分类: C/C++
2010-07-03 13:30:25
Abstract
In shared memory multiprocessor architectures, such as SMPs, threads can be used to implement parallelism. Historically, hardware vendors have implemented their own proprietary versions of threads, making portability a concern for software developers. For UNIX systems, a standardized C language threads programming interface has been specified by the IEEE POSIX 1003.1c standard. Implementations that adhere to this standard are referred to as POSIX threads, or Pthreads.
The tutorial begins with an introduction to concepts, motivations, and design considerations for using Pthreads. Each of the three major classes of routines in the Pthreads API are then covered: Thread Management, Mutex Variables, and Condition Variables. Example codes are used throughout to demonstrate how to use most of the Pthreads routines needed by a new Pthreads programmer. The tutorial concludes with a discussion of LLNL specifics and how to mix MPI with pthreads. A lab exercise, with numerous example codes (C Language) is also included.
Level/Prerequisites: Ideal for those who are new to parallel programming with threads. A basic understanding of parallel programming in C is assumed. For those who are unfamiliar with Parallel Programming in general, the material covered in would be helpful.
Pthreads Overview
UNIX PROCESS
THREADS WITHIN A UNIX PROCESS
Pthreads Overview
Pthreads Overview
For example, the following table compares timing results for the fork() subroutine and the pthreads_create() subroutine. Timings reflect 50,000 process/thread creations, were performed with the time utility, and units are in seconds, no optimization flags.
Note: don't expect the sytem and user times to add up to real time, because these are SMP systems with multiple CPUs working on the problem at the same time. At best, these are approximations run on local machines, past and present.
Platform
fork()
pthread_create()
real
user
sys
real
user
sys
AMD 2.3 GHz Opteron (16cpus/node)
12.5
1.0
12.5
1.2
0.2
1.3
AMD 2.4 GHz Opteron (8cpus/node)
17.6
2.2
15.7
1.4
0.3
1.3
IBM 4.0 GHz POWER6 (8cpus/node)
9.5
0.6
8.8
1.6
0.1
0.4
IBM 1.9 GHz POWER5 p5-575 (8cpus/node)
64.2
30.7
27.6
1.7
0.6
1.1
IBM 1.5 GHz POWER4 (8cpus/node)
104.5
48.6
47.2
2.1
1.0
1.5
INTEL 2.4 GHz Xeon (2 cpus/node)
54.9
1.5
20.8
1.6
0.7
0.9
INTEL 1.4 GHz Itanium2 (4 cpus/node)
54.5
1.1
22.2
2.0
1.2
0.6
Platform
MPI Shared Memory Bandwidth
(GB/sec)
Pthreads Worst Case
Memory-to-CPU Bandwidth
(GB/sec)
AMD 2.3 GHz Opteron
1.8
5.3
AMD 2.4 GHz Opteron
1.2
5.3
IBM 1.9 GHz POWER5 p5-575
4.1
16
IBM 1.5 GHz POWER4
2.1
4
Intel 2.4 GHz Xeon
0.3
4.3
Intel 1.4 GHz Itanium 2
1.8
6.4
Pthreads Overview
Shared Memory Model:
Thread-safeness:
The Pthreads API
Routine Prefix
Functional Group
pthread_
Threads themselves and miscellaneous subroutines
pthread_attr_
Thread attributes objects
pthread_mutex_
Mutexes
pthread_mutexattr_
Mutex attributes objects.
pthread_cond_
Condition variables
pthread_condattr_
Condition attributes objects
pthread_key_
Thread-specific data keys
pthread_rwlock_
Read/write locks
pthread_barrier_
Synchronization barriers
Compiling Threaded Programs
Compiler / Platform
Compiler Command
Description
IBM
AIX
xlc_r / cc_r
C (ANSI / non-ANSI)
xlC_r
C++
xlf_r -qnosave
xlf90_r -qnosave
Fortran - using IBM's Pthreads API (non-portable)
INTEL
Linux
icc -pthread
C
icpc -pthread
C++
PathScale
Linux
pathcc -pthread
C
pathCC -pthread
C++
PGI
Linux
pgcc -lpthread
C
pgCC -lpthread
C++
GNU
Linux, AIX
gcc -pthread
GNU C
g++ -pthread
GNU C++
Thread Management
(thread,attr,start_routine,arg)
(status)
(attr)
(attr)
Creating Threads:
Question: After a thread has been created, how do you know when it will be scheduled to run by the operating system?
