JPH10124392A - Method for optimizing cache competition miss of fixed area - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce competition misses between addresses in a fixed area by preparing the information of rspective using physical pages, the information of respective groups and secondary cache page information and executing optimization. SOLUTION: The information 25 of respective using physical pages, the information 26 of respective groups and secondary cache page information 27 are prepared by using access state data 11 and system constitution data 12 as input information to execute optimization. A dummy area reducing pages as less as possible is inserted into a position immediately before or after a group as trial, the number of competition misses generated due to the insertion result is calculated and an inserting position and the number of pages to be inserted are found out so that the number of competition misses is less than a target value. When the allocation of physical addresses in the fixed area is changed at the time of system generation in accordance with the calculated results, cache competition misses between fixed areas can be minimized. When the number of competition misses arrives at a fixed value, optimization is stopped.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optimization method for minimizing contention errors in a cache memory (hereinafter abbreviated as cache) in a fixed area of a physical addressing cache memory of a central processing unit (hereinafter abbreviated as CPU). Especially electronic calculators, word processors, electronic exchanges,
The present invention relates to a method for optimizing a fixed area cache conflict miss of a CPU in an information processing apparatus such as a workstation.

[0002]

2. Description of the Related Art Generally, a CPU in an information processing apparatus has a high-speed cache, and the cache can use the following two types of space.

(1) Fixed area: An area in which the relationship between a logical address and a physical address is fixed in hardware.

(2) Floating area: The relationship between a logical address and a physical address is determined by an OS (operating system) and is an area that is floating in hardware.

[0005] On the other hand, a cache usually has a size of several tens of kilobytes (hereinafter abbreviated as KB) to several megabytes (hereinafter abbreviated as MB), and includes 16 bytes (hereinafter abbreviated as B) to several hundreds. Some caches have many rows of size B and are accessed by using a portion of the physical address to determine the number of the row in the physical addressing cache. This cache is called a physical addressing cache. At this time,
The size of the address space is usually several tens MB or more,
The size of the cache is much smaller,
Even if the addresses are different, the same row may be accessed. In this case, the content of the line is different from the intended one and cannot be used. Therefore, it is necessary to reload the content from a memory or the like. This is called a cache contention miss, and is a major factor in lowering CPU performance.

As a method of minimizing the contention error, in the floating area, the same row is prevented from being concentratedly used by changing the relationship between the logical address and the physical address (other than the unused area). Various methods have been developed to minimize contention mistakes. However, for the fixed area, since there is a characteristic that the relationship between the logical address and the physical address cannot be changed, conflict errors do not occur between the addresses in the fixed area, or even if they occur,
As a result, effective optimization has not been realized.

[0007]

However, when the actual use condition of the fixed area is analyzed by trace data or the like, a conflict mistake may occur between the addresses in the fixed area, and the performance degradation due to the conflict mistake may occur. But 10
There are examples that reach ~ 40%.

In the present invention, although the fixed area has the characteristic that the relationship between the logical address and the physical address cannot be changed, the fixed area can be changed by using the "means for solving the problem" described below. And an optimization method for minimizing contention mistakes between addresses within the same.

[0009]

According to the present invention, the following three general properties in a fixed area of an information processing apparatus such as an electronic computer, a word processor, an electronic exchange, and a workstation are utilized. (Property 1) In the fixed area, the relationship between the physical address and the logical address is fixed in hardware. Therefore, the only way to reduce contention errors between a plurality of addresses in the fixed area is to move the logical address (both the physical address). (Property 2) In a system having a fixed area, generally,
The fixed area is a kernel area of the OS (an area of the basic OS). There are numerous programs and data for the functions of the kernel, the sum of which usually amounts to 4-16 MB. The portion of them used during online services is generally only a small part. (Property 3) Since the kernel area includes a large number of functions, between each program, a repair area usually called a patch area and a large number of dummy areas for future expansion are included. In general, taking these factors into account, the logical addresses of the kernel area are specified collectively when the system is generated. Therefore, although the relationship between the logical address and the physical address cannot be changed in units of one page like a floating area, moving the logical address space (and physical address space) by several pages in a certain unit is Generally easy.

In the optimization method of the present invention, the unit is called a "group", and the minimum unit of the size of the group is the size of the physical addressing cache in order to avoid a contention mistake. Immediately before or after the group, insert a dummy area of as few pages as possible as a trial, calculate the number of conflict mistakes generated from the inserted result, and make sure that the number of conflict mistakes is less than the target value. Find the insertion position and the number of pages to be inserted. Then, according to those results, the assignment of the logical address of the fixed area is changed at the time of system generation, thereby minimizing cache conflict miss between the fixed areas. In these calculations, information on used logical addresses and physical addresses obtained from trace data in an actual system is used.

