JPS603751A - Data processing system - Google Patents

Abstract

PURPOSE:To perform a data flow processing without confusing operands with each other by providing an instruction taking-out part, an instruction memory, an instruction firing checking part, an operand memory, and an instruction executing part in a processor element. CONSTITUTION:An address converting part 7 sends a taking-out token FET to an instruction taking-out part 10, and the instruction taking-put part 10 sends a firing token FIT to an instruction firing checking part 12, and the checking part 12 sends an execution permitting token ET to an instruction executing part 14, and the executing part 14 sends a result token RT to the address converting part 7. The result token RT has a color (indentifier), a destination address, and an operand, and the color is given by a scheduler 2. The destination address consists of a function name (program name), a logical page address, and an address in a logical page. The name of the processor element is added to the result token RT sent from the address converting part 7 to a network 6.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field of the Invention] The present invention relates to a data processing method that allows an operand corresponding to a firing token to be retrieved from an operand memory without causing confusion of operands in a data flow machine. It is.

[Prior Art and Problems] Dataflow machines having the following characteristics have already been proposed.

1. It has a memory (operand memory) that makes the operand that arrives first wait until all the operands required for the instruction are available, and when the operand of a certain instruction arrives in the operand memory, a key is added to the operand. to find other operands required by the instruction, and if all operands required by the instruction are available,
It has a mechanism to send the operands to the instruction execution unit, and if they are not ready, to make the operands wait in the operand memory.

2. Furthermore, each operand of an instruction is assigned an address (
Destination address) is added, and when the 0th instruction becomes executable, %res is added to that instruction and sent to the instruction execution section for execution, and then the execution result is sent to the operand at that address. It has a mechanism to extract the next instruction and destination address to be used as a command.

3. When a function is called multiple times at the same time, in order to share the function code and process the functions at the same time, the operands used in the function called at one place are different from the operands used in the function called at a certain place. In order to distinguish the operands used when the same function is called and processed in different locations, an identifier (
color) to the operand and use the following command,
Pass the actual arguments and results of the function O (+) CALL L...Obtain a new color,
Send to LLINK command.

(II) LLINK: Adds a new color to the actual argument of the function and sends it to the function input.

(Ill) 0LINK...The return address of the function execution result (return address) and the current color (return color) are used as operands, a new color is added to them, and the result is sent to the OUT command.

(IV) OUT: Adds a return address to the function execution result and sends it to the return address.

Examples of such data flow machines include the DDDP of Oki Electric and the one proposed in Japanese Patent Application No. 57-231898. Furthermore, in the method of using the destination address as the operand memory key, unless a function call is made, the destination address of each instruction is unique. Can be used as However, when making a function call, when executing OUT, the destination address (OU
In the case of T, the destination address is called the return address) is 0L
The destination address cannot be used as a key because it is sent as an operand from INK, and if the same function is called multiple times within one function, when LLINK is executed, the execution results will go to the same destination, so the destination address is unique. If you use the destination address as a key, multiple LLIs will be generated.
Confusion between NK operands occurs. Therefore, when making a function call, data flow cannot be processed simply by using the destination address as the key of the operand memory.

[Purpose of the invention]

The present invention aims to eliminate the above-mentioned drawbacks, and is to provide a data processing method that enables dataflow processing to be performed without causing confusion between operands in a dataflow machine that is capable of making function calls. The purpose is to provide a method.

[Structure of the invention]

Therefore, the data processing method of the present invention includes a plurality of processor elements, a network for communicating between the processor elements, and a CALL instruction for acquiring a new color and sending it to the LLINK instruction and the 0LINK instruction. Adds a new color to the actual arguments of the function and sends them to the function entry address, and uses the return address indicating the return destination of the function execution result and the current color as operands, and adds the new color to it. This is a data processing method in a data flow machine that is designed to be able to execute the 0LINK instruction that sends the result of the function to the OUT instruction, and the OUT instruction that adds a return signal to the function execution result and sends it to the return address. The processor element includes an instruction fetching section, an instruction memory, an instruction firing inspection section, an operand memory, and an instruction execution section, and the instruction fetching section executes the instruction when the fetching token arrives. Based on the destination address of the token, the instruction code that uses the operand of the fetch token, the operand number indicating the operand number of the instruction, and the destination address of the execution result of the next instruction are extracted from the instruction memory and The combination of color and operand is sent as an ignition token to the instruction ignition inspection section, and the instruction code is OUT instruction or LL.
When the INK instruction is indicated, a special identifier is retrieved from the instruction memory and added to the ignition token, and the instruction ignition checker determines whether the instruction is OUT or LLINK when the ignition token arrives. In this case, the operand with the same key is searched from the operand memory using the color of the ignition token and a special identifier as a key, and if it is neither the OUT instruction nor the LLINK instruction, the color of the ignition token and the destination address are searched. The operand with the same key is searched from the above operand memory, and if the operands necessary for the instruction are available, the destination address, instruction code, and
The operands are sent together as executable 1--kun to the instruction execution unit 9, and if they are not complete, the operand, color, destination address, and operand number in the ignition token are stored in the operand memory, and an OUT instruction is sent. Or, in the case of an LLINK instruction, a special identifier is also stored in the operand memory, and the instruction execution unit executes the instruction based on the instruction code and operand, and stores the color, destination address, and execution result. It is characterized in that it is configured to be output as a result token.

