JP2535655B2 - Dictionary search method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Purpose] Outline Industrial field of application Conventional technology Problems to be solved by the invention Means for solving the problem Action Example Effect of the invention [Summary] For example, incremental decomposition type Ziv-Lempel Regarding the dictionary search method used at the time of encoding, for the purpose of searching the dictionary at high speed, in the dictionary search method for searching the character string represented by the input reference number and character, the dictionary is a candidate An index that divides the element into a plurality of groups based on the extension character and stores the reference number of the candidate element belonging to each group and the identification information corresponding to this candidate element, corresponding to the reference number and the information about the extension character. And a list storing the reference numbers of any of the other candidate elements belonging to the same group as the corresponding candidate element and the identification information corresponding to this candidate element, corresponding to the reference number. , At first, from the index corresponding to the input reference number and the information about the input character, and thereafter, from the list corresponding to the reference number output by the dictionary, output the reference number and the identification information to the dictionary. Based on the identification information read from the dictionary, the detection means for detecting the candidate element in which the input character is added as an extended character, and the reference number read from the dictionary. And a determination unit that determines whether or not there is a candidate element that does not exist, and the reading unit, the detection unit, and the determination unit operate independently.

[Industrial applications]

The present invention relates to a dictionary search method used in Ziv-Lempel encoding of an incremental decomposition type which is a kind of universal code.

In recent years, various types of data such as character codes, vector information, and image information have been handled by computers, and the amount of data handled has been increasing rapidly.

When storing or transmitting such a huge amount of data, it is desirable to omit the redundant part contained in the data and compress the data amount. Therefore, there is a demand for a method of efficiently compressing data regardless of the type of data.

The universal coding method has a feature that it can be applied to the compression of various data described above, because it is not necessary to define a code table in advance.

Here, in the present specification, a word unit of data is referred to as a “character”, and data of a plurality of consecutive words is referred to as a “character string”.
Called.

The Ziv-lempel code is a typical method of the above-mentioned universal code (see Munakata "Ziv-Lempel").
Data compression method of empel ", Information Processing, Vol.26, No.1,1985), universal type algorithm and incremental decomposition type algorithm are proposed. Furthermore, as an improvement of the universal type algorithm, LZSS code (TCBell, “Bet
ter OPM / L Text Compression ", IEEE Trans.on Commun.,
Vol.COM-34, No. 12, Dec. 1986), and as an improvement of the incremental decomposition type algorithm, LZW code (TAWelch, “AT
echnique for High-Performance Data Compression ", C
omputer, June 1984).

Among these encoding methods, the LZW code has come to be used for file compression of a storage device because of its high-speed processing and its simple algorithm.

[Conventional technology]

The incremental decomposition type algorithm decomposes an input character string into a series of components formed by adding one character as an increment to a subsequence already registered in the dictionary, and this sequence of components is registered as the registered subsequence. The input character string is encoded by representing it by the reference number corresponding to and the increment. In addition, the above-mentioned components are registered in the dictionary as new subsequences, and are used in the subsequent encoding processing.

Further, in the LZW code, the increment described above is incorporated in the next subsequence.

Hereafter, for the sake of simplicity, as input string, "a", "b",
Character string consisting of three characters "c""ababcbababaaaaa ...
・ When "" (see Fig. 4 (a)) is input,
This LZW encoding method will be described.

In this case, the reference characters "1", "2", and "3" are given to the above-mentioned three characters "a", "b", and "c" to register them in the dictionary, and the encoding process is started.

First, read the first character (for example, the character "a") of the input character string, search for this character from the dictionary, and refer to the reference number (for example, "1") corresponding to this character as the code corresponding to the focused character string. Let ω.

After that, the second and subsequent characters of the input character string are sequentially read, and this character is designated as the extended character K corresponding to the increment described above, and is represented by the combination (ωK) of the above-mentioned code ω and this extended character K. Subsequence (ωK) (hereinafter, combination (ω
K) is referred to as a subsequence expression) is searched from the dictionary. When the corresponding subsequence (ωK) is retrieved, the reference number corresponding to the above-mentioned subsequence (ωK) is set as a new code ω, the next character of the input character string is read, and the above-described processing is repeated. .

In this way, the character string to be encoded is sequentially extended one character at a time, and this character string is searched sequentially from the dictionary, so that the input character string of the input character string is selected from the substrings registered in the dictionary. The substring that matches the part of interest the longest is searched, and the reference number corresponding to this substring is
It is output as the corresponding code ω. Further, at this time, the partial sequence in which the extended character K is added to the partial sequence (ω) corresponding to the reference number ω is a combination of the reference number ω and the extended character K (ω
K), given a reference number and registered in the dictionary as a new subsequence.

