Qorivva MPC5xxx/SPC5xx Debugger and NEXUS Trace
57
©1989-2021 Lauterbach GmbH
On-chip timestamps are controlled with the command
.
Trace Based Run-time Measurement for off-chip Parallel NEXUS
For parallel NEXUS trace ports, the NEXUS adapter stores one NEXUS trace message per trace record.
Additionally there is only little and rather constant delay between event and message output, therefore the
tool generated timestamps
will yield in very exact results. The resolution of the timestamps mostly
depends on the time the trace port needs to send out a single message, i.e. the trace port width (number of
MDOs). As a rough estimate, the resolution will be 2~3 MCKO clock cycles for trace ports with >=12 MDOs,
and 8~16 MCKO clock cycles for trace ports with 2 or 4 MDOs.
If a better resolution is required,
processor generated timestamps
can be used for parallel NEXUS trace
ports, if supported by the processor. The resolution of the processor generated timestamp is 1/<core
frequency>. Please be aware that processor generated timestamps add 25~35% to the required trace port
bandwidth, which may cause message loss due to on-chip message FIFO overflows.
Trace Based Run-time Measurement for off-chip Aurora NEXUS
Due to the nature of NEXUS via high-speed serial trace port, up to three messages can be stored in one
trace record. Due to this, run-time measurements with
tool generated timestamps
will result in less
precision on sub-function level or when trace filtering is performed. Run-time measurements on function-
level, as well as average measurements usually do not lose precision. It is recommended to not use trace
filtering if a run-time measurement is performed.
If a better resolution is required,
processor generated timestamps
can be used for Aurora NEXUS trace
ports, if supported by the processor. The resolution of the processor generated timestamp is 1/<core
frequency>. Please be aware that processor generated timestamps add 25~35% to the required trace port
bandwidth, which may cause message loss due to on-chip message FIFO overflows.
Trace Based Run-time Measurement for on-chip Trace / Trace-to-memory
In this configuration, the generated trace data is directly stored on the processor, either in the small on-chip
trace buffer or the bigger emulation memory (if emulation device available). The debugger transfers the
recorded trace data through JTAG to the PC after the recording took place. Therefore
tool generated
timestamps
are not available.
Run-time measurements using the on-chip trace buffer are only possible if the processor supports
processor generated timestamps
. Processor generated timestamps increase the amount of generated
trace data by 25~35%. Unlike the off-chip trace solutions, the trace bandwidth to the on-chip trace buffer is
usually sufficient enough to transfer the timestamp information without message loss. Due to the limited
amount of buffer size (up to 2MBytes on emulation devices), recording the timestamp information will
noticeably reduce the amount of trace events stored in the on-chip trace buffer.
