The ordered clause works like this: different threads execute concurrently until they encounter the ordered region, which is then executed sequentially in the same order as it would get executed in a serial loop. This still allows for some degree of concurrency, especially if the code section outside the ordered region has substantial run time. … Read more
The random number generator rand(3) uses global state variables (hidden in the (g)libc implementation). Access to them from multiple threads leads to cache issues and also is not thread safe. You should use the rand_r(3) call with seed parameter private to the thread: long i; unsigned seed; #pragma omp parallel private(seed) { // Initialise the … Read more
I manged to get SIMD working with the Cholesky decomposition. I did this using loop tiling as I have used before in matrix multiplication. The solution was not trivial. Here are the times for a 5790×5790 matrix on my 4 core/ 8 HT Ivy Bridge system (eff = GFLOPS/(peak GFLOPS)): double floating point peak GFLOPS … Read more
You could allocate the big array inside the parallel region, where you can query about the actual number of threads being used: int *hista; #pragma omp parallel { const int nthreads = omp_get_num_threads(); const int ithread = omp_get_thread_num(); #pragma omp single hista = new int[nbins*nthreads]; … } delete[] hista; For better performance I would advise … Read more
People in HPC usually say that one thread is usually not enough to saturate a single memory link, the same usually being true for network links as well. Here is a quick and dirty OpenMP enabled memsetter I wrote for you that fills with zeros twice 2 GiB of memory. And here are the results … Read more
This is the first time I’m answering my own question but it seems appropriate. Based on hirschhornsalz answer for prefix sum on 16 bytes simd-prefix-sum-on-intel-cpu I have come up with a solution for using SIMD on the first pass for 4, 8, and 16 32-bit words. The general theory goes as follows. For a sequential … Read more
Try using Homebrew‘s llvm: brew install llvm You then have all the llvm binaries in /usr/local/opt/llvm/bin. Compile the OpenMP Hello World program. Put omp_hello.c /****************************************************************************** * FILE: omp_hello.c * DESCRIPTION: * OpenMP Example – Hello World – C/C++ Version * In this simple example, the master thread forks a parallel region. * All threads in … Read more
I think you can use std::vector with OpenMP most of the time and still have good performance. The following code for example fills std::vectors in parallel and then combines them in the end. As long as your main loop/fill function is the bottleneck this should work well in general and be thread safe. std::vector<int> vec; … Read more
Yes it is possible to do an array reduction with OpenMP. In Fortran it even has construct for this. In C/C++ you have to do it yourself. Here are two ways to do it. The first method makes private version of S for each thread, fill them in parallel, and then merges them into S … Read more
The clock function measures cpu time, the time you spend actively on the CPU, the OMP function measures the time as it has passed during execution, two completely different things. Your process seems to be blocked in waiting somewhere.