv->p_clock);
+  return 0;
+}
+
+/* half-stubbed version as we don't really well support signals */
+int
+pthread_kill (pthread_t t, int sig)
+{
+  struct _pthread_v *tv;
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  tv = __pthread_get_pointer (t);
+  if (!tv || t != tv->x || tv->in_cancel || tv->ended || tv->h == NULL
+      || tv->h == INVALID_HANDLE_VALUE)
+  {
+    pthread_mutex_unlock (&mtx_pthr_locked);
+    return ESRCH;
+  }
+  pthread_mutex_unlock (&mtx_pthr_locked);
+  if (!sig)
+    return 0;
+  if (sig < SIGINT || sig > NSIG)
+    return EINVAL;
+  return pthread_cancel(t);
+}
+
+unsigned
+_pthread_get_state (const pthread_attr_t *attr, unsigned flag)
+{
+  return (attr->p_state & flag);
+}
+
+int
+_pthread_set_state (pthread_attr_t *attr, unsigned flag, unsigned val)
+{
+  if (~flag & val)
+    return EINVAL;
+  attr->p_state &= ~flag;
+  attr->p_state |= val;
+
+  return 0;
+}
+
+int
+pthread_attr_init (pthread_attr_t *attr)
+{
+  memset (attr, 0, sizeof (pthread_attr_t));
+  attr->p_state = PTHREAD_DEFAULT_ATTR;
+  attr->stack = NULL;
+  attr->s_size = 0;
+  return 0;
+}
+
+int
+pthread_attr_destroy (pthread_attr_t *attr)
+{
+  /* No need to do anything */
+  memset (attr, 0, sizeof(pthread_attr_t));
+  return 0;
+}
+
+int
+pthread_attr_setdetachstate (pthread_attr_t *a, int flag)
+{
+  return _pthread_set_state(a, PTHREAD_CREATE_DETACHED, flag);
+}
+
+int
+pthread_attr_getdetachstate (const pthread_attr_t *a, int *flag)
+{
+  *flag = _pthread_get_state(a, PTHREAD_CREATE_DETACHED);
+  return 0;
+}
+
+int
+pthread_attr_setinheritsched (pthread_attr_t *a, int flag)
+{
+  if (!a || (flag != PTHREAD_INHERIT_SCHED && flag != PTHREAD_EXPLICIT_SCHED))
+    return EINVAL;
+  return _pthread_set_state(a, PTHREAD_INHERIT_SCHED, flag);
+}
+
+int
+pthread_attr_getinheritsched (const pthread_attr_t *a, int *flag)
+{
+  *flag = _pthread_get_state(a, PTHREAD_INHERIT_SCHED);
+  return 0;
+}
+
+int
+pthread_attr_setscope (pthread_attr_t *a, int flag)
+{
+  return _pthread_set_state(a, PTHREAD_SCOPE_SYSTEM, flag);
+}
+
+int
+pthread_attr_getscope (const pthread_attr_t *a, int *flag)
+{
+  *flag = _pthread_get_state(a, PTHREAD_SCOPE_SYSTEM);
+  return 0;
+}
+
+int
+pthread_attr_getstack (const pthread_attr_t *attr, void **stack, size_t *size)
+{
+  *stack = (char *) attr->stack - attr->s_size;
+  *size = attr->s_size;
+  return 0;
+}
+
+int
+pthread_attr_setstack (pthread_attr_t *attr, void *stack, size_t size)
+{
+  attr->s_size = size;
+  attr->stack = (char *) stack + size;
+  return 0;
+}
+
+int
+pthread_attr_getstackaddr (const pthread_attr_t *attr, void **stack)
+{
+  *stack = attr->stack;
+  return 0;
+}
+
+int
+pthread_attr_setstackaddr (pthread_attr_t *attr, void *stack)
+{
+  attr->stack = stack;
+  return 0;
+}
+
+int
+pthread_attr_getstacksize (const pthread_attr_t *attr, size_t *size)
+{
+  *size = attr->s_size;
+  return 0;
+}
+
+int
+pthread_attr_setstacksize (pthread_attr_t *attr, size_t size)
+{
+  attr->s_size = size;
+  return 0;
+}
+
+static void
+test_cancel_locked (pthread_t t)
+{
+  struct _pthread_v *tv = __pth_gpointer_locked (t);
+
