JPS59158459A - Structure memory recovery system - Google Patents

Abstract

PURPOSE:To realize a structure again in high speed by moving a data section corresponding to a header section in use at present to a continuous area of another heap memory and using the other heap memory as a data accumulating area of the structure so as to recover the memory used already. CONSTITUTION:A header memory 110 accumulates the header section of the structure and heap memories 120, 121 accumulate a data section of the structure to a continuous area. Suppose that the continuous area accumnlating the data section of a new structure is not ensured when one of the two memories 120 and 121 is used as the data section accumulating area of the structure. In this case, a strcture memory managing unit 130 moves a data section corresponding to the header section in use at present in the header section of the structure accumulated in the memory 110 to the continuous area of the other heap memory and then, the other heap memory is used as the data section accumulating area of the structure.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to the implementation of a prolog structure and a memory cell recovery method.

In recent years, as interest in intelligent systems has increased, predicate logic language prologs have been attracting attention. One of the reasons for this is that a structure composed of a variable number of elements can be scaled.

For example, in the prolog, the state in which the basket contains ``apples'', ``no l'', and ``oranges'' can be expressed as follows using a structure with 3 elements per basket.

Basket (apples, pears, oranges) Furthermore, the fact that Abe has this basket can be described as follows using a structure with a structure ratio of 'has' and two elements.

Hold (fat, basket (apples, pears, oranges)).

In prolog, data can be easily described by using a structure with a variable number of elements.

On the other hand, when a prolog is executed, a large number of structures are dynamically created and destroyed, so a large amount of memory cells are required as an area for the structures. Therefore, in order to increase the usage efficiency of memory cells, we carry out so-called garbage reflection (abbreviated as GC), which collects used memory cells and uses them again to realize the structure. be exposed. However, the processing cost required for garpage collection is large, and may take 30 to 50 parts or more of the total burial cost.The present invention aims to realize and structure a structure that can process such GC at high speed. This provides a method for recovering body memory.

Before describing embodiments of the present invention, a conventional method for realizing a structure and its problems will be discussed.

One of the well-known implementation methods for realizing a structure is a method of connecting a decimal number of fixed-length areas with pointers.

FIG. 1 shows a conceptual diagram when expressing a structure (fat section, basket (apple, pear, orange)) using a binary tree list cell 10. The greatest feature of this method is that the unit for securing memory cells is a fixed length, so there is no need for continuous unused areas. That is,
It is only necessary to connect fixed areas (two consecutive memory cells in a binary tree list) with a pointer, which does not require the work of collecting used memory cells by garbage collection.

However, the method of connecting fixed-length areas with pointers has the problem of large overhead when accessing the elements of the structure. For example, to access the nth element of a structure, the pointer must be traced n times,
This causes memory access bottlenecks.

As another method for solving the drawbacks of the above method, a method is known in which the elements of the structure are arranged in a continuous area and each element is accessed by offset from the header section. Figure 2 is a diagram representing the structure of this method (fat section, basket (apple, pear, tangerine)). In Figure 2, 20 is '! 44 structure name area, 2nd element number area, 22
23 represents a structure header area, and 23 represents a data area.

In this method, the structure has a horizontal body header area 22 consisting of a structure subject and the number of elements, and a data area 23 for storing each element of the structure, which are allocated as continuous areas. Therefore, when including a structure as an element, the pointer to the structure header area 22 may be used as the element. Furthermore, elements can be accessed at high speed by using offsets from the structure header area 22.

However, this method has a problem in that the overhead of garbage flexing is large. That is,
This method requires structures to be stored in continuous areas of variable length.

Further, due to the property of the structure, there is a possibility that one structure may be shared by a plurality of structures.

