Method and arrangement for streaming data profiling
View Patent ↗A circuit arrangement includes a plurality of functional units each of which comprises a plurality of data processing modules and a local controller. The plurality of data processing modules run a common system clock and are connected by a streaming data bus running a handshake-type streaming data transfer protocol. A profiling module of the circuit arrangement assesses control signals tapped at predefined interfaces of the streaming data bus during real time operation, for determining link performance and communication patterns for profiling and debugging purposes, and hence constitutes a simple and low cost approach for assessing intra-component and inter-component link performance and communication patterns on large SoCs. A method for profiling data flow for use in such a circuit arrangement is also provided.
1. A circuit arrangement comprising a plurality of functional units, each unit comprising a plurality of data processing modules and a local controller, said plurality of data processing modules running a common system clock and being connected by a streaming data bus running a handshake-type streaming data transfer protocol which comprises three control signals including first and second handshake-type control signals and a third control signal which marks a beginning and an end of a frame comprised of a logical group of data elements within the data stream,
wherein at least one of said functional units further comprises a profiling module which receives at least one set of said first, second, and third control signals tapped at a predefined link of said streaming data bus during real time operation, for determining link performance and communication patterns of said streaming data stream for profiling and debugging purposes;
wherein said profiling module comprises at least one link selection means, a respective profiler unit connected to each of said at least one link selection means and adapted to receive the system clock signal and the set of control signals tapped from the selected link to extract link performance parameters therefrom, and a software accessible set of control registers for defining a link to be assessed, a link performance acquisition mode, and a time period of the control signals to be assessed, and for storing link performance output parameters from said profiler units; and
wherein said profiler unit comprises a first counter for determining a number of frames that passed since a last clear operation, a second counter for determining a duration of the sampled frame, a third counter for determining a number of data beats of the sampled frame, and a counter control logic for controlling said first, second, and third counters for a selected frame to be sampled.
2. The circuit arrangement of claim 1 , wherein said profiler unit further comprises a first register connected to an output of the first counter, a first comparator connected between an output of the second counter and a second register, and second comparator connected between an output of the third counter and a third register, said counter control logic being adapted to configure said counters, comparators, and registers for a selected frame to be sampled and in function of a selected acquisition mode.
3. A method for profiling a data flow of streaming data for use in a circuit arrangement which comprises a plurality of functional units, each functional unit comprising a plurality of data processing modules and a local controller, said plurality of data processing modules running a common system clock and being connected by a streaming data bus running a handshake-type streaming data transfer protocol which comprises three control signals including first and second handshake-type control signals and a third control signal which marks a beginning and an end of a frame comprised of a logical group of data elements within the data stream, the method comprising the steps of:
providing a tapping line from a plurality of links of said streaming data bus between any pair of said data processing modules;
selecting a link to be assessed; selecting a link performance acquisition mode;
tapping said first, second, and third control signals from the selected link to be assessed;
determining a set of link performance parameters from said first, second, and third control signals and said common system clock for a predefined time period of the control signals; and
wherein the step of determining a set of link performance parameters comprises counting a number of frames that passed since a last clear operation, counting a number of clock ticks for a sampled frame, and counting a number of data beats of the sampled frame.
4. The method of claim 3 , wherein the step of determining a set of link performance parameters comprises, for a selected frame, determining minimum and maximum duration of the first control signal sampled asserted.
5. The method of claim 3 , wherein the step of determining a set of link performance parameters comprises, for a selected frame, determining minimum and maximum duration of the first control signal sampled de-asserted.
6. The method of claim 3 , wherein the step of determining a set of link performance parameters comprises, for a selected frame, determining minimum and maximum duration of the second control signal sampled asserted.
7. The method of claim 3 , wherein the step of determining a set of link performance parameters comprises, for a selected frame, determining minimum and maximum duration of the second control signal sampled de-asserted.
8. The method of claim 3 , wherein the step of determining a set of link performance parameters comprises determining a minimum or a maximum frame gap and the frame before the gap.