projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 99b7c2b) | patch
Revert "Currently the control of the eval-method is mixed with fast-math."
authorZahira Ammarguellat <zahira.ammarguellat@intel.com>
Thu, 9 Mar 2023 14:24:01 +0000 (09:24 -0500)
committerZahira Ammarguellat <zahira.ammarguellat@intel.com>
Fri, 10 Mar 2023 19:44:06 +0000 (14:44 -0500)
commit2f1264260b37e9bd79737181e459ae20e10c5fea
tree25fb7d8dd2a632ad4d613b51d8e19da911c30f42tree | snapshot
parent99b7c2bad3e5898353223c4682dce8ce31a69f30commit | diff
Revert "Currently the control of the eval-method is mixed with fast-math."

Setting __FLT_EVAL_METHOD__ to -1 with fast-math will set
__GLIBC_FLT_EVAL_METHOD to 2 and long double ends up being used for
float_t and double_t. This creates some ABI breakage with various C libraries.
See details here: https://github.com/llvm/llvm-project/issues/60781

This reverts commit bbf0d1932a3c1be970ed8a580e51edf571b80fd5.
clang/include/clang/Lex/Preprocessor.h diff | blob | history
clang/lib/Lex/PPMacroExpansion.cpp diff | blob | history
clang/lib/Lex/Preprocessor.cpp diff | blob | history
clang/lib/Sema/SemaAttr.cpp diff | blob | history
clang/test/CodeGen/X86/fexcess-precision.c diff | blob | history
clang/test/CodeGen/eval-method-fast-math.cpp [deleted file] blob | history
clang/test/Preprocessor/flt_eval_macro.cpp diff | blob | history
Domain: System / Toolchain;