Calculation apparatus

This invention relates to arithmetic device, especially relates to arithmetic device which outputs the arithmetic result according to input conditions.

Common arithmetic device has build-in arithmetic-logic unit which responds to arithmetic program when input exist, and then acquires the output.

Nevertheless, common arithmetic device need repeat the arithmetic corresponding to arithmetic equation according to arithmetic program, so the arithmetic speed is low.

In general, for example, when program deals with CASE declaration, CPU compares conditional declaration temporarily and outputs consistence result after checking whether conditional declaration is consistent with input. Therefore, when there are n conditions, it requires at most n comparisons of conditional declaration.

Therefore, because CPU need execute CASE declarations in turn, common program algorithms have disadvantages that, for example, processing time lengthens, etc. In particular, when inconsistent CASE declarations in conditional equation repeat again and again, comparison times of conditional branches goes up, so processing speed goes down on a large scale.

In view of the above disadvantages, this invention aims to develop an arithmetic device which can execute high-speed processing.

Arithmetic device in this invention exports arithmetic result corresponding to the input, including: memory means that stores the arithmetic result corresponding to input at the address corresponding to input, and control device which refreshes the storage synchronously with the arithmetic.

According to this invention, arithmetic result can be acquired as soon as input is read to address, which can execute high-speed processing. Besides, to refresh arithmetic result synchronously with the arithmetic can deal with a variety of arithmetic equations with small memory capacity.

Besides, through memory means by which output are corresponding to data address, arithmetic device of this invention stores arithmetic equation which is selected according to appearance ratio. Input of arithmetic equation is corresponding to address, and output corresponding to data.

According to this invention, mnemonic stores arithmetic equation according to frequency, so the arithmetic equation selected by learning can execute arithmetic.

FIGURES EXPLANATION

FIG. 1 is the block diagram of implementation example 1 of this invention.

FIG. 2 is the input structure of implementation example 1 of this invention.

FIG. 3 is table data structure of implementation example 1 of this invention.

FIG. 4 is mnemonic data structure of implementation example 1 of this invention.

FIG. 5 is the processing flow diagram of read control module of implementation example 1 of this invention.

FIG. 6 is the processing flow diagram of write control module of implementation example 1 of this invention.

FIG. 7 explains the application example of implementation example 1 of this invention.

FIG. 8 is the block diagram of implementation example 2 of this invention.

FIG. 9 explains the application example of implementation example 2 of this invention.

FIG. 10 (A)˜(C) are the block diagrams of application examples of this invention.

FIG. 11 is the block diagram of implementation example 3 of this invention.

FIG. 12 is driver functional block diagram of implementation example 3 of this invention.

FIG. 13 is the data structure of arithmetic equation table implementation example 3 of this invention.

FIG. 14 is the data structure of statistic table of implementation example 3 of this invention.

FIG. 15 is the block diagram of arithmetic board of implementation example 3 of this invention.

FIG. 16 is mnemonic data structure of implementation example 3 of this invention.

FIG. 17 is the flow diagram of monitoring processing of implementation example 3 of this invention.

FIG. 18 is the flow diagram of learning processing of implementation example 3 of this invention.

FIG. 19 is the flow diagram of data refresh processing of implementation example 3 of this invention.

FIG. 20 is the flow diagram of extrapolation processing of implementation example 3 of this invention.

FIG. 21 is the flow diagram of extrapolation processing of implementation example 3 of this invention.

FIG. 22 is the flow diagram of interpolation processing of implementation example 3 of this invention.

FIG. 23 is the flow diagram of approximation processing of implementation example 3 of this invention.

FIG. 24 is the flow diagram of fixation processing of implementation example 3 of this invention.

FIG. 25 is the flow diagram of by-pass processing of implementation example 3 of this invention.

FIG. 26 is the flow diagram of feedback processing of implementation example 3 of this invention.

FIG. 27 is the block diagram of modified example of arithmetic board of implementation example 3 of this invention.

FIG. 28 is the block diagram of other modified example of arithmetic board of implementation example 3 of this invention.

FIG. 1 is the block diagram of implementation example 1 of this invention.

