diff options
Diffstat (limited to 'lib/std')
-rw-r--r-- | lib/std/private/syslocks.nim | 234 | ||||
-rw-r--r-- | lib/std/private/threadtypes.nim | 176 | ||||
-rw-r--r-- | lib/std/threads.nim | 276 |
3 files changed, 686 insertions, 0 deletions
diff --git a/lib/std/private/syslocks.nim b/lib/std/private/syslocks.nim new file mode 100644 index 000000000..dff339e9e --- /dev/null +++ b/lib/std/private/syslocks.nim @@ -0,0 +1,234 @@ +# +# +# Nim's Runtime Library +# (c) Copyright 2012 Andreas Rumpf +# +# See the file "copying.txt", included in this +# distribution, for details about the copyright. +# + +# Low level system locks and condition vars. + +{.push stackTrace: off.} + +when defined(windows): + type + Handle = int + + SysLock* {.importc: "CRITICAL_SECTION", + header: "<windows.h>", final, pure.} = object # CRITICAL_SECTION in WinApi + DebugInfo: pointer + LockCount: int32 + RecursionCount: int32 + OwningThread: int + LockSemaphore: int + SpinCount: int + + SysCond* {.importc: "RTL_CONDITION_VARIABLE", header: "<windows.h>".} = object + thePtr {.importc: "ptr".} : Handle + + proc initSysLock*(L: var SysLock) {.importc: "InitializeCriticalSection", + header: "<windows.h>".} + ## Initializes the lock `L`. + + proc tryAcquireSysAux(L: var SysLock): int32 {.importc: "TryEnterCriticalSection", + header: "<windows.h>".} + ## Tries to acquire the lock `L`. + + proc tryAcquireSys*(L: var SysLock): bool {.inline.} = + result = tryAcquireSysAux(L) != 0'i32 + + proc acquireSys*(L: var SysLock) {.importc: "EnterCriticalSection", + header: "<windows.h>".} + ## Acquires the lock `L`. + + proc releaseSys*(L: var SysLock) {.importc: "LeaveCriticalSection", + header: "<windows.h>".} + ## Releases the lock `L`. + + proc deinitSys*(L: var SysLock) {.importc: "DeleteCriticalSection", + header: "<windows.h>".} + + proc initializeConditionVariable( + conditionVariable: var SysCond + ) {.stdcall, noSideEffect, dynlib: "kernel32", importc: "InitializeConditionVariable".} + + proc sleepConditionVariableCS( + conditionVariable: var SysCond, + PCRITICAL_SECTION: var SysLock, + dwMilliseconds: int + ): int32 {.stdcall, noSideEffect, dynlib: "kernel32", importc: "SleepConditionVariableCS".} + + + proc signalSysCond*(hEvent: var SysCond) {.stdcall, noSideEffect, + dynlib: "kernel32", importc: "WakeConditionVariable".} + + proc broadcastSysCond*(hEvent: var SysCond) {.stdcall, noSideEffect, + dynlib: "kernel32", importc: "WakeAllConditionVariable".} + + proc initSysCond*(cond: var SysCond) {.inline.} = + initializeConditionVariable(cond) + proc deinitSysCond*(cond: var SysCond) {.inline.} = + discard + proc waitSysCond*(cond: var SysCond, lock: var SysLock) = + discard sleepConditionVariableCS(cond, lock, -1'i32) + +elif defined(genode): + const + Header = "genode_cpp/syslocks.h" + type + SysLock* {.importcpp: "Nim::SysLock", pure, final, + header: Header.} = object + SysCond* {.importcpp: "Nim::SysCond", pure, final, + header: Header.} = object + + proc initSysLock*(L: var SysLock) = discard + proc deinitSys*(L: var SysLock) = discard + proc acquireSys*(L: var SysLock) {.noSideEffect, importcpp.} + proc tryAcquireSys*(L: var SysLock): bool {.noSideEffect, importcpp.} + proc releaseSys*(L: var SysLock) {.noSideEffect, importcpp.