jax
16371553 - [Mosaic GPU] Second attempt at fixing `test_tcgen05_collective_mma`.

Commit
162 days ago
[Mosaic GPU] Second attempt at fixing `test_tcgen05_collective_mma`. We need to make sure operands in SMEM have been loaded on both blocks before issuing the MMA instruction. We also pass `orders_tensor_core=True` to cluster barrier arrive/wait calls so that it also applies to TensorCore operations. PiperOrigin-RevId: 796837258
References
Author
Parents
Loading