In this way, the character string shown in FIG. 4 (a) is decomposed into underlined partial strings in the figure, and corresponds to each partial string as shown in FIG. 4 (b). Code "1",
“2”, “4”, ... Are output. Also, FIG. 4 (c)
Table 1 shows an example of the dictionary created in Table 1 for the correspondence relationship between the input character string and the partial string registered in the dictionary.

The dictionary created during the above-described LZW encoding process has this structure as shown in FIG. 5, and each element of the dictionary corresponds to each node of the dictionary tree. There is.
In Fig. 5, the numbers shown in brackets at each node are
The reference number of the corresponding dictionary element is shown.

Here, when searching for subsequences in the above-described encoding process, the elements registered in the dictionary are searched sequentially,
Since the processing time becomes long, the hash method is applied to the dictionary search process to speed up the process.

In the hash method, a function (hash function) for finding the address of the storage location of this element x from the element x of the set S consisting of character strings is defined, and the element x is stored at the address found by this hash function. It is a thing. The address obtained by the above-mentioned hash function is called a hash address.

For example, the reference number ω and the extended character K described above are represented by a binary number, and this may be a hash address of a combination (ωK). However, in this case, it is necessary to allocate a huge capacity of the dictionary.

Therefore, an external hash method is used in which a list storing elements having the same hash address is provided for each hash address. In the external hash method, as shown in FIG. 6, a corresponding list is shown by searching the index section with a hash address. Further, each list stores identification information corresponding to each element and a pointer indicating the storage location of the next element. In this way, the elements having the same hash address are linked by the index and the list so that they can be searched sequentially.

For example, the reference number ω described above is used as a hash address, and the top address of the list that stores a partial sequence in which one character is added to the partial sequence corresponding to the reference number ω is stored in this hash address. Further, the partial list corresponding to the node corresponding to the “child” of the node corresponding to the above-mentioned reference number ω may be sequentially stored in the corresponding list. In this way, the connection relationship between the candidate elements represented by the combination of the reference number ω and the one letter K may be shown. Further, in this case, the extended character K of each element may be stored in the list as corresponding identification information.

FIG. 7 is a flowchart showing the encoding operation when the external hash method is used for searching the dictionary.

As described above, the dictionary is initialized to include at least the first character of the input character string, and the variable n is set to the reference number given to the substring registered next. For example, the reference numbers "1", "2" assigned to the characters "a", "b", "c",
"3" is stored as a hash address in the dictionary, and the variable n
Set the numerical value "4" to. Here, let N _max be the maximum number of subsequences that can be registered in the dictionary, define an array First, an array Next, and an array Ext that are each composed of N _max elements, and set the initial value “0” for all the elements of these arrays. To set. This disposition First corresponds to the index part shown in FIG. 6, and the array Nex
The t and the array Ext correspond to a list. Therefore, the array
In the i-th component First [i] of First, a number indicating the component of the array Next that is the head of the list corresponding to the node with the reference number i is set. Also, the i-th component Ext of the array Ext
The extended character K of the element of the dictionary indicated by the reference number i is set in [i]. Also, the i-th component Next of the array Next
In [i], a pointer indicating an element corresponding to the “sibling” of the element with the reference number i is set.

Next, the first character K is read, the reference number corresponding to this character K is set in the variable i, and the encoding process is started.

First, as the extended character K, the next character of the input character string is read (step 701), and if there is a character to be read next, an affirmative decision is made in step 702, and dictionary search processing is started.

In this case, the variable i is saved to another variable ω, the initial value “0” is set to the variable j (step 703), and then the array Next indicated by the value of the component First [i] corresponding to the variable i The number of the component of is set to the variable i (step 704).

When it is determined in step 705 that the variable i is not the numerical value “0” (negative determination), the element stored in the corresponding list is set as a candidate element and the search process in this list is started.

In this case, the component indicating the extended character of the corresponding candidate element
Ext [i] is compared with the extended character K (step 706). In the case of negative determination in step 706, step 707
In, the pointer of the next candidate element set in the component Next [i] is set as a new variable i, and the process returns to step 705.
In this way, steps 705 to 707 are repeated to search for the corresponding list.

In step 706, when the component Ext [i] = K (affirmative determination), it is determined that the partial string that matches the input character string is registered in the dictionary, and the process returns to step 701 and the next character is returned. Is read and the character string with this character added is encoded.

