feat: integrate NEP CUDA kernels into MD runner with complete GPU compute path - #7526

Open

lijianing-sudo wants to merge 27 commits into

deepmodeling:developfrom

Audrey-777:fix/cuda-gpu-integration

Commits on May 30, 2026

Add NEP CUDA postprocess prototype
Donetella
committed
fd2f72c
add auth.json to log in codex
Donetella
committed
1ae5850
do some test and remove root file
dyzheng
committed
a0a9daf
test by GPU
dyzheng
committed
54c0cc8

Commits on Jun 5, 2026

Add DP NEP CUDA acceleration refactor
dyzheng
committed
2548ef8

Commits on Jun 24, 2026

Add NEP CUDA core compute (4 kernels: descriptor+ANN, radial force, angular force, ZBL) with 15 device functions and fine-grained cudaEvent timing. Comprehensive report v2 with phase 2 details and …
dyzheng
committed
2dfad46

Commits on Jun 26, 2026

fix: CMake NEP CUDA build + GPU compute path integration

lijianing-sudo
and
claude
committed
4c7ae34
fix: force CUDA C++14 standard for GCC 11 compat

lijianing-sudo
and
claude
committed
4206398
fix: move CUDA C++14 flags BEFORE enable_language(CUDA)

lijianing-sudo
and
claude
committed
82593d5
fix: match NEP API - ann→annmb, w0/b0/w1 are const double*[94]

lijianing-sudo
and
claude
committed
5e81919
fix: add nep_cuda_compute declaration to .cuh header

lijianing-sudo
and
claude
committed
e6a2c69
fix: use extern declaration instead of including .cuh

lijianing-sudo
and
claude
committed
5d1394d
fix: correct nep_force_angular_kernel call arguments

lijianing-sudo
and
claude
committed
4151996
fix: add missing stdio.h/stdlib.h for fprintf/stderr/exit
lijianing-sudo
committed
bebca14
fix: auto-detect GCC-10 for CUDA 11.x host compiler

lijianing-sudo
and
claude
committed
482522c
fix: use CMAKE_CUDA_HOST_COMPILER instead of -ccbin flag

lijianing-sudo
and
claude
committed
66ae0e2
docs: clarify w0[0] memory layout assumption in GPU path

lijianing-sudo
and
claude
committed
9886ab2
test: add 2x2x2 HfO2 supercell (192 atoms) for GPU benchmark

lijianing-sudo
and
claude
committed
8e49abe
test: add 4x4x4 HfO2 supercell (1536 atoms) for GPU scaling

lijianing-sudo
and
claude
committed
8c4cc5e
fix: move neighbor_nep.h include from header to .cpp

lijianing-sudo
and
claude
committed
72db88e
chore: clean up test INPUTs, remove device gpu, fix to NVE

lijianing-sudo
and
claude
committed
9a4d4c6
fix: guard neighbor_nep.h include with __NEP macro

lijianing-sudo
and
claude
committed
bfbdc16

Commits on Jun 28, 2026

Merge branch 'develop' into fix/cuda-gpu-integration
lijianing-sudo
authored
aed739b

Commits on Aug 8, 2026