JP2018019138A - Information processing apparatus, information processing system and information processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an information processing apparatus for encoding data represented by a floating-point number by dividing it into a formal value having a required accuracy and a formal value having a higher accuracy. SOLUTION: A separation means of an information processing apparatus separates data in a floating point number representation into an exponential value and a formal value, and a difference means extracts a difference between a predetermined exponential value and an exponential value of each data. , The dividing means divides the formal value of each data into the upper first formal value and the lower second formal value according to the difference, and the first coding means is the predetermined exponential value. And the first formal value are encoded, and the second coding means encodes the difference of each data and the second formal value. [Selection diagram] Fig. 1

Description

  The present invention relates to an information processing apparatus, an information processing system, and an information processing program.

  In Patent Document 1, a floating-point format digital signal sample X in which each sample has a polarity, an 8-bit exponent part, and a 23-bit mantissa part is rounded by an integer conversion unit and converted into an integer-format digital signal sample Y. The sequence of the digital signal samples Y is losslessly encoded by the compression unit to output a code string Ca, the digital signal samples Y are converted into the digital signal samples X ′ in the floating point format by the floating point conversion unit, and the digital signal samples X ′ And a digital signal sample X are obtained by a subtracting unit, the difference signal is losslessly encoded, and a code string Cb is output.

  Patent Document 2 discloses that even when the exponent value of a floating-point data value (the data value intended for lossless compression) is different from the exponent value of a predicted value used for encoding, the compression rate is not reduced. The purpose is to enable lossless compression (encoding) of data values, and the encoding device is a floating-point number value represented by using an exponent value and a mantissa value and is a target value to be encoded. A data value storage unit for storing a certain data value, and a data related value storage unit for storing a data related value which is a floating point number value used for encoding the data value stored in the data value storage unit; Adjusting the mantissa value of the data related value based on the difference between the exponent value of the data related value stored in the data related value storage unit and the exponent value of the data value stored in the data value storage unit. Adjusted mantissa raw to generate adjusted mantissa A specific exponent value based on the exponent value of the data value stored in the data value storage unit, and the adjustment value generated by the mantissa value of the data value and the adjustment mantissa value generation unit A code that generates a specific mantissa value based on a numerical value, and generates an encoding target value that is a floating-point value and a value to be encoded based on the specific exponent value and the specific mantissa value An encoding target value generation unit and a data encoded value generation unit that generates a data encoded value by encoding the encoding target value generated by the encoding target value generation unit are disclosed.

  In Patent Document 3, a method and apparatus for compressing signal samples uses block floating-point representation, and the number of bits per mantissa is determined by the sample of the maximum absolute value in the group. Define a group of signal samples with a fixed number of samples per group, where the largest absolute value sample in the group defines an exponent value that corresponds to the number of bits to represent the largest sample value, and the exponent value is coded Can form exponent tokens, and the exponent differences between successive exponent values can be coded individually or jointly, with the samples in the group mapped to corresponding mantissas, each mantissa being an exponent By having the number of bits based on the value and removing the LSB depending on the exponent value, a mantissa with fewer bits is generated and the feedback control is compressed It is disclosed that monitors the Bit Rate and / or quality requirements standards.

  Patent Document 4 relates to data compression and decompression, and more specifically, to provide a method and apparatus for compressing and decompressing signal data, and obtaining signal data, Determine the block length of multiple data blocks into which data is divided, determine the index of the data block, use the index of the data block to compress the signal data contained in the data block For compressing signal data, including forming a mantissa sequence of data blocks and forming a compressed data block by using the block length, exponent, and mantissa sequence of the data block By constructing a variable length data block adapted to the dynamic characteristics of the signal data Method for compressing signal data to increase the compression ratio of the signal data is disclosed.

  Patent Document 5 provides an information transmission apparatus that performs encoding that does not require a constant for decoding by a decoder, performs efficient compression encoding, and can easily control the amount of data. For the purpose, the information transmission apparatus divides the input signal into a plurality of bands on the frequency axis and stores it for a certain period of time, separates the stored signal into an exponent part and a mantissa part, and encodes each of them by encoding them. An output band division encoding device, and an output signal of the band division encoding device, and a band division decoding device for decoding the original signal and decoding the original signal, respectively, A frequency band divider that divides the input signal into multiple bands on the frequency axis, a memory that stores the output signal of the frequency band divider for a certain period of time, and a maximum value of the signal that is stored for a certain period of time is detected. Effective top An effective high-order digit detector for detecting the constant, a constant storage for storing a predetermined constant for each band, and an exponent part for each band using a predetermined constant and the output of the effective high-order digit detector. An exponent part determiner, a first encoder that calculates a mantissa part of the signal stored for a certain period of time using the exponent part, encodes the mantissa part, and outputs an encoded signal; and a first encoder If the bit amount generated in step 1 is calculated for each band and the predetermined bit amount is exceeded or even if all the signals stored for a certain period of time are not encoded, the encoded truncation signal And a second encoder that outputs a specific code word when the encoded truncation signal is received, and the encoded signal is output until the encoded truncation signal is received. And the output of the first encoder and the second The first encoder outputs a non-equal length code that can be instantaneously decoded, and encodes when an encoded truncation signal is input. Is disclosed.

International Publication No. 2004/098066 Japanese Patent No. 5619326 Special table 2013-508867 gazette JP2013-251895A Japanese Patent Laid-Open No. 03-154443

As a result of processing such as numerical calculation, there is data of floating point number representation. For example, visualization (visualization) and further numerical calculation may be performed on the processing result. The required accuracy may differ between the visualization and the numerical calculation. Specifically, this is a case where low (quantized) accuracy is used for visualization and high accuracy is used for numerical calculation. In the technique described in the above-mentioned patent document, since compression and decompression are performed with an arbitrary accuracy, data can be decoded only when the decompression is performed with the accuracy specified at the time of compression.
The present invention relates to an information processing apparatus, an information processing system, and an information processing system, which are divided into a mantissa with a required accuracy and a mantissa with a higher accuracy when encoding data in a floating-point number representation. The purpose is to provide a program.

The gist of the present invention for achieving the object lies in the inventions of the following items.
The invention according to claim 1 is a separating means for separating floating point expression data into an exponent value and a mantissa value, a difference means for extracting a difference between a predetermined exponent value and an exponent value of each data, and the difference And dividing means for dividing the mantissa of each data into an upper first mantissa and a lower second mantissa, and the first exponent value and the first mantissa are encoded. And a second encoding unit that encodes the difference of each data and the second mantissa.

  The invention according to claim 2 is the information processing apparatus according to claim 1, wherein a predetermined exponent value is the maximum exponent value in the data of a plurality of floating-point numbers to be processed.

  According to a third aspect of the present invention, the dividing means uses, as the second mantissa value, the value of the number of digits in the lower order of the mantissa value of the floating-point number data, and sets the remaining upper digits as the first mantissa value. The information processing apparatus according to claim 1, wherein the information processing apparatus is a numerical value.

  According to a fourth aspect of the present invention, there is provided a second dividing unit that divides the first mantissa value, and a third code that encodes a mantissa value other than the highest mantissa value divided by the second dividing unit. The first encoding means encodes the predetermined exponent value and the most significant mantissa divided by the second dividing means. It is an information processing apparatus as described in any one.

  The invention of claim 5 further comprises prediction error calculation means for calculating an error between the first mantissa value or the second mantissa value and a prediction value, and the first encoding means or the second code. The information processing apparatus according to claim 1, wherein the converting unit encodes the error.

  The invention according to claim 6 is a first decoding means for decoding a predetermined exponent value and a first mantissa value encoded by the first encoding means of the information processing apparatus according to claim 1, 2. The second decoding means for decoding the difference and the second mantissa encoded by the second encoding means of the information processing apparatus according to claim 1, the first decoding means and the second decoding Means for adding the first mantissa value and the second mantissa value in accordance with the data decoded from the means, and conversion means for converting the processing result by the adding means into a floating-point number representation. An information processing apparatus that does not perform processing by the second decoding unit or the adding unit in accordance with required accuracy.

  According to a seventh aspect of the present invention, there is provided a separating means for separating floating point number representation data into an exponent value and a mantissa value, a difference means for extracting a difference between a predetermined exponent value and an exponent value of each data, and the difference And dividing means for dividing the mantissa of each data into an upper first mantissa and a lower second mantissa, and the first exponent value and the first mantissa are encoded. Encoding means, an encoding device having second encoding means for encoding the difference between each data and a second mantissa, and a predetermined encoding encoded by the first encoding means A first decoding means for decoding the exponent value and the first mantissa value; a second decoding means for decoding the difference encoded by the second encoding means and a second mantissa value; In accordance with the data decoded from the second decoding means and the second decoding means. An adder for adding the value and the second mantissa, and a converter for converting the processing result of the adder into a floating-point number representation, according to the accuracy required for the decoded data, An information processing system having a decoding device that does not perform processing by the decoding unit or the adding unit.

