Repository navigation

GPU stack 5/8: Poisson/diffusion solve with fewer passes over the vectors - #786

Merged

max-models merged 59 commits into

gpu-stack-4-kronecker-massfrom

gpu-stack-5-diffusion-solve

Oct 24, 2026

Merged

GPU stack 5/8: Poisson/diffusion solve with fewer passes over the vectors#786
max-models merged 59 commits into
gpu-stack-4-kronecker-massfrom
gpu-stack-5-diffusion-solve

Commits on Oct 16, 2026
Commits on Oct 23, 2026
- CuPy backend: host/device conversions in post-processing; math.sqrt for host scalars in Species
  
  max-models
  and
  claude
  authored
  6bb532f
- CUDA version of the mass-matrix assembly kernel kernel_3d_mat
  
  max-models
  and
  claude
  authored
  5e2dced
- Parity cases for kernel_3d_mat
  
  max-models
  and
  claude
  authored
  b586cde
- Apply separable mass matrices as Kronecker products of 1D banded matrices
  
  max-models
  and
  claude
  authored
  4147d1a
- Kronecker mass: launch the band-pass kernel through CachedLaunch
  
  max-models
  and
  claude
  authored
  f766830
- Merge branch 'gpu-stack-4-kronecker-mass' into gpu-stack-5-diffusion-solve
  max-models
  committed
  9aeec98
- CuPy backend: host/device conversions in post-processing; math.sqrt for host scalars in Species
  
  max-models
  and
  claude
  authored
  d0a8f31
- CUDA version of the mass-matrix assembly kernel kernel_3d_mat
  
  max-models
  and
  claude
  authored
  5c099f3
- Parity cases for kernel_3d_mat
  
  max-models
  and
  claude
  authored
  bcbb46a
- Apply separable mass matrices as Kronecker products of 1D banded matrices
  
  max-models
  and
  claude
  authored
  bbe66ed
- Kronecker mass: launch the band-pass kernel through CachedLaunch
  
  max-models
  and
  claude
  authored
  07fd091
- Merge branch 'gpu-stack-4-kronecker-mass' into gpu-stack-5-diffusion-solve
  max-models
  committed
  986c7ce
- CuPy backend: host/device conversions in post-processing; math.sqrt for host scalars in Species
  
  max-models
  and
  claude
  authored
  f1cbb2d
- CUDA version of the mass-matrix assembly kernel kernel_3d_mat
  
  max-models
  and
  claude
  authored
  ef927d1
- Parity cases for kernel_3d_mat
  
  max-models
  and
  claude
  authored
  c1f8875
- Apply separable mass matrices as Kronecker products of 1D banded matrices
  
  max-models
  and
  claude
  authored
  890b6d9
- Kronecker mass: launch the band-pass kernel through CachedLaunch
  
  max-models
  and
  claude
  authored
  df53602
- Merge branch 'gpu-stack-4-kronecker-mass' into gpu-stack-5-diffusion-solve
  max-models
  committed
  6f686c1
- CuPy backend: host/device conversions in post-processing; math.sqrt for host scalars in Species
  
  max-models
  and
  claude
  authored
  f63a560
- CUDA version of the mass-matrix assembly kernel kernel_3d_mat
  
  max-models
  and
  claude
  authored
  1a3939e
- Parity cases for kernel_3d_mat
  
  max-models
  and
  claude
  authored
  361b541
- Apply separable mass matrices as Kronecker products of 1D banded matrices
  
  max-models
  and
  claude
  authored
  62afbad
- Kronecker mass: launch the band-pass kernel through CachedLaunch
  
  max-models
  and
  claude
  authored
  1a0f477
- Merge branch 'gpu-stack-4-kronecker-mass' into gpu-stack-5-diffusion-solve
  max-models
  committed
  674017a
- CuPy backend: host/device conversions in post-processing; math.sqrt for host scalars in Species
  
  max-models
  and
  claude
  authored
  035b8a9
- CUDA version of the mass-matrix assembly kernel kernel_3d_mat
  
  max-models
  and
  claude
  authored
  565eea9
- Parity cases for kernel_3d_mat
  
  max-models
  and
  claude
  authored
  08ea857
- Apply separable mass matrices as Kronecker products of 1D banded matrices
  
  max-models
  and
  claude
  authored
  715cc5c
- Kronecker mass: launch the band-pass kernel through CachedLaunch
  
  max-models
  and
  claude
  authored
  c78a6f7
- Merge branch 'gpu-stack-4-kronecker-mass' into gpu-stack-5-diffusion-solve
  max-models
  committed
  38f9cb4
Commits on Oct 24, 2026
- Merge branch 'gpu-stack-4-kronecker-mass' into gpu-stack-5-diffusion-solve
  max-models
  committed
  766a868

Commits

Commits on Oct 16, 2026

Commits on Oct 23, 2026

Commits on Oct 24, 2026