According to the present invention, as described above, physical addresses (and corresponding logical addresses) are moved in units of groups, and therefore, even if the relationship between logical addresses and physical addresses cannot be changed, It is possible to find the logical address at the time of system generation that minimizes the occurrence of contention errors, and it is possible to greatly reduce contention errors between fixed areas, which had been abandoned in the past, and greatly improve system performance. is there.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows an outline of input / output and a block diagram of a method for optimizing a cache contention miss in a fixed area of a physical addressing cache (hereinafter referred to as a secondary cache) according to an embodiment of the present invention. The figure is shown. The figure shows the relationship between the input information 1, the optimization program, the optimization data 2, and the output information 3, and the information on each used physical page which is each block in the optimization program and the optimization data 2. 21, information processing 22 of each group, secondary cache page information processing 23, optimization processing 24, information 25 of each used physical page, information 26 of each group, and secondary cache page information 27. Show the relationship.

First, input information 1 is access state data 1
1 and system configuration data 12. The access state data 11 is a file storing information used in the optimization method of the present invention, out of information created by tracing or emulation of the entire system. Many are stored in the order given. The information of the system configuration data 12 is the system configuration data used by the optimization method of the present invention in the system to be optimized, and is a file storing the information shown in FIG.

The optimizing program and the optimizing data 2 are used for each physical page information processing 21 and each group of information processing 22.
, Secondary cache page information processing 23, optimization processing 24, information 25 of each used physical page, and information 2 of each group.
6 and each block of the secondary cache page information 27.

When each used physical page information processing 21 is activated, each used physical page information processing 21 first inputs the system configuration data 12 to the internal information area (register, etc.).
Then, while reading the access state data 11, information 25 of each used physical page is created. The information 25 of each physical page to be used is an “array of structures” which is the information of FIG. 4 as one structure, and is created and arranged for the number of pages in which it is used, and is used. Information about all pages.

Thereafter, each used physical page information processing 21 transfers control to each group of information processing 22. Information processing 22 for each group
When the control is passed, the system configuration data 12 set in the internal information area and the information 25 of each used physical page
Is used to create information 26 for each group. Information 2 for each group
6, the information shown in FIG. 5 is made into one structure, and
This is an "array of structures" created and arrayed for each group.
This is information on all groups, and a value obtained by adding 1 to a subscript (starting from 0) of the array of the structure represents a group number of the information of the structure. After that, the information processing 22 of each group passes control to the secondary cache page information processing 23.

When the control is passed, the secondary cache page information processing unit 23 uses the system configuration data 12 set in the internal information area and the information 25 of each used physical page to store the secondary cache page information 27. Create At this time, the contents of the secondary cache page information 27 indicate the state of the contention miss of the secondary cache before the optimization is performed, and the information of FIG. , One page is assumed to be 4096B). This is an array of structures in which the number of pages [secondary cache size / page size (= 4096B)] in the secondary cache is arrayed. The subscript indicates the relative page number in the secondary cache. Thereafter, control is passed to the optimization processing 24.

When the control is passed to the optimization processing 24, (1) first, it is determined whether or not optimization is necessary for each page in the secondary cache by referring to the secondary cache page information 27. If optimization is not necessary, the group information is output to the output information 3 as an analysis result, and optimization processing 2
The processing of 4 is finished. If optimization is required, go to (2). (2) When optimization is required, information 25 of each used physical page, information 26 of each group, and secondary cache page information 27
Is copied into the optimization processing 24 to determine which of the patterns in FIG. 7 is the race condition, and various changes are made in accordance with the pattern to try to optimize. (3) If the optimization is successful, the contents of the changed copy are stored in the information 25 of each used physical page and the information 26 of each group.
By returning to the secondary cache page information 27,
The optimization for one page in the next cache is completed, and the process returns to (1). (3 ′) If the optimization fails, the information 25 of each used physical page, the information 26 of each group, and the secondary cache page information 27
Is again copied (returned) inside the optimization processing 24, and the optimization is tried again with different degrees of change. The maximum number of repetitions of this optimization attempt is [((the number of page conflicts existing in the page in the secondary cache to be optimized at that time)) − 1] × (total number of pages in the secondary cache) -1) It is performed once, and if the optimization still does not succeed, the optimization is stopped, a message is output, and the contents of the information 26 of each group at that time are output to a file of the output information 3 which is the analysis result. And finish all the processing.

Next, the processing of the optimization program and the optimization data 2 will be described for each block therein.

