libpll 0.3.2

Changes

• Optional per-rate category scalers for protein and generic kernels
• Hardware detection for APPLE builds defaults to assembly code
• Improved fix for negative p-matrix values
• Fixed derivatives computation for Lewis/Felsenstein ascertainment bias correction
• Fixed set_tipchars() for ascertainment bias correction with non-DNA sequences
• Fixed excessive memory allocation when compressing site patterns
• Fixed issue with PHYLIP parsing when header ends with CRLF

Notes

1. The file libpll-0.3.2-linux-x86_64.tar.gz is a pre-compiled binary of libpll.
2. The file libpll-src-0.3.2.tar.gz contains the libpll source code and does not require autotools installed.
3. Files 0.3.2.tar.gz and 0.3.2.zip are generated by GitHub, and require autotools installed (as when cloning the repository).

libpll 0.3.1

Changes

• Checks for older versions of clang and gcc to use assembly instructions
  for cpu features detection
• Include guards for pll.h
• Correct updating of padded eigen-decomposition arrays for models with a
  number of states not being a power of two
• Changed to the usage of builtin functions for cpu features detection
• Check for x86intrin.h

Notes

1. The file libpll-0.3.1-linux-x86_64.tar.gz is a pre-compiled binary of libpll.
2. The file libpll-src-0.3.1.tar.gz contains the libpll source code and does not require autotools installed.
3. Files v0.3.1.tar.gz and 0.3.1.zip are generated by GitHub, and require autotools installed (as when cloning the repository).

libpll 0.3.0

Changes

• Run-time detection of cpu features
• Vectorized (AVX) computation of 20-state transition probability matrices
• Faster tip-inner kernels for 20-state models
• Improved AVX vectorization of derivatives
• Faster PHYLIP parser
• vectorized scaling for 20-state and arbitrary-state models
• AVX2 vectorizations for partials, likelihood and derivatives
• Unweighted parsimony functions including SSE, AVX and AVX2 vectorizations
• Randomized stepwise addition
• Portable functions for parsing trees from a C-string
• Optional per-rate category scalers to prevent numerical underflows on large
  trees
• Setting of identity matrix if all exponentiations of eigenvalues multiplied
  by branch length and rate are approximately equal to one
• Re-entrant cross-platform pseudo-random number generator
• Wrapper tree structures
• Custom exporting of tree structures using a callback function
• Support for median category rates in discrete gamma model
• Fixed derivatives computation
• Fixed parsing of branch lengths in newick trees
• Fixed computation of invariant sites
• Multiplication of log-likelihood with pattern weight after scaling term
• Added destructors for eliminating memory leaks when tree parsing fails
• Fixed sumtable computation when having multiple substitution matrices
• Fixed ascertainment bias computation
• Fixed per-site log-likelihood computation
• Fixed uninitialized values in testing framework

Notes

1. The file libpll-0.2.0-linux-x86_64.tar.gz is a pre-compiled binary of libpll.
2. The file libpll-src-0.3.0.tar.gz contains the libpll source code and does not require autotools installed.
3. Files v0.3.0.tar.gz and 0.3.0.zip are generated by GitHub, and require autotools installed (as when cloning the repository).