JP3732530B2 - Data flow computer execution control device - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to an execution control apparatus for a data flow computer that performs asynchronous parallel execution control.
[0002]
[Prior art]
The data flow computer expresses a program as a data flow graph composed of arcs indicating paths between nodes, and executes a calculation by flowing data values called tokens while updating the graph structure.
[0003]
Nodes include operation nodes and control nodes. The operation node receives the token value, performs the given operation, and outputs the result as the token value. The control node controls the route through which the token flows.
[0004]
The data flow computer includes a data driven execution control device and a request driven execution control device as execution control devices.
[0005]
Data driven execution control device:
The data driving execution control apparatus performs processing on all nodes of the data flow graph by the following method.
[0006]
1. Causes the node to wait until a token reaches all input arcs.
[0007]
2. After the tokens are arranged on the input arc, the function given to the node is executed. This is called node firing. At this time, the token on the input arc is removed.
[0008]
3. The execution result is output as a token on the output arc.
[0009]
By firing each node according to the execution method described above, the entire graph is executed in asynchronous parallel.
[0010]
A configuration example of a conventional data drive execution control device is shown in FIG. The function of each means in FIG. 5 will be described below.
(Ignition control means 10C)
The token arrival monitoring function 12 detects and fires a node with tokens on all input arcs. The node classification function 13 classifies the fired node into an operation node and a control node, and sends a control signal to the corresponding instruction execution means. A node having an operation instruction is called an operation node, and a node having a control instruction is called a control node.
(Calculation instruction execution means 20C)
The input token processing function 21A erases all tokens on the input arc of the fired node, and sends the token value to the operation execution function 22. The operation execution function 22 performs the operation given by the operation instruction on the token value and sends the result to the token output function 23C. The token output function 23C substitutes the calculation result for the token value and places the token on the output arc of the node that fired the token.
(Control command execution means 30C)
The control instruction decode function 31 identifies the instruction of the control node and sends it to the corresponding instruction execution function. In the instruction execution function, control is performed according to a given node type. There are the following types of control nodes.
[0011]
(Distribution node (D) 32C) 1 input n output node. Delete the token on the input arc and place the token with the copied value on all output arcs.
[0012]
(Branch node (SW) 33) A node with two inputs and two outputs. When the value of the token on the second input arc is true, the token on the first input arc is placed on the first output arc. If false, place the token on the first input arc on the second output arc.
[0013]
(Procedure call node (CALL) 34A) This is a node with two inputs and one output. The token on the second input arc is copied, the procedure start node of the function is obtained from the function name stored as the value of the token on the first input talk, and placed on the first input arc of that node. In addition, a token having a value corresponding to the name of the output arc is generated. A procedure return node of the function is obtained from the function name stored as the token value on the first input arc, and the generated token is placed on the first input arc of the node. Erase the token on the input arc.
[0014]
(Procedure start node (ENTRY) 36) This is a node with one input and one output. Place a copy of the token on the input arc on the output arc. Erase the token on the input arc.
[0015]
(Procedure return node (RETURN) 35C) This is a node with 2 inputs and 0 outputs. Copy the value on the second input arc to generate a token. As the token value on the first input arc, the generated token is placed in the designated arc.
[0016]
FIG. 6 shows an execution example. In the figure, a black circle represents a token, a white circle node represents an operation node, and a white circle and cross node represent a distribution node. A shaded node represents a node that has finished execution. The operation will be described by paying attention to the bold line nodes in the figure.
[0017]
1. The token arrival monitoring function 12 does not ignite the thick line node because the tokens are not aligned on the input arc (FIGS. 6 (1) to (3)).
[0018]
2. After the tokens are arranged on the input arc (FIG. 6 (4)), the token arrival monitoring function 12 fires the node.
[0019]
3. The node classification function 13 identifies the fired node as an operation node, and sends a control signal to the operation instruction execution means 20C.
[0020]
4). The input token processing function 21A removes tokens on the input arc.
[0021]
5). Calculation is performed by the calculation execution function 22.
[0022]
6). The token output function 23 places a token on the output arc (FIG. 6 (5)).
[0023]
By executing as described above, the entire graph is executed.