  The invention according to claim 8 is a computer that separates floating point representation data into an exponent value and a mantissa value, and a difference unit that extracts a difference between a predetermined exponent value and an exponent value of each data; , Dividing means for dividing the mantissa of each data into an upper first mantissa and a lower second mantissa according to the difference, and encoding the predetermined exponent value and the first mantissa An information processing program for functioning as a first encoding unit that performs the above-described process and a second encoding unit that encodes the difference between each data and the second mantissa.

  According to a ninth aspect of the invention, the computer decodes a predetermined exponent value and a first mantissa value encoded by the first encoding means of the computer on which the information processing program according to the eighth aspect is executed. And a second decoding means for decoding the difference encoded by the second encoding means of the computer executing the information processing program according to claim 8 and the second mantissa value. , Adding means for adding a first mantissa value and a second mantissa value in accordance with data decoded from the first decoding means and the second decoding means, and a processing result by the adding means in a floating-point number representation This is an information processing program that functions as a conversion means for converting to, and does not allow the second decoding means or the adding means to perform processing according to the accuracy required for the decoded data.

  According to the information processing apparatus of the first aspect, when encoding the data in the floating-point number representation, the encoding is performed by dividing the data into a mantissa with a required accuracy and a mantissa with a higher accuracy.

  According to the information processing apparatus of the second aspect, the predetermined exponent value can be the maximum exponent value in the data of a plurality of floating point numbers to be processed.

  According to the information processing apparatus of claim 3, the second mantissa value is a value of the number of lower digits of the difference of the mantissa value of the floating-point data, and the remaining mantissa number is the first mantissa value. Value.

  According to the information processing apparatus of the fourth aspect, the first mantissa value can be divided and encoded.

  According to the information processing apparatus of the fifth aspect, an error from the predicted value can be encoded.

  According to the information processing apparatus of the sixth aspect, depending on the accuracy required for the decoded data, the processing by the second decoding unit or the adding unit can be prevented.

  According to the information processing system of the seventh aspect, when encoding the data of the floating point number representation, it is divided into a mantissa with a required accuracy and a mantissa with a higher accuracy.

  According to the information processing program of the eighth aspect, when encoding the data of the floating point number representation, it is divided into a mantissa with a required accuracy and a mantissa with a higher accuracy.

  According to the information processing program of the ninth aspect, depending on the accuracy required for the decoded data, the processing by the second decoding unit or the adding unit can be prevented.

It is a conceptual module block diagram about the structural example of 1st Embodiment. It is explanatory drawing which shows the system configuration example using this Embodiment. It is a flowchart which shows the process example by 1st Embodiment. It is a flowchart which shows the process example by 1st Embodiment. It is a flowchart which shows the process example by 1st Embodiment. It is explanatory drawing which shows the example of a numerical calculation result. It is explanatory drawing which shows the data structure example of a process result table. It is explanatory drawing which shows the data structure example of a process result table. It is explanatory drawing which shows the example of a data structure of a process result table (block). It is explanatory drawing which shows the example of a data structure of an exponent value table. It is explanatory drawing which shows the example of a data structure of a mantissa table. It is explanatory drawing which shows the example of a data structure of a difference index value table. It is explanatory drawing which shows the example of a data structure of the mantissa table of effective precision. It is explanatory drawing which shows the example of a data structure of the mantissa table of invalid precision. It is explanatory drawing which shows the example of a data structure of a prediction error table. It is explanatory drawing which shows the example of a data structure of the encoding data of a basic code. It is explanatory drawing which shows the data structure example of the encoding data of a difference code. It is explanatory drawing which shows the example of a data structure of the encoding data of a basic code. It is explanatory drawing which shows the example of a data structure of an exponent value table. It is explanatory drawing which shows the example of a data structure of the mantissa table of effective precision. It is explanatory drawing which shows the example of a data structure of an effective precision data table. It is explanatory drawing which shows the example of a data structure of an effective precision data table. It is explanatory drawing which shows the example of a data structure of an exponent value table. It is explanatory drawing which shows the example of a data structure of the mantissa table of effective precision. It is explanatory drawing which shows the example of a data structure of an effective precision data table. It is explanatory drawing which shows the example of a data structure of an effective precision data table. It is a conceptual module block diagram about the structural example of 2nd Embodiment. It is explanatory drawing which shows the process example by 2nd Embodiment. It is explanatory drawing which shows the example of a data structure of the encoding data of the basic code 1. FIG. It is explanatory drawing which shows the example of a data structure of the encoding data of the basic code m (m> = 2). It is a conceptual module block diagram about the structural example of 3rd Embodiment. It is explanatory drawing which shows the process example by 3rd Embodiment. It is a notional module block diagram about the structural example of 4th Embodiment. It is a flowchart which shows the process example by 4th Embodiment. It is a flowchart which shows the process example by 4th Embodiment. It is a block diagram which shows the hardware structural example of the computer which implement | achieves this Embodiment.

Hereinafter, examples of various preferred embodiments for realizing the present invention will be described with reference to the drawings.
<First Embodiment>
FIG. 1 is a conceptual module configuration diagram of a configuration example according to the first embodiment.
The module generally refers to components such as software (computer program) and hardware that can be logically separated. Therefore, the module in the present embodiment indicates not only a module in a computer program but also a module in a hardware configuration. Therefore, the present embodiment is a computer program for causing these modules to function (a program for causing a computer to execute each procedure, a program for causing a computer to function as each means, and a function for each computer. This also serves as an explanation of the program and system and method for realizing the above. However, for the sake of explanation, the words “store”, “store”, and equivalents thereof are used. However, when the embodiment is a computer program, these words are stored in a storage device or stored in memory. This means that control is performed so as to be stored in the apparatus. Modules may correspond to functions one-to-one, but in mounting, one module may be configured by one program, or a plurality of modules may be configured by one program, and conversely, one module May be composed of a plurality of programs. The plurality of modules may be executed by one computer, or one module may be executed by a plurality of computers in a distributed or parallel environment. Note that one module may include other modules. Hereinafter, “connection” is used not only for physical connection but also for logical connection (data exchange, instruction, reference relationship between data, etc.). “Predetermined” means that the process is determined before the target process, and not only before the process according to this embodiment starts but also after the process according to this embodiment starts. Also, if it is before the target processing, it is used in accordance with the situation / status at that time or with the intention to be decided according to the status / status up to that point. When there are a plurality of “predetermined values”, they may be different values, or two or more values (of course, including all values) may be the same. In addition, the description of “do B when A” is used to mean “determine whether or not A and do B when A”. However, the case where it is not necessary to determine whether or not A is excluded.
In addition, the system or device is configured by connecting a plurality of computers, hardware, devices, and the like by communication means such as a network (including one-to-one correspondence communication connection), etc., and one computer, hardware, device. The case where it implement | achieves by etc. is included. “Apparatus” and “system” are used as synonymous terms. Of course, the “system” does not include a social “mechanism” (social system) that is an artificial arrangement.
In addition, when performing a plurality of processes in each module or in each module, the target information is read from the storage device for each process, and the processing result is written to the storage device after performing the processing. is there. Therefore, description of reading from the storage device before processing and writing to the storage device after processing may be omitted. Here, the storage device may include a hard disk, a RAM (Random Access Memory), an external storage medium, a storage device via a communication line, a register in a CPU (Central Processing Unit), and the like.

  The information processing system according to the present embodiment includes an encoding device 100 and a decoding device 150. The encoding device 100 encodes (also referred to as compression) the floating point number data group 105. The decoding device 150 decodes the code (also referred to as decompression) and generates a floating point number data group 195.

In the explanation, terms are defined.
The precision is a digit indicated by the lowest value of the mantissa part in the floating-point number. For example, 1.2345e-05 is 1.2345e-05 = 123445e-09, and the accuracy is e-09.
The mantissa value is an integer value composed of a mantissa part.
The exponent value is an integer value composed of an exponent part.
Effective precision is any precision specified by the user. In the present embodiment, the lowest precision in the block (the largest exponent value in the block of floating point data) is used as the effective precision.
Invalid accuracy is higher than effective accuracy.
The maximum exponent value in the block is the maximum exponent value in the block.
The difference index value is a difference between the maximum index value in the block and the index value.
The difference effective precision mantissa is a difference between the effective precision mantissa and the predicted value.
The basic code is data obtained by encoding a maximum exponent value in a block and a mantissa value of effective precision.
The basic code 1 is data obtained by encoding a mantissa value having the lowest precision among the maximum exponent value in the block and the mantissa values having effective precision divided.
The basic code m (m ≧ 2) is data obtained by encoding a mantissa value other than the lowest accuracy among the mantissa values with effective precision divided into the maximum exponent value in the block.
The basic code a is data obtained by encoding a prediction error between a maximum exponent value in a block and a mantissa value with effective precision.
The difference code is data obtained by encoding a difference exponent value and an invalid precision mantissa.

