csgo-2018-source/public/vstdlib/jobthread.h
2021-07-24 21:11:47 -07:00

1471 lines
48 KiB
C++
Raw Permalink Blame History

//========== Copyright <20> 2005, Valve Corporation, All rights reserved. ========
//
// Purpose: A utility for a discrete job-oriented worker thread.
//
// The class CThreadPool is both the job queue, and the
// worker thread. Except when the main thread attempts to
// synchronously execute a job, most of the inter-thread locking
// on the queue.
//
// The queue threading model uses a manual reset event for optimal
// throughput. Adding to the queue is guarded by a semaphore that
// will block the inserting thread if the queue has overflown.
// This prevents the worker thread from being starved out even if
// not running at a higher priority than the master thread.
//
// The thread function waits for jobs, services jobs, and manages
// communication between the worker and master threads. The nature
// of the work is opaque to the Executer.
//
// CJob instances actually do the work. The base class
// calls virtual methods for job primitives, so derivations don't
// need to worry about threading models. All of the variants of
// job and OS can be expressed in this hierarchy. Instances of
// CJob are the items placed in the queue, and by
// overriding the job primitives they are the manner by which
// users of the Executer control the state of the job.
//
//=============================================================================
#include <limits.h>
#include "tier0/threadtools.h"
#include "tier1/refcount.h"
#include "tier1/utllinkedlist.h"
#include "tier1/utlvector.h"
#include "tier1/functors.h"
#include "vstdlib/vstdlib.h"
#ifndef JOBTHREAD_H
#define JOBTHREAD_H
#ifdef AddJob // windows.h print function collisions
#undef AddJob
#undef GetJob
#endif
#ifdef VSTDLIB_DLL_EXPORT
#define JOB_INTERFACE DLL_EXPORT
#define JOB_OVERLOAD DLL_GLOBAL_EXPORT
#define JOB_CLASS DLL_CLASS_EXPORT
#else
#define JOB_INTERFACE DLL_IMPORT
#define JOB_OVERLOAD DLL_GLOBAL_IMPORT
#define JOB_CLASS DLL_CLASS_IMPORT
#endif
#if defined( _WIN32 )
#pragma once
#endif
//-----------------------------------------------------------------------------
//
//-----------------------------------------------------------------------------
class CJob;
//-----------------------------------------------------------------------------
//
//-----------------------------------------------------------------------------
enum JobStatusEnum_t
{
// Use negative for errors
JOB_OK, // operation is successful
JOB_STATUS_PENDING, // file is properly queued, waiting for service
JOB_STATUS_INPROGRESS, // file is being accessed
JOB_STATUS_ABORTED, // file was aborted by caller
JOB_STATUS_UNSERVICED, // file is not yet queued
};
typedef int JobStatus_t;
enum JobFlags_t
{
JF_IO = ( 1 << 0 ), // The job primarily blocks on IO or hardware
JF_BOOST_THREAD = ( 1 << 1 ), // Up the thread priority to max allowed while processing task
JF_SERIAL = ( 1 << 2 ), // Job cannot be executed out of order relative to other "strict" jobs
JF_QUEUE = ( 1 << 3 ), // Queue it, even if not an IO job
};
enum JobPriority_t
{
JP_LOW,
JP_NORMAL,
JP_HIGH,
JP_IMMEDIATE,
JP_NUM_PRIORITIES,
// Priority aliases for game jobs
JP_FRAME = JP_NORMAL,
JP_FRAME_SEGMENT = JP_HIGH,
};
#define TP_MAX_POOL_THREADS 64
struct ThreadPoolStartParams_t
{
ThreadPoolStartParams_t( bool bIOThreads = false, unsigned nThreads = (unsigned)-1, int *pAffinities = NULL, ThreeState_t fDistribute = TRS_NONE, unsigned nStackSize = (unsigned)-1, int iThreadPriority = SHRT_MIN )
: bIOThreads( bIOThreads ), nThreads( nThreads ), nThreadsMax( -1 ), fDistribute( fDistribute ), nStackSize( nStackSize ), iThreadPriority( iThreadPriority )
{
bExecOnThreadPoolThreadsOnly = false;
#if defined( DEDICATED ) && IsPlatformLinux()
bEnableOnLinuxDedicatedServer = false; // by default, thread pools don't start up on Linux DS
#endif
bUseAffinityTable = ( pAffinities != NULL ) && ( fDistribute == TRS_TRUE ) && ( nThreads != (unsigned)-1 );
if ( bUseAffinityTable )
{
// user supplied an optional 1:1 affinity mapping to override normal distribute behavior
nThreads = MIN( TP_MAX_POOL_THREADS, nThreads );
for ( unsigned int i = 0; i < nThreads; i++ )
{
iAffinityTable[i] = pAffinities[i];
}
}
}
int nThreads;
int nThreadsMax;
ThreeState_t fDistribute;
int nStackSize;
int iThreadPriority;
int iAffinityTable[TP_MAX_POOL_THREADS];
bool bIOThreads : 1;
bool bUseAffinityTable : 1;
bool bExecOnThreadPoolThreadsOnly : 1;
#if defined( DEDICATED ) && IsPlatformLinux()
bool bEnableOnLinuxDedicatedServer : 1;
#endif
};
//-----------------------------------------------------------------------------
//
// IThreadPool
//
//-----------------------------------------------------------------------------
typedef bool (*JobFilter_t)( CJob * );
//---------------------------------------------------------
// Messages supported through the CallWorker() method
//---------------------------------------------------------
enum ThreadPoolMessages_t
{
TPM_EXIT, // Exit the thread
TPM_SUSPEND, // Suspend after next operation
};
//---------------------------------------------------------
#ifdef Yield
#undef Yield
#endif
abstract_class IThreadPool : public IRefCounted
{
public:
virtual ~IThreadPool() {};
//-----------------------------------------------------
// Thread functions
//-----------------------------------------------------
virtual bool Start( const ThreadPoolStartParams_t &startParams = ThreadPoolStartParams_t() ) = 0;
virtual bool Stop( int timeout = TT_INFINITE ) = 0;
//-----------------------------------------------------
