KernelBench operators
Faster GPU kernels for PyTorch operators, exactly as correct as the reference (placeholder)
Comparing changes
Choose two branches to see what changed, then open a pull request to bring the work across.
Able to merge. These branches can be merged automatically.
Sign in to open a pull request.
8 commits
+16 −0 routes/autotune-search/README.md added View file
| @@ -0,0 +1,16 @@ | ||
| 1 | +--- | |
| 2 | +tags: search | |
| 3 | +--- | |
| 4 | +# autotune-search | |
| 5 | + | |
| 6 | +Search tile sizes and warps per problem shape. | |
| 7 | + | |
| 8 | +## What happened | |
| 9 | + | |
| 10 | +- 2026-10-08: 1.61 speed-up (+0.61 on the baseline) | |
| 11 | +- 2026-10-12: 1.58 speed-up (+0.58 on the baseline) | |
| 12 | +- 2026-10-15: 1.74 speed-up (+0.74 on the baseline) | |
| 13 | + | |
| 14 | +## Next | |
| 15 | + | |
| 16 | +Bring the CUDA matmul tiling in. | |
+17 −0 routes/cuda-tiling/README.md added View file
| @@ -0,0 +1,17 @@ | ||
| 1 | +--- | |
| 2 | +issue: 2 | |
| 3 | +status: done | |
| 4 | +tags: systems | |
| 5 | +--- | |
| 6 | +# cuda-tiling | |
| 7 | + | |
| 8 | +Shared-memory tiling for matmul and conv in CUDA. | |
| 9 | + | |
| 10 | +## What happened | |
| 11 | + | |
| 12 | +- 2026-09-19: 1.15 speed-up (+0.15 on the baseline) | |
| 13 | +- 2026-09-27: 1.29 speed-up (+0.29 on the baseline) | |
| 14 | + | |
| 15 | +## Next | |
| 16 | + | |
| 17 | +Merge into the fused-attention route. | |