JP2735255B2 - Hatching treatment method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention provides high-speed execution of hashing by performing vector processing by a pipeline type vector computer or by performing parallel processing by a parallel type computer. The purpose is to enable.

[Conventional technology]

Hashing is a method of registering well-known data with a key in a table and a method of searching the table, which are often used because data can be searched at high speed. There are two types of hashing, open hashing and bucket hashing, both of which are described in the following documents.

(1) The Art of Computer Programming, Volume 3, Sorting and Searching, Addison Wesley, pp. 506-549 (1973) (The Art of Computer Programming, Vol3, Sorting a
nd Searching, Addison Wesley, pp. 506-549, 1973).

Document (1) is based on the premise that hashing is sequentially processed.

[Problems to be solved by the invention]

Although it is conceivable to use a vector computer to execute hashing faster than a general-purpose computer, in the method described in reference (1), registration of a key in a table is performed by vector processing or parallel processing. Could not be executed at high speed. Hereinafter, the method of Reference (1) is referred to as “conventional technology”.

The reason why the conventional method does not allow vector processing or parallel processing even when registering multiple keys at once,
This will be described with reference to FIG. Hereinafter, a table for registering keys according to general terms will be referred to as a hash table. FIG. 2 shows the process of registering four keys included in the array 211 in the hash table 212. When registering the keys 201 and 202 included in the array 211, the hash value 22 of the key 201 is registered.
1 and the hash value 222 of the key 202 are to be registered in the hash table as subscript values, but the hash value 221 and the hash value 2
22 can collide with, or become unidentified.

In the conventional method, the calculation regarding the key 201 and the calculation regarding the key 202 are sequentially performed as shown in FIG. Therefore, when the key 202 is registered, it is detected that a collision has occurred. Therefore, the subscript value 223 can be obtained by recalculation and registered. However, due to this sequentiality, the registration of each key cannot be performed by vector processing or parallel processing, and therefore cannot be performed by vector processing or parallel processing. Also the key
If vector processing or parallel processing is forcibly performed so that registration of 201 and registration of key 202 are performed in parallel, collision cannot be detected properly, and processing is terminated without any key being registered. I will.

[Means for solving the problem]

In order to solve the above-described problem, the calculation of a plurality of keys and the registration in the hash table are performed in parallel, and the collision process is performed in the following procedure.

(1) A collision between a key already registered in the hash table before registration of the plurality of keys and the plurality of keys may be performed by vector processing or parallel processing before registering the plurality of keys. , Ie, by vector processing or parallel processing.

(2) A collision that occurs when registering a plurality of keys described above, after writing them into a hash table, reads out a plurality of values stored in the table, and matches the corresponding plurality of keys. This is detected by judging whether this is done by vector processing or parallel processing.

(3) The key for which the collision is detected in (1) and (2) is registered again in a different entry in the hash table. Since a collision may occur again, the operation of (2) is repeated while changing entries.

[Action]

If the collision process is performed in the above procedure, (1) When a collision between a key registered in advance in the hash table and a plurality of keys to be newly registered occurs, the plurality of keys are registered in the previously registered key. (2) When a collision occurs between the plurality of keys, it is determined which of the keys was not properly written in due to the collision. Since the repeated registration processing is performed until they are detected and registered properly, all of the plurality of keys can be registered in the hash table at high speed by vector processing or parallel processing.

〔Example〕

An embodiment of the present invention will be described with reference to FIGS. 1, 3, and 5. FIG. Although this embodiment relates to open hashing, bucket hashing can be easily realized with only a few changes.

FIG. 1 is a flow chart showing a procedure of including a plurality of keys as elements, inputting array data in a main memory, and registering the array data in a hash table in the main memory. Each of the steps shown in FIG. 1 is sequentially executed by using an arithmetic unit and a vector register of a vector computer. In each step, parallel processing, ie, pipeline processing, chaining, and parallel execution of a plurality of instructions are performed. I am doing it.

FIG. 3 shows the data generated when the procedure shown in FIG. 1 is executed by inputting an array K 211 composed of four keys, and the procedure. The size of the hash table 212 is 6. Normally, a much larger value is used as the size of the hash table, but here a small value is used for the sake of simplicity. In most cases, keys and data are registered in pairs in the hash table, but only keys are registered in the hash table 212. In the hash table 212, a key 103 is registered in advance.

FIG. 5 is a diagram showing the configuration of the vector computer in the present embodiment. That is, it is composed of a main storage device 501, a vector register group 502, a mask register group 503, and a pipeline operation unit group 504. Then, loading and storing can be performed from the main storage device 501 to an arbitrary register of the vector / register group 502 by pipeline. Further, data is input from an arbitrary register of the vector register group 502 and an arbitrary register of the mask register group 503 to an arbitrary arithmetic unit of the arithmetic unit group 503 to perform a pipeline operation, and a result is stored in the vector register group 502. It can be stored in any register. In the present embodiment, the main storage device 501
The key layout K 211 and the hash table 212 are stored in advance.