On the other hand, when the value of the component First [i] or the component Next [i] corresponding to the variable i is “0”, step 70
It becomes affirmative judgment in 5. In this case, it indicates that other candidate elements linked to the subsequence of reference number i are not registered in the dictionary. In this case, it is determined that the corresponding substring is not registered in the dictionary, and the processing from step 708 onward is performed.

Here, as described above, every time the corresponding partial string is searched from the dictionary, the reference number corresponding to the partial string searched in step 703 is saved in the variable ω. Therefore, the reference number saved in this variable ω indicates the registered substring that matches the input character string the longest, and the code corresponding to this reference number ω is output (step 708), Register new substring.

First, the value of the variable n is set to the variable i, the variable n is incremented, and the component Ex corresponding to the variable i is set.
The extended character K is set in t [i] (step 709).

Next, it is determined whether or not the value of the variable j is "0" (step 710). If the determination is affirmative, the variable i is set in the component First [ω] (step 711), and the reference number ω Defines a list corresponding to. On the other hand, in the case of a negative determination in step 710, the variable i is set in the component Next [j] (step 712) and a new “sibling” is added to the corresponding list.

Thus, after the registration process is completed, the reference number corresponding to the extended character K is set as the variable i (step 71
3) Return to step 701 and repeat the above-mentioned processing. When there are no more characters to be read, a negative judgment is made in step 702, the variable ω at that time is output as a corresponding code (step 714), and the processing ends. To do.

[Problems to be Solved by the Invention]

By the way, in the above-described conventional method, in the list search process, a concatenation determination process for determining whether or not there is a concatenated candidate element, a match detection process for detecting a candidate character that matches the input extended character. , And a reading process for setting the next pointer and reading from the dictionary are sequentially performed. In this way, if the list is sequentially manipulated by software and the number of candidate elements is large,
In particular, it takes a long time to perform a search process for a subsequence. Therefore, the encoding processing speed is about several tens of KB / s, and the transfer speed (several hundred KB / s to several MB / s) to a magnetic tape device or magnetic disk device.
There was a problem that the encoding process could not be performed in real time according to s).

On the other hand, if the data compression apparatus is configured by using each independent element for each step of the above-described encoding processing, the encoding processing can be speeded up, but the circuit scale becomes large and the cost is high. There is a drawback that

Here, in the above-mentioned conventional example, for simplicity, 3
Although the case of encoding a character string made up of characters has been described, the actual character string is composed of many characters. Therefore, normally, in the dictionary search process, the list corresponding to a certain reference number is handed down, the candidate elements corresponding to "siblings" are sequentially read out, the matching element is detected, and the presence / absence of a connected element is detected. Takes the longest time to detect.

The present invention was created in view of such a point, and an object of the present invention is to provide a dictionary search system that searches a dictionary at high speed.

[Means for solving the problem]

 FIG. 1 is a block diagram of the principle of the present invention.

(I) Invention of Claim 1 In the drawing, a character string represented by an input reference number and a character from among different character strings registered in the dictionary 110 corresponding to the respective reference numbers The dictionary 110 in the dictionary search method for searching for is to divide a candidate element obtained by adding a different character as an expansion character to each of the character strings to which reference numbers are given, and divide the candidate element into a plurality of groups based on the expansion place, Corresponding to the reference number and the extended character, the index 111 for storing the reference number given to any of the candidate elements belonging to each group and the identification information corresponding to this candidate element, and the reference number And a list 112 that stores the reference number given to any of the other candidate elements belonging to the same group as the candidate element given the reference number and the identification information corresponding to this candidate element. .

The reading means 121 first outputs the reference number and the identification information stored in the index 111 corresponding to the input reference number and the information regarding the input character, and thereafter, the reference number output by the dictionary 110. In response to, the dictionary 110 is instructed to output the reference number and the identification information stored in the list 112.

The detection means 122 detects a candidate element in which the input character is added as an extended character based on the identification information read from the dictionary 110, and outputs the detection result as a search result.

The determination means 123 determines whether or not there is a candidate element that has not been read based on the reference number read from the dictionary 110, and outputs this determination result as a search result.

As a whole, the reading means 121, the detecting means 122, and the judging means 12
3 and 3 are configured to operate independently.

(Ii) Invention of Claim 2 The reading means 121 in the invention of claim 2 activates the reading operation for the dictionary 110 at fixed time intervals, and in the dictionary search system according to the invention of claim 1, the detecting operation by the detecting means 122. Read-out means 121
The read operation is performed in parallel with.

