JPH0348952A - Memory control system and matrix, table, file and information processing systems using memory control system - Google Patents

Abstract

PURPOSE:To shorten the processing time and to simplify a program by processing only the necessary areas when the large scale data is processed omitting the processing time required for the unnecessary areas. CONSTITUTION:An access detection request table 51 is added to a page table 14. Then the table 51 is set opposite to the address space of each user program 2 and each entry of the table 51 is opposite to each page of the program 2 respectively. In each entry of the table 51 and a valid/invalid field 17, the requests are received from the programs 2 and a flag is set by a memory control program 1. When an interruption is applied to the program 1 from an address converter 13 due to the entry invalidation, the program starts an access detection interruption process 3 to the program 2. As a result, the page corresponding to the relevant entry is initialized and therefore the initialization processing time, etc., required to the unnecessary areas is omitted. Thus, the processing time is shortened and the program is simplified.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention is useful when processing large-scale data, such as initializing only the area to be accessed or storing only the updated area. Processing unnecessary areas,
The present invention relates to a memory management method capable of reducing processing time by eliminating special determination processing, and matrix processing, table processing, five-file processing, and information processing method using the memory management method. [Prior art] Regarding conventional memory management methods, for example, see "Operating System A (1977)" published by the Japan Computer Association.
6) pp. 149-+91. As is clear from the above literature, in the conventional memory management method,
A memory management program within the operating system manages addresses used by each user using an address conversion device that converts logical addresses into physical addresses. As a result, the Honeycomb program was able to provide an extremely large virtual memory space without being limited by the size of real memory. FIG. 15 is a functional block diagram of a conventional memory management system. l is the memory management program, 2 is the user program, 1
2 is a secondary storage device such as a magnetic disk; 13 is an address conversion device gi that converts a logical address into a physical address; 14
is a page table. The memory management program 1 sets a page table 14 corresponding to the address space of the user programmer 2, which shows the correspondence between logical addresses in the address space and physical addresses in the real memory in units of pages. By referring to and updating this page table 14, the user program address space is managed, and an address space larger than the actual memory size is provided to the user program 2. As shown in FIG. 15, this page table 14 includes a valid/invalid field 17, an update information field 18, and a physical address field 19. The valid/invalid field 17 indicates whether or not the entry is valid for each entry corresponding to each page, and the update information field 18 is set with information indicating whether the page has been updated, and the physical address is set in the update information field 18. Field 19 indicates the physical address on real memory allocated to the corresponding page. Note that here, OX 12000. OX1
The address is expressed in hexadecimal notation such as 5000. When there is an access from the user program 2, the address translation device N13 refers to the page table entry corresponding to the address for each access, and if the valid/invalid field 17 of the entry is valid, the physical address field 19 is The value is converted into a physical address, and the address is accessed. In the case of an update access, an update flag indicating that there has been an update access is set in the update information field 18 and stored. If the valid/invalid field 17 of the corresponding page table entry is invalid, the address translation device fR13 causes a page fault (missing page) interrupt to the memory management program 1. As a result, the memory management program 1 By searching for a free memory area that is not allocated anywhere in the memory 11, storing its physical address in the physical address field 19 of the entry that caused the interrupt in the table 14, and enabling the valid/invalid field 17. , and after allocating the real memory to the corresponding page, the processing of the user program 2 is restarted. In this case, if there is no free memory area that is not allocated anywhere, another page to which real memory in the table 14 is allocated is swapped out to the secondary storage device 2 as needed, and that memory is Assign to page faulted page. (Problem to be solved by the invention) In this way, with conventional memory management methods, user programs are not limited by the memory size and can obtain a large address space. In this case, unnecessary processing may be performed such as initializing all address areas or storing data in all address areas on a secondary storage device after data processing. In addition, in order to avoid processing unnecessary areas, it was necessary to check the address to be accessed each time. Figures 16 and 17 show the initialization process in the conventional memory management method. FIG. 16(a) shows a method of initializing all areas, and FIG. 16(b) shows a method of checking whether initialization has been completed for each area to be processed. In Fig. 17(a), the entire area that has been updated is stored.
In b), the information is stored in each updated area. That is, in the conventional method, the entire area is processed or a check is performed each time the area is processed, which takes an extremely long time in either case. First, in FIG. 16(a), after initializing all areas in advance (step 103Q), data processing is sequentially performed on arbitrary areas (step 102a), and then in FIG.