// Functions for any thread
//-----------------------------------------------------
virtual unsigned GetJobCount() = 0;
virtual int NumThreads() = 0;
virtual int NumIdleThreads() = 0;
//-----------------------------------------------------
// Pause/resume processing jobs
//-----------------------------------------------------
virtual int SuspendExecution() = 0;
virtual int ResumeExecution() = 0;
//-----------------------------------------------------
// Offer the current thread to the pool
//-----------------------------------------------------
virtual int YieldWait( CThreadEvent **pEvents, int nEvents, bool bWaitAll = true, unsigned timeout = TT_INFINITE ) = 0;
virtual int YieldWait( CJob **, int nJobs, bool bWaitAll = true, unsigned timeout = TT_INFINITE ) = 0;
virtual void Yield( unsigned timeout ) = 0;
bool YieldWait( CThreadEvent &event, unsigned timeout = TT_INFINITE );
bool YieldWait( CJob *, unsigned timeout = TT_INFINITE );
//-----------------------------------------------------
// Add a native job to the queue (master thread)
// See AddPerFrameJob below if you want to add a job that
// wants to be run before the end of the frame
//-----------------------------------------------------
virtual void AddJob( CJob * ) = 0;
//-----------------------------------------------------
// Add an function object to the queue (master thread)
//-----------------------------------------------------
virtual void AddFunctor( CFunctor *pFunctor, CJob **ppJob = NULL, const char *pszDescription = NULL, unsigned flags = 0 ) { AddFunctorInternal( RetAddRef( pFunctor ), ppJob, pszDescription, flags ); }
//-----------------------------------------------------
// Change the priority of an active job
//-----------------------------------------------------
virtual void ChangePriority( CJob *p, JobPriority_t priority ) = 0;
//-----------------------------------------------------
// Bulk job manipulation (blocking)
//-----------------------------------------------------
int ExecuteAll( JobFilter_t pfnFilter = NULL ) { return ExecuteToPriority( JP_LOW, pfnFilter ); }
virtual int ExecuteToPriority( JobPriority_t toPriority, JobFilter_t pfnFilter = NULL ) = 0;
virtual int AbortAll() = 0;
//-----------------------------------------------------
// Add a native job to the queue (master thread)
// Call YieldWaitPerFrameJobs() to wait only until all per-frame jobs are done
//-----------------------------------------------------
virtual void AddPerFrameJob( CJob * ) = 0;
//-----------------------------------------------------
// Add an arbitrary call to the queue (master thread)
//
// Avert thy eyes! Imagine rather:
//
// CJob *AddCall( <function>, [args1, [arg2,]...]
// CJob *AddCall( <object>, <function>, [args1, [arg2,]...]
// CJob *AddRefCall( <object>, <function>, [args1, [arg2,]...]
// CJob *QueueCall( <function>, [args1, [arg2,]...]
// CJob *QueueCall( <object>, <function>, [args1, [arg2,]...]
//-----------------------------------------------------
#define DEFINE_NONMEMBER_ADD_CALL(N) \
template <typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *AddCall(FUNCTION_RETTYPE (*pfnProxied)( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
if ( !NumIdleThreads() ) \
{ \
pJob = GetDummyJob(); \
FunctorDirectCall( pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
AddFunctorInternal( CreateFunctor( pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob ); \
} \
\
return pJob; \
}
//-------------------------------------
#define DEFINE_MEMBER_ADD_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *AddCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
if ( !NumIdleThreads() ) \
{ \
pJob = GetDummyJob(); \
FunctorDirectCall( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
AddFunctorInternal( CreateFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob ); \
} \
\
return pJob; \
}
//-------------------------------------
#define DEFINE_CONST_MEMBER_ADD_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *AddCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) const FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
if ( !NumIdleThreads() ) \
{ \
pJob = GetDummyJob(); \
FunctorDirectCall( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
AddFunctorInternal( CreateFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob ); \
} \
\
return pJob; \
}
//-------------------------------------
#define DEFINE_REF_COUNTING_MEMBER_ADD_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *AddRefCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
if ( !NumIdleThreads() ) \
{ \
pJob = GetDummyJob(); \
FunctorDirectCall( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
AddFunctorInternal( CreateRefCountingFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob ); \
} \
\
return pJob; \
}
//-------------------------------------
#define DEFINE_REF_COUNTING_CONST_MEMBER_ADD_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *AddRefCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) const FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
if ( !NumIdleThreads() ) \
{ \
pJob = GetDummyJob(); \
FunctorDirectCall( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
AddFunctorInternal( CreateRefCountingFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob ); \
} \
\
return pJob; \
}
//-----------------------------------------------------------------------------
#define DEFINE_NONMEMBER_QUEUE_CALL(N) \
template <typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *QueueCall(FUNCTION_RETTYPE (*pfnProxied)( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
