JP2765058B2 - Stream operation completion detection method and apparatus - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a data flow processor in which a memory unit and an operation unit are connected by a pipeline bus, and the operation order is controlled by a data driving method. The present invention relates to a stream computation end detection method for controlling the computation of a token and a device therefor.

[Conventional technology]

Conventionally, as a data flow type processor,
There is a technology described in Japanese Patent Application Laid-Open No. 58-70360, and as a product thereof, there is a μPD7281 which is a data flow processor manufactured by NEC Corporation.

The μPD7281 has a configuration as shown in FIG.
A token, which is a unit of data input to the data flow processor from the external bus, includes a data value, a link table address for referring to the link table 92 after input, and a module indicating the data flow processor in which the token is to be processed. With numbers. Token input unit 91
Input the token internally if the module number of the token passing through the external bus matches the number of the data flow processor; otherwise, the token output unit
Output to the external bus as it is through 97. The input token refers to the link table 92 based on the link table address of the token, and obtains a function table address for referencing the function table 93 and a link table address for referencing the link table 92 next time. After that, it is sent to the function table 93.

The token refers to the function table address in the function table 93, and refers to / updates the management information in the data memory 94. At the same time, the processing code indicating the processing contents in the processing unit 96, the access address of the data memory 94, and the like. Get.
The token is sent from there to the data memory 94, where it waits for the operand of the other party of the binary operation or reads out the constant for the constant operation as necessary. The queue memory 95 is a memory for temporarily holding the token when the processing unit 96 is processing the previous token and cannot input the next token.If the processing unit 96 is not busy, the token is processed from the queue memory 95. The data is sent to the unit 96, and according to the processing code, one of the addition, subtraction, multiplication, logical operation, shift, comparison, bit inversion, priority encoding, diversion, numerical value generation, copy, operation using the internal register, etc. of the integer data is performed. Receive one process. The processed token is sent to the link table 92, and is referred again by the link table address. In the same manner, the circuit goes around the internal ring bus until the external bus output instruction is executed, and receives necessary processing for the data value.

If the token is to be output to an external bus by an external bus output instruction after being processed by the processing unit 96, the processing unit 96 receives the last operation processing in the data flow processor, and then the link table 92 Fetches a link table address to be held as an external bus token, and fetches a processing code indicating a module number and output to be held as an external bus token in the function table 93, and then inputs the queue to the queue memory 95 via the data memory 94. . If the processing code of the token indicates an output at the exit of the queue memory 95, the token is sent from the queue memory 95 to the token output unit 97, and the module number and the link table address obtained as described above are obtained. After being transformed into the form of an external bus token by using, the data is output to the external bus.

As described above, in the conventional data flow processor, a method is used in which a plurality of tokens to be subjected to the same processing are given the same token identifier (referred to as a link table address) and are continuously transmitted. Such a continuous flow is called a stream.

In a conventional processor, when performing cumulative addition of stream data, the following method is used as shown in FIG. 7. That is, the method shown in FIG. A unit is provided with a register, and each time a token is input by inputting stream data, the register performs cumulative addition.

In the method shown in FIG. 7 (b), the addition is decomposed into a combination of binomial additions, the program is programmed in a tournament system so that the sum is finally obtained, and then the stream data is distributed and input.

The method shown in FIG. 7 (c) uses the “convo” of μPD7281.
Instruction ", the stream data input to the operation node and the partial sum calculated so far are divided by an odd-numbered number, and two terms are added using the addition instruction below,
After performing the operation of returning the result to the input arc the number of times corresponding to the length of the stream given in advance, the cumulative addition result is output.

By such a method, the data flow processor performs the cumulative addition of the stream data.

[Problems to be solved by the invention]

However, the method of performing the cumulative addition of the stream data in the data flow processor has the following problems.

In the method shown in FIG. 7 (a), that is, the method using a register, the register in the processing unit is used, so that the cumulative addition at a plurality of locations cannot be performed at the same time. Required.

In the method shown in FIG. 7 (b), that is, the method based on the tournament system, a program for defining the data flow must be prepared so that the tournament system can be added to data of a predetermined length.

