trunk/84e524623ea4754a748936bf1ba6ecaaa92c3ae6: add kernel shared memory get_graph_data() (#196527)
- PyTorch: 814 events in the last 90 days
- PyTorch: 803th Release in the last 90 days
- Previous: earlier the same day · viable/strict/1789170654: [MPS] Migrate softmax to metal kernels (#196071)
What happened
add kernel shared memory get_graph_data() shared_mem_bytes is the launch's dynamic shared memory and comes from the node params, right next to the dimensions already read there. It stays best-effort like the kernel-name lookup beside it: a failed params query reports None rather than raising, since a dump is a diagnostic and a missing number should not cost the caller the whole graph. Test Plan: LD_LIBRARY_PATH=/usr…
Summary assembled by rule from the sources below