[Embodiments of the invention]

The present invention will be explained below with reference to the drawings. ◇Figure 1 is a diagram showing an overview of a dataflow machine, Figure 2 is a diagram showing an example of a dataflow graph, and Figure 3 is a diagram showing an example of a dataflow graph. Figure 4 shows the format of the result token, Figure 5 shows the format of the ignition token, Figure 6 shows an example.
The figure shows the format of an executable token, FIG. 7 is a diagram explaining the operation of the instruction fetching section, FIG. 8 is a diagram explaining the operation of the instruction firing inspection section, and FIG. 9 is a diagram explaining the operation of the instruction execution section. diagram showing,
FIG. 10 is a diagram showing executable tokens and result tokens of the CALL instruction, FIG. 11 is a diagram showing executable tokens and result tokens of the LLINK instruction, and FIG. 12 is a diagram showing the 0LINK instruction.
FIG. 13 is a diagram showing executable tokens and result tokens of the K instruction, and FIG. 13 is a diagram showing executable tokens and result tokens of the OUT instruction.

In FIG. 1, 1-0 to l-n are processor elements, 2 is a scheduler, 3 is a host computer, 4 is a display, 5 is a DASD, and 6 is a network. Processor element 1-i (i
=0, 1=-n> has a configuration as shown in FIG. The processor element will be explained in detail later.

When the scheduler 2 receives a CALL instruction to call a function, it sends a new color (COL OR) and the name of the processor element that will execute the function to the issuing processor element. host computer 3
controls the entire dataflow machine and performs input/output processing. The network 6 is for sending tokens sent from a processor element, scheduler, or host computer to a target processor element, scheduler, or host computer.

FIG. 2 shows an example of a dataflow graph. In FIG. 2, the arrows are called arcs and the rectangular frames are called operation nodes. The data flow graph in Figure 2 multiplies the data of arc 4.5, sets this result to arc 6, and the data of arc 6 and the data on arc 7 (the numerical value "io
”).The number appended to the arc indicates the address within the logical page.

FIG. 3 is a diagram showing an example of a processor element. In FIG. 3, 7 is an address translation section, 8 is an address translation memory, 9 is a wait memory, 1o is an instruction fetch section, 11 is an instruction memory, 12 is an instruction firing check section, 13 is an operand memory, and FET is a fetch token. , FIT is the ignition token, ET is the executable token, ](T indicates the result token.

The address conversion section 7 converts the fetch token FET into the instruction fetch section 1.
0, the instruction retrieval unit 1o sends the ignition token FIT as shown in FIG. Then, the instruction execution section 14 sends the result talk/IIT as shown in FIG. 4 to the address conversion section 7.

The result token RT has a color, a destination address and an operand. The color is given by scheduler 2, and the destination address is the function name (program
It consists of "name), discussion-'°7 code" and address in the discussion-71-page. Although not shown, the name of the processor element is added to the result token RT sent to the network 6.

If the received result token RT should be sent to the network 6, the address conversion unit 7 sends it to the network 6, and if the received result token RT is to be processed internally, the address conversion unit 7 sends it to the network 6. See. Information indicating which program and which logical page is stored in each physical page of the instruction memory 11 is written in the address translation memory 8 . The address translation unit 7 stores the result token RT in the address translation memory 8.
If there is a destination logical page address, convert the logical destination address in the result token RT into a physical destination address indicating the address of the instruction memory, send this as an extraction token FET to the instruction fetching unit 10, and perform address conversion. If the destination logical page address of the result token RT does not exist in the memory 8, the result token RT is stored in the standby memory 9 and a page load request is issued to the host computer 3. When the desired logical page body is sent from the host computer 3, the logical page body is stored in the instruction memory 11, and then the corresponding result token RT is taken out from the standby memory 9 by the address translation unit 7, and the address It is converted and sent to the instruction fetching section 10. The initial large sword token sent from network 6 is the result token) (has the same format as T, and the extraction token FET is the result token) for other configurations, except that the destination address is a physical one. It is the same as token RT.

