[WIP] GPU acceleration - #403

Draft

lmdiazangulo wants to merge 38 commits into

devfrom

performance-nvfortran

Commits on Apr 29, 2026

Add NVIDIA HPC compiler support to CI and CMake presets

lmdiazangulo
and
Copilot
committed

Commits on May 11, 2026

Merge remote-tracking branch 'origin/dev' into performance-nvfortran
lmdiazangulo
committed
Trying with cuda fortran
lmdiazangulo
committed
CUDA works, but slower
lmdiazangulo
committed
Remove OpenACC, implement persistent CUF device memory
lmdiazangulo
committed
Optimize CUF GPU: cache coefficients on device, reduce per-step transfer by 60%
lmdiazangulo
committed
CPU optimizations: collapse(2) on CPML/MUR, remove duplicate calls
lmdiazangulo
committed
GPU profiling: identify cudaMemcpy bottleneck, add pinned alloc flag
lmdiazangulo
committed
GPU: keep fields on device between timesteps, eliminate per-step cudaMemcpy
lmdiazangulo
committed
GPU: add gpu_destroy, clean up CPML scaffolding, fix link error
lmdiazangulo
committed
Add benchmark suite: automated benchmark script with CUDA-Fortran and CPU modes
lmdiazangulo
committed
Add CPML left-boundary GPU kernels (Phase 1)
lmdiazangulo
committed

Commits on May 12, 2026

Commits on May 13, 2026