HTTPS clone URL
Subversion checkout URL
Latest commit e5dd314 stamatak -fixed incorrect computation of lower bound for the recomputation app… …
…roach, it used the natural logartithm, now it uses the correct base 2 logarithm
RAxML-Light Readme For detailed documentation please refer to the RAxML-Light manual included in this archive: RAxML-Light-Manual.pdf Quick-start: To compile the various versions: The distribution comes with six Makefiles for the sequential, Pthreads, and MPI-based versions. The sequential, Pthreads and MPI versions either use SSE3 128-bit wide SIMD (Single Instruction Multiple Data) or AVX 256-bit wide SIMD instructions which are offered by all recent AMD and Intel x86 architectures. The default compiler is gcc, but you may experiment with Intel icc or the Portland PGI compiler (although those compilers have not been tested by me). The parallel MPI version works with Intel icc and gcc and so far I have tested the MVAPICH2 (gcc and icc) and OpenMPI (gcc) compilers. To compile the sequential versions, type: make -f Makefile.SSE3.gcc make -f Makefile.AVX.gcc This will generate executables called raxmlLight/raxmlLight-AVX. If you want to compile the Pthreads version next, first type “rm *.o” in your terminal to remove the object files generated for compiling and linking the sequential program and type: make -f Makefile.SSE3.PTHREADS.gcc make -f Makefile.AVX.PTHREADS.gcc This will produce executables called raxmlLight-Pthreads and raxmlLight-Pthreads-AVX respectively Then, to compile the MPI version, first type “rm *.o” again and then type: make -f Makefile.SSE3.MPI make -f Makefile.AVX.MPI which will generate executables called raxmlLight-MPI/raxmlLight-MPI-AVX. For this you will need to have a MPI compiler (called mpicc) installed on your system/cluster. If you don't know what a MPI compiler is, just talk to your local geek. For testing purposes under Ubuntu Linux, it is probably easiest to install OpenMPI. On our local cluster here at HITS using the 64-bit icc compiler v11.1 and MVAPICH2 yielded the best performance. It is really worth playing around with different compiler and MPI implementations, because this can yield performance differences of up to 30%. WARNING: For performance reasons the number of partitions is hard-coded in the MPI version of the code. This is admittedly not very elegant, but we had to sacrifice flexibility for performance. Thus, when running RaxML-Light, you may sometimes see it abort with the following or a similar error message: raxmlLight-MPI: fineGrainMpi.c: startFineGrainMpi: Assertion `tr->NumberOfModels == 2' failed. In such a case you will have to edit source file axml.h, and amend the following line: #define NUM_BRANCHES 1 to read: #define NUM_BRANCHES X where X is the number of partitions in the dataset you want to analyze. Thereafter, please re-compile the MPI code and everything should work. For details on command line options please refer to the manual.