The method shown in FIG. 7C, that is, the method based on the convo instruction cannot be used unless the length of the stream is previously known by the convo instruction, and lacks flexibility.

As described above, in the conventional data flow processor, when a stream data of an arbitrary length is flowed to one arc in a data flowgram and their cumulative addition is taken, it can be used in parallel at a plurality of locations and corresponds to an arbitrary length. There is no means for performing the cumulative addition, and there is a problem in that the processing is inefficient regardless of which method is used. This is because there is no means for detecting the end of an arbitrary-length stream input to the conventional data flow processor and detecting the end of the operation.

The present invention provides a method for accelerating the cumulative addition of a variable-length stream by flowing a token with an end mark as the end of stream data and detecting the end of the token, and a data flow processor realizing the method. It is intended to be.

[Means for solving the problem]

In the first invention, processing is performed by flowing a token having a link table address, which is an identifier for distinguishing each other, data to be processed, and an end mark indicating the end of a stream, into an internal bus as a data unit. In a stream operation completion detection method for detecting the completion of an operation on a data stream composed of a plurality of tokens by a data flow processor having a data memory, the termination state obtained as a result of the immediately preceding determination processing for each token input The value of the flag, the value of the write flag storing whether or not the data was written to the data memory immediately before, and the end mark of the token input this time are input. Whether to write the data value of the input token to the data memory as an operation for the token And the value of the termination state flag as a result of the current determination processing, the value of the write flag as a result of the current determination processing, the value of the link table address of the token to be output as a result, and the value of the token to be output as the result. Determine the value of the end mark and the operation code that specifies the operation at the subsequent stage of the token to be output as a result, and determine the end of the stream operation by giving different link table addresses during and at the end of the operation on the stream. It is characterized by:

The second invention has an internal memory that performs processing by flowing a token having an identifier for distinguishing each other, data to be processed, and an end mark indicating the end of a stream as an data unit to an internal bus. In the stream operation end detecting device of the data flow processor, an input token register for temporarily holding an incoming token, an operation code field referred to by an address in the token register and specifying an operation, and data for accessing the internal memory. A code table having a memory address field, a state holding unit accessed and held by an address in the token register, which holds and stores an end state flag field and a write flag field as internal states for realizing the end detection operation; The input token register and the previous An end detection controller that receives the contents of the code table and the state holding unit as input, updates the contents of the state holding unit, and determines a value of an output token; and links the input token register by a signal of the end detection controller. And an incrementer for generating a link table address field of the output token from the value of the table address field.

[Action]

When the end-of-stream detection method of the present invention is used, an end mark is given in advance to the last token of the stream to be subjected to cumulative addition.

If there are two consecutive token inputs, the end-of-stream detection device writes the former into memory and reads and adds it when the latter arrives. When a stream is input, an operation of adding two adjacent data streams is performed, and the partial sum is output as a token and returned as an input stream to the original instruction. In each addition, the length of the input stream is reduced by one.

The stream operation completion detection device holds the value of the token termination mark processed last time in the termination status flag, and continues the above-described operation until both this and the token termination mark input this time become “1”. When both become "1", it is determined that the stream length has become 1 and the data at that time has another link table address as a result of the cumulative addition of the entire original stream data. Output by token (ie from another location).

According to the present invention, by detecting the end of a stream according to the above-described operation, a variable-length and efficient stream cumulative addition can be realized.

〔Example〕

 Embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a diagram showing the basic configuration of a stream operation end detecting device according to the present invention, FIG. 2 is a diagram showing the configuration of a data flow processor having the stream operation end detecting device, and FIG. It is a figure showing a format.

The data flow processor shown in FIG. 2 performs processing by flowing a token having an identifier for distinguishing each other, data to be processed, and an end mark indicating the end of a stream to an internal bus as a unit of data. A data flow processor having an internal memory, comprising: a bus token input unit 11, a link table 12, a stream operation end detecting device 1, a data memory 14, and a buffer queue 15.
, Processing unit 16 and bus token output unit
17 and.

