Shanghai Zhaoxin Semiconductor Co., Ltd.

Via Alliance Semiconductor Co., Ltd.

Tsinghua University

Chinese Academy of Medical Sciences

Xiaoyang Li

Yan Wang

Zongpu Qi

Zheng Wang

Lin Li

Jing Chen

Jiin Lai

Weilin Wang

Wei Zhao

Zhenhua Huang

Fangfang Wu

Di Hu

Jin Yu

Juanli Song

Tao Li

Chen Chen

Lei Meng

Xuehua Han

Shican Luo

Dongxue Gao

Kuan-Jui Ho

Mengchen Yang

Yanliang Liu

Zhiqiang Hui

Haiquan Zhang

Zandong Han

Dong Du

Jiandong Jiang

Haipeng Zhou

Zhenxiong Zhao

Shurong Ma

Jiawen Shou

A performance analysis system includes a picker module and a calculation circuit. The picker module is placed in the processing device to capture a plurality of pieces of time information of a unit circuit of each of a plurality of tasks in the processing device during total execution time of processing the plurality of tasks. The calculation circuit performs an interval analysis operation on the time information. The interval analysis operation includes: calculating an overlap period between a current task and a previous task; and counting time occupied by the unit circuit during the total execution time of processing the tasks by the processing device according to a relation between the current time interval of the current task corresponding to the unit circuit and the overlap period.

Performance analysis system for analyzing processing performance of processing device and method thereof

A micro-processor circuit and a method of performing neural network operation are provided. The micro-processor circuit is suitable for performing neural network operation. The micro-processor circuit includes a parameter generation module, a compute module and a truncation logic. The parameter generation module receives in parallel a plurality of input parameters and a plurality of weight parameters of the neural network operation. The parameter generation module generates in parallel a plurality of sub-output parameters according to the input parameters and the weight parameters. The compute module receives in parallel the sub-output parameters. The compute module sums the sub-output parameters to generate a summed parameter. The truncation logic receives the summed parameter. The truncation logic performs a truncation operation based on the summed parameter to generate a plurality of output parameters of the neural network operation.

Micro-processor circuit and method of performing neural network operation

A micro-processor circuit and a method of performing neural network operation are provided. The micro-processor circuit is suitable for performing neural network operation. The micro-processor circuit includes a parameter generation module, a compute module and a compare logic. The parameter generation module receives in parallel a plurality of input parameters and a plurality of weight parameters of the neural network operation. The parameter generation module generates in parallel a plurality of sub-output parameters according to the input parameters and the weight parameters. The compute module receives in parallel the sub-output parameters. The compute module sums the sub-output parameters to generate a summed parameter. The compare logic receives the summed parameter. The compare logic performs a comparison operation based on the summed parameter to generate a plurality of output parameters of the neural network operation.

A processing system includes at least one core, at least one accelerator function unit (AFU), a microcontroller, and a memory access unit. The AFU and the core share a plurality of virtual addresses to access a memory. The microcontroller is coupled between the core and the AFU. The core develops and stores a task in one of the virtual addresses. The microcontroller analyzes the task and dispatches the task to the AFU. The AFU accesses the virtual address indicating where the task is stored through the memory access unit to executes the task.

Processing system for dispatching tasks and memory access method thereof

A processing system includes at least one core, a plurality of accelerator function unit (AFU) and a memory access unit. The memory access unit includes at least one pipeline resource and an arbitrator. The core develops a plurality of tasks. Each of the AFU is used to execute at least one of the tasks which corresponds to several memory access requests. The arbitrator selects one of the AFUs using a round-robin method at each clock period to transmit a corresponding memory access request of the selected AFU to the pipeline resource, so that the selected AFU executes the memory access request through the pipeline resource to read or write data related to the task.

Processing system with round-robin mechanism and its memory access method

A microprocessor for neural network computing having a mapping table, a microcode memory, and a microcode decoding finite-state machine (FSM) is disclosed. According to the mapping table, a macroinstruction is mapped to an address on the microcode memory. The microcode decoding FSM decodes contents which are retrieved from the microcode memory according to the address, to get microinstructions involving at least one microinstruction loop that is repeated to operate a datapath to complete the macroinstruction.

Microprocessor for neural network computing and processing method of macroinstruction

A processing system includes a core, at least one accelerator function unit (AFU) and an accelerator interface. The core is utilized to develop at least one task. The AFU is utilized to execute the task. The accelerator interface is arranged between the core and the AFU to receive an accelerator interface instruction transmitted by the processing core and instruct the AFU to execute the task according to the accelerator interface instruction.