[Action]

(I) Invention of Claim 1 In the invention of Claim 1, the candidate element obtained by adding a different character to each of the character strings given reference numbers as an expansion character is based on the above-mentioned expansion character. It is divided into multiple groups. In addition, in the index 111,
Corresponding to the reference number and the information about the extended character, the reference number given to any of the above-mentioned candidate elements belonging to each group and the identification information corresponding to this candidate element are stored. Further, in the list 112, corresponding to the reference number, the reference number given to any of the other candidate elements belonging to the same group as the candidate element given this reference number, and the identification information corresponding to this candidate element. And are stored.

The index 111 described above corresponds to the index part of the external hash method, and the list 112 corresponds to the list of the external hash method. Further, the reference numbers stored in the index 111 and the list 112 also serve as a pointer indicating the storage location of the next candidate element, which indicates the connection relationship of the candidate elements belonging to the same group. Further, as the identification information, the extended character in each character string may be stored.

Initially, the reading means 121 instructs the dictionary 110 to output the reference number and the identification information stored in the index 111 corresponding to the input reference number and the information about the input character. Further, thereafter, the reading means 121 causes the dictionary 1
The output of the reference number and the identification information stored in list 112 corresponding to the reference number output by 10 is instructed.

In this way, following the index 111, the identification information (for example, extended characters) corresponding to the candidate elements linked by the above-mentioned pointers is sequentially read from the list 112.

Therefore, the detection means 122 may output that the corresponding candidate element is detected as the search result when the extended character read as the identification information from the dictionary 110 and the input character match.

Further, when the reference number output from the dictionary 110 does not correspond to the character string registered in the dictionary 110, the determining unit 123 determines that there is no unread candidate element, and searches this determination result. You can output it as a result.

In the invention of claim 1, the candidate element corresponding to each reference number is divided into a plurality of groups based on the information regarding the extended character, and the group corresponding to the input reference number and the information regarding the input character is formed. The corresponding character string is detected from the candidate elements to which it belongs. Therefore, it is possible to reduce the number of times the reading unit 121 reads the candidate elements from the dictionary 110 and the number of times the detecting unit 122 performs the matching detection operation, and it is possible to perform the search process at high speed. Further, since the reading means 121, the detecting means 123, and the judging means 124 operate independently of each other, like the conventional method,
It is not necessary to wait for the end of the previous processing, and the dictionary search processing can be speeded up.

(Ii) Invention of Claim 2 In the invention of Claim 2, the reading means 121 performs a reading operation for the dictionary 110 at predetermined time intervals, and the detecting operation by the detecting means 122 and the judging operation by the judging means 123 are performed. , Is performed in parallel with the reading means by the reading means 121.

For example, the reading unit 121 described above may perform the reading operation at each time required for the reading operation from the dictionary 110. Normally, the detection means is compared with the time required for the read operation.
It is considered that the detection operation by 122 and the determination operation by the determination means 123 are completed in a short time. By performing the reading operation and the detection operation and the determination operation in parallel, the search processing can be pipelined and processed.

Therefore, according to the second aspect of the present invention, the number of times the read operation and the match detection operation are performed is reduced, and the read operation and the match detection operation and the determination operation are pipelined to be processed, so that the dictionary 110 It is possible to speed up the character string search processing.

〔Example〕

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 2 shows the configuration of a data compression apparatus to which the dictionary search method according to an embodiment of the present invention is applied.

Here, the correspondence between FIG. 1 and the embodiment will be described.

 The dictionary 110 corresponds to the dictionary 230.

 The index 111 corresponds to the index unit 231.

 The list 112 corresponds to the list unit 232.

The reading unit 121 corresponds to the candidate element holding unit 241, the timing control circuit 244, and the flip-flop (FF) 245a.

The detection means 122 includes an extended character register 261 and a comparison circuit 262.
Is equivalent to.

The determination means 123 corresponds to the NOR circuit (NOR) circuit 243.

Assuming that there is the above correspondence, the configuration and operation of the embodiment will be described below.

In FIG. 2, 201 is a microprocessor (MPU)
202 indicates an input port, 230 indicates a dictionary, 240 indicates a dictionary search circuit, and 205 indicates an output port. The MPU 201, the input port 202, the dictionary 230, the dictionary search circuit 240, and the output port 205 described above are connected to each other via a bus 206.

Also, the character string input via the input port 202 is M
LZW encoded by PU201, through output port 205,
The data is transferred to and accumulated in a magnetic disk device (not shown) or the like.