), data processing is performed multiple times (step 102b).
, 102c, --.), each time a system determines whether the area to be processed from now on has been initialized (steps 104b, 104b, 102c, --.).
4c, ... --), if the initialization has not been completed, first complete the initialization processing (step 103b,
103c. ), when performing data processing or storing updated results, etc., in FIG. 17(a), after performing the updating process (step 1lla), the updated part is of course , memorize all areas including parts that are not updated (
In step 113a), and in ff117 diagram (b),
Each time update processing is performed (steps l1lb, l1lc, .
. . . ), stores the updated area (steps 115b, 115c, . . . ), and finally processes only the updated area based on the stored information (step 113b). . As described above, in the conventional memory management system, it is necessary to check the address to be accessed every time data is processed in order to process unnecessary areas or omit processing in unnecessary areas. As a result, not only does the judgment process take time proportional to the number of times the data is processed, but many judgment processes are inserted in the middle of the program, making it difficult to create a program and making it difficult to determine the flow of processing. There was also. The purpose of the present invention is to solve these conventional problems, and when processing large-scale data, only necessary areas need to be processed, reducing the time for processing unnecessary areas and the processing time for judgment.
An object of the present invention is to provide a memory management method, matrix, table, file processing, and information processing method that can shorten processing time and simplify program creation. [Means for Solving the Problems] In order to achieve the above object, one memory management method of the present invention is as follows:
(b) When a user program issues an access detection request in advance to a memory management program for any area to be accessed in the user program's address space, the memory management program receives the access detection request. The feature is that an area is stored, and when a user program accesses each page in the area after the above request, an interrupt is caused to the user program. (b) When an arbitrary page in the address space of a user program is specified and a request is made from the user program, the memory management program checks whether there is any update access to the page after the request is made. Another feature is that information regarding the presence or absence of update access for each page is stored and sent to the user program when a send request is received from the user program. Further, the information processing method of the present invention is provided in (c) an information processing system including an operating system that manages an address space accessed by a user program. After the user program issues an access detection request to the operating system in advance to a certain area in the address space of the user program. A feature of the present invention is that when a part of the area is accessed, only the accessed part of the area is processed by receiving an interrupt from the operating system as to which part of the area has been accessed. Furthermore, in the file method of the present invention, (d) when performing file processing, the user program requests the operating system to detect access to an area reserved for the whole or a part of the file, and As interrupt processing when an access to an area is detected, a function is registered that converts the file data as necessary and loads it to the corresponding page, and performs transaction processing while converting and loading only the accessed part into memory. Furthermore, before processing the transaction, a request is issued to detect the presence or absence of update access to the area, the transaction results are processed separately into the updated part and the unupdated part, and the original file or another file is processed. It is characterized by the use of a memory management method that stores data in . Further, in the table processing method of the present invention, (e) when performing table processing, the user program receives an interrupt when registering data without initializing the entire area of the table, and The feature is that it uses a memory management method that initializes only a portion of the table required for interrupt processing. Furthermore, in the matrix processing method of the present invention, when the user program performs calculation processing of a large sparse matrix in which only a portion differs from other elements, the user program converts the target matrix into a set of submatrices in page units. By expressing it as A feature of this method is that it uses a memory management method that selectively performs matrix operations on only necessary submatrices. [Operation] In the present invention, for any page in the user program's address space, the user program determines whether or not to cause an interrupt to the user program when an access to that page occurs. Request in advance from (
(Access detection function for initialization) If the requested page is accessed, an interrupt is generated to the user program, and after initialization, the program is restarted. In addition, information on whether any page in the user program's address space has been accessed for update after a certain point is retained, and this information can be cleared from the user program at any time to respond to requests. Accordingly, the information is returned to the user program, or the user program can start a specified process depending on whether there is an update (update access detection function). [Example] Hereinafter, an example of the present invention will be described in detail with reference to the drawings. FIG. 3 is an overall configuration diagram of a memory management system showing an embodiment of the present invention. The memory management system of the present invention includes a central processing unit IO;
It includes a memory 11, a secondary storage device 12, and an address translation device 13. The memory 1 is either stored on the memory chip from the beginning or is stored on the secondary storage device m! an operating system program 5 that manages the entire system by being loaded from the file area 15 of 12 and executed by the central processing unit 10; and a memory management program l that constitutes a part of the operating system program 5. A page table 14 necessary for memory management, an access detection request table 51 related to the present invention,
Update information table 52, pseudo update information table 53, temporary update information table 54, and each user program 2
is stored. The memory management program 1 uses management information such as a page table 14% access detection request table 51 and an update information table 52 necessary for memory management to allocate memory to each user program 2 and manage memory access. , address translation device f! !
