trunk/d4258aa05fc98e7852a6c78350d44e3fa7bdb2ab: Extend inline asm hop for multiple-return (#193434)
- PyTorch: 118 events in the last 90 days
- PyTorch: 117th Release in the last 90 days
- Previous: earlier the same day · trunk/9f3a72a53845f3abdda815ba2815b9446beac828: Skip bolt optimization for libtorch.so (#194941)
What happened
Human-authored context: Extend inline asm hop for multiple-return. AI-generated summary from Codex, reviewed by the author: Allow inline_asm_elementwise to accept a tuple of output dtypes and return multiple tensors from one assembly invocation. Inductor represents each result independently while Triton codegen shares the underlying invocation through kernel CSE. Mixed output dtypes and pack > 1 are supported, while…
Summary assembled by rule from the sources below