Kai Londenberg

[BUG] Illegal CUDA shared memory access in SM90 GEMM TMA Warpspecialized at ClusterBarrier::init

7

### Bug description When running the provided code as a standalone executable, a CUDA illegal memory access is reported. Using compute-sanitizer, I could pinpoint this to an illegal shared memory...

[Inductor cutlass backend] Fix cutlass_utils.get_max_alignment() for strided layouts.

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #121492 * #124928 * __->__ #124930 * #124929 Fixes cutlass_utils.get_max_alignment() which was so far not checking the alignment properly. Basically the method so...

[Inductor max autotune] Make autotune_select_algorithm more robust

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #124928 * #125406 This diff makes sure that a custom exception is thrown when no valid choices remain during autotuning. This allows...

[Inductor cutlass backend] Enabled nonzero workspace and Cutlass StreamK

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #124928 * __->__ #125406 Enable nonzero workspace and Cutlass StreamK for Inductor Cutlass GEMM ops. This is a simpler rewrite of my original...