Processing system and heterogeneous processor acceleration method

The invention introduces a method for accelerating compression, performed by configuration logic of a compression accelerator, containing: obtaining an input parameter from a processor core; obtaining a configuration setting from a compression parameter table according to the input parameter; configuring hardware coupled between a first buffer and a second buffer to form a data transmission path according to the input parameter, wherein the first buffer stores raw data; and transmitting the configuration setting to devices on the data transmission path for processing the raw data to generate compressed data and storing the compressed data in the second buffer.

Methods for accelerating compression and apparatuses using the same

An accelerated compression method and apparatus are provided. The accelerated compression apparatus includes a look-ahead buffer, a string matching processing pipeline and a control circuit. The string to be compressed extracted from the data register is stored to the look-ahead buffer. A string to be compressed includes Q characters, and a repeat flag is stored in the look-ahead buffer for each character in the string to be compressed. P instances are issued in parallel in each issue cycle. When all the characters included in P substrings corresponding to the P instances are identical to each other, the control circuit sets the repeat flags of the start characters corresponding to the last (P&#x2212;1) instances among the P instances to a set state. An instance in which the repeat flag of any character of the P instances is not set to the set state is sent to the string matching processing pipeline for a matching operation, and an instance in which the repeat flags of all the characters are set to the set state is prevented from being sent to the string matching processing pipeline.

Accelerated compression method and accelerated compression apparatus

A microprocessor provides at least two storage areas and uses a datapath for Booth multiplication. According to a first and second field of a microinstruction, the datapath gets multiplicand number supply data from the first storage area and multiplier number supply data from the second storage area. The datapath operates according to a word length indicated in a third field of the microinstruction. The datapath gets multi-bit acquisitions for Booth multiplication from the multiplier number supply data. The datapath divides the multiplicand number supply data into multiplicand numbers according to the word length, and performs Booth multiplication on the multiplicand numbers based on the multi-bit acquisitions to get partial products. According to the word length, the datapath selects a part of the partial products to be shifted and added for generation of a plurality of products.

Microprocessor with booth multiplication

A microprocessor with dynamically adjustable bit width is provided, which has a bit width register, a datapath, a statistical register, and a bit width adjuster. The bit width register stores at least one bit width. The datapath operates according to the bit width stored in the bit width register to acquire input operands from received data and process input operands. The statistical register collects calculation results of the datapath. The bit width adjuster adjusts the bit width stored in the bit width register based on the calculation results collected in the statistical register.

Microprocessor with dynamically adjustable bit width for processing data

A microprocessor with Booth multiplication, in which several acquisition registers are used. In a first word length, a first acquisition register stores an unsigned ending acquisition of a first multiplier number carried in multiplier number supply data, and a third acquisition register stores a starting acquisition of a second multiplier number carried in the multiplier number supply data. In a second word length that is longer than the first word length, a fourth acquisition register stores a middle acquisition of a third multiplier number carried in the multiplier number supply data. A partial product selection circuit is required for selection of a partial product, to get the partial product from Booth multiplication based on the third acquisition register (corresponding to the first word length) or based on the fourth acquisition register (corresponding to the second word length).

A microprocessor having the capability of executing a micro-instruction for series calculation is provided. The microprocessor includes an instruction decoder and an execution circuit for series calculation. The micro-instruction whose source operands correspond to an undetermined number x and a plurality of coefficients ato a(for xto x) is decoded by the instruction decoder. Based on x and ato a, the execution circuit for series calculation includes at least one multiplier for calculating exponentiation values of x (e.g. x), and includes at least one MAU (multiply-and-accumulate unit) for combining x, the exponentiation values of x, and the coefficients ato afor the series calculation.

Microprocessor with circuit for series calculation

An accelerated compression method and apparatus are provided. The accelerated compression apparatus includes a look-ahead memory, a string matching processing pipeline and a control circuit. According to an issue pointer, the control circuit issues sub-strings of a string to be compressed from the look-ahead memory to the string matching processing pipeline for a matching operation to obtain a matched length and a matched offset. The control circuit determines a new retiring position according to the matched length corresponding to a retire pointer. When the new retiring position exceeds an issuing position pointed by the issue pointer, the control circuit resets the string matching processing pipeline.

A chipset with a near-data processing (NDP) engine, which uses the NDP engine to perform a command transformation and thereby to generate an input and output (I/O) command to operate a peripheral device connected to the chipset. The chipset further has a traffic control module. The chipset receives a request to operate the peripheral device, and the traffic control module directs the request to the NDP engine to be transformed into the I/O command. The NDP engine may implement a file system, or achieve acceleration of a database or may be operated to cope with a remote direct memory access packet.

Growing community of inventors

Beijing, China

Xiaoyang Li