Thread Attributes:
Terminating Threads:
Example Code - Pthread Creation and Termination
#include#include #define NUM_THREADS 5 void *PrintHello(void *threadid) { long tid; tid = (long)threadid; printf("Hello World! It's me, thread #%ld!\n", tid); pthread_exit(NULL); } int main (int argc, char *argv[]) { pthread_t threads[NUM_THREADS]; int rc; long t; for(t=0; t
Thread Management
Question: How can you safely pass data to newly created threads, given their non-deterministic start-up and scheduling?
Example 1 - Thread Argument Passing
long *taskids[NUM_THREADS]; for(t=0; tExample 2 - Thread Argument Passing
This example shows how to setup/pass multiple arguments via a structure. Each thread receives a unique instance of the structure.
struct thread_data{ int thread_id; int sum; char *message; }; struct thread_data thread_data_array[NUM_THREADS]; void *PrintHello(void *threadarg) { struct thread_data *my_data; ... my_data = (struct thread_data *) threadarg; taskid = my_data->thread_id; sum = my_data->sum; hello_msg = my_data->message; ... } int main (int argc, char *argv[]) { ... thread_data_array[t].thread_id = t; thread_data_array[t].sum = sum; thread_data_array[t].message = messages[t]; rc = pthread_create(&threads[t], NULL, PrintHello, (void *) &thread_data_array[t]); ... }Example 3 - Thread Argument Passing (Incorrect)
This example performs argument passing incorrectly. It passes the address of variable t, which is shared memory space and visible to all threads. As the loop iterates, the value of this memory location changes, possibly before the created threads can access it.
int rc; long t; for(t=0; tThread Management
Joining and Detaching Threads
Routines:
(threadid,status)
(threadid)
(attr,detachstate)
(attr,detachstate)
Joining:
- "Joining" is one way to accomplish synchronization between threads. For example:
- The pthread_join() subroutine blocks the calling thread until the specified threadid thread terminates.
- The programmer is able to obtain the target thread's termination return status if it was specified in the target thread's call to pthread_exit().
- A joining thread can match one pthread_join() call. It is a logical error to attempt multiple joins on the same thread.
- Two other synchronization methods, mutexes and condition variables, will be discussed later.
Joinable or Not?
Detaching:
- When a thread is created, one of its attributes defines whether it is joinable or detached. Only threads that are created as joinable can be joined. If a thread is created as detached, it can never be joined.
- The final draft of the POSIX standard specifies that threads should be created as joinable.
- To explicitly create a thread as joinable or detached, the attr argument in the pthread_create() routine is used. The typical 4 step process is:
- Declare a pthread attribute variable of the pthread_attr_t data type
- Initialize the attribute variable with pthread_attr_init()
- Set the attribute detached status with pthread_attr_setdetachstate()
- When done, free library resources used by the attribute with pthread_attr_destroy()
- The pthread_detach() routine can be used to explicitly detach a thread even though it was created as joinable.
- There is no converse routine.
Recommendations:
- If a thread requires joining, consider explicitly creating it as joinable. This provides portability as not all implementations may create threads as joinable by default.
- If you know in advance that a thread will never need to join with another thread, consider creating it in a detached state. Some system resources may be able to be freed.
Example: Pthread Joining
Example Code - Pthread Joining
This example demonstrates how to "wait" for thread completions by using the Pthread join routine. Since some implementations of Pthreads may not create threads in a joinable state, the threads in this example are explicitly created in a joinable state so that they can be joined later.