[0024]
Request driven execution control device:
The request driving execution control device selects only the node on the route to the designated node and performs calculation. This execution control apparatus performs processing on all the nodes of the data flow graph by the following method.
[0025]
1. (A) The node is made to wait until a request signal (demand) arrives on the output arc.
[0026]
(B) After the demand reaches the output arc, the demand is output to the input arc.
[0027]
(C) The node waits until the token reaches all input arcs.
[0028]
2. After the tokens are aligned on the input arc, fire the node. At this time, the token on the input arc is removed.
[0029]
3. The calculation result is output to the output arc.
[0030]
FIG. 7 shows an execution example. In the figure, a white circle node represents an operation node, a white circle and cross node represent a distribution node, a black circle represents a token, and a circle with right-bottom eight hatching represents a demand. Also, the nodes with left-down hatching indicate the nodes that have received demand, and the shaded nodes indicate the nodes that have fired. When each node is executed individually as described above, the operation of the entire graph is as follows.
[0031]
1. The demand is propagated from the designated node to the input side from the output side of each node on the graph (FIG. 7 (1) to (5)).
[0032]
2. When all demands have propagated to the start node, a token is sent from the start node to the output arc (FIG. 7 (6)).
[0033]
3. Data driven execution is performed on a subgraph composed of nodes that have received demand (FIGS. 7 (7) to (10)).
[0034]
As described above, only the node receiving the demand is selectively executed.
[0035]
Colored token method:
In a data flow computer, there is a colored token method as a multiple execution method under a multiple execution environment. A tag called color is attached to the token, and a color is set for each execution environment. The execution control by the colored token method is realized by changing the items 1, 2 and 3 of the data drive execution control device and the items 1 (c), 2 and 3 of the request drive execution control device as follows.
[0036]
1. The node is made to wait until tokens of the same color arrive on all input arcs (the same applies to 1 (c)).
[0037]
2. After the tokens with the same color are aligned on the input arc, the node is fired. At this time, tokens having the same color on the input arc are removed.
[0038]
3. The result is output to the output arc. At this time, the token color on the input arc is copied and output.
[0039]
Using the above execution method, one graph can be executed in parallel under a plurality of execution environments (by color) without affecting each other.
[0040]
FIG. 8 shows a configuration example of an execution control apparatus using a conventional colored token method. The function of each means in FIG. 8 will be described below.
(Ignition control means 10D)
The colored token arrival monitoring function 16 detects and fires a node having tokens of the same color on all input arcs. The node classification function 13 classifies the fired node into an operation node and a control node, and sends a control signal to the corresponding instruction execution means. (Calculation instruction execution means 20D)
The input token processing function 21B deletes all tokens having the same color on the input arc of the fired node, and sends the token value to the calculation execution function 22. The operation execution function 22 performs the operation given by the operation instruction on the token value, and sends the result to the token output function 23D. The token output function 23D substitutes the calculation result for the token value, copies the erased token color from the input arc, and places the token on the output arc.
(Control command execution means 30D)
The control instruction decode function 31 identifies the instruction of the control node and sends it to the corresponding instruction execution function. In the instruction execution function, control is performed according to a given node type. There are the following types of control nodes.
[0041]
(Distribution node (D) 32B) 1 input n output node. Erasing tokens on the input arc and placing tokens with copied values and colors on all output arcs.
[0042]
(Branch node (SW) 33) A node with two inputs and two outputs. When the value of the token on the second input arc is true, the token on the first input arc is placed on the first output arc. If false, place the token on the first input arc on the second output arc.
[0043]
(Procedure call node (CALL) 34C) is a node with two inputs and one output. When firing, find the color that is not used by any token on the data flow graph. A token is generated based on the token value on the second input arc and the newly generated color. The function name is obtained from the token value on the first input arc, and the generated token is placed on the first input arc of the procedure start node. In addition, a token with the generated color is generated using a pair of the color of the token on the input arc and the name of the output arc as a value. The procedure return node of the function is obtained from the function name stored as the value of the token on the first input arc, and the generated token is placed on the first input arc of that node. Erase the token on the input arc.
