Compiling network descriptions to multiple platforms
A method of generating executable code for a target platform in a neural network includes receiving a spiking neural network description. The method also includes receiving platform-specific instructions for one or more target platforms. Further, the method includes, generating executable code for the target platform(s) based on the platform-specific instructions and the network description.
1. A computer-implemented method of generating executable code for a target platform in a neural network, comprising:
receiving a spiking neural network description;
receiving platform-specific optimizations for a plurality of target platforms, the platform-specific optimizations comprising cache coherence for multi-processor target platforms, and the platform-specific optimizations based on a partition of the neural network across the plurality of target platforms;
generating executable code for at least one target platform of the plurality of target platforms by processing the platform-specific optimizations for the plurality of target platforms with the network description; and
deploying the executable code to the at least one target platform.
2. The method of claim 1 , further comprising partitioning the neural network.
3. The method of claim 2 , in which the neural network partitioning is user specified.
4. The method of claim 2 , in which the partitioning is function specified.
5. The method of claim 2 , in which the neural network is partitioned based at least in part on feature restrictions.
6. The method of claim 2 , in which the neural network is partitioned based at least in part on a fault tolerance policy.
7. The method of claim 2 , in which the neural network is partitioned based at least in part on simulation statistics.
8. The method of claim 2 , in which the neural network is partitioned based at least in part on power consumption.
9. An apparatus for generating executable code for a target platform in a neural network, comprising:
a memory; and
at least one processor coupled to the memory, the at least one processor being configured:
to receive a spiking neural network description;
to receive platform-specific optimizations for a plurality of target platforms, the platform-specific optimizations comprising cache coherence for multi-processor target platforms, and the platform-specific optimizations based on a partition of the neural network across the plurality of target platforms;
to generate executable code for at least one target platform of the plurality of target platforms by processing the platform-specific optimizations for the plurality of target platforms with the network description; and
to deploy the executable code to the at least one target platform.
10. The apparatus of claim 9 , in which the at least one processor is further configured to partition the neural network.
11. The apparatus of claim 10 , in which the neural network partitioning is user specified.
12. The apparatus of claim 10 , in which the neural network partitioning is function specified.
13. The apparatus of claim 10 , in which the neural network is partitioned based at least in part on feature restrictions.
14. The apparatus of claim 10 , in which the neural network is partitioned based at least in part on a fault tolerance policy.
15. The apparatus of claim 10 , in which the neural network is partitioned based at least in part on simulation statistics.
16. The apparatus of claim 10 , in which the neural network is partitioned based at least in part on power consumption.
17. An apparatus generating executable code for a target platform in a neural network, comprising:
means for receiving a spiking neural network description;
means for receiving platform-specific optimizations for a plurality of target platforms, the platform-specific optimizations comprising cache coherence for multi-processor target platforms, and the platform-specific optimizations based on a partition of the neural network across the plurality of target platforms;
means for generating executable code for at least one target platform of the plurality of target platforms by processing the platform-specific optimizations for the plurality of target platforms with the network description; and
means for deploying the executable code to the at least one target platform.
18. A non-transitory computer-readable medium having encoded thereon program code for generating executable code for a target platform in a neural network, the program code being executed by a processor and comprising:
program code to receive a neural network description;
program code to receive platform-specific instructions for a plurality of target platforms, the platform-specific optimizations comprising cache coherence for multi-processor target platforms, and the platform-specific optimizations based on a partition of the neural network across the plurality of target platforms;
program code to generate executable code for at least one target platform of the plurality of target platforms by processing the platform-specific optimizations for the plurality of target platforms with the network description; and
program code to deploy the executable code to the at least one target platform.