} + + proc initSysCond*(L: var SysCond) = discard + proc deinitSysCond*(L: var SysCond) = discard + proc waitSysCond*(cond: var SysCond, lock: var SysLock) {. + noSideEffect, importcpp.} + proc signalSysCond*(cond: var SysCond) {. + noSideEffect, importcpp.} + proc broadcastSysCond*(cond: var SysCond) {. + noSideEffect, importcpp.} + +else: + type + SysLockObj {.importc: "pthread_mutex_t", pure, final, + header: """#include <sys/types.h> + #include <pthread.h>""".} = object + when defined(linux) and defined(amd64): + abi: array[40 div sizeof(clong), clong] + + SysLockAttr* {.importc: "pthread_mutexattr_t", pure, final + header: """#include <sys/types.h> + #include <pthread.h>""".} = object + when defined(linux) and defined(amd64): + abi: array[4 div sizeof(cint), cint] # actually a cint + + SysCondObj {.importc: "pthread_cond_t", pure, final, + header: """#include <sys/types.h> + #include <pthread.h>""".} = object + when defined(linux) and defined(amd64): + abi: array[48 div sizeof(clonglong), clonglong] + + SysCondAttr {.importc: "pthread_condattr_t", pure, final + header: """#include <sys/types.h> + #include <pthread.h>""".} = object + when defined(linux) and defined(amd64): + abi: array[4 div sizeof(cint), cint] # actually a cint + + SysLockType = distinct cint + + proc initSysLockAux(L: var SysLockObj, attr: ptr SysLockAttr) {. + importc: "pthread_mutex_init", header: "<pthread.h>", noSideEffect.} + proc deinitSysAux(L: var SysLockObj) {.noSideEffect, + importc: "pthread_mutex_destroy", header: "<pthread.h>".} + + proc acquireSysAux(L: var SysLockObj) {.noSideEffect, + importc: "pthread_mutex_lock", header: "<pthread.h>".} + proc tryAcquireSysAux(L: var SysLockObj): cint {.noSideEffect, + importc: "pthread_mutex_trylock", header: "<pthread.h>".} + + proc releaseSysAux(L: var SysLockObj) {.noSideEffect, + importc: "pthread_mutex_unlock", header: "<pthread.h>".} + + when defined(ios): + # iOS will behave badly if sync primitives are moved in memory. In order + # to prevent this once and for all, we're doing an extra malloc when + # initializing the primitive. + type + SysLock* = ptr SysLockObj + SysCond* = ptr SysCondObj + + when not declared(c_malloc): + proc c_malloc(size: csize_t): pointer {. + importc: "malloc", header: "<stdlib.h>".} + proc c_free(p: pointer) {. + importc: "free", header: "<stdlib.h>".} + + proc initSysLock*(L: var SysLock, attr: ptr SysLockAttr = nil) = + L = cast[SysLock](c_malloc(csize_t(sizeof(SysLockObj)))) + initSysLockAux(L[], attr) + + proc deinitSys*(L: var SysLock) = + deinitSysAux(L[]) + c_free(L) + + template acquireSys*(L: var SysLock) = + acquireSysAux(L[]) + template tryAcquireSys*(L: var SysLock): bool = + tryAcquireSysAux(L[]) == 0'i32 + template releaseSys*(L: var SysLock) = + releaseSysAux(L[]) + else: + type + SysLock* = SysLockObj + SysCond* = SysCondObj + + template initSysLock*(L: var SysLock, attr: ptr SysLockAttr = nil) = + initSysLockAux(L, attr) + template deinitSys*(L: var SysLock) = + deinitSysAux(L) + template acquireSys*(L: var SysLock) = + acquireSysAux(L) + template tryAcquireSys*(L: var SysLock): bool = + tryAcquireSysAux(L) == 0'i32 + template releaseSys*(L: var SysLock) = + releaseSysAux(L) + + # rlocks + var SysLockType_Reentrant* {.importc: "PTHREAD_MUTEX_RECURSIVE", + header: "<pthread.h>".}: SysLockType + proc initSysLockAttr*(a: var SysLockAttr) {. + importc: "pthread_mutexattr_init", header: "<pthread.h>", noSideEffect.