[0044]
(Procedure start node (ENTRY) 36) This is a node with one input and one output. Place a copy of the token on the input arc on the output arc. Erase the token on the input arc.
[0045]
(Procedure return node (RETURN) 35D) This is a node with 2 inputs and 0 outputs. The color before the procedure call is obtained from the token on the first input arc, and the value on the second input arc is copied to generate a token. Place the generated token in the arc specified as the value of the token on the first input arc.
[0046]
An actual execution example is shown in FIG. In the figure, black circles (K, a) represent tokens of value K color a. Here, an execution example is shown with attention paid to one node.
[0047]
1. The colored token arrival monitoring function 16 detects that tokens of the same color (i) are arranged on all the input arcs, and fires the node (FIG. 9 (3)).
[0048]
2. The node classification function 13 identifies the fired node as an operation node, and sends a control signal to the operation command means 20D.
[0049]
3. The input token processing function 21B removes the token of color (i) on the input arc.
[0050]
4). The calculation execution function performs calculation 22.
[0051]
5). The token output function 23D outputs a token having the same color (i) to the output arc (FIG. 9 (4)).
[0052]
[Problems to be solved by the invention]
For example, in an educational support program, you may want to give more problems to learners who do not have basic knowledge, and let learners who have more basic knowledge select some of those problems. . Here, the difference in basic knowledge is regarded as a difference in level, and it is assumed that the more basic knowledge there is, the higher the level.
[0053]
High level learners do not need much basic knowledge. In contrast, learners with lower levels must acquire basic knowledge with more problems. In other words, it is considered that a problem for a learner at a low level is prepared by adding a problem for acquiring basic knowledge required by a learner at a low level to a problem required by a learner at a high level. In such a case, the problem that is required by a high-level learner is a subset of the problems that are required by a low-level learner.
[0054]
Here, let us consider expressing a problem presentation program in which a function for presenting a certain problem is a node and the order of presentation is an arc. When problem presentations corresponding to a plurality of levels are performed, if a function for selectively executing each node is given for each level, one problem presentation program may be created and shared at each level. If there is no such function, a separate program must be created for each level. As described above, when a program is created for each level, when a program corresponding to a certain level is corrected, the same correction must be made to a program below that level. At this time, when the common part is corrected, the number of programs to be corrected further increases. In order to solve such a problem, a function for selectively executing a node for each level is required.
[0055]
As a method for realizing a function of selectively executing a node for each level, a method using a conditional branch is conceivable. That is, by adding a branch program in front of the node and sending only tokens that satisfy the condition to the node, the node can be selectively executed according to the level. In this method, a branch program must be assigned to all nodes. Therefore, there is a problem that the branch program needs to be corrected when the problem presentation level is changed. As described above, in the method using the conditional branch, when the program is created / modified, it is necessary to create / modify a location related to the conditional branch, which increases the burden on the program creator. In order to solve such a problem, the data flow computer needs to have a function of selectively executing a node for each level.
[0056]
In the data driven execution control device, execution is advanced by firing when tokens are arranged on the input arc. In this apparatus, means for selecting an execution path is provided by a branch instruction and a procedure call instruction. In the request drive execution control device, in addition to the selection based on the above command, the demand can be propagated, and the node to which the demand has reached can be selectively executed. However, these devices do not have a function of selectively firing a node according to the level.
[0057]
The colored token model is a technology that expresses the difference in execution environment by the color of the token, detects the difference in execution environment at the time of execution, and advances the execution by using the token of the same execution environment. This method is not a technique corresponding to the difference in level, and the token is executed for the entire graph or all the nodes to which the demand has been propagated according to the color of the token. Therefore, it does not have a function of selectively firing a node according to the level.
[0058]
An object of the present invention is to provide an execution control device for a data flow computer that selectively executes nodes according to levels.