FIG. 8 is a diagram showing an example of an optimization program according to an embodiment of the present invention and information processing of each used physical page 21 of the optimization data 2.
3 shows a flowchart of the block. FIG. 8 shows step 1
001 to Step 1013 will be described.

FIG. 9 shows a flowchart (part 1) of the block of the information processing 22 of each group of the optimization program and the optimization data 2 according to one embodiment of the present invention. FIG. 9 shows step 1
020 to step 1030 will be described.

FIG. 10 is a flowchart (part 2) of the block of the information processing 22 of each group of the optimization program and the optimization data 2 according to one embodiment of the present invention. FIG. 10 shows steps 1040 to 1051.

FIG. 11 is a flowchart (part 1) of a block of the secondary cache page information processing 23 of the optimization program and the optimization data 2 according to an embodiment of the present invention.
FIG. 11 shows steps 1060 to 1068.

FIG. 12 is a flowchart (part 2) of a block of the secondary cache page information processing 23 of the optimization program and the optimization data 2 according to the embodiment of the present invention.
FIG. 12 shows steps 1070 to 1075.

FIG. 13 shows a flowchart (No. 1) of an optimization program and an optimization process 24 for optimizing data 2 according to an embodiment of the present invention. FIG. 13 shows step 1
Steps 080 to 1091 will be described.

FIG. 14 is a flowchart (part 2) of an optimization program and an optimization process 24 for optimizing data 2 according to an embodiment of the present invention. FIG. 14 shows step 1
095 to step 1109 will be described.

FIG. 15 is a flowchart (part 3) of an optimization program and an optimization process 24 for optimizing data 2 according to an embodiment of the present invention. FIG. 15 shows step 1
Steps 110 to 1116 will be described.

FIG. 16 is a flowchart (part 4) of an optimization program and an optimization process 24 for optimizing data 2 according to an embodiment of the present invention. FIG. 16 shows step 1
Steps 1201 to 1131 will be described.

[Explanation of Blocks of Each Used Physical Page Information Processing 21 of Optimized Program and Optimized Data 2]
The blocks of the used physical page information processing 21 of the optimization program and the optimization data 2 will be described.

When the optimizing program and the optimizing data 2 are started, a block of each used physical page information processing 21 is started, and in the flowchart of FIG. 8, an array of all variables and structures, which are internal information of the optimizing, is shown. Is cleared to 0. (Step 1001) Next, the system configuration data 12 shown in FIG. 3 is input and set as variables (variable names are the same as the items in FIG. 3) which are internal information (Step 100).
2).

Next, from the access state data 11, FIG.
(Step 1003).

At this time, if the file of the access state data 11 is at the end, EOF (end of file) is returned, so that the end of the file can be known. Since the access state data 11 usually has data of tens of thousands or more, the process goes to step 1005 because the end of the file is not reached here (step 105).
04).

By comparing the input logical address with the start logical address of the No. 35 fixed area in FIG. 3 of the system configuration data 12 and the last logical address of the No. 36 fixed area, the input logical address is the address of the fixed area. It is checked whether there is any address, and if it is not the address of the fixed area, it is not processed and is ignored, and the process returns to step 1003 to read the next access state data.
Go to (Step 1005).

The physical page address is obtained by setting the lower 12 bits of the physical address to 0, the logical product of the physical address and (the size of the secondary cache −1) is obtained, and the result is divided by the size of the row. Thus, the line number is obtained (step 1006), and all physical page addresses in the array of the structure of the information 25 of each used physical page in FIG. 4 are checked to see if the same physical page address is present (step 1006). Step 1007).

If there is, go to step 1010 to set information there. If not, go to 1013 to set the information on the free physical page (step 1
008).

If there is the same physical page address,
It is checked whether or not the corresponding line of the line use presence / absence information of the used physical page information is unused (0) (step 1010). If not, the next access state data 11 is checked. Go to step 1003, if not used, go to step 1012 (step 1011), add 1 to the corresponding row use information and the number of used rows, and return to step 1003 to check the next access state data 11. (Step 1012).

If there is no identical physical page address,
A physical page address is set in the first empty structure (“the number of used physical page information” of the used physical page information) in the array of the used physical page information 25, and the number of used rows is set to 1, and After setting 1 to the row use presence information and adding 1 to the variable "number of used physical page information" indicating the first free space, the process returns to step 1003 to check the next access state data 11 (step 1003). 101
3).

The above processing is repeated, and when the processing of all the access state data 11 is completed, the end of the file is detected (step 1004). In this case, the "each used physical page information processing 21" ends, and the next Control is passed to the processing of the block (step 1020). As described above, the creation of the "information 25 of each used physical page" is completed, and at this time, the number of information that has been created is set in the information number of used physical pages.