Arithmetic device 1 of this implementation example consists of mnemonic 2 , write control module 3 , read control module 4 and table 5 . Arithmetic device 1 of this implementation example sends out reconfiguration instruction every time arithmetic device 1 has input, and mnemonic 2 remaps according to table 5 . At this time, to suppress reconfiguration instruction according to input through write control module 3 and read control module 4 can inhibit remap.

Mnemonic 2 is composed of free read/write mnemonic. Mnemonic 2 writes arithmetic result through write control module 3 , and reads the stored arithmetic result through read control module 4 .

According to instructions of read control module 4 , write control module 3 reads out arithmetic result from table 5 and stores in mnemonic 2 .

Equation input is fed to read control module 4 through input end Tin. Input is fed to read control module 4 as address of mnemonic 2 . Mnemonic 2 reads out the data corresponding to address of read control module 4 as arithmetic result.

In addition, write control module 3 and read control module 4 are composed of programs and are stored in computer where arithmetic device of this implementation example is installed. Besides, table 5 is formed in permanent storage.

Here, explain the input in detail.

FIG. 2 is the input structure of implementation example 1 of this invention.

Input is the same as address of mnemonic 2 . Input consists of upper address ADDi 0 and lower address ADD 0 j . Read control module 4 feeds input lower address ADD 0 j as address to mnemonic 2 . Besides, read control module 4 controls write control module 3 according to upper address ADDi 0 .

According to instructions of write control module 3 , write control module 3 , from table 5 , reads out the data in memory block Ai corresponding to upper address ADDi 0 and writes into mnemonic 2 .

Here, explain data structure of table 5 in detail.

FIG. 3 is table data structure of implementation example 1 of this invention.

Table 5 consists of m memory blocks A 1 ˜Am. Among blocks A 1 ˜Am, Ai has the same data capacity as mnemonic 2 .

In memory block A 1 are stored n data D 1 l ˜D 1 n corresponding to upper address ADD 10 . In memory block A 2 are stored n data D 2 l ˜D 2 n corresponding to upper address ADD 20 . Likewise, data are stored in each memory block till memory block Am.

Data D 1 l -D 1 n , D 2 l -D 2 n , . . . Dml-Dmn stored in table 5 are the arithmetic results of arithmetic equation with respective input and are stored after prior arithmetic.

Among data D 1 l -D 1 n , D 2 l -D 2 n , . . . Dml-Dmn stored in table 5 , data Dil-Din in memory block Ai are stored in mnemonic 2 .

Next, explain data structure of mnemonic 2 in detail.

FIG. 4 is mnemonic data structure of implementation example 1 of this invention.

Mnemonic 2 stores data Dil-Din at address ADD 0 l ˜ADD 0 n . Mnemonic 2 stores data according to the lower address ADD 0 j fed by read control module 4 .

Next, explain the actions of read control module 4 in detail.

FIG. 5 is the processing flow diagram of read control module of implementation example 1 of this invention.

Read control module 4 feeds input ADDij through input end Tin at step S 1 - 1 , and at step S 1 - 2 , checks whether data Dij stored in mnemonic 2 is just the data requested by input ADDij.

If data stored in mnemonic 2 is just the data requested by input ADDij at step S 1 - 2 , then at step S 1 - 5 , data D 0 j stored in mnemonic 2 at address ADD 0 j are read out. If the data requested by input ADD(−) don't exist in data stored in mnemonic 2 at step S 1 - 2 , then at step S 1 - 3 , write control module 3 is instructed to write the data string including Dij requested by input ADDij into mnemonic 2 . According to instructions of read control module 4 , write control module 3 writes the requested data string into mnemonic 2 from table 5 as explained below. After write control module 3 refreshes the data string of mnemonic 2 , message of refresh finished is fed to read control module 4 .

Step S 1 - 4 is to check whether data in mnemonic 2 have been refreshed. According to message of write control module 3 , if it is determined that the data in mnemonic 2 have been refreshed at step S 1 - 4 , then at step S 1 - 5 , data D 0 j stored in mnemonic 2 at address ADD 0 j are read out.