[0059]
[Means for Solving the Problems]
  The execution control device of the data flow computer of the present invention is
  A node with a numerical value called the environmental level,Express a program with a data flow graph composed of arcs that connect nodes,Token with a numerical value called execution environment levelIn the execution control device of the data flow computer that executes the calculation by updating the value of the value based on the data flow graph,
  Token propagates along the path on the data flow graphDoThe arrival of the token at the input arc is monitored, and a token having the specified execution environment level is received from a node having the environment level equal to or higher than the execution environment level.Node X connected by arcWhen all the input arcs heading toNode X concernedFiring control means for identifying a command associated with
  ConcernedNode XIs determined to be an operation instruction by the firing control means, an operation is performed by inputting the arrived token, the operation result is substituted into a token value, and the token is stored in an environment higher than the execution environment level of the token. Arithmetic instruction execution means for propagating on an output arc toward a node having a level;
  ConcernedNode XIs identified as a control command by the ignition control means,Node XAnd a control instruction execution means for changing the path of the token that has arrived in accordance with the contents of the instruction and propagating the token on an output arc toward a node having the environment level equal to or higher than the execution environment level of the token.
[0060]
[Action]
The present invention is characterized in that an execution control device of a data flow computer has an environment-dependent firing function and an environment-dependent distribution function.
[0061]
The environment-dependent firing function means that when a token having a certain execution environment level m reaches a node n, all input arcs (referred to as environment input arcs) from the node having the environment level of m or higher to the node n are called. This is a function for igniting node n when tokens of execution environment level m are prepared. The environment-dependent distribution function is a function for outputting a token on all output arcs (referred to as environment output arcs) directed to nodes having an environment level equal to or higher than the token execution environment level.
[0062]
The environment-dependent firing function causes a node to wait until all tokens are arranged on all environment input arcs, and fires a node after tokens are arranged on all environment input arcs. The environment-dependent distribution function is a function for outputting a token giving an execution environment level on all environment output arcs. Therefore, when a certain execution environment level m is given to a token and program execution is started, the token is propagated only between nodes having an environment level of m or higher and only nodes having an environment level of m or higher are executed. Can do.
[0063]
【Example】
Next, embodiments of the present invention will be described with reference to the drawings.
[0064]
Let N be the set of nodes. For node n, tag (n) is assigned as the environment level. Further, the execution environment level m is assigned to the token. Tag the environment level of the node connected before the i-th input arc of node n_in(N, i), tag the environmental level of the node connected to the tip of the j-th output arc of node n_out (N, j). At this time, m ≦ tag_inThe arc extending from the node having the environment level satisfying (n, i) to the node n is the environment input arc of the node n, and m ≦ tag from the node n._out An arc extending to a node having an environment level that satisfies (n, j) is an environment output arc.
[0065]
FIG. 1 is a block diagram showing a configuration of a data driven execution control apparatus having an environment dependent firing function and an environment dependent distribution function. Each function is described below.
(Ignition control means 10A)
The input arc selection function 10 selects an input arc having a node having an environment level equal to or higher than the token execution environment level as an environment input arc from the node input arc. The token arrival monitoring function 12 detects and fires a node on which all tokens are arranged on all environment input arcs. The node classification function 13 classifies the fired node into an operation node and a control node, and sends a control signal to the corresponding instruction execution means.
(Calculation instruction execution means 20A)
The input token processing function 21A erases all tokens on the environment input arc of the fired node, and sends the token value to the operation execution function 22. The calculation execution function 22 performs a given calculation on the token value and sends the result to the token output function 23. The token output function 23A substitutes the calculation result for the token value, and places the token given the execution environment level on the environment output arc.
(Control command execution means 30A)
The control instruction decode function 31 identifies the instruction of the control node and sends it to the corresponding instruction execution function. In the instruction execution function, control is performed according to a given node type. There are the following types of control nodes.
[0066]
(Distribution node (D) 32A) 1 input n output node. Delete the token on the input arc, and place the token with the copied value and execution environment level on all environment output arcs.
[0067]
(Branch node (SW) 33) A node with two inputs and two outputs. When the value of the token on the second input arc is true, the token on the first input arc is placed on the first output arc. If false, place the token on the first input arc on the second output arc.
[0068]
(Procedure call node (CALL) 34A) This is a node with two inputs and one output. The token on the second input arc is copied, the procedure start node of the function is obtained from the function name stored as the value of the token on the first input arc, and placed on the first input arc. In addition, a token having a value corresponding to the name of the output arc is generated. The procedure return node of the function is obtained from the function name stored as the value of the token on the first input arc, and the generated token is placed on the first input arc of that node. Erase the token on the input arc.