 The procedure of FIG. 1 will be described with reference to FIG.

[1] First, step 101 is executed. That is,
The hash value of each element K (i) of the key array K 211 is obtained in the arithmetic unit of the vector computer, and is substituted for each element HV (i) of the array HV301. The value of the array HV301 is a value greater than or equal to 0 and less than the size of the hash table, and is held on the vector register of the vector computer.

[2] Next, step 102 is executed. That is, the hash table HT212 is drawn using each element HV (i) (i = 1,..., N) of the array HV301 as a subscript. In the case of FIG. 3, N = 4
It is. The entry HT of the hash table HT 212 (HV (i))
It is determined whether or not the value obtained from (i = 1,..., N) is a value indicating that it is not registered (represented by * in the figure). When the result of the determination is obtained as a logical value and an array having the logical value as an element is generated, an array 303 is obtained.
Since i = 1,2,4 is not registered, the value of the element of the corresponding array 303 is true, but for i = 3, the hash table
Since the value 103 already registered in the HT 212 is accessed, the value of the corresponding element of the array 303 becomes false.

Only for i for which the value of the element of the array 303 is true, the element HV (i) of the key array K 211 is stored in the hash table HT 212 as a subscript. That is, it is assigned to the entry HT (HV (i)) of the hash table HT212. That is, the value 353 of the first element, the value 621 of the second element, and the value 911 of the fourth element of the key array K 211 are substituted. Since the value 353 and the value 911 are substituted into the fifth element of the hash table HT 212 by this value, the value of the fifth element is 911. As a result, the hash table HT becomes like 212e.

[3] Next, step 103 is executed. That is, each element K (i) of the key array K 211 (i = 1,..., N)
Is determined to be equal to the value included in the entry HT (HV (i)) of the hash table HT 212e. Each element ENT (i) of the array ENT 304 as a logical value using the result of the determination.
Substitute for When K (1) is compared with the value 911, it is found that the value is not equal, so the value of ENT (1) is false. K
(2), K (4)
The values of ENT (2) and ENT (4) are each true. K
Since (3) is compared with the value 103, it is known that there is no difference, and the value of ENT (3) becomes false.

[4] Next, step 104 is executed. That is, the number of unregistered elements can be reduced by counting the number of elements of the array ENT 304 having a value of false in the mask count instruction (eg, the VMCO instruction in Hitachi S-810) of the vector computer. First, substitute for the variable NR. The value of the variable NR is 2.

[5] Next, step 105 is executed. That is, by controlling the vector compression storage instruction (either the VSTC instruction or the VSTDC instruction in S-810) in the array ENT 304, each element HV of the array HV 301 is controlled.
(J) By storing only the unregistered ones of HV (i) (i = 1,..., N) in (j = 1,..., NR), the array HV
Will look like 301a. That is, since the first element ENT (1) of the array NET 304 is a value indicating that it is not registered, that is, false, the value 5 of the first element of the array 301 is stored in the first element of the array 301a. Since ENT (2) is a value indicating that it has been registered, that is, true, the value 3 of the second element of the array 301 is omitted. Since the value of ENT (3) is false,
The value 1 of the third element of the array 301 is stored in the second element of the array 301a. Since the value of ENT (4) is true, the value 5 of the fourth element of the array 301 is discarded. Each element K (j) of array K 211
In (j = 1,..., NR), only the unregistered values of K (i) (i = 1,..., N) are stored. Therefore, the array K
The value of the first element of 211a is 353, and the value of the second element is 415.

[6] Next, step 106 is executed. That is, it is determined whether the value of the variable NR is 0. If it is 0, the registration has been completed, and the execution of the procedure ends.
In this case, since the value of the variable NR is not 0, that is, there is a key whose registration has not been completed, the processing is continued.

[7] Next, step 107 is executed. That is, the value of the variable NR is executed for the variable N. The value of N becomes 2.

[8] Subsequently, step 108 is executed. That is, using the value HV (i) of each element of the array HV 301a,
(I) +1) is obtained by hash table size (i = 1,... N), and each element HV of array HV 301b is obtained.
(I) Substitute (i = 1,... N).

[9] Next, step 102 is executed again. That is, the hash table HV212e is drawn using each element HV (i) (i = 1,... N) of the array HV301b as a subscript. H which is the result
It is determined whether or not T (HV (i)) (i = 1,..., N) is a value indicating that it has not been registered. When the semi-determined result is obtained as a logical value and an array having the logical value as an element is obtained, an array 305 is generated. Since neither i = 1 or 2 has been registered, the value of the corresponding element of the array 305 is true.

The entry HT of the hash table HT 212e of the key array K 211a only for i for which the value of the element of the array 305 is true.
(HV (i)). In this case, both the first element 353 and the second element 415 of the key array are substituted. As a result, the hash table HT becomes like 212f.