AddFunctorInternal( CreateFunctor( pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob, NULL, JF_QUEUE ); \
return pJob; \
}
//-------------------------------------
#define DEFINE_MEMBER_QUEUE_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *QueueCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
AddFunctorInternal( CreateFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob, NULL, JF_QUEUE ); \
return pJob; \
}
//-------------------------------------
#define DEFINE_CONST_MEMBER_QUEUE_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *QueueCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) const FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
AddFunctorInternal( CreateFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob, NULL, JF_QUEUE ); \
return pJob; \
}
//-------------------------------------
#define DEFINE_REF_COUNTING_MEMBER_QUEUE_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *QueueRefCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
AddFunctorInternal( CreateRefCountingFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob, NULL, JF_QUEUE ); \
return pJob; \
}
//-------------------------------------
#define DEFINE_REF_COUNTING_CONST_MEMBER_QUEUE_CALL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N> \
CJob *QueueRefCall(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ) const FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
CJob *pJob; \
AddFunctorInternal( CreateRefCountingFunctor( pObject, pfnProxied FUNC_FUNCTOR_CALL_ARGS_##N ), &pJob, NULL, JF_QUEUE ); \
\
return pJob; \
}
FUNC_GENERATE_ALL( DEFINE_NONMEMBER_ADD_CALL );
FUNC_GENERATE_ALL( DEFINE_MEMBER_ADD_CALL );
FUNC_GENERATE_ALL( DEFINE_CONST_MEMBER_ADD_CALL );
FUNC_GENERATE_ALL( DEFINE_REF_COUNTING_MEMBER_ADD_CALL );
FUNC_GENERATE_ALL( DEFINE_REF_COUNTING_CONST_MEMBER_ADD_CALL );
FUNC_GENERATE_ALL( DEFINE_NONMEMBER_QUEUE_CALL );
FUNC_GENERATE_ALL( DEFINE_MEMBER_QUEUE_CALL );
FUNC_GENERATE_ALL( DEFINE_CONST_MEMBER_QUEUE_CALL );
FUNC_GENERATE_ALL( DEFINE_REF_COUNTING_MEMBER_QUEUE_CALL );
FUNC_GENERATE_ALL( DEFINE_REF_COUNTING_CONST_MEMBER_QUEUE_CALL );
#undef DEFINE_NONMEMBER_ADD_CALL
#undef DEFINE_MEMBER_ADD_CALL
#undef DEFINE_CONST_MEMBER_ADD_CALL
#undef DEFINE_REF_COUNTING_MEMBER_ADD_CALL
#undef DEFINE_REF_COUNTING_CONST_MEMBER_ADD_CALL
#undef DEFINE_NONMEMBER_QUEUE_CALL
#undef DEFINE_MEMBER_QUEUE_CALL
#undef DEFINE_CONST_MEMBER_QUEUE_CALL
#undef DEFINE_REF_COUNTING_MEMBER_QUEUE_CALL
#undef DEFINE_REF_COUNTING_CONST_MEMBER_QUEUE_CALL
private:
virtual void AddFunctorInternal( CFunctor *, CJob ** = NULL, const char *pszDescription = NULL, unsigned flags = 0 ) = 0;
//-----------------------------------------------------
// Services for internal use by job instances
//-----------------------------------------------------
friend class CJob;
virtual CJob *GetDummyJob() = 0;
public:
virtual void Distribute( bool bDistribute = true, int *pAffinityTable = NULL ) = 0;
virtual bool Start( const ThreadPoolStartParams_t &startParams, const char *pszNameOverride ) = 0;
virtual int YieldWaitPerFrameJobs( ) = 0;
};
//-----------------------------------------------------------------------------
JOB_INTERFACE IThreadPool *CreateNewThreadPool();
JOB_INTERFACE void DestroyThreadPool( IThreadPool *pPool );
//-------------------------------------
JOB_INTERFACE void RunThreadPoolTests();
//-----------------------------------------------------------------------------
JOB_INTERFACE IThreadPool *g_pThreadPool;
#ifdef _X360
JOB_INTERFACE IThreadPool *g_pAlternateThreadPool;
#endif
//-----------------------------------------------------------------------------
// Class to combine the metadata for an operation and the ability to perform
// the operation. Meant for inheritance. All functions inline, defers to executor
//-----------------------------------------------------------------------------
DECLARE_POINTER_HANDLE( ThreadPoolData_t );
#define JOB_NO_DATA ((ThreadPoolData_t)-1)
class CJob : public CRefCounted1<IRefCounted, CRefCountServiceMT>
{
public:
CJob( JobPriority_t priority = JP_NORMAL )
: m_status( JOB_STATUS_UNSERVICED ),
m_ThreadPoolData( JOB_NO_DATA ),
m_priority( priority ),
m_flags( 0 ),
m_pThreadPool( NULL ),
m_CompleteEvent( true ),
m_iServicingThread( -1 )
{
}
//-----------------------------------------------------
// Priority (not thread safe)
//-----------------------------------------------------
void SetPriority( JobPriority_t priority ) { m_priority = priority; }
JobPriority_t GetPriority() const { return m_priority; }
//-----------------------------------------------------
void SetFlags( unsigned flags ) { m_flags = flags; }
unsigned GetFlags() const { return m_flags; }
//-----------------------------------------------------
void SetServiceThread( int iServicingThread ) { m_iServicingThread = (char)iServicingThread; }
int GetServiceThread() const { return m_iServicingThread; }
void ClearServiceThread() { m_iServicingThread = -1; }
//-----------------------------------------------------
// Fast queries
//-----------------------------------------------------
bool Executed() const { return ( m_status == JOB_OK ); }
bool CanExecute() const { return ( m_status == JOB_STATUS_PENDING || m_status == JOB_STATUS_UNSERVICED ); }
bool IsFinished() const { return ( m_status != JOB_STATUS_PENDING && m_status != JOB_STATUS_INPROGRESS && m_status != JOB_STATUS_UNSERVICED ); }
JobStatus_t GetStatus() const { return m_status; }
//-----------------------------------------------------
// Try to acquire ownership (to satisfy). If you take the lock, you must either execute or abort.
