Repository navigation
Open
Labels
enhancementNew feature or requestNew feature or requestperformanceMeasured runtime, memory, scaling, or build performanceMeasured runtime, memory, scaling, or build performancewasmWebAssembly runtime, compilers, and benchmarksWebAssembly runtime, compilers, and benchmarks
Description
Parent: #486
Depends on: #487, #469, #470, #481
Goal
Add a profiling optimizing WebAssembly tier for hot functions and loops, including bounds/alias/range optimization, inlining, and allocation/GC specialization.
Scope
- Build typed SSA with explicit memory/table/global/GC/exception/atomic effects and exact source/trap origins.
- Add constant folding, GVN/CSE, DCE, range analysis, loop optimization, redundant bounds-check elimination, address strength reduction, load/store forwarding, and block layout.
- Inline direct and guarded indirect/ref calls with code-size/compile-cost budgets; specialize JS import/export conversions and common runtime helpers.
- Optimize multi-memory/memory64/table/GC object operations only with correct alias, growth, relocation, and trap guards.
- Add OSR/tiering/deopt or safe replacement from baseline code, preserving exact Wasm stack/locals/handlers/GC roots.
- Coordinate vectorization with Wasm: lower fixed-width and relaxed SIMD to native vectors #488 and tiering/memory policy with JIT: optimize tiering policy, warmup, and compilation cost #481/Memory: bound heap, cache, and executable-code footprint #480.
No-workaround rules
- No hoisting across a possible trap/effect when observable order changes, stale memory base after growth, lost integer wrap/NaN behavior, assumed import purity, or unbounded compile/code growth.
Acceptance
- IR verifier and pass-specific positive/negative tests cover traps, aliases, growth, exceptions, GC, atomics, and recovery.
- Baseline/optimizer/interpreter differentials cover every optimized family on all declared backends.
- Full conformance, no-GIL, moving-GC, TSan, OOM/fault, streaming, and artifact-lifetime gates pass.
- Perf: add causal attribution and exact-parent regression infrastructure #461 reports compile time, code bytes, eliminated checks/operations, OSR/deopt, runtime calls, and memory.
- Expanded Bench: define the representative dependency-free performance matrix #460 Wasm rows meet Roadmap: make WebAssembly native and faster than JSC #486 goals without cold/compile/RSS regressions hidden by warmed throughput.
Foundation
Reuse the verified optimization/effect principles from #469, not JavaScript-specific value semantics.
Metadata
Metadata
Assignees
Labels
enhancementNew feature or requestNew feature or requestperformanceMeasured runtime, memory, scaling, or build performanceMeasured runtime, memory, scaling, or build performancewasmWebAssembly runtime, compilers, and benchmarksWebAssembly runtime, compilers, and benchmarks
Activity