At this time, the MPU 21 is configured to instruct the dictionary search circuit 240 to search, from the dictionary 230, the part of interest of the input character string (hereinafter referred to as the target partial string).

Here, in the LZW encoding method, a subsequence is represented by a reference number ω corresponding to the subsequence searched up to that point, and the last one character K added as an extension character to this subsequence. Therefore, the dictionary search circuit 240 selects the substring to which the expansion character K is added from the candidate elements obtained by adding one different character as the expansion character to the beginning part of the focused substring except the expansion character K. You can search for.

At this time, the subsequence of interest belongs to a group of candidate elements to which a character having the most significant bit equal to the most significant bit K _m of the above-mentioned extended character K is added as an extended character. Therefore, if the above-mentioned candidate element is divided into two groups based on the most significant bit of the extended character in each, and the candidate element belonging to the group corresponding to the most significant bit K _m is searched, You can search.

The configuration of the dictionary 230 when the candidate elements are divided into two groups and searched based on the most significant bit of the extended character will be described below.

As shown in FIG. 3, the dictionary 230 is composed of 3N consecutive storage areas, and the dictionary 230 outputs the content of the storage area designated by the read address according to the read signal introduced. Is configured to.

Of the 3N storage areas described above, the storage area in which the upper 2 bits of the address are “00” and the storage area in which the address is “01” are the index unit 23 corresponding to the index of the external hash method described above.
Assigned to 1. Further, the storage area in which the upper 2 bits of the address are “10” is assigned to the list unit 232 corresponding to the collection of the list of the external hash method. Further, each storage area of the index unit 231 and the list unit 232 is a pointer unit that stores a pointer indicating a storage location of a candidate element connected to a subsequence whose reference number is the lower portion excluding the upper 2 bits of each address. It is composed of an identification information section that stores identification information corresponding to the above-mentioned candidate elements.

Here, the upper 2 bits of the address of the index unit 231 are "0.
0 "of the region to the pointer unit (hereinafter, referred to as index 231 _00), the portion most significant bit subsequence of the reference numbers the lower part of the address obtained by adding one character is" 0 " The column reference number may be stored as a pointer. The area of the upper two bits of the address "0" to the pointer unit (hereinafter, referred to as index 231 _01), the most significant bit subsequence of the reference numbers the lower part of the address is "1"
The reference number of the partial sequence obtained by adding 1 character is stored as a pointer. In this way, the candidate element obtained by adding one character to the subsequence of each reference number as an extended character is divided into the index part 231 ₀₀ and the index part 231 ₀₁ according to the most significant bit of each extended character. Stored.

In addition, the pointer part of each area of the list part 232 is assigned to the substring indicated by the corresponding reference number, the part excluding the extension character and the most significant bit of the extension character, and the other parts assigned to different substrings. The reference number thus obtained may be stored as a pointer.

Further, the extended character in each candidate element may be stored as identification information in the identification information part of each storage area of the index part 231 and the list part 232.

In this way, the index unit 231 and the list unit 232 show the mutual connection relationship between the candidate elements belonging to the same group in which the most significant bits of the extended characters are the same. For example,
In FIG. 3, the connection relationship between candidate elements (indicated by reference numbers ω ₁ and ω ₂ in the figure) in which an expansion character whose most significant bit is logical “0” is added to the subsequence of reference number “1” and The connection relationship of candidate elements (indicated by reference numbers ω _i and ω _j in the figure) in which an extended character having the highest bit of logic “1” is added to the subsequence of reference number “2” is shown.

Further, an initial value “0” is set in the contents of the storage areas of the index unit 231 and the list unit 232 described above when the encoding process is started.

The dictionary search circuit 240 includes a candidate element holding unit 241 that holds information about candidate elements read from the dictionary 230, and an element whose extended character matches the extended character K input from the MPU 201 among the introduced candidate elements. Match detection unit 242 for detecting
A NOR circuit 243, a timing control unit 244 for controlling the timing of the operation of each of these units, and two flip-flops (FF) 245a and 245b.

In the candidate element holding unit 241 described above, via the bus 206,
The data output by the index unit 231 and the list unit 232 of the dictionary 230 is introduced. The candidate element holding unit 241 is composed of two address registers 251, 252 and a candidate character register 253.

Of the data introduced via the bus 206, the pointer portion is introduced into the address register 251, and the identification information portion is introduced into the candidate character register 253. Also, address register 2
The part of the output of 51 other than the most significant bit is introduced into the address register 252 and is held once.

These registers are configured to store the introduced data in response to the load signal introduced from the timing control circuit 244.

The match detection unit 242 compares the extended character register 261 that stores the input extended character K with the extended character K stored in the extended character register 261 and the candidate character stored in the candidate character register 253 described above. It is composed of a comparison circuit 262. Further, the comparison circuit 262 is configured to output a logical “1” when the above-mentioned extended character K and the candidate character match, and the output of this comparison circuit 262 is sent to the MPU 201 as a match detection signal. Has been introduced.

Further, the NOR circuit 243 includes the address register 25 described above.
The output of 1 is introduced, and the output of this NOR circuit 243 is M
Introduced in PU201.

The timing control circuit 244 is configured to control the FF 245a and generate and output a read signal and a load signal in response to an instruction from the MPU 201. This read signal is input to the dictionary 230 described above, and the load signal is the address register 251, 252 of the candidate element holding unit 241.
Is input to the candidate character register 253.

The output of the FF 245a described above is introduced to the FF 245b and is also supplied to the dictionary 230 as the most significant bit of the read address. The output of the address register 251 of the candidate element holding unit 241 described above is supplied to the dictionary 230 as a lower part of the read address.

Hereinafter, a search operation of the dictionary 230 by the dictionary search circuit 240 will be described.

For example, when performing the encoding processing of the character string shown in FIG. 4 (a), first, the first character "a" is read, and this character "a" is set as the head part of the target subsequence Find the corresponding reference number (eg "1"). After that, the next character “b” is read as an extended character K, and the result obtained by adding the most significant bit of this character “b” to the above-mentioned reference number is input to the address register 251 as a hash address. Further, the above-mentioned character “b” may be input as the expanded character K into the expanded character register 261 and the timing control circuit 246 of the dictionary search circuit 240 may be instructed to start the search operation.

In response to the search start instruction described above, the timing control circuit
The 244 first resets the output of the FF 245a to a logical "0",
Then, the read signal is introduced into the dictionary 230. This allows
A logical "0" is introduced as the most significant bit of the read address, and the index part 231 of the dictionary 230 is selected. In addition, according to the most significant bit of the contents of the address register 251 described above,
Either the index portion 231 ₀₀ and index 231 ₀₁ is selected, the data of the storage area corresponding to the reference number ω of the index portion 231 which is selected is introduced to the dictionary retrieval circuit 240 via the bus 206. In this way, the extended character of the first candidate element and the pointer indicating another candidate element connected to this candidate element are read from the index unit 231 and the candidate character register 253 of the dictionary search circuit 240 is read. Address registers 251 and respectively.

The timing control circuit 244 outputs the above-mentioned read signal, and after the time (read cycle time) τ required for the data read operation from the dictionary 230 elapses, the address register 251, 252 and the candidate character register of the candidate element holding unit 241.
Introduce load signal to 253 and.

In response to this load signal, the address register 251 stores the pointer read from the pointer section of the corresponding storage area of the index section 231. Further, at this time, the timing control circuit 244 sets the logic “1” in the FF 245a. Therefore, thereafter, the most significant bit of the read address becomes a logical “1”, the list section 232 of the dictionary 230 is selected, and the storage area of the list section 232 corresponding to the above-mentioned pointer is changed to
The next candidate element connected to the first candidate element described above is read.

After that, the timing control circuit 244 repeats the operation of outputting the read signal and then outputting the load signal after the lapse of the read cycle time τ described above, unless otherwise instructed by the MPU 201. In this way, the candidate elements connected by the pointer are sequentially read from the list unit 232. Further, the address register 252 holds the reference number read as a pointer immediately before, and the FF 245b holds the most significant bit of the read address designated in the immediately previous read operation.

As described above, the candidate elements belonging to the group corresponding to the most significant bit K _m of the extended character K are sequentially read out every read cycle time τ.

Here, the above-described match detection unit 242 is configured as the candidate element holding unit 2
Operates independently of 41. Therefore, in parallel with the above-described read operation, the comparison circuit 262 of the match detection unit 242 compares the character read previously and stored in the candidate character register 253 with the extended character K. .

As described above, the comparison circuit 262 outputs the logic "1" as the match detection result and inputs it to the MPU 201 when the candidate character and the extended character match. Therefore, when the logical "1" is introduced as the match detection result, the MPU 201 determines that the corresponding subsequence has been searched,
The interrupt processing described below may be performed.