13 is a device that converts a logical address in the address space of a user program into a physical address in memory;
By referring to the page table 14 that shows the correspondence between the logical address and the physical address on a page-by-page basis,
The memory management program 1 performs address translation every time the user program 2 accesses memory, and the memory management program 1 uses the function of the address translation device R13 to convert the swap area 16 on the secondary storage device 12 in the same way as before. memory 1
A large-scale virtual memory is realized by using the memory 11 as a destination for expelling pages in the memory 11 when there is no more free memory area in the memory 11. FIGS. 6 and 7 are flowcharts of data processing initialization processing using the access detection function of the present invention and data processing update processing using the update access detection function of the present invention. In the present invention, detection of access to any page can be controlled from a user program. Therefore, as shown in FIG. 6, without initializing the work area in advance, a request is simply issued to detect access to the area (step 101), and if the work area is initialized, If we assume that the data is processed as is (step 10)
2) It is possible to initialize only the pages in the work area accessed by the interrupt request in the form of an interrupt at the time of the first access (step 103). Next, in the present invention, for storage processing after data processing, etc., as shown in FIG. Request to clear. Submit it to the memory management program (step 110).
), and in the data update process (step 111), the process is performed without paying any attention to where in the work area to update or not, and after the process is completed, the updated information of each page in the work area is stored in the memory. Retrieve from management (step 112
), processing such as storage is performed only on the updated page (step +13).Thereby, the user program does not need to perform processing to determine which part of the work area has been accessed during data processing. Furthermore, since initialization and storage processing of unnecessary parts can be omitted, processing time can be shortened. 1 and 2 are explanatory diagrams of the operation of the present invention. Of these, FIG. 1 is a diagram showing an example of an access detection function for initialization processing, etc., and FIG. 2 is a diagram showing an example of an update access detection function for update processing. ff! 1, the contents are new compared to the conventional one in that an access detection request table 51 is provided in addition to the page table I4, and an access detection interrupt function is activated by the memory management program l. The operation of the address translation device 13 and the address translation device [13
The relationship between the page table 14 and the page table 14 is exactly the same as the conventional one, so a description thereof will be omitted. In order to realize the access detection function of the present invention, the access detection request table 51 is converted into a page table! Add to 4. The access detection request table 51 exists corresponding to the address space of each user program 2, and the table 5
] corresponds to each page in the address space of each user program 2. Each entry in the table 51 first receives a request from the user program 2, and
The memory management program 1 sets the 'present' and 'absent' flags. Furthermore, the memory management program 1 rewrites the valid flag in the valid/invalid field on the page table 14 corresponding to 'Yes' on the access detection request table 51 to an invalid flag. As a result, when the address translation device 13 refers to the page table 14 and executes address translation, if the corresponding entry is invalid, an interrupt occurs to the memory management program l. There is. Then, when the address translation device W113 generates an interrupt due to an invalid entry (page fault), the memory management program 1 first activates the access detection interrupt processing 3 for the user program 2, thereby detecting the page corresponding to the entry. Executes initialization. In other words, for an entry set as 'Yes' on the access detection request table 51, when the data processing function group 4 in the user program 2 accesses the page 20 in the corresponding user program address space, the entry is The access detection interrupt processing function 3 in the user program 2 that generates an access detection interrupt and initializes the corresponding page 20 is started, and after the processing is completed, the data processing interval number group 4 is interrupted. Next, the memory management program 1 executes the process of restarting the process.
A detailed explanation will be given of how the access detection request table 51 is actually managed and how the above processing is executed. FIG. 8 is a flowchart of processing when a user program of the present invention requests registration/cancellation of access detection, and FIG. 9 is a flowchart of processing when a page fault occurs by the access detection function of the present invention. In the access detection request registration process of FIG. 8(a), when the memory management program l receives an access detection registration request from the user program 2, it detects access to an entry in the access detection request table 51 corresponding to the requested page. Set to “Yes” to indicate that
0), then set the valid/invalid field 17 of the page table entry corresponding to the requested page to invalid, and when the corresponding page is accessed, the address translation device Fa+3
The page fault interrupt is sent to memory management program 1.