Through the above procedure, the arithmetic results of requested input are acquired.

Next, explain the actions of write control module 3 in detail.

FIG. 6 is the processing flow diagram of write control module of implementation example 1 of this invention.

At step S 2 - 1 , write control module 3 checks whether read control module 4 sends out refresh instruction; if no refresh instruction, end up processing directly.

Besides, if read control module 4 sends out refresh instruction at step S 2 - 1 , then at step S 2 - 2 , reads out from table 5 the data string corresponding to upper address ADDi 0 of read control module 4 message and at step S 2 - 3 , stores into mnemonic 2 . After write control module 3 writes the data into mnemonic 2 , message of refresh finished is fed to read control module 4 .

Through the above procedure, the data string corresponding to input ADDij are wrote into mnemonic 2 .

Through the above procedure, write the data string corresponding to input ADDij into mnemonic 2 .

In addition, check up FALSE and TRUE of arithmetic result. When arithmetic result is FALSE, read out from table 5 next data string and store into mnemonic 2 .

Here, explain the application example of this implementation example.

FIG. 7 explains the application example of implementation example 1 of this invention.

This is an application example of arithmetic executed by M programs P 1 -PM with arithmetic device 1 of this implementation example.

Program P 1 performance is applied after mnemonic 2 stores arithmetic result of program P 1 . Program P 2 performance is applied after mnemonic 2 stores arithmetic result of program P 2 . Likewise, program PM performance is applied after mnemonic 2 stores arithmetic result of program PM.

In addition, use a mnemonic system to get arithmetic result in this implementation example; it is also feasible to use multiple systems.

Besides, data stored in mnemonic 2 are refreshed according to input in this implementation example; it is also feasible to give priority to assigning arithmetic result of the most used arithmetic equation to mnemonic 2 after arithmetic result is output statistically. This structure may reduce data shifting of mnemonic 2 to the least. In addition, write control module 5 and table 5 may be deleted, arithmetic result of the most used arithmetic equation may be only stored in mnemonic 2 in advance, and other arithmetic equations are to be processed by programs and CPU. This structure may execute efficient arithmetic with small memory capacity.

FIG. 8 is the block diagram of implementation example 2 of this invention. Assign the same symbols to the same parts as FIG. 1 , leaving out explanation.

Arithmetic device 10 of this implementation example consists of N mnemonics 2 - 1 ˜ 2 -N, write control module 11 , read control module 12 and table 5 .

Write control module 11 controls data write into N mnemonics 2 - 1 ˜ 2 -N.

Read control module 12 reads out arithmetic result from N mnemonics 2 - 1 ˜ 2 -N according to input.

FIG. 9 explains the application example of implementation example 2 of this invention.

In this implementation example, arithmetic result of program P 1 is wrote into mnemonic 2 - 1 , arithmetic result of program P 2 is wrote into mnemonic 2 - 2 . Likewise, arithmetic result of program PN is wrote into mnemonic 2 -N.

Arithmetic result is exported from mnemonic 2 - 1 in program P 1 application, arithmetic result is exported from mnemonic 2 - 2 in program P 1 application. Likewise, arithmetic result is exported from mnemonic 2 -N in program PN application.

Thus, N programs P 1 -PN can be executed without refreshing data in mnemonics 2 - 1 ˜ 2 -N. Therefore, time need not be changed and execution can be high-speed.

Accoding to this implementation example, arithmetic results can be exported in parallel when multiple programs are applied.

In addition, data stored in mnemonic are acquired by table in the above implementation examples 1 and 2. Yet it is also feasible to write arithmetic result into mnemonic after mnemonic data are calculated by CPU in turn, thus table can be omitted.

In addition, here explain the application examples of arithmetic devices 1 and 10 of implementation example 1 and 2.

FIG. 10 is the block diagram of application examples of this invention. FIG. 10(A) is the application example included in CPU, FIG. 10(B) is the application example connecting with mnemonic collecting bar, and FIG. 10(C) is the application example connecting with PCI collecting bar.

