Patent Application
20050172103
1. Field of the Invention
The present invention relates to an array-type computer processor including a state control unit and a data-path unit that includes a plurality of processor elements and a plurality of switch elements, which are arranged in a matrix.
2. Description of the Related Art
The current processor units that can perform any of various data processes include the products in practical use that are referred to as the so-called CPU (Central Processing Unit) and MPU (Micro Processor Unit).
The data-processing system using such a processor unit stores in the memory device the various object codes, in which a plurality of operation instructions are described, and the various processed data. The processor unit can data read in order the operation instructions and processed data from memory device to serially perform a plurality of data processes.
A single processor unit can thus perform various data processes. The data processes, however, need to perform serially in order the plurality of data processes, and for each of the serial data processes, the processor unit needs to data read the operation instructions from the memory device, thereby making it hard to perform complicated data processes at high speed.
On the other hand, if only one data process is to be performed, logical circuits may be formed in hardware for performing the data process, without the necessity of the processor unit data reading in order the plurality of operation instructions from the memory device to perform serially in order the plurality of data processes. The complicated data processes can thus be performed at high speed, but performing only one data process, of course.
That is, the data-processing system that can switch between any object codes can perform various data processes, but not at high speed due to the fixed hardware configuration. On the other hand, the logical circuits in hardware can perform the data processes at high speed, but only one data process due to the fixed object code.
To achieve the above-described problems, the applicants invented an array-type computer processor as the processor unit that can change the hardware configuration according to the software. This array-type computer processor includes a number of small-scale processor elements and switch elements arranged in a matrix, and a data-path unit and state control unit provided in parallel.
Each of the plurality of processor elements can perform individually a data process according to each operation instruction that is individually data set. Each of the plurality of switch elements can switch control each of the connections between the plurality of processor elements according to the each operation instruction that is individually data set.
The array-type computer processor can thus switch the operation instructions of the plurality of processor elements and the plurality of switch elements to change the hardware configuration to perform the various data processes.
A number of small-scale processor elements as hardware can perform simple data processes in parallel so as to perform complicated data processes at high speed as a whole.
For each operation cycle, the state control unit sequentially switches according to the object code the context of the operation instructions for the plurality of processor elements and the plurality of switch elements as described above, so that the array-type computer processor can continuously perform the parallel processes according to the object code (see, for example, Japanese Patent No. 3269526, Japanese application patent laid-open publication No. 2000-138579, Japanese application patent laid-open publication No. 2000-224025, Japanese application patent laid-open publication No. 2000-232354, Japanese application patent laid-open publication No. 2000-232162, Japanese application patent laid-open publication No. 2003-076668, Japanese application patent laid-open publication No. 2003-099409, and “Introduction to the Configurable, Highly Parallel Computer”, Lawrence Snyder, Purdue University, “IEEE Computer, vol.15, No.1, January 1982, pp47-56”).
A data-processing system is also in practical use that includes the plurality of data-processing devices connected in parallel for sharing the complicated data processes. Such a data-processing system includes a homogeneity-connected type with the plurality of data-processing devices of the same structure connected, and a heterogeneity-connected type with the plurality of data-processing devices of different structures connected.
The homogeneity-connected type of the data-processing system shares one data process in the plurality of data-processing devices of the same structure, thereby allowing for the data process at high parallelism. The heterogeneity-connected type of the data-processing system shares one data process in the plurality of data-processing devices of different types, thereby allowing each data-processing device to perform its special data process. The applicants proposed the heterogeneity-connected type of the data-processing system as described above which includes the combination of the general MPU and the array-type computer processor (see, for example, Japanese application patent laid-open publication 2003-196248).
The array-type computer processor as described above can actually be used after storing data of a computer program of the object code in a program memory, and connecting the program memory to the array-type computer processor via a system bus and the like. The array-type computer processor then obtains and holds data of the computer program from the external program memory, and operates according to the data-held computer program.
The array-type computer processor can data store the computer program in a limited storage capacity, however, the user of the array-type computer processor can create any kind of the computer programs, so that the data volume of the computer program can exceed the storage capacity of the array-type computer processor.
To prevent this, it may be ensured that the array-type computer processor has an enough storage capacity. This may, however, enlarge the circuit scale of the array-type computer processor to increase the circuit size and manufacturing cost, so that a small volume of the computer program may waste the storage capacity of array-type computer processor.
The present invention was accomplished in light of the above-described problems, and aims to provide an array-type computer processor which can perform the operation according to a computer program even if the data volume of the computer program is over the storage capacity.
The array-type computer processor according to the present invention includes, a data-path unit, a state control unit, and a code-obtaining means. The data-path unit comprises a plurality of processor elements and a plurality of switch elements, which are arranged in a matrix. A plurality of the processor elements perform each of the data processes according to an instruction code in which data are described in a computer program for each of the plurality of operation states that are sequentially transferred. A plurality of the switch elements can switch control, according to the instruction code, each of the connections between the plurality of the processor elements. The state control unit sequentially transfers, for each operation state, the contexts of the instruction code for each operation state of the data-path unit, according to the instruction codes and properly inputted event data.
Note, however, that the code-obtaining means obtains data of, from the external program memory that stores data of the computer program, a predetermined number of cooperative partial instruction codes of the operation states along with the corresponding partial instruction codes of the contexts, so that the state control unit operates with temporarily holding only a predetermined number of the instruction codes of the operation states whose data are obtained, and the data-path unit operates with temporarily holding only a predetermined number of the instruction codes of the contexts whose data are obtained. Every time the state control unit and data-path unit complete their operations with the temporarily-held instruction codes, the code-obtaining means obtains data of the instruction codes of the subsequent operation states and contexts, so that the array-type computer processor can perform the operations according to the computer program, even if the data volume of the computer program exceeds the storage capacity.