[0069]
(Procedure start node (ENTRY) 36) This is a node with one input and one output. Place a copy of the token on the input arc on the output arc. Erase the token on the input arc.
[0070]
(Procedure return node (RETURN) 35A) 2 input 0 output node. A token is generated by copying the value of the token on the second input arc and the execution environment level. Place the generated token in the arc designated as the token on the first input arc.
[0071]
FIG. 2 shows an execution example. T1, t2, t3 (t1 <t2 <t3) are used as the environmental level. In the figure, a black circle represents a token, a white circle node represents an operation node, and a white circle and cross node represent a distribution node. A shaded node represents a node that has been executed. The number at the upper left of the node represents the node number, and the number at the upper right represents the environmental level. An arc extending from node x to y is expressed as x → y. The arc numbers are 1, 2, ... from the left.
[0072]
Here, the operation will be described in the case where the token of the execution environment level t2 is propagated. First, paying attention to the node 4, the operation will be described.
[0073]
1. At i = 2, 3, t2 ≦ tag_in(4, i). Therefore, the input arc selection function 11 identifies the arcs 2 → 4, 3 → 4 as environment input arcs.
[0074]
2. When tokens are aligned on these environmental input arcs (FIG. 2 (2)), the node 4 is fired.
[0075]
3. The node classification function 13 identifies the node 4 as an operation node and sends a control signal to the operation instruction execution means 20A.
[0076]
4). Thereafter, the input token processing function 21A removes the token on the input arc.
[0077]
5). The calculation execution function 22 performs calculation.
[0078]
6). Finally, the token output function 23A has t2 ≦ tag_out 4 → 5 from (4, 1) is determined as an environment output arc, and a token of execution environment level t2 is output (FIG. 2 (3)).
[0079]
Next, attention is paid to the node 5.
[0080]
1. The input arc selection function 11 is t2 ≦ tag_inFrom (5, 1), it is determined that arc 4 → 5 is an environmental input arc. Since the tokens are arranged on the environment input arc, the node 5 is fired.
[0081]
2. The node classification function 13 sends a control signal to the control instruction execution means 30A, and the control instruction decode function 31 determines that it is a distribution node.
[0082]
3. At j = 1, 2, t2 ≦ tag_out (5, j) is satisfied. Therefore, the distribution node determines that arcs 5 → 6, 5 → 7 are environment output arcs, and outputs a token of execution environment level t2 on these arcs (FIG. 2 (4)).
[0083]
As described above, the data driven execution control device having the environment dependent firing function and the environment dependent distribution function executes.
[0084]
Example of request-driven execution control device:
Demand propagation is the same as in [Prior Art]. A partial graph composed of the node to which the demand is propagated and its input arc is defined as a data flow graph to be executed. This graph is executed by a data driven execution control device having an environment dependent firing function and an environment dependent distribution function.
[0085]
Application to data-driven execution controller by colored token method:
FIG. 3 is a block diagram illustrating a configuration of a data driven execution control device (colored token method) having an environment dependent firing device and an environment dependent distribution function. Each function is described below.
(Ignition control means 10B)
The firing control means 10B performs the following processing for each node.
[0086]
1. The color detection function 14 detects all colors of tokens on the input arc.
[0087]
2. The color-specific input arc selection function 15 selects an unprocessed color among the colors detected by the color detection function 14. Next, an input arc having a node having an environment level equal to or higher than the execution environment level of the selected token is selected as the environment input arc from the input arcs of the nodes.
[0088]
3. The token arrival monitoring function 12 detects and fires a node having tokens having the same color as the selected color on all environment input arcs. If ignition is not performed and there is an unprocessed color among the colors detected by the color detection function 14, the process returns to 1.
[0089]
4). The node classification function 13 classifies the fired node into an operation node and a control node, and sends a control signal to the corresponding instruction execution means. If there is an unprocessed color among the colors detected by the color detection function 14, the process returns to 1.