As shown in the example of FIG. 1, the encoding apparatus 100 includes a block division module 110, a floating-point number separation module 115, an intra-block maximum exponent value calculation module 120, a difference exponent value calculation module 125, a mantissa division module 130, The first encoding module 135 and the second encoding module 140 are provided.
The block division module 110 is connected to the floating point number separation module 115. The block division module 110 receives the floating point number data group 105 and divides the floating point number data group 105 into blocks. Here, the block is a set of data of a predetermined number.
The floating point number separation module 115 is connected to the block division module 110, the intra-block maximum exponent value calculation module 120, and the mantissa division module 130. The floating-point number separation module 115 separates floating-point number representation data into an exponent value and a mantissa value. In the first embodiment, the floating point number data group 105 divided into blocks by the block dividing module 110 is separated into an exponent value and a mantissa value.

The intra-block maximum exponent value calculation module 120 is connected to the floating point number separation module 115 and the difference exponent value calculation module 125. The intra-block maximum exponent value calculation module 120 extracts the maximum exponent value from a plurality of target floating-point data. In the first embodiment, the maximum exponent value (intra-block maximum exponent value) is extracted from the floating-point number data group 105 divided into blocks by the block division module 110.
The difference index value calculation module 125 is connected to the intra-block maximum index value calculation module 120 and the mantissa division module 130. The difference index value calculation module 125 extracts a difference between a predetermined index value and the index value of each data.
Here, as the “predetermined exponent value”, the maximum exponent value in the data of a plurality of target floating-point numbers (the maximum exponent value as a result of processing by the in-block maximum exponent value calculation module 120) Also good. In this case, the difference (difference index value) between the maximum index value in the block and the index value of each data in the block is calculated.

The mantissa division module 130 is connected to the floating point number separation module 115, the difference exponent value calculation module 125, the first encoding module 135, and the second encoding module 140. The mantissa division module 130 divides the mantissa of each data into a higher first mantissa and a lower second mantissa according to the difference that is the processing result of the difference index value calculation module 125. That is, the mantissa value is divided into effective precision and invalid precision. Since the mantissa value is divided into valid precision and invalid precision, the precision of data can be controlled at the time of decoding.
Further, the mantissa division module 130 uses the value of the number of lower-order differences of the mantissa value of the floating-point number data as the second mantissa value, and the value of the remaining upper digits as the first mantissa value. It is good.

The first encoding module 135 is connected to the mantissa division module 130 and outputs a basic code 142. The first encoding module 135 encodes the above-described predetermined exponent value and the first mantissa value. Specifically, the first encoding module 135 encodes the intra-block maximum exponent value and effective significand value of the block.
The first encoding module 135 encodes the predetermined exponent value, the difference, and the first mantissa value when the difference as the processing result by the difference index value calculation module 125 is not 0. Also good.
The second encoding module 140 is connected to the mantissa division module 130 and outputs a difference code 144. The second encoding module 140 encodes the difference between each data and the second mantissa value. Specifically, the second encoding module 140 encodes the difference index value of the block and the invalid precision mantissa.

  As illustrated in the example of FIG. 1, the decoding device 150 includes a first decoding module 155, a second decoding module 160, an addition module 165, a conversion module 170, and a block integration module 175. The decoding device 150 does not cause the second decoding module 160 or the addition module 165 to perform processing according to the accuracy required for the decoded data. That is, when only the data decoded by the first decoding module 155 is used, the processing by the second decoding module 160 and the addition module 165 is unnecessary.

The first decoding module 155 is connected to the addition module 165 and receives the basic code 142. The first decoding module 155 decodes the predetermined exponent value and the first mantissa value encoded by the first encoding module 135 of the encoding device 100. Specifically, the first decoding module 155 decodes the intra-block maximum exponent value and the effective precision mantissa value from the basic code 142.
The second decoding module 160 is connected to the addition module 165 and receives the difference code 144. The second decoding module 160 decodes the difference encoded by the second encoding module 140 of the encoding device 100 and the second mantissa. Specifically, the second decoding module 160 decodes the difference exponent value and the invalid precision mantissa from the difference code 144.

The addition module 165 is connected to the first decoding module 155, the second decoding module 160, and the conversion module 170. The addition module 165 adds the first mantissa value and the second mantissa value according to the data decoded by the first decoding module 155 and the second decoding module 160. That is, the decoding result of the difference code 144 is added to the decoding result of the basic code 142.
The conversion module 170 is connected to the addition module 165 and the block integration module 175. The conversion module 170 converts the processing result by the addition module 165 into a floating-point number representation.
The block integration module 175 is connected to the conversion module 170 and outputs a floating point number data group 195. The block integration module 175 integrates the blocks and generates a floating point number data group 195. That is, the reverse process of the block division module 110 is performed. When both the basic code 142 and the differential code 144 are decoded, the floating point number data group 105 and the floating point number data group 195 are the same (so-called lossless compression / decompression). When only the basic code 142 is decoded, the accuracy of the floating point number data group 195 is less than or equal to the accuracy of the floating point number data group 105 (so-called lossy compression / decompression).

  The encoding apparatus 100 divides input data into arbitrary blocks. Then, the data in the block is separated into effective accuracy and invalid accuracy, and the effective accuracy data is encoded as a basic code, and the invalid accuracy data is encoded as a differential code. At the time of decoding, the decoding device 150 decodes only the basic code according to the user's application to obtain effective precision data (irreversible compression / decompression), or decodes the basic code and the difference code together to obtain invalid precision data. It is also possible to obtain (reversible compression / decompression).

FIG. 2 is an explanatory diagram showing a system configuration example using the present embodiment.
The super computer 210A, the encoding device 100B in the super computer 210B, the encoding device 100A in the information processing device 220, the decoding device 150D in the information processing device 230D, and the decoding device 150E in the information processing device 230E connect the communication line 290. Are connected to each other. The communication line 290 may be wireless, wired, or a combination thereof, and may be, for example, the Internet or an intranet as a communication infrastructure. Further, the function by the super computer 210 may be realized as a cloud service.
The information processing device 220, the information processing device 230A, the information processing device 230B, and the information processing device 230C are connected to each other.
The super computer 210B has an encoding device 100B. The information processing apparatus 220 includes an encoding apparatus 100A. The information processing device 230A includes a decryption device 150A. The information processing device 230B includes a decryption device 150B. The information processing device 230C has a decryption device 150C. The information processing device 230D has a decryption device 150D. The information processing device 230E has a decryption device 150E.

For example, the super computer 210 performs numerical calculation and outputs data in a floating-point number representation. As this data, for example, there is data obtained by simulating atomic motion in a specific environment. As a method of utilizing this data, there are two uses, for example, data visualization and input to another numerical calculation. At this time, the required accuracy may differ between visualization and numerical calculation. Specifically, this is a case where low (quantized) accuracy is used for visualization and high accuracy is used for numerical calculation.
The supercomputer 210 </ b> A transmits the floating-point number data group 105 as a processing result to the information processing apparatus 220. The encoding device 100A of the information processing device 220 performs encoding, and transmits the basic code 142 and the difference code 144 to the information processing device 230A, the information processing device 230B, and the information processing device 230C. Since the information processing device 230A performs the atomic motion visualization process, the decoding device 150A decodes only the basic code 142 and performs the floating point number data group 195 visualization process. This is because the accuracy up to the difference code 144 is not necessary for the visualization process. Since the information processing device 230B performs further numerical calculation by the analysis processing, the decoding device 150B decodes the basic code 142 and the differential code 144 to obtain the floating-point number data group 195 (data having the same accuracy as the floating-point number data group 105). ) Is used for numerical calculation. Since the information processing device 230C further performs numerical calculation by analysis processing together with the visualization processing, when performing the visualization processing, the decoding device 150C decodes only the basic code 142 and performs the visualization processing of the floating point number data group 195. . When analysis processing is performed, the basic code 142 and the differential code 144 are decoded, and numerical calculation using the floating-point number data group 195 is performed.
The super computer 210B encodes the processing result of the super computer 210B with the encoding device 100B and transmits the result to the information processing device 230D and the information processing device 230E. Since the information processing device 230D performs visualization processing of atomic motion, the decoding device 150D decodes only the basic code 142 and performs visualization processing of the floating point number data group 195. Since the information processing device 230E further performs numerical calculation by analysis processing, the decoding device 150E decodes the basic code 142 and the differential code 144 and performs numerical calculation using the floating point number data group 195.

