https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124391
Bug ID: 124391
Summary: Surprising rounding of std::float16_t literals
Product: gcc
Version: 15.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: janschultke at googlemail dot com
Target Milestone: ---
https://godbolt.org/z/rMK5sqdr7
auto f16 = 512.749999999f16;
auto f32 = 4194304.749999999f32;
This generates the following:
"f16":
.value 24578
"f32":
.long 1249902593
Evidently, 512.749999999f16 is being rounded UP to 513/0x6002, despite
512.5/0x6001 being the mathematically closer nearest representable value.
This behavior is inconsistent with the rounding of std::float32_t.
4194304.749999999f32 is mathematically closer to 4194304.5/0x4a800001 than to
4194305/0x4a800002, so it gets rounded DOWN.
This behavior is also inconsistent with Clang, which rounds 512.749999999f16
DOWN to 512.5/0x6001.