[Explanation of Blocks of Information Processing 22 of Each Group of Optimization Program and Optimization Data 2] When control is passed to the blocks of information processing 22 of each group of optimization program and optimization data 2, first, In order to facilitate the processing of the group information, all the physical page addresses of the information 25 of each used physical page are referred to, and the information 25 of each used physical page is referred to.
Are rearranged in the order of physical page addresses (step 1020).

In order to perform group information processing, the variable (internal information; hereinafter omitted) “number of processed physical pages processed” and the variable “group information number” are set to 0, and the variable “virtual group number 1” is set to al.
l Initially set to F (0xffffffff) (step 102
1).

Next, the result obtained by dividing the physical page address of the 0th used physical page of the information 25 of each used physical page by the size of the secondary cache is set to the variable “virtual group number 2” (step 1022). .

The values of the virtual group number 1 and the virtual group number 2 are used to discriminate whether the group number of the immediately preceding used physical page is equal to the group number of the currently used physical page being processed. Information. Then, first, the group class of the information of the 0-th group (FIG. 5) is 1 representing the head use group.
(Refer to No. 250 in FIG. 5), and set “the value obtained by rounding down the physical page address to the boundary of the group (virtual group number 2 × size of secondary cache)” for the old start address and the current start address (step). 1023).

Since the creation of one group of information has been completed, 1 is added to the number of group information, and the process proceeds to step 1025 to perform the processing of the next group (step 1024).

In the processing of the next group, first, the contents of the virtual group number 2 which is the information of the currently used physical page group is changed to the virtual group number 1 which is the information of the immediately preceding physical page group. Transfer (step 1025).

In order to see the information of the next used physical page, 1 is added to the used physical page processed number (step 102).
6), it is checked whether or not the used physical page processed number is equal to the used physical page information number to see whether or not all used physical page information 25 has been processed (step 102).
7) Since it is not equal here, the process goes to step 1029 to check the information 25 of the next used physical page (step 1028).

Here, the value obtained by dividing the physical page address of the "used physical page processed number" -th used physical page by the size of the secondary cache is set as the virtual group number 2 (step 1029), and the virtual group number is set. It is checked whether or not it is equal to 1. If it is equal, the used physical page currently being processed is in the same group as the previous used physical page, so there is no need to create new group information. Returning to step 1026 to perform the processing, if not equal, the group is not the same as the previous used physical page, so the process goes to step 1040 to create a new group of information (step 1030).

In creating new group information, it is first checked whether or not the virtual group number 2 is equal to "the virtual group number 1 plus 1". Since there is a group having no used physical page (hereinafter abbreviated as an unused group) among the groups to be used, first, the processing goes to step 1043 to perform processing of the unused group. Since there is no unused group among groups to be created, the process goes to step 1041 to create a used group (step 1040).

In the creation of the use group, it is not known at that time whether the use group is the first half use group or the second half use group. Therefore, first, the “group information number” -th group information (newly created) In the group class of “group”, 2 (refer to No. 250 in FIG. 5) representing (“the first half use group meaning that this group is a use group before the sacrifice unused group”) is set for the time being, and the old top address and The current head address is set to “value obtained by rounding down the physical page address to the boundary of the group (virtual group number 2 × secondary cache size)” (step 1041).
Since the information of the new group has been created, 1 is added to the number of group information, and the process returns to step 1025 to process the information 25 of the next used physical page (step 1042).

In the processing of the unused group, first, 3 (see No. 250 in FIG. 5) representing the first half unused group is set as the "group information number" -th group information (information of a newly created group). For the start address and the current start address, “(virtual group number 1 + 1) × the size of the secondary cache” which is “the address immediately after the immediately preceding group” is set, and the size of the group is calculated later. To the old last address and the current last address,
"Address of the page immediately before the next usage group to be created"
"(Virtual group number 2 x size of secondary cache)-
4096) (Note: 4096 is the size of one page) "(step 1043), and a variable" maximum empty group length "representing the maximum size of the unused group created so far.
Is smaller than the size of the unused group currently being created (old end address-old start address) (step 1044).

If it is not smaller, go to step 1048 to end the processing of the unused group, and if smaller, go to step 1046 to update the maximum empty group length and the like (step 1045).

In updating the maximum empty group length or the like, the maximum empty group length is the size of the unused group that is currently being processed (the old last address−
The old head address is set (step 1046), the number of group information is set in the variable “maximum empty group number”, and the process goes to step 1048 to end the processing of the unused group (step 1047).

When the processing of the unused group is completed, 1 is added to the number of group information, and the process returns to step 1041 to create information of the used group immediately after the unused group (step 104).
8).