The various means referred in the present invention may be any means that is provided to realize its function, such as dedicated hardware for providing a predetermined function, a data-processing device being provided with a predetermined function by a computer program, a predetermined function provided in a data-processing device by a computer program, and a combination thereof.
The various means referred in the present invention may not necessarily be individually independent. A plurality of means may be formed as one member, one means may form a part of any other means, a part of one means may overlap with a part of any other means, and the like.
The data-processing device referred in the present invention may be any hardware that can data read a computer program to perform a corresponding data process, such as hardware including MPU as a main body to which are connected various devices such as ROM, RAM (Random Access Memory), and I/F (Interface) unit.
The event data referred in the present invention may be the data by which the state control unit and data-path unit and the like communicate various information to each other to perform various operations. The event data includes, for example, a predetermined code by which the data-path unit informs to the state control unit of the operation state being transferred to the next stage, a predetermined code by which the state control unit informs to the code-obtaining means of the initial condition or the current operation state or the impossible continuation of the operation transfer, and a predetermined code by which the code-obtaining means informs to the state control unit of the starting the operation.
The array-type computer processor according to the present invention can perform the operation according to the computer program, even if the data volume of the computer program exceeds the storage capacity, so that the array-type computer processor can adapt to various volumes of the computer programs without increasing the circuit scale uselessly.
The above and other objects, features, and advantages of the present invention will become apparent from the following description with reference to the accompanying drawings, which illustrate examples of the present invention.
One embodiment according to the present invention will be described below with reference to drawings. Data-processing system 1000 in this embodiment includes, as shown in
Data-processing system 1000 also includes program memory 302 dedicated for storing a computer program of array-type computer processor 100, and program memory 303 dedicated for storing computer program of MPU 200. These memories connect to external bus 300.
Array-type computer processor 100 data reads its own computer program from program memory 302 and performs a data process according to the computer program. At the same time, data-path unit 106 data processes the inputted processed data and output it. Data-path unit 106 generates event data according to the data process.
MPU 200 includes hardware such as I/F circuit, processor core, and internal register (not shown), and operates according to the computer program stored in program memory 303 to logically form various means as various functions, such as data-input means, data-processing means, data-storage means, and data-output means.
The data-input means corresponds to a function where the processor core recognizes the input data of the I/F circuit according to the computer program. The data-input means can input the processed data and event data. The data-processing means corresponds to a function where the processor core performs data processes. The data-processing means can data process the inputted processed data according to the computer program and event data.
The data-storage means corresponds to a function where the processor core stores the processed data into the internal register. The data-storage means temporarily stores various data such as processed data. The data-output means corresponds to a function where the processor core controls the data output of I/F circuit. The data-output means outputs the processed processed data and event data.
Note, however, that MPU 200 in data-processing system 1000 receives the event data and at least part of the processed data from array-type computer processor 100, and generates a new event data according to at least part of the processed data, and outputs at least part of the processed data and the newly generated event data to array-type computer processor 100.
Array-type computer processor 100 includes I/F circuit 101, processor core 102, memory controller 103 as a generation circuit and a virtual-recognition means, read multiplexer 104 as a data-read circuit, and the like. Processor core 102 includes, as shown in
Data-path unit 106 includes, as shown in
As shown in
I/F unit 101 includes, as shown in
Protocol-control unit 131 has a data-set bus protocol in common with external bus 300. Protocol-control unit 131 communicates various data with external bus 300 according to the bus protocol and communicates various data with memory-access unit 132 via task-switching circuit 150 in a simpler manner.
Memory-access unit 132 transmits, as shown in
Synchronization-control circuit 133 temporarily holds the event data that MPU 200 inputs via external bus 300 to Protocol-control unit 131, and temporarily holds the event data inputted by state control unit 105.
As shown in
Memory controller 103 transmits various data transmitted by Memory-access unit 132 in I/F unit 101 to state control unit 105 and data-path unit 106 in processor core 102. Read multiplexer 104 reads the held data in state control unit 105 and data-path unit 106 to transmit the data to Memory-access unit 132.
More particularly, as shown in
Instruction decoder 138 connects via instruction bus 142 to memory controller 103.
Instruction decoder 138 also connects to transient table memory 139 and instruction memory 140. Transient table memory 139 connects to state memory 141.
As mentioned above, read multiplexer 104 reads the held data in state control unit 105 and data-path unit 106. Each of memories 139 to 141 in state control unit 105, thus, connects to read multiplexer 104 via data bus 143. Processor/switch elements 107, 108 in data-path unit 106 connects to read multiplexer 104 via m/nb data bus 109, 110.
Processor elements 107 are also, as shown in
Further, address buses 144 in Y columns connect to one instruction decoder 138 in state control unit 105. Address bus 144 for each column connects to memory-control circuit 111 in processor element 107 in X rows.
The computer program of array-type computer processor 100, which is data stored in program memory 302, is data described as a context in which are sequentially switched the instruction codes of a plurality of processor elements 107 and a plurality of switch elements 108, which elements are arranged in a matrix in data-path unit 106. The instruction codes of state control unit 105 for switching the context for each operation cycle are data described as sequentially-transferred operation states. The relative relations between the plurality of operation states that are sequentially transferred are data described as transition rules.
In state control unit 105, therefore, instruction decoder 138 decodes the computer program that is data read from program memory 302 by code-obtaining circuit 150, and instruction memory 140 stores data of the instruction code, and transition table memory 139 stores data of transition rules of a plurality of operation states, which is described in detail later.
