Intel Software Network - feed

comparison cilk++, openmp, pthreads first results

Thu, 05 Nov 2009 04:10:30 -0800

Hello,

Here the first results of my comparison of cilk++ and openmp, on my application which is 3D meshgeneration. I have done here the first part which is octree-refinement + inserting triangles. The reference is a pthreaded version of the code.
Please consider, that these are first numbers. The difference between cilk++ and openmp is also due to the used compiler-version (but cilk is done only for the 4.2.4).

best regards

Ferdinand

PTU Error with OpenGL apps and new NVIDIA drivers

Wed, 04 Nov 2009 09:51:15 -0800

Running on a Redhat 5 64-bit linux system with NVIDIA Quadro gfx, using NVIDIA 185.18.14 or newer drivers, ptu profiling (vtssrun) on any OpenGL app gives the error:

2714602318157736 : 2a95f18680 : CRITICAL : 'dlsym' for '_Exit' error Bad file descriptoropen log '(null)': Bad address
write log: Bad file descriptor
2714602318197872 : 2a95f18680 : CRITICAL : 'dlsym' for '_exit' faild. Error Bad file descriptoropen log '(null)': Bad address
write log: Bad file descriptor

over and over again.

If I roll back to 173.14.12, the profiling works as expected

PTU Error with OpenGL apps and new NVIDIA drivers

Wed, 04 Nov 2009 09:40:40 -0800

Running on Redhat 5 64-bit linux machines (tried on several different ones) with NVIDIA driver 185.18.14 or later, if I try to profile (vtssrun) any OpenGL app using ptu 3.2, I receive the following errors:

2714602317306492 : 2a95f18680 : CRITICAL : 'dlsym' for '_Exit' error Bad file descriptoropen log '(null)': Bad address
write log: Bad file descriptor
2714602317342532 : 2a95f18680 : CRITICAL : 'dlsym' for '_exit' faild. Error Bad file descriptoropen log '(null)': Bad address
write log: Bad file descriptor

If I roll back to 173.x.x drivers (specifically 173.14.12) the profiling works as expected

User undo action in TM

Wed, 04 Nov 2009 05:42:38 -0800

STM ABI says that function registered by calling _ITM_addUserUndoAction() is executed if a transaction is aborted or retried to allow the user code to undo any operations.

Is it possible to get pointer to _ITM_abortReason struct or to detect if transaction was aborted by user or retried inside this user-defined function?

cilk++ does not support 64bit programming ?

Tue, 03 Nov 2009 15:10:58 -0800

I find my cilk++ code can not be compiler wtih vs 2008 and intel c++ if I choose win64 platform.

openmp vs cilk++

Mon, 02 Nov 2009 19:18:49 -0800

which one is better ?
easy to use, performance ?

Problems with cilk++

Mon, 02 Nov 2009 12:29:24 -0800

Hello,

I was downloading and testing the clik++ compiler. I am very exited about this, since the first results are showing good performance and a nice coding style too. Now I was moving further, porting my standard application to cilk++. At the beginning this is an octree, where I insert (a lot of) triangles. It took me a long time to find out, but now I ended up with a state, where running the code with g++ gives the correct results. Running the code with cilk++ (only changing the compiler, no cilk_for, or any other parallel stuff used) gives wrong results. If I switch off the optimization in cilk++ it again works (even with the parallel loops). The problem is, if I cannot use the optimization, this is very slow and it makes no sense to use it.

Are there some known compiler problems? How can I proceed, I would still love to do the cilk-typed coding, since this would simplyfy a lot.

best regards

Ferdinand

Intel STM Compiler barrier optimizations

Wed, 21 Oct 2009 11:55:41 -0700

Hi -

The TM Compiler and Runtime ABI specification 1.0.1 states that the compiler could generate code without some barriers. In such a case, the compiler places speficic requirements on the runtime. (e.g. code without "after write" barriers requires an in-place update STM).

My understanding is that the Intel STM Compiler v3.0 implements this ABI specification but I am wondering whether it also makes the above relaxations (omitting barriers).

Thanks,

-Haris

Intel Adaptive Spike-Based Solver

Mon, 19 Oct 2009 08:51:49 -0700

Hello,

I have a question about Intel Spike solver. I want to solve matrix ecology2 (http://www.cise.ufl.edu/research/sparse/matrices/McRae/ecology2.html) from florida matrix collection using SPIKE. What kind of solving strategy (RSS, DFS, OIS) should I choose? The matrix is solved on a 48-nodes cluster with 8 cores on each node.

Intel STM C/C++ Contention Manager

Tue, 13 Oct 2009 16:04:37 -0700

Hi all,

Does anyone know whether or not it is possible to override the Intel compiler's transaction contention manager with a custom implementation? I am basically looking to modify the behaviour in which Intel STM aborts one transaction over another.

Any help would be greatly appreciated.