When the above processing is repeated and the creation of group information is completed for all the used physical pages, the number of processed used physical pages becomes equal to the number of used physical page information (at step 1028). The process proceeds to step 1050 in order to perform processing of the unused sacrifice group.

Of the groups after the maximum empty group number, 5 indicating the latter half use group is set for the group class of the use group, and 6 indicating the latter half unused group is set for the group class of the unused group. By setting 4 indicating a sacrifice unused group to the group class of the group of the group number, the information of the group is completed (step 1050), and the physical page addresses of the information (FIG. 4) of all the used physical pages, The group numbers of the information of all the used physical pages are determined by using the old head address of all the used groups and the information of “old head address + secondary cache size−4096” indicating the end address of the group. To complete the processing of the blocks of the information processing 22 of each group,
Control is passed to the process of the next block (step 1060) (step 1051).

[2 of optimization program and optimization data 2]
Description of Block of Next Cache Page Information Processing 23] When control is passed to the block of the secondary cache page information processing 23 of the optimization program and the optimization data 2, first, the used physical page processed number is initialized to 0. (Step 1060).

Next, the following processes (2) to (8) are performed on the information of the (1) “used physical processed number” -th used physical page. (Step 1061) The following processes (2) to (8) are performed
Since it is limited to the processing of the information of the “used physical processed number” number used physical page, [“the used physical processed number” number]
Is omitted.

(2) The logical product of the physical page address and (the size of the secondary cache -1) is calculated, and the result is calculated as 40
By dividing by 96, the secondary cache relative page number of the used physical page is obtained and set to the variable “secondary cache relative page number” (step 1062).

(3) The following processes (4) to (8) are performed on the secondary cache page information of the “secondary cache relative page number” obtained in (2) above (step 106).
3).

In the following (4) to (8),
Since the processing is limited to the processing of the “secondary cache relative page number” -th secondary cache page information, all the description of “[secondary cache relative page number] th” is omitted.

(4) The subscript of the used physical page (equal to the number of used physical processes) is added to the subscript list of the used page of the secondary cache page information (the subscript of the used page is added to the “duplicate number” in the subscript list). Then, by adding 1 to the page duplication number, the fact that the used physical page uses the secondary cache page information is registered in the secondary cache page information (step 1064).

(5) For all the rows of “the row in which the row usage information of the used physical page information is used (not 0)”, the “secondary cache page information row” corresponding to the row number is used. 1 is added to the “number of times of use”, and if the addition result is 1, 1 is further added to the number of used lines of the secondary cache page information to reduce the usage status of the used physical page line by 2 The next cache page information is registered (step 1065).

(6) By adding 1 to the used physical page processed number, the processing of the used physical page ends (step 1066).

(7) It is checked whether or not the number of used physical page processes is equal to the number of used physical page information (step 106).
7).

(8) If they are not equal, the process returns to step 1061 to process the next physical page used (return to (1) above). If they are the same, the page and line reference status of the information of all used physical pages is reflected in the secondary cache page information. Next, in order to process the conflicting page of the secondary cache page information, step 1070 [described below] is performed. (9)] (step 1068).

(9) First, the processing of the competing page is as follows.
Variables “Number of processed secondary cache pages” and “Number of all conflicts”
Is initialized to 0 (step 1070).

Next, the following processes (11) to (14) are performed on the (10) “number of processed secondary cache pages” -th secondary cache page information. (Step 107
1) Note that in the following (11) to (14), the processing is limited to the processing of the “number of processed secondary cache pages” -th secondary cache page information. Is omitted.

(11) A line in which the number of uses of the secondary cache page information per line is 2 or more causes a conflict error. Therefore, the following process (12) is performed for each line in which the conflict error occurs. Do. (Step 1072) In the following (12), since the processing is limited to the processing of a certain row in which a conflict error has occurred, the description of [the row in which the conflict error has occurred] is omitted.

(12) Information on the used pages (which may cause a conflict error) is sequentially obtained from the used page subscript list, and for each used page, it is determined whether the above line is used or not. , Check if the used page already exists in the subscript list of the competing page in the secondary cache page information, and if not, add it to the subscript list of the competing page. One is added to the number of page conflicts and the variable “total number of conflicts” (step 1073).

(13) One is added to the number of processed secondary cache pages (step 1074).

(14) The size of the secondary cache is set to 409
By checking whether or not the number divided by 6 is equal to the number of processed secondary cache pages, it is checked whether or not all the secondary cache page information has been processed. Step 1071 [the above (1)
0)], and if they are equal (all have been processed), the processing of this block is terminated, and the flow proceeds to step 1080 to pass control to the optimization processing (step 1075).