+  if (!tv || tv->in_cancel || tv->ended != 0 || (tv->p_state & PTHREAD_CANCEL_ENABLE) == 0)
+    return;
+  if ((tv->p_state & PTHREAD_CANCEL_ASYNCHRONOUS) == 0)
+    return;
+  if (WaitForSingleObject(tv->evStart, 0) != WAIT_OBJECT_0)
+    return;
+  pthread_mutex_unlock (&tv->p_clock);
+  _pthread_invoke_cancel();
+}
+
+int
+pthread_setcancelstate (int state, int *oldstate)
+{
+  _pthread_v *t = __pthread_self_lite ();
+
+  if (!t || (state & PTHREAD_CANCEL_ENABLE) != state)
+    return EINVAL;
+
+  pthread_mutex_lock (&t->p_clock);
+  if (oldstate)
+    *oldstate = t->p_state & PTHREAD_CANCEL_ENABLE;
+  t->p_state &= ~PTHREAD_CANCEL_ENABLE;
+  t->p_state |= state;
+  test_cancel_locked (t->x);
+  pthread_mutex_unlock (&t->p_clock);
+
+  return 0;
+}
+
+int
+pthread_setcanceltype (int type, int *oldtype)
+{
+  _pthread_v *t = __pthread_self_lite ();
+
+  if (!t || (type & PTHREAD_CANCEL_ASYNCHRONOUS) != type)
+    return EINVAL;
+
+  pthread_mutex_lock (&t->p_clock);
+  if (oldtype)
+    *oldtype = t->p_state & PTHREAD_CANCEL_ASYNCHRONOUS;
+  t->p_state &= ~PTHREAD_CANCEL_ASYNCHRONOUS;
+  t->p_state |= type;
+  test_cancel_locked (t->x);
+  pthread_mutex_unlock (&t->p_clock);
+
+  return 0;
+}
+
+void _fpreset (void);
+
+#if defined(__i386__)
+/* Align ESP on 16-byte boundaries. */
+__attribute__((force_align_arg_pointer))
+#endif
+unsigned __stdcall
+pthread_create_wrapper (void *args)
+{
+  unsigned rslt = 0;
+  struct _pthread_v *tv = (struct _pthread_v *)args;
+
+  _fpreset();
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  pthread_mutex_lock (&tv->p_clock);
+  _pthread_once_raw(&_pthread_tls_once, pthread_tls_init);
+  TlsSetValue(_pthread_tls, tv);
+  tv->tid = GetCurrentThreadId();
+  pthread_mutex_unlock (&tv->p_clock);
+
+
+  if (!setjmp(tv->jb))
+    {
+      intptr_t trslt = (intptr_t) 128;
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      /* Call function and save return value */
+      if (tv->func)
+        trslt = (intptr_t) tv->func(tv->ret_arg);
+      pthread_mutex_lock (&mtx_pthr_locked);
+      tv->ret_arg = (void*) trslt;
+      /* Clean up destructors */
+      _pthread_cleanup_dest(tv->x);
+    }
+  else
+    pthread_mutex_lock (&mtx_pthr_locked);
+
+  pthread_mutex_lock (&tv->p_clock);
+  rslt = (unsigned) (size_t) tv->ret_arg;
+  /* Make sure we free ourselves if we are detached */
+  if (tv->evStart)
+    CloseHandle (tv->evStart);
+  tv->evStart = NULL;
+  if (!tv->h)
+    {
+      tv->valid = DEAD_THREAD;
+      pthread_mutex_unlock (&tv->p_clock);
+      pthread_mutex_destroy (&tv->p_clock);
+      push_pthread_mem (tv);
+      tv = NULL;
+      TlsSetValue (_pthread_tls, tv);
+    }
+  else
+    {
+      pthread_mutex_unlock (&tv->p_clock);
+      pthread_mutex_destroy (&tv->p_clock);
+      /* Reinitialise p_clock, since there may be attempts at
+         destroying it again in __dyn_tls_thread later on. */
+      tv->p_clock = PTHREAD_MUTEX_INITIALIZER;
+      tv->ended = 1;
+    }
+  while (pthread_mutex_unlock (&mtx_pthr_locked) == 0)
+   Sleep (0);
+  _endthreadex (rslt);
+  return rslt;
+}
+
+int
+pthread_create (pthread_t *th, const pthread_attr_t *attr, void *(* func)(void *), void *arg)