//-----------------------------------------------------
bool TryLock() { return m_mutex.TryLock(); }
void Lock() { m_mutex.Lock(); }
void Unlock() { m_mutex.Unlock(); }
//-----------------------------------------------------
// Thread event support (safe for NULL this to simplify code )
//-----------------------------------------------------
bool WaitForFinish( uint32 dwTimeout = TT_INFINITE ) { if (!this) return true; return ( !IsFinished() ) ? g_pThreadPool->YieldWait( this, dwTimeout ) : true; }
bool WaitForFinishAndRelease( uint32 dwTimeout = TT_INFINITE ) { if (!this) return true; bool bResult = WaitForFinish( dwTimeout); Release(); return bResult; }
CThreadEvent *AccessEvent() { return &m_CompleteEvent; }
//-----------------------------------------------------
// Perform the job
//-----------------------------------------------------
JobStatus_t Execute();
JobStatus_t TryExecute();
JobStatus_t ExecuteAndRelease() { JobStatus_t status = Execute(); Release(); return status; }
JobStatus_t TryExecuteAndRelease() { JobStatus_t status = TryExecute(); Release(); return status; }
//-----------------------------------------------------
// Terminate the job, discard if partially or wholly fulfilled
//-----------------------------------------------------
JobStatus_t Abort( bool bDiscard = true );
virtual char const *Describe() { return "Job"; }
private:
//-----------------------------------------------------
friend class CThreadPool;
JobStatus_t m_status;
JobPriority_t m_priority;
CThreadMutex m_mutex;
unsigned char m_flags;
char m_iServicingThread;
short m_reserved;
ThreadPoolData_t m_ThreadPoolData;
IThreadPool * m_pThreadPool;
CThreadEvent m_CompleteEvent;
#if defined( THREAD_PARENT_STACK_TRACE_ENABLED )
void * m_ParentStackTrace[THREAD_PARENT_STACK_TRACE_LENGTH];
#endif
private:
//-----------------------------------------------------
CJob( const CJob &fromRequest );
void operator=(const CJob &fromRequest );
virtual JobStatus_t DoExecute() = 0;
virtual JobStatus_t DoAbort( bool bDiscard ) { return JOB_STATUS_ABORTED; }
virtual void DoCleanup() {}
};
//-----------------------------------------------------------------------------
class CFunctorJob : public CJob
{
public:
CFunctorJob( CFunctor *pFunctor, const char *pszDescription = NULL )
: m_pFunctor( pFunctor )
{
if ( pszDescription )
{
Q_strncpy( m_szDescription, pszDescription, sizeof(m_szDescription) );
}
else
{
m_szDescription[0] = 0;
}
}
virtual JobStatus_t DoExecute()
{
(*m_pFunctor)();
return JOB_OK;
}
const char *Describe()
{
return m_szDescription;
}
private:
CRefPtr<CFunctor> m_pFunctor;
char m_szDescription[16];
};
//-----------------------------------------------------------------------------
// Utility for managing multiple jobs
//-----------------------------------------------------------------------------
class CJobSet
{
public:
CJobSet( CJob *pJob = NULL )
{
if ( pJob )
{
m_jobs.AddToTail( pJob );
}
}
CJobSet( CJob **ppJobs, int nJobs )
{
if ( ppJobs )
{
m_jobs.AddMultipleToTail( nJobs, ppJobs );
}
}
~CJobSet()
{
for ( int i = 0; i < m_jobs.Count(); i++ )
{
m_jobs[i]->Release();
}
}
void operator+=( CJob *pJob )
{
m_jobs.AddToTail( pJob );
}
void operator-=( CJob *pJob )
{
m_jobs.FindAndRemove( pJob );
}
void Execute( bool bRelease = true )
{
for ( int i = 0; i < m_jobs.Count(); i++ )
{
m_jobs[i]->Execute();
if ( bRelease )
{
m_jobs[i]->Release();
}
}
if ( bRelease )
{
m_jobs.RemoveAll();
}
}
void Abort( bool bRelease = true )
{
for ( int i = 0; i < m_jobs.Count(); i++ )
{
m_jobs[i]->Abort();
if ( bRelease )
{
m_jobs[i]->Release();
}
}
if ( bRelease )
{
m_jobs.RemoveAll();
}
}
void WaitForFinish( bool bRelease = true )
{
for ( int i = 0; i < m_jobs.Count(); i++ )
{
m_jobs[i]->WaitForFinish();
if ( bRelease )
{
m_jobs[i]->Release();
}
}
if ( bRelease )
{
m_jobs.RemoveAll();
}
}
void WaitForFinish( IThreadPool *pPool, bool bRelease = true )
{
pPool->YieldWait( m_jobs.Base(), m_jobs.Count() );
if ( bRelease )
{
for ( int i = 0; i < m_jobs.Count(); i++ )
{
m_jobs[i]->Release();
}
m_jobs.RemoveAll();
}
}
private:
CUtlVectorFixed<CJob *, 16> m_jobs;
};
//-----------------------------------------------------------------------------
// Job helpers
//-----------------------------------------------------------------------------
#define ThreadExecute g_pThreadPool->QueueCall
#define ThreadExecuteRef g_pThreadPool->QueueRefCall
#define BeginExecuteParallel() do { CJobSet jobSet
#define EndExecuteParallel() jobSet.WaitForFinish( g_pThreadPool ); } while (0)
#define ExecuteParallel jobSet += g_pThreadPool->QueueCall
#define ExecuteRefParallel jobSet += g_pThreadPool->QueueCallRef
//-----------------------------------------------------------------------------
// Work splitting: array split, best when cost per item is roughly equal
//-----------------------------------------------------------------------------