In Figure 3 (1) and (2), structure A and structure C are connected to structure B.
FIG. 3 is a conceptual diagram showing the transition of a heap area due to garbage collection when sharing a heap area. In Figure 3,
30 is heap memory, 31 is the area of the structure in use, 3
2 indicates the area of the used structure. Also, Figure 3 (1)
represents the state of the heap memory before performing garbage collection, and FIG. 3(2) represents the state of heap memory after performing garbage collection. Garbage collection according to this method is realized as follows. First, mark the structure currently being used for squid. This is accomplished by marking structures that are directly and indirectly referenced from the stack in use by the following algorithm.

(1) For all structures pointed to from the stack (2
).

(2) If the body is already marked as in use, do nothing; if not, mark the structure as in use, and if the structure contains further structures as elements, y
Repeat the process (2) for all elementary structures.

Structures that are not marked as in use by the above algorithm can be reclaimed as used space.

However, simply connecting used areas with a pointer as in the structure implementation method using fixed areas does not ensure a continuous area as the area for the next new structure. Therefore, in order to secure an unused area on a continuous area, as shown in FIGS. I have to collect 32.

The problem with this method is that when relocating a structure, all pointers pointing to the relocated structure must be rewritten.

The purpose of the present invention is to solve the problems of the conventional method described above, and to realize a structure in which access to elements is fast and the overhead of one casion of the structure at the time of gouvage flexion is minimized. K provides a structure memory recovery method that makes it possible.

The above purpose is to store the header part of the structure in a structure consisting of a fixed length header part which becomes the IJ of the structure and a variable length data part which stores each element of the structure. When using a header memory, two heap memories that multiply the data part of the structure into a continuous area, and one of the two heap memories as the storage area for the data part of the structure. When it becomes impossible to secure a continuous area to store the data part of a new structure,
Among the header parts of the structure stored in the header memory, the data part corresponding to the header part currently in use is moved to a continuous area of the other heap memory, and from then on, the data part of the structure is stored in the other heap memory. This is achieved by a structure memory recovery method that includes a structure memory management unit that performs the operation of using the memory as an area.

Next, specific examples of the present invention will be described.

FIG. 4 is a block diagram of a structure memory recovery device which is one of the embodiments of the present invention. In the diagram, 110 is a memo! +, 111 is a mark pit area, 112 is a structural body temperature area, 113 is an element number area, and 114 is a pointer area to a data area in which elements are stored. Also, 120,1
21 is a heap memory having the same logical address space, 122 is a data area for structures in use, 123 is a data area for used structures, and 124 is an unused area; these 122 to 124 are heap memories. The memory of either memory 120 or 121 that is currently in use will be replaced.

Furthermore, 130 is a structure memory management unit;
31 is a heap memory switch that indicates the heap memory currently in use; 132 is a start address register that indicates the start address of the unused area 124; and 133 is the unused area 124.
is the final address register of . 150 is a stack memory that stores primary information used to execute the prologue. The structure memory management unit 130 includes the header memory 110 . Said heap memory 120°12
1 and has the ability to instruct the marking of the structure currently in use from the pointer to the structure header stored in the stack memory 150, and 140 represents the previous f structure memory. Management unit 130
．． It represents an internal bus that couples the stack memory 150 and an external processing unit (not shown).

In this example, 1 bit is prepared in the mark pit area 111, 11 bits in the structural body temperature area 112, 8 bits in the number of elements area 113, and 16 bits in the pointer area 114 (9). Further, the header memory 110 has 9 KB, and the heap memories 120 and 121 each have 128 KB out of 20 bits including the tag bit out of 4 bits.

Therefore, in this embodiment, 2000 structures each having a maximum of 256 elements can be handled.

In addition, the structure memory management unit 130 may include, for example, a commercially available 16-bit microprocessor and a 4KB high-speed memory I
It is composed of C. Further, the structure memory management unit 130 includes a 1-bit flip-flop as a heap memory switch 131, and an optical end address register 132. 16 as the terminal address register 133
It has two bit-width registers. The stack memory 150 is composed of 32 KB of 16 bit wide memory.