[Explanation of optimization program and optimization data 2 optimization processing 24 block] When control is transferred to the optimization program and optimization data 2 optimization processing 24 block, first, (1) It is checked whether the total number of conflicts is larger than 0, and if it is larger, optimization is necessary.
Go to 81, and if 0, optimization is unnecessary, so go to Step 1130 (Step 1080). If optimization is necessary, first, the variable “number of processed secondary cache pages” is initialized to 0 (step 1081), and the process proceeds to step 1082.

(2) “Number of processed secondary cache pages”
It is checked whether or not the page contention number of the second secondary cache page information is greater than 0 (step 1082). If it is large, the secondary cache page is first subjected to optimization at step 1085 [optimization]. (4)], and if it is not larger than 0, the optimization is not necessary for the secondary cache page, so the flow proceeds to Step 1084 “(3) below” (Step 1083).

(3) If optimization is not required for the secondary cache, 1 is added to the number of processed secondary cache pages in order to process the next secondary cache page, and step 1082 “ 2) "(Step 1084).

(4) When the secondary cache needs to be optimized, if the secondary cache page information is successfully optimized, at least the page of the secondary cache page information is calculated from the total number of conflicts. Since the number of page conflicts should decrease by the number of conflicts, a value obtained by subtracting the number of page conflicts of the secondary cache page information (the “number of processed secondary cache pages”) from the total number of conflicts is obtained. The target number of competitions for optimization is set (step 1085).

(5) If the optimization fails, it is necessary to restore each information and perform another optimization again. Therefore, before the optimization, first, the information of each used physical page and the secondary The cache page information and all the information in each group are copied to the internal information area (step 1086).

(6) Step 1 which is the subsequent optimization processing
In steps 087 (7) to 1127 (30), processing is performed on the information of each used physical page copied above, the secondary cache page information, and the information of each group (the description of “copied area” is omitted). (Step 10)
87).

(7) When there are N competing pages, N-1
If the conflicting page is optimized (moved), the conflict should be resolved. Therefore, in order to determine which conflicting page is not moved (exempt from optimization), the "number of processed secondary cache pages" The subscripts are obtained from the subscript list of the competing pages of the information by the number of page conflicts of the secondary cache page information of the second cache page information, and the group class of each group information is obtained. It is checked which one of the "patterns" matches (step 1088).

(8) If the pattern is No. 1 (only the first use group or the first half use group), the contending page with the smallest group number is excluded from the optimization target from FIG. 7, and immediately before the remaining contending page group Go to step 1091 [(10)] because it is necessary to insert a dummy page (step 1089).

(9) When the pattern is N02 (only the second half group), the conflicting page with the largest group number is excluded from the optimization target from FIG. 7, and immediately after the remaining conflicting page group,
Since it is necessary to embed a dummy page for each, go to step 1095 [(11)]. In other cases, the page with the smallest group number is excluded from the optimization target, and among the remaining competing pages, Since it is necessary to insert a dummy page immediately before the first half use group and immediately after the second half use group, the process goes to step 1091 [(10)] (step 1090).

(10) The above (7) [Step 1088]
Among the competing pages with the subscripts obtained in step 2, only the competing page having the smallest group number is removed, and the remaining competing pages are defined as competing pages to be optimized, and the process proceeds to step 1096 (12) (step 1091).

(11) The above (7) [Step 1088]
Of the conflicting pages with the subscripts obtained in step 2, only the conflicting page having the largest group number is removed, and the remaining conflicting pages are defined as the conflicting pages to be optimized, and the process proceeds to step 1096 (12) (step 1095).

(12) The number of competing pages to be optimized is checked and set to the variable “optimizing competing page number” (step 1).
096).

(13) Array of internal information “number of inserted pages [0] to number of inserted pages [optimized contention page number−1]”
As the number of inserted pages for the first optimization
Step 11 to initialize and start optimization
Go to 00 (14) (step 1097).

(14) Optimization is performed by the number of optimization conflicting pages,
First, the variable “target number” is initialized to 0 (step 1100).

(15) The group class of the group of conflicting pages to be optimized for the processing target is checked (step 1101). If the group is the first-half use group, the process goes to step 1103 (16); Go to 1105 (18) (step 1102).

(16) If the group is the first-half use group, add the number of pages to be inserted [processing target number] to the number of pages to be inserted immediately before the information of the group (step 1103) (here, “add” instead of “setting”). Is performed in order to take over the result of the past optimization success. If the optimization has never been successful in the past, the number of pages immediately before the information of the group is 0 in the initial state, If it succeeds, it may be a value other than 0, so it is necessary to inherit that value.)

