RMA WG 10 18 2018

Agenda

(Naveen) The main update: addition of semantic check for overlapping the destination buffer and signal register buffer (sig_addr and dest may not be same or overlapping data objects).
(Jim) Data object is not defined in a precise way. Could use "multiple bytes" or "non-overlapping".
(Jim) Not a good idea to follow the broadcast spec descriptions.
(Dave) "Overlapping" is also used for communication/computation so might add "memory" with it.

(Dave) "restrict" keyword is not in C++ standard. Might have to remove it.
(Jim) Or specify its limitation.
(Dave) Will wait for Nick's opinion as he proposed "restrict" initially.
(Anshuman) Suggest to use template.
(Jim) SOS has used generic binding for c++.
(Dave) (Reading the spec).
(Jim) (Some minor suggestions on the text:) variable declaration outside a loop, a comma position, etc.
(Dave) Wait set might appear new and strange to a reader in a later paragraph.
(Jim) Reader should be able to read it twice to understand it.
(Jim) There is an inconsistency between API description and return value description of shmem_test_any regarding the return value.
(Dave) Will do reading in the next spec meeting.