FIG. 3 is a flowchart illustrating a processing example according to the first exemplary embodiment (encoding device 100).
In step S302, the block division module 110 divides the block.
In step S304, the floating point number separation module 115 separates the exponent value e and the mantissa value m.
In step S306, the intra-block maximum exponent value calculation module 120 calculates the intra-block maximum exponent value f.
In step S308, the difference index value calculation module 125 calculates a difference index value Δe.
In step S310, the mantissa division module 130 divides the mantissa m. Details of the processing in step S310 will be described later using the flowchart shown in the example of FIG.
In step S312, the first encoding module 135 performs encoding of basic data by compression A.
In step S314, the second encoding module 140 encodes the difference data by compression B.

FIG. 4 is a flowchart showing an example of processing by the first embodiment (the mantissa division module 130).
In step S402, the mantissa division module 130 calculates the mantissa value m1 with effective accuracy. Specifically, the mantissa value m1 is calculated by m1 = [m / 10 ^∧ Δe].
In step S404, the mantissa division module 130 calculates an invalid precision mantissa m2. Specifically, the mantissa value m2 is calculated by m2 = m−m1 * 10 △ ^Δe .

FIG. 5 is a flowchart illustrating a processing example according to the first exemplary embodiment (decoding device 150).
In step S502, the first decryption module 155 decrypts the basic data by the decompression A.
In step S504, the second decoding module 160 decodes the code data by the expansion B.
In step S506, the addition module 165 adds data.
In step S508, the conversion module 170 generates a floating point number.
In step S510, the block integration module 175 integrates the blocks.

Hereinafter, the processing by each module will be described with a specific example. However, in order to simplify the explanation, the number of data is small and simple numerical values are used.
FIG. 6 is an explanatory diagram illustrating an example of numerical calculation results. FIG. 6 shows a state in which one particle in a specific environment moves on the X axis. The horizontal axis represents time [sec], and the vertical axis represents the X coordinate value. FIG. 7 is data in which the X-coordinate values of the particles in FIG. 6 are arranged in chronological order in a floating-point number format with five significant digits. FIG. 7 is an explanatory diagram showing an example of the data structure of the processing result table 700. The processing result table 700 has a time [sec] column 710 and an X coordinate value [mm] column 720. The time [sec] column 710 stores time [sec]. The X coordinate value [mm] column 720 stores the X coordinate value [mm].
In this example, the X coordinate value in FIG. 7 is used as input data (floating point number data group 105). Hereinafter, the X coordinate value of the particle for each time is referred to as an element.

Processing performed by the encoding device 100 will be described.
The block division module 110 divides input data (floating point number data group 105) into blocks. In this example, as shown in the example of FIG. 8, the data is divided into five elements (block 810, block 820) with reference to the time axis. In the following, description will be given focusing on block 810.

例えば、Ｆ（ｔ）＝１．１１１１ｅ^＋０２のとき、ｅ（ｔ）＝２及びｍ（ｔ）＝１１１１１となる。式（１）より、図９の例に示す注目ブロックの浮動小数点数は、図１０の例に示す指数値と図１１の例に示す仮数値に分離される。
図９は、処理結果テーブル（ブロック）９００のデータ構造例を示す説明図である。処理結果テーブル（ブロック）９００は、時間［ｓｅｃ］欄９１０、Ｘ座標値［ｍｍ］欄９２０を有している。時間［ｓｅｃ］欄９１０は、時間［ｓｅｃ］を記憶している。Ｘ座標値［ｍｍ］欄９２０は、Ｘ座標値［ｍｍ］を記憶している。
図１０は、指数値テーブル１０００のデータ構造例を示す説明図である。指数値テーブル１０００は、時間［ｓｅｃ］欄１０１０、指数値欄１０２０を有している。時間［ｓｅｃ］欄１０１０は、時間［ｓｅｃ］を記憶している。指数値欄１０２０は、指数値を記憶している。
図１１は、仮数値テーブル１１００のデータ構造例を示す説明図である。仮数値テーブル１１００は、時間［ｓｅｃ］欄１１１０、仮数値欄１１２０を有している。時間［ｓｅｃ］欄１１１０は、時間［ｓｅｃ］を記憶している。仮数値欄１１２０は、仮数値を記憶している。 The floating point number separation module 115 separates the floating point number of the X coordinate value of the block 810 into a mantissa value and an exponent value. The time t is assumed to be N significant digits. Assuming that the floating point number data at time t is F (t), F (t) is expressed by the exponent value e (t) and the mantissa value m (t) as shown in Expression (1).

For example, when F (t) = 1.1111e ⁺ 02, e (t) = 2 and m (t) = 11111. From the expression (1), the floating point number of the target block shown in the example of FIG. 9 is separated into the exponent value shown in the example of FIG. 10 and the mantissa value shown in the example of FIG.
FIG. 9 is an explanatory diagram showing an example of the data structure of the processing result table (block) 900. The processing result table (block) 900 has a time [sec] column 910 and an X coordinate value [mm] column 920. The time [sec] column 910 stores time [sec]. The X coordinate value [mm] column 920 stores the X coordinate value [mm].
FIG. 10 is an explanatory diagram showing an example of the data structure of the exponent value table 1000. The exponent value table 1000 has a time [sec] column 1010 and an exponent value column 1020. The time [sec] column 1010 stores time [sec]. The exponent value column 1020 stores an exponent value.
FIG. 11 is an explanatory diagram showing an example of the data structure of the mantissa table 1100. The mantissa table 1100 has a time [sec] column 1110 and a mantissa column 1120. The time [sec] column 1110 stores time [sec]. The mantissa column 1120 stores mantissa values.

図１０の例では、ブロック最大指数値は、ｔ＝４のときでｆ＝３となる。 The intra-block maximum exponent value calculation module 120 calculates the maximum value from the exponent values in the block of interest and sets it as the intra-block maximum exponent value. The maximum exponent value f in the block is defined by the equation (2).

In the example of FIG. 10, the block maximum exponent value is f = 3 when t = 4.

式（３）より、注目ブロックの差分指数値△ｅ（ｔ）を図１２に示す。図１２は、差分指数値テーブル１２００のデータ構造例を示す説明図である。差分指数値テーブル１２００は、時間［ｓｅｃ］欄１２１０、差分指数値欄１２２０を有している。時間［ｓｅｃ］欄１２１０は、時間［ｓｅｃ］を記憶している。差分指数値欄１２２０は、差分指数値を記憶している。差分指数値欄１２２０内の値は、ブロック内最大指数値（３）から、図１０の例に示した指数値テーブル１０００の指数値欄１０２０内の値を減算したものである。 The difference index value calculation module 125 calculates a difference index value from the maximum index value in the block and each index value of the block. The difference index value Δe (t) at time t is defined by equation (3).

From equation (3), the difference index value Δe (t) of the block of interest is shown in FIG. FIG. 12 is an explanatory diagram showing an example of the data structure of the difference index value table 1200. The difference index value table 1200 has a time [sec] field 1210 and a difference index value field 1220. The time [sec] column 1210 stores time [sec]. The difference index value column 1220 stores a difference index value. The value in the difference index value column 1220 is obtained by subtracting the value in the index value column 1020 of the index value table 1000 shown in the example of FIG. 10 from the maximum index value (3) in the block.

よって、図１１の例に示す注目ブロックの仮数値ｍ（ｔ）は、有効精度の仮数値ｍ１（ｔ）（図１３）、無効精度の仮数値ｍ２（ｔ）（図１４）に分割される。
図１３は、有効精度の仮数値テーブル１３００のデータ構造例を示す説明図である。有効精度の仮数値テーブル１３００は、時間［ｓｅｃ］欄１３１０、有効精度の仮数値欄１３２０を有している。時間［ｓｅｃ］欄１３１０は、時間［ｓｅｃ］を記憶している。有効精度の仮数値欄１３２０は、有効精度の仮数値（上位の仮数値）を記憶している。図１４は、無効精度の仮数値テーブル１４００のデータ構造例を示す説明図である。無効精度の仮数値テーブル１４００は、時間［ｓｅｃ］欄１４１０、無効精度の仮数値欄１４２０を有している。時間［ｓｅｃ］欄１４１０は、時間［ｓｅｃ］を記憶している。無効精度の仮数値欄１４２０は、無効精度の仮数値（下位の仮数値）を記憶している。 The mantissa division module 130 divides the mantissa into a mantissa with effective precision and a mantissa with invalid precision. The effective precision mantissa value m1 (t) and the invalid precision mantissa value m2 (t) at time t are defined by equations (4) and (5), respectively.

