JPH02126317A - Data format converting system - Google Patents

Abstract

PURPOSE:To attain the high speed conversion of floating point form data to prescribed fixed point format data by shifting the bits of only the number of shifts with a mantissa part as an index part and with the number of cut-off digits as the number of shifts by fixed point format data. CONSTITUTION:An adding means 1 prepares a sum 6 with a value 5 to invert the value of an index part 4 and respective digits of the binary expression of the number of the digits of the prescribed digits or below and obtains the value related to the number of shifts. A selecting means 2 identifies the value of the sum 6, and at the time of being the size in the prescribed range, the value composed of the bit column of the prescribed digit or below of the sum 6 is outputted as the number 7 of controls. At the time of being smaller than the prescribed range, the prescribed value to obtain the result of the underflow is outputted as the number 7 of controls and at the time of being larger than the range, the prescribed value to obtain the result of the overflow is outputted as the number 7 of controls respectively. A memory 3 has respective storing positions corresponding to the number 7 of the controls concerning all respective values capable of obtaining by a mantissa part 8. By the address specified by the mantissa part 8 and the number 7 of the controls, the necessary data format converting result 9 is held at the storing position read from the memory 3.

Description

[Detailed Description of the Invention] [Summary] Regarding a conversion method from floating point format data to fixed point format data in data processing, an economical method that can convert floating point format data to predetermined fixed point format data at high speed is provided. For the purpose of data format conversion method, in data format conversion from floating point format data consisting of a mantissa and exponent part (4) to a fixed point format with truncated below a predetermined digit, the value of the exponent part and the below predetermined digit are calculated. an addition means for generating a sum with a value obtained by inverting each digit of a binary representation of the number of digits, and when the size of the sum is within a predetermined range, a value consisting of a bit string of a predetermined digit or less of the sum; a selection means for setting a predetermined value as a control number if it is smaller than the range, and another predetermined value if it is larger than the range; and each memory corresponding to the control number for each possible value of the mantissa. A memory having a location is provided, and data read from the memory at an address determined by the mantissa part and the control number is the format conversion result of the floating point format data corresponding to the mantissa part and the control number. Configure.

[Industrial application field]

The present invention relates to a data format conversion method for converting floating point format data to fixed point format data in data processing.

[Prior Art and Problems to be Solved by the Invention] In data processing in which numerical values are handled by a computer, floating point format data input as numerical data is often converted to fixed point format data for convenience of processing.

Here, floating-point format data is data in which a numerical value is expressed by a mantissa part and an exponent part, each having a predetermined bit length, as is well known, and for example, by the expression, [mantissa part] [Exponent part] power] It is defined that the mantissa part has a decimal point to the right of the least significant bit or to the left of the most significant bit representing the absolute value, and the exponent part is an integer.

On the other hand, fixed-point format data, as is well known, is data that represents an integer (that is, a number with a decimal point to the right of the least significant bit) using a predetermined bit length. Note that this expression is, for example, a representation format in which a certain number of lower digits of an integer are truncated,
That is, it can also be treated as representing [an integer represented by data]×[2 to the power of a fixed number].

When converting the data format from floating point format to fixed point format as described above, shift the mantissa part to the left or right by the number of bits determined by the value represented by the exponent part of the floating point format as described above. Fixed-point format data is obtained by executing processing to bring the decimal point to the required fixed position, but this processing extracts the exponent part from the data and determines the shift amount and direction from the exponent part. If the mantissa is shifted and an overflow occurs as a result, processing such as detecting it and taking necessary measures is required, resulting in processing that takes a relatively long time to execute.

For this reason, we have developed an economical method for cases where high speed is required, such as when converting floating-point format data collected in real-time processing into fixed-point format data in real time and using it as input data for image processing. A system that satisfies these requirements is desired.

An object of the present invention is to provide an economical data format conversion method that can convert floating point format data into predetermined fixed point format data at high speed.

[Means to solve the problem]

FIG. 1 is a block diagram showing the configuration of the present invention.

The figure shows the configuration of a data format conversion method for converting floating point format data to predetermined fixed point format data.
1 is an addition means that outputs the sum of the exponent part 4 of the floating point format data and the inverted value 5 of the binary representation of a predetermined number of digits, and 2 is a selection that generates the control number 7 from the sum 6 output by the addition means 1. 3 is a mantissa part 8 of floating point format data and a selection means 2
This memory is configured to read out data 9 as a result of data format conversion by using the control number 7 outputted by the controller as an address input.

[For production]

When converting floating point format data with no bits in the mantissa and nc bits in the exponent to data in fixed point format with n0 bits in which the lower digits of the number of digits represented by the no bits (hereinafter referred to as the number of truncated digits) are truncated. Assuming that the decimal point of the mantissa is to the right of the least significant bit, for fixed-point format data, the mantissa is shifted by the number of bits equal to the number of shifts, with the number of shifts equal to "exponent part - number of digits to be truncated" (however, if the number of shifts is If it is positive, shift it to the left; if it is negative, shift it to the right).

In this case, if the number of shift bits is n0 or more, in the case of a left shift, it will overflow unconditionally regardless of the bit pattern of the mantissa, and a correct result cannot be obtained, and in the case of a right shift, Since it will underflow unconditionally and the result will be 0, the number of shifts will be −(no −1) ~+
Only within the range of 2n,, -1 up to (no-1),
It is possible to obtain a result with a valid numerical value depending on the value of each mantissa, and if it is smaller than this range, the result is unconditionally 0, and if it is larger than this range, the result is not obtained. should be made unconditional. The invention allows for a reduction in the memory requirements for holding results on this basis.

