[all-commits] [llvm/llvm-project] 3464b0: [AMDGPU][Verifier] Mark calls to entry functions a...

[Shilei Tian via All-commits]

  Branch: refs/heads/users/shiltian/dont-allow-call-to-amdgpu-kernel
  Home:   https://github.com/llvm/llvm-project
  Commit: 3464b0622a9c1ce6c6a67fc8936472516e599a66
      https://github.com/llvm/llvm-project/commit/3464b0622a9c1ce6c6a67fc8936472516e599a66
  Author: Shilei Tian <i at tianshilei.me>
  Date:   2025-04-09 (Wed, 09 Apr 2025)

  Changed paths:
    M llvm/include/llvm/IR/CallingConv.h
    M llvm/lib/IR/Verifier.cpp
    M llvm/test/CodeGen/AMDGPU/attributor-flatscratchinit.ll
    R llvm/test/CodeGen/AMDGPU/call-to-kernel-undefined.ll
    R llvm/test/CodeGen/AMDGPU/call-to-kernel.ll
    A llvm/test/Verifier/call-to-non-callable-functions.ll

  Log Message:
  -----------
  [AMDGPU][Verifier] Mark calls to entry functions as invalid in the IR verifier

For AMDGPU, calls to entry functions are invalid. Previously, due to certain
limitations, this restriction was not enforced by the IR verifier. These
limitations have now been resolved, enabling us to enforce this check.

Adding target-dependent checks directly into the IR verifier is not ideal.
However, a cleaner solution, such as a dedicated target-dependent IR verifier,
is underway (e.g., https://github.com/llvm/llvm-project/pull/123609). Once that
or similar code is merged, we can move this check accordingly.



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