Important

This is a highly experimental tool, historically developed for internal use. It shall be considered as early alpha.

PML Molino

A prototype Graphical User Interface to explore PML analysis results and trace data.

Setup

• From the repository, install all required Python dependencies using poetry:

  poetry install --with dev

• From a package, install using Python pip

  pip install molino-0.4-py3-none-any.whl

Importing data

The GUI requires a compliant database in the root of the project named transactions.sqlite. This implies importing observation traces, collected on target, and corresponding PML interference analysis results. This is service is provided by the importer.ipynb view. Alternatively, example importers are included for both under the molino/importers package, with their usage under molino/tests/test_db.py.

Observations traces

The default observation trace importer (molino/importers/alf.py) supports the alftrace format. Thealf` trace format captures for each multi-transaction and initiator the number of cycles and the number of load/store instructions as a transaction count. The reference measurement for each transaction, in isolation, is provided in the file.

• TODO: Document the trace format

PML Interference analysis

The default PML analysis importer (molino/importers/pml.py) supports the txt output format. The txt format captures a list of transactions with their classification and arity captured in the filename. A file matching *_free_<arity:int>.txt (respectively *_itf_<arity:int>.txt) captures multi-transactions free from interference (respectively suffering from interference).

Entries are formatted as ||-separated (multi-)transaction names, one per line, each surrounded by <>. As an example:

< C0_RdCo_BK1 || C1_Rd_MPIC >

Usage

You can start a server using the molino_dashboard command which will serve the different views.

• TODO Document importer view
• TODO Document transactions view