Arithmetic devices 1 and 10 which can be combined in CPU 21 are demonstrated in FIG. 10(A) and so can directly connect with internal collecting bar 22 in CPU 21 , therefore, arithmetic devices 1 and 10 can read/write with high clock and operate with high speed.

Besides, arithmetic devices 1 and 10 may also be composed of wafer different from CPU 31 , and connect with mnemonic 32 to operate. With this structure, big area wafer can enlarge memory capacity of mnemonic 2 . Therefore, adaptation to arithmetic result can be reduced and arithmetic devices 1 and 10 can operate with high speed.

In addition, it is also feasible to connect arithmetic devices 1 and 10 with PCI collecting bar 43 set by CPU 41 through bridge circuit 42 . With this structure, arithmetic devices 1 and 10 can be installed in PC board, which easily adapts to personal computer on market; memory capacity of mnemonic 2 in the form of circuit board can also be enlarged. In addition, explain below how to install them in PCI board in detail.

PCI board which realizes arithmetic device 1 of this implementation example consists of mnemonic and PCI target collecting bar controller. For example, mnemonic is composed of RAM which has 4096-word address space and 12-bit collecting-main-width bus. PCI target collecting bar controller is the interface of PCI collecting bar and mnemonic.

When editing algorithm of PCI board mnemonic, right item(input) of arithmetic equation are set as mnemonic address, and left item(arithmetic results) of arithmetic equation are stored in mnemonic. When executing algorithm, right item(input) of arithmetic equation are fed to mnemonic address, and left item(arithmetic results) of arithmetic equation are exported from mnemonic.

In addition, explain the processing in detail.

For example, explain how to deal with the algorithm below. Switch(a)

In this case, mnemonic address An consistent with CASE declaration condition b[n] are edited. In execution, arithmetic result x[n] can be acquired from mnemonic with declaration condition b[n] input at mnemonic address.

With such processing, when dealing with the above CASE declaration of n branches, as long as write declaration condition b[n] is wrote to mnemonic address, arithmetic result x[n] can be exported. Therefore, arithmetic result can be acquired with high speed. Besides, hardware is used together with conditional branches to operate, so responses are fixed and can be processed instantaneously.

As explained above, according to this invention, arithmetic result, through writing input to address, can be exported in no time to accelerate the processing. Besides, to adapt arithmetic result synchronously with the arithmetic may deal with a variety of arithmetic equations with small memory capacity.

Next, explain implementation example 3 of this invention.

FIG. 11 is the block diagram of implementation example 3 of this invention.

Arithmetic device 100 of this implementation example may be realized by personal computer. Arithmetic device 100 consists of driver 102 , arithmetic equations table 103 , statistic table 104 and arithmetic board 105 .

Software driver 102 is installed in personal computer, and operates through personal computer CPU. Driver 102 operates according to input of application program 111 .

FIG. 12 is driver functional block diagram of implementation example 3 of this invention.

Driver 102 consists of monitoring processing 121 , learning processing 122 , extrapolation processing 123 , by-pass processing 124 and feedback processing 125 .

Monitoring processing 121 is to monitor the input of application program 111 . Learning processing 122 is to make arithmetic board 105 store arithmetic equation through learning. Extrapolation processing 123 is to extrapolate the output when output corresponding to input doesn't exist. By-pass processing 124 is to make arithmetic board 105 store arithmetic equation through application program 111 . Feedback processing 125 is to make arithmetic board 105 export feedback to input.

Access to arithmetic board 105 according to input of arithmetic board 105 . Output are acquired from arithmetic board 105 and then are fed to application program 111 . Besides, driver 102 fabricates statistic table 104 according to input of arithmetic board 105 , and makes arithmetic board 105 store data in arithmetic equations table 103 according to statistic table 104 .

As for multiple arithmetic equations f 1 ˜fn, arithmetic equations table 103 stores respective output corresponding to input.

FIG. 13 is the data structure of arithmetic equation table of implementation example 3 of this invention.

Statistic table 104 is to manage the respective appearance ratio of multiple arithmetic equations f 1 -fn.

Arithmetic equations table 103 is shown in FIG. 12 , it stores output d 1 ˜dm corresponding to input A 1 ˜Am as to multiple arithmetic equations f 1 ˜fn.