State control unit 105 sequentially transfers the operation state according to the transition rules in transition table memory 139, and generates instruction pointers for the plurality of processor elements 107 and instruction pointers for the plurality of switch elements 108, according to the instruction code in instruction memory 140.
The transition rules temporarily held in transition table memory 139 disclose the current operation state, which is then temporarily held in state memory 141. Further, instruction memory 140 stores data of the plurality of instruction codes according to the plurality of operation states, so that memory controller 103 transmits the plurality of address data to state control unit 105 accordingly.
Instruction bus 142 transmits to state control unit 105 the instruction codes in which also encoded and data set the address data of processor element 107 at which the instruction codes are data stored. Instruction decoder 138 decodes the address data and transmits it to processor element 107 in one column selected from address buses 144 in Y columns.
At the same time, instruction memory 112 in processor element 107 stores data of an instruction code with memory controller 103 selecting one of instruction buses 142 in X rows and transmitting the instruction code. In this way, the instruction code and address data are transmitted to one processor element 107, and then the instruction code is data stored in one address space of instruction memory 112 according to the address data.
As shown
Instruction memory 112 temporarily holds the instruction codes of processor element 107 and switch element 108, which codes are data read from program memory 302, so that instruction codes of processor element 107 and switch element 108 are assigned with the instruction pointers supplied by state control unit 105. Instruction decoder 113 decodes the instruction codes assigned with the instruction pointers, and controls the operations of switch element 108, internal variable wiring, m/nb ALU 117 and 118, and the like.
Mb bus 109 transmits mb of “8 (bit)” processed data, and nb bus 110 transmits nb of “1 (bit)” processed data, so that switch element 108 controls the connections between a plurality of processor elements 107 through m/nb buses 109, 110 according to the operation control by instruction decoder 113.
More particularly, switch element 108 has bus connector 121 to which mb buses 109 and nb buses 110 communicate in four directions. Switch element 108 controls the connections between such a plurality of communicated mb buses 109 and the connections between the plurality of communicated nb buses 110.
In array-type computer processor 100, state control unit 105 thus sequentially switches the context of data-path unit 106 for each operation cycle according to a computer program that is data set in program memory 302. For each stage, a plurality of processor elements 107 operate in parallel in data processes that can be set individually.
Input-control circuit 122, as shown in
Output-control circuit 123 controls the connections of the data inputs from mb register file 115 and mb ALU 117 to mb bus 109, and the connections of the data inputs form nb register file 116 and nb ALU118 to nb bus 110.
The internal variable wiring of processor element 107 controls the connections between mb register file 115 and mb ALU 117 and the connections between nb register file 116 and nb ALU 118 inside processor element 107, according to the operation control by instruction decoder 113.
Mb register file 115, according to the connection relation controlled by the internal variable wiring, temporarily holds the processed data of mb that is input from mb bus 109 and the like, and outputs the data to mb ALU 117 and the like. Nb register file 116, according to the connection relation controlled by the internal variable wiring, temporarily holds the processed data of nb that is input from nb bus 110 and the like, and outputs the data to nb ALU 118 and the like.
Mb ALU 117 performs with mb processed data the data process according to the operation control by instruction decoder 113. Nb ALU 118 performs with nb processed data the data process according to the operation control by instruction decoder 113. The m/nb data process can thus be appropriately performed according to the bit number of the processed data.
The process results in data-path unit 106 are fed back, if desired, to state control unit 105 as event data. State control unit 105, according to the inputted event data, transfers the operation state to the next-stage operation state, and switches the context of data-path unit 106 to the next-stage context.
In principle, array-type computer processor 100 in this embodiment can data read the instruction codes from program memory 302 and temporarily hold the instruction codes in state control unit 105 and data-path unit 106, to allow state control unit 105 and data-path unit 106 to operate according to the instruction code, as described above.
State control unit 105 and data-path unit 106 can, however, temporarily hold only a predetermined number of instruction codes at a time, so that state control unit 105 and data-path unit 106 may not temporarily hold all of a set of the instruction codes that are data stored in program memory 302 as a computer program.
In this case, code-obtaining circuit 150 in I/F unit 101 obtains data of a predetermined number of cooperative partial instruction codes of the operation states along with the corresponding partial instruction codes of the contexts from program memory 302, so that state control unit 105 operates with temporarily holding only a predetermined number of the data-obtained instruction codes of the operation states, and data-path unit 106 operates with temporarily holding only a predetermined number of the data-obtained instruction codes of the contexts.
Every time state control unit 105 and data-path unit 106 complete their operations with the temporarily-held instruction codes, code-obtaining circuit 150 obtains data of the instruction codes of the subsequent operation states and contexts, so that array-type computer processor 100 can perform a set of operations even if the processor can only hold portions of the computer program.
More specifically, code-obtaining circuit 150 may include ASIC (Application Specific Integrated Circuit), for example. Code-obtaining circuit 150 may also include, as shown in
If state control unit 105 detects the initial state temporarily holding no instruction codes, or the operation completion at a predetermined number of instruction codes temporarily held, state control unit 105 then outputs the event data for impossible continuation to event-input circuit 151 in code-obtaining circuit 150. Event-input circuit 151 in code-obtaining circuit 150 corresponds to, for example, I/O (Input/Output) port and the like, and receives the event data for impossible continuation that is inputted from state control unit 105.
State-obtaining circuit 152, when event-input circuit 151 receives the event data for impossible continuation, then obtains data of the current operation state from state memory 141 in state control unit 105, and obtains from data-path unit 106 the event data assigning the next-stage operation state. Operation-obtaining circuit 153, according to the operation states and event data that are data obtained by state-obtaining circuit 152, obtains data of a predetermined number of the instruction codes of subsequent operation states by a predetermined algorithm from program memory 302.
