[PATCH] D69771: [APFloat] Handle exponent underflow correctly
Ehud Katz via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Nov 18 23:35:13 PST 2019
ekatz marked 2 inline comments as done.
ekatz added inline comments.
================
Comment at: llvm/include/llvm/ADT/APFloat.h:146
/// A signed type to represent a floating point numbers unbiased exponent.
- typedef signed short ExponentType;
+ typedef int32_t ExponentType;
----------------
efriedma wrote:
> What's the effect of this on the size of APFloat on a 64-bit target? On a 32-bit target?
`APFloat` remains the same size. I added a note about that in the description. Maybe it is not clear enough.
`detail::IEEEFloat` is the only record that has a field of `ExponentType` type. It remains the size of 24 bytes and alignment of 8 bytes (in both 32 bit and 64 bit targets).
================
Comment at: llvm/lib/Support/APFloat.cpp:994
+ // Save the addend now, before we change our properties, as it may be equal to
+ // 'this' pointer.
+ if (addend)
----------------
efriedma wrote:
> Does this have some functional effect? If it does, is there a test?
Now that I think of it, I'll just separate it into another patch. It is a bug, revealed from the tests I've made.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D69771/new/
https://reviews.llvm.org/D69771
More information about the llvm-commits
mailing list