+{
+  HANDLE thrd = NULL;
+  int redo = 0;
+  struct _pthread_v *tv;
+  unsigned int ssize = 0;
+  pthread_spinlock_t new_spin_keys = PTHREAD_SPINLOCK_INITIALIZER;
+  unsigned thrAddr; /* Dummy variable to pass a valid location to _beginthreadex (Win98). */
+
+  if (attr && attr->s_size > UINT_MAX)
+    return EINVAL;
+
+  if ((tv = pop_pthread_mem ()) == NULL)
+    return EAGAIN;
+
+  if (th)
+    *th = tv->x;
+
+  /* Save data in pthread_t */
+  tv->ended = 0;
+  tv->ret_arg = arg;
+  tv->func = func;
+  tv->p_state = PTHREAD_DEFAULT_ATTR;
+  tv->h = INVALID_HANDLE_VALUE;
+  /* We retry it here a few times, as events are a limited resource ... */
+  do
+    {
+      tv->evStart = CreateEvent (NULL, 1, 0, NULL);
+      if (tv->evStart != NULL)
+	break;
+      Sleep ((!redo ? 0 : 20));
+    }
+  while (++redo <= 4);
+
+  tv->p_clock = PTHREAD_MUTEX_INITIALIZER;
+  replace_spin_keys (&tv->spin_keys, new_spin_keys);
+  tv->valid = LIFE_THREAD;
+  tv->sched.sched_priority = THREAD_PRIORITY_NORMAL;
+  tv->sched_pol = SCHED_OTHER;
+  if (tv->evStart == NULL)
+    {
+      if (th)
+       memset (th, 0, sizeof (pthread_t));
+      push_pthread_mem (tv);
+      return EAGAIN;
+    }
+
+  if (attr)
+    {
+      int inh = 0;
+      tv->p_state = attr->p_state;
+      ssize = (unsigned int)attr->s_size;
+      pthread_attr_getinheritsched (attr, &inh);
+      if (inh)
+	{
+	  tv->sched.sched_priority = __pthread_self_lite ()->sched.sched_priority;
+	}
+      else
+	tv->sched.sched_priority = attr->param.sched_priority;
+    }
+
+  /* Make sure tv->h has value of INVALID_HANDLE_VALUE */
+  _ReadWriteBarrier();
+
+  thrd = (HANDLE) _beginthreadex(NULL, ssize, pthread_create_wrapper, tv, 0x4/*CREATE_SUSPEND*/, &thrAddr);
+  if (thrd == INVALID_HANDLE_VALUE)
+    thrd = 0;
+  /* Failed */
+  if (!thrd)
+    {
+      if (tv->evStart)
+	CloseHandle (tv->evStart);
+      pthread_mutex_destroy (&tv->p_clock);
+      replace_spin_keys (&tv->spin_keys, new_spin_keys);
+      tv->evStart = NULL;
+      tv->h = 0;
+      if (th)
+        memset (th, 0, sizeof (pthread_t));
+      push_pthread_mem (tv);
+      return EAGAIN;
+    }
+  {
+    int pr = tv->sched.sched_priority;
+    if (pr <= THREAD_PRIORITY_IDLE) {
+	pr = THREAD_PRIORITY_IDLE;
+    } else if (pr <= THREAD_PRIORITY_LOWEST) {
+	pr = THREAD_PRIORITY_LOWEST;
+    } else if (pr >= THREAD_PRIORITY_TIME_CRITICAL) {
+	pr = THREAD_PRIORITY_TIME_CRITICAL;
+    } else if (pr >= THREAD_PRIORITY_HIGHEST) {
+	pr = THREAD_PRIORITY_HIGHEST;
+    }
+    SetThreadPriority (thrd, pr);
+  }
+  ResetEvent (tv->evStart);
+  if ((tv->p_state & PTHREAD_CREATE_DETACHED) != 0)
+    {
+      tv->h = 0;
+      ResumeThread (thrd);
+      CloseHandle (thrd);
+    }
+  else
+    {
+      tv->h = thrd;
+      ResumeThread (thrd);
+    }
+  Sleep (0);
+  return 0;
+}
+
+int
+pthread_join (pthread_t t, void **res)
+{
+  struct _pthread_v *tv = __pth_gpointer_locked (t);
+  pthread_spinlock_t new_spin_keys = PTHREAD_SPINLOCK_INITIALIZER;
+
+  CHECK_OBJECT(tv, ESRCH);
+
+  if ((tv->p_state & PTHREAD_CREATE_DETACHED) != 0)