That is, array-type computer processor 100 in this embodiment operates with state control unit 105 that transfers the operation states to the next-stage operation state, and transfers the contexts of data-path unit 106 to the next-stage contexts, as described above. In this case, data-path unit 106 performs the data process with the data-set contexts, and then outputs to state control unit 105 the event data assigning the next-stage operation state to transfer to the next-stage contexts.
According to the input event data, state control unit 105 then transfers its own operation states to the next-stage condition, and transfers the contexts of data-path unit 106 to the next-stage contexts. Thus, when array-type computer processor 100 stops its operation and the current operation state of state control unit 105 and the event data for data-path unit 106 are obtained, the next-stage operation state is disclosed.
For example, as shown in
At the initial state shortly after the power supply of array-type computer processor 100 is turned on, state memory 141 of state control unit 105 holds data of no current operation states, of course, so that state-obtaining circuit 152 can data detect the operation states not being data held.
The operation-obtaining circuit then automatically obtains data of a set of four instruction codes from “0” from program memory 302. In this case, as shown in
Context-detection circuit 154 detects data of a predetermined number of contexts corresponding to the operation states of the instruction codes that are data obtained by operation-obtaining circuit 153. Context-obtaining circuit 156 obtains data of the instruction codes of the contexts whose data are detected by context-detection circuit 154, from program memory 302.
Code-setting circuit 157 sets data of in state control unit 105 and data-path unit 106 the instruction codes of the operation states and contexts whose data are obtained by context-obtaining circuit 153 and operation-obtaining circuit 156. Event-output circuit 158 outputs, after code-setting circuit 157 completes the setting data of of the instruction codes, to state control unit 105 the event data for starting the operation.
Relation-generation circuit 159 data generates the correspondence relations between the contexts whose data are detected by context-detection circuit 154 and the operation states. Relation-setting circuit 161 also sets data of the correspondence relations whose data are generated by relation-generation circuit 159 in state control unit 105. That is, instruction memory 112 in processor element 107 temporarily holds the instruction code for each of the plurality of contexts, and the instruction code and memory address can have any relationship between them.
Because state control unit 105 assigns the instruction code of the contexts with the memory address in instruction memory 112, relation-generation circuit 159 data generates the correspondence relations between the contexts and operation states, which is necessary for the assignment, and relation-setting circuit 161 sets data of the relations in state control unit 105.
State control unit 105 starts, upon input of the event data for starting the operation, to sequentially transfer the operation states according to the instruction code and transition rules whose data are stored, and sequentially transfers the contexts of data-path unit 106 for each of the operation states that are sequentially transferred according to the data-set correspondence relations.
Duplication-detection circuit 162 detects the duplication between the plurality of the last-time contexts and the plurality of this-time contexts whose data are detected by context-detection circuit 154. Update-control circuit 163 eliminates the data obtainment from program memory 302 by context-obtaining circuit 156, and the setting data of the instruction codes of the duplicate contexts detected by duplication-detection circuit 162 in data-path unit 106 by code-setting circuit 157.
For example, the last-time four contexts of “0→1→(2, 3)”, as shown in
For ease of illustration,
In the above-described configurations of data-processing system 1000 in this embodiment, MPU 200 acts as a main processor and array-type computer processor 100 acts as a coprocessor, so that array-type computer processor 100 data and MPU 200 can data process cooperatively.
Array-type computer processor 100 and MPU 200 can data read their own computer programs from program memories 302 and 303, and perform corresponding process operations to perform data processes using the processed data inputted via data line 301.
After the data processes are performed, array-type computer processor 100 and MPU 200 output the processed data to data line 301.
The computer program of array-type computer processor 100 is data described as a context in which the instruction codes of a plurality of processor elements 107 and a plurality of switch elements 108 are sequentially switched. The computer program of array-type computer processor 100 is also data described as an operation state in which the instruction code of state control unit 105 sequentially transfers to switch the context for each operation cycle.
In array-type computer processor 100 that operates according to such a computer program, state control unit 105 sequentially transfers the operation state, and sequentially transfers the context of data-path unit 106 for each operation cycle. For each operation cycle, thus, a plurality of processor elements 107 operate in parallel in individually settable data processes and a plurality of switch elements 108 switch control the connections between the plurality of processor elements 107.
In this case, the process results in data-path unit 106 are fed back, if required, to state control unit 105 as the event data. State control unit 105 thus transfers, according to the inputted event data, the operation state to the next-stage operation state, and switches the context of data-path unit 106 to the next-stage context.
As described above, array-type computer processor 100 in this embodiment can data read the instruction code from program memory 302, and temporarily holds the instruction code in state control unit 105 and data-path unit 106, to allow state control unit 105 and data-path unit 106 to operate according to the instruction code.
In array-type computer processor 100 in this embodiment, however, state control unit 105 and data-path unit 106 can temporarily hold only a predetermined number of instruction codes at a time, so that state control unit 105 and data-path unit 106 may not temporarily hold all of a set of the instruction codes whose data are stored in program memory 302 as a computer program.
In such a case, code-obtaining circuit 150 in I/F unit 101 obtains data of a predetermined number of cooperative partial instruction codes of the operation states along with the corresponding partial instruction codes of the contexts from program memory 302, so that state control unit 105 operates with temporarily holding only a predetermined number of the instruction codes of the operation states whose data are obtained, and data-path unit 106 operates with temporarily holding only a predetermined number of the instruction codes of the contexts whose data are obtained.
Every time state control unit 105 and data-path unit 106 complete their operations with the temporarily-held instruction codes, code-obtaining circuit 150 obtains data of the instruction codes of subsequent operation states and contexts, so that array-type computer processor 100 can perform a set of operations even if it can only hold portions of the computer program.