When the data processing function group 4 of the user program 2 accesses the corresponding page 20 in the address space, the access detection request registration process is completed (step 121). In the memory management method of the present invention, the conventional page fault processing is partially modified to
Perform the processing shown in the figure. That is, in the page fault processing shown in FIG. 9, first, as in the conventional case, it is checked whether the faulted page is a page that the user program 2 is allowed to access (step 130), and if the page is not allowed, then , performs error handling (step 136), and if allowed, allocates memory if real memory has not yet been allocated (step +31) and sets the valid/invalid field 1 of the corresponding page table entry.
7 is set to "enabled" (step 132), conventionally,
Although the processing has been completed up to this point, the memory management method of the present invention further executes the following processing, namely, checking the entry in the access detection request table 51 corresponding to the faulted page (step 133). , if ' is present, clear the information of the access detection request table entry (step 134), cause an access detection interrupt to the user program 2,
The interrupt processing function 3 of the user program is started (step 135), and after the function returns, the page fault interrupt processing is ended, and the processing of the interrupted data processing function group 4 is restarted. Further, FIG. 8(b) shows processing in response to a request from the user program 2 to cancel an access detection request for a page for which an access detection request has already been issued. When a release request is issued, the entry in the access detection request table 51 corresponding to the page for which the release request has been made is cleared and set to "no" state (step 122), and the process ends. FIGS. 5(a) and 5(b) are format diagrams of the access detection request table of the present invention. In the embodiment shown in FIG. 1, an access detection request table 51 is used that stores the presence or absence of an access detection request from the user program 2 for each page.
1 may be formatted as shown in FIGS. 5(a) and 5(b). That is, as shown in FIG. 5(a), the area indicates the start address and size, and is defined in a wider area than the page. You can also. Moreover, as shown in FIG. 5(b),
The area may be defined as a start address and an end address. In addition, here, the address is 0x2000...■6
It represents a base number. Next, the update access detection function of the present invention will be explained in detail. FIG. 2 is a diagram showing the operation of the update access detection function of the present invention. In this figure, the difference from the conventional memory management method is that an update information table 52 is provided, and an address conversion device! 213 performs set/reset operations not only on the page table 14 but also on the update information table 52. That is, the address translation device fR1
3, in response to an update access to the address space from the user program 2, update access is performed not only to the update information field 18 of the corresponding entry in the page table 14 but also to the corresponding entry in the update information table 52. Set information indicating that there was. Each entry in this update information table 52 corresponds to each user program 2.
corresponds to each page of the address space. In response to a request from the user program 2, the memory management program 1 clears an arbitrary entry to an ``unupdated'' state, and also serves to read the information and pass it to the user program 2. Address conversion device fi! Step 13 checks all operations in which the user program 2 accesses the address space, and
Upon update access, the entry corresponding to the accessed page in the update information table 52 is set to the ``update'' state. Therefore, the user program 2 issues a request to the memory management program l to clear the necessary entries in the update information table 52 at the necessary time, and
By reading the previously cleared entry in the update information table 52 after executing a certain data process, the update status of each page after an arbitrary point in time can be obtained. By the way, in the above embodiment, not only the update information field 18 of the page table 14 but also the update information field 5
The update access detection function is realized by using address translation device [3] which also sets update access information in [2]. The reason why it is necessary to set update access information separately in this way is that even in the conventional method, the update information field 18 of the page table 14 needs to be used for information for evicting and releasing pages; This is because the update information table 14 must be used for the update access detection function in the present invention. In the embodiment shown in FIG. m2, an address translation device 13 that controls both the page table 14 and the update information table 52 is used. However, in other embodiments, without using a special address translation device 13. It will be explained below that there is also a method of implementing the present invention using the address translation device 5213 which operates in a conventional manner. That is, update field 18 of page table 14
This method implements the update access detection function of the present invention using a conventional address translation device v113 that sets update access information only in one location. FIG. 4 is a configuration diagram of an access detection request table of an update access detection function showing another embodiment of the present invention;
FIG. 0 is a processing flowchart of the update access detection function in FIG. As is clear from FIG. 4, the address translation device 13 is
Since the update access information is set only in the update information field of the page table 14, in this embodiment, instead of the update information table 52. Pseudo update information table 53 and temporary update information table 54
and will be newly established. Like the update information table 52, each entry in each of these tables 53 and 54 corresponds to each page of the address space of each user program. To explain each function of these tables 53 and 54, the information in the update information table 52 in FIG. handle. In addition, the temporary update information table 54 and the page table 14
Each update information field 18 in the ORL corresponds to the information in the update information field in the page table 14 in FIG. FIG. 18 is a diagram showing the operation time chart of the table in FIG. 4 in comparison with the operation chart of the table in FIG. 2. The upper two charts 901 and 902 in this figure
indicates the operation chart of the table in FIG. 2, and the bottom three charts 903, 904, and 905 indicate the operation chart of FIG. 4. M(X, Y) in FIG. 0 is the table corresponding to FIG. Indicates that information on whether there has been update access to the corresponding page between time X and time Y is held. Also, time T indicates the time when memory was allocated to the corresponding page, and T is the time when a request to clear update information for the page was received from program 2, and T is the time when the page was removed from the memory and the memory was released.
