GitLab

Update the register-based memhub to the last specifications

See merge request !267

Laurent Petre authored May 12, 2022 and Camilla Galloni committed May 17, 2022

More specifically, we follow the new system where the magic register is
independent of the address table (i.e. always the last register in the
GEM block address space) and filled with a status code describing the
outcome of the last transaction.

Implement a RAW file reader

See merge request !263

The reader class reads one event at a time from a RAW file. The reader
is able to automatically detect if the file is uncompressed or ZSTD
compressed. In the later case, it is decompressed on the fly.

Usability gemrpc updates

See merge request !229

Add a cyclic mode to the latency scan

Closes #211

See merge request !266

Allow setting a common VFAT threshold for physics runs

Closes #210

See merge request !265

Fix doxigen

Optimize the S-curve scan speed

See merge request !264

Also enable the LTO only for release/optimized builds.

The speed up is the results of two operations:

1. Minimize the costly LMDB register lookups

2. Avoid re-writing the same registers with the same value

This new LMDB organization provides a much neater system. First of all,
the API is significantly clean-up, not relying on string parsing and
magical indexes anymore. Additionally, a significant performance
improvement has been seen on register lookups.

Such an implementation simplifies the code path and allows two
improvements:

* implement atomic Read-Modify-Write transactions;
* reduce the number of Lock-Unlock operations, potentially improving the
  performances.

The 8b10b comma character is sent once per orbit (3564 BX), on BC0.

After a trigger links reset, one must ensure to wait enough time before
checking for link errors to ensure that the data stream is properly
aligned.

Implement S-curve scan

Closes #130

See merge request !216

Misc local readout application updates

Closes #205 and #204

See merge request !262

This includes:

* The FED ID register used in the DAQ data format
* The source and destination MAC addresses used by the local readout

Note the slightly different behavior between the µTCA and ATCA
(including the CVP13) based systems. For the former the FED ID is
limited to 12-bits and includes the slot number while for the latter the
FED ID is a 32-bits number and the slot number is irrelevant.

Such a feature is required to keep the switch FIB up-to-date and
optimize the data flow in the network.

The current policy is the following:

* Only C code is allowed in the data plane (i.e. the code running on the
  lcores)

* C++ code is allowed in control plane (e.g. control threads,
  initialization routines,...)

* Rename the "source" global variable into "queue". The new wording
  better matches the scope and purpose of the structure.
* Add file Doxygen comments.
* Fix comments formatting.

This is required for installation on the p5 cluster where Python 3.6.8
is provided by the python36 package from the @epel repository while the
standard CentOS 7.9 installation provide Python 3.6.8 in the python3
package from the @updates repository.

The DPDK framework is already provided in the official @extra repository
under the name dpdk-*. When trying to install custom packages with the
same name via Yum repositories, the packages can be ignored due to
repository priorities or protections.

This is the case on the p5 cluster. In agreement with the sysadmins, we
chose the clean(-ish) solution to rename the packages as gem-dpdk-*.

Update OptoHybrid register names to last firmware version

See merge request !259

Fix RPM installation at p5

See merge request !258