} + proc setSysLockType*(a: var SysLockAttr, t: SysLockType) {. + importc: "pthread_mutexattr_settype", header: "<pthread.h>", noSideEffect.} + + # locks + proc initSysCondAux(cond: var SysCondObj, cond_attr: ptr SysCondAttr = nil) {. + importc: "pthread_cond_init", header: "<pthread.h>", noSideEffect.} + proc deinitSysCondAux(cond: var SysCondObj) {.noSideEffect, + importc: "pthread_cond_destroy", header: "<pthread.h>".} + + proc waitSysCondAux(cond: var SysCondObj, lock: var SysLockObj): cint {. + importc: "pthread_cond_wait", header: "<pthread.h>", noSideEffect.} + proc signalSysCondAux(cond: var SysCondObj) {. + importc: "pthread_cond_signal", header: "<pthread.h>", noSideEffect.} + proc broadcastSysCondAux(cond: var SysCondObj) {. + importc: "pthread_cond_broadcast", header: "<pthread.h>", noSideEffect.} + + when defined(ios): + proc initSysCond*(cond: var SysCond, cond_attr: ptr SysCondAttr = nil) = + cond = cast[SysCond](c_malloc(csize_t(sizeof(SysCondObj)))) + initSysCondAux(cond[], cond_attr) + + proc deinitSysCond*(cond: var SysCond) = + deinitSysCondAux(cond[]) + c_free(cond) + + template waitSysCond*(cond: var SysCond, lock: var SysLock) = + discard waitSysCondAux(cond[], lock[]) + template signalSysCond*(cond: var SysCond) = + signalSysCondAux(cond[]) + template broadcastSysCond*(cond: var SysCond) = + broadcastSysCondAux(cond[]) + else: + template initSysCond*(cond: var SysCond, cond_attr: ptr SysCondAttr = nil) = + initSysCondAux(cond, cond_attr) + template deinitSysCond*(cond: var SysCond) = + deinitSysCondAux(cond) + + template waitSysCond*(cond: var SysCond, lock: var SysLock) = + discard waitSysCondAux(cond, lock) + template signalSysCond*(cond: var SysCond) = + signalSysCondAux(cond) + template broadcastSysCond*(cond: var SysCond) = + broadcastSysCondAux(cond) + +{.pop.} diff --git a/lib/std/private/threadtypes.nim b/lib/std/private/threadtypes.nim new file mode 100644 index 000000000..a1cdf21dc --- /dev/null +++ b/lib/std/private/threadtypes.nim @@ -0,0 +1,176 @@ +include system/inclrtl + +const hasSharedHeap* = defined(boehmgc) or defined(gogc) # don't share heaps; every thread has its own + +when defined(windows): + type + Handle* = int + SysThread* = Handle + WinThreadProc* = proc (x: pointer): int32 {.stdcall.} + + proc createThread*(lpThreadAttributes: pointer, dwStackSize: int32, + lpStartAddress: WinThreadProc, + lpParameter: pointer, + dwCreationFlags: int32, + lpThreadId: var int32): SysThread {. + stdcall, dynlib: "kernel32", importc: "CreateThread".} + + proc winSuspendThread*(hThread: SysThread): int32 {. + stdcall, dynlib: "kernel32", importc: "SuspendThread".} + + proc winResumeThread*(hThread: SysThread): int32 {. + stdcall, dynlib: "kernel32", importc: "ResumeThread".} + + proc waitForSingleObject*(hHandle: SysThread, dwMilliseconds: int32): int32 {. + stdcall, dynlib: "kernel32", importc: "WaitForSingleObject".} + + proc waitForMultipleObjects*(nCount: int32, + lpHandles: ptr SysThread, + bWaitAll: int32, + dwMilliseconds: int32): int32 {. + stdcall, dynlib: "kernel32", importc: "WaitForMultipleObjects".} + + proc terminateThread*(hThread: SysThread, dwExitCode: int32): int32 {. + stdcall, dynlib: "kernel32", importc: "TerminateThread".