Techniques related to packing pieces of data having variable bit lengths to serial packed data using a graphics processing unit and a central processing unit are discussed. Such techniques include executing bit shift operations for the pieces of data in parallel via execution units of the graphics processing unit and packing the bit shifted pieces of data via the central processing unit.

In various examples, a VPU and associated components may be optimized to improve VPU performance and throughput. For example, the VPU may include a min/max collector, automatic store predication functionality, a SIMD data path organization that allows for inter-lane sharing, a transposed load/store with stride parameter functionality, a load with permute and zero insertion functionality, hardware, logic, and memory layout functionality to allow for two point and two by two point lookups, and per memory bank load caching capabilities. In addition, decoupled accelerators may be used to offload VPU processing tasks to increase throughput and performance, and a hardware sequencer may be included in a DMA system to reduce programming complexity of the VPU and the DMA system. The DMA and VPU may execute a VPU configuration mode that allows the VPU and DMA to operate without a processing controller for performing dynamic region based data movement operations.

According to one embodiment, a processor includes an instruction decoder to decode a first instruction to gather data elements from memory, the first instruction having a first operand specifying a first storage location and a second operand specifying a first memory address storing a plurality of data elements. The processor further includes an execution unit coupled to the instruction decoder, in response to the first instruction, to read contiguous a first and a second of the data elements from a memory location based on the first memory address indicated by the second operand, and to store the first data element in a first entry of the first storage location and a second data element in a second entry of a second storage location corresponding to the first entry of the first storage location.

A vector processor is disclosed including a variety of variable-length instructions. Computer-implemented methods are disclosed for efficiently carrying out a variety of operations in a time-conscious, memory-efficient, and power-efficient manner. Methods for more efficiently managing a buffer by controlling the threshold based on the length of delay line instructions are disclosed. Methods for disposing multi-type and multi-size operations in hardware are disclosed. Methods for condensing look-up tables are disclosed. Methods for in-line alteration of variables are disclosed.

Methods, systems, and apparatus for a matrix multiply unit implemented as a systolic array of cells are disclosed. Each cell of the matrix multiply includes: a weight matrix register configured to receive a weight input from either a transposed or a non-transposed weight shift register; a transposed weight shift register configured to receive a weight input from a horizontal direction to be stored in the weight matrix register; a non-transposed weight shift register configured to receive a weight input from a vertical direction to be stored in the weight matrix register; and a multiply unit that is coupled to the weight matrix register and configured to multiply the weight input of the weight matrix register with a vector data input in order to obtain a multiplication result.

Methods and systems relating to improved processing architectures with pre-staged instructions are disclosed herein. A disclosed processor includes an instruction memory, at least one functional processing unit, a bus, a set of instruction registers configured to be loaded, using the bus, with a set of pre-staged instructions from the instruction memory, and a logic circuit configured to provide the set of pre-staged instructions from the set of instruction registers to the at least one functional processing unit in response to receiving an instruction from the instruction memory.

The present disclosure relates to a data processing method and device, and related products. The product may include a control unit. The control unit may include an instruction caching unit, an instruction processing unit, and a storage queue unit. The instruction caching unit is configured to store a calculation instruction associated with an artificial neural network computation. The instruction processing unit may be configured to parse the calculation instruction to obtain a plurality of computation instructions. The storage queue unit may be configured to store an instruction queue, where the instruction queue may include a plurality of computation instructions or calculation instructions to be executed in a sequence of the queue. By adopting the above method, the present disclosure may improve a computation efficiency of the related products when performing a neural network model computation.

A Reduced Instruction Set Computer (“RISC”) supporting large-word operations in a computing environment is disclosed. In one implementation, in response to receiving one or more control signals from a central processing unit (“CPU”), a set of operations are executed on a state of a special purpose execution unit (“SPU”) having a plurality of SPU registers, the SPU being associated with the CPU and the state of the SPU having word widths of one or more of the plurality of registers being greater in size than word widths of a plurality of CPU registers of a computing system and a set of state-master bits to synchronize the state of the SPU and a state of the CPU. The results of the set of operations are stored in the plurality of CPU registers or an alternative set of the plurality of SPU registers.

Provided is a processor that is used for limited purposes such as preprocessing of raw data and that has a small circuit scale and high program processing efficiency, wherein an instruction block includes a 2-bit opcode. The processor can move to a branch destination or perform an operation by using an immediate bit accompanying the instruction block, by assigning a branch flag or an immediate instruction determination bit corresponding to the opcode.

A Reduced Instruction Set Computer (“RISC”) supporting large-word operations in a computing environment is disclosed. In one implementation, in response to receiving one or more control signals from a central processing unit (“CPU”), a set of operations are executed on a state of a special purpose execution unit (“SPU”) having a plurality of SPU registers, the SPU being associated with the CPU and the state of the SPU having word widths of one or more of the plurality of registers being greater in size than word widths of a plurality of CPU registers of a computing system and a set of state-master bits to synchronize the state of the SPU and a state of the CPU. The results of the set of operations are stored in the plurality of CPU registers or an alternative set of the plurality of SPU registers.