Intel® Moderncode for Parallel Architectures
Support for developing parallel programming applications on Intel® Architecture.

Cannot hold locks at a thread boundary

pkroy
Beginner
163 Views
Hi all,


I am using cilk++ in my application. While using cilkscreen i get the following message :
Cannot hold locks at a thread boundary

What does it mean ???

I am doing something like this :

void my_Recursive_fn()
{

if ( base_case){
work0
return;
}
else
{
work1
spawn my_Recursive_fn();
my_Recursive_fn();
sync;
}
return;
}



0 Kudos
3 Replies
William_Leiserson__I
163 Views
Quoting - pkroy
Hi all,


I am using cilk++ in my application. While using cilkscreen i get the following message :
Cannot hold locks at a thread boundary

What does it mean ???

I am doing something like this :

void my_Recursive_fn()
{

if ( base_case){
work0
return;
}
else
{
work1
spawn my_Recursive_fn();
my_Recursive_fn();
sync;
}
return;
}




Hi Pkroy,

This error is an indication that you are holding a lock across a spawn or sync. Cilkscreen complains about this for a couple of reasons:

1. The thread that is running just after a spawn or sync statement is not necessarily the same one that was running beforehand. This means that the acquire() and release() functions may be called on different threads, which is not permitted on certain kinds of locks on Windows or on pthread mutexes.

2. You can deadlock, inadvertantly. Consider the following example:

void f (lock) {
acquire(lock);
do_work();
release(lock);
}

void g (lock) {
spawn f();
acquire(lock);
do_work();
sync;
release(lock);
}

In this case, if the acquire() in g() is executed before the acquire() in f(), then the program will deadlock because the release() in g() cannot be executed until after f() has returned and g() has been able to sync. But f() will never return.

---

For these reasons, cilkscreen will generate an error if you are holding a lock across a spawn or sync boundary.

-wml
pkroy
Beginner
163 Views
Hi wml,
Thanks for an illustrative example ...
pkroy
Hi Pkroy,

This error is an indication that you are holding a lock across a spawn or sync. Cilkscreen complains about this for a couple of reasons:

1. The thread that is running just after a spawn or sync statement is not necessarily the same one that was running beforehand. This means that the acquire() and release() functions may be called on different threads, which is not permitted on certain kinds of locks on Windows or on pthread mutexes.

2. You can deadlock, inadvertantly. Consider the following example:

void f (lock) {
acquire(lock);
do_work();
release(lock);
}

void g (lock) {
spawn f();
acquire(lock);
do_work();
sync;
release(lock);
}

In this case, if the acquire() in g() is executed before the acquire() in f(), then the program will deadlock because the release() in g() cannot be executed until after f() has returned and g() has been able to sync. But f() will never return.

---

For these reasons, cilkscreen will generate an error if you are holding a lock across a spawn or sync boundary.

-wml

William_Leiserson__I
163 Views
No problem. :)
Reply