#pragma warning(push)
#pragma warning(disable:4389)
#pragma warning(disable:4018)
#pragma warning(disable:4701)
#define DEFINE_NON_MEMBER_ITER_RANGE_PARALLEL(N) \
template <typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N, typename ITERTYPE1, typename ITERTYPE2> \
void IterRangeParallel(FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( ITERTYPE1, ITERTYPE2 FUNC_SEPARATOR_##N FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ), ITERTYPE1 from, ITERTYPE2 to FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
const int MAX_THREADS = 16; \
int nIdle = g_pThreadPool->NumIdleThreads(); \
ITERTYPE1 range = to - from; \
int nThreads = min( nIdle + 1, range ); \
if ( nThreads > MAX_THREADS ) \
{ \
nThreads = MAX_THREADS; \
} \
if ( nThreads < 2 ) \
{ \
FunctorDirectCall( pfnProxied, from, to FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
ITERTYPE1 nIncrement = range / nThreads; \
\
CJobSet jobSet; \
while ( --nThreads ) \
{ \
ITERTYPE2 thisTo = from + nIncrement; \
jobSet += g_pThreadPool->AddCall( pfnProxied, from, thisTo FUNC_FUNCTOR_CALL_ARGS_##N ); \
from = thisTo; \
} \
FunctorDirectCall( pfnProxied, from, to FUNC_FUNCTOR_CALL_ARGS_##N ); \
jobSet.WaitForFinish( g_pThreadPool ); \
} \
\
}
FUNC_GENERATE_ALL( DEFINE_NON_MEMBER_ITER_RANGE_PARALLEL );
#define DEFINE_MEMBER_ITER_RANGE_PARALLEL(N) \
template <typename OBJECT_TYPE, typename FUNCTION_CLASS, typename FUNCTION_RETTYPE FUNC_TEMPLATE_FUNC_PARAMS_##N FUNC_TEMPLATE_ARG_PARAMS_##N, typename ITERTYPE1, typename ITERTYPE2> \
void IterRangeParallel(OBJECT_TYPE *pObject, FUNCTION_RETTYPE ( FUNCTION_CLASS::*pfnProxied )( ITERTYPE1, ITERTYPE2 FUNC_SEPARATOR_##N FUNC_BASE_TEMPLATE_FUNC_PARAMS_##N ), ITERTYPE1 from, ITERTYPE2 to FUNC_ARG_FORMAL_PARAMS_##N ) \
{ \
const int MAX_THREADS = 16; \
int nIdle = g_pThreadPool->NumIdleThreads(); \
ITERTYPE1 range = to - from; \
int nThreads = min( nIdle + 1, range ); \
if ( nThreads > MAX_THREADS ) \
{ \
nThreads = MAX_THREADS; \
} \
if ( nThreads < 2 ) \
{ \
FunctorDirectCall( pObject, pfnProxied, from, to FUNC_FUNCTOR_CALL_ARGS_##N ); \
} \
else \
{ \
ITERTYPE1 nIncrement = range / nThreads; \
\
CJobSet jobSet; \
while ( --nThreads ) \
{ \
ITERTYPE2 thisTo = from + nIncrement; \
jobSet += g_pThreadPool->AddCall( pObject, pfnProxied, from, thisTo FUNC_FUNCTOR_CALL_ARGS_##N ); \
from = thisTo; \
} \
FunctorDirectCall( pObject, pfnProxied, from, to FUNC_FUNCTOR_CALL_ARGS_##N ); \
jobSet.WaitForFinish( g_pThreadPool ); \
} \
\
}
FUNC_GENERATE_ALL( DEFINE_MEMBER_ITER_RANGE_PARALLEL );
//-----------------------------------------------------------------------------
// Work splitting: competitive, best when cost per item varies a lot
//-----------------------------------------------------------------------------
template <typename T>
class CJobItemProcessor
{
public:
typedef T ItemType_t;
void Begin() {}
// void Process( ItemType_t & ) {}
void End() {}
};
template <typename T>
class CFuncJobItemProcessor : public CJobItemProcessor<T>
{
public:
void Init(void (*pfnProcess)( T & ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL )
{
m_pfnProcess = pfnProcess;
m_pfnBegin = pfnBegin;
m_pfnEnd = pfnEnd;
}
//CFuncJobItemProcessor(OBJECT_TYPE_PTR pObject, void (FUNCTION_CLASS::*pfnProcess)( ITEM_TYPE & ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL );
void Begin() { if ( m_pfnBegin ) (*m_pfnBegin)(); }
void Process( T &item ) { (*m_pfnProcess)( item ); }
void End() { if ( m_pfnEnd ) (*m_pfnEnd)(); }
protected:
void (*m_pfnProcess)( T & );
void (*m_pfnBegin)();
void (*m_pfnEnd)();
};
template <typename T, class OBJECT_TYPE, class FUNCTION_CLASS = OBJECT_TYPE >
class CMemberFuncJobItemProcessor : public CJobItemProcessor<T>
{
public:
void Init( OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( T & ), void (FUNCTION_CLASS::*pfnBegin)() = NULL, void (FUNCTION_CLASS::*pfnEnd)() = NULL )
{
m_pObject = pObject;
m_pfnProcess = pfnProcess;
m_pfnBegin = pfnBegin;
m_pfnEnd = pfnEnd;
}
void Begin() { if ( m_pfnBegin ) ((*m_pObject).*m_pfnBegin)(); }
void Process( T &item ) { ((*m_pObject).*m_pfnProcess)( item ); }
void End() { if ( m_pfnEnd ) ((*m_pObject).*m_pfnEnd)(); }
protected:
OBJECT_TYPE *m_pObject;
void (FUNCTION_CLASS::*m_pfnProcess)( T & );
void (FUNCTION_CLASS::*m_pfnBegin)();
void (FUNCTION_CLASS::*m_pfnEnd)();
};
template <typename T>
class CLoopFuncJobItemProcessor : public CJobItemProcessor<T>
{
public:
void Init(void (*pfnProcess)( T*, int, int ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL )
{
m_pfnProcess = pfnProcess;
m_pfnBegin = pfnBegin;
m_pfnEnd = pfnEnd;
}
void Begin() { if ( m_pfnBegin ) (*m_pfnBegin)(); }
void Process( T* pContext, int nFirst, int nCount ) { (*m_pfnProcess)( pContext, nFirst, nCount ); }
void End() { if ( m_pfnEnd ) (*m_pfnEnd)(); }
protected:
void (*m_pfnProcess)( T*, int, int );
void (*m_pfnBegin)();