First, FIG. 2 showing the configuration of the data flow processor and FIG. 3 showing the token format inside the data flow processor.
The operation of the data flow processor using the present invention will be described with reference to the drawings. This data flow processor is composed of units shown by rectangles in FIG. 2. Each unit exchanges a token in synchronization with a pipeline clock supplied in common from the outside, and operates.

In FIG. 2, the data flow processor operates according to a token input from the outside via the bus token input unit 11. The token in the link table 12 includes a link table address 63 for identifying the token in a data section 64 having a data value to be processed as shown in FIG. 3 (a), and an end for indicating the end of the stream. With mark 62. In the stream, a plurality of tokens having the same link table address flow while maintaining the order of each other, and arithmetic processing is performed using the fact that the relative positional relationship between the tokens in the plurality of tokens is invariable. The end mark has a value of "1" for the last token and "0" otherwise.

The token is a processing unit shown in FIG.
Before receiving the operation at 16, the link table address to be held after the operation and the code table address 73 for referring to the stream operation end detecting device 1 are stored in the link table 12.
Fetch with By this reference, the token having the format shown in FIG. 3 (b) is sent to the stream operation end detecting device 1 and undergoes a stream operation end detecting operation as described later. The token is then transferred to data memory 14 as needed.
Along with the two operands obtained by the memory read in step (1) and the operation code used in the processing unit 16 (FIG. 3 (d)), the data is sent to the processing unit 16 through the buffer queue 15, where the operation processing according to the operation code is performed. Receive. After the arithmetic processing, the link table 12 is transmitted via the signal line 103 according to the link table address.
And the same operation continues, or is output to the outside of the data flow processor as a result token via the bus token output unit 17.

Next, the configuration and operation of the stream calculation end detecting device 1 will be described with reference to FIGS.

The stream operation completion detecting device 1 includes an input token register 4 for temporarily storing an incoming token, an operation code 30 which is referred to by an address in the token register 4 and specifies an operation, and a data memory address for accessing an internal memory. A code table 5 having 32; a state holding unit 6 which holds and stores an end state flag 36 and a write flag 37 as internal states for realizing the end detection operation; An end token controller 8 that receives the contents of the input token register 4, the code table 5, and the state holding unit 6, updates the contents of the state holding unit 6, and determines the value of the output token. The signal of the output token is converted from the value of the link table address of the input token register 4 by the signal. It consists incrementer 9 for generating a table address 71.

In such a stream operation end detection device 1,
From the link table 12 in the preceding stage shown in FIG.
The token input to the stream operation end detecting device 1 through 1 is latched in the input token register 4 of the stream operation end detecting device 1 shown in FIG. 1 in synchronization with the pipeline cycle of the data flow processor. . In FIG. 1, the register is divided and written so as to correspond to each bit field in the token. Code table 5 based on the code table address 73 of the input token
And the state holding unit 6 are referred to.

The code table 5 is a table memory having a plurality of fields shown in FIG. 3 (e), and the row referred to here is the device for detecting the stream operation end of the token 1, the data memory 14 and the processing unit 16 shown in FIG. The operation in is specified. The operation code 30 shown in FIG. 1 is an instruction code indicating an operation in the stream operation end detecting device 1 and the processing unit 16 shown in FIG. The stream operation completion detection device 1 executes operations such as token waiting and data memory read / write that the data flow processor has in the background art in addition to the stream addition instruction described in detail below. The data memory address 32 shown in FIG. 1 gives an address when accessing the data memory by the stream addition instruction.

The state holding unit 6 is a table having a plurality of rows as shown in FIG. 3 (f), and the referenced row holds the internal state of the stream operation completion detection device to which the token refers. The termination state flag 36 basically indicates whether or not the termination token has been processed at the time of the previous instruction execution of the line.
The write flag 37 indicates whether the data memory was last written or read.

