Repository navigation

[ROCm] Fused 4-bit SIMT GEMM - #1979

Merged

matthewdouglas merged 13 commits into

bitsandbytes-foundation:mainfrom

sstamenk:rdna_simt

Jul 7, 2026

Merged

[ROCm] Fused 4-bit SIMT GEMM#1979
matthewdouglas merged 13 commits into
bitsandbytes-foundation:mainfrom
sstamenk:rdna_simt

Commits on Jun 18, 2026
- Implement SIMT kernel support for RDNA
  sstamenk
  committed
  abc147c
- CUDA path fix
  sstamenk
  committed
  552b2eb
Commits on Jun 19, 2026
- Use LDS LUT uniformly across architectures and data-types
  sstamenk
  committed
  7641ded
- Restore .gitignore
  sstamenk
  committed
  483a733
Commits on Jun 23, 2026
- Optimize HIP fp16 SIMT accumulation
  sstamenk
  committed
  6dbc24d
Commits on Jul 1, 2026
- Instinct optimizations
  sstamenk
  committed
  9bf5c10
- Fix linting
  sstamenk
  committed
  f422447
Commits on Jul 3, 2026
- Make changes ROCm < 7 compatible
  sstamenk
  committed
  2858734
- Cleaup
  sstamenk
  committed
  be3df71
- Use __GFX9__ for warp size checks
  sstamenk
  committed
  38e2323
- Drop HIP version compatibility gating
  sstamenk
  committed
  f076b93
- Merge branch 'main' into rdna_simt
  sstamenk
  authored
  8ffb5ef
- Forward declare HIP bf16 structs
  sstamenk
  committed
  ca84430