Time T is the time when the memory was reallocated to the page, and the lowercase letters represent any time between these times. In order to implement conventional virtual storage, as shown in chart 901, update information for the page after the previous memory allocation is required when evicting the page from memory. In addition, in order to realize a mechanism for returning update information after a certain point in the present invention to the user program, the chart 902
As shown in FIG. 2, it is necessary to hold update information for the page since summer at the time T specified by the user program so that it can be returned whenever requested by the user program. 21st! In the embodiment 1, this is achieved using an address translation device RA13 that records update accesses for each page in both the update information field 18 of the page table 14 and the update information table 52. However, in this embodiment, since a conventional address translation device is used that stores update access information only in the update information field 18 of the page table 14, the 41st! Chart I table 9
03,904.905, the update information field of the conventional page table in FIG. 2 - the update information field of the page table OR pseudo-update It is an information table, and ff! The relationship of update information table in FIG. 2 = update information field of page table OR temporary update information table is established. A method of managing the update information field 18, pseudo update information table 53, and temporary update information table 54 of the page table 14 will be explained with reference to diagram m10. 1O (a) and (b) are processes based on requests from the user program 2, and FIGS. 1O (c) and (d) are processes based on requests from the conventional virtual storage system. First, FIG. 10(a) shows processing when the memory management program l receives a request from the user program 2 to clear update access information for a certain page. This corresponds to the processing at time T in FIG. In the embodiment of FIG. 2, the corresponding update information table 5
However, in the embodiment shown in FIG. 4, the corresponding entry in the pseudo update information table 53 is cleared (step 140), and the update information field of the page table entry corresponding to the requested page is cleared. 18 is ORed with the corresponding temporary update information table 54 entry and stored (step 141), and the update information field 18 of the page table entry is cleared (step 142), thereby updating the page table entry. Only the contents of the temporary update information table that copied the information remain, and all other table contents are cleared. Next, FIG. 1O(b) shows processing when the user program 2 issues a read request for update information of a certain page to the memory management program 1. As mentioned above, the update information table 52 of the fJs2 diagram is
Since the pseudo update information table 53 and the update information field 18 of the page table 14 are in an OR1 relationship, in response to a request to read update information, the entry of the pseudo update information table 53 corresponding to the requested page and the corresponding page OR with the information in the update information field of the table entry
The updated information for the page from time T shown in FIG. 18 to the present time is returned to the user program 2 (step 150). Next, FIG. 51510 (C) shows the real memory release process for each page. This corresponds to the processing at time T in FIG. When memory is released, when memory is reallocated the first time and then released again, it can now be determined whether the corresponding page has been updated between the current memory release and the next memory release. Therefore, it is necessary to clear the update information field 18 of the corresponding page table 14. To round it off, first, the information in the update information field 18 of the page table entry corresponding to the page whose real memory is to be released is
It is ORed with the corresponding entry in the pseudo update information table 53 and stored (step 160). As a result, even if the page table 14 is cleared,
The update information registered there is copied and saved in the pseudo update information table 53.Next, check whether the page has never been swapped out or the corresponding update information field 18 and the corresponding update information. Check both of the entry information in the temporary update information table 54 and check whether either of them indicates that the corresponding page has been updated (step +
61), that is, in the conventional virtual memory system, by checking only the update information field 18 of the page table entry, it is possible to check whether the page has been updated between the previous memory release and the current memory release. In contrast, in this embodiment, both the page table entry and the corresponding entry in the temporary update information table 54 are checked. Thereby, it is possible to check both the update information from before the update access information clear request was made from the user program 2 and the update information that occurred after that. If the determination process in step 161 is true, that is, if the page has never been swapped out, or if the page has been updated since the previous memory release, the page is moved to the swap area of the secondary storage device 12. Step 16
2), Update information field of the corresponding page table entry! 8 and the entries in the temporary update information table 54 are cleared (step 163), and the memory allocated to that page is released (step 164).