According to the present invention, all a bodies are accessed via the structure header, so there is at most one pointer pointing to the data area of the structure. Therefore, even if the data part is relocated, rewriting the pointer only needs to change the pointer part of the structure header (10), which can greatly reduce the overhead of the structure header. In addition, data area partitions can be processed at high speed using two heaps, and two heaps 1i'i
Since V are placed in the same address space, the bit width of the pointer part of the structure header only needs to be enough to point to one heap file.

Next, the structure memory recovery method in this embodiment will be explained with reference to the drawings.

FIG. 5 is a conceptual diagram illustrating the structure memory recovery method in this embodiment. However, in this embodiment, a header in which the mark bit order 111 is '1' indicates that the header is in use, and when the heap memory switch 131 is /O/, the heap memory switch 131 indicates that the heap memory 120 is in use. When 131 is '1', it indicates that heap memory 121 is in use.

When no structure is defined at all, the value of the mark pit area 111 of all (11) structure headers is 70', and the heap memory switch 131 is set to '0', and the tip address register 132 is set to the tip of the heap memory 120. The end address register 133 is the heap memory 12
It points to the end address of 0.

When the structure memory management unit 130 is requested to create a new structure from an external processing unit (for example, a prolog processor, not shown), the structure memory management unit 130 operates as follows.

For example, consider the case where structures (apples, pears, oranges) are placed on the heap. First, add the number of elements 3 to the value of the end address register 132, and determine whether the value exceeds the value of the end address register 133. If it does not exceed the mark bit 111 from the header memory 110.
Select one header area where is '0', assign a value representing the structure I basket to the structure header 112 of the header area, assign the number of elements 3 to the element number area 113, and set the current value to the pointer area 114. Assigning address 02) pointed to by the tip address register 132 and incrementing the address of the tip address register 132 by 1, the $ afforestation element 'apple',
The values representing 'None' and 'Orange' are stored in the heap memory 120.
and returns the address of the header area to the external processing unit as the address of the structure.

As the value of the structure, the value of the structure header is returned as the address, so it has 6 structures (fat, basket, apple,
A structure that includes a structure as an element, such as None, Orange))', stores a pointer to the structure header in the data area, as shown in FIG.

In Figure 5, the structure basket (Hanako, basket (apple, pear 1.I/can)) shares the structure basket (apple, pear, orange), and the stack memory 150 has a structure (thickness). Part, basket (apple, pear, orange)) and structure (13) Represents a state in which only the holding (hanako, basket (apple, pear, orange)) is pointed.

In this state, furthermore, the structure Human (Abe, Hanako) Add.

Similarly, the structure memory management unit 130 compares the value of the end address register 132 plus the number of elements 2 and the value of the end address register 133, but if the value of the end address register 133 is smaller. If so, the garbage collection process described below will be $+! III
J. When the garbage flexion is activated, all the values of mark bits 111 are first set to /Q/. Next, each structure header placed in the stack memory 150 is marked as an in-use structure header according to the following algorithm.

(1) After setting the mark bit 111 to /1', search for the number of elements from the address pointed to by the pointer 114,
Perform the operation (2) on the element whose value is the structure header.

(14) (2) If the value of the mark bit 111 of the corresponding structure header is '1', the process of (1) is repeated for the structure header.If it is '0', the process (1) is repeated for that structure header.

As a result, in the header memory IIO, only the mark pit area 111 of the structure header currently in use is set to ''.

Next, the value of the leading address register 132 is set to the leading address of the heap memory 121, and the data area 122 in use is relocated according to the following algorithm.

(1) For each structure header in the header memory 110, if the value of the mark pit area is '1', process (2). Once all structure headers have been processed, process (3) is performed.

(2) The contents of the heap area 120 from the address pointed to by the pointer area 114 of the structure header to the address plus the value of the element value area 110 are transferred from the address pointed to by the tip address register 132 to a continuous area of the heap area 121. Then, the candy of the number of elements branch 110 is added to the value of the tip address register 132 (15).

