https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127974
Bug ID: 127974
Summary: Float'Valid causes an unnecessary spill
Product: gcc
Version: 16.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: ada
Assignee: unassigned at gcc dot gnu.org
Reporter: liam at liampwll dot com
CC: dkm at gcc dot gnu.org
Target Milestone: ---
Created attachment 65880
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65880&action=edit
fix
Right now Float'Valid will always cause a spill. This is trivial to fix, as
shown in the attached patch. This does change the behaviour a tiny bit, but it
seems like the old behaviour that could change the sign when calling 'Valid is
incorrect, especially because it will cause a crash if the float is in rodata.
I found this while trying to improve performance of some float-heavy code where
I'd like to leave checks turned on.
As an aside, 'Exponent and 'Fraction are also really slow since they force a
non-inlined call to Decompose, but adding a fast path to them is a bit
involved.