} + + proc setThreadAffinityMask*(hThread: SysThread, dwThreadAffinityMask: uint) {. + importc: "SetThreadAffinityMask", stdcall, header: "<windows.h>".} + +elif defined(genode): + const + GenodeHeader* = "genode_cpp/threads.h" + type + SysThread* {.importcpp: "Nim::SysThread", + header: GenodeHeader, final, pure.} = object + GenodeThreadProc* = proc (x: pointer) {.noconv.} + + proc initThread*(s: var SysThread, + env: GenodeEnv, + stackSize: culonglong, + entry: GenodeThreadProc, + arg: pointer, + affinity: cuint) {. + importcpp: "#.initThread(@)".} + + +else: + when not (defined(macosx) or defined(haiku)): + {.passl: "-pthread".} + + when not defined(haiku): + {.passc: "-pthread".} + + const + schedh = "#define _GNU_SOURCE\n#include <sched.h>" + pthreadh* = "#define _GNU_SOURCE\n#include <pthread.h>" + + when not declared(Time): + when defined(linux): + type Time = clong + else: + type Time = int + + when (defined(linux) or defined(nintendoswitch)) and defined(amd64): + type + SysThread* {.importc: "pthread_t", + header: "<sys/types.h>" .} = distinct culong + Pthread_attr* {.importc: "pthread_attr_t", + header: "<sys/types.h>".} = object + abi: array[56 div sizeof(clong), clong] + elif defined(openbsd) and defined(amd64): + type + SysThread* {.importc: "pthread_t", header: "<pthread.h>".} = object + Pthread_attr* {.importc: "pthread_attr_t", + header: "<pthread.h>".} = object + else: + type + SysThread* {.importc: "pthread_t", header: "<sys/types.h>".} = int + Pthread_attr* {.importc: "pthread_attr_t", + header: "<sys/types.h>".} = object + type + Timespec* {.importc: "struct timespec", header: "<time.h>".} = object + tv_sec*: Time + tv_nsec*: clong + + proc pthread_attr_init*(a1: var Pthread_attr): cint {. + importc, header: pthreadh.} + proc pthread_attr_setstack*(a1: ptr Pthread_attr, a2: pointer, a3: int): cint {. + importc, header: pthreadh.} + proc pthread_attr_setstacksize*(a1: var Pthread_attr, a2: int): cint {. + importc, header: pthreadh.} + proc pthread_attr_destroy*(a1: var Pthread_attr): cint {. + importc, header: pthreadh.} + + proc pthread_create*(a1: var SysThread, a2: var Pthread_attr, + a3: proc (x: pointer): pointer {.noconv.}, + a4: pointer): cint {.importc: "pthread_create", + header: pthreadh.} + proc pthread_join*(a1: SysThread, a2: ptr pointer): cint {. + importc, header: pthreadh.} + + proc pthread_cancel*(a1: SysThread): cint {. + importc: "pthread_cancel", header: pthreadh.} + + type CpuSet* {.importc: "cpu_set_t", header: schedh.} = object + when defined(linux) and defined(amd64): + abi: array[1024 div (8 * sizeof(culong)), culong] + + proc cpusetZero*(s: var CpuSet) {.importc: "CPU_ZERO", header: schedh.} + proc cpusetIncl*(cpu: cint; s: var CpuSet) {. + importc: "CPU_SET", header: schedh.} + + when defined(android): + # libc of android doesn't implement pthread_setaffinity_np, + # it exposes pthread_gettid_np though, so we can use that in combination + # with sched_setaffinity to set the thread affinity. + type Pid* {.importc: "pid_t", header: "<sys/types.h>".} = int32 # From posix_other.nim + + proc setAffinityTID*(tid: Pid; setsize: csize_t; s: var CpuSet) {. + importc: "sched_setaffinity", header: schedh.} + + proc pthread_gettid_np*(thread: SysThread): Pid {. + importc: "pthread_gettid_np", header: pthreadh.