Therefore, the mantissa value m (t) of the block of interest shown in the example of FIG. 11 is divided into the mantissa value m1 (t) (FIG. 13) with effective accuracy and the mantissa value m2 (t) (FIG. 14) with invalid accuracy. .
FIG. 13 is an explanatory diagram showing an example of the data structure of the significand value table 1300 of effective accuracy. The effective accuracy mantissa table 1300 includes a time [sec] column 1310 and an effective accuracy mantissa column 1320. The time [sec] column 1310 stores time [sec]. The effective precision mantissa column 1320 stores the effective precision mantissa (high order mantissa). FIG. 14 is an explanatory diagram showing an example of the data structure of the invalid precision mantissa table 1400. The invalid precision mantissa table 1400 includes a time [sec] field 1410 and an invalid precision mantissa field 1420. The time [sec] column 1410 stores time [sec]. The invalid precision mantissa field 1420 stores invalid precision mantissa values (lower mantissa values).

有効精度の仮数値ｍ１（ｔ）と予測した有効精度の仮数値ｐ（ｔ）の時間ｔにおける予測誤差△ｍ１（ｔ）は式（７）で定義する。

Note that the mantissa division module 130 (prediction error calculation module 3135 in a third embodiment to be described later) may generate a prediction value of a significand with effective accuracy and calculate a prediction error. With this processing, the code amount of the basic code can be reduced when the same mantissa value continues over time. Here, the estimated value p (t) of the mantissa with effective accuracy at time t is defined by equation (6). Note that another calculation method may be used, such as defining the predicted value p (t) as a value that minimizes the sum of squared errors with a plurality of effective precision mantissa values.

The prediction error Δm1 (t) at the time t between the effective precision mantissa value m1 (t) and the predicted effective precision mantissa value p (t) is defined by equation (7).

From the equation (7), the prediction error Δm1 (t) is shown in the example of FIG. In this configuration, the prediction error Δm1 (t) is encoded instead of the mantissa value m1 (t).
FIG. 15 is an explanatory diagram showing an example of the data structure of the prediction error table 1500.
The prediction error table 1500 includes a time [sec] column 1510 and a prediction error column 1520 of a significand value of effective accuracy. The time [sec] column 1510 stores time [sec]. The effective accuracy mantissa prediction error column 1520 stores an effective accuracy mantissa prediction error.

The processing of the first encoding module 135 and the second encoding module 140 and the encoded data (basic code 142 and differential code 144) will be described. The first encoding module 135 outputs encoded data obtained by encoding the intra-block maximum exponent value and the mantissa value of effective precision as the basic code 142 (FIG. 16). The second encoding module 140 outputs encoded data obtained by encoding the invalid precision mantissa value and the exponent value residual as a differential code 144 (FIG. 17). When predicting a mantissa value with effective accuracy, the basic code encodes the intra-block maximum exponent value and the prediction error of the mantissa value with effective accuracy (FIG. 18). The basic code and the difference code are encoded losslessly using conventional methods such as Huffman code and arithmetic code, respectively.
FIG. 16 is an explanatory diagram showing a data structure example of encoded data 1600 of a basic code (an example of a structure of encoded data of a basic code (when effective value mantissa is not predicted)). The encoded data 1600 of the basic code is composed of an intra-block maximum exponent value 1610 and an effective precision mantissa 1620.
FIG. 17 is an explanatory diagram illustrating a data structure example (difference code encoded data structure example) of differential code encoded data 1700. The encoded data 1700 of the differential code includes a differential exponent value 1710 and an invalid precision mantissa 1720.
FIG. 18 is an explanatory diagram showing an example of the data structure of encoded data 1800 of the basic code (example of the structure of encoded data of the basic code (when predicting a significand with effective precision)). The encoded data 1800 of the basic code is composed of an intra-block maximum exponent value 1810 and an effective precision mantissa prediction error 1820.

Processing performed by the decoding device 150 will be described.
First, a method for acquiring data with effective accuracy will be described. The first decoding module 155 decodes the basic code at time t to obtain the maximum exponent value f ′ within the block of interest and the mantissa value m1 ′ (t) of effective accuracy. The exponent value e ′ (t) and the mantissa value m ′ (t) at time t are defined by Equation (8) and Equation (9).

式（１０）及び式（１１）より、注目ブロックの指数値ｅ’（ｔ）及び仮数値ｍ’（ｔ）を図１９の例、図２０の例に示す。
図１９は、指数値テーブル１９００のデータ構造例を示す説明図である。指数値テーブル１９００は、時間［ｓｅｃ］欄１９１０、指数値欄１９２０を有している。時間［ｓｅｃ］欄１９１０は、時間［ｓｅｃ］を記憶している。指数値欄１９２０は、指数値を記憶している。
図２０は、有効精度の仮数値テーブル２０００のデータ構造例を示す説明図である。有効精度の仮数値テーブル２０００は、時間［ｓｅｃ］欄２０１０、有効精度の仮数値欄２０２０を有している。時間［ｓｅｃ］欄２０１０は、時間［ｓｅｃ］を記憶している。有効精度の仮数値欄２０２０は、有効精度の仮数値を記憶している。 When the prediction error of the mantissa value with effective precision is encoded, from the prediction value p ′ (t) at time t and the prediction error Δm1 ′ (t) of mantissa value with effective accuracy obtained by decoding the basic code. The mantissa value m ′ (t) is calculated. The predicted value p ′ (t) and the mantissa value m ′ (t) are defined by Expression (10) and Expression (11).

From the formula (10) and the formula (11), the index value e ′ (t) and the mantissa value m ′ (t) of the block of interest are shown in the example of FIG. 19 and the example of FIG.
FIG. 19 is an explanatory diagram showing an example of the data structure of the exponent value table 1900. The exponent value table 1900 has a time [sec] column 1910 and an exponent value column 1920. The time [sec] column 1910 stores time [sec]. The exponent value column 1920 stores an exponent value.
FIG. 20 is an explanatory diagram showing an example of the data structure of the significand value table 2000 of effective accuracy. The effective accuracy mantissa table 2000 has a time [sec] field 2010 and an effective accuracy mantissa field 2020. The time [sec] column 2010 stores time [sec]. The effective accuracy mantissa column 2020 stores effective accuracy mantissa values.

式（１２）より、有効精度のデータを図２１の例に示す。
図２１は、有効精度データテーブル２１００のデータ構造例を示す説明図である。有効精度データテーブル２１００は、時間［ｓｅｃ］欄２１１０、Ｘ座標値［ｍｍ］欄２１２０を有している。時間［ｓｅｃ］欄２１１０は、時間［ｓｅｃ］を記憶している。Ｘ座標値［ｍｍ］欄２１２０は、Ｘ座標値［ｍｍ］を記憶している。 A case where the processing by the second decoding module 160 and the addition module 165 is not necessary will be described.
The conversion module 170 converts the maximum exponent value in the block and the mantissa value of effective precision into a floating point number. The time t is assumed to be N significant digits. Assuming that the floating point number data at time t is F ′ (t), F ′ (t) is expressed as shown in Expression (12).

The data of effective accuracy is shown in the example of FIG.
FIG. 21 is an explanatory diagram showing an example of the data structure of the effective accuracy data table 2100. The effective accuracy data table 2100 has a time [sec] column 2110 and an X coordinate value [mm] column 2120. The time [sec] column 2110 stores time [sec]. The X coordinate value [mm] column 2120 stores the X coordinate value [mm].

The block integration module 175 integrates the divided blocks to obtain effective accuracy data. FIG. 22 shows an expansion result of floating-point number data with effective precision of input data. The accuracy of each block is unified as e + 03.
FIG. 22 is an explanatory diagram showing an example of the data structure of the effective accuracy data table 2200. The effective accuracy data table 2200 has a time [sec] column 2210 and an X coordinate value [mm] column 2220. The time [sec] column 2210 stores time [sec]. The X coordinate value [mm] column 2220 stores the X coordinate value [mm].

Next, a method for decompressing invalid precision data will be described. As a premise, it is assumed that the basic code is decoded by the first decoding module 155 to obtain data of effective accuracy.
Let Δe ″ (t) be the difference index value at time t, and m2 ″ (t) be the invalid precision mantissa. The second decoding module 160 restores the difference exponent value Δe ″ (t) and the invalid precision mantissa m2 ″ (t) from the difference code 144. Then, the addition module 165 adds the invalid precision difference exponent value Δe ″ (t) and the invalid precision mantissa m2 ″ (t) to the effective precision data, and the addition module 165 adds the effective precision data. The exponent value e (t) '' and the effective precision mantissa value m '' (t) are restored from the equations (13) and (14).