#include#include #include #define NUM_THREADS 4 void *BusyWork(void *t) { int i; long tid; double result=0.0; tid = (long)t; printf("Thread %ld starting...\n",tid); for (i=0; i<1000000; i++) { result = result + sin(i) * tan(i); } printf("Thread %ld done. Result = %e\n",tid, result); pthread_exit((void*) t); } int main (int argc, char *argv[]) { pthread_t thread[NUM_THREADS]; pthread_attr_t attr; int rc; long t; void *status; /* Initialize and set thread detached attribute */ pthread_attr_init(&attr); pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE); for(t=0; t Thread Management
Stack Management
Routines:
(attr, stacksize)
(attr, stacksize)
(attr, stackaddr)
(attr, stackaddr)
Preventing Stack Problems:
- The POSIX standard does not dictate the size of a thread's stack. This is implementation dependent and varies.
- Exceeding the default stack limit is often very easy to do, with the usual results: program termination and/or corrupted data.
- Safe and portable programs do not depend upon the default stack limit, but instead, explicitly allocate enough stack for each thread by using the pthread_attr_setstacksize routine.
- The pthread_attr_getstackaddr and pthread_attr_setstackaddr routines can be used by applications in an environment where the stack for a thread must be placed in some particular region of memory.
Some Practical Examples at LC:
- Default thread stack size varies greatly. The maximum size that can be obtained also varies greatly, and may depend upon the number of threads per node.
Node
Architecture
#CPUs
Memory (GB)
Default Size
(bytes)AMD Opteron
8
16
2,097,152Intel IA64
4
8
33,554,432Intel IA32
2
4
2,097,152IBM Power5
8
32
196,608IBM Power4
8
16
196,608IBM Power3
16
16
98,304
Example: Stack Management
Example Code - Stack Management
This example demonstrates how to query and set a thread's stack size.
#include#include #define NTHREADS 4 #define N 1000 #define MEGEXTRA 1000000 pthread_attr_t attr; void *dowork(void *threadid) { double A[N][N]; int i,j; long tid; size_t mystacksize; tid = (long)threadid; pthread_attr_getstacksize (&attr, &mystacksize); printf("Thread %ld: stack size = %li bytes \n", tid, mystacksize); for (i=0; i Thread Management
Miscellaneous Routines
()
(thread1,thread2)
- pthread_self returns the unique, system assigned thread ID of the calling thread.
- pthread_equal compares two thread IDs. If the two IDs are different 0 is returned, otherwise a non-zero value is returned.
- Note that for both of these routines, the thread identifier objects are opaque and can not be easily inspected. Because thread IDs are opaque objects, the C language equivalence operator == should not be used to compare two thread IDs against each other, or to compare a single thread ID against another value.
(once_control, init_routine)
- pthread_once executes the init_routine exactly once in a process. The first call to this routine by any thread in the process executes the given init_routine, without parameters. Any subsequent call will have no effect.
- The init_routine routine is typically an initialization routine.
- The once_control parameter is a synchronization control structure that requires initialization prior to calling pthread_once. For example:
pthread_once_t once_control = PTHREAD_ONCE_INIT;
Mutex Variables
Overview
- Mutex is an abbreviation for "mutual exclusion". Mutex variables are one of the primary means of implementing thread synchronization and for protecting shared data when multiple writes occur.
- A mutex variable acts like a "lock" protecting access to a shared data resource. The basic concept of a mutex as used in Pthreads is that only one thread can lock (or own) a mutex variable at any given time. Thus, even if several threads try to lock a mutex only one thread will be successful. No other thread can own that mutex until the owning thread unlocks that mutex. Threads must "take turns" accessing protected data.