} + + proc setAffinity*(thread: SysThread; setsize: csize_t; s: var CpuSet) = + setAffinityTID(pthread_gettid_np(thread), setsize, s) + else: + proc setAffinity*(thread: SysThread; setsize: csize_t; s: var CpuSet) {. + importc: "pthread_setaffinity_np", header: pthreadh.} + + +const + emulatedThreadVars* = compileOption("tlsEmulation") +# we preallocate a fixed size for thread local storage, so that no heap +# allocations are needed. Currently less than 16K are used on a 64bit machine. +# We use `float` for proper alignment: +const nimTlsSize {.intdefine.} = 16000 +type + ThreadLocalStorage* = array[0..(nimTlsSize div sizeof(float)), float] + PGcThread* = ptr GcThread + GcThread* {.pure, inheritable.} = object + when emulatedThreadVars: + tls*: ThreadLocalStorage + else: + nil + when hasSharedHeap: + next*, prev*: PGcThread + stackBottom*, stackTop*: pointer + stackSize*: int + else: + nil + +const hasAllocStack* = defined(zephyr) # maybe freertos too? + +type + Thread*[TArg] = object + core*: PGcThread + sys*: SysThread + when TArg is void: + dataFn*: proc () {.nimcall, gcsafe.} + else: + dataFn*: proc (m: TArg) {.nimcall, gcsafe.} + data*: TArg + when hasAllocStack: + rawStack*: pointer + +proc `=copy`*[TArg](x: var Thread[TArg], y: Thread[TArg]) {.error.} diff --git a/lib/std/threads.nim b/lib/std/threads.nim new file mode 100644 index 000000000..5726a5cdc --- /dev/null +++ b/lib/std/threads.nim @@ -0,0 +1,276 @@ +# +# +# Nim's Runtime Library +# (c) Copyright 2012 Andreas Rumpf +# +# See the file "copying.txt", included in this +# distribution, for details about the copyright. +# + +## Thread support for Nim. +## +## Nim's memory model for threads is quite different from other common +## programming languages (C, Pascal): Each thread has its own +## (garbage collected) heap and sharing of memory is restricted. This helps +## to prevent race conditions and improves efficiency. See `the manual for +## details of this memory model <manual.html#threads>`_. +## +## Examples +## ======== +## +## .. code-block:: Nim +## +## import std/locks +## +## var +## thr: array[0..4, Thread[tuple[a,b: int]]] +## L: Lock +## +## proc threadFunc(interval: tuple[a,b: int]) {.thread.} = +## for i in interval.a..interval.b: +## acquire(L) # lock stdout +## echo i +## release(L) +## +## initLock(L) +## +## for i in 0..high(thr): +## createThread(thr[i], threadFunc, (i*10, i*10+5)) +## joinThreads(thr) +## +## deinitLock(L) + +import std/private/[threadtypes] +export Thread + +import system/ansi_c + +when defined(nimPreviewSlimSystem): + import std/assertions + +when defined(genode): + import genode/env + + +when hasAllocStack or defined(zephyr) or defined(freertos): + const + nimThreadStackSize {.intdefine.} = 8192 + nimThreadStackGuard {.intdefine.} = 128 + + StackGuardSize = nimThreadStackGuard + ThreadStackSize = nimThreadStackSize - nimThreadStackGuard +else: + const + StackGuardSize = 4096 + ThreadStackMask = + when defined(genode): + 1024*64*sizeof(int)-1 + else: + 1024*256*sizeof(int)-1 + + ThreadStackSize = ThreadStackMask+1 - StackGuardSize + + +when defined(gcDestructors): + proc allocThreadStorage(size: int): pointer = + result = c_malloc(csize_t size) + zeroMem(result, size) +else: + template allocThreadStorage(size: untyped): untyped = allocShared0(size) + +#const globalsSlot = ThreadVarSlot(0) +#sysAssert checkSlot.int == globalsSlot.int + +# Zephyr doesn't include this properly without some help +when defined(zephyr): + {.emit: """/*INCLUDESECTION*/ + #include <pthread.h> + """.