Repository navigation

cuda-oxide: move SIMT toolchain pin to stable 1.99.1 - #1609

Open

nihalpasham wants to merge 104 commits into

NVIDIA:mainfrom

nihalpasham:feat/stabilize-toolchain-1.99.1

Open

cuda-oxide: move SIMT toolchain pin to stable 1.99.1#1609
nihalpasham wants to merge 104 commits into
NVIDIA:mainfrom
nihalpasham:feat/stabilize-toolchain-1.99.1

Commits on Sep 18, 2026
Commits on Oct 1, 2026
- feat(bench): autotune cuBLASLt baselines and record B200 results
  nihalpasham
  committed
  ffffcbe
Commits on Oct 11, 2026
- merge: migrate CuTe translation to the CUDA Rust layout
  nihalpasham
  committed
  14e6f4d
- feat: add CUTLASS primitives export profile
  nihalpasham
  committed
  2574f6c
- feat: select primitives profile through CUTLASS codegen
  nihalpasham
  committed
  257713c
- feat: add typed primitive memory and tensor maps
  nihalpasham
  committed
  5e49461
- feat: add explicit primitive resource lifecycles
  nihalpasham
  committed
  7453e9a
- feat: expose typed TCGen05 primitive instructions
  nihalpasham
  committed
  d475be9
- feat: expose typed TMA and synchronization primitives
  nihalpasham
  committed
  fd7e801
- test: verify primitives API through the CUTLASS compiler
  nihalpasham
  committed
  9c684c1
- feat(primitives): port persistent GEMMs and add SM120 MXFP4 support
  nihalpasham
  committed
  a101bcc
- docs(primitives): record RTX 5090 GEMM validation and timing
  nihalpasham
  committed
  787b13f
- perf(primitives): avoid checked shared-region result carriers
  nihalpasham
  committed
  f6e5519
- fix(translation): adapt examples and tests to rebased main
  nihalpasham
  committed
  0809425
- feat(examples): add C++ blockscale primitives comparison
  nihalpasham
  committed
  77fcde6
- refactor(examples): colocate blockscale GEMM references
  nihalpasham
  committed
  c8d3337
- Extract typed CUDA primitives into cuteir/ctm-rs
  nihalpasham
  committed
  96e7aa1
- Rename the blockscale Rust example to blockscale_gemm_ctm
  nihalpasham
  committed
  96bf070
- Move the FP16 CTM example into cuteir/examples
  nihalpasham
  committed
  1755d95
- Add a matching CUTLASS CTM blockscale reference
  nihalpasham
  committed
  5626ed6
- Move the CUTLASS CTM reference beside the Rust CTM example
  nihalpasham
  committed
  d108a74
- Improve FP16 CTM codegen and migrate CuTe primitives
  nihalpasham
  committed
  3a00962
- Name FP16 benchmark implementations explicitly
  nihalpasham
  committed
  53cc755
Commits on Oct 12, 2026
- Add SM120a attention examples and refine CTM primitives
  nihalpasham
  committed
  01364c5
- Compare FMHA examples against CuTeDSL by default
  nihalpasham
  committed
  fc51afd
- Add validated SM100 CuTe-RS attention examples
  nihalpasham
  committed
  238a2ef
Commits on Oct 14, 2026
- Modularize and document the CuTe implementation
  nihalpasham
  committed
  d642dcb
- Route SM100 CuTe attention primitives through CTM
  nihalpasham
  committed
  ab462fd
Commits on Oct 15, 2026
Commits on Oct 16, 2026
- Express Qwen GPU kernels through typed CuTe abstractions
  nihalpasham
  committed
  f1aa992
- Modularize CuTe IR crates and preserve defining identities
  nihalpasham
  committed
  05a4257
- Merge upstream SM100 CTM work into CuTe refactor
  nihalpasham
  committed
  c128512
- Organize CuTe models and document Qwen workloads
  nihalpasham
  committed
  1f87f3c
- Reduce Qwen projection and Talker execution overhead
  nihalpasham
  committed
  29466ea
- Clarify CuTe translation docs and clean local artifacts
  nihalpasham
  committed
  f5f0e95
Commits on Oct 17, 2026
Commits on Oct 18, 2026
Commits on Oct 19, 2026
- docs(cuda-oxide): add visual CuTeIR README and CuTe in Rust book section
  nihalpasham
  committed
  38f9940
- Merge remote-tracking branch 'origin/exp/cuteir-translation' into exp/cuteir-translation
  nihalpasham
  committed
  916662a
Commits on Oct 20, 2026
- feat(cuteir): add true NVFP4 (E2M1+E4M3) support and qwen3_8b_cute model
  nihalpasham
  committed
  2d62c43
Commits on Oct 21, 2026
Commits on Oct 22, 2026