FIG. 7 shows the operation of the instruction fetching section 1o. When the fetch token FET arrives, the instruction fetch unit 1o extracts the instruction code that uses the operand from the destination address of the token, the number of the operand of the instruction (operand number), and the execution of the next instruction. The resulting destination address is taken out of the instruction memory 11, combined with the color and operand, is made into a firing token FIT, and sent to the instruction firing inspection section 12. If the search command is 0UT1LiLINK, the special identifier is also retrieved from the IM and added to the ignition token FIT. Although not shown, the number of operands is also taken out from the instruction memory 11, and the ignition token FIT
is added to. An unused address in the instruction memory 11 is used as the special identifier.

FIG. 8 explains the operation of the instruction firing inspection section 12. When the ignition token FIT arrives, the instruction firing inspection unit 12 searches the operand memory 13 for an operand whose color and special identifier are the same if the instruction is 0UT1LLINK, and for other instructions, it searches for an operand whose color and destination address are the same. Search operand memory 13 for operands that are the same. If all the operands necessary for the instruction are available (it is also said that the instruction has been fired), the color, destination address, instruction code, and operands are all sent to the instruction execution unit 14 as an executable token ET. If not, the operand, color, destination address, and operand number in the ignition token FIT are stored in the operand memory 13. However, for OUT and LLINK instructions, a special identifier is stored in the operand memory 13 instead of the destination address.

FIG. 9 explains the operation of the instruction execution section 14.

The instruction execution unit 14 executes the instruction using the instruction code and operand, and sends the color, destination address, and execution result together as a result token l (T) to the address conversion unit 7.

The instruction execution unit 14 may change the color and destination address of the executable token depending on the instruction code, and use the changed color and destination address as the color and destination address of the result token RT.

FIG. 1O shows executable tokens and result tokens for a CALL instruction. In addition, processor Niremen) 1
When the instruction execution unit 14 of -i receives the executable token ET of the CALL instruction, it does not perform any processing and sends it to the address translation unit 7, which sends it to the scheduler 2 via the network 6. send. Scheduler 2 can be considered to have the same configuration as processor element l-i. The operands of the CALL instruction include the name PEP of the processor element that executes the calling function F and the name f of the called function. When a CALL instruction is executed, a result token RT is created consisting of a color, a destination address, and a new color υ. The color and destination address of the resulting token It_T of the CALL instruction are unchanged. Scheduler 2 is a color
If a memory (not shown) is provided, this color memory is referred to and an unused color is set as a new color.

The new color of the result token RT corresponds to the operand, and although not shown, the name PEf of the processor element that executes the function f is also used as an operand, CA, L.
It is added to the result token of the L instruction. The result token RT of the CALL instruction is sent to processor element PEF. Note that PEf may be the same as PEp.

FIG. 11 shows the executable and result tokens of the LLINK instruction. When the LLINK executable token ET is sent, the processor
The instruction execution unit 14 of the EF sets the value of the new color of the executable token ET as the value of a part of the color of the result token RT, and sets the function entry address of the executable token ET to the result token R.
Fill in the destination address field of T, write the operand of executable token ET (actual argument of function f) in the operand field of result token RT, and further write result token R (not shown).
The processor element PEf is added to the beginning of T and sent to the address conversion unit 7. The result of this LLINK, RT, is sent to processor element PEf via the network.

FIG. 12 shows the executable and result tokens of 0LINK. When the 0LINK executable token ET is sent, the processor
The instruction execution unit 4 of F takes the new color of the executable token ET as the value of part of the color of the result token RT, leaves the address of OUT as is, and sets the color of the executable token ET and the return address as the result token. 11 Fill in the operand field of T, and further add processor elementization PEf to the beginning of the result token RT (not shown) and send it to the address conversion unit 7. The result token RT of this 0LINK is
Processor element PEf via network 6
sent to. Note that when the instruction memory 11 is read and accessed using the destination address of the result token RT of the CALL instruction, the LLINK instruction and the 0LINK instruction are taken out from the instruction memory 11.

