https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106316
sandra at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |sandra at gcc dot gnu.org
--- Comment #3 from sandra at gcc dot gnu.org ---
I see a related example in the OpenMP 5.2.2 examples document at 6.14.7,
"Declare Target Directive with device_type Clause". I agree with Jakub,
"declare target" is supposed to be doing this, not "declare variant". I can't
find anything in the documentation for "declare variant" that indicates that it
should be doing any inference about the variant function definition based only
on the name of the function being used in a "declare variant" directive.
OTOH, it does seem that GCC is not honoring the host/nohost thing on "declare
target" either -- contrary to the introductory paragraph of the section that
says
"Only a device version of the procedure foo() is made available", it appears in
the .optimized dump for both host and offload target.