+    return EINVAL;
+  if (pthread_equal(pthread_self(), t))
+    return EDEADLK;
+
+  /* pthread_testcancel (); */
+  if (tv->ended == 0 || (tv->h != NULL && tv->h != INVALID_HANDLE_VALUE))
+    WaitForSingleObject (tv->h, INFINITE);
+  CloseHandle (tv->h);
+  if (tv->evStart)
+    CloseHandle (tv->evStart);
+  tv->evStart = NULL;
+  /* Obtain return value */
+  if (res)
+    *res = tv->ret_arg;
+  pthread_mutex_destroy (&tv->p_clock);
+  replace_spin_keys (&tv->spin_keys, new_spin_keys);
+  push_pthread_mem (tv);
+
+  return 0;
+}
+
+int
+_pthread_tryjoin (pthread_t t, void **res)
+{
+  struct _pthread_v *tv;
+  pthread_spinlock_t new_spin_keys = PTHREAD_SPINLOCK_INITIALIZER;
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  tv = __pthread_get_pointer (t);
+
+  if (!tv || !TEST_HANDLE(tv->h))
+    {
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      return ESRCH;
+    }
+
+  if ((tv->p_state & PTHREAD_CREATE_DETACHED) != 0)
+    {
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      return EINVAL;
+    }
+  if (pthread_equal(pthread_self(), t))
+    {
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      return EDEADLK;
+    }
+  if(tv->ended == 0 && WaitForSingleObject(tv->h, 0))
+    {
+      if (tv->ended == 0)
+        {
+	      pthread_mutex_unlock (&mtx_pthr_locked);
+	      /* pthread_testcancel (); */
+	      return EBUSY;
+	    }
+    }
+  CloseHandle (tv->h);
+  if (tv->evStart)
+    CloseHandle (tv->evStart);
+  tv->evStart = NULL;
+
+  /* Obtain return value */
+  if (res)
+    *res = tv->ret_arg;
+  pthread_mutex_destroy (&tv->p_clock);
+  replace_spin_keys (&tv->spin_keys, new_spin_keys);
+
+  push_pthread_mem (tv);
+
+  pthread_mutex_unlock (&mtx_pthr_locked);
+  /* pthread_testcancel (); */
+  return 0;
+}
+
+int
+pthread_detach (pthread_t t)
+{
+  int r = 0;
+  struct _pthread_v *tv = __pth_gpointer_locked (t);
+  HANDLE dw;
+  pthread_spinlock_t new_spin_keys = PTHREAD_SPINLOCK_INITIALIZER;
+
+  pthread_mutex_lock (&mtx_pthr_locked);
+  if (!tv || !TEST_HANDLE(tv->h))
+    {
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      return ESRCH;
+    }
+  if ((tv->p_state & PTHREAD_CREATE_DETACHED) != 0)
+    {
+      pthread_mutex_unlock (&mtx_pthr_locked);
+      return EINVAL;
+    }
+  /* if (tv->ended) r = ESRCH; */
+  dw = tv->h;
+  tv->h = 0;
+  tv->p_state |= PTHREAD_CREATE_DETACHED;
+  _ReadWriteBarrier();
+  if (dw)
+    {
+      CloseHandle (dw);
+      if (tv->ended)
+	{
+	  if (tv->evStart)
+	    CloseHandle (tv->evStart);
+	  tv->evStart = NULL;
+	  pthread_mutex_destroy (&tv->p_clock);
+	  replace_spin_keys (&tv->spin_keys, new_spin_keys);
+	  push_pthread_mem (tv);
+	}
+    }
+  pthread_mutex_unlock (&mtx_pthr_locked);
+
+  return r;
+}
+
+static int dummy_concurrency_level = 0;
+
+int
+pthread_getconcurrency (void)
+{
+  return dummy_concurrency_level;
+}
+
+int
+pthread_setconcurrency (int new_level)
+{
+  dummy_concurrency_level = new_level;
+  return 0;
+}
+
+int
+pthread_setname_np (pthread_t thread, const char *name)
+{
+  struct _pthread_v *tv;
+  char *stored_name;
+
+  if (name == NULL)
+    return EINVAL;
+
