TFD Draft#127

Draft

evasnow1992 wants to merge 25 commits intoNVIDIA-Digital-Bio:mainfrom

evasnow1992:evax/tfd

Commits on Apr 2, 2026

Add TFD (Torsion Fingerprint Deviation) CPU implementation
evasnow1992
committed
Add TFD GPU CUDA kernels for dihedral and pairwise TFD computation
evasnow1992
committed
Add TFD GPU orchestration layer and unified public API
evasnow1992
committed
Add TFD Python bindings and test suite
evasnow1992
committed
Add multi-quartet support for ring and symmetric torsions in TFD
evasnow1992
committed
Parallelize RDKit preprocessing in TFD batch builds
evasnow1992
committed
Add TFD benchmark script comparing RDKit, CPU, and GPU backends
evasnow1992
committed
Add NVTX instrumentation to TFD and nsys profiling script
evasnow1992
committed
Refactor TFDSystemHost: remove redundant fields and optimize bond reuse
evasnow1992
committed
Fix Morgan fingerprint invariants and ring torsion deviation to match RDKit
evasnow1992
committed
Add skipGpuWorkItems to buildTFDSystem for CPU path optimization
evasnow1992
committed
Add molecule preprocessing script and pickle-based caching for benchmarks
evasnow1992
committed
Refactor GPU kernels: MolDescriptor dispatch, block-per-molecule TFD
evasnow1992
committed
Add return_type option to Python TFD API and optimize GPU result extraction
evasnow1992
committed
Add Python tests for numpy and tensor return types in TFD API
evasnow1992
committed
Update TFD reference values for RDKit 2025.09.5
evasnow1992
committed
Simplify TFD API layering: direct CPU/GPU generator routing
evasnow1992
committed
Move TFDGenerator tests to dedicated test_tfd.cpp, add backend validation
evasnow1992
committed
Remove redundant list copies in Python TFD result handling
evasnow1992
committed
Document floating-point tiebreak limitation in findCentralBond
evasnow1992
committed
Add --pkl-file and --verify-only flags to tfd_bench.py
evasnow1992
committed
Commit message:
evasnow1992
committed
Optimize CPU TFD result path: zero-copy numpy arrays, add NVTX instrumentation
evasnow1992
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

TFD Draft#127

TFD Draft#127
evasnow1992 wants to merge 25 commits intoNVIDIA-Digital-Bio:mainfrom
evasnow1992:evax/tfd

Commits on Apr 2, 2026

Add TFD (Torsion Fingerprint Deviation) CPU implementation

Add TFD GPU CUDA kernels for dihedral and pairwise TFD computation

Add TFD GPU orchestration layer and unified public API

Add TFD Python bindings and test suite

Add multi-quartet support for ring and symmetric torsions in TFD

Parallelize RDKit preprocessing in TFD batch builds

Add TFD benchmark script comparing RDKit, CPU, and GPU backends

Add NVTX instrumentation to TFD and nsys profiling script

Refactor TFDSystemHost: remove redundant fields and optimize bond reuse

Fix Morgan fingerprint invariants and ring torsion deviation to match RDKit

Add skipGpuWorkItems to buildTFDSystem for CPU path optimization

Add molecule preprocessing script and pickle-based caching for benchmarks

Refactor GPU kernels: MolDescriptor dispatch, block-per-molecule TFD

Add return_type option to Python TFD API and optimize GPU result extraction

Add Python tests for numpy and tensor return types in TFD API

Update TFD reference values for RDKit 2025.09.5

Simplify TFD API layering: direct CPU/GPU generator routing

Move TFDGenerator tests to dedicated test_tfd.cpp, add backend validation

Remove redundant list copies in Python TFD result handling

Document floating-point tiebreak limitation in findCentralBond

Add --pkl-file and --verify-only flags to tfd_bench.py

Commit message:

Optimize CPU TFD result path: zero-copy numpy arrays, add NVTX instrumentation

Commits on Apr 3, 2026

Optimize TFD batch build and simplify C++ API surface

Fix copyright years and ruff formatting for CI