v1.2.9

• Fix segfault triggered by length-zero sequences (#244)
• Fix memory detection problem for some outdated MacOS versions (#232)
• Fix an incorrect assertion in unitig graph refreshing (#237)
• Added --verbose to output full log to the screen

v1.2.8

• Add intermediate megahit_core_popcnt for CPUs that have ABM but not BMI2 (thanks to @sjaenick)
• Allow new assembly task with --continue

v1.2.7

• Symbol link megahit_core_no_hw_accel to megahit_toolkit for backward compatibility (resolved #226)
• Better logging and for memory adjustment during SDBG building
• Attempt to continue SDBG building even if user-specified memory size is not sufficient

v1.2.6

• Refactored and fixed a bug in local assembler
• Refactored megahit script
• Obtain total memory size from os.sysconf (thanks to @Finesim97 )
• Fixed segmentation fault in Mac OS with clang 4.0 (thanks to @apcamargo)
• Added --cleaning-rounds and --disconnect-ratio options for more flexible graph cleaning control

v1.2.5-beta

• Fixed a bug that causes higher memory usage in seq2sdbg
• Refactor on sequence sorters, edge I/O and contig I/O

v1.2.4-beta

• Fixed a few memory leaks
• Use std::vector to replace malloc in SDBG builders
• Try to fix potential problem caused by benign data race in unitig graph refreshing
• Faster by using phmap and xxh3

v1.2.3-beta

• Refactored sequence readers
• Fixed a bug in SDBG building of large k-mer sizes

v1.2.2-beta

• Automatically detect POPCNT/BMI2 and select the correct megahit_core binary

v1.2.1-beta

• Added --no-hw-accel option for users whose CPUs do not support BMI2/POPCNT
• Added --test option for testing
• Compilable with CMake 2.8 and g++4.8

v1.2.0-beta

Heavily refactored the whole project:

• Remove GPU support
• Use cmake
• Use sparsepp to replace IDBA's hash map for better performance in both speed and memory efficiency
• Use pdep instruction to speed up rank and select
• Rewrite unitig graph
• Rewrite the iterate-edge component
• Rewrite the SDBG library, except for the builder
• Fixed a bug which may involve too many reads into local assembly

The changes result in a faster and more memory-efficient tool, but have little effect on assembly quality.