More particularly, at the startup of array-type computer processor 100 in this embodiment, various memories 139 to 141 in state control unit 105 hold no data, of course, so that state control unit 105 detects the initial state where the instruction codes are not temporarily held and outputs the event data for impossible continuation to code-obtaining circuit 150 in I/F circuit 101.
Then, in code-obtaining circuit 150, as shown in
According to the detected initial state and the inputted event data, operation-obtaining circuit 153 then obtains data of a predetermined number of the instruction codes of subsequent operation states from program memory 302. In the case of the initial state, a predetermined number of instruction codes starting at the first operation state whose data are obtained.
Context-detection circuit 154 then detects data of a predetermined number of contexts according to the operation states of the data-obtained instruction codes. Context-obtaining circuit 156 obtains data of the instruction codes of the data-detected contexts from program memory 302.
For example, as shown in
After the instruction codes of the operation states and contexts are data obtained as described above, code-setting circuit 157 sets data of the instruction codes in state control unit 105 and data-path unit 106. The relation-generation circuit data generates the correspondence relations between the data-detected contexts and the operation states. Relation-setting circuit 161 then also sets data of the data-generated correspondence relations in state control unit 105.
Upon completion of setting data of the above-described instruction codes and correspondence relations, event-output circuit 158 outputs to state control unit 105 the event data for starting the operation. State control unit 105 then starts to sequentially transfer the operation states according to the instruction codes and transition rules whose data are stored, and sequentially transfer the contexts of data-path unit 106 for each of the operation states that are sequentially transferred according to the data-set correspondence relations. When sequentially transferring the operation states as described above, state control unit 105 holds data of the current operation state in state memory 141.
If state control unit 105 and data-path unit 106 only data hold a predetermined number of a great number of instruction codes of the computer program as described above, when array-type computer processor 100 operates as described above, a condition may occur where the operations according to the instruction codes cannot be continued.
State control unit 105 again outputs the event data for impossible continuation to code-obtaining circuit 150. After receiving the input of the event data for impossible continuation, code-obtaining circuit 150 obtains data of the current operation state from state memory 141 in state control unit 105, and obtains from data-path unit 106 the event data assigning the next-stage operation state.
The event data of data-path unit 106 assigns the next-stage operation state of state control unit 105 as described above. Code-obtaining circuit 150 thus specifies the next-stage operation state from the event data and the current operation state of state control unit 105, and obtains data of a predetermined number of instruction codes starting at the next-stage operation state from program memory 302.
If, for example, among the eight instruction codes of “0 to 7” of a computer program, as shown in
Note, however, that code-obtaining circuit 150 detects, when obtaining data of the subsequent instruction codes as described above, the duplication between the plurality of the last-time contexts and the plurality of this-time contexts, and eliminates the data obtainment from program memory 302 and the setting data of in data-path unit 106 of the instruction codes of the duplication-detected contexts.
For example, for the last-time four contexts of “0→1→(2, 3)” as shown in
Code-obtaining circuit 150 obtains data of the instruction codes of the operation states and contexts as described above, and then sets data of the instruction codes in state control unit 105 and data-path unit 106. Note, however, that the setting data of the duplication-detected instruction codes is omitted, as described above, and the new instruction codes are overwritten at the positions of the last-time unnecessary instruction codes.
If, for example, the last-time four contexts of “0→1→(2, 3)” are updated to the this-time four contexts of “4→7→1→3”, as shown in
Code-obtaining circuit 150 also data generates the correspondence relations between the data-detected contexts and the operation states, and sets data of the correspondence relations in state control unit 105. After completing the setting data of the instruction codes and correspondence relations as described above, code-obtaining circuit 150 outputs to state control unit 105 the event data for starting the operation.
State control unit 105 thus starts to sequentially transfer the operation states according to the instruction codes and transition rules that are data stored, and sequentially transfer the contexts of data-path unit 106 for each of the operation states that are sequentially transferred according to the data-set correspondence relations. In array-type computer processor 100 in this embodiment, every time state control unit 105 and data-path unit 106 stop their operations with the data-set instruction codes, code-obtaining circuit 150 continues its operations as described above.
In array-type computer processor 100 in this embodiment, every time state control unit 105 and data-path unit 106 stop their operations with the data-set instruction codes, code-obtaining circuit 150 continues its operations as described above, so that even if state control unit 105 and data-path unit 106 can only data hold portions of a number of instruction codes of the computer program, a set of process operations corresponding to the computer program can be performed.
Furthermore, when code-obtaining circuit 150 only updates a predetermined number of the instruction codes of operation states and contexts, duplication is detected between the plurality of the last-time contexts and the plurality of this-time contexts, and the data obtainment from program memory 302 and the setting data of in data-path unit 106 of the instruction codes of the contexts whose duplication is detected are eliminated. The instruction codes can thus update data with less process burden and required time, thereby providing lower power consumption and higher process speed of array-type computer processor 100.
Further, in array-type computer processor 100 in this embodiment, code-obtaining circuit 150 as hardware is built in that realizes the process operation for obtaining data of portions of a set of a number of instruction codes at a time from program memory 302 and setting data of them in state control unit 105 and data-path unit 106 as described above, so that array-type computer processor 100 can alone perform the above-described operations at high speed.
The present invention is not limited to the embodiments described above, and various changes can be made without departing from the spirit and scope thereof. For example, the above-described embodiments illustrate data-processing system 1000 including array-type computer processor 100, MPU 200, and program memories 302 and 303, which are connected via external bus 300. Also possible, however, is a data-processing system (not shown) including only array-type computer processor 100 and program memory 302, and the like.
