Abstract: Disclosed herein are vector index registers in vector processors that each store multiple addresses for accessing multiple positions in vectors. It is known to use scalar index registers in vector processors to access multiple positions of vectors by changing the scalar index registers in vector operations. By using a vector indexing register for indexing positions of one or more operand vectors, the scalar index register can be replaced and at least the continual changing of the scalar index register can be avoided.

Abstract: An arithmetic apparatus used for a neural network includes a plurality of digital-time conversion circuits connected in series and a time-digital conversion circuit connected to a last digital-time conversion circuit in the series. Each of the digital-time conversion circuits is configured to delay a first input time signal by a variable amount, delay a second input time signal by a fixed amount, and output the delayed first and second input time signals respectively as either first and second output time signals or second and first output time signals, in accordance with the input data. The time-digital conversion circuit is configured to generate a digital output signal by comparing first and second output time signals from the last digital-time conversion circuit.

Abstract: This device comprises a fast sampler comprising: a truncated table associating with truncated random numbers rmsb coded on Nmsb bits, the only sample k for which, whatever the number rlsb belonging to the interval [0; 2Nr?Nmsb?1], the following condition is met: F(k?1)<(rmsb, rlsb)?F(k), where: (rmsb, rlsb) is the binary number coded on Nr bits and the Nmsb most significant bits of which are equal to the truncated random number rmsb and the (Nr?Nmsb) least significant bits of which are equal to the number rlsb, Nmsb is an integer number lower than Nr, a module for searching for a received truncated random number rmsb in the truncated table, and able to transmit the sample k, associated, by the truncated table, with the received truncated random number rmsb, by way of random number drawn according to the probability distribution ?.

Abstract: A streaming engine in a system receives a first set of stream parameters into a queue to define a first stream along with an indication of either a queue mode of operation or a speculative mode of operation for the first stream. Acquisition of the first stream then begins. At some point, a second set of stream parameters is received into the queue to define a second stream. When the queue mode of operation was specified for the first stream, the second set of parameters is queued and acquisition of the second stream is delayed until completion of acquisition of the first stream. When the speculative mode of operation was specified for the first stream, acquisition of the first stream is canceled upon receipt of the second set of stream parameters and acquisition of the second stream begins immediately.

Abstract: A random number generator includes a static random number generator, a dynamic entropy source, a counter and a combining circuit. The static random number generator includes an initial random number pool and a static random number pool to output a static random number sequence from one thereof the initial random number pool and the static random number pool. The dynamic entropy source is used to generate a dynamic entropy bit. The counter is used to generate a dynamic random number sequence according to the dynamic entropy bit. The combining circuit is used to output a true random number sequence to a lively random number pool according to the static random number sequence and the dynamic random number sequence. The static random number pool is updated when the lively random number pool is fully updated.

Abstract: Systems, devices, and methods related to a Deep Learning Accelerator and memory are described. An integrated circuit may be configured to perform at least computations on matrix operands and configured with: random access memory configured to store instructions executable by the Deep Learning Accelerator and store matrices of an Artificial Neural Network; a connection between the random access memory and the Deep Learning Accelerator; and an interface to a memory controller. The interface may be configured to facilitate access to the random access memory by the memory controller. In response to an indication provided in the random access memory, the Deep Learning Accelerator may execute the instructions to apply input that is stored in the random access memory to the Artificial Neural Network, generate output from the Artificial Neural Network, and store the output in the random access memory.

Abstract: A system, processor, programming product and/or method including: an instruction dispatch unit configured to dispatch instructions of a compare immediate-conditional branch instruction sequence; and a compare register having at least one entry to hold information in a plurality of fields. Operations include: writing information from a first instruction of the compare immediate-conditional branch instruction sequence into one or more of the plurality of fields in an entry in the compare register; writing an immediate field and the ITAG of a compare immediate instruction into the entry in the compare register; writing, in response to dispatching a conditional branch instruction, an inferred compare result value into the entry in the compare register; comparing a computed compare result value to the inferred compare result value stored in the entry in the compare register; and not execute the compare immediate instruction or the conditional branch instruction.

Abstract: A processing unit having a register file includes: a plurality of registers each having a write enable input configured to receive a write enable signal and a write data input connected to a write data path of the processing unit and configured to write data values from the write data path for storage in a register when the write enable signal is asserted; write circuitry configured in a normal mode of operation to assert the write enable signal of a respective one of the registers to cause operational data values to be written to that register from the write data path; and data cleansing circuitry configured to control a data cleansing mode in which write enable signals of all registers in the register file are simultaneously asserted to cause cleansing data values to be simultaneously written to all registers in the register file from the write data path.

Abstract: A neural network device includes a shift register circuit, a control circuit, and a processing circuit. The shift register circuit includes registers configured to, in each cycle of cycles, transfer stored data to a next register and store new data received from a previous register to a current register. The control circuit is configured to sequentially input data of input activations included in an input feature map into the shift register circuit in a preset order. The processing circuit, includes crossbar array groups that receive input activations from at least one of the registers and perform a multiply-accumulate (MAC) operation with respect to the received input activation and weights, is configured to accumulate and add at least some operation results output from the crossbar array groups in a preset number of cycles to obtain an output activation in an output feature map.

Abstract: A modular multiplication circuit includes a main operation circuit, a look-up table, and an addition unit. The main operation circuit updates a sum value and a carry value according to 2iA corresponding to a first operation value A and m bits of a second operation value B currently under operation, m is a positive integer, i is from 0 to m?1. The look-up table records values related to a modulus, and selects one of the values as a look-up table output value according to the sum value. The addition unit updates the sum value and the carry value according to the look-up table output value and outputs the updated sum value and the updated carry value to the main operation circuit. The modular multiplication circuit updates the sum value and the carry value in a recursive manner by using m different bits of the second operation value B.