(Calculation instruction execution means 20B)
The input token processing function 21B deletes all tokens having the same color on the environment input arc of the fired node, and sends the token value to the operation execution function 22. The operation execution function 22 performs the given operation on the token value and sends the result to the token output function 23B. The token output function 23B substitutes the calculation result for the token value, and places the token given the erased token color and execution environment level on the environment output arc.
(Control command execution means 30B)
The control instruction decode function 31 identifies the instruction of the control node and sends it to the corresponding instruction execution function. In the instruction execution function, control is performed according to a given node type. There are the following types of control nodes.
[0090]
(Distribution node (D) 32B) 1 input n output node. Delete the tokens on the input arc, and place all the tokens whose values, colors, and execution environment levels are copied onto the environment output arc.
[0091]
(Branch node (SW) 33) A node with two inputs and two outputs. When the value of the token on the second input arc is true, the token on the first input arc is placed on the first output arc. If false, place the token on the first input arc on the second output arc.
[0092]
(Procedure call node (CALL) 34B) This is a 2-input 1-output node. When firing, find the color that is not used by any token on the data flow graph. A token is generated based on the token value on the second input arc, the execution environment level, and the newly generated color. The function name is obtained from the token value on the first input arc, and the generated token is placed on the first input arc of the procedure start node. Also, a token is generated based on an execution environment level and a newly generated color, with a pair of the name of the token color on the input arc and the name of the output arc as a value. The procedure return node of the function is obtained from the function name stored as the value of the token on the first input arc, and the generated token is placed on the first input arc of that node. Erase the token on the input arc.
[0093]
(Procedure start node (ENTRY) 36) This is a node with one input and one output. Place a copy of the token on the input arc on the output arc. Erase the token on the input arc.
[0094]
(Procedure return node (RETURN) 35B) This is a node with 2 inputs and 0 outputs. The color before the procedure call is obtained from the token on the first input arc, and the token value on the second input arc and the execution environment level are copied to generate a token. Place the generated token in the arc specified as the value of the token on the first input arc.
[0095]
FIG. 4 shows an execution example. As environmental levels, t1, t2, and t3 (t1 <t2 <t3) are used. The number at the upper left of the node represents the node number, and the number at the upper right represents the environmental level. The black circle token has the color a and is given t1 as the execution environment level. The black square token has the color b and is given t2 as the execution environment level. White circle nodes represent calculation nodes, and white circle cross nodes represent distribution nodes. An arc extending from node x to y is expressed as x → y. The arc numbers are 1, 2, ... from the left. First, the operation will be described by paying attention to the node 4.
[0096]
1. The color detection function 14 detects the presence of colors a and b.
[0097]
2. The color-specific input arc selection function 15 first knows that the execution environment level is t1 from the tag of the token of color a. In node 4, at i = 1, 2, 3, t1 ≦ tag_in(4, i). Accordingly, the color-specific input arc selection function 15 determines that arcs 1 → 4, 2 → 4, 3 → 4 are environment input arcs.
[0098]
3. The token arrival monitoring function 12 does not fire the node 4 yet because the tokens of the color a are not prepared on the environment input arc (FIG. 4 (1)).
[0099]
4). Next, the color-specific input arc selection function 15 knows that the execution environment level is t2 from the tag of the token of color b. At i = 2, 3, t2 ≦ tag_in(4, i). Therefore, the color-specific input arc selection function 15 determines arcs 2 → 4 and 3 → 4 as environment input arcs.
[0100]
5). The token arrival monitoring function 12 fires the node 4 because the tokens of the color b are arranged on the environment input arc (FIG. 4 (1)).
[0101]
6). The node classification function 13 identifies the node 4 as an operation node and sends a control signal to the operation instruction execution means 20B.
[0102]
7). The input token processing function 21B removes the token of color b on the environmental input arc.
[0103]
8). The calculation execution function 22 performs calculation.
[0104]
9. The token output function 23B is t2 ≦ tag_out From (4, 1), it is determined that arc 4 → 5 is an environment output arc, the color of the input token is copied, and a token having execution environment level t2 is output to arc 4 → 5 (FIG. 4 (2)).
[0105]
Next, attention is paid to the node 5.
