镜像站点 · 本页由第三方 GitHub 只读镜像提供,非 GitHub 官方站点,不接受任何登录或凭据输入。前往 github.com
Skip to content

JIT: lower environment reads and coercive operators inside OSR regions #1016

Closed

Description

@glennmichael123

Parents: #466, #469
Performance evidence: #473
Follows: #1014, #1015

Executed evidence

After e11e1a1 compacts the exact frozen representativeJson inner loop from 147 function-wide nodes to 43 region nodes, all loop lowerers reach operation selection and reject. The exact region inventory contains two load_var nodes (the JSON reads) and one bit_and node (the & 65535 update). Both already have exact effect frame states, exceptional targets, input arity, native runtime ABI dispatch, and straight-line lowering; appendBlockOperations simply omits them from region lowering and falls through to UnsupportedChunk.

Scope

  • Stage zero-input global environment reads through the existing exact effect-state/native-operation path inside selected regions.
  • Stage the already-supported unary/coercive and two-input operator families through rooted region scratch inputs, preserving operand order and exceptional targets.
  • Reuse the narrow per-op native ABI and its exact deopt state; do not hide the bytecode interpreter behind a region call.
  • Mark numeric results only from the existing required-numeric proof and validate them on return.

Acceptance

  • Focused OSR tests execute a global read and coercive binary operator with exact result, exception, step, and side-exit recovery behavior.
  • Moving-GC stack maps root every staged managed input and rewrite recovery state.
  • The exact representativeJson function publishes a native region or exposes a newly measured, separately filed boundary.
  • Forced JIT off/on output and checksums agree; no-GIL and TSan remain race-free.
  • Exact-parent attribution proves increased native transfer without a frozen control regression.

Activity

  1. glennmichael123 commented on Oct 2, 2026

    @glennmichael123
    ContributorAuthor

    Completed by 0efe89e, with reusable callback attribution added in a4d47ae. Current real-GC loop tests pass 14/14 and the no-GIL plus unsuppressed TSan seeded optimizer differential passes 3/3. Frozen production attribution preserves checksum 324952086, publishes two optimizer artifacts, records 17,113 OSR entries/deoptimizations, and reduces scored VM dispatch from 23,184,279 to 9,224,978. The new histogram reconciles exactly to 1,440,643 runtime callbacks: load_var 480,214; get_prop 480,215; call_with_this 480,214. This proves the environment/coercion staging is live and identifies the next independent stable-global/property/call residency boundary in #1024.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions