Issue 241706
Summary ldexp(ldexp(x, a), b) != ldexp(x, a + b) if subnormals may round
Labels floating-point
Assignees
Reporter ZERICO2005
    https://github.com/llvm/llvm-project/blob/main/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp#L3524-L3555

`ldexp` is always exact unless overflow/underflow occurs, or if a subnormal result is rounded. This means that ldexp(ldexp(x, a), b) != ldexp(x, a + b) since the former may round subnormals twice and the latter may round a subnormal just once. The transformation is probably ok under `assoc`
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to