The above-described embodiments also illustrate code-obtaining circuit 150 that intervenes between protocol-control unit 131 and memory-access unit 132. This code-obtaining circuit 150 can, however, reside at various positions where its functions can be realized. The above-described embodiments also illustrate code-obtaining circuit 150 that has hardware of various means provided as various circuits 151 to 163. For example, however, the combination of the microprocessor and software can provide some or all of the circuits in code-obtaining circuit 150.
MPU 200 with software can also provide some or all of various circuits 151 to 163 in code-obtaining circuit 150. MPU 200 provides the function of code-obtaining circuit 150 at lower speed, but can provide the function of code-obtaining circuit 150 according to the computer program of MPU 200 that is data stored in program memory 303, so that MPU 200 can easily provide the function without any change of the structure of array-type computer processor 100.
For example, MPU 200 can provide all functions of code-obtaining circuit 150 according to the computer program that is data stored in program memory 303 as the computer program for allowing MPU 200 to perform such as receiving the event data for impossible continuation that is inputted by state control unit 105 in array-type computer processor 100; upon input of the event data for impossible continuation, obtaining data of the current operation state from state control unit 105 and obtaining from data-path unit 106 the event data for assigning the next-stage operation state; according to the operation states and event data that are data obtained, obtaining data of from program memory 302 by a predetermined algorithm a predetermined number of the instruction codes of subsequent operation states; detecting data of a predetermined number of contexts according to the operation states of the data-obtained instruction codes; obtaining data of from program memory 302 the instruction codes of the data-detected contexts; setting data of in state control unit 105 and data-path unit 106 the instruction codes of the operation states and contexts that are data obtained; detecting the duplication between the plurality of the last-time contexts and the plurality of this-time contexts that are data detected; eliminating the data obtainment and setting data of of the instruction codes of the duplicate-detected contexts; data generating the correspondence relations between the data-detected contexts and the operation states; also setting data of in state control unit 105 the data-generated correspondence relations; and upon completion of setting data of the instruction codes and correspondence relations, outputting to state control unit 105 the event data for starting the operation.
A dedicated circuit such as ASIC (not shown) that connects to array-type computer processor 100 via external bus 300 and the like can also provide some or all functions of various circuits 151 to 163 in code-obtaining circuit 150. Such a dedicated circuit can also be integrated with program memory 302 in array-type computer processor 100 (not shown).
As illustrated in the above-described embodiments, code-obtaining circuit 150 data updates the instruction codes of state control unit 105 and data-path unit 106, by omitting the data obtainment and setting data of the instruction codes of the duplicate contexts between the last time and this time. All instruction codes can also be data updated each time, however, without detecting duplication or omitting process operations.
This case does not provide less process burden or required time for the data update of the instruction code. This case, however, eliminates the detection of the duplicate instruction codes and the control of omitting operations, thereby make it possible to eliminate the formation of the duplication-detection circuit 162 and update-control circuit 163 to reduce the process burden and required time for the formation.
That is, if a number of duplications are expected to occur in the instruction codes a predetermined number of which are data updated at a time, the duplication-detection circuit 162 and update-control circuit 162 are preferably provided. If few duplications are expected to occur in the instruction codes a predetermined number of which are data updated at a time, no duplication-detection circuits 162 or update-control circuits 162 are preferably provided.
As illustrated in the above-described embodiments, the context can data update with less process burden and required time, by providing duplication-detection circuit 162 and update-control circuit 163 in code-obtaining circuit 150 to detect only the duplicate contexts between this time and the last time to omit the data obtainment and setting data of the instruction codes.
Code-obtaining circuit 150 can also include, however, an update-storage means for data registering the various update techniques of the contexts and their update cost, and a technique-detection means for detecting data of the update technique that minimizes the total of the update cost from the combination of the plurality of the last-time contexts and the plurality of this-time contexts (both means are not shown), to allow the contexts of data-path unit 106 to be data updated with an update technique that is data detected by the code-setting circuit 157 by the operation control of the update-control circuit 163.
The above-described update-storage means can be provided, for example, by a memory circuit that stores data of the various update techniques as a computer program. The technique-detection means can be provided, for example, by a micro processor that data reads a predetermined computer program from the memory circuit according to the mounted computer program and performs the process operations (both are not shown).
The first update technique can be a technique in which, for example, the context-obtaining circuit 156 obtains data of only different instruction codes between the last-time and this-time contexts, and the code-setting circuit 157 sets data of only the difference between this-time and the last-time instruction codes. If there are a plurality of different instruction codes between the last-time and this-time contexts, the above update technique can also select a combination that minimizes the total of the update cost of the difference.
The second update technique can be a technique in which, in the predetermined combination of the last-time and this-time contexts, the last-time context in data-path unit 106 is initialized, and then the code-setting circuit 157 sets data of this-time context.
The third update technique can be a technique in which in the predetermined combination of the plurality of the last-time and this-time contexts, the corresponding last-time contexts are overwritten with other last-time contexts, and then the context-obtaining circuit 156 obtains data of only the instruction codes different from those of this-time context, and the code-setting circuit 157 sets data of the different instruction codes.
More specifically, for the last-time four contexts of “0→1→(2, 3)” as shown in
The context includes, however, the instruction codes of processor elements 107 of a matrix with X rows and Y columns, so that the context of “0” and the context of “4” may have, for example, common instruction codes in most processor elements 107. In this case, the first update technique can be used to data obtain and data store only the difference instruction codes to update the context of “0” to the context of “4”, thereby reducing its process burden and increasing the operation speed.