Abstract: A semiconductor memory device includes a plurality of memory bank groups configured to be accessed in parallel; an internal memory bus configured to receive external data from outside the plurality of memory bank groups; and a first computation circuit configured to receive internal data from a first memory bank group of the plurality of memory bank groups during each first period of a plurality of first periods, receive the external data through the internal memory bus during each second period of a plurality of second periods, the second period being shorter than the first period, and perform a processing in memory (PIM) arithmetic operation on the internal data and the external data during each second period.

Abstract: Systems and methods are provided to perform multiply-accumulate operations of multiple data types in a systolic array. One or more processing elements in the systolic array can include a shared multiplier and one or more adders. The shared multiplier can include a separate and/or a shared circuitry where the shared circuitry can perform at least a part of integer multiplication and at least a part of non-integer multiplication. The one or more adders can include one or more shared adders or one or more separate adders. The shared adder can include a separate and/or a shared circuitry where the shared circuitry can perform at least a part of integer addition and at least a part of non-integer addition.

Abstract: An instruction simulation device and a method thereof are provided. The simulation device includes a monitor, which is configured to determine whether a ready-for-execution instruction is an instruction under a new/extended instruction set sharing the same instruction set architecture as that of the processor. If the ready-for-execution instruction is an extended instruction, it is converted into a simulation program which consists of a compatible instruction sequence further composed of at least one native instruction of the processor or a compatible instruction recognizable/executable by the processor. An execution result of the extended instruction is simulated by executing the simulation program, thereby extending the service life of an electronic appliance embodied with the disclosed simulation device therein.

Abstract: This application describes hybrid hardware accelerators, systems, and apparatus for performing various computations in neural network applications using the same set of hardware resources. An example accelerator may include weight selectors, activation input interfaces, and a plurality of Multiplier-Accumulation (MAC) circuits organized as a plurality of MAC lanes Each of the plurality of MAC lanes may be configured to: receive a control signal indicating whether to perform convolution or vector operations; receive one or more weights according to the control signal; receive one or more activations according to the control signal; and generate output data based on the one or more weights and the one or more input activations according to the control signal and feed the output data into an output buffer. Each of the plurality of MAC lanes includes a plurality of multiplier circuits and a plurality of adder-subtractor circuits.

Abstract: A compute-in-memory bitcell is provided that includes a pair of cross-coupled inverter for storing a stored bit. The compute-in-memory bitcell includes a logic gate for multiplying the stored bit with an input vector bit. An output node for the logic gate connects to a second plate of a capacitor. A first plate of the capacitor connects to a read bit line.

Abstract: A random number generation device includes: a plurality of first uniform random number generators configured to respectively generate a plurality of first uniform random numbers; a plurality of first normal random number generators configured to respectively generate a plurality of first normal random numbers based on the plurality of first uniform random numbers; a plurality of second uniform random number generators configured to perform a logical operation on bit values of two or more of the first uniform random numbers to respectively generate a plurality of second uniform random numbers; and at least one second normal random number generator configured to generate at least one second normal random number based on the plurality of second uniform random numbers.

Abstract: In one embodiment of the present invention, a programmable vision accelerator enables applications to collapse multi-dimensional loops into one dimensional loops. In general, configurable components included in the programmable vision accelerator work together to facilitate such loop collapsing. The configurable elements include multi-dimensional address generators, vector units, and load/store units. Each multi-dimensional address generator generates a different address pattern. Each address pattern represents an overall addressing sequence associated with an object accessed within the collapsed loop. The vector units and the load store units provide execution functionality typically associated with multi-dimensional loops based on the address pattern. Advantageously, collapsing multi-dimensional loops in a flexible manner dramatically reduces the overhead associated with implementing a wide range of computer vision algorithms.

Abstract: A signal processing apparatus comprises an operation processing part that performs operation processing on data represented in the two's complement representation and a storage processing part that performs storage processing on data represented in a second representation format as a data representation format, and in the second representation format, a data value is identical to one in the two's complement representation when the value is positive or zero, and all the bits lower than the most significant bit that indicates the sign in the two's complement representation are inverted when a data value is negative.

Abstract: An apparatus and method are provided for inhibiting instruction manipulation. The apparatus has execution circuitry for performing data processing operations in response to a sequence of instructions from an instruction set, and decoder circuitry for decoding each instruction in the sequence in order to generate control signals for the execution circuitry. Each instruction comprises a plurality of instruction bits, and the decoder circuitry is arranged to perform a decode operation on each instruction to determine from the value of each instruction bit, and knowledge of the instruction set, the control signals to be issued to the execution circuitry in response to that instruction. An input path to the decoder circuitry comprises a set of wires over which the instruction bits of each instruction are provided.

Abstract: A processor includes a front-end with an instruction set that operates at a first bit width and a floating point unit coupled to receive the instruction set in the processor that operates at the first bit width. The floating point unit operates at a second bit width and, based upon a bit width assessment of the instruction set provided to the floating point unit, the floating point unit employs a shadow-latch configured floating point register file to perform bit width reconfiguration. The shadow-latch configured floating point register file includes a plurality of regular latches and a plurality of shadow latches for storing data that is to be either read from or written to the shadow latches. The bit width reconfiguration enables the floating point unit that operates at the second bit width to operate on the instruction set received at the first bit width.