[0106]
1. The color detection function 14 detects the presence of the color b.
[0107]
2. The color-specific input arc selection function 15 knows that the execution environment level is t2 from the tag of the token of color b. t2 ≦ tag_inSince (5, 1) is satisfied, the color-specific input arc selection function 15 determines that the arc 4 → 5 is an environmental input arc.
[0108]
3. The color-specific input arc selection function 15 fires the node 5 because the tokens of the color b are arranged on the environment input arc (FIG. 4 (2)).
[0109]
4). The node classification function 13 sends a control signal to the control command execution means 30B. The control instruction decode function 31 determines this node as a distribution node.
[0110]
5). At j = 1, t2 ≦ tag_out Since (5, j), the distribution node determines that arcs 5 → 6, 5 → 7 are environmental output arcs. The token having the execution environment level t2 copied by copying the color of the input token is output on these environment output arcs ((3) in FIG. 4).
[0111]
As described above, the data driven execution control device (using the colored token model) having the environment dependent firing function and the environment dependent distribution function executes.
[0112]
Example of request-driven execution control method with colored token model:
Demand propagation is the same as in [Prior Art]. A partial graph composed of the node to which the demand is propagated and its input arc is defined as a data flow graph to be executed. This partial graph is executed by a data driven execution control device (using a colored token model) having an environment dependent firing function and an environment dependent distribution function.
[0113]
【The invention's effect】
As described above, the present invention gives an execution environment level to a token and an environment level to a node, and the execution control device has an environment-dependent firing function and an environment-dependent distribution function. Even if the calculation is performed at the same time shared at the environmental level, there is an effect that only the minimum necessary calculation corresponding to each environmental level can be performed independently of each other.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration example of a data driven execution control apparatus having an environment dependent firing function and an environment dependent distribution function.
FIG. 2 is a diagram illustrating an example of calculation path control using an environmental level.
FIG. 3 is a block diagram showing a configuration example (colored token method) of a data driven execution control device having an environment dependent firing function and an environment dependent distribution function.
FIG. 4 is a diagram illustrating an example of firing control according to an environmental level in a colored token scheme.
FIG. 5 is a diagram illustrating a configuration example of a conventional data drive execution control device.
FIG. 6 is a diagram illustrating an example of data driving execution.
FIG. 7 is a diagram illustrating an example of request drive execution.
FIG. 8 is a block diagram showing a configuration example (colored token) of a conventional data driven execution device.
FIG. 9 is a diagram illustrating an example of firing by a colored token.
[Explanation of symbols]
10A, 10B, 10C, 10D Ignition control means
20A, 20B, 20C, 20D arithmetic instruction execution means
30A, 30B, 30C, 30D Control command execution means
11 Input arc selection function
12 Token arrival monitoring function
13 Node classification function
14 color detection function
15 Input arc selection function for each color
16 Colored token arrival monitoring function
21A, 21B Input token processing function
22 Calculation execution function
23A, 23B, 23C, 23D Token output function
31 Control instruction decode function
32A, 32B, 32C Distribution node (D)
33 Distribution node (SW)
34A, 34B, 34C Procedure call node (CALL)
35A, 35B, 35C, 35D Procedure return node (RETURN)
36 Procedure Start Node (ENTRY)

Claims (1)

A program is represented by a data flow graph composed of nodes assigned with numerical values called environment levels and arcs connecting between the nodes, and the values of tokens assigned numerical values called execution environment levels are expressed in the data flow graph. In the execution control device of the data flow computer that executes the calculation by updating based on
Token monitoring the arrival of the input arcs of the middle tokens propagating along a path on the data flow graph, the token with the specified said execution environment level, with the environmental level above said execution environment level Firing control means for identifying an instruction associated with the node X when all the input arcs from the node toward the node X connected by the arc are aligned;
If the node X is identified as an operation instruction by the firing control means, an incoming token is input to perform an operation, the operation result is substituted into a token value, and the token is set to the execution environment level of the token. Arithmetic instruction execution means for propagating on the output arc toward the node having the above environmental level,
When the node X is identified as a control command by the firing control means, the path of the token that has arrived is changed according to the command content of the node X , and the token is set to the environment level equal to or higher than the execution environment level of the token. Having control command execution means for propagating on the output arc toward the node with
Data flow computer execution control device.

Images