Statistic table 104 manages the appearance frequencies e 1 ˜en of multiple arithmetic equations f 1 ˜fn. Appearance frequencies e 1 ˜en are corresponding to appearance times of arithmetic equations f 1 ˜fn, being added [1] after every arithmetic. In addition, add the result multiplied by coefficients [0.5] or [0.4] to rarely used arithmetic equations. Besides, add the result multiplied by coefficients [1.5] or [1.4] to frequently used arithmetic equations or important arithmetic equations.

For example, arithmetic board 105 is a device which connects PCI collecting bar, ISA collecting bar, USB and PCMCIA card port of personal computer, and exports the arithmetic result of requested arithmetic equation according to input of driver 102 .

FIG. 15 is the block diagram of arithmetic board of implementation example 3 of this invention.

Arithmetic board 105 consists of internal interface 131 , mnemonic controller 132 , mnemonic 133 and external interface 134 . Internal interface 131 is the interface of the above collecting bars of personal computer. Mnemonic controller 132 controls data read/write of mnemonic 133 according to instructions from internal interface 131 . Mnemonic 133 stores the data of arithmetic equation fx among multiple arithmetic equations f 1 ˜fn stored in arithmetic equations table 103 . Besides, mnemonic 133 sets input of application program 111 as address, and reads out data according to input at the set address.

FIG. 16 is mnemonic data structure of implementation example 3 of this invention.

Mnemonic 133 is shown in FIG. 16 , it stores data d 1 ˜dm at address A 1 ˜Am. Address A 1 ˜Am are corresponding to input of arithmetic equation fx, and stored data d 1 ˜dm are corresponding to output of arithmetic equation fx

External interface is the interface of external devices.

Next, explain the actions of driver 102 in detail.

Firstly, explain monitoring processing 121 of driver 102 .

FIG. 17 is the flow diagram of monitoring processing of implementation example 3 of this invention.

Monitoring processing 121 includes steps S 11 - 1 ˜S 11 - 3 .

Step S 11 - 1 is to monitor input of application program 111 . Step S 11 - 2 is to check whether input of application program 111 is the requested input by arithmetic.

At step S 11 - 2 , step S 11 - 3 is executed if input of application program 111 is just the requested input by arithmetic. Step S 11 - 3 is to start learning processing 122 and extrapolation processing 123 .

Besides, at step S 11 - 2 , step S 11 - 4 is executed if input of application program 111 is not the requested input by arithmetic.

Step S 11 - 4 is to check whether input of application program 111 is the requested input by by-pass processing 124 . At step S 11 - 4 , step S 11 - 5 is executed if input of application program 111 is just the requested input by by-pass processing 124 . Step S 11 - 5 is to start by-pass processing 124 .

Besides, at step S 11 - 4 , step S 11 - 6 is executed if input of application program 111 is not the requested input by by-pass processing 124 . Step S 11 - 6 is to check whether input of application program (−)1 is the requested input by feedback processing 12 .

At step S 11 - 6 , step S 11 - 7 is executed if input of application program 111 is just the requested input by feedback processing 125 . Step S 11 - 7 is to start feedback processing 125 .

FIG. 18 is the flow diagram of learning processing of implementation example 3 of this invention.

Learning processing 121 includes steps S 12 - 1 ˜S 12 - 4 .

Step S 12 - 1 is to monitor output. Step S 12 - 2 is to check whether monitor result and output of step S 12 - 1 are [TRUE].

At step S 12 - 2 , step S 12 - 3 is executed if output of step S 12 - 1 is determined to be [TRUE]. Step S 12 - 3 is to feed output to statistic table 104 . For example, when output arithmetic equation according to application program 111 input is as the arithmetic equation fx shown in FIG. 4 , appearance ratio dx in FIG. 4 is changed into (dx+1).

Next, explain data refresh processing.

FIG. 19 is the flow diagram of data refresh processing of implementation example 3 of this invention.

Data refresh processing in step S 12 - 4 includes steps S 13 - 1 ˜S 13 - 7 .

