perf(TDDFT): Add CUDA acceleration for snap_psibeta_half function (Useful information about largely improves the snap_psibeta_half function)#6808

Merged

mohanchen merged 13 commits intodeepmodeling:developfrom

dzzz2001:cuda_tddft_claude

Dec 28, 2025

Commits on Dec 26, 2025

feat(tddft): Add CUDA acceleration for snap_psibeta with constant memory grids
dzzz2001
committed
refactor(snap_psibeta_atom_batch_gpu): add timer and simplify code structure
dzzz2001
committed
remove snap_psibeta_neighbor_batch_gpu
dzzz2001
committed
perf(gpu): optimize snap_psibeta_atom_batch_kernel with loop restructuring
dzzz2001
committed
perf(gpu): optimize compute_ylm_gpu with atan2 and sincos
dzzz2001
committed
make the code more concise
dzzz2001
committed
perf(cuda): optimize compute_ylm_gpu with template parameters
dzzz2001
committed
perf(cuda): use warp shuffle for reduction in snap_psibeta kernel
dzzz2001
committed
refactor(cuda): improve snap_psibeta_kernel code organization and documentation
dzzz2001
committed
refactor(td_nonlocal_lcao): use runtime check for GPU/CPU branch selection
dzzz2001
committed
refactor(snap_psibeta): unify CUDA error checking macros
dzzz2001
committed
refactor(snap_psibeta_kernel): use ModuleBase constants
dzzz2001
committed
refactor(snap_psibeta): use ModuleBase::WARNING_QUIT for error handling
dzzz2001
committed