[CUDA] Use monotonic ordering for __nvvm_atom* builtins (#185822)

CUDA's __nvvm_atom* builtins are expected to produce atomic operations
with relaxed ordering. However, Clang lowered tham as atomicrmw and cmpxchg
with the default seq_cst ordering. That mismatch went unnoticed because
until recently NVPTX back end was unable to lower all atomic instructions correctly,
and despite using `cst_seq` ordering in IR we ended up generating the intended
PTX instructions with relaxed ordering, It worked well enough until
https://github.com/llvm/llvm-project/pull/179553 implemented correct NVPTX
atomic lowering.
That, in turn, caused severe performance regression for the code that
relied on these builtins.

Thanks to @akshayrdeodhar for figuring out what happened.

Switching __nvvm_atom* builtins to generate atomic instructions with
monotonic ordering matches the expected semantics of the builtins,
and restores performance of the generated code.

See:
https://github.com/llvm/llvm-project/pull/179553#issuecomment-4035193968
5 files changed
tree: 6d4cddf4aabcd0c7b524070305800537e2b38652
  1. .ci/
  2. .github/
  3. bolt/
  4. clang/
  5. clang-tools-extra/
  6. cmake/
  7. compiler-rt/
  8. cross-project-tests/
  9. flang/
  10. flang-rt/
  11. libc/
  12. libclc/
  13. libcxx/
  14. libcxxabi/
  15. libsycl/
  16. libunwind/
  17. lld/
  18. lldb/
  19. llvm/
  20. llvm-libgcc/
  21. mlir/
  22. offload/
  23. openmp/
  24. orc-rt/
  25. polly/
  26. runtimes/
  27. third-party/
  28. utils/
  29. .clang-format
  30. .clang-format-ignore
  31. .clang-tidy
  32. .git-blame-ignore-revs
  33. .gitattributes
  34. .gitignore
  35. .mailmap
  36. CODE_OF_CONDUCT.md
  37. CONTRIBUTING.md
  38. LICENSE.TXT
  39. pyproject.toml
  40. README.md
  41. SECURITY.md
README.md

The LLVM Compiler Infrastructure

OpenSSF Scorecard OpenSSF Best Practices libc++

Welcome to the LLVM project!

This repository contains the source code for LLVM, a toolkit for the construction of highly optimized compilers, optimizers, and run-time environments.

The LLVM project has multiple components. The core of the project is itself called “LLVM”. This contains all of the tools, libraries, and header files needed to process intermediate representations and convert them into object files. Tools include an assembler, disassembler, bitcode analyzer, and bitcode optimizer.

C-like languages use the Clang frontend. This component compiles C, C++, Objective-C, and Objective-C++ code into LLVM bitcode -- and from there into object files, using LLVM.

Other components include: the libc++ C++ standard library, the LLD linker, and more.

Getting the Source Code and Building LLVM

Consult the Getting Started with LLVM page for information on building and running LLVM.

For information on how to contribute to the LLVM project, please take a look at the Contributing to LLVM guide.

Getting in touch

Join the LLVM Discourse forums, Discord chat, LLVM Office Hours or Regular sync-ups.

The LLVM project has adopted a code of conduct for participants to all modes of communication within the project.