PMU bandwidth report

The PMU Bandwidth report helps you identify routing bottlenecks. If there is a significant difference between the expected amount of data flowing through a route and the amount that’s actually measured on chip, there might be congestion in the preceding buffers.

Locate the report

The report is provided to you

In .XLSX format at /reports/collated_report.xslx in your output folder in the 'PMU Bandwidth' worksheet.
As a standalone CSV file at at /reports/bandwidth_report.csv in your output folder.

Read the output data

Column Name	Meaning
source_unit_name	The name of the PMU from which the buffer originates
event_name	The event name associated with the instrumentation counter
required_avg_bandwidth	The provisioned average bandwidth
bandwidth_vec/cycles	The bandwidth through a buffer, as measured on the RDU
slowdown_factor	The ratio of required_avg_bandwidth and bandwidth_vec/cycles
manhattan_distance	The distance between the buffer and its destination, measured by horizontal + vertical lengths
number_iterations	The number of iterations that PMU was run for, in reference to its parent
number_iterations_global	The number of iterations that PMU was run for, in reference to the whole section
bank_conflict_count	The raw instrumentation counter count

Column Name

Meaning

source_unit_name

The name of the PMU from which the buffer originates

event_name

The event name associated with the instrumentation counter

required_avg_bandwidth

The provisioned average bandwidth

bandwidth_vec/cycles

The bandwidth through a buffer, as measured on the RDU

slowdown_factor

The ratio of required_avg_bandwidth and bandwidth_vec/cycles

manhattan_distance

The distance between the buffer and its destination, measured by horizontal + vertical lengths

number_iterations

The number of iterations that PMU was run for, in reference to its parent

number_iterations_global

The number of iterations that PMU was run for, in reference to the whole section

bank_conflict_count

The raw instrumentation counter count

Interpret the data

If there’s a significant difference between what is expected as the amount of data flowing through a route and what’s actually measured on chip, congestion might be occurring in the preceding buffers.

Example screenshots

The SambaTune GUI displays the table of PMU bandwidth for every stage in every section executed on the RDU. Here are examples of table reports available on the SambaTune GUI: