Abstract: Systems and methods for optimizing data flow in an integrated circuit includes creating a task graph based on transforming an optimized network graph for a neural network application, wherein creating the task graph includes: enumerating a plurality of distinct tasks based on a decomposition of each of a plurality of network operations of the optimized network graph; and allocating a data buffer to each of pairs of dependent tasks of the plurality of distinct tasks based on the decomposition of each of the plurality of network operations of the optimized network graph; encoding a token-informed task scheduler based on a composition of the task graph, wherein the encoding the token-informed task scheduler includes: programming the token-informed task scheduler to cause an execution of the plurality of distinct tasks based on identifying a state of a respective data buffer between each of the pairs of dependent tasks.

Abstract: Systems and methods for optimizing data flow in an integrated circuit includes creating a task graph based on transforming an optimized network graph for a neural network application, wherein creating the task graph includes: enumerating a plurality of distinct tasks based on a decomposition of each of a plurality of network operations of the optimized network graph; and allocating a data buffer to each of pairs of dependent tasks of the plurality of distinct tasks based on the decomposition of each of the plurality of network operations of the optimized network graph; encoding a token-informed task scheduler based on a composition of the task graph, wherein the encoding the token-informed task scheduler includes: programming the token-informed task scheduler to cause an execution of the plurality of distinct tasks based on identifying a state of a respective data buffer between each of the pairs of dependent tasks.

Abstract: A system and method for minimizing a total physical size of data buffers for executing an artificial neural network (ANN) on an integrated circuit includes implementing a buffer-sizing simulation based on sourcing a task graph of the ANN, wherein: (i) the task graph includes a plurality of distinct data buffers, wherein each of the plurality of distinct data buffers is assigned to at least one write operation and at least one read operation; (ii) the buffer-sizing simulation, when executed, computes an estimated physical size for each of a plurality of distinct data buffers for implementing the artificial neural network on a mixed-signal integrated circuit; and (iii) configuring the buffer-sizing simulation includes setting simulation parameters that include buffer-size minimization parameters and buffer data throughput optimization parameters; and generating an estimate of a physical size for each of the plurality of distinct data buffers based on the implementation of the buffer-sizing simulation.

Abstract: A system and method for minimizing a total physical size of data buffers for executing an artificial neural network (ANN) on an integrated circuit includes implementing a buffer-sizing simulation based on sourcing a task graph of the ANN, wherein: (i) the task graph includes a plurality of distinct data buffers, wherein each of the plurality of distinct data buffers is assigned to at least one write operation and at least one read operation; (ii) the buffer-sizing simulation, when executed, computes an estimated physical size for each of a plurality of distinct data buffers for implementing the artificial neural network on a mixed-signal integrated circuit; and (iii) configuring the buffer-sizing simulation includes setting simulation parameters that include buffer-size minimization parameters and buffer data throughput optimization parameters; and generating an estimate of a physical size for each of the plurality of distinct data buffers based on the implementation of the buffer-sizing simulation.

Abstract: Systems and methods for optimizing data flow in an integrated circuit includes creating a task graph based on transforming an optimized network graph for a neural network application, wherein creating the task graph includes: enumerating a plurality of distinct tasks based on a decomposition of each of a plurality of network operations of the optimized network graph; and allocating a data buffer to each of pairs of dependent tasks of the plurality of distinct tasks based on the decomposition of each of the plurality of network operations of the optimized network graph; encoding a token-informed task scheduler based on a composition of the task graph, wherein the encoding the token-informed task scheduler includes: programming the token-informed task scheduler to cause an execution of the plurality of distinct tasks based on identifying a state of a respective data buffer between each of the pairs of dependent tasks.

Abstract: Systems and methods for improving a computational performance of a mixed-signal integrated circuit includes identifying a suboptimal graph component of a computation graph of a subject application, wherein: (i) the computation graph comprises a plurality of graphical nodes representing computational operations and a plurality of graphical edges representing data dependencies between the graphical nodes, and (ii) the suboptimal graph component comprises a subset of the plurality of graphical nodes and the plurality of graphical edges that do not satisfy an optimal operation threshold; at compile time, selectively applying an optimizing transformation to the suboptimal graph component based on attributes of a first activation function within the suboptimal graph component, wherein the optimization transformation, when applied, transforms the suboptimal graph component to an optimal graph component that satisfies the optimal operation threshold; and reconstructing the computation graph using the optimal graph co

Abstract: Systems and methods for optimizing data flow in an integrated circuit includes creating a task graph based on transforming an optimized network graph for a neural network application, wherein creating the task graph includes: enumerating a plurality of distinct tasks based on a decomposition of each of a plurality of network operations of the optimized network graph; and allocating a data buffer to each of pairs of dependent tasks of the plurality of distinct tasks based on the decomposition of each of the plurality of network operations of the optimized network graph; encoding a token-informed task scheduler based on a composition of the task graph, wherein the encoding the token-informed task scheduler includes: programming the token-informed task scheduler to cause an execution of the plurality of distinct tasks based on identifying a state of a respective data buffer between each of the pairs of dependent tasks.