), On the other hand, if the determination process (step 161) is false, that is, if the page was swapped out in the previous memory release and the page has not been updated since then, only the memory release is executed (step 164). . Next, FIG. 10(d) shows processing when real memory is allocated to a certain page. This corresponds to the processing at time T or T in FIG. In this case, the update information field of the corresponding page table entry is exactly the same as the conventional method! All you need to do is clear step 8 (step +70). By performing the processing shown in FIG. 10, update access according to the present invention can be achieved using a conventional address translation device that sets update information access information for a page only in the update information field 18 of a page table entry. Both detection functions can be realized. Next, another embodiment of the update information reading method of the present invention will be described. In the embodiment shown in FIG. 2, the memory management program 1 reads the update information of each page stored in the update information table 52 in response to a request from the user program 2, and returns it to the user program 2. Instead of returning update information for each page, the memory management program l that has received a request from the user program 2 refers to the update information table 52 and determines whether each page has been updated or not. Each location Fl depending on whether it is updated or not updated! For example, upon request from user program 2, memory '1' may be activated.
The 7F11 program 1 can start the program 8 function when there is an update access to the page, and start the program B function when there is no update access. Next, a modification of the embodiment of the present invention will be described. In the embodiments shown in FIGS. 1, 2, and 4, the access detection request table 51, update information table 52, pseudo update information table 53, and temporary update information table 54 are provided as independent tables. , instead of doing so, each entry may be one field of the page table 14, that is, the page table 14 includes, in addition to the valid/invalid field 17, the update information field 18, and the physical address 19, the access detection request. field and an update information field (or a pseudo update information field and a temporary update information field). Figures fjitt to Figure 14 show examples of how the present invention works, and Figure 11 shows one 51! FIG. 2 is a conceptual diagram of file processing according to Akira. Now, as shown in FIG. 11, consider the case where a file 15a stored in the file area 15 in the secondary storage device 12 is to be updated. First, a work area m is defined on the address space of the user program corresponding to the size of the file 15a (shaded area), an access detection request is issued to the memory management program 1 for this area, and As the access interrupt processing, a process (+03d) is registered that reads the part corresponding to the accessed page from the file 15a, performs conversion if necessary, and then loads it into the corresponding page. When the area m of the part is accessed, a reservation is made to perform necessary conversion on the data of the file 15a and then load it into that area. Then, if you execute data processing (102d), there will still be file 15.
When a page in which the data of a is not loaded is accessed, an interrupt occurs, and only the accessed page can be converted and loaded into the corresponding area. Furthermore, before data processing, the update access information p of all pages in the work area is cleared, and after the access interrupt processing is completed, by clearing the update access information p of the page that caused the access M interrupt, After data processing, processing (180) including storage in a separate file 15b depending on whether or not each page in this work area has been updated can be performed. That is, after data processing, the update access information is read out and only the updated part p is selectively processed, or another process is performed depending on whether there is an update, or, for example, only the updated part is processed. File 15a
You can store it while converting it back to , or save it in another file 1
5b. FIG. 12 is a conceptual diagram of a name management table management method using the present invention. As a method of use other than that shown in FIG. 11, it can also be used for initialization processing of a name management table used by a compiler or the like. That is, as shown in FIG. 12, Oη performs the name management table reference/registration process (+02e).
