Repository navigation

Add Metal and OpenCL GPU backends - #1

Open

robtaylor wants to merge 186 commits into

mainfrom

metal-backend

Open

Add Metal and OpenCL GPU backends#1
robtaylor wants to merge 186 commits into
mainfrom
metal-backend

Commits on Jan 2, 2026
- Add Metal backend for Apple Silicon GPU acceleration
  
  robtaylor
  and
  claude
  committed
  72c6c28
- Add OpenCL backend for portable GPU acceleration
  
  robtaylor
  and
  claude
  committed
  5d0c03b
- Add GitHub Actions CI with GPU testing infrastructure
  
  robtaylor
  and
  claude
  committed
  6bd4dce
- Add cuDSS-style block CSR interface for solver creation
  robtaylor
  committed
  90530f0
- Add LU factorization support for single-block matrices
  
  robtaylor
  and
  claude
  committed
  5958fea
- Add UMFPACK comparison tests for LU factorization validation
  
  robtaylor
  and
  claude
  committed
  e5683db
Commits on Jan 3, 2026
- Fix multi-block LU factorization Schur complement updates
  
  robtaylor
  and
  claude
  committed
  49f264a
- Add LDL^T factorization for symmetric indefinite matrices
  
  robtaylor
  and
  claude
  committed
  9b6f91d
- Address code review feedback for LDL^T implementation
  
  robtaylor
  and
  claude
  committed
  a06a4a0
- Relax OpenCL double precision tolerance for CI
  
  robtaylor
  and
  claude
  committed
  508f01e
Commits on Feb 23, 2026
Commits on Feb 24, 2026
- [cmake] Make llvm-ar optional, fall back to CMAKE_AR
  robtaylor
  committed
  e4b9fe3
- [cmake] Improve llvm-ar error message with install instructions
  robtaylor
  committed
  dabaf6b
Commits on Feb 25, 2026
- Add GPU vs UMFPACK comparison tests and async dispatch infrastructure
  robtaylor
  committed
  3f0d77c
- Fuse saveGemm dispatches into batched kernel for Metal LU
  robtaylor
  committed
  8ed0870
- Fix shared buffer race condition in batched saveGemm
  robtaylor
  committed
  8dd9e8b
- Add cuDSS vs BaSpaCho nsys profiling benchmark
  robtaylor
  committed
  4a503fb
- Fix runner labels: use nvidia-runner-1 without self-hosted
  robtaylor
  committed
  8104204
- Add cmake and build deps to CUDA self-hosted runner
  robtaylor
  committed
  df242d3
- Fix sparse_elim_straight kernel, add missing Metal kernels & per-kernel tests
  robtaylor
  committed
  b2cdc43
- Use NVIDIA CUDA container for GPU CI jobs
  robtaylor
  committed
  730501b
- Fix cuDSS profiling workflow: correct version, add nsys install
  robtaylor
  committed
  debb3b4
- Fix nsight-systems install: use versioned package name
  robtaylor
  committed
  155ec40
- Switch CUDA containers to Ubuntu 24.04
  robtaylor
  committed
  3a27154
- Simplify cuDSS install: use apt package on Ubuntu 24.04
  robtaylor
  committed
  11f7dc8
- Fix unused parameter warning in CUDA trsmUpperRight
  robtaylor
  committed
  2a20518
- Update CudssBenchmarkTest for cuDSS 0.7 API
  robtaylor
  committed
  ef425c5
- Fix nsys profile step: remove invalid cd in container
  robtaylor
  committed
  690a7f9
- Fix FastSymbolic tests
  robtaylor
  committed
  f59fc6e
Commits on Feb 27, 2026
- Add fast symbolic analysis using CHOLMOD
  robtaylor
  committed
  acbc8d9
- Add supernode merging and level-set scheduling
  robtaylor
  committed
  a09b15e
- Address code review findings for Milestones 1 & 2
  robtaylor
  committed
  709d5f3
- Integrate supernode merging and level-set scheduling into solver pipeline
  robtaylor
  committed
  e6206c5
- Address code review findings for Milestone 3
  robtaylor
  committed
  5425f30
- Fix division-by-zero UB in RandomScalarMatrix test
  robtaylor
  committed
  d8575d6
- Cache FetchContent dependencies in CI workflows
  robtaylor
  committed
  a4b2238
- Add CI performance regression detection for GPU backends
  robtaylor
  committed
  ff06a00
- Add cuDSS benchmark comparison and CI performance summaries
  robtaylor
  committed
  6bbbe10
- Replace Metal potrf/trsm CPU fallbacks with MPS GPU implementations
  robtaylor
  committed
  c2b751a
- Address code review: cache status buffer, add edge-case tests
  robtaylor
  committed
  4796175
- Implement batched Metal factorization, solve, and benchmarks
  robtaylor
  committed
  09b144e
- Address code review: fix solveLt test, add potrf status checks
  robtaylor
  committed
  78bbda8
Commits on Feb 28, 2026
Commits on Mar 1, 2026
Commits on Mar 2, 2026
- Port LU sparse elimination kernels to CUDA
  robtaylor
  committed
  43150f8
- Fix lu_bench GPU mirror re-allocation in refinement loop
  robtaylor
  committed
  bab6b75
- Batch Metal solve dispatches: 7.6x solve speedup
  robtaylor
  committed
  5c82c1e
- Batch pivot upload in Metal solve: eliminate per-lump sync
  robtaylor
  committed
  7eda82c
- Optimize Metal Cholesky factor: three-tier potrf routing + batch commands
  robtaylor
  committed
  dabeed2
- GPU-resident pivots: eliminate per-lump commitAndWait in Metal LU factor
  robtaylor
  committed
  fb86c5c
- Document Level-Set Concurrent Dispatch investigation (Step 6)
  