In this case, the MPU 201 first sets the timing control circuit 24
Instruct 4 to stop the read operation and read the reference number held in the address register 252. Here, this reference number corresponds to the partial string read out immediately before, that is, the partial string determined by the match detection unit 242 to match the target partial string. In this case, the MPU 201 reads the next one character of the input character string as a new extended character K, inputs it to the extended character register 261, and stores the most significant bit K _m of this extended character K and the above-mentioned reference number. The address is input to the address register 251, and the dictionary search circuit 240 is instructed to start the search processing of the corresponding partial sequence.

In this way, each time a substring that matches the substring of interest is searched, the next one character is added as an extended character K to the searched substring to extend the substring of interest. The subsequence is searched and the encoding operation is continued.

In addition, the NOR circuit 243 also operates independently, like the match detection unit 242. Therefore, in parallel with the above-described read operation, the NOR circuit 243 determines whether or not there is a candidate element to be connected depending on whether or not the valid pointer other than the initial value “0” is stored in the address register 251. Is determined.

Therefore, when the NOR circuit 243 outputs a logical "1" as the determination result and it is determined that there is no connected candidate element, the MPU 201 may perform the interrupt process described below.

First, the MPU 201 instructs the timing control circuit 244 of the dictionary search circuit 240 to stop the read operation, and outputs the reference number ω corresponding to the last searched substring as a code. Next, the MPU 201 reads the reference number held in the address register 252 and the output of FF245b, and the FF24b
Registration processing of a new subsequence is performed according to the output of 5b.

For example, if the output of FF245b is logic "0", MUP2
In 01, the most significant bit of the write address is set to “0” and a new partial string is registered in the index unit 231. In this case, the extension string K is added to the substring corresponding to the above-mentioned reference number ω in the storage area of the index part 231 indicated by the most significant bit K _m of the extension character K and the above-mentioned reference number. The reference number ω _n thus obtained may be stored as a pointer, and the extended character K may be stored as identification information. In this way, the first candidate element belonging to the group corresponding to the above-mentioned reference number ω and the most significant bit K _m of the extended character K is registered.

On the other hand, when the output of FF245b is logic "1", MPU201
Registers a new partial string in the list section 232 with the most significant bit of the write address set to "1". In this case, in the storage area of the list unit 232 corresponding to the above-mentioned pointer,
The reference number ω _n described above may be stored as a pointer, and the extended character K may be stored as identification information. In this way, the above-mentioned reference number ω and the most significant bit K _m of the extended character K are
New candidate elements belonging to the group corresponding to and are registered.

After that, the above-described extended character K is set as the head portion of the target substring, and the next one character of the input character string is set as the new extended character K, and the encoding operation may be continued.

As described above, the index part 231 and the list part 232 of the dictionary 230.
By the pointer of each storage area of and, the connection relationship of the candidate elements in which the most significant bit of the extended character in each candidate element is equal, the candidate element according to the most significant bit of the extended character,
Divide into two groups and search.

Here, when the probability that a logic "1" appears as the most significant bit of a character and the probability that a logic "0" appears are the same, the two groups divided as described above have the same number of candidate elements. Are believed to belong. Therefore, by dividing the candidate elements into two groups and performing the search as described above, it is possible to reduce the number of candidate elements to be searched in the search processing to about half, and It is possible to detect at high speed.

Further, the timing control circuit 244 supplies a read signal to the dictionary 230 at every read cycle time τ, and
The read address created based on the pointer read from the dictionary 230 is supplied to the dictionary 230. Also, the match detection unit 242
And the NOR circuit 243 are operated independently of each other, and the dictionary 230
By performing the match detection operation and the connection determination operation in parallel with the read operation from, the connected candidate elements can be searched for without going through the MPU 201, and the read processing, the match detection processing, and the connection processing can be performed. It is possible to process the determination processing and the pipeline processing.

As a result, it is possible to significantly reduce the time required for the candidate element search processing, as compared with the conventional case where all candidate elements are searched through the MPU.

As described above, by using the simple circuit as shown in FIG. 2, it is possible to speed up the search process of the character string from the dictionary 230, and the time required for the search process of the dictionary can be shortened. It is possible to speed up the conversion processing. In this case, the encoding speed can be set to the same level as the transfer speed to the magnetic disk device, and the encoded data can be transferred to the magnetic disk device or the like in real time.

Further, the MPU 201 may perform the above-described interrupt processing according to the outputs of the match detection unit 242 and the NOR circuit 243, and does not need to operate at a high speed.

In addition, in the above-mentioned embodiment, the case where the candidate element is divided into two groups based on the most significant bit of the extended character of each candidate element has been described, but the number of divisions of the group and the method of division are not limited. , It is applicable as long as it is divided into a plurality of groups based on the extended character of each candidate element and is searched for in each group.