If, as described above, the context of “0, 2” is data updated to the context of “4, 7” only by the difference instruction code, the total update cost can be compared between the “0→4, 2→7” and “0→7, 2→4” to select a lower update cost to further reduce the process burden and increase the operation speed.
For the context of “0” being data updated to the context of “4” as described above, the instruction codes of the context of “0” are data set in all the processor elements 107 of a matrix with X rows and Y columns, but the instruction codes of the context of “4” may only be data set in portions of the processor elements 107 of a matrix with X rows and Y columns.
In this case, however, the processor elements 107 that do not data set the instruction codes of this-time context of “4” will hold the last-time instruction codes, which need to be initialized. The instruction codes are initialized, as in the case where the instruction codes are data stored, by serially selecting and performing at a time one of processor elements 107 of a matrix with X rows and Y columns, so that the update cost is the same as in the case where the instruction codes are data stored.
The instruction codes of all processor elements 107 of a matrix with X rows and Y columns can thus be formed to be initialized at a time to use the second update technique to initialize the context of “0” and then newly data store the context of “4”, thereby reducing the update cost.
For the context of “0” in “0→1→(2, 3)” being data updated to the context of “4” in “4→7→1→3”, most instruction codes are different between “0” and “4”, and most instruction codes are common between “1” that is not data updated and “4” that is data updated, for example.
In such a case, the context of “1” in “0→1→(2, 3)” is overwritten on the contexts of “0” at a time, and then the context of “1” is data updated to the context of “4” only by the difference instruction codes, thereby making it possible to reduce its process burden and increase the operation speed.
Existing array-type computer processor 100 does not have a function of overwriting the held contexts on other contexts at a time as described above. The instruction codes are initialized, as in the case where the instruction codes are data stored, by serially selecting and performing at a time one of processor elements 107 of a matrix with X rows and Y columns, so that the initialization of all processor elements 107 in data updating the contexts as described above may increase the update cost.
The above-described update technique thus needs to reduce the update cost using a function of initializing the instruction codes of processor elements 107 of a matrix with X rows and Y columns at a time, and a function of overwriting the held contexts on other contexts at a time. A specific description is given below of the hardware structure for easily and quickly initializing and overwriting the contexts.
As shown in
State control unit 105 is arranged to also generate a set signal “set” for disabling the address data that selects at a time one of processor elements 107 of a matrix with X rows and Y columns and for selecting at a time all processor elements 107 of a matrix with X rows and Y column, and a select signal “sel” for selecting one of the Z contexts that are temporarily held in data-path unit 106 as the instruction codes.
As shown in
Setting-all circuit 170 comprises logical circuits including such as AND gates 171 and 172, OR gates 173, and selector circuits 174 to 176. Setting-all circuit 170 can data store the instruction code “data” for each externally-inputted context into one layer of instruction memories 112 as mentioned above, as well as initialize for each context the instruction codes that are temporarily held in instruction memory 112, and overwrite at a time, on the instruction codes of any context, the instruction codes of any context that are temporarily held in instruction memory 112.
Specifically, AND gate 171 receives the address data “x, y” that selects at a time one of processor elements 107 of a matrix with X rows and Y columns. Z AND gates 172 receive the output signal from AND gate 171 and the address data “z” of each layer of instruction memories 112.
The output signals from AND gates 172 are inputted to the each layer of instruction memory 112. The address data “x, y” thus selects one of processor elements 107 of a matrix with X rows and Y columns, and the address data “z” selects one layer of Z layers of instruction memories 112 in processor element 107.
Note, however, that the output signal from AND gate 171, together with the above-described set signal “set”, is input to OR gate 173, so that the set signal “set” generated will disable the address data “x, y” and select one layer of instruction memories 112 corresponding to the address data “z” in all processor elements 107.
Z selector circuits 174 each have one input terminal to which each layer of instruction memory 112 connects individually, and the other input terminal to which initial value “def” is inputted through ground and the like. Selector circuit 174 receives externally the above-described address data “z” as a control signal to usually data output the instruction codes of one layer of instruction memories 112 connected to selector circuit 174. If, however, selector circuit 174 receives externally the address data “z” of one layer of instruction memories 112 connected to selector circuit 174, it data outputs the initial value “def”.
The output signals from these Z selector circuits 174 are inputted to selector circuit 175. The select signal “sel” is also externally inputted to selector circuit 175. Selector circuit 175 selects one layer from Z layers of instruction memories 112 according to the externally-inputted select signal “sel”.
The output signal from selector circuit 175, together with the instruction code “data”, is inputted to selector circuit 176. The above-described set signal “set” is also externally inputted to selector circuit 176. Selector circuit 176 selects, according to the set signal “set”, the externally-inputted instruction code “data” or the instruction code that is data outputted from selector circuit 175, so that the selected instruction codes are data stored in one layer of instruction memories 112 that is selected as described above.
For example, one of the plurality of contexts can be initialized at a time in all the processor elements 107 as follows. With all processor elements 107 selected according to the set signal “set”, one layer of instruction memories 112 is selected according to the address data “z”.
Selector circuit 174 then data outputs the initial value “def” according to the address data “z”. If the above-described one layer of instruction memories 112 is selected according to the select signal “sel”, selector circuit 175 then selects the initial value “def”.
Selector circuit 176 also selects this initial value “def” according to the set signal “set”, and stores data of the initial value “def” in the selected one layer of instruction memories 112 as described above. The instruction codes of a particular context can thus be initialized at a time in all the processor elements 107.
The last-time predetermined context can be overwritten on the last-time other contexts at a time as follows. With all processor elements 107 selected according to the set signal “set”, one layer of instruction memories 112 to be overwritten with a context is selected according to the address data “z”.