Abstract: Systems and methods for improving a computational performance of a mixed-signal integrated circuit includes identifying a suboptimal graph component of a computation graph of a subject application, wherein: (i) the computation graph comprises a plurality of graphical nodes representing computational operations and a plurality of graphical edges representing data dependencies between the graphical nodes, and (ii) the suboptimal graph component comprises a subset of the plurality of graphical nodes and the plurality of graphical edges that do not satisfy an optimal operation threshold; at compile time, selectively applying an optimizing transformation to the suboptimal graph component based on attributes of a first activation function within the suboptimal graph component, wherein the optimization transformation, when applied, transforms the suboptimal graph component to an optimal graph component that satisfies the optimal operation threshold; and reconstructing the computation graph using the optimal graph co

Abstract: The present invention provides devices for screening the skin of an individual in real time using a region-fusion based segmentation with narrow band graph partitioning algorithm to analyze and classify a region of interest of the skin as benign or malignant. Also, provided is a method for screening the skin of an individual using the devices described herein. In addition the present invention provides a digital processor-implemented system for classifying a region of interest on the skin and a processor readable medium having processor-executable instructions to perform skin cancer detection.

Abstract: The present invention provides devices for screening the skin of an individual in real time using a region-fusion based segmentation with narrow band graph partitioning algorithm to analyze and classify a region of interest of the skin as benign or malignant. Also, provided is a method for screening the skin of an individual using the devices described herein. In addition the present invention provides a digital processor-implemented system for classifying a region of interest on the skin and a processor readable medium having processor-executable instructions to perform skin cancer detection.

Abstract: A method of register allocation in complier using a computer instruction set having tiered instructions that accesses differing numbers of registers makes a first preliminary register allocation attempt using an initially specified register set for each instruction. If this fails, the method identifies instructions having an initially specified limited register having a variable not register allocatable. The method makes a second preliminary register allocation attempt except using a less restrictive register set for the identified instructions. This method employs a next less restrictive register set and re-attempts preliminary register allocations for instructions with more than two levels of register restriction.

Abstract: In order to call a sub-routine, a software program first calls an intermediate or stub sub-routine. The stub sub-routine is provided with the address of the sub-routine and the return address. The stub sub-routine saves the states of the selected system resources and calls the sub-routine. After executing the sub-routine, the process returns to the stub sub-routine where the states of the resources, previously saved, are restored. The stub sub-routine returns the process to the return address. By up-dating the stub sub-routine, existing programs can be made backward compatible without the need for revising the software program. When the stub sub-routine is used with a plurality of software programs, the total code length can be reduced by placing the state-saving function only in the stub sub-routine.

Abstract: This invention is useful in a very long instruction word data processor that fetches a predetermined plural number of instructions each operation cycle. A predetermined one of these instructions is used as a special header. This special header has a unique encoding different from any normal instruction. When decoded this special header instructs decode hardware to decode this fetch packet in a special way. In one embodiment a bit field in the header signals the decode hardware whether to decode each instruction word normally or in an alternative way. The header may include extension opcode bits corresponding to each of the other instruction slots. In another embodiment another bit field signals whether to decode an instruction field as one normal length instruction or as two half-length instructions.

Abstract: This invention employs a 16-bit instruction set that has a subset of the functionality of the 32-bit instruction set. In this invention 16-bit instructions and 32-bit instructions can coexist in the same fetch packet. In the prior architecture 32-bit instructions may not span a 32-bit boundary. The 16-bit instruction set is implemented with a special fetch packet header that signals whether the fetch packet includes some 16-bit instructions. This fetch packet header also has special bits that tell the hardware how to interpret a particular 16-bit instruction. These bits essentially allow overlays on the whole or part of the 16-bit instruction space. This makes the opcode space larger permitting more instructions than with a pure 16-bit opcode space.

Abstract: A method of operating a multiple execution unit microprocessor in a software pipelined loop is disclosed. This method allows the microprocessor to respond to interrupt requests and other runtime conditions during execution of a software pipelined loop utilizing multiple assignment of registers. In one embodiment, the method comprises branching out of the software pipelined loop, upon occurrence of an interrupt request, to an interrupt epilog that consumes in-flight register values and sets the interrupt return pointer to the address of an interrupt prolog. The interrupt is then taken. The interrupt prolog is a subset of the loop prolog, and restores the processor to an operational state equivalent to one that would have existed had the interrupt not been taken. Loop execution is then resumed without data loss or corruption.