(3) Set the value of the heap memory switch 131 to '1' to indicate that the heap memory 121 is being used as a data storage area for the structure.

FIG. 6 is a conceptual diagram showing the state after relocation is performed using the algorithm. Since the unused data area is compressed by relocation, the unused area 124 can be obtained as a continuous area. Also, in the header memory 110, the value of the mark pit area of the used structure header is returned to 'O', so that the header area can be reused.

After that, just add the structure as you normally would. However, if the necessary data area and header area cannot be secured even after garbage collection, an error message is output and processing is interrupted.

Furthermore, when the heap memory 121 runs out of space, the data area in use on the heap memory 121 is similarly relocated to the heap memory 120 (16), and the heap memory switch 131 is set to '0'. Burp.

[Brief explanation of drawings]

Figure 1 is a conceptual diagram showing an example of realizing a structure using a fixed @bin, Figure 2 is a conceptual diagram showing an example of realizing a structure using a continuous region of considerable length, Figure 3 is a conceptual diagram showing an example of garbage collection when a structure is realized using a continuous area of variable length, and Figure 4 is an example of the structure memory collection device according to the present invention. This is a diagram. Further, FIGS. 5 and 6 are conceptual diagrams explaining the Garpage correction method of the present invention. In the figure, 20 is a structure person q! 21 is the element value area, 22 is the header area of the structure, 23 is the data area of the structure, 30 is the heap memory, 31 is the area of the structure in use, and 32 is the area of the used structure. In addition, 110 is a header memory of a structure processing device which is an embodiment of the present invention, 111 is a mark bit 411, (17
) 112 is the structure water region, 113 is the 'g prime number region, 114
Pointer area to the data area, 120 is the first heap memory, 121 is the second heap memory, 122 is the data area in use, 123 is the used data area, 124 is the unused wavelet, 130 is the structure memory management unit, 131
represents the heap memory switch, 132 is the tip address register, 133 is the end address register, 140 is the internal data bus, and 150 is the stack memory. Year Patent Application No. 0330
No. 65 No. 2, Name of the invention Structure memory recovery method 3, Relationship with the amended person's case Applicant 5-33-1 Shiba, Iwa-ku, Tokyo (423) NEC Corporation Representative Tadahiro Sekimoto 4; Agent address: Sumitomo Sanda Building, 37-8 Shiba 5-chome, Minato-ku, Tokyo 108 (Contact address: NEC Corporation Patent Department) 5. Scope of claims of the specification to be amended, column 6, Contents of the amendment Correct the range as shown in the attached sheet 〇 (Attachment) Claims Structure En) For a structure consisting of a fixed-length header section that becomes IJ and a variable-length data section that stores each element of the structure. , a header memory that stores the header part of the structure, and 2 that stores the data part of the structure in a continuous area.
When one heap memory and one of the two heap memories are used as the storage area for the data part of a structure, it becomes impossible to secure a continuous area to store the data part of a new structure. Then, among the header parts of the structure stored in the header memory, the data part corresponding to the header part currently in use is moved to a contiguous area of the other heap memory, and from then on, the data of the structure is transferred to the other heap memory. A structure memory recovery method characterized by comprising a structure memory management unit that operates to use the structure memory as a partial storage area. -・Complete 1. Pa315-

Claims (1)

[Claims] For a structure consisting of a fixed-length header part that is an entry of the structure and a variable-length data part that stores each element of the structure, a header memory that stores the header part of the structure, and a header memory that stores the header part of the structure, and a When two heap memories are used to store the data part of a structure in a continuous area, and one of the two heap memories is used as the storage area for the data part of a structure, the data of the new structure is When a contiguous area for storing the data is secured, the data part corresponding to the header part currently in use among the header parts of the structure stored in the header memory is moved to a contiguous area in the other heap memory, and from then on, A structure memory recovery method characterized by comprising a structure memory management unit that operates to use the other heap memory as a data storage area of a structure.