The end detection controller 8 calculates the operation code field of the line specified by the code table address 73 on the input token register 4 in the code table 5, the termination mark 72 from the input token register 4, and the input token The state held in the line designated by the code table address 73 on the register is input, and the state transition described above is performed. The state of the result is written back to the same line of the state holding unit 6, and the end mark 82 of the token output from the stream operation end detecting device 1, the data memory code 8
3. Generate and output each field of the operation code 85. Also, it outputs a signal 86 to the incrementer 9.

The input / output of the end detection controller and the update of the state holding unit follow Table 1 below.

However, the first half of the "operation code" describes the operation in the data memory, and the second half describes the operation in the processing unit. “Write” writes to data memory, “read”
Is read from data memory, "dmnop" is no effect on data memory (NOP), "add" is the addition of two operands in the processing unit, "nop" is the NOP output of the first operand in the processing unit, “ERROR” indicates a transition to an error state. When the write “write” is performed in the data memory, the token disappears in the data memory, so that the corresponding output token term is not written. When the incrementer signal 86 shown in FIG. 1 is "1", the incrementer 9 outputs a value obtained by adding "1" to the input, and otherwise outputs the input as it is.

The stream operation completion detecting device 1 forms an output token as described above and sends it to the subsequent data memory 14 via the signal line 102 in FIG. FIG. 3 (c) shows the format of the token sent from the stream operation completion detecting device 1 to the data memory 14. The data section 80 has the same data as when the token was input from the synchronization queue, and the rest is determined by the operation of the stream operation completion detection device 1.

The data memory code 83 passes (I) the input token without accessing the data memory. For the output token, the right operand is set to “0” (code dmnop). (II) The data memory 14 is read using the value of the data memory address 84 as an address, and the read value is sent as the right operand 66 (code read). ), (III) Write the data value of the token in the data section 80 to the data memory using the value of the data memory address 84 as the address, and erase the token there (code writ
e) Specify either The data memory address 84 indicates an access address when the token reads and writes data in the data memory 14 shown in FIG.

The operation code 85 is fetched from the code table 5 of the stream operation end detecting device 1 shown in FIG. 1, and specifies the operation of the stream operation end detecting device and the operation instruction in the processing unit 16.

The operation of the present invention, ie, the cumulative addition of stream data, according to the stream operation end detection method of the first invention will be described with reference to FIG. As shown in FIG. 4, a data sequence to be cumulatively added is input as stream data to a node 20 that executes a “stream addition instruction”. Only the last token T4 of the stream data is a token whose end mark is "1". The stream addition instruction has a termination state flag and a write flag therein. When a token is input, the stream addition instruction performs a determination according to the flowchart of FIG. And perform an operation determined based on the result. However, in FIG. 5, a square with two output branches indicates that an operation of determining whether the value inserted therein is "1" or "0" is performed. The other rectangles describe the new state of the state holding unit and the end mark of the output token among the operations performed as a result. “ES
M "," ESF ", and" WF "represent the end mark of the input token, the end state flag of the state holding unit, and the write flag, respectively.

Hereinafter, this operation will be described by taking as an example an operation when a stream having a length of 4 is input. The initial value of the state holding unit in the stream operation completion detection device corresponding to the node that executes the stream addition instruction is such that the end state flag is “0” and the write flag is “0”.

(1) First, when the leading token T1 enters this node with the terminating mark "0", the code table 5 and the state holding unit shown in FIG. 1 are referred to by the code table address of the token. By this input, the row of a in Table 1 which is the operation table of the end detection controller shown earlier is executed, and the data of the token is written to the data memory in accordance with this. The token disappears in the data memory.

(2) The end mark of the second token T2 is also "0"
However, since the write flag is "1", the row of Table 1 is executed, and the data read from the data memory and the data held by the input token are added by the processing unit. The result is output by a token T5 having the given link table address, which joins the original arm 21 via the arc 22 in FIG.

(3) Token T5 having a partial sum of tokens 1 and 2 obtained in (2) above, and third token T3 of the original stream
Merges at the entrance of the node 20 shown in FIG. Whichever arrives earlier, the data for the one who arrived earlier is shown in Table 1a
Is written to the data memory according to the row of (1), and the one arriving later reads it out according to the row of Table 1b and adds both data. This gives a token T6 with the third partial sum of the stream, which is also
Input from 21 to node 20.