- Mutexes can be used to prevent "race" conditions. An example of a race condition involving a bank transaction is shown below:
Thread 1
Thread 2
BalanceRead balance: $1000
$1000Read balance: $1000
$1000Deposit $200
$1000Deposit $200
$1000Update balance $1000+$200
$1200Update balance $1000+$200
$1200- In the above example, a mutex should be used to lock the "Balance" while a thread is using this shared data resource.
- Very often the action performed by a thread owning a mutex is the updating of global variables. This is a safe way to ensure that when several threads update the same variable, the final value is the same as what it would be if only one thread performed the update. The variables being updated belong to a "critical section".
- A typical sequence in the use of a mutex is as follows:
- Create and initialize a mutex variable
- Several threads attempt to lock the mutex
- Only one succeeds and that thread owns the mutex
- The owner thread performs some set of actions
- The owner unlocks the mutex
- Another thread acquires the mutex and repeats the process
- Finally the mutex is destroyed
- When several threads compete for a mutex, the losers block at that call - an unblocking call is available with "trylock" instead of the "lock" call.
- When protecting shared data, it is the programmer's responsibility to make sure every thread that needs to use a mutex does so. For example, if 4 threads are updating the same data, but only one uses a mutex, the data can still be corrupted.
Mutex Variables
Creating and Destroying Mutexes
Routines:
(mutex,attr)
(mutex)
(attr)
(attr)
Usage:
- Mutex variables must be declared with type pthread_mutex_t, and must be initialized before they can be used. There are two ways to initialize a mutex variable:
- Statically, when it is declared. For example:
pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;- Dynamically, with the pthread_mutex_init() routine. This method permits setting mutex object attributes, attr.
The mutex is initially unlocked.
- The attr object is used to establish properties for the mutex object, and must be of type pthread_mutexattr_t if used (may be specified as NULL to accept defaults). The Pthreads standard defines three optional mutex attributes:
- Protocol: Specifies the protocol used to prevent priority inversions for a mutex.
- Prioceiling: Specifies the priority ceiling of a mutex.
- Process-shared: Specifies the process sharing of a mutex.
Note that not all implementations may provide the three optional mutex attributes.
- The pthread_mutexattr_init() and pthread_mutexattr_destroy() routines are used to create and destroy mutex attribute objects respectively.
- pthread_mutex_destroy() should be used to free a mutex object which is no longer needed.
Mutex Variables
Locking and Unlocking Mutexes
Routines:
(mutex)
(mutex)
(mutex)
Usage:
- The pthread_mutex_lock() routine is used by a thread to acquire a lock on the specified mutex variable. If the mutex is already locked by another thread, this call will block the calling thread until the mutex is unlocked.
- pthread_mutex_trylock() will attempt to lock a mutex. However, if the mutex is already locked, the routine will return immediately with a "busy" error code. This routine may be useful in preventing deadlock conditions, as in a priority-inversion situation.
- pthread_mutex_unlock() will unlock a mutex if called by the owning thread. Calling this routine is required after a thread has completed its use of protected data if other threads are to acquire the mutex for their work with the protected data. An error will be returned if:
- If the mutex was already unlocked
- If the mutex is owned by another thread
- There is nothing "magical" about mutexes...in fact they are akin to a "gentlemen's agreement" between participating threads. It is up to the code writer to insure that the necessary threads all make the the mutex lock and unlock calls correctly. The following scenario demonstrates a logical error:
Thread 1 Thread 2 Thread 3 Lock Lock A = 2 A = A+1 A = A*B Unlock Unlock
Question: When more than one thread is waiting for a locked mutex, which thread will be granted the lock first after it is released?
Example: Using Mutexes
Example Code - Using Mutexes
This example program illustrates the use of mutex variables in a threads program that performs a dot product. The main data is made available to all threads through a globally accessible structure. Each thread works on a different part of the data. The main thread waits for all the threads to complete their computations, and then it prints the resulting sum.