Then, an access detection request is issued to the table area, an access detection interrupt is generated when the table area is accessed, and only the necessary part is initialized by this interrupt (103e). Note that n is an uninitialized area. , q is the table month area. FIG. 13 is an explanatory diagram of a method of allocating address space when performing matrix operations using the present invention, and FIG. 14 is a flowchart of the matrix operation processing method of FIG. 13. Another example of use is for sparse large-scale matrix operations. As an example of matrix multiplication, first divide the two matrices A and B to be operated on, and the storage location C after the operation, into square submatrices in page units and store them in the address space, as shown in Figure 13. Place it in Although FIG. 13 shows only matrix A, the same is true for matrix B and C. Access detection requests are issued to these areas in the 0th order, and the access interrupt processing function is as follows. A function that clears the corresponding page to 0! IQ (step 190), and also clear all update access information for pages in these areas (step 19, that is, create a 6th order 2 matrix A starting from a state with no update information, Insert non-zero elements into B (step 1
92), non-O elements are significant digits other than zero. After that, update access information of pages corresponding to matrices A and B is obtained from the memory management program (step 193
), and their updated access information aiJ+ b i
When J, from the following equation, only when al) lbkJ is not O, submatrix operation A111BkJ is performed to calculate C1J. Then, only when the calculated C1J is not 0, it is actually written into the C1J area (step 194). Thereby, the calculation result can be obtained by reading the CIJ. Furthermore, this C1J is transferred to A, where the non-O element writing process (step 194) has been completed
It is also possible to consider it as B and repeat the calculations from step 193 onwards. As described above, in this embodiment, the user program can detect access to any page in the address space in the form of an interrupt, and can also detect that each page is updated after a certain point. You can know whether it is true or not. Therefore, by using this for various processing. The user program does not need to perform any special determination processing, and only needs to process the necessary areas, reducing processing time and facilitating program creation. [Effects of the Invention] As explained above, according to the present invention, when performing large-scale data processing, only the necessary area can be initialized without initializing the entire work area, and the data Even if you do not have to remember all the updated locations during processing, you can obtain update information for each page in the work area from the memory management program after processing, so you can perform processes such as storing only updated pages in a file. It is possible to do this.

[Brief explanation of drawings]

FIG. 1 is a functional block diagram of a page access detection section showing an embodiment of the present invention, FIG. 2 is a functional block diagram of a page update access detection section showing an embodiment of the present invention, and FIG. 3 is a functional block diagram of a page access detection section showing an embodiment of the present invention. Overall configuration diagram of the applied information processing system, Part 4
The figure is a functional block diagram of the main part of the page update access detection unit showing another embodiment of the present invention, Figure 5 is another configuration diagram of the access detection request table of Figure 1, and Figure f56 is the access detection function of the present invention. Flowchart of initialization processing using . FIG. 7 is a flowchart of the update process using the update access detection function of the present invention, FIG. 8 is a flowchart when requesting registration/cancellation of access detection from a user program in the present invention, and FIG. 9 is a flowchart of the access detection of the present invention. 10 is a flowchart of the function when a page fault occurs. FIG. 10 is a processing flowchart of the update access detection function of FIG. 4. FIG. 11 is a conceptual diagram of file processing showing an application example of the present invention.
The figure is a processing conceptual diagram of a name management table management method showing an applied example of the present invention, and Fig. 13 is an explanatory diagram of an address space allocation method for matrix operations showing an applied example of the present invention.
l is a processing flowchart of matrix operations in m13rI4, Fig. 15 is a functional block diagram showing a conventional virtual memory management method, and Figs. 16 and tjS17 are initialization processing and update processing using the conventional memory management method. FIG. 18 is a time chart for starting each table in FIGS. 2 and 4. 1: Memory management program S2 user program, 3
: access detection interrupt processing function, 4: data processing function group, 5: operating system program, lO:
central processing unit, ll: memory, 12: secondary storage device, 1
3 Near address translation device, 14: Page table, I5:
File area, 16: Swap area, 17 2-page table valid/invalid field, 18 2-page table update information field, 19: Physical address field of page table, 20: Access detection target page, 5
1: Access detection request table, 52: Update information table, 53: Pseudo update information table, 54 Temporary update information table. Memory Fig. 5 (a) Fig. 8 (a) Fig. 8 (Part 2) Fig. 1 (Part 3) (d) When real memory is allocated Fig. 1 (aJ Conventional method 1 (Conventional method) 2

Claims (1)

[Scope of Claims] 1. In an information processing system equipped with a memory management program that divides and manages an address space accessed by a user program in page units, a user program can access any part of the address space of the user program. When a request to detect access to an area is sent to a memory management program in advance, the memory management program stores the area for which the access detection request was made, and after the request, the user program A memory management method characterized by causing an interrupt to the user program when a page is accessed. 2. In an information processing system equipped with an operating system that manages an address space accessed by a user program, the user program issues a request to the operating system to detect access to a certain area in the address space of the user program in advance. After that, when a part of the area is accessed, only the accessed part of the area is processed by receiving an interrupt from the operating system as to which part has been accessed. 3. In an information processing system equipped with a memory management program that manages the address space accessed by a user program by dividing it into page units, specify any page in the address space of the user program and access the address space from the user program. When a request is made, the memory management program memorizes whether there has been an update access to the page since the time of the request, and when there is a send request from the user program, it stores information regarding whether there is an update access for each page. A memory management method characterized by sending information to the user program. 4. Instead of sending the information regarding the presence or absence of update access for each page to the user program, when there is a request from the user program to send information regarding the presence or absence of update access for each page, the memory management program transmits the information to the user program. 4. The memory management system according to claim 3, wherein the process specified by the user program is activated in response to whether or not there is an update specified by the user program. 5. In the memory management system according to claim 1, the memory management program provides an access detection request table corresponding to each page of the address space of each user program, and receives access detection requests for pages specified by each user program. When the address translation device receives the request, it registers the presence or absence of the detection request in the corresponding entry of the access detection request table, and changes the corresponding page table entry on the page table referenced by the address translation device.
