=============================================================================== QuickProbs version 1.02, April 2014 Adam Gudys, Sebastian Deorowicz Silesian University of Technology, Gliwice, Poland adam.gudys@polsl.pl, sebastian.deorowicz@polsl.pl GPL version 3.0 applies. =============================================================================== 1. Running QuickProbs QuickProbs is a variant of MSAProbs algorithm customised for graphics processors. Therefore, it supports properly all MSAProbs command line arguments. These are: -o, --outfile specify the output file name (STDOUT by default) -num_threads specify the number of threads used, and otherwise detect automatically; in QuickProbs it applies only to stage IV of the algorithm as it has been reimplemented in OpenMP; the parameter has no influance on stages I and III as they were customised for GPU execution -clustalw use CLUSTALW output format instead of FASTA format -c, --consistency REPS use 0 <= REPS <= 5 (default: 2) passes of consistency transformation -ir, --iterative-refinement REPS use 0 <= REPS <= 1000 (default: 10) passes of iterative-refinement -v, --verbose report progress while aligning (default: off) -annot FILENAME write annotation for multiple alignment to FILENAME -a, --alignment-order print sequences in alignment order rather than input order (default: off) -version print out version of QuickProbs Additionally, QuickProbs is equipped with three new parameters. Two former allows user to select on which OpenCL platform and device GPU code is to be run. The last one is a switch enabling QuickProbs accurate mode. -p number of OpenCL platform -d number of OpenCL device -q run QuickProbs in the accurate mode In order to list all available OpenCL devices and platforms please run QuickProbs without any parameters. First execution of QuickProbs on particular platform and device may take longer than following executions. This is because package needs to compile all OpenCL kernels. 2. OpenCL environment on AMD platforms. In order to optimally utilise computational power of AMD graphics processors some environment variables have to be set. They are described below. 1. GPU_MAX_WORKGROUP_SIZE For unknown reasons maximum size of workgroup for AMD devices reported by OpenCL is lower than physical hardware limit. E.g. for Radeon 7970 it reports maximum size of 256 while real limit is 1024. This environment variable allows one to override default setting. Read in device documentation maximum physical size of workgroup and set the variable to this value. 2. GPU_MAX_ALLOC_PERCENT This value indicate maximum percentage size of single OpenCL buffer related to total size of GPU global memory. Tests in the paper were performed with maximum buffer size equal to 600MB. On the device with 3GB of memory it corresponds to GPU_MAX_ALLOC_PERCENT equal to 20.