Details
- Reviewers
philnik - Group Reviewers
Restricted Project - Commits
- rG81a3828dbd99: [libcxx] <experimental/simd> excluded long double for 32-bits x86 temporarily
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
libcxx/test/std/experimental/simd/test_utils.h | ||
---|---|---|
44 | It should be quite enough to just check for __i386__ here - I don't know of any case where __i686__ is defined but __i386__ isn't. FWIW, the previous condition of checking for #ifdef __MINGW32__ did cover 64 bit mingw too which I presume wasn't intended; the __MINGW32__ define is set for all mingw targets, even on 64 bit (just like _WIN32 is defined for all win32 api targets, regardless of whether they're 32 or 64 bit). |
Thanks. I've tested with just __i386__, and I can confirm I no longer get test failures in 32-bit build with this change. The 64-bit test run is still ongoing but I think it's unlikely for this to break it.
@Joy12138, could you merge it now? This is blocking us from testing new snapshots of LLVM.
I don't know anything about precommit CI. We're basically building it using C++ compiler set to i686-pc-linux-gnu-clang++ that's symlink to the amd64 build of clang (and via the filename magic forces -m32).
It should be quite enough to just check for __i386__ here - I don't know of any case where __i686__ is defined but __i386__ isn't.
FWIW, the previous condition of checking for #ifdef __MINGW32__ did cover 64 bit mingw too which I presume wasn't intended; the __MINGW32__ define is set for all mingw targets, even on 64 bit (just like _WIN32 is defined for all win32 api targets, regardless of whether they're 32 or 64 bit).