Machine-learning enhanced compiler

1. A method comprising:

generating a netlist for a circuit design;

predicting, by applying a first machine learning model to the netlist, a first amount of time that it will take to finish compiling the circuit design;

based on the first amount of time exceeding a first threshold, predicting, by applying a second machine learning model to the netlist, a first place and route strategy; and

adjusting a logic of the circuit design in accordance with the first place and route strategy.

2. The method of claim 1 , further comprising:

predicting, by applying a third machine learning model to the adjusted logic, a second amount of time that it will take to finish compiling the circuit design;

based on the second amount of time exceeding a second threshold, predicting, by applying a fourth machine learning model to the adjusted logic, a second place and route strategy; and

placing components of the circuit design in accordance with the second place and route strategy.

3. The method of claim 2 , further comprising:

predicting, by applying a fifth machine learning model to the placed components, a third amount of time that it will take to finish compiling the circuit design;

based on the third amount of time exceeding a third threshold, predicting, by applying a sixth machine learning model to the placed components, a third place and route strategy; and

routing the components in accordance with the third place and route strategy.

4. The method of claim 3 , further comprising generating a bitstream in accordance with the routing of the components.

5. The method of claim 1 , wherein:

the first and second machine learning models are part of a base model generated using base data; and

the base model was selected from a group comprising the base model, a tuned model generated using training data generated by compiling a plurality of circuit designs, and a hybrid model generated using the training data and the base data.

6. The method of claim 5 , further comprising selecting the base model in response to a determination that a first distance between the netlist and the circuit design is less than a second distance between the netlist and the training data used to generate the tuned model and a third distance between the netlist and the training data and the base data used to generate the hybrid model.

7. The method of claim 6 , further comprising determining the first distance by:

applying a first weight to a distance between a first feature in the netlist and a first feature in the base data to produce a first weighted distance;

applying a second weight to a distance between a second feature in the netlist and a second feature in the base data to produce a second weighted distance; and

summing the first weighted distance and the second weighted distance.

8. The method of claim 1 , further comprising launching a separate compilation of the circuit design, wherein adjusting the logic of the circuit design in accordance with the first place and route strategy occurs in the separately launched compilation.

9. An apparatus comprising:

a memory; and

a hardware processor communicatively coupled to the memory, the hardware processor configured to:

receive an instruction to compile a circuit design;

in response to the instruction, generate a netlist for the circuit design;

predict, by applying a first machine learning model to the netlist, a first amount of time that it will take to finish compiling the circuit design; and

based on the first amount of time exceeding a first threshold, predict, by applying a second machine learning model to the netlist, a first place and route strategy; and

adjust a logic of the circuit design in accordance with the first place and route strategy.

10. The apparatus of claim 9 , the hardware processor further configured to:

predict, by applying a third machine learning model to the adjusted logic, a second amount of time that it will take to finish compiling the circuit design;

based on the second amount of time exceeding a second threshold, predict, by applying a fourth machine learning model to the adjusted logic, a second place and route strategy; and

place components of the circuit design in accordance with the second place and route strategy.

11. The apparatus of claim 10 , the hardware processor further configured to:

predict, by applying a fifth machine learning model to the placed components, a third amount of time that it will take to finish compiling the circuit design;

based on the third amount of time exceeding a third threshold, predict, by applying a sixth machine learning model to the placed components, a third place and route strategy; and

route the components in accordance with the third place and route strategy.

12. The apparatus of claim 11 , the hardware processor further configured to generate a bitstream in accordance with the routing of the components.

13. The apparatus of claim 9 , wherein:

the first and second machine learning models are part of a base model generated using base data; and

the base model was selected from a group comprising the base model, a tuned model generated using training data generated by compiling a plurality of circuit designs, and a hybrid model generated using the training data and the base data.