From the equations (13) and (14), the index value e ″ (t) and the mantissa value m ″ (t) of the block of interest are shown in the example of FIG. 23 and the example of FIG.
FIG. 23 is an explanatory diagram of an exemplary data structure of the exponent value table 2300. The exponent value table 2300 has a time [sec] column 2310 and an exponent value column 2320. The time [sec] column 2310 stores time [sec]. The exponent value column 2320 stores an exponent value.
FIG. 24 is an explanatory diagram showing an example of the data structure of the significand value table 2400 with effective accuracy. The effective accuracy mantissa table 2400 includes a time [sec] column 2410 and an effective accuracy mantissa column 2420. The time [sec] column 2410 stores time [sec]. The effective accuracy mantissa field 2420 stores effective accuracy mantissa values.

図２５は、有効精度データテーブル２５００（復号されたデータ）のデータ構造例を示す説明図である。有効精度データテーブル２５００は、時間［ｓｅｃ］欄２５１０、Ｘ座標値［ｍｍ］欄２５２０を有している。時間［ｓｅｃ］欄２５１０は、時間［ｓｅｃ］を記憶している。Ｘ座標値［ｍｍ］欄２５２０は、Ｘ座標値［ｍｍ］を記憶している。 The conversion module 170 converts the maximum exponent value in the block and the mantissa value of effective precision into a floating point number. Assuming that the floating point number data at time t is F ″ (t), F ″ (t) is expressed as in Expression (15). From the equation (15), the example of FIG. 25 is obtained as invalid precision data.

FIG. 25 is an explanatory diagram showing an example of the data structure of the effective accuracy data table 2500 (decoded data). The effective accuracy data table 2500 has a time [sec] column 2510 and an X coordinate value [mm] column 2520. The time [sec] column 2510 stores time [sec]. The X coordinate value [mm] column 2520 stores the X coordinate value [mm].

The block integration module 175 integrates the divided blocks and obtains invalid precision data. FIG. 26 shows an example of the decoding result of the invalid precision floating point number data of the input data.
FIG. 26 is an explanatory diagram showing an example of the data structure of the effective accuracy data table 2600 (decoded data). The effective accuracy data table 2600 has a time [sec] column 2610 and an X coordinate value [mm] column 2620. The time [sec] column 2610 stores time [sec]. The X coordinate value [mm] column 2620 stores the X coordinate value [mm].

  The difference in compression and expansion between the present embodiment and the conventional method will be described. For example, let us consider a case where the input of numerical calculation different from the case of visualizing the data in FIG. At this time, it is assumed that the data up to the accuracy of e + 03 is required for visualization, and the accuracy of the original data is required for the numerical calculation. The conventional method needs to compress and decompress the input data with the accuracy of e + 03 for visualization, and further compress and decompress with the accuracy of the original data for numerical calculation. In this embodiment, if the input data is compressed only once, the accuracy of e + 03 and the accuracy of the original data can be expanded without performing compression again.

  A difference in code amount between the present embodiment and the conventional method will be described. For example, the number of block divisions of input data is set to 1. The total code amount of encoded data when the conventional method is used is the sum of the code amount of “effective precision code” and the code amount of “invalid precision code”, and the code amount when this embodiment is used. The total code amount of the digitized data is the sum of the code amount of “basic code 142” and the code amount of “difference code 144”. Here, the code amount of “effective precision code” and the code amount of “basic code 142” are the same in principle. On the other hand, the code amount of “invalid precision code” is encoded data obtained by encoding all input data, and the code amount of “differential code 144” is encoded data obtained by encoding a part of the input data. Therefore, the code amount of “difference code 144” is smaller than the code amount of “invalid precision code”. Therefore, when compressing and decompressing data with a plurality of precisions, the total code amount of the encoded data of the present embodiment is smaller than the total code amount of the encoded data of the conventional method.

<Second Embodiment>
FIG. 27 is a conceptual module configuration diagram of a configuration example according to the second embodiment.
In addition, the same code | symbol is attached | subjected to the site | part of the same kind as the above-mentioned embodiment, and the overlapping description is abbreviate | omitted (hereinafter the same).
The encoding apparatus 2700 includes a block division module 110, a floating-point number separation module 115, an intra-block maximum exponent value calculation module 120, a difference exponent value calculation module 125, a mantissa value division module 130, an effective precision division module 2735, and 1-1. Encoding module 2740, first-second encoding module 2745, and second encoding module 140. The encoding device 2700 is obtained by adding an effective accuracy division module 2735, a 1-2 encoding module 2745, and the like to the encoding device 100.

The block division module 110 is connected to the floating point number separation module 115 and receives the floating point number data group 105.
The floating point number separation module 115 is connected to the block division module 110, the intra-block maximum exponent value calculation module 120, and the mantissa division module 130.
The intra-block maximum exponent value calculation module 120 is connected to the floating point number separation module 115 and the difference exponent value calculation module 125.
The difference index value calculation module 125 is connected to the intra-block maximum index value calculation module 120 and the mantissa division module 130.
The mantissa division module 130 is connected to the floating point number separation module 115, the difference exponent value calculation module 125, and the effective precision division module 2735.
The effective precision division module 2735 is connected to the mantissa division module 130, the 1-1 encoding module 2740, the 1-2 encoding module 2745, and the second encoding module 140. The effective accuracy division module 2735 divides the first mantissa. Specifically, the effective accuracy division module 2735 divides the effective accuracy into arbitrary accuracy. Thereby, the effective accuracy can be divided into a plurality of accuracy.
The 1-1 encoding module 2740 is connected to the effective precision division module 2735 and outputs the basic code 1: 2742. The 1-1 encoding module 2740 encodes the above-described predetermined exponent value and the most significant mantissa divided by the effective precision dividing module 2735. Specifically, the 1-1 encoding module 2740 encodes the maximum exponent value in the block and the mantissa value having the lowest precision divided by the minimum precision.
The 1-2 encoding module 2745 is connected to the effective precision division module 2735 and outputs the basic code m: 2742-m. The 1-2 encoding module 2745 encodes a mantissa value other than the most significant mantissa divided by the effective precision division module 2735. Specifically, the 1-2 encoding module 2745 encodes an effective precision mantissa.
The second encoding module 140 is connected to the effective accuracy division module 2735 and outputs a difference code 144.

The decoding device 2750 includes a 1-1 decoding module 2755, a 1-2 decoding module 2760, a second decoding module 160, an addition module 165, a conversion module 170, and a block integration module 175.
The 1-1st decoding module 2755 is connected to the addition module 165 and receives the basic code 1: 2742. The 1-1 decoding module 2755 is a predetermined exponent value encoded by the 1-1 encoding module 2740 of the encoding device 2700 and the most significant mantissa value divided by the effective precision dividing module 2735. Is decrypted. Specifically, the 1-1st decoding module 2755 decodes the mantissa having the lowest precision among the mantissas having effective precision divided from the maximum exponent value in the block from the basic code 1: 2742.
The 1-2 decoding module 2760 is connected to the addition module 165 and receives the basic code m: 2742-m. The 1-2 decoding module 2760 encodes a mantissa value other than the highest-order mantissa value divided by the effective precision division module 2735 encoded by the 1-2 encoding module 2745 of the encoding device 2700. To do. Specifically, the 1-2 decoding module 2760 decodes a mantissa value other than the lowest accuracy among the mantissa values of effective precision divided from the basic code m: 2742-m.
The second decoding module 160 is connected to the addition module 165 and receives the difference code 144.
The addition module 165 is connected to the 1-1 decoding module 2755, the 1-2 decoding module 2760, the second decoding module 160, and the conversion module 170.
The conversion module 170 is connected to the addition module 165 and the block integration module 175.
The block integration module 175 is connected to the conversion module 170 and outputs a floating point number data group 195.

FIG. 28 is an explanatory diagram illustrating a processing example according to the second exemplary embodiment.
The effective precision division module 2735 has a mantissa division module 2835 with effective precision.
The effective precision mantissa division module 2835 is connected to the 1-1 encoding module 2740 and the 1-2 encoding module 2745. The effective precision mantissa division module 2835 accepts an N-digit effective precision mantissa 2810, divides the N-digit effective accuracy mantissa 2810 into k (2 ≦ k ≦ M), and outputs the first mantissa 2810. The numerical value 2837 is passed to the 1-1 encoding module 2740 and the m-th (m = 1... K) mantissa 2839 is passed to the 1-2 encoding module 2745.
The 1-1 encoding module 2740 is connected to the effective precision mantissa division module 2835 of the effective accuracy division module 2735, receives the first mantissa value 2837 from the effective precision mantissa division module 2835, and The code 1: 2742 is output.
The 1-2 encoding module 2745 is connected to the effective precision mantissa division module 2835 of the effective accuracy division module 2735, and the mth (m = 1... K) mth (m = 1... K) from the effective accuracy mantissa division module 2835. The mantissa value 2839 is received, and the basic code 2: 2842-2, the basic code 3: 2842-3, the basic code k-1: 2842-k-1, and the basic code k: 2842-k are output.