A page fault interrupt is generated when the page is accessed, and if the page is accessed and the real memory is unallocated when the page fault occurs, real memory is allocated and then the above access is performed. A memory management method characterized in that, by referring to a corresponding entry in a detection request table, when an access detection request presence is registered for the page, a program interrupt is caused to a user program that has issued a detection request. 6. In the memory management method according to claim 3, the memory management program provides an update information table for recording update access for each page in the address space of each user program, A memory management method characterized in that any entry in the update information table is cleared at any time upon request from each user program. 7. In an information processing system equipped with an address translation device that sets and registers the fact that the page has been updated in an area corresponding to the page when even a part of the page is updated,
The address translation device is characterized in that it simultaneously updates and registers two tables: a page table and an update information table that stores the presence or absence of update access to a page in the address space of a user program. A memory management method that uses 8. In the memory management system according to claim 3, when using an address translation device that registers only one piece of update information in the corresponding page table entry for update access to each page, in addition to the page table, A pseudo update information table and a temporary update information table are provided corresponding to the address space of each user program, and update information on the page table is provided when real memory is allocated to the corresponding page and when the page is swapped out. , and also clears the above pseudo update information table and temporary update information table first. When receiving an update access detection request for a certain page from the user program, clears the corresponding entry in the above update information table and at the same time clears the temporary update information table. After ORing the corresponding update information in the page table with the corresponding entry in the information table, clear the update information in the page table, and if you want to allocate the real memory allocated to a certain page to another program, first The update information of the page table entry corresponding to the page is ORed with the corresponding entry of the above pseudo update information table, and the update information of the page table entry and the corresponding entry of the temporary update information table are checked, and one of them is set. If the page has been swapped out, or if the page has never been swapped out, after swapping out the page, clear the corresponding update information in the page table and temporary update information table to free up memory. When there is an update information retrieval request from a user program, it refers to the update information in the page table corresponding to the user program and the part of the pseudo update information table corresponding to the page requested by the user program, and retrieves both. A memory management method characterized by ORing information and sending it to the user program. 9. In the memory management method according to claim 5, the access detection request table is held in the form of area information for which an access detection request has been made, instead of storing the presence or absence of an access detection request for each page. Characteristic memory management method. 10. When performing file processing, the user program requests the operating system to detect access to an area reserved for the whole or part of the file, and performs interrupt processing when detecting access to the area. , register a function that converts the file data as necessary and loads it to the corresponding page, performs transaction processing while converting and loading only the accessed part into memory, and before performing the transaction processing, 4. A method according to claim 1 or 3, wherein a request is issued to detect the presence or absence of an update access to an area, and the transaction result is processed separately into an updated part and a non-updated part and stored in the original file or another file. A file processing method characterized by using a memory management method. 11. When performing table processing, the user program receives an interrupt when registering data without first initializing the entire area of the table, and processes only the part of the table necessary for processing the interrupt. A table processing method characterized in that the memory management method according to claim 2 is used for initialization. 12. When performing a large sparse matrix calculation process in which only one part is different from other elements, the user program expresses the target matrix in the form of a set of submatrices in page units, and first initializes the entire area. By writing only the different elements, you can initialize only the necessary submatrices, and when performing operations between matrices, only the updated submatrices can be detected and only the necessary submatrices can be selectively selected. A matrix processing method characterized by using the memory management method according to claim 2 or 3, which performs matrix operations.