} + + +# We jump through some hops here to ensure that Nim thread procs can have +# the Nim calling convention. This is needed because thread procs are +# ``stdcall`` on Windows and ``noconv`` on UNIX. Alternative would be to just +# use ``stdcall`` since it is mapped to ``noconv`` on UNIX anyway. + + + +proc onThreadDestruction*(handler: proc () {.closure, gcsafe, raises: [].}) = + ## Registers a *thread local* handler that is called at the thread's + ## destruction. + ## + ## A thread is destructed when the `.thread` proc returns + ## normally or when it raises an exception. Note that unhandled exceptions + ## in a thread nevertheless cause the whole process to die. + nimThreadDestructionHandlers.add handler + + +{.push stack_trace:off.} +when defined(windows): + proc threadProcWrapper[TArg](closure: pointer): int32 {.stdcall.} = + nimThreadProcWrapperBody(closure) + # implicitly return 0 +elif defined(genode): + proc threadProcWrapper[TArg](closure: pointer) {.noconv.} = + nimThreadProcWrapperBody(closure) +else: + proc threadProcWrapper[TArg](closure: pointer): pointer {.noconv.} = + nimThreadProcWrapperBody(closure) +{.pop.} + +proc running*[TArg](t: Thread[TArg]): bool {.inline.} = + ## Returns true if `t` is running. + result = t.dataFn != nil + +proc handle*[TArg](t: Thread[TArg]): SysThread {.inline.} = + ## Returns the thread handle of `t`. + result = t.sys + +when hostOS == "windows": + const MAXIMUM_WAIT_OBJECTS = 64 + + proc joinThread*[TArg](t: Thread[TArg]) {.inline.} = + ## Waits for the thread `t` to finish. + discard waitForSingleObject(t.sys, -1'i32) + + proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = + ## Waits for every thread in `t` to finish. + var a: array[MAXIMUM_WAIT_OBJECTS, SysThread] + var k = 0 + while k < len(t): + var count = min(len(t) - k, MAXIMUM_WAIT_OBJECTS) + for i in 0..(count - 1): a[i] = t[i + k].sys + discard waitForMultipleObjects(int32(count), + cast[ptr SysThread](addr(a)), 1, -1) + inc(k, MAXIMUM_WAIT_OBJECTS) + +elif defined(genode): + proc joinThread*[TArg](t: Thread[TArg]) {.importcpp.} + ## Waits for the thread `t` to finish. + + proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = + ## Waits for every thread in `t` to finish. + for i in 0..t.high: joinThread(t[i]) + +else: + proc joinThread*[TArg](t: Thread[TArg]) {.inline.} = + ## Waits for the thread `t` to finish. + discard pthread_join(t.sys, nil) + + proc joinThreads*[TArg](t: varargs[Thread[TArg]]) = + ## Waits for every thread in `t` to finish. + for i in 0..t.high: joinThread(t[i]) + +when false: + # XXX a thread should really release its heap here somehow: + proc destroyThread*[TArg](t: var Thread[TArg]) = + ## Forces the thread `t` to terminate. This is potentially dangerous if + ## you don't have full control over `t` and its acquired resources. + when hostOS == "windows": + discard TerminateThread(t.sys, 1'i32) + else: + discard pthread_cancel(t.sys) + when declared(registerThread): unregisterThread(addr(t)) + t.dataFn = nil + ## if thread `t` already exited, `t.core` will be `null`. + if not isNil(t.core): + deallocThreadStorage(t.core) + t.core = nil + +when hostOS == "windows": + proc createThread*[TArg](t: var Thread[TArg], + tp: proc (arg: TArg) {.thread, nimcall.