[all-commits] [llvm/llvm-project] e3ba4c: DAG: Materialize a directly-lowered gc.relocate in...

Matt Arsenault via All-commits all-commits at lists.llvm.org
Fri Sep 11 02:16:49 PDT 2026


  Branch: refs/heads/users/arsenm/codegen/statepoint-relocate-eh-edge-liveness
  Home:   https://github.com/llvm/llvm-project
  Commit: e3ba4c8677108c306a3bdfc9fa152055f937ce6a
      https://github.com/llvm/llvm-project/commit/e3ba4c8677108c306a3bdfc9fa152055f937ce6a
  Author: Matt Arsenault <Matthew.Arsenault at amd.com>
  Date:   2026-09-11 (Fri, 11 Sep 2026)

  Changed paths:
    M llvm/include/llvm/CodeGen/FunctionLoweringInfo.h
    M llvm/lib/CodeGen/SelectionDAG/StatepointLowering.cpp
    A llvm/test/CodeGen/X86/statepoint-relocate-eh.ll

  Log Message:
  -----------
  DAG: Materialize a directly-lowered gc.relocate in its own block

Fixes machine verifier errors with -early-live-intervals.

A NoRelocate statepoint value is a directly-lowered leaf value. When its
gc.relocate is in another block, it was made available by exporting it from the
statepoint's block. For an invoke this defines the vreg after the call, which
does not dominate the use reached along the unwind edge, giving an invalid live
range (or a silent read of an undefined register without -verify-machineinstrs).

Since the leaf has no chain or operands, rematerialize it at the gc.relocate
instead of exporting it. undef already rematerializes there. A gc.relocate
value is always a pointer, so only the integer-constant and frame-index leaves
are handled.

Co-authored-by: Claude (Claude-Opus-4.8)



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications


More information about the All-commits mailing list