Repository navigation

[SC26 SCC] NTU Team Supernova's GPU optimizations to MFC - #2093

Closed

aliencaocao wants to merge 10 commits into

MFlowCode:masterfrom

aliencaocao:sc26-pr

Closed

[SC26 SCC] NTU Team Supernova's GPU optimizations to MFC#2093
aliencaocao wants to merge 10 commits into
MFlowCode:masterfrom
aliencaocao:sc26-pr

Commits on Oct 19, 2026
Commits on Oct 20, 2026
- m_rhs: keep the classic path out of CUDA-graph capture
  
  aliencaocao
  and
  claude
  committed
  e28b830
- m_chemistry: pass the diffusion-flux direction offsets as scalars
  
  aliencaocao
  and
  claude
  committed
  a1266fc
- m_start_up: size the CUDA device heap for the fused kernels
  
  aliencaocao
  and
  claude
  committed
  e038f4d
- tuning: accept src/common solver modules in MFC_REGISTER_CAPS
  
  aliencaocao
  and
  claude
  committed
  51ac66b
- HLLC: gate the ADC work arrays and second kernel emission on hypoelasticity
  
  aliencaocao
  and
  claude
  committed
  94a021b
- Validate MFC_REGISTER_CAPS at build time against the case
  
  aliencaocao
  and
  claude
  committed
  6b54fe0
- Take sc26's IB marker atomic-max, igr guards and mixed-precision scratch
  
  aliencaocao
  and
  claude
  committed
  32a2760