diff --git a/doc/user/chapters/parallel_worker.org b/doc/user/chapters/parallel_worker.org index b18937ae9b1f01008caf0586f3dc3488cafe5563..0245daa62c1f75284d1e8368e764c2b9d3d87bef 100644 --- a/doc/user/chapters/parallel_worker.org +++ b/doc/user/chapters/parallel_worker.org @@ -12,7 +12,7 @@ workers which execute a single parallel task (see [[https://files.inria.fr/starp To use this functionnality: - StarPU must be compiled with the configure option =--enable-parallel-worker= -- Chameleon automatically detects if the StarPU parallel workers are available or not and can exploit them, but you need to force Chameleon to be linked with multi-threaded BLAS library if you want the parallel workers to be able to do parallel BLAS calls. To do that, you must add =-DCHAMELEON_KERNELS_MT=ON= to your cmake line. +- Chameleon automatically detects if the StarPU parallel workers are available or not and can exploit them, but you need to force Chameleon to be linked with a multi-threaded BLAS library if you want the parallel workers to be able to do parallel BLAS calls. To do that, you must add =-DCHAMELEON_KERNELS_MT=ON= to your cmake line. Below are given some examples to use the couple Chameleon/StarPU to enable parallel tasks to be run concurrently. For now, this is only available for a few subset of tasks that are used in the Cholesky decomposition (=POTRF=, =TRSM=, =SYRK=, =HERK=, and =GEMM=) but all other algorithms using these kernels benefit from it.