One layer of instruction memories 112 to be overwritten on a context is then selected according to the select signal “sel”, so that selector circuit 176 also selects the instruction codes of the selected context according to the set signal “set”. The selected instruction codes are data stored in the one layer of instruction memories 112 that is selected according to the address data “z”. The last-time predetermined context is thus overwritten on the last-time other contexts at a time in all processor elements 107.
In the above-described setting-all circuit 170, if the set signal “set” selects selector circuit 175, the externally-inputted instruction code “data” is disabled. The instruction code “data” can thus be used as a control signal for selector circuit 175 to eliminate the generation of the dedicated select signal “sel”.
The above-described embodiments illustrate that program memory 302 stores data of only one computer program of array-type computer processor 100, and array-type computer processor 100 obtains data of only one computer program from program memory 302 to perform only one process operation.
It is also possible, however, that program memory 302 stores data of a plurality of computer programs of array-type computer processor 100, and array-type computer processor 100 obtains data of the plurality of computer programs from program memory 302 in a time-sharing manner to perform a plurality of process operations in parallel in a time-sharing manner.
In this case, instruction memories 140 and 112 in state control unit 105 and data-path unit 106 temporarily hold the instruction codes of corresponding operation states and contexts of the plurality of computer programs. And when state control unit 105 and data-path unit 106 are operating with one of the plurality of computer programs that is temporarily held, code-obtaining circuit 150 obtains data of a predetermined number of the cooperative partial instruction codes of the operation states of other computer programs, along with the corresponding partial instruction codes of the contexts.
For example, if a computer program “A” includes contexts and operation states of “a1 to a4”, a computer program “B” includes contexts and operation states of “b1 to b4”, and instruction memories 140 and 112 in state control unit 105 and data-path unit 106 each temporarily hold at a time four operation states and four contexts, instruction memories 140 and 112 first temporarily hold “a1 and a2” for the process operations of state control unit 105 and data-path unit 106.
In running of “a1 and a2”, the free space of instruction memories 140 and 112 temporarily hold “b1 and b2”. After running of “a1 and a2” is complete, running of “b1 and b2” starts. In running of “b1 and b2”, “a1 and a2” in instruction memories 140 and 112 are data updated to “a3 and a4”. After running of “b1 and b2” is complete, running of “a3 and a4” starts.
In running of “a3 and a4”, “b1 and b2” in instruction memory 140 and 112 are updated to “b3 and b4”. After running of “a3 and a4” is complete, running of “b3 and b4” starts. In this way, array-type computer processor 100 can perform in a time-sharing manner the process operations in a plurality of computer programs, thereby performing the process operations of the plurality of computer programs at overall high efficiency.
More specifically, to achieve the parallel operations with a plurality of computer programs as described above, code-obtaining circuit 150 is provided with an operation stopping circuit as an operation-stopping means, a stop-obtaining circuit as a stop-obtaining means, a stop holding circuit as a stop holding means, and a switch-setting circuit as a switch-setting means (not shown).
The operation-stopping circuit stops the operations of state control unit 105 and data-path unit 106 upon input of the event data for program switching in event-input circuit 151. The stop-obtaining circuit obtains the operation state of stopped state control unit 105 and the processed data of stopped data-path unit 106.
The stop-holding circuit temporarily holds for each of the plurality of computer programs the data-obtained operation state and processed data. The switch-setting circuit data reads, upon completion of the temporary hold, from the stop-holding circuit the operation state and processed data of other computer programs, and then sets data of the operation state and processed data in state control unit 105 and data-path unit 106.
Event-output circuit 158 outputs, upon completion of the setting data of, to state control unit 105 event data for starting the operation. State control unit 105 starts, upon input of the event data for starting the operation, to sequentially transfer the operation states. The event data for program switching as described above can also serve as, for example, the event data for impossible continuation as described above.
The various circuits as described above can be provided in code-obtaining circuit 150 to allow array-type computer processor 100 to perform in parallel the process operations with a plurality of computer programs in a time-sharing manner. The combination of the micro processor and computer program can provide, of course, some or all of the above-described various circuits, and so can external MPU 200 and dedicated circuits.
Additionally, array-type computer processor 100 can also perform in parallel the process operations with a plurality of computer programs in a time-sharing manner as described above with priorities set for the plurality of computer programs. More specifically, in the above-described case, code-obtaining circuit 150 is provided with a priority-detection circuit as a priority-detection means, and a code-increasing-and-decreasing circuit as a code-increasing-and-decreasing means (not shown).
The priority-detection circuit accumulates the frequency of use which becomes the computer program priority for each portion of the instruction codes of the plurality of that computer programs that is temporarily held in instruction memory 140 and instruction memory 112, to detect the priority for the plurality of computer programs of which only portion of the instruction codes are temporarily held by instruction memory 140 and instruction memory 112
The code-increasing-and-decreasing circuit preferentially deletes the instruction code with a lower frequency of use, for example, by increasing and decreasing among a plurality of computer programs according to the priorities the number of the instruction codes only portion of which are temporarily held in instruction memory 140 and instruction memory 112.
If, for example, computer programs “A” and “B” are given, and instruction memories 140 and 112 in state control unit 105 and data-path unit 106 each temporarily hold at a time four operation states and four contexts, as described above, instruction memory 140 and 112 can each temporarily hold three contexts and three operation states of computer program “A” and temporarily hold one context and one operation state of “B”, according to the priority. In this way, array-type computer processor 100 can run a plurality of computer programs in parallel at still higher efficiency.
While preferred embodiments of the present invention have been described using specific terms, such description is for illustrative purposes only, and it is to be understood that changes and variations may be made without departing from the spirit or scope of the following claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2004-026800 | Feb 2004 | JP | national |
