The SKIP_PAYLOAD stage of decoders would not read the last word
(last_word_in = '1').
The <NAME>_MEMBER_END stage was removed from encoders/decoders, since it
could happen that the last aggregated member of a last collection entry
could also be the last type member overall, which would miss toggling
the "last_word_out" correctly. The logic was instead put inside the last
member stage (instead of defining a separate MEMEBR_END stage).
A Flag was not reset in a sub-branch that is taken when PUSH_MODE=FALSE
& DURABILITY=TRANSIENT_LOCAL, that prevented the RTPS writer from ever
sending HEARTBEATs.
All the necessary "glue" logic to convert ROS Data to a form that the
DDS/RTPS back-end can use is implemented in packages.
The ROS Node discovery information is statically generated in packages
(similar to the RTPS Participant Data), and a special dds writer is
implemented (ros_static_discovery_writer) that has this static data as
its only payload sample.
Some definitions are moved out of rtps_config_package to prevent
circular package dependency.
Since the Type Specific user facing entities did not actually wrap the
DDS entities, but connected to them through port signals, a more
semantically correct name would be "interface", since they are the user
facing interface of the DDS entities.
Since we disabled the 'expectsInlineQoS' Flag due to Cyclone DDS
compatibility, we no longer have access to the lifespan of the remote
writers.
The Endpoint Match Format was changed to also send the Lifespan Duration
from the rtps_builtin_endpoint.
ACK HANDLING
------------
If there are no remote reliable reader endpoints the new Cache Changes
are immediately ACKed/removed from the HC.
Internal Global ACK SN is set to SEQUENCENUMBER_UNKNOWN, and
only initialized on first reliable remote reader endpoint.
If all remote reliable reader endpints are unmatched the internal Global
ACK SN is set to SEQUENCENUMBER_UNKNOWN and all Cache Changes are
ACKed/removed from the HC.
PUSH MODE
---------
Historical DATA is no longer send on remote endpoint match, if PUSH mode
is FALSE.
TESTBENCH
---------
Level 1 Test2 of rtps_writer was split into 2 tests.
Apparently Cyclone DDS does not support Readers with the inline QoS
Flag.
Change rtps_test_package to generate only necessary PIDs (e.g. no
PID_EXPECTS_INLINE_QOS for writers)
The rtps_handler was droping SequenceNumberSets where the 0 bit was set.
The rtps_handler Test1 was also extended a bit to test for more invalid
variants.
Until now the rtps_builtin_endpoint was using the last parsed Locator as
the Locator of choice.
The rtps_builtin_endpoint was extended with a PREFER_MULTICAST boolean
generic that allows to influence which locators are used.
Test1 of rtps_builtin_endpoint was split in two, to test both settings
of the generic.
The FIND_NEXT_INSTANCE stage of the dds_reader was doing a 128-bit
comparison (Key Hash) in 1 clock cycle.
The comparison was split among 4 clock cycles.
The PARTICIPANT_STALE_CHECK stage of the rtps_builtin_endpoint was doing
4 consecutive 64-bit comparisons in the same clock cycle.
The stage was further sub-split to make one 64-bit comparison per clock cycle.
Simple Dual Port (Read and Write Port), and True Dual Port RAM
implementations, together with their respective Altera implementations
were added.
The 'arch' Architectures should have the same behaviour as the Altera
Implementations (single_port_ram was modified to achieve that)
Add a Key Hash Generator that just generates a Key Hash with the
verbatim contents of the Type Key Fields.
This is the case for all Types with a combined Key Field size less than
16 Bytes.
The REGISTER_INSTANCE and LOOKUP_INSTANCE DDS Writer Operations return
an Instance Handle. Until now this instance Handle was returned via the
Data output interface of the dds_writer (ready_out, valid_out,
data_out_last_word_out).
This was changed, so that the dds_writer directly outputs the Instance
Handle via a dedicated port while asserting the done Signal.
This has the result that the <TYPENAME>_writer_wrapper no longer
"exports" data out signals from the dds_writer.
