tqchen commented on PR #741: URL: https://github.com/apache/tvm-ffi/pull/741#issuecomment-5546651689
## Speed Measured on TVM's `StructuralMap`-based substitute pass, split/fuse fixture, `-O3`, ns/node, medians of 11 pinned process runs. Both sides are self-consistent git states built identically. | | before | after | change | |---|---:|---:|---:| | hooked `StructuralMap` (replace) | 39.35 | **28.58** | **−27.4%** | Cross-confirmed on a standalone harness that mirrors the TIR node hierarchy — non-final `Expr` base with 64 child slots and a `ty` field, a `FreeVar`-kind `Var` taking an identity remap on every match, a `PrimExpr` view type, and binary ops carrying the result-type guard. Its engine cost over its own floor tracks TVM's to within 1% (+15.51 vs +15.33 ns/node), so the improvement is reproducible without a TVM build. The harness is not part of this PR; it lives on `bench-structural-mutate-harness`. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
