Skip to content

NewGVN is likely screwing up in the presence of lifetime.start and end #30859

@llvmbot

Description

@llvmbot
Bugzilla Link 31511
Version trunk
OS All
Blocks #30343
Reporter LLVM Bugzilla Contributor

Extended Description

MemorySSA is smart enough to understand that lifetime.start and lifetime.end don't actually alias anything.

However, this means when we make replacements of loads with stores, etc, we probably have situations where we are extending the life part the lifetime.start or end on either side.

I'll work up some testcases and test them

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugzillaIssues migrated from bugzillallvm:GVNGVN and NewGVN stages (Global value numbering)

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions