https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95590
Bug ID: 95590
Summary: OpenACC 'attach' behavior if already attached to
different data
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Keywords: openacc
Severity: normal
Priority: P3
Component: libgomp
Assignee: unassigned at gcc dot gnu.org
Reporter: tschwinge at gcc dot gnu.org
CC: jakub at gcc dot gnu.org, jules at gcc dot gnu.org
Target Milestone: ---
Created attachment 48709
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=48709&action=edit
'mdc-3_.c'
Consider the 'mdc-3_.c' test case I'm attaching, in particular its 'data_work =
data_2' case, and then OpenACC 2.6, 2.7.2. "Data Clause Actions", "Attach
Action", which states that "If the _attachment counter_ for _var_ ['data_work']
is nonzero [true; has been attached to 'data_1'] and the pointer in device
memory already points to the device copy of the data in _var_ [false; still
points to 'data_1_d'], [not relevant]. Otherwise, the pointer in device memory
is _attached_ to the device copy of the data by initiating an update for the
pointer in device memory to point to the device copy of the data and setting
the _attachment counter_ for the pointer _var_ to one. I read that to mean
that after 'data_work = data_2' has been set, the 'acc_attach((void **)
&data_work)' should then "drop" the attachment of the device copy of
'data_work' to currently 'data_1_d', and "re-attach" to the new '*data_work',
that is, to 'data_2_d', and re-initialize the _attachment counter_ to one.
(PGI 19.10 seems to agree with my interpretation, for what that's worth.)
