Cilk assertion failed: *h

[Ruben_Perez]

Hello,

I am observing the following assertion, and am unsure of its meaning

sched.c:643: cilk assertion failed: *h

Though I have not replicated it in a smaller test case, what I do is define a lambda function, and then run through a cilk for loop which calls another function with the lambda function as an argument. The called function does call the lambda itself, but the error occurs when the called function is returning. I am using icpc 12.0.2 20110112. Once I simplify my code down to a postable size which demonstrates the issue, I'll post that, but in the meantime any advice would be great.

Thanks,

Ruben

[Ruben_Perez]

After trying to make a smaller test case, I get the following error instead:

W12: frame won undo-detach race with flags 8001

It occurs when a a cilk_spawn my_lamba_func call is included(though the error does not occur the first time it is executed) The test code is attached at the end of this pose

To give a bit more background on what I am doing, I am attempting to use lambda's to replace the old cilk5 inlets in an parallel alpha beta search program. Since this would probably be used in game scenarios where time is of the essence, I use progressive deepening to search the game tree to depth 0,1 ...n as time allows. At the beginning of each search, I first search the best known sub tree with cilk spawn my_lambda(search(sub_tree)), the search the others in a cilk_for loop.

Interestingly enough, the search works search up to depth 2 or 3, but crashes with the error above afterwards. If I do not spawn the first search of the best known sub tree, this error does not occur(at least in the test program)

In my full original Alpha beta search code, the undo-detach error occurred almost all the time, and I would occasionally see the assetion failed error. It was after I tried serializing access to the lambda function with mutex's that I saw the cilk assertion failed error only. In the test case, even serializing calls to the lambda function cause the undo-detach error. Sorry for the long post, but any ideas or suggestions are greatly appreciated.

test-program.zip

[Brandon_H_Intel]

Hi Rubin,

I'm taking a look at the attached test case. I'll let you know if I can reproduce the problem you're seeing.