#include#include #include /* The following structure contains the necessary information to allow the function "dotprod" to access its input data and place its output into the structure. */ typedef struct { double *a; double *b; double sum; int veclen; } DOTDATA; /* Define globally accessible variables and a mutex */ #define NUMTHRDS 4 #define VECLEN 100 DOTDATA dotstr; pthread_t callThd[NUMTHRDS]; pthread_mutex_t mutexsum; /* The function dotprod is activated when the thread is created. All input to this routine is obtained from a structure of type DOTDATA and all output from this function is written into this structure. The benefit of this approach is apparent for the multi-threaded program: when a thread is created we pass a single argument to the activated function - typically this argument is a thread number. All the other information required by the function is accessed from the globally accessible structure. */ void *dotprod(void *arg) { /* Define and use local variables for convenience */ int i, start, end, len ; long offset; double mysum, *x, *y; offset = (long)arg; len = dotstr.veclen; start = offset*len; end = start + len; x = dotstr.a; y = dotstr.b; /* Perform the dot product and assign result to the appropriate variable in the structure. */ mysum = 0; for (i=start; i Serial version
Pthreads versionCondition Variables
Overview
- Condition variables provide yet another way for threads to synchronize. While mutexes implement synchronization by controlling thread access to data, condition variables allow threads to synchronize based upon the actual value of data.
- Without condition variables, the programmer would need to have threads continually polling (possibly in a critical section), to check if the condition is met. This can be very resource consuming since the thread would be continuously busy in this activity. A condition variable is a way to achieve the same goal without polling.
- A condition variable is always used in conjunction with a mutex lock.
- A representative sequence for using condition variables is shown below.
Main Thread
- Declare and initialize global data/variables which require synchronization (such as "count")
- Declare and initialize a condition variable object
- Declare and initialize an associated mutex
- Create threads A and B to do work
Thread A
- Do work up to the point where a certain condition must occur (such as "count" must reach a specified value)
- Lock associated mutex and check value of a global variable
- Call pthread_cond_wait() to perform a blocking wait for signal from Thread-B. Note that a call to pthread_cond_wait() automatically and atomically unlocks the associated mutex variable so that it can be used by Thread-B.
- When signalled, wake up. Mutex is automatically and atomically locked.
- Explicitly unlock mutex
- Continue
Thread B
- Do work
- Lock associated mutex
- Change the value of the global variable that Thread-A is waiting upon.
- Check value of the global Thread-A wait variable. If it fulfills the desired condition, signal Thread-A.
- Unlock mutex.
- Continue
Main Thread
Join / Continue
Condition Variables
Creating and Destroying Condition Variables
Routines:
(condition,attr)
(condition)
(attr)
(attr)
Usage:
- Condition variables must be declared with type pthread_cond_t, and must be initialized before they can be used. There are two ways to initialize a condition variable:
- Statically, when it is declared. For example:
pthread_cond_t myconvar = PTHREAD_COND_INITIALIZER;- Dynamically, with the pthread_cond_init() routine. The ID of the created condition variable is returned to the calling thread through the condition parameter. This method permits setting condition variable object attributes, attr.
- The optional attr object is used to set condition variable attributes. There is only one attribute defined for condition variables: process-shared, which allows the condition variable to be seen by threads in other processes. The attribute object, if used, must be of type pthread_condattr_t (may be specified as NULL to accept defaults).
Note that not all implementations may provide the process-shared attribute.
- The pthread_condattr_init() and pthread_condattr_destroy() routines are used to create and destroy condition variable attribute objects.
- pthread_cond_destroy() should be used to free a condition variable that is no longer needed.
Condition Variables
Waiting and Signaling on Condition Variables
Routines:
(condition,mutex)
(condition)
(condition)
Usage:
- pthread_cond_wait() blocks the calling thread until the specified condition is signalled. This routine should be called while mutex is locked, and it will automatically release the mutex while it waits. After signal is received and thread is awakened, mutex will be automatically locked for use by the thread. The programmer is then responsible for unlocking mutex when the thread is finished with it.