Step S 13 - 1 is to set variable n to be [1]. Step S 13 - 2 is to refer to statistic table 104 . Step S 13 - 3 is to check whether appearance ratio dn of arithmetic equation fn in statistic table 104 is under set value D.

At step S 13 - 3 , step S 13 - 4 is executed if appearance ratio dn is under the set value D, that is, application frequency of arithmetic equation fn is small. Step S 13 - 4 is to delete arithmetic equation fn from statistic table 104 .

At step S 13 - 3 , step S 13 - 5 is directly executed if appearance ratio dn of arithmetic equation fn is over the set value D. Step S 13 - 5 is to set variable n to be (n+1)

Step S 13 - 6 is to check whether variable n is the prior set constant N. At step S 13 - 6 , come back to step S 13 - 2 and repeat steps S 13 - 2 ˜S 13 - 5 if variable n is not the constant N. At step S 13 - 6 , step S 13 - 7 is executed if variable n is not the constant N.

Step S 13 - 7 is to write data of arithmetic equation fxmax which has the maximum appearance ratio dx in statistic table 104 into mnemonic 133 .

Through the above procedure, the probability of adapting data in mnemonic 133 is greatly reduced when input of application program exists.

Next, explain extrapolation processing 123 .

FIG. 20 is the flow diagram of extrapolation processing of implementation example 3 of this invention.

Extrapolation processing 123 includes steps S 14 - 1 ˜S 14 - 3 .

Step S 14 - 1 is to monitor output. Step S 14 - 2 is to check whether monitor result and output of step S 14 - 1 are [FALSE]. At step S 14 - 2 , step S 14 - 3 is executed if output of S 14 - 1 is determined to be [FALSE]

Explain the extrapolation processing.

FIG. 21 is the flow diagram of extrapolation processing of implementation example 3 of this invention.

Extrapolation processing includes steps S 15 - 1 ˜S 15 - 7 .

Step S 15 - 1 is to check whether is input of other arithmetic equations. Step S 15 - 2 is executed if it is not input of other arithmetic equations.

Step S 15 - 2 is to check whether output fx(m+1) is in mnemonic 133 . At step S 15 - 2 , step S 15 - 3 is executed if output fx(m+1) is determined to be in mnemonic 133 .

Step S 15 - 3 is to check whether output fx(m−1) is in mnemonic 133 . At step S 15 - 3 , step S 15 - 4 is executed if output fx(m−1) is determined to be in mnemonic 133 , that is, output fx(m+1) and fx(m−1) are determined to be in existence.

Step S 15 - 4 is interpolation processing which interpolates output fx(m+1) and fx(m−1) to acquire output fx(m).

At step S 15 - 2 , step S 15 - 5 is executed if output fx(m+1) is determined to be not in existence. Step S 15 - 5 is to check whether output fx(m−1) exists.

At step S 15 - 5 , step S 15 - 6 is executed if output fx(m−1) is determined to be in existence, that is, only output fx(m−1) is determined to be in existence. Besides, at step S 15 - 3 , step S 15 - 6 is executed if output fx(m−1) is determined to be not in existence, that is, only output fx(m+1) is determined to be in existence.

Step S 15 - 6 is approximation processing which approximately processes output fx(m+1) or fx(m−1) in existence and output fx(m).

At step S 15 - 5 , step S 15 - 7 is executed if output fx(m−1) is determined to be not in existence. Step S 15 - 7 is fixation processing which sets output fx(m) as prior fixed output fx(M).

Next, explain interpolation processing in step S 15 - 4 .

FIG. 22 is the flow diagram of interpolation processing of implementation example 3 of this invention.

Interpolation processing includes steps S 16 - 1 ˜S 16 - 2 .

Step S 16 - 1 is to read output fx(m−1) and fx(m+1) from mnemonic 133 . Step S 16 - 2 is to export fx(m) after output fx(m−1) and fx(m+1) are placed in equation (1) below.

{ fx ( m− 1)+ fx ( m+ 1)}/2   (1)

Interpolation processing of this implementation example sets the intermediate value of output fx(m−1) and fx(m+1) to be output fx(m).

Next, explain approximation processing in step S 15 - 6 .

