Bug-basher - #260
Conversation
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
CUDA's headers select implementations from _MSC_VER. Clava defined _MSC_VER to 1, forcing the legacy MSVC paths in crt/math_functions.hpp (int-returning isinf/isnan, glibc-style __signbitl calls) that conflict with clang's __DEVICE__ declarations. Define a real MSVC version (1930) plus -D_NATIVE_WCHAR_T_DEFINED (keeps clang's builtin __stddef_wchar_t.h from typedef'ing wchar_t), -D_LIBCPP_OBJECT_FORMAT_COFF (clang's CUDA host pass defines __ELF__ on Windows, so libc++ missed the COFF guard and emitted an aligned_storage<..., 16384> above the 8192-byte Windows alignment cap) and -D_ALLOW_UNSUPPORTED_LIBCPP (CUDA's crt/host_defines.h rejects libc++ on x86-64). Drop -fms-compatibility (emulates pre-VS2015, making char16_t/char32_t non-keywords and breaking libc++ 18 headers) and the no-op -D_LIBCPP_MSVCRT. Also exclude Standard.CUDA from the '-std' fallback so passing the CUDA standard no longer resurrects the rejected '-std=cuda' flag, and give the CUDA legacy tests a 60s timeout: parsing the builtin CUDA resources regularly exceeds the 5s vitest default.
The test takes ~14s on macOS runners and has been timing out at the 10s limit in every CI run.
|
|
@codex review |
|
Codex Review: Didn't find any major issues. Can't wait for the next one! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
1 participant
No description provided.