(4) Assuming that the fourth token T4 of the stream arrives earlier than the token T6 having the partial sum of these three data, since it has the end mark "1", the data is stored in the data memory according to the row of Table 1e. And the termination status flag is set to "1".

(5) Subsequently, when the other token T6 arrives at the node 20 following the token T4, the row of Table 1 d is executed. The value of the data memory is read and added, and the end state flag of the stream operation end detection device becomes "1". The end mark of the token T7 having the addition result is “1”, and the token T7 is input to the node 20 from the arc 21 via the arc 22.

(6) When the token T7 is input to the node 20, Table g
Is selected, whereby the data of the token T7 is output without any operation. However, the incrementer in the stream operation completion detection device operates, and the link table address “1” of the output token is added. As a result, the output token is output from the arc 23 different from the other cases, and has an accumulation result of four data of the stream.

In the above description, the order of arrival at the time of merging of two tokens is assumed. However, any other order will work correctly. Although the length of the stream is set to 4 as an example, the same operation is performed for a stream having an arbitrary length of 2 or more, and a correct cumulative addition result can be obtained.

〔The invention's effect〕

As described above, stream data having an arbitrary length can be calculated by using the “stream addition instruction” in which the method of the present invention is implemented by the device of the present invention to calculate the stream with the end mark at the end of the stream. Has the effect of realizing the cumulative addition of.

In this method, since no register is used, this instruction can be used in parallel at a plurality of locations, and a stream whose length is not known in advance can be processed.

[Brief description of the drawings]

FIG. 1 is a basic configuration diagram showing an example of a stream operation end detection device of the present invention, FIG. 2 is a configuration diagram of a data flow processor having the stream operation end detection device, and FIG. 3 is a token used for describing the present invention. FIG. 4 is a diagram showing the format of a table, FIG. 4 is a data flow graph describing the operation of a data flow processor when the present invention is used, FIG. FIG. 6 is a configuration diagram of a conventional data flow processor, and FIG. 7 is a flow graph showing a method of realizing the cumulative addition of stream data in the conventional data flow processor. 1 stream operation end detecting device 4 input token register 5 code table 6 state holding unit 8 end detecting controller 9 incrementer

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 15/82 JOIS (JICST)

Claims (2)

(57) [Claims] Processing is performed by flowing a token having a link table address as an identifier for distinguishing each other, data to be processed, and an end mark indicating the end of a stream to an internal bus as a unit of data. In a stream operation completion detection method for detecting the completion of an operation on a data stream composed of a plurality of tokens by a data flow processor having a data memory, a termination state flag obtained as a result of a previous determination process for each token input , The value of a write flag that stores whether or not the data was written to the data memory immediately before, and the end mark of the token that has been input this time. As an operation for whether or not to write the data value of the input token to the data memory , The value of the termination state flag as a result of the current determination processing, the value of the write flag as the result of the current determination processing, the value of the link table address of the token to be output as a result, and the end of the token to be output as the result Determining the end of stream operation by determining the value of the mark and the operation code that specifies the operation at the subsequent stage of the token to be output as a result, and giving a different link table address during and at the end of the operation on the stream A stream operation end detection method. 2. Data having an internal memory, which is processed by flowing a token having an identifier for distinguishing each other, data to be processed, and an end mark indicating a stream end to an internal bus as a data unit. An input token register for temporarily holding an incoming token, an operation code field that is referred to by an address in the token register and specifies an operation, and a data memory for accessing the internal memory. A code table having an address field, a state holding unit that holds and stores an end state flag field and a write flag field as internal states for realizing the end detection operation, and is accessed by an address in the token register; Input token register and said An end detection controller that receives a code table and the contents of the state holding unit as input, updates the contents of the state holding unit, and determines a value of an output token, and a link table of the input token register according to a signal of the end detection controller. A stream operation end detecting device, comprising: an incrementer for generating a link table address field of an output token from a value of the address field.