[10] Next, step 103 is executed again. That is, each element K (i) of the key array K 211a (i = 1,...)
N) is the entry HT (HV
It is determined whether the values match as the values included in (i)). The result of the determination is substituted into each element ENT (i) of the array ENT 306 as a logical value. K (1) value and value 353, K
The value of (2) is compared with the value 415, and it is found that they are all equal. Therefore, the values of ENT (1) and ENT (2) are both true.

[11] Next, step 104 is executed again. That is, by counting the number of elements of the array ENT 304 having a value of false in the mask count instruction of the vector calculator, the number of unregistered elements is collected,
And The value of NR is 0.

[12] Next, step 105 is executed. That is, by controlling the vector compression storage instruction by masking in the array ENT 306, each element HV (j) (j =
.. NR) store only unregistered HV (i) (i = 1,... N). The array HV is an empty array, that is, an array having zero elements. Each element K (j) of the array K 211a (j = 1,
.. NR) only the unregistered ones of K (i) (i = 1,... N) are stored. Therefore, the array K also becomes an empty array.

[13] Next, step 106 is executed. That is, it is determined whether the value of the variable NR is 0. Since the value of the variable NR is 0, that is, the registration has been completed, the execution of the procedure ends. In the hash table HT 212f obtained as an output, all input keys are registered.

 This is the end of the description of the embodiment, and a supplementary explanation will be given.

In the above embodiment, only the key is registered in the hash table. However, in general, a key and data are paired and registered in a hash table. In that case, each entry of the hash table may be divided into two, and a key and data may be stored in each. When the array containing the keys is compressed (step 105), it is necessary to also compress the array containing the data.

Further, in the above embodiment, whether the key has been registered properly is determined by comparing the key registered in the hash table with the key included in the key layout. Therefore, when the key layout includes a plurality of keys having the same value, the operation does not work properly. In order to operate properly in such a case, it is sufficient to register in accordance with the procedure shown in FIG. 4 instead of step 102 in FIG.

Further, in the above embodiment, the execution by the pipeline type vector computer is premised, but the parallel computer can perform the same processing as the pipeline type vector computer. FIG. 6 shows a parallel computer capable of executing the procedure of FIG. In this parallel computer, the main storage units of the main storage unit group 601 are connected to the processing units of the processing unit group 602 by buses, respectively, and each bus is connected by a data exchange network 603. In the case of a parallel computer,
Although it is necessary to synchronize the processing units so that each step of FIG. 1 is executed sequentially, in the parallel computer of FIG. 6, synchronization is achieved by the synchronization bus 604. The key layout K 211 and the hash table 212 correspond to the sixth
As shown in the figure, the data is distributed and stored in the main storage device group in advance.

〔The invention's effect〕

In a vector computer such as the Hitachi S-810, most of the operations in each of the above steps can be executed by vector instructions. Or, in the case of more than that, registration to the hash table can be performed several times to several tens times faster than the conventional method. The effect of synchronization can also be obtained in a parallel computer.

[Brief description of the drawings]

FIG. 1 is a flow chart showing a procedure for registering a hash table in an embodiment of the present invention, that is, a procedure for inputting an array including a plurality of keys as elements and registering the array in a hash table by vector processing. It is. FIG. 2 is an explanatory diagram of a problem of the conventional method. That is, 4
The table shows the data generated when registration in the hash table is performed by scalar processing in which an array consisting of individual keys is input, and the procedure. FIG. 3 shows the data generated when the procedure shown in FIG. 1 is executed by inputting an array consisting of four keys in the embodiment, that is, when data is registered in the hash table by vector processing, The procedure is closed. FIG. 4 is a flow chart showing a procedure instead of a part of FIG. 1 in the supplementary embodiment. FIG. 5 is a diagram showing a configuration of a vector computer according to the present embodiment. FIG. 6 is a diagram showing a configuration of a parallel computer in a supplementary embodiment.

Claims (1)

(57) [Claims] 1. A hashing method for hashing a plurality of keys to calculate an address of a hash table on a storage device and storing the plurality of keys at the address, wherein the calculation relating to the plurality of keys and the hashing are performed. A plurality of keys are simultaneously stored in a table by vector processing or parallel processing, and a collision process at the time of the storage is performed. (1) Before storing the plurality of keys, the keys are already stored in the hash table. The collision between the key and the plurality of keys is detected by vector processing or parallel processing before storing the plurality of keys.
(2) When a collision occurs between the plurality of keys at the time of storage, after storing them in the hash table, a plurality of values stored in the hash table are read out and the corresponding plurality of keys are read out. A second step of detecting whether or not the key matches with a vector processing or a parallel processing, and (3) a key in which a collision is detected in the first and second steps is different from the hash table. A hashing processing method performed by the third step of storing the data in the address again.