(17) For each of the groups after the group and before the sacrifice unused group, the number of pages to be inserted [processing target number] × 409 is added to the current head address of the group information.
6 and then go to step 1107 (20) (step 1104) (again, the setting is not performed here, but the addition is performed because the value has already been added to the current start address due to optimization of another group, etc.). In some cases, the result is taken over, and the result is omitted.

(18) If the group is not the first-half use group, the number of pages to be inserted [the number to be processed] is added to the number of pages to be inserted immediately after the information of the group (step 1105).

(19) The number of inserted pages from the current start address of the group information for all groups before the group and after the sacrificial unused group, and from the current end address for the sacrificial unused group, respectively. [Process target] x 40
96 is subtracted, and the procedure goes to step 1107 (20) (step 1106).

(20) In order to process the next group of competing pages, 1 is added to the processing target (step 110).
7) It is checked whether the result is equal to the number of competing pages to be optimized (step 1108).

(21) If they are equal, the processing of all the groups of competing pages to be processed has been completed.
Go to 10, if not equal, there is a conflict page to be processed that has not yet been processed, so step 1101 (15)
And the process is repeated (step 1109).

(22) When the processing of all the contended page groups to be processed is completed, the current head address and the old head address of the information of the group of the group number are respectively applied to all the information of each used physical page. Is obtained, and if the difference is not 0, it is optimized (the address has been moved)
Therefore, 1 indicating that the physical address has been changed by the optimization is set to the page attribute of the information of the used physical page, and (old physical address + difference) is set to the new physical page address (step 1110). .

(23) Next, the secondary cache page information is all cleared to 0 (step 1111). For all the information of each used physical page, if the page attribute = 1, the new physical address and the page attribute = If 0, the old physical address is used, and steps 1060 to 10
By performing the same processing as in step 75 and recreating each piece of secondary cache page information, a new total number of conflicts is obtained, and the total number of conflicts is set to "new total number of conflicts" (step 1112).

(24) It is checked whether or not the new total number of competitions is equal to or less than the target number of competitions. If not, the optimization has failed. Go to (25) and if
Since the optimization is successful, in order to optimize contention for the next secondary cache page, step 1114 (3
Go to 1) (step 1113).

(25) If the optimization fails, in order to redo the optimization under another condition, the used physical pages of the copy source copied in step 1086, the secondary cache page information, and the information of each group are read. All are again copied to the internal area to return to the state prior to the failed optimization described above (step 1116).

(26) In order to change the condition, first, the number of optimized conflicting pages −1 is initially set in a variable “number of repetitions” (step 1120).

(27) Next, 1 is added to the number of inserted pages [the number of repetitions] (step 1121), and it is checked whether the number of inserted pages is equal to or larger than (the size of the secondary cache 次 4096) (step 1121). 1122).

(28) If not, step 1100 [the above (1)
4)], the next optimization is started, and if it is more than that, there is no point in increasing the number of inserted pages, so step 1124 is performed to increase another inserted page.
Go to (29) (step 1123).

(29) In order to increase another insertion page,
First, it is checked whether the number of repetitions is 0, and if it is 0, all inserted pages can no longer be increased, so that the step 1125 is performed to stop the optimization.
Go to (31) and if it is not 0, it is possible to increase the number of other inserted pages, so step 1126 (30)
Go to (Step 1124).

(30) If the number of inserted pages can be increased, the number of inserted pages [the number of repetitions], which is the current number of inserted pages, is set to 1 in the initial state (step 11).
26), subtracting 1 from the number of repetitions, and step 1121
By returning to [(27)], another process of increasing the number of inserted pages is performed (step 1127).

(31) If it is not possible to increase all the inserted pages, an optimization stop message is output to the output information 3, and then the optimization result up to the stop of the optimization is output in step 1131. Go to (Step 11
25).

(32) The above (24) (step 111)
If the optimization success is detected in 3), the new total number of competitions is set to the total number of competitions (step 1114), and step 1086 is performed.
All of the used physical pages, secondary cache page information, and information of each group of the copy destination at the time of copying are written back to the copy source information for use in the next optimization, and thereafter, the process returns to step 1080. (Step 1115).

(33) In step 1080, if the total number of conflicts is larger than 0, further optimization is required.
The next optimization is performed by repeatedly executing step 1082 [(2)] and subsequent steps. If the total number of conflicts is 0, all optimization has succeeded. Go to 1130 (34) (step 1080).

(34) If all the optimizations have succeeded, a message indicating that the optimization has been completed is output to the output information 3 and then the procedure goes to step 1131 (35) (step 113).
0).

(35) Finally, all information of each group is output to the output information 3 as an analysis result, and the optimization program 2 is stopped (step 1131).

