v1.1.16 — inline argument forwarding (#228 secondary)#230
Merged
Conversation
…spill+reload Tier 0 follow-up to #228's secondary finding. inline_calls_in_block spilled EVERY param to a fresh temp; when the arg was a bare `local.get K` this produced a `local.set TEMP; … local.get TEMP` round-trip. simplify_locals cleans that in straight-line callers but BAILS on any control flow (lib.rs ~8905), so in CF callers (the gust hot path) the copy survived. Forward a trailing `local.get K` arg directly into the inlined body when the callee does NOT write that param (callee_param_writes). Sound: K isn't rewritten before the inlined body, and the body never writes K (param not written; callee locals remap to a disjoint range above param_start_idx). Forwardable args are a top-suffix (only bare local.gets pop contiguously off the value stack). remap_locals_in_block now takes a per-param target array. Z3 verify-or-revert stays the backstop. Tests: test_inline_arg_forwarding_no_redundant_copy (2 local.get args → 0 spills); test_inline_arg_forwarding_skips_written_param (written param stays spilled, no caller-local clobber). 416 lib tests green; multisite + mix still inline + validate. Refs #228 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Forward a bare `local.get K` argument directly into the inlined body when the callee doesn't write that param, instead of spilling to a temp + reloading. Cleans the redundant arg-copy that survived in control-flow callers (where simplify_locals bails). Z3-gated; sound (callee_param_writes guards written params). No behavior change beyond cleaner inlined code. Refs #228 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Tier 0 of the post-v1.1.15 inliner feature loop. Z3-gated, no behavior change beyond cleaner inlined code.
#228 secondary finding: inlining spilled EVERY callee param to a fresh temp, so a bare
local.get Karg becamelocal.set TEMP; … local.get TEMP.simplify_localscleans that in straight-line callers but bails on any control flow, so in CF callers (the gust hot path) the copy survived.Fix: forward a trailing
local.get Karg directly into the inlined body when the callee does NOT write that param (callee_param_writes). Sound — K isn't rewritten before the inlined body, and the body never writes K (param unwritten; callee locals remap to a disjoint range). Forwardable args are a value-stack top-suffix;remap_locals_in_blocknow takes a per-param target array. The Z3 verify-or-revert stays the correctness gate.Validation: new
test_inline_arg_forwarding_no_redundant_copy(2local.getargs → 0 spills) +test_inline_arg_forwarding_skips_written_param(written param stays spilled, no caller-local clobber). 416 lib tests green; multisite + mix repros still inline + validate. Merge only through passing required checks; the two known-infra reds (Rocq-upstream, Verification-Gate comment-403) don't block.Refs #228