To do this, first, the adding means 1 shown in FIG. 1 generates a sum 6 of the value of the exponent part 4 and the value 5 obtained by inverting each digit of the binary representation of the number of digits below a predetermined number, which is related to the shift number. The selection means 2 identifies the value of the sum 6, and if the sum 6 is within a predetermined range, the control number 7 selects a value consisting of a bit string of a predetermined digit or less of the sum 6 as being within the range. Output as . Further, if it is smaller than the range, a predetermined value for obtaining the underflow result is output, and if it is larger than the range, another predetermined value for obtaining the overflow result is output as the control number 7, respectively.

The memory 3 has storage locations corresponding to the control number 7 for all possible fractions of the mantissa part 8, and the required data is stored in the storage location read from the memory 3 according to the address determined by the mantissa part 8 and the control number 7. Configure to retain format conversion results.

With the above processing method, it is possible to obtain conversion result data directly from memory at high speed without the need for a shift operation of the mantissa, and the memory capacity required for this can be reduced to 2no -1 for each bit pattern of the mantissa. It can be made economical by compressing it into words.

〔Example〕

Figure 2 shows floating point format data consisting of a 7-bit mantissa indicated by DIN and a 5-bit exponent not concluded by GIN, and is truncated by the number of 5-bit digits indicated by Gl.
FIG. 2 is a block diagram showing an example of a configuration of a data format conversion method for converting data into bit fixed-point format data.

First, GlN and the inverted value of each bit of G are inputted to the adder 10, which serves as an adding means, and the addition is performed, resulting in an addition result G consisting of a 5-bit sum 11 and a 1-bit carry 12.
Get Aoo. GADD is clearly G, NG* + (
2'-1), so G, N-GR and G are the shift numbers.
The relationship with ADD is shown in the table below.

Table odor "i', GAIIE + color (25-1) Sochi binary 11 (in expression, subtracting (11111) is the number of bits to be shifted, and shifting more than 7 bits to the right or left will result in underflow or overflow. It is clear that the shift amount becomes even larger outside the range of the table, and valid conversion results are obtained within the range of right or left shift of 6 bits in the table.

Therefore, it is sufficient to detect a shift number of 6 bits or less, but the control number is determined for the range shown in the table and for smaller and larger cases, taking into consideration the address boundaries of the memory configuration, which will be described later.

For this purpose, the identification circuit 13 constitutes a selection means together with the selector 14, detects that GADD is within the range shown in the table, and controls the selector 14 with its output to obtain the control number S. L.I.
? Determine.

That is, when GADD is within the range of the table, the output of the identification circuit 13 becomes 1, the A side of the selector 14 is selected, and the lower 4 bits of GADD are set to S. Output as U. Also, if GAD is outside the range shown in the table, the identification circuit 1
3 becomes 0, and the B side of the selector 14 is selected.

The lower 3 bits of the lower 4 bits of G Ag2O are connected to the B side as carry bits. Therefore, if the range is smaller than the range in the table, it is expressed in binary (1000), and if it is larger, it is expressed as (1000).
0111) to S. It will be output as U.

S out is connected to the lower part of the mantissa part DIN and is stored in memory 1.
5 address input. Therefore, for each bit pattern of the mantissa, its mantissa and S. ur 'c Each word of the memory 15 read at the concatenated address has its mantissa part S. By storing in advance the result of shifting according to the number of shifts corresponding to the LITO value, data converted to fixed-point format is obtained by reading out a word in the memory 15.

In this case, the word corresponding to underflow is stored with all bits O, and the word corresponding to overflow is displayed with all bits set to 1, for example. Note that even if the number of shifts is within 6 bits to the right or left, underflow or overflow may occur depending on the bit pattern of the mantissa, and in such cases, the corresponding word may also have all 0s or 6 bits left. Memorize all 1.

〔Effect of the invention〕

As is clear from the above description, the present invention has a significant industrial effect in that conversion from floating point format data to fixed point format data in data processing can be processed at high speed with an economical configuration.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of the present invention, and FIG. 2 is a block diagram showing an embodiment of the invention. In the figure, ■ is an addition means, 2 is a selection means, 3.15 is a memory, 4 is an exponent part, 5 is a predetermined number of digits, 6 is a sum, 8 is a mantissa part, 10 is an adder, 14 is a selector binary representation 7 is a control number, 9 is data, 13 is an identification circuit, 3'''Ida is a block diagram showing the configuration of the present invention.FIG. 7 is a block diagram showing an embodiment of the present invention.

Claims (1)

[Claims] In data format conversion from floating point format data consisting of a mantissa part (8) and an exponent part (4) to a fixed point format with truncated below a predetermined digit, the value of the exponent part (4) and an addition means (1) for generating a sum (6) with a value (5) obtained by inverting each digit of a binary representation of the number of digits equal to or less than the predetermined number of digits; Selection means (2) that sets a value consisting of a bit string of a predetermined digit or less of the sum if it is within the range, a predetermined value if it is smaller than the range, and another predetermined value if it is larger than the range, respectively, as the control number (7). and a memory (3) having each storage location corresponding to the control number (7) for each possible value of the mantissa (8).
and data (9) read from the memory (3) at an address determined by the mantissa part (8) and the control number (7).
A data format conversion method, characterized in that the data format conversion method is configured to make the format conversion result of the floating point format data corresponding to the mantissa part and the control number.