+  tv = __pth_gpointer_locked (thread);
+  if (!tv || thread != tv->x || tv->in_cancel || tv->ended || tv->h == NULL
+      || tv->h == INVALID_HANDLE_VALUE)
+    return ESRCH;
+
+  stored_name = strdup (name);
+  if (stored_name == NULL)
+    return ENOMEM;
+
+  if (tv->thread_name != NULL)
+    free (tv->thread_name);
+
+  tv->thread_name = stored_name;
+  SetThreadName (tv->tid, name);
+
+  if (_pthread_set_thread_description != NULL)
+    {
+      mbstate_t mbs = {0};
+      size_t required_size = mbsrtowcs(NULL, &name, 0, &mbs);
+      if (required_size != (size_t)-1)
+        {
+          wchar_t *wname = malloc((required_size + 1) * sizeof(wchar_t));
+          if (wname != NULL)
+            {
+              mbsrtowcs(wname, &name, required_size + 1, &mbs);
+              _pthread_set_thread_description(tv->h, wname);
+              free(wname);
+            }
+        }
+    }
+  return 0;
+}
+
+int
+pthread_getname_np (pthread_t thread, char *name, size_t len)
+{
+  struct _pthread_v *tv;
+  size_t thread_name_len;
+
+  if (name == NULL)
+    return EINVAL;
+
+  tv = __pth_gpointer_locked (thread);
+  if (!tv || thread != tv->x || tv->in_cancel || tv->ended || tv->h == NULL
+      || tv->h == INVALID_HANDLE_VALUE)
+    return ESRCH;
+
+  if (len < 1)
+    return ERANGE;
+
+  if (tv->thread_name == NULL)
+    {
+      name[0] = '\0';
+      return 0;
+    }
+
+  thread_name_len = strlen (tv->thread_name);
+  if (thread_name_len >= len)
+    return ERANGE;
+
+  memcpy (name, tv->thread_name, thread_name_len + 1);
+  return 0;
+}
diff --git a/vendor/winpthreads/src/thread.h b/vendor/winpthreads/src/thread.h
new file mode 100644
index 00000000..90f9cd39
--- /dev/null
+++ b/vendor/winpthreads/src/thread.h
@@ -0,0 +1,76 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef WIN_PTHREAD_H
+#define WIN_PTHREAD_H
+
+#include <setjmp.h>
+/* internal header files */
+#include "rwlock.h"
+
+#define LIFE_THREAD 0xBAB1F00D
+#define DEAD_THREAD 0xDEADBEEF
+#define EXCEPTION_SET_THREAD_NAME ((DWORD) 0x406D1388)
+
+typedef struct _pthread_v _pthread_v;
+struct _pthread_v
+{
+    unsigned int valid;
+    void *ret_arg;
+    void *(* func)(void *);
+    _pthread_cleanup *clean;
+    int nobreak;
+    HANDLE h;
+    HANDLE evStart;
+    pthread_mutex_t p_clock;
+    int cancelled : 2;
+    int in_cancel : 2;
+    int thread_noposix : 2;
+    unsigned int p_state;
+    unsigned int keymax;
+    void **keyval;
+    unsigned char *keyval_set;
+    char *thread_name;
+    pthread_spinlock_t spin_keys;
+    DWORD tid;
+    int rwlc;
+    pthread_rwlock_t rwlq[RWLS_PER_THREAD];
+    int sched_pol;
+    int ended;
+    struct sched_param sched;
+    jmp_buf jb;
+    struct _pthread_v *next;
+    pthread_t x; /* Internal posix handle.  */
+};
+
+typedef struct __pthread_idlist {
+  struct _pthread_v *ptr;
+  pthread_t id;
+} __pthread_idlist;
+
+int _pthread_tryjoin(pthread_t t, void **res);
+void _pthread_setnobreak(int);
+int  __pthread_shallcancel(void);
+WINPTHREAD_API struct _pthread_v * __pth_gpointer_locked (pthread_t id);
+int _pthread_delay_np_ms (DWORD to);
+
+#endif
