Published by atillack almost 3 years ago
This release enables running on Apple M1 and adds the Boron atom type in addition to:
Published by atillack about 3 years ago
This release fixes two bugs leading to "rapid unscheduled program terminations" under certain circumstances (see PR #153 and issue #152 ):
Published by atillack about 3 years ago
This release improves the performance of the Cuda path to finally be on par with the OpenCL version (see PR #151 for details) and adds some small fixes to correctly deal with relative paths in grid map files.
Published by atillack about 3 years ago
This release contains major feature updates and code optimizations:
OVERLAP=ON
to parallelize running on multiple GPUs. During parallel execution, load balancing is achieved by each waiting thread grabbing the next available GPU. GPUs to run on can be chosen individually with the --device
, -D
argument, i.e. -D 1,3,5
, or all available GPUs can be used with -D all
.--filelist
, -B
argument (as an alternative to the file list file) to dock all ligands with the specified grid maps, flex res, and other settings (example: autodock_gpu_128wi -M receptor.maps.fld -B ligands*.pdbqt
)adgpu_analysis
tool that can also be compiled when no DEVICE
(and hence no Cuda or OpenCL) is specified during compilation to have --xml2dlg
functionality without needing an accelerator framework.--rlige 1
has been extended to full dlg-style energy output and is also added to reference ligand output in dlg output.--xraylfile
) as well input ligand output in dlg when specified.pdbqt*
(note the *
at the end) are now interpreted as the reference ligand for successively specified ligandsPublished by atillack over 3 years ago
This updated release fixes a bug with the file list in cases with no resnames specified.
AD-GPU was erroneously bailing out silently (observed in issue #145). The correct behavior, to use the ligand pdbqt file base name, has been restored.
Published by atillack over 3 years ago
This release fixes a bug introduced with the contact analysis in PR #131
Description of bug fix:
#MACROMOLECULE
) when no contact analysis is requestedPublished by diogomart over 3 years ago
CUDA has been enabled.
Gradient evals are now faster, for both OpenCL and CUDA.
Published by diogomart over 3 years ago
New features:
intnbp_r_eps
from autodock4)autostop
, which is enabled by defaultPublished by diogomart about 4 years ago
In the previous Solis-Wets version all genes were updated at each iteration. Here, the probability of updating each gene decreases with the number of genes. Furthermore, the magnitude of gene updates was scaled down and also decreases with the number of genes. Simple ligands for which the E50 are below 128k evals see no improvements but complex ligands see 8x improvements (e.g., E50 reductions from 8M to 1M).
The stopping criterion of Solis-Wets was incorporated into ADADELTA, leading to about 2x better performance (E50) for several ligands. Complex ligands, for which the E50 is at 8M or larger, see no improvements.
Two command line options are introduced:
-autostop 1
automatically stops the docking job when convergence is reached.-heuristics 1
sets the maximum number of evaluations of each LGA run (-nev
) based on the number of rotatable bonds.