For example, the candidate element may be divided into four groups based on the upper 2 bits of the extended character, and the candidate element with many "1" s in the extended character bit pattern and the candidate element with many "0" s may be divided. You may divide. In addition, a logical "1" or a logical "0" is assigned to each of the extended character bit patterns.
It is also possible to prepare a look-up table (LUT) for storing and to divide the output of this LUT into a group having a logical “1” and a group having a logical “0”. In short, the division may be performed so that the number of candidate elements belonging to a plurality of groups is almost the same.

When a part of the extended character is included in the hash address as in the above-described embodiment, only the portion not included in the extended character hash address is stored in the dictionary 230 as the identification information of each candidate element. Just store it. As a result, it is possible to prevent an increase in the capacity of the memory used as the dictionary 230 except for the overlapping portion between the hash address and the identification information.

Further, in the above-described embodiment, the case where the present invention is applied to the data compression device has been described. However, the present invention is not limited to this, and when a dictionary having a tree structure is configured using the external hash method and this dictionary is searched, If applicable, it can be applied.

〔The invention's effect〕

As described above, according to the first aspect of the present invention, the number of times the reading operation and the coincidence detecting operation are performed is reduced, and the reading means, the detecting means, and the determining means are operated independently of each other, whereby the dictionary search processing is performed. Can be speeded up, and the coding process can be speeded up.

According to the second aspect of the present invention, the number of times the read operation and the match detection operation are performed is reduced, and the read operation and the match detection operation and the determination operation are pipelined and processed. Can be further speeded up.

[Brief description of drawings]

FIG. 1 is a block diagram of the principle of the present invention, FIG. 2 is a block diagram of a data compression device according to an embodiment of the present invention, FIG. 3 is a diagram showing a structure of a dictionary according to the embodiment, and FIG. 4 is LZW encoding. FIG. 5 is an explanatory diagram of the method, FIG. 5 is a diagram showing the structure of the dictionary, FIG. 6 is an explanatory diagram of the external hash method, and FIG. 7 is a flow chart showing a conventional encoding operation. In the figure, 110 is a dictionary, 111 is an index, 112 is a list, and 121.
Is a reading means, 122 is a detecting means, 123 is a determining means, 201 is a microprocessor, 202 is an input port, 205 is an output port, 206 is a bus, 230 is a dictionary, 231 is an index part, 232 is a list part, and 240 is a dictionary. A search circuit, 241 is a candidate element holding unit, 242 is a match detection unit, 243 is a NOR (NOR) circuit, 244 is a timing control circuit, 245 is a flip-flop (FF), 251,2
52 is an address register, 253 is a candidate character register, 261 is an extended character register, and 262 is a comparison circuit.

 ─────────────────────────────────────────────────── --- Continuation of the front page (72) Hirotaka Chiba, Inventor Hirotaka Chiba 1015 Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Within Fujitsu Limited (56) References JP-A-2-85927 (JP, A) JP-A-2-132556 (JP, A)

Claims (2)

(57) [Claims] 1. A dictionary for searching a character string represented by an input reference number and a character from different character strings registered in a dictionary (110) corresponding to a reference number given to each. In the search method, the dictionary (110) divides a candidate element obtained by adding one different character as an extended character to each of the character strings given the reference numbers into a plurality of groups based on the extended character. And an index (111) for storing the reference number given to any of the candidate elements belonging to each group and the identification information corresponding to this candidate element, corresponding to the reference number and the information about the extended character. In correspondence with the reference number, the reference number given to any of the other candidate elements belonging to the same group as the candidate element given the reference number and the identification information corresponding to this candidate element are stored. First, the output of the reference number and the identification information stored in the index (111) corresponding to the input reference number and the information about the input character is performed. The output of the reference number and the identification information stored in the list (112) corresponding to the reference number output by the dictionary (110) is output to the dictionary (11
0) and a reading means (121) for instructing, and based on the identification information read from the dictionary (110), a candidate element in which an input character is added as an extended character is detected, and this detection result is used as a search result. Based on the reference number read from the dictionary (110) and the detection means (122) for outputting as a determination result, and a determination for outputting the determination result as a search result. Means (123)
A dictionary retrieval system comprising: and a reading means (121), a detecting means (122), and a judging means (123) which operate independently of each other. 2. The reading means (121) starts a reading operation for the dictionary (110) at a predetermined time interval, and the reading operation by the detecting means (122) and the judging operation by the judging means (123) are performed. 2. The dictionary search system according to claim 1, wherein the reading operation is performed in parallel with the reading operation by the reading means (121).