  robtaylor
  and
  claude
  committed
  4e7fcf2
- CPU BLAS fallback for Metal dense LU: 8.8x factor speedup
  robtaylor
  committed
  bdd0f77
Commits on Mar 3, 2026
Commits on Mar 4, 2026
- Fix CUDA batched GEMM: atomicAdd for correctness + thread utilization
  robtaylor
  committed
  abded36
- CUDA LU: CPU BLAS dense fallback + lazy readValue cache (237ms → 3ms)
  robtaylor
  committed
  d8f703c
Commits on Mar 5, 2026
Commits on Mar 8, 2026
- Metal: remove CPU BLAS fallbacks from factor + batched factor
  robtaylor
  committed
  f2c5307
- Metal: remove CPU Eigen fallbacks from LU solve methods
  robtaylor
  committed
  0bddeeb
- Metal: add 5 GPU kernels for Cholesky dense solve
  robtaylor
  committed
  572a2ac
- CUDA: remove cpuBlasMode_ infrastructure
  robtaylor
  committed
  71c7212
- GPU maxAbsDiag reduction: replace CPU readValue loop with GPU kernel
  robtaylor
  committed
  a9b55c8
- Metal: remove inter-phase CPU sync from solve paths
  robtaylor
  committed
  c0aca15
- Remove BackendRef from public API
  robtaylor
  committed
  7111507
- Update docs for pure GPU backend architecture
  robtaylor
  committed
  4502b9b
Commits on Mar 9, 2026
- feat: Graph compatibility preparations for LU factorization
  robtaylor
  committed
  7b5f2c9
- feat: Replace cuSOLVER getrf with graph-capture-compatible LU kernel
  robtaylor
  committed
  f196193
Commits on Mar 10, 2026
Commits on Mar 11, 2026
Commits on Mar 12, 2026
Commits on Mar 13, 2026
Commits on Mar 14, 2026
- perf: Replace MPS LU with custom Metal kernel to eliminate encoder transitions
  robtaylor
  committed
  2ddb608
- refactor: Remove BASPACHO_METAL_PROFILE per-kernel profiling infrastructure
  robtaylor
  committed
  6747ecd
Commits on Mar 15, 2026
Commits on Mar 16, 2026
Commits on Mar 17, 2026
Commits on Mar 23, 2026
Commits on Mar 24, 2026
- feat: add SpruxFFISolver — encapsulated LU solver for circuit simulation FFI
  robtaylor
  committed
  7b5533d
- fix: support SPRUX_METALLIB_PATH env var for runtime metallib discovery
  robtaylor
  committed
  5d2152a
Commits on Mar 25, 2026
Commits on Mar 26, 2026

Commits

Commits on Jan 2, 2026

Commits on Jan 3, 2026

Commits on Feb 23, 2026

Commits on Feb 24, 2026

Commits on Feb 25, 2026

Commits on Feb 27, 2026

Commits on Feb 28, 2026

Commits on Mar 1, 2026

Commits on Mar 2, 2026

Commits on Mar 3, 2026

Commits on Mar 4, 2026

Commits on Mar 5, 2026

Commits on Mar 8, 2026

Commits on Mar 9, 2026

Commits on Mar 10, 2026

Commits on Mar 11, 2026

Commits on Mar 12, 2026

Commits on Mar 13, 2026

Commits on Mar 14, 2026