void (*m_pfnEnd)();
};
template <typename T, class OBJECT_TYPE, class FUNCTION_CLASS = OBJECT_TYPE >
class CLoopMemberFuncJobItemProcessor : public CJobItemProcessor<T>
{
public:
void Init( OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( T*, int, int ), void (FUNCTION_CLASS::*pfnBegin)() = NULL, void (FUNCTION_CLASS::*pfnEnd)() = NULL )
{
m_pObject = pObject;
m_pfnProcess = pfnProcess;
m_pfnBegin = pfnBegin;
m_pfnEnd = pfnEnd;
}
void Begin() { if ( m_pfnBegin ) ((*m_pObject).*m_pfnBegin)(); }
void Process( T *item, int nFirst, int nCount ) { ((*m_pObject).*m_pfnProcess)( item, nFirst, nCount ); }
void End() { if ( m_pfnEnd ) ((*m_pObject).*m_pfnEnd)(); }
protected:
OBJECT_TYPE *m_pObject;
void (FUNCTION_CLASS::*m_pfnProcess)( T*, int, int );
void (FUNCTION_CLASS::*m_pfnBegin)();
void (FUNCTION_CLASS::*m_pfnEnd)();
};
#pragma warning(push)
#pragma warning(disable:4189)
template <typename ITEM_TYPE, class ITEM_PROCESSOR_TYPE, int ID_TO_PREVENT_COMDATS_IN_PROFILES = 1>
class CParallelProcessor
{
public:
CParallelProcessor()
{
m_pItems = m_pLimit= 0;
}
void Run( ITEM_TYPE *pItems, unsigned nItems, int nChunkSize = 1, int nMaxParallel = INT_MAX, IThreadPool *pThreadPool = NULL )
{
if ( nItems == 0 )
return;
#if defined(_X360)
volatile int ignored = ID_TO_PREVENT_COMDATS_IN_PROFILES;
#endif
m_nChunkSize = nChunkSize;
if ( !pThreadPool )
{
pThreadPool = g_pThreadPool;
}
m_pItems = pItems;
m_pLimit = pItems + nItems;
int nJobs = nItems - 1;
if ( nJobs > nMaxParallel )
{
nJobs = nMaxParallel;
}
if (! pThreadPool ) // only possible on linux
{
DoExecute( );
return;
}
int nThreads = pThreadPool->NumThreads();
if ( nJobs > nThreads )
{
nJobs = nThreads;
}
if ( nJobs > 0 )
{
CJob **jobs = (CJob **)stackalloc( nJobs * sizeof(CJob **) );
int i = nJobs;
while( i-- )
{
jobs[i] = pThreadPool->QueueCall( this, &CParallelProcessor<ITEM_TYPE, ITEM_PROCESSOR_TYPE, ID_TO_PREVENT_COMDATS_IN_PROFILES>::DoExecute );
}
DoExecute();
for ( i = 0; i < nJobs; i++ )
{
jobs[i]->Abort(); // will either abort ones that never got a thread, or noop on ones that did
jobs[i]->Release();
}
}
else
{
DoExecute();
}
}
ITEM_PROCESSOR_TYPE m_ItemProcessor;
private:
void DoExecute()
{
if ( m_pItems < m_pLimit )
{
#if defined(_X360)
volatile int ignored = ID_TO_PREVENT_COMDATS_IN_PROFILES;
#endif
m_ItemProcessor.Begin();
ITEM_TYPE *pLimit = m_pLimit;
int nChunkSize = m_nChunkSize;
for (;;)
{
ITEM_TYPE *pCurrent = m_pItems.AtomicAdd( nChunkSize );
ITEM_TYPE *pLast = MIN( pLimit, pCurrent + nChunkSize );
while( pCurrent < pLast )
{
m_ItemProcessor.Process( *pCurrent );
pCurrent++;
}
if ( pCurrent >= pLimit )
{
break;
}
}
m_ItemProcessor.End();
}
}
CInterlockedPtr<ITEM_TYPE> m_pItems;
ITEM_TYPE * m_pLimit;
int m_nChunkSize;
};
#pragma warning(pop)
template <typename ITEM_TYPE>
inline void ParallelProcess( ITEM_TYPE *pItems, unsigned nItems, void (*pfnProcess)( ITEM_TYPE & ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CFuncJobItemProcessor<ITEM_TYPE> > processor;
processor.m_ItemProcessor.Init( pfnProcess, pfnBegin, pfnEnd );
processor.Run( pItems, nItems, 1, nMaxParallel );
}
template <typename ITEM_TYPE, typename OBJECT_TYPE, typename FUNCTION_CLASS >
inline void ParallelProcess( ITEM_TYPE *pItems, unsigned nItems, OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( ITEM_TYPE & ), void (FUNCTION_CLASS::*pfnBegin)() = NULL, void (FUNCTION_CLASS::*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CMemberFuncJobItemProcessor<ITEM_TYPE, OBJECT_TYPE, FUNCTION_CLASS> > processor;
processor.m_ItemProcessor.Init( pObject, pfnProcess, pfnBegin, pfnEnd );
processor.Run( pItems, nItems, 1, nMaxParallel );
}
// Parallel Process that lets you specify threadpool
template <typename ITEM_TYPE>
inline void ParallelProcess( IThreadPool *pPool, ITEM_TYPE *pItems, unsigned nItems, void (*pfnProcess)( ITEM_TYPE & ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CFuncJobItemProcessor<ITEM_TYPE> > processor;
processor.m_ItemProcessor.Init( pfnProcess, pfnBegin, pfnEnd );
processor.Run( pItems, nItems, 1, nMaxParallel, pPool );
}
template <typename ITEM_TYPE, typename OBJECT_TYPE, typename FUNCTION_CLASS >
inline void ParallelProcess( IThreadPool *pPool, ITEM_TYPE *pItems, unsigned nItems, OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( ITEM_TYPE & ), void (FUNCTION_CLASS::*pfnBegin)() = NULL, void (FUNCTION_CLASS::*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CMemberFuncJobItemProcessor<ITEM_TYPE, OBJECT_TYPE, FUNCTION_CLASS> > processor;
processor.m_ItemProcessor.Init( pObject, pfnProcess, pfnBegin, pfnEnd );
processor.Run( pItems, nItems, 1, nMaxParallel, pPool );
}
// ParallelProcessChunks lets you specify a minimum # of items to process per job. Use this when
// you may have a large set of work items which only take a small amount of time per item, and so
// need to reduce dispatch overhead.
