GPU Cluster Pair Performance - #30

Merged

rafaelravedutti merged 59 commits into

mainfrom

Aug 6, 2026

Commits on May 27, 2026

Add j-slice and imask for superclusters kernels
rafaelravedutti
committed
8827d50
WHen GPU_J_SLICES is 1, avoid atomics
rafaelravedutti
committed
e9e99ce

Commits on Jun 8, 2026

Merge branch 'main' into gpu_cp_perf
rafaelravedutti
committed
7218d75
Add geometric mixing rule for remaining cases
rafaelravedutti
committed
d676a6b

Commits on Jun 10, 2026

Add double cutoff and pruning to VL cases
rafaelravedutti
committed
6fa795d
Update TODO.md
rafaelravedutti
committed
9f7227e

Commits on Jun 11, 2026

Add todo task for fixing bug for SVE2 case
rafaelravedutti
committed
e480312

Commits on Jun 23, 2026

Add tabulated (spline interpolation) force kernels for LJ
rafaelravedutti
committed
3ccee27
Adjust allocations and string matching
rafaelravedutti
committed
190d565

Commits on Jun 24, 2026

Update .md files
rafaelravedutti
committed
85fa594
Add preliminary (ref) version of Cluster Pair kernels with tabulated potentials
rafaelravedutti
committed
293d8bc
For now, only use ref tabulated kernels for CP
rafaelravedutti
committed
68d3274
Fix error when not using reference kernel
rafaelravedutti
committed
3aa5537
Run clang-format
rafaelravedutti
committed
67ab32a
Add CLAUDE.md (experimental)
rafaelravedutti
committed
a9d97e3

Commits on Jun 29, 2026

Add CSR format for neighbor lists
rafaelravedutti
committed
e9cfa21
Add config.mk changes and fix nbN reference
rafaelravedutti
committed
c77ff5b
Update TODO.md
rafaelravedutti
committed
f862ac3

Commits on Jun 30, 2026

Add displacement-based strategy to trigger neighbor-lists rebuilding
rafaelravedutti
committed
f217e49

Commits on Jul 6, 2026

Add experimental SIMD-compressed LJ kernels
rafaelravedutti
committed
e7835ff
Update TODO.md
rafaelravedutti
committed
104509f

Commits on Jul 7, 2026

Update TODO.md and GCC flags
rafaelravedutti
committed
1653ffc
Fix test skipping for tabulated kernels
rafaelravedutti
committed
b50affe
Adjust stubbed cases and make them work for GPUs
rafaelravedutti
committed
d3d4072
By default, set LJ_TABLE_INDEX to rsq
rafaelravedutti
committed
a9665e8
Fix reneighbor interval parameter output
rafaelravedutti
committed
e6e0d66
Add -mprefer-vector-width=512 option to GCC when building for AVX512
rafaelravedutti
committed
a7c9d9a
Include LJ_COMB_RULE and USE_SIMD_KERNEL references into the binary tag
rafaelravedutti
committed
bc5db95

Commits on Jul 13, 2026

Commits on Jul 14, 2026

Include GPU setupPbc
rafaelravedutti
committed
cd4ab7c
Include GPU setupPbc
rafaelravedutti
committed
b4b0f2b

Commits on Jul 23, 2026

Fix cluster-pair reallocation when using CSR
rafaelravedutti
committed
a4b0f9e

Commits on Jul 30, 2026

Add tail loop split for VL SIMD kernels
rafaelravedutti
committed
5df75fc
Provide matching flags for CLANG, GCC and ICX
rafaelravedutti
committed
eacf184

Commits on Aug 3, 2026

Fix GPU cluster-pair neighbor-list corruption and SIMD portability bugs
rafaelravedutti
committed
bc5a5ab

Commits on Aug 4, 2026

Run clang-format
rafaelravedutti
committed
79c71d7
Fix issues with super-cluster version
rafaelravedutti
committed
0ad2fee
Merge remote-tracking branch 'origin/main' into gpu_cp_perf
rafaelravedutti
committed
d1bbb36
Only use reverseGhostForcesCPU for half_neigh cases
rafaelravedutti
committed
b7224c3
Fix soa_neighbors buffer allocation for GPUs
rafaelravedutti
committed
a565a90
Adjust PBC for GPU kernels
rafaelravedutti
committed
500dfb5
Fix SVE intrinsics with FP32 for Cluster Pair case
rafaelravedutti
committed
ef519ee

Commits on Aug 5, 2026

Replace _z suffix by _x to get rid of redundant movprfx instructions
rafaelravedutti
committed
b33c081
Update TODO.md
rafaelravedutti
committed
4c9826f
Run clang-format
rafaelravedutti
committed
38c91c6
Add Vector-Length Agnostics variant for VL SIMD kernels and fix old LJ_COMB_NONE references
rafaelravedutti
committed
c58e5e4
Add tests for OpenMP
rafaelravedutti
committed
4d27728

Commits on Aug 6, 2026