  The encoding device 2700 is obtained by adding an effective accuracy division module 2735 for dividing the effective accuracy to an arbitrary accuracy to the encoding device 100. By dividing the effective accuracy into a plurality of accuracy and encoding, the user can obtain the required accuracy according to the application. Let M be the number of significant digits of the significand of effective precision. A mantissa of effective accuracy is divided into k (2 ≦ k ≦ M) with arbitrary accuracy, and a basic code m (m = 1... K) is generated. Here, the divided mantissa values with effective accuracy are called m-th mantissa values in order from the lowest accuracy. The basic code 1: 2742 is composed of encoded data of the maximum exponent value in the block and the first mantissa value. For example, the encoded data 2900 of the basic code 1 shown in FIG. 29 shows the data structure of the basic code 1: 2742, and is composed of an intra-block maximum exponent value 2910 and a first mantissa value 2920. The basic code m: 2742-m (m ≧ 2) is composed of encoded data of the mth mantissa value. For example, the encoded data 3000 of the basic code m (m ≧ 2) shown in FIG. 30 shows the data structure of the basic code m: 2742-m. The basic code m is reversibly encoded using a conventional method such as a Huffman code or an arithmetic code. As a result, the decoding device 2750 adds the basic code m: 2742−m (m ≧ 2) after decompressing the data with effective precision decoded with the basic code 1: 2742, and thus is higher without re-encoding. Accurate data can be decompressed in stages.

<Third Embodiment>
FIG. 31 is a conceptual module configuration diagram of a configuration example according to the third embodiment.
The encoding device 3100 includes a block division module 110, a floating point number separation module 115, an intra-block maximum exponent value calculation module 120, a difference exponent value calculation module 125, a mantissa value division module 130, a prediction error calculation module 3135, a first 1-a. Encoding module 3140 and second encoding module 140. The encoding device 3100 is obtained by adding a prediction error calculation module 3135 to the encoding device 100. In the description of the first embodiment, the processing described by using the expressions (6), (7), FIG. 15 and FIG. 18 is performed.
The block division module 110 is connected to the floating point number separation module 115 and receives the floating point number data group 105.
The floating point number separation module 115 is connected to the block division module 110, the intra-block maximum exponent value calculation module 120, and the mantissa division module 130.
The intra-block maximum exponent value calculation module 120 is connected to the floating point number separation module 115 and the difference exponent value calculation module 125.
The difference index value calculation module 125 is connected to the intra-block maximum index value calculation module 120 and the mantissa division module 130.
The mantissa division module 130 is connected to the floating point number separation module 115, the difference index value calculation module 125, and the prediction error calculation module 3135.
The prediction error calculation module 3135 is connected to the mantissa division module 130, the 1-a encoding module 3140, and the second encoding module 140. The prediction error calculation module 3135 calculates an error between the first mantissa value or the second mantissa value and the prediction value. Specifically, the prediction error calculation module 3135 calculates a prediction error between the mantissa value of the effective accuracy and the prediction value. Thus, the compression rate can be increased when the same mantissa value continues in the data string.
The 1-a encoding module 3140 is connected to the prediction error calculation module 3135 and outputs a basic code a3142. The 1-a encoding module 3140 encodes the predetermined exponent value and the error calculated by the prediction error calculation module 3135. Specifically, the 1-a encoding module 3140 encodes the intra-block maximum exponent value and the prediction error.
The second encoding module 140 is connected to the prediction error calculation module 3135 and outputs a difference code 144. The second encoding module 140 encodes the error calculated by the prediction error calculation module 3135.

The decoding device 3150 includes a 1-a decoding module 3155, a second decoding module 160, an effective precision mantissa restoration module 3160, an addition module 165, a conversion module 170, and a block integration module 175.
The 1-a decoding module 3155 is connected to an effective precision mantissa restoration module 3160 and receives a basic code a3142. The 1-a decoding module 3155 decodes the predetermined exponent value encoded by the 1-a encoding module 3140 of the encoding device 3100 and the error calculated by the prediction error calculation module 3135. Specifically, the 1-a decoding module 3155 decodes the intra-block maximum exponent value and the prediction error from the basic code a: 3142.
The effective precision mantissa restoration module 3160 is connected to the 1-a decoding module 3155 and the addition module 165. The effective precision mantissa restoration module 3160 restores the effective precision mantissa from the prediction error.
The second decoding module 160 is connected to the addition module 165 and receives the difference code 144.
The addition module 165 is connected to the mantissa restoration module 3160 with effective precision, the second decoding module 160, and the conversion module 170.
The conversion module 170 is connected to the addition module 165 and the block integration module 175.
The block integration module 175 is connected to the conversion module 170 and outputs a floating point number data group 195.

FIG. 32 is an explanatory diagram illustrating a processing example according to the third exemplary embodiment.
The prediction error calculation module 3135 is connected to the 1-a-th encoding module 3140 and receives a significand value 3210 of effective accuracy.
The 1-a encoding module 3140 is connected to the prediction error calculation module 3135 and outputs a basic code a3142.
The prediction error calculation module 3135 calculates a prediction error between the effective precision mantissa value and the prediction value. As a method for generating a predicted value, for example, a method using an encoded effective precision mantissa value as a predicted value, or a value that minimizes the sum of squares of errors with a plurality of encoded effective accuracy mantissa values is used as a predicted value. A method used as the above may be used. The prediction error is losslessly encoded using a conventional method such as Huffman code or arithmetic code. If an appropriate prediction method is used according to the characteristics of data, the compression rate can be increased.

<Fourth embodiment>
FIG. 33 is a conceptual module configuration diagram of a configuration example according to the fourth embodiment.
In the above-described embodiment, the floating-point number data group 105 is divided by the block division module 110, but it is not necessarily divided into blocks. It suffices if a predetermined exponent value (for example, the maximum exponent value) can be extracted from the target floating-point number data group 105.

The encoding device 3300 includes a floating point number separation module 115, a maximum exponent value calculation module 3320, a difference exponent value calculation module 125, a mantissa division module 130, a first encoding module 135, and a second encoding module 140. doing.
The floating point number separation module 115 is connected to the maximum exponent value calculation module 3320 and the mantissa division module 130 and receives the floating point number data group 105.
The maximum exponent value calculation module 3320 is connected to the floating point number separation module 115 and the difference exponent value calculation module 125. The maximum exponent value calculation module 3320 extracts the maximum exponent value in the target floating point number data group 105.
The difference index value calculation module 125 is connected to the maximum index value calculation module 3320 and the mantissa division module 130.
The mantissa division module 130 is connected to the floating point number separation module 115, the difference exponent value calculation module 125, the first encoding module 135, and the second encoding module 140.
The first encoding module 135 is connected to the mantissa division module 130 and outputs a basic code 142.
The second encoding module 140 is connected to the mantissa division module 130 and outputs a difference code 144.

The decoding device 3350 includes a first decoding module 155, a second decoding module 160, an addition module 165, and a conversion module 170.
The first decoding module 155 is connected to the addition module 165 and receives the basic code 142.
The second decoding module 160 is connected to the addition module 165 and receives the difference code 144.
The addition module 165 is connected to the first decoding module 155, the second decoding module 160, and the conversion module 170.
The conversion module 170 is connected to the addition module 165 and outputs a floating point number data group 195.

FIG. 34 is a flowchart illustrating a processing example according to the fourth exemplary embodiment (encoding device 3300).
In step S3402, the floating point number separation module 115 separates the exponent value e and the mantissa value m.
In step S3404, the maximum exponent value calculation module 3320 calculates the maximum exponent value f.
In step S3406, the difference index value calculation module 125 calculates a difference index value Δe.
In step S3408, the mantissa division module 130 divides the mantissa m. The detailed processing content of step S3408 has been described above with reference to the flowchart shown in the example of FIG.
In step S3410, the first encoding module 135 performs encoding of basic data using compression A.
In step S <b> 3412, the second encoding module 140 encodes difference data using compression B.
The flowchart shown in the example of FIG. 34 is obtained by deleting step S302 of the flowchart shown in the example of FIG. 3 and correcting step S306 to step S3404.

FIG. 35 is a flowchart illustrating a processing example according to the fourth exemplary embodiment (decoding device 3350).
In step S3502, the first decryption module 155 decrypts the basic data by the decompression A.
In step S3504, the second decoding module 160 decodes the code data by the expansion B.
In step S3506, the addition module 165 adds data.
In step S3508, the conversion module 170 generates a floating point number.
The flowchart shown in the example of FIG. 35 is obtained by deleting step S510 of the flowchart shown in the example of FIG.