template <typename ITEM_TYPE>
inline void ParallelProcessChunks( ITEM_TYPE *pItems, unsigned nItems, void (*pfnProcess)( ITEM_TYPE & ), int nChunkSize, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CFuncJobItemProcessor<ITEM_TYPE> > processor;
processor.m_ItemProcessor.Init( pfnProcess, NULL, NULL );
processor.Run( pItems, nItems, nChunkSize, nMaxParallel );
}
template <typename ITEM_TYPE, typename OBJECT_TYPE, typename FUNCTION_CLASS >
inline void ParallelProcessChunks( ITEM_TYPE *pItems, unsigned nItems, OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( ITEM_TYPE & ), int nChunkSize, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CMemberFuncJobItemProcessor<ITEM_TYPE, OBJECT_TYPE, FUNCTION_CLASS> > processor;
processor.m_ItemProcessor.Init( pObject, pfnProcess, NULL, NULL );
processor.Run( pItems, nItems, nChunkSize, nMaxParallel );
}
template <typename ITEM_TYPE, typename OBJECT_TYPE, typename FUNCTION_CLASS >
inline void ParallelProcessChunks( IThreadPool *pPool, ITEM_TYPE *pItems, unsigned nItems, OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( ITEM_TYPE & ), int nChunkSize, int nMaxParallel = INT_MAX )
{
CParallelProcessor<ITEM_TYPE, CMemberFuncJobItemProcessor<ITEM_TYPE, OBJECT_TYPE, FUNCTION_CLASS> > processor;
processor.m_ItemProcessor.Init( pObject, pfnProcess, NULL, NULL );
processor.Run( pItems, nItems, nChunkSize, nMaxParallel, pPool );
}
template <class CONTEXT_TYPE, class ITEM_PROCESSOR_TYPE>
class CParallelLoopProcessor
{
public:
CParallelLoopProcessor()
{
m_nIndex = m_nLimit = 0;
m_nChunkCount = 0;
m_nActive = 0;
}
void Run( CONTEXT_TYPE *pContext, int nBegin, int nItems, int nChunkCount, int nMaxParallel = INT_MAX, IThreadPool *pThreadPool = NULL )
{
if ( !nItems )
return;
if ( !pThreadPool )
{
pThreadPool = g_pThreadPool;
}
m_pContext = pContext;
m_nIndex = nBegin;
m_nLimit = nBegin + nItems;
nChunkCount = MAX( MIN( nItems, nChunkCount ), 1 );
m_nChunkCount = ( nItems + nChunkCount - 1 ) / nChunkCount;
int nJobs = ( nItems + m_nChunkCount - 1 ) / m_nChunkCount;
if ( nJobs > nMaxParallel )
{
nJobs = nMaxParallel;
}
if ( !pThreadPool ) // only possible on linux
{
DoExecute( );
return;
}
int nThreads = pThreadPool->NumThreads();
if ( nJobs > nThreads )
{
nJobs = nThreads;
}
if ( nJobs > 0 )
{
CJob **jobs = (CJob **)stackalloc( nJobs * sizeof(CJob **) );
int i = nJobs;
while( i-- )
{
jobs[i] = pThreadPool->QueueCall( this, &CParallelLoopProcessor<CONTEXT_TYPE, ITEM_PROCESSOR_TYPE>::DoExecute );
}
DoExecute();
for ( i = 0; i < nJobs; i++ )
{
jobs[i]->Abort(); // will either abort ones that never got a thread, or noop on ones that did
jobs[i]->Release();
}
}
else
{
DoExecute();
}
}
ITEM_PROCESSOR_TYPE m_ItemProcessor;
private:
void DoExecute()
{
m_ItemProcessor.Begin();
for (;;)
{
int nIndex = m_nIndex.AtomicAdd( m_nChunkCount );
if ( nIndex < m_nLimit )
{
int nCount = MIN( m_nChunkCount, m_nLimit - nIndex );
m_ItemProcessor.Process( m_pContext, nIndex, nCount );
}
else
{
break;
}
}
m_ItemProcessor.End();
--m_nActive;
}
CONTEXT_TYPE *m_pContext;
CInterlockedInt m_nIndex;
int m_nLimit;
int m_nChunkCount;
CInterlockedInt m_nActive;
};
template < typename CONTEXT_TYPE >
inline void ParallelLoopProcess( IThreadPool *pPool, CONTEXT_TYPE *pContext, int nStart, int nCount, void (*pfnProcess)( CONTEXT_TYPE*, int, int ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelLoopProcessor< CONTEXT_TYPE, CLoopFuncJobItemProcessor< CONTEXT_TYPE > > processor;
processor.m_ItemProcessor.Init( pfnProcess, pfnBegin, pfnEnd );
processor.Run( pContext, nStart, nCount, 1, nMaxParallel, pPool );
}
template < typename CONTEXT_TYPE, typename OBJECT_TYPE, typename FUNCTION_CLASS >
inline void ParallelLoopProcess( IThreadPool *pPool, CONTEXT_TYPE *pContext, int nStart, int nCount, OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( CONTEXT_TYPE*, int, int ), void (FUNCTION_CLASS::*pfnBegin)() = NULL, void (FUNCTION_CLASS::*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelLoopProcessor< CONTEXT_TYPE, CLoopMemberFuncJobItemProcessor<CONTEXT_TYPE, OBJECT_TYPE, FUNCTION_CLASS> > processor;
processor.m_ItemProcessor.Init( pObject, pfnProcess, pfnBegin, pfnEnd );
processor.Run( pContext, nStart, nCount, 1, nMaxParallel, pPool );
}
template < typename CONTEXT_TYPE >
inline void ParallelLoopProcessChunks( IThreadPool *pPool, CONTEXT_TYPE *pContext, int nStart, int nCount, int nChunkSize, void (*pfnProcess)( CONTEXT_TYPE*, int, int ), void (*pfnBegin)() = NULL, void (*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelLoopProcessor< CONTEXT_TYPE, CLoopFuncJobItemProcessor< CONTEXT_TYPE > > processor;
processor.m_ItemProcessor.Init( pfnProcess, pfnBegin, pfnEnd );
processor.Run( pContext, nStart, nCount, nChunkSize, nMaxParallel, pPool );
}
template < typename CONTEXT_TYPE, typename OBJECT_TYPE, typename FUNCTION_CLASS >
inline void ParallelLoopProcessChunks( IThreadPool *pPool, CONTEXT_TYPE *pContext, int nStart, int nCount, int nChunkSize, OBJECT_TYPE *pObject, void (FUNCTION_CLASS::*pfnProcess)( CONTEXT_TYPE*, int, int ), void (FUNCTION_CLASS::*pfnBegin)() = NULL, void (FUNCTION_CLASS::*pfnEnd)() = NULL, int nMaxParallel = INT_MAX )
{
CParallelLoopProcessor< CONTEXT_TYPE, CLoopMemberFuncJobItemProcessor<CONTEXT_TYPE, OBJECT_TYPE, FUNCTION_CLASS> > processor;
processor.m_ItemProcessor.Init( pObject, pfnProcess, pfnBegin, pfnEnd );
processor.Run( pContext, nStart, nCount, nChunkSize, nMaxParallel, pPool );
}
template <class Derived>
class CParallelProcessorBase
{
protected:
typedef CParallelProcessorBase<Derived> ThisParallelProcessorBase_t;
typedef Derived ThisParallelProcessorDerived_t;
public:
CParallelProcessorBase()
{
m_nActive = 0;
}
protected:
void Run( int nMaxParallel = INT_MAX, int threadOverride = -1 )
{
