Cost management against requirements for the generation of a NoC
Example implementations as described herein are directed to systems and methods for processing a NoC specification for a plurality of performance requirements of a NoC, and generating a plurality of NoCs, each of the plurality of NoCs meeting a first subset of the plurality of performance requirements. For each of the plurality of NoCs, the example implementations involve presenting a difference between an actual performance of the each of the plurality of NoCs and each performance requirement of a second subset of the plurality of performance requirements and one or more costs for each of the plurality of NoCs.
1. An apparatus, comprising:
a processor, configured to:
process a Network on Chip (NoC) specification for a plurality of performance requirements of a NoC;
generate each of a plurality of NoCs in a form of a Register Transfer Level (RTL), each of the plurality of NoCs meeting a first subset of the plurality of performance requirements;
for each of the plurality of NoCs, present a difference between an actual performance of the each of the plurality of NoCs and each performance requirement of a second subset of the plurality of performance requirements and one or more costs for each of the plurality of NoCs; and
for a selection of a NoC of the plurality of NoCs for construction, incorporating the RTL associated with the selected NoC into hardware.
2. The apparatus of claim 1 , wherein the plurality of performance requirements comprises bandwidth requirements and latency requirements for a plurality of sets of traffic of the NoC specification.
3. The apparatus of claim 1 , wherein the plurality of performance requirements comprises performance requirements of selected one or more subsets of traffic of the NoC specification.
4. The apparatus of claim 1 , wherein the plurality of performance requirements comprises multiple System on Chip (SoC) or NoC use profiles.
5. The apparatus of claim 1 , wherein the one or more costs comprises at least one of area, power consumption, and wiring congestion.
6. The apparatus of claim 1 , wherein the presenting the difference and the one or more costs comprises providing a visualization for the plurality of NoCs based on a function incorporating performance and cost.
7. The apparatus of claim 6 , wherein the function is a gradient of performance and the one or more costs.
8. The apparatus of claim 1 , wherein the processor is further configured to utilize a machine learning based method configured to select a NoC from the plurality of NoCs for generation based on performance versus the one or more costs.
9. The apparatus of claim 8 , wherein the machine learning based method is configured to select a NoC based on a computation of a gradient between the performance and the costs, and another function for NoCs having the same gradient.
10. The apparatus of claim 1 , wherein the processor is configured to present the difference by providing a graphical display comparing performance versus the one or more costs.
11. A method, comprising:
processing a NoC specification for a plurality of performance requirements of a NoC;
generating each of a plurality of NoCs in a form of a Register Transfer Level (RTL), each of the plurality of NoCs meeting a first subset of the plurality of performance requirements;
for each of the plurality of NoCs, presenting a difference between an actual performance of the each of the plurality of NoCs and each performance requirement of a second subset of the plurality of performance requirements and one or more costs for each of the plurality of NoCs; and
for a selection of a NoC of the plurality of NoCs for construction, incorporating the RTL associated with the selected NoC into hardware.
12. The method of claim 11 , wherein the plurality of performance requirements comprises bandwidth requirements and latency requirements for a plurality of sets of traffic of the NoC specification.
13. The method of claim 11 , wherein the plurality of performance requirements comprises performance requirements of selected one or more subsets of traffic of the NoC specification.
14. The method of claim 11 , wherein the plurality of performance requirements comprises multiple SoC/NoC use profiles.
15. The method of claim 11 , wherein the one or more costs comprises at least one of area, power consumption, and wiring congestion.
16. The method of claim 11 , wherein the presenting the difference and the one or more costs comprises providing a visualization for the plurality of NoCs based on a function incorporating performance and cost.
17. The method of claim 11 , wherein the function is a gradient of performance and the one or more costs.
18. The method of claim 11 , further comprising utilizing a machine learning based method configured to select a NoC from the plurality of NoCs for generation based on performance versus the one or more costs.
19. The method of claim 18 , wherein the machine learning based method is configured to select a NoC based on a computation of a gradient between the performance and the costs, and another function for NoCs having the same gradient.
20. The method of claim 11 wherein the presenting the difference comprises providing a graphical display comparing performance versus the one or more costs.