trunk/3376c132fd39be20457f93ed99a15e03a11d7da3: [inductor] AOTI cpp wrapper: support invoke_subgraph (#194703) (#194703)
- PyTorch: 414 events in the last 90 days
- PyTorch: 411th Release in the last 90 days
- Previous: earlier the same day · trunk/312ef3ee1cd45fbee99f2e4e8f6035f05ceeebed: Add BF16x9 precision mode for CUDA FP32 matmul (#195301)
What happened
Summary: CppWrapperCpu.codegen_invoke_subgraph was a bare raise NotImplementedError , so any torch.compiler.nested_compile_region under AOTInductor died at codegen: torch._inductor.exc.InductorError: NotImplementedError: codegen invoke_subgraph is not implemented for cpp wrapper cpp_wrapper_cpu.py:2813 Two parts. 1. Implement CppWrapperCpu.codegen_invoke_subgraph . The class already has a complete subgraph-inlining…
Summary assembled by rule from the sources below