FIG. 13 shows the executable and result tokens of OUT. Upon receiving the LLINK result token RT, processor element PEf performs the operation of function f based on its actual arguments.

When the execution result is created, the processor
The OUT instruction is executed by Ef. When the executable token ET of OUT is sent, the instruction execution unit 14 of the processor PEf sends the executable token E
Write the return color value of T in the color part of the result token RT, write the return address of the executable token ET in the destination address part of the result token RT, and write the operand (result of the function) of the executable token ET as the result. As an operand of the token RT, and although not shown, the result token ItT
The name of the processor (PEF) is added to the beginning of the file and sent to the address conversion unit 7. The resultant token of the OTJ T with is sent via the network 6 to the processor element PEF.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, in a data flow machine that can perform function calls, instructions are stored in OU as a key to operand memory.
If the instruction is neither T nor LLINK, the destination address and color are used; if the instruction is OUT or LLINK, a special identifier and color are used to distinguish it from the operands of other instructions within the function. Therefore, data flow processing can be performed without causing confusion of operands.

[Brief explanation of drawings]

Figure 1 is a diagram showing an overview of the data book --- q>:y.
FIG. 2 is a diagram showing an example of a data flow graph. Fig. 3 is a diagram showing an example of a processor element, Fig. 4 is a diagram showing an example of a processor element.
5 shows the format of the ignition token, FIG. 6 shows the format of the executable token, FIG. 7 shows the operation of the instruction fetching unit, and FIG. 8
9 is a diagram illustrating the operation of the instruction execution unit. Figure 2g10 is a diagram illustrating the executable token and result token of the CALL instruction. FIG. 12 is a diagram showing executable tokens and result tokens. FIG. 12 is a diagram showing executable tokens and result tokens of the 0LINK instruction. FIG. 13 is a diagram showing executable tokens of the OUT instruction. 1-0 to 1-n...processor element, 2
...Scheduler, 3...Host! t n 44'
4, 4...Display, 5...DASD, 6...
...Network, 7...Address translation unit, 8...
Address conversion memory, 9...Waiting memory, 10...Instruction fetching unit, 11...Instruction memory, 12...Instruction firing inspection unit, 13 River operand memory, FET...Fetching token, FIT... ...Ignition token, ET...Executable token. 9Q

Claims (1)

[Claims] It has multiple processor elements and a network for communication between the processor elements, and acquires a new color and uses the LLINK command and 0LINK
CALL command to send to the command, and L to add new colors to the actual arguments of the function and send them to the function entry address.
The LINK instruction, the return address indicating the return destination of the function execution result, and the current color are used as operands, and a new color is added to the 0LINK instruction, which is sent to the OUT instruction.The 0LINK instruction adds the return color to the function execution result. A data processing method in a dataflow machine capable of executing an OUT instruction to send the OUT instruction to a return address, the processor element comprising: an instruction fetching unit;
The instruction fetching section includes an instruction memory, an instruction firing inspection section, an operand memory, and an instruction execution section, and when the fetching token arrives, the instruction fetching section uses the operand of the fetching token based on the destination address of the token. The instruction code to be executed, the operand number indicating the operand number of the instruction, and the destination address of the execution result of the next instruction are retrieved from the instruction memory, and these, together with the color and operand, are used as the ignition token to execute the above instruction. In addition to transmitting the command to the ignition inspection section, if the retrieved instruction code indicates an OUT command, LT, or INK command, a special identifier is retrieved from the command memory and added to the ignition token, and the command is fired. When the ignition token arrives, if the instruction is OUT, T, or LINK, the inspection unit searches for an operand with the same key from the operand memory using the color of the ignition token and a special identifier as a key, and issues an OUT instruction. and LLINK instructions, the trap searches the above operand memory for an operand with the same key using the color of the firing token and the destination address as keys, and if the operands required by the instruction are available, then Sends the color, destination address, instruction code, and operand together as an executable token to the instruction execution unit, and if they are not complete, stores the operand, color, destination address, and operand number in the ignition token in the operand memory. together with O.U.
In the case of a T instruction or a LLINK instruction, a special identifier is also stored in the operand memory, and the instruction execution unit executes the instruction using the instruction code and operand, and also stores the color, destination address, and execution result. A data processing method characterized in that the data processing method is configured to output a result token as a result token.