}, + param: TArg) = + ## Creates a new thread `t` and starts its execution. + ## + ## Entry point is the proc `tp`. + ## `param` is passed to `tp`. `TArg` can be `void` if you + ## don't need to pass any data to the thread. + t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) + + when TArg isnot void: t.data = param + t.dataFn = tp + when hasSharedHeap: t.core.stackSize = ThreadStackSize + var dummyThreadId: int32 + t.sys = createThread(nil, ThreadStackSize, threadProcWrapper[TArg], + addr(t), 0'i32, dummyThreadId) + if t.sys <= 0: + raise newException(ResourceExhaustedError, "cannot create thread") + + proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = + ## Pins a thread to a `CPU`:idx:. + ## + ## In other words sets a thread's `affinity`:idx:. + ## If you don't know what this means, you shouldn't use this proc. + setThreadAffinityMask(t.sys, uint(1 shl cpu)) + +elif defined(genode): + var affinityOffset: cuint = 1 + ## CPU affinity offset for next thread, safe to roll-over. + + proc createThread*[TArg](t: var Thread[TArg], + tp: proc (arg: TArg) {.thread, nimcall.}, + param: TArg) = + t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) + + when TArg isnot void: t.data = param + t.dataFn = tp + when hasSharedHeap: t.stackSize = ThreadStackSize + t.sys.initThread( + runtimeEnv, + ThreadStackSize.culonglong, + threadProcWrapper[TArg], addr(t), affinityOffset) + inc affinityOffset + + proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = + {.hint: "cannot change Genode thread CPU affinity after initialization".} + discard + +else: + proc createThread*[TArg](t: var Thread[TArg], + tp: proc (arg: TArg) {.thread, nimcall.}, + param: TArg) = + ## Creates a new thread `t` and starts its execution. + ## + ## Entry point is the proc `tp`. `param` is passed to `tp`. + ## `TArg` can be `void` if you + ## don't need to pass any data to the thread. + t.core = cast[PGcThread](allocThreadStorage(sizeof(GcThread))) + + when TArg isnot void: t.data = param + t.dataFn = tp + when hasSharedHeap: t.core.stackSize = ThreadStackSize + var a {.noinit.}: Pthread_attr + doAssert pthread_attr_init(a) == 0 + when hasAllocStack: + var + rawstk = allocThreadStorage(ThreadStackSize + StackGuardSize) + stk = cast[pointer](cast[uint](rawstk) + StackGuardSize) + let setstacksizeResult = pthread_attr_setstack(addr a, stk, ThreadStackSize) + t.rawStack = rawstk + else: + let setstacksizeResult = pthread_attr_setstacksize(a, ThreadStackSize) + + when not defined(ios): + # This fails on iOS + doAssert(setstacksizeResult == 0) + if pthread_create(t.sys, a, threadProcWrapper[TArg], addr(t)) != 0: + raise newException(ResourceExhaustedError, "cannot create thread") + doAssert pthread_attr_destroy(a) == 0 + + proc pinToCpu*[Arg](t: var Thread[Arg]; cpu: Natural) = + ## Pins a thread to a `CPU`:idx:. + ## + ## In other words sets a thread's `affinity`:idx:. + ## If you don't know what this means, you shouldn't use this proc. + when not defined(macosx): + var s {.noinit.}: CpuSet + cpusetZero(s) + cpusetIncl(cpu.cint, s) + setAffinity(t.sys, csize_t(sizeof(s)), s) + +proc createThread*(t: var Thread[void], tp: proc () {.thread, nimcall.}) = + createThread[void](t, tp) + +when not defined(gcOrc): + include system/threadids |