int i = g_pThreadPool->NumIdleThreads();
if ( nMaxParallel < i)
{
i = nMaxParallel;
}
while( i -- > 0 )
{
if ( threadOverride == -1 || i == threadOverride - 1 )
{
++ m_nActive;
ThreadExecute( this, &ThisParallelProcessorBase_t::DoExecute )->Release();
}
}
if ( threadOverride == -1 || threadOverride == 0 )
{
++ m_nActive;
DoExecute();
}
while ( m_nActive )
{
ThreadPause();
}
}
protected:
void OnBegin() {}
bool OnProcess() { return false; }
void OnEnd() {}
private:
void DoExecute()
{
static_cast<Derived *>( this )->OnBegin();
while ( static_cast<Derived *>( this )->OnProcess() )
continue;
static_cast<Derived *>(this)->OnEnd();
-- m_nActive;
}
CInterlockedInt m_nActive;
};
//-----------------------------------------------------------------------------
// Raw thread launching
//-----------------------------------------------------------------------------
inline uintp FunctorExecuteThread( void *pParam )
{
CFunctor *pFunctor = (CFunctor *)pParam;
(*pFunctor)();
pFunctor->Release();
return 0;
}
inline ThreadHandle_t ThreadExecuteSoloImpl( CFunctor *pFunctor, const char *pszName = NULL )
{
ThreadHandle_t hThread;
hThread = CreateSimpleThread( FunctorExecuteThread, pFunctor );
if ( pszName )
{
ThreadSetDebugName( hThread, pszName );
}
return hThread;
}
inline ThreadHandle_t ThreadExecuteSolo( CJob *pJob ) { return ThreadExecuteSoloImpl( CreateFunctor( pJob, &CJob::Execute ), pJob->Describe() ); }
template <typename T1>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1 ), pszName ); }
template <typename T1, typename T2>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2 ), pszName ); }
template <typename T1, typename T2, typename T3>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2, T3 a3 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2, a3 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2, a3, a4 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2, a3, a4, a5 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5, T6 a6 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2, a3, a4, a5, a6 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5, T6 a6, T7 a7 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2, a3, a4, a5, a6, a7 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
inline ThreadHandle_t ThreadExecuteSolo( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5, T6 a6, T7 a7, T8 a8 ) { return ThreadExecuteSoloImpl( CreateFunctor( a1, a2, a3, a4, a5, a6, a7, a8 ), pszName ); }
template <typename T1, typename T2>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2 ), pszName ); }
template <typename T1, typename T2, typename T3>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2, T3 a3 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2, a3 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2, a3, a4 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2, a3, a4, a5 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5, T6 a6 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2, a3, a4, a5, a6 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5, T6 a6, T7 a7 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2, a3, a4, a5, a6, a7 ), pszName ); }
template <typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
inline ThreadHandle_t ThreadExecuteSoloRef( const char *pszName, T1 a1, T2 a2, T3 a3, T4 a4, T5 a5, T6 a6, T7 a7, T8 a8 ) { return ThreadExecuteSoloImpl( CreateRefCountingFunctor(a1, a2, a3, a4, a5, a6, a7, a8 ), pszName ); }
//-----------------------------------------------------------------------------
inline bool IThreadPool::YieldWait( CThreadEvent &theEvent, unsigned timeout )
{
CThreadEvent *pEvent = &theEvent;
return ( YieldWait( &pEvent, 1, true, timeout ) != TW_TIMEOUT );
}
inline bool IThreadPool::YieldWait( CJob *pJob, unsigned timeout )
{
return ( YieldWait( &pJob, 1, true, timeout ) != TW_TIMEOUT );
}
//-----------------------------------------------------------------------------
inline JobStatus_t CJob::Execute()
{
if ( IsFinished() )
{
return m_status;
}
AUTO_LOCK( m_mutex );
AddRef();
JobStatus_t result;
switch ( m_status )
{
case JOB_STATUS_UNSERVICED:
case JOB_STATUS_PENDING:
{
// Service it
m_status = JOB_STATUS_INPROGRESS;
#if defined( THREAD_PARENT_STACK_TRACE_ENABLED )
//replace thread parent trace with job parent
{
CStackTop_ReferenceParentStack stackTop( m_ParentStackTrace, ARRAYSIZE( m_ParentStackTrace ) );
result = m_status = DoExecute();
}
#else
result = m_status = DoExecute();
#endif
DoCleanup();
m_CompleteEvent.Set();
break;
}
case JOB_STATUS_INPROGRESS:
AssertMsg(0, "Mutex Should have protected use while processing");
// fall through...
case JOB_OK:
case JOB_STATUS_ABORTED:
result = m_status;
break;
default:
AssertMsg( m_status < JOB_OK, "Unknown job state");
result = m_status;
}
Release();
return result;
}
//---------------------------------------------------------
inline JobStatus_t CJob::TryExecute()
{
// TryLock() would only fail if another thread has entered
// Execute() or Abort()
if ( !IsFinished() && TryLock() )
{
// ...service the request
Execute();
Unlock();
}
return m_status;
}
//---------------------------------------------------------
inline JobStatus_t CJob::Abort( bool bDiscard )
{
if ( IsFinished() )
{
return m_status;
}
AUTO_LOCK( m_mutex );
AddRef();
JobStatus_t result;
switch ( m_status )
{
case JOB_STATUS_UNSERVICED:
case JOB_STATUS_PENDING:
{
result = m_status = DoAbort( bDiscard );
if ( bDiscard )
DoCleanup();
m_CompleteEvent.Set();
}
break;
case JOB_STATUS_ABORTED:
case JOB_STATUS_INPROGRESS:
case JOB_OK:
result = m_status;
break;
default:
AssertMsg( m_status < JOB_OK, "Unknown job state");
result = m_status;
}
Release();
return result;
}
//-----------------------------------------------------------------------------
#endif // JOBTHREAD_H