- The pthread_cond_signal() routine is used to signal (or wake up) another thread which is waiting on the condition variable. It should be called after mutex is locked, and must unlock mutex in order for pthread_cond_wait() routine to complete.
- The pthread_cond_broadcast() routine should be used instead of pthread_cond_signal() if more than one thread is in a blocking wait state.
- It is a logical error to call pthread_cond_signal() before calling pthread_cond_wait().
Proper locking and unlocking of the associated mutex variable is essential when using these routines. For example:
- Failing to lock the mutex before calling pthread_cond_wait() may cause it NOT to block.
- Failing to unlock the mutex after calling pthread_cond_signal() may not allow a matching pthread_cond_wait() routine to complete (it will remain blocked).
Example: Using Condition Variables
Example Code - Using Condition Variables
This simple example code demonstrates the use of several Pthread condition variable routines. The main routine creates three threads. Two of the threads perform work and update a "count" variable. The third thread waits until the count variable reaches a specified value.
#include#include #include #define NUM_THREADS 3 #define TCOUNT 10 #define COUNT_LIMIT 12 int count = 0; int thread_ids[3] = {0,1,2}; pthread_mutex_t count_mutex; pthread_cond_t count_threshold_cv; void *inc_count(void *t) { int i; long my_id = (long)t; for (i=0; i LLNL Specific Information and Recommendations
This section describes details specific to Livermore Computing's systems.
Implementations:
- All LC production systems include a Pthreads implementation that follows draft 10 (final) of the POSIX standard. This is the preferred implementation.
- Implementations differ in the maximum number of threads that a process may create. They also differ in the default amount of thread stack space.
Compiling:
- LC maintains a number of compilers, and usually several different versions of each - see the web page.
- The compiler commands described in the section apply to LC systems.
- Additionally, all LC IBM compilers are aliased to their thread-safe command. For example, xlc really uses xlc_r. This is only true for LC IBM systems.
Mixing MPI with Pthreads:
- Programs that contain both MPI and Pthreads are common and easy to develop on all LC systems.
- Design:
- Each MPI process typically creates and then manages N threads, where N makes the best use of the available CPUs/node.
- Finding the best value for N will vary with the platform and your application's characteristics.
- For IBM SP systems with two communication adapters per node, it may prove more efficient to use two (or more) MPI tasks per node.
- In general, there may be problems if multiple threads make MPI calls. The program may fail or behave unexpectedly. If MPI calls must be made from within a thread, they should be made only by one thread.
- Compiling:
- Use the appropriate MPI compile command for the platform and language of choice
- Be sure to include the required flag as in the table above (-pthread or -qnosave)
- MPICH is not thread safe
- An example code that uses both MPI and Pthreads is available below. The serial, threads-only, MPI-only and MPI-with-threads versions demonstrate one possible progression.
- (for IBM SP)
Topics Not Covered
Several features of the Pthreads API are not covered in this tutorial. These are listed below. See the section for more information.
- Thread Scheduling
- Implementations will differ on how threads are scheduled to run. In most cases, the default mechanism is adequate.
- The Pthreads API provides routines to explicitly set thread scheduling policies and priorities which may override the default mechanisms.
- The API does not require implementations to support these features.
- Keys: Thread-Specific Data
- As threads call and return from different routines, the local data on a thread's stack comes and goes.
- To preserve stack data you can usually pass it as an argument from one routine to the next, or else store the data in a global variable associated with a thread.
- Pthreads provides another, possibly more convenient and versatile, way of accomplishing this through keys.
- Mutex Protocol Attributes and Mutex Priority Management for the handling of "priority inversion" problems.
- Condition Variable Sharing - across processes
- Thread Cancellation
- Threads and Signals
- Synchronization constructs - barriers and locks
Pthread Library Routines Reference
For convenience, an alphabetical list of Pthread routines, linked to their corresponding man page, is provided below.
This completes the tutorial.
Please complete the online evaluation form - unless you are doing the exercise, in which case please complete it at the end of the exercise.