diff --git a/vendor/winpthreads/src/version.rc b/vendor/winpthreads/src/version.rc
new file mode 100644
index 00000000..19c2d1e7
--- /dev/null
+++ b/vendor/winpthreads/src/version.rc
@@ -0,0 +1,67 @@
+/*
+   Copyright (c) 2011 mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#include <winver.h>
+#include "wpth_ver.h"
+
+#if defined(__MINGW64__)
+# define WPTH_VERSIONINFO_COMMENT "GNU C build -- MinGW-w64 64-bit"
+#elif defined(__MINGW32__)
+# define WPTH_VERSIONINFO_COMMENT "GNU C build -- MinGW-w64 32-bit"
+#else
+# define WPTH_VERSIONINFO_COMMENT "MSVC build"
+#endif
+
+VS_VERSION_INFO VERSIONINFO
+  FILEVERSION    WPTH_VERSION
+  PRODUCTVERSION WPTH_VERSION
+  FILEFLAGSMASK  VS_FFI_FILEFLAGSMASK
+#ifdef _DEBUG
+  FILEFLAGS      VS_FF_DEBUG
+#else
+  FILEFLAGS      0
+#endif
+  FILEOS         VOS_NT_WINDOWS32
+  FILETYPE       VFT_DLL
+BEGIN
+    BLOCK "StringFileInfo"
+    BEGIN
+        BLOCK "040904b0"
+        BEGIN
+            VALUE "FileDescription", "POSIX WinThreads for Windows"
+            VALUE "ProductVersion", WPTH_VERSION_STRING
+            VALUE "FileVersion", WPTH_VERSION_STRING
+            VALUE "InternalName", "libwinpthread-" WPTH_VERSION_MAJOR_STRING ".dll"
+            VALUE "OriginalFilename", "libwinpthread-" WPTH_VERSION_MAJOR_STRING ".dll"
+            VALUE "CompanyName", "MinGW-W64 Project. All rights reserved."
+            VALUE "LegalCopyright", "Copyright (C) MinGW-W64 Project Members 2010-2023"
+            VALUE "Licence", "MIT AND BSD-3-Clause"
+            VALUE "Info", "https://www.mingw-w64.org/"
+	    VALUE "Comment", WPTH_VERSIONINFO_COMMENT
+        END
+    END
+    BLOCK "VarFileInfo"
+    BEGIN
+        VALUE "Translation", 0x409, 1200
+    END
+END
+
diff --git a/vendor/winpthreads/src/wpth_ver.h b/vendor/winpthreads/src/wpth_ver.h
new file mode 100644
index 00000000..ea36c1e1
--- /dev/null
+++ b/vendor/winpthreads/src/wpth_ver.h
@@ -0,0 +1,30 @@
+/*
+   Copyright (c) 2011-2016  mingw-w64 project
+
+   Permission is hereby granted, free of charge, to any person obtaining a
+   copy of this software and associated documentation files (the "Software"),
+   to deal in the Software without restriction, including without limitation
+   the rights to use, copy, modify, merge, publish, distribute, sublicense,
+   and/or sell copies of the Software, and to permit persons to whom the
+   Software is furnished to do so, subject to the following conditions:
+
+   The above copyright notice and this permission notice shall be included in
+   all copies or substantial portions of the Software.
+
+   THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+   IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+   FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+   AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+   LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+   FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
+   DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef __WPTHREADS_VERSION__
+#define __WPTHREADS_VERSION__
+
+#define WPTH_VERSION 1,0,0,0
+#define WPTH_VERSION_STRING "1, 0, 0, 0"
+#define WPTH_VERSION_MAJOR_STRING "1"
+
+#endif