All relevant instantiations and testcases are changed accordingly
Until now it was assumed that a serialized key is the PLAIN_CDR2
Big Endian encoding of a <TYPENAME>KeyHolder Object [DDS_XTYPES v1.3, 7.6.8]
(I.e. the same blob that computes the MD5 key hash).
Taking other DDS implementations as reference (e.g. Cyclone DDS), it
seems they are using a normal Payload containing a KeyHolder(<TYPENAME>)
Object [DDS_XTYPES v1.3, 7.2.2.4.7] as the serialized key.
The Key Holder Template (together with the Type1 and Type2
implementations) were updated to reflect this change.
A bug fix were the Key Hash was not reset on a 'PUSH_SERIALIZED_KEY'
opcode was also fixed (together with the testbench).
Add Entity that connects an input and output FWFT_FIFO to a Avalon MM
Interface.
NOTE: The Implementation has no sync process. It relies on all FIFO
Signals being synchronous. (May affect timing closure)
Using seperately compiled Libraries we interconnect two systems, and
test their communication and interaction.
A bug in rtps_builtin_endpoint was fixed (were if only the SUB data was
to be sent, it was never actually sent).
Templates with the pre-defined code hull for the reader_wrapper,
writer_wrapper, and key_holder were created.
A "cookbook" of how to expand this TEMPLATES to create type specific IDL
conversions was also added.
The "code generation" is kept general to allow a code generator to
automatically produce this wrappers in the future.
The writer_wrapper was modified to allow changing the used endian via
generic. This allows to test Little Endian handling in simulations.
The Test2 are basically Test1 with Little Endian
By connecting the output of the writer_wrapper to the input of the
reader_wrapper we can effectively test the encoding/decoding.
Since the writer_wrapper only uses Big Endian we cannot test Little
Endian handling.
Previously the simaple info data was valid for exectly 1 clock cycle,
in which the user had to pull the 'get_data' signal if he wanted the
associated payload.
This change allows the user to control how long the sample info data is
shown.
The inline-QoS of PDP Packets is parsed, and Participants/Endpoints can
be unmatched by Unregister/Dispose Status Info Updates.
Test 1,2,3 was extended to test this functionality.
NOTE: PID_KEY_HASH is a requirement (at least for unmatching Endpoints),
as serialized key payloads are unsupported.
The GAP parsing logic was backported from RTPS Reader, so that the RTPS
Builtin Endpoint can parse all kind of GAP without making assumptions.
The Test 4 of the RTPS Builtin Endpoint was extended to test HAERTBEAT and GAP Handling of the RTPS
Builting endpoint, and Test 2 of the RTPS Reader was extended with an
additional Test (GAP with next expected SN not marked in bitmap)
Until now the ACKNACK Response of the RTPS Builtin Endpoint had an empty
Bitmap, effectively ACKing SNs, but never NACKing any (i.e. not
requesting any).
Similar to the RTPS Endpoints we request the next 32 SNs on every
ACKNACK Response.
The ParticipantMessageData SN Handling was also implemented.
Backport Memory Controller and Memory FSM from RTPS/DDS Endpoints to
RTPS Builtin Endpoint.
The Memory is now using a single linked list and the FSM uses Frame Field
Flags.
The main FSM uses check_time to initiate stale checks, and the stale
checks are done in the main FSM.
Testbench was modified to accomodate the changes (Previous L0 Test4 was
removed and integrated in L0 Test 1).
The REMOVE_WRITER DDS Operation of DDS Reader now generates Samples if
the Instance State changes.
Updated testbench.
Added checks for DATA_AVAILABLE communication status.
Test Sample Rejected Status Handling.
Test GET_SAMPLE_REJECTED_STATUS DDS Operation.
Extend Sample Rejected Status Kind by REJECTED_BY_PAYLOAD_MEMORY_LIMIT
Test the DDS READ, TAKE, READ_NEXT_SAMPLE, TAKE_NEXT_SAMPLE,
READ_INSTANCE, TAKE_INSTANCE, READ_NEXT_INSTANCE, TAKE_NEXT_INSTANCE
Operations.
Backport fixes to previous testbenches.
Change handling of DISPOSE Samples of unknown Instances.
