The address information is now very bursty a first in

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: traced, or just particular routines. Trigger conditions can be specified, and the trace can be collected before or after the trigger or with the trigger in the centre of the trace. Data accesses can be selected to be included in the trace or not, and the trace may collect just the address of the data access, or just the data itself, or both. Embedded trace options As noted above, the Embedded Trace Macrocell may be synthesized in several different configurations allowing the functionality of the unit to be traded off against cost (measured in terms of the numbers of gates and pins used). The minimal system requires 5 pins to issue pipeline information and 4 pins to issue data (in addition to the 5-pin JTAG interface). This is sufficient for execution tracing, but will only support very limited data tracing and has restricted triggering and filter ing capabilities. A 9-byte FIFO is used to smooth the data transfer rate, and the hard ware cost of this implementation is approximately 15 Kgates. A maximal system uses 5 pins to issue pipeline information and 16 pins to issue data (again, in addition to the 5-pin JTAG interface). It is capable of tracing the flow of execution and all but the very-worst-case data activity. A 40-byte FIFO is used to smooth the data flow, and the hardware cost is approximately 50 Kgates. Signal processing support 239 Between these two extremes several intermediate configurations are possible. All allow for external inputs (that is, inputs from other logic on the chip) to control the trace triggering, and for triggering from the EmbeddedlCE breakpoint logic. Trace overflow With all of the implementations of the Embedded Trace Macrocell there are some circumstances where the trace FIFO buffer can overflow. The unit can be configured either to stall the processor or to discontinue tracing when this happens. In either case, real-time tracing is lost, although only temporarily while the FIFO drains. Fundamentally what has happened is the information bandwidth has exceeded the capability of the compression algorithm to reduce it to match the b...
View Full Document

This document was uploaded on 10/30/2011 for the course CSE 378 380 at SUNY Buffalo.

Ask a homework question - tutors are online