[0106]

As described above, according to the present invention, there is an effect that optimization of contention error in a fixed area, which has been neglected in the past, can be optimized.

Further, by using the access status data 11 and the system configuration data 12 as input information, information 25 of each used physical page, information 26 of each group, and secondary cache page information 27 are created and optimized. And the number of contention errors in the fixed area in the secondary cache can be minimized.

In the fixed area, the relationship between the physical address and the logical address is fixed. Therefore, the logical address in which the dummy pages corresponding to the number of inserted pages are inserted at each insertion position indicated by the output information 3 obtained above. If specified at the time of system generation, the number of contention misses in the secondary cache, which has been minimized by the above-mentioned optimization, can be realized on an actual system.

[Brief description of the drawings]

FIG. 1 is a diagram showing input / output outlines and blocks of a “method for optimizing a cache conflict miss in a fixed area of a secondary cache” according to an embodiment of the present invention.

FIG. 2 is a configuration diagram of access state data according to an embodiment of the present invention.

FIG. 3 is a configuration diagram of system configuration data according to an embodiment of the present invention.

FIG. 4 is a configuration diagram of information of each used physical page according to an embodiment of the present invention.

FIG. 5 is a configuration diagram of information of each group according to an embodiment of the present invention.

FIG. 6 is a configuration diagram of secondary cache page information according to an embodiment of the present invention.

FIG. 7 is a diagram illustrating a relationship between a pattern of a competition group and an optimization method according to an embodiment of the present invention.

FIG. 8 is a flowchart of each used physical page information processing 21 of the optimization program and the optimization data 2 according to the embodiment of the present invention.

FIG. 9 is a flowchart (part 1) of information processing 22 of each group of the optimization program and the optimization data 2 according to an embodiment of the present invention.

FIG. 10 is a flowchart (part 2) of the information processing 22 of each group of the optimization program and the optimization data 2 according to an embodiment of the present invention.

FIG. 11 is a flowchart (part 1) of the secondary cache page information processing 23 of the optimization program and the optimization data 2 according to the embodiment of the present invention.

FIG. 12 is a flowchart (part 2) of the secondary cache page information processing 23 of the optimization program and the optimization data 2 according to the embodiment of the present invention.

FIG. 13 is a flowchart (part 1) of an optimization program 24 and an optimization process 24 of the optimization data 2 according to the embodiment of the present invention.

FIG. 14 is a flowchart (part 2) of the optimization program 24 and the optimization processing 24 of the optimization data 2 according to the embodiment of the present invention.

FIG. 15 is a flowchart (part 3) of the optimization processing 24 of the optimization program and the optimization data 2 according to the embodiment of the present invention.

FIG. 16 is a flowchart (part 4) of the optimization processing 24 of the optimization program and the optimization data 2 according to the embodiment of the present invention.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 input information 2 optimization program and optimization data 3 output information 11 access status data 12 system configuration data 13 file name information 21 each used physical page information processing block 22 each group information processing block 23 secondary cache page information processing Block 24 Block of optimization processing 25 Block of information of each used physical page 26 Block of information of each group 27 Block of secondary cache page information

Claims (1)

[Claims] A central processing unit in an information processing apparatus has a high-speed cache memory, and the cache memory is a set of a large number of rows, and the row number is determined using a part of information of a physical address. Is fixed when the address space in which the cache memory can be used includes a fixed area defined as "an area in which the relationship between a logical address and a physical address is fixed by hardware" In an optimization method for minimizing contention mistakes in an area cache memory, a step of dividing the fixed area into units each of which is an integral multiple of the size of a physical addressing cache memory; Accordingly, a procedure for inserting a dummy area of one page, and all pages in a state where the dummy area is inserted. Calculating the number of contention misses of the physical addressing cache memory by reflecting the usage status of the row of the page, and calculating the number of contention misses as "a page contention between different pages for each physical addressing cache memory page. The result of subtracting the number of page conflicts relating to one page in the physical addressing cache memory that is being optimized from the total number of conflicts, which is the sum of the numbers, is less than the target conflict number. If there is, there is one in the physical addressing cache memory that is performing the optimization process.
A procedure for obtaining an optimal solution for the group with respect to pages; and if the calculated number of conflict errors is larger than the target number of conflicts, the number of pages to be inserted immediately before or immediately after the group is 1
A procedure of increasing the number of pages and repeating the calculation and comparison, and the number of inserted pages is a value obtained by subtracting 1 from the total number of pages of the physical addressing cache memory for all the groups to be optimized. Reached,
A cache conflict in a fixed area, comprising: a step of stopping the optimization; and a step of repeating the above for all pages having a contention miss in the physical addressing cache memory if the optimization is not stopped. How to optimize for mistakes.