FIG. 23 is the flow diagram of approximation processing of implementation example 3 of this invention.

Approximation processing includes steps S 17 - 1 and S 17 - 2 .

Step S 17 - 1 is to read output fx(m−1) or fx(m+1) from mnemonic 133 . Step S 17 - 2 is to export output fx(m−1) or fx(m+1) as output of fx(m).

Next, explain fixation processing in step S 15 - 7 .

FIG. 24 is the flow diagram of fixation processing of implementation example 3 of this invention.

Fixation processing includes steps S 18 - 1 .

Step S 18 - 1 is to set the prior constant x(M) to be output fx(m), which may fix output fx(m).

Next, explain by-pass processing 124 .

By-pass processing 124 is to by-pass learning processing 23 and directly write arithmetic equation fA of application program 111 into mnemonic 133 .

FIG. 25 is the flow diagram of by-pass processing of implementation example 3 of this invention.

By-pass processing includes steps S 19 - 1 ˜-S 19 - 3 .

Step S 19 - 1 is to write arithmetic equation fA from application program 111 . Step S 19 - 2 is to write arithmetic equation fA from application program 111 into mnemonic 133 .

Step S 19 - 3 is to write data of application program 111 into mnemonic 133 , and feed mnemonic 133 output to application program 111 . Besides, at this time, learning processing is in halted state.

Next, explain feedback processing 125 .

FIG. 26 is the flow diagram of feedback processing of implementation example 3 of this invention.

Feedback processing 125 includes steps S 20 - 1 ˜-S 20 - 4 .

Step S 20 - 1 is to determine to feed all of mnemonic 133 output data back to mnemonic 133 address, or feed part of mnemonic 133 output data back to mnemonic 133 address. Step S 20 - 1 determines according to instructions from application program 111 .

At step S 20 - 1 , step S 20 - 2 is executed if application program 111 instructs to feed all of mnemonic 133 output data back to mnemonic 133 address. Step S 20 - 2 is to instruct arithmetic board 105 to feed all of mnemonic 133 output data back to mnemonic 133 address.

At step S 20 - 1 , step S 20 - 3 is executed if application program 111 instructs to feed part of mnemonic 133 output data back to mnemonic 133 address. Step S 20 - 3 is to instruct arithmetic board 105 to feed part of mnemonic 133 output data back to mnemonic 133 address.

After all or part of mnemonic 133 output data are instructed to feedback to mnemonic 133 address at step S 20 - 2 and step S 20 - 3 , step S 20 - 4 is executed. Step S 20 - 4 is to feed data of application program 111 to arithmetic board 105 .

At step S 20 - 2 , if arithmetic board 105 instructs to feed all of mnemonic 133 output data back to mnemonic 133 address, output of mnemonic controller 132 and mnemonic 133 can both be controlled as address input of mnemonic 133 . Therefore, dynamic arithmetic may be high-speed.

In addition, this implementation example explains mnemonic 133 is set to be one segment, yet multiple segments is also feasible

FIG. 27 is the block diagram of modified example of arithmetic board of implementation example 3 of this invention. Assign the same symbols to the same parts as FIG. 15 , leaving out explanations.

Arithmetic board 140 of this modified example connects n mnemonics 133 - 1 ˜ 133 -n with mnemonic controlled 132 , so arithmetic equations can be wrote into n mnemonics 133 - 1 ˜ 133 -n in parallel.

According to this modified example, n mnemonics 133 - 1 ˜ 133 -n can store arithmetic equations in parallel.

FIG. 28 is the block diagram of other modified examples of arithmetic board of implementation example 3 of this invention. Assign the same symbols to the same parts as FIG. 15 , leaving out explanations.

Arithmetic board 150 of this modified example is formed by n mnemonics 133 - 1133 -n in series with data output as address input.

According to this modified example, one input can generates arithmetic results of n arithmetic equations

As explained above, in this invention, arithmetic equation is stored in mnemonic according to frequency, so arithmetic can be executed by arithmetic equation selected by learning.

Besides, this invention is not confined to the above examples; a variety of modified examples within this invention scope can adapt to this invention.