  Note that the hardware configuration of the computer on which the program according to the present embodiment is executed is a general computer as illustrated in FIG. 36, specifically, a personal computer, a computer that can be a server, or the like. That is, as a specific example, the CPU 3601 is used as a processing unit (calculation unit), and the RAM 3602, ROM 3603, and HD 3604 are used as storage devices. As the HD 3604, for example, a hard disk or an SSD (Solid State Drive) may be used. Block division module 110, floating point number separation module 115, intra-block maximum exponent value calculation module 120, difference exponent value calculation module 125, mantissa division module 130, first encoding module 135, second encoding module 140, 1st decoding module 155, 2nd decoding module 160, addition module 165, conversion module 170, block integration module 175, effective precision division module 2735, 1-1 encoding module 2740, 1-2 encoding Module 2745, 1-2 decoding module 2760, effective precision mantissa division module 2835, prediction error calculation module 3135, 1-a encoding module 3140, 1-a decoding module 3155, effective accuracy temporary Numerical restoration module CPU 3601 for executing a program such as 160, a RAM 3602 for storing the program and data, a ROM 3603 for storing a program for starting the computer, a floating-point number data group 105 to be encoded, a basic User for HD 3604, which is an auxiliary storage device (may be a flash memory or the like) that stores code 142, differential code 144, and decoded floating point number data group 195, and a keyboard, mouse, touch screen, microphone, etc. An accepting device 3606 that accepts data based on the operation, an output device 3605 such as a CRT, a liquid crystal display, and a speaker, a communication line interface 3607 for connecting to a communication network such as a network interface card, and the like. And a bus 3608 for exchanging data. A plurality of these computers may be connected to each other via a network.

Among the above-described embodiments, the computer program is a computer program that reads the computer program, which is software, in the hardware configuration system, and the software and hardware resources cooperate with each other. Is realized.
Note that the hardware configuration shown in FIG. 36 shows one configuration example, and the present embodiment is not limited to the configuration shown in FIG. 36, and is a configuration capable of executing the modules described in this embodiment. I just need it. For example, some modules may be configured with dedicated hardware (for example, Application Specific Integrated Circuit (ASIC), etc.), and some modules are in an external system and connected via a communication line Alternatively, a plurality of systems shown in FIG. 36 may be connected to each other via a communication line and cooperate with each other. In particular, in addition to personal computers, portable information communication devices (including mobile phones, smartphones, mobile devices, wearable computers, etc.), information appliances, robots, copiers, fax machines, scanners, printers, multifunction devices (scanners, printers, An image processing apparatus having two or more functions such as a copying machine and a fax machine) may be incorporated.

The program described above may be provided by being stored in a recording medium, or the program may be provided by communication means. In that case, for example, the above-described program may be regarded as an invention of a “computer-readable recording medium recording the program”.
The “computer-readable recording medium on which a program is recorded” refers to a computer-readable recording medium on which a program is recorded, which is used for program installation, execution, program distribution, and the like.
The recording medium is, for example, a digital versatile disc (DVD), which is a standard established by the DVD Forum, such as “DVD-R, DVD-RW, DVD-RAM,” and DVD + RW. Standard “DVD + R, DVD + RW, etc.”, compact disc (CD), read-only memory (CD-ROM), CD recordable (CD-R), CD rewritable (CD-RW), Blu-ray disc ( Blu-ray (registered trademark) Disc), magneto-optical disk (MO), flexible disk (FD), magnetic tape, hard disk, read-only memory (ROM), electrically erasable and rewritable read-only memory (EEPROM (registered trademark)) )), Flash memory, Random access memory (RAM) SD (Secure Digital) memory card and the like.
Then, the whole or a part of the program may be recorded on the recording medium for storage or distribution. Also, by communication, for example, a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a wired network used for the Internet, an intranet, an extranet, or a wireless communication It may be transmitted using a transmission medium such as a network or a combination of these, or may be carried on a carrier wave.
Furthermore, the program may be a part or all of another program, or may be recorded on a recording medium together with a separate program. Moreover, it may be divided and recorded on a plurality of recording media. Further, it may be recorded in any manner as long as it can be restored, such as compression or encryption.

DESCRIPTION OF SYMBOLS 100 ... Encoding apparatus 110 ... Block division | segmentation module 115 ... Floating-point number separation module 120 ... Intra-block maximum exponent value calculation module 125 ... Difference exponent value calculation module 130 ... Mantissa value division module 135 ... 1st encoding module 140 ... 1st 2. Coding module 150 ... Decoding device 155 ... First decoding module 160 ... Second decoding module 165 ... Addition module 170 ... Conversion module 175 ... Block integration module 210 ... Supercomputer 220 ... Information processing device 230 ... Information processing device 290 ... Communication line 2700 ... Encoding device 2735 ... Effective precision division module 2740 ... 1-1 encoding module 2745 ... 1-2 encoding module 2750 ... Decoding device 2755 ... 1-1 decoding module 760 ... 1-2 decoding module 2835 ... Effective precision mantissa division module 3100 ... Encoding device 3135 ... Prediction error calculation module 3140 ... 1-a encoding module 3150 ... Decoding device 3155 ... 1-a Decoding module 3160 ... Effective precision mantissa restoration module 3300 ... Encoding device 3320 ... Maximum exponent value calculation module 3350 ... Decoding device

Claims (9)

Separating means for separating floating point number representation data into an exponent value and a mantissa value;
A difference means for extracting a difference between a predetermined exponent value and an exponent value of each data;
A dividing means for dividing the mantissa value of each data into an upper first mantissa value and a lower second mantissa value according to the difference;
First encoding means for encoding the predetermined exponent value and the first mantissa value;
An information processing apparatus comprising: a second encoding unit that encodes the difference between each data and a second mantissa. As a predetermined exponent value, the maximum exponent value in the data of a plurality of target floating-point numbers,
The information processing apparatus according to claim 1. The dividing means sets the value of the number of lower digits of the mantissa value of the floating-point number data as the second mantissa value, and sets the value of the remaining upper digits as the first mantissa value.
The information processing apparatus according to claim 1 or 2. Second dividing means for dividing the first mantissa value;
A third encoding means for encoding a mantissa value other than the highest-order mantissa value divided by the second dividing means;
The first encoding means encodes the predetermined exponent value and the most significant mantissa divided by the second dividing means;
The information processing apparatus according to any one of claims 1 to 3. Prediction error calculation means for calculating an error between the first mantissa value or the second mantissa value and a prediction value;
The first encoding means or the second encoding means encodes the error;
The information processing apparatus according to any one of claims 1 to 4. A first decoding means for decoding a predetermined exponent value and a first mantissa encoded by the first encoding means of the information processing apparatus according to claim 1;
A second decoding means for decoding the difference encoded by the second encoding means of the information processing apparatus according to claim 1 and a second mantissa;
Adding means for adding a first mantissa value and a second mantissa value according to the data decoded from the first decoding means and the second decoding means;
Conversion means for converting the processing result by the addition means into a floating-point number representation;
According to the accuracy required for the decoded data, do not perform the processing by the second decoding means or the adding means,
Information processing device. Separating means for separating floating point number representation data into an exponent value and a mantissa value;
A difference means for extracting a difference between a predetermined exponent value and an exponent value of each data;
A dividing means for dividing the mantissa value of each data into an upper first mantissa value and a lower second mantissa value according to the difference;
First encoding means for encoding the predetermined exponent value and the first mantissa value;
An encoding device having second encoding means for encoding the difference of each data and a second mantissa;
First decoding means for decoding a predetermined exponent value and a first mantissa value encoded by the first encoding means;
Second decoding means for decoding the difference encoded by the second encoding means and the second mantissa;
Adding means for adding a first mantissa value and a second mantissa value according to the data decoded from the first decoding means and the second decoding means;
Conversion means for converting the processing result by the addition means into a floating-point number representation;
According to the accuracy required for the decoded data, do not perform the processing by the second decoding means or the adding means,
An information processing system having a decoding device. Computer
Separating means for separating floating point number representation data into an exponent value and a mantissa value;
A difference means for extracting a difference between a predetermined exponent value and an exponent value of each data;
A dividing means for dividing the mantissa value of each data into an upper first mantissa value and a lower second mantissa value according to the difference;
First encoding means for encoding the predetermined exponent value and the first mantissa value;
An information processing program for functioning as a second encoding means for encoding the difference between each data and a second mantissa. Computer
First decoding means for decoding a predetermined exponent value and a first mantissa value encoded by a first encoding means of a computer on which the information processing program according to claim 8 is executed;
A second decoding means for decoding the difference encoded by the second encoding means of the computer executing the information processing program according to claim 8 and the second mantissa;
Adding means for adding a first mantissa value and a second mantissa value according to the data decoded from the first decoding means and the second decoding means;
Function as conversion means for converting the processing result by the addition means into a floating-point number representation;
According to the accuracy required for the decoded data, do not perform the processing by the second decoding means or the adding means,
Information processing program.

Images