https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122767
Alexandre Oliva <aoliva at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Ever confirmed|0 |1
Status|UNCONFIRMED |ASSIGNED
Assignee|unassigned at gcc dot gnu.org |aoliva at gcc dot
gnu.org
Last reconfirmed| |2025-11-20
--- Comment #2 from Alexandre Oliva <aoliva at gcc dot gnu.org> ---
Created attachment 62855
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62855&action=edit
patch that makes HARD_REG_SET hashing insensitive to word size and endianness
This first-cut blunt patch allows a 64-to-32-bit stage1, 32-bit stage[23]
x86_64 toolchain to pass the bootstrap compare test, that otherwise fails for
cp/call.o.
If we settled on a canonical endianness (the target's?), normalization could be
done on a 32-bit word-by-word basis instead, possibly involving byte reversal.
But depending on endianness, we can bypass normalization altogether, and simply
hash a prefix of HARD_REG_SET, which might justify making that the normalized
form. WDYT?
