JPH08101919A - Numerical converter and device for turning coordinate value to integer - Google Patents

Abstract

PURPOSE: To constitute a numerical converter in a small scale without additing limitation to texture coordinates and without deteriorating accuracy. CONSTITUTION: This numerical converter is constituted of floating point registers 10a-10i, a maximum exponent detector 11, mantissa shifting devices 12a-12i and encoding devices 13a-13i and converts floating point numbers held in the floating point registers 10a-10i to integers holding a relative size by detecting the maximum exponent of the floating point numbers held in the floating point registers 10a-10i in the maximum exponent detector 11, shifting a mantissa by the differences of the maximum exponent and respective exponents in the mantissa shifting devices 12a-12i and performing encoding in the encoding device 13a-13i.

Description

Detailed Description of the Invention

[0001]

The present invention relates to a three-dimensional computer
The present invention relates to a numerical conversion device and a coordinate value integer conversion device that perform texture mapping processing, which is a method of dramatically improving the expressiveness of a graphics system.

[0002]

2. Description of the Related Art Texture mapping is used to draw triangles and squares in computer graphics.
The image data prepared in advance is attached to the surface of the image to generate a highly realistic image. As an example, FIG. 7 shows that a quadrangle subjected to texture mapping is rotated.

Generally, texture mapping is performed in units of triangles at the time of drawing, and is processed as follows. 1) For each vertex of the triangle, texture coordinates (S,
T, Q) is given. 2) Texture coordinates (s, t, q) of points inside the triangle
Is obtained by linear interpolation from the texture coordinates of the vertices. 3) The position (u, v) on the image data is u = s / q,
Obtained by dividing v = t / q.

[0004] General computer graphics
Many systems are composed of two parts, a geometry part 100 and a rendering part 101 as shown in FIG.
The geometry unit 100 performs geometrical calculations such as coordinate conversion, lighting calculation, and clipping process, and the rendering unit 101 performs processes such as line generation and triangle filling. It is general that the geometry unit 100 performs floating point calculation, converts the result into an integer and transfers the result to the rendering unit 101, and the rendering unit 101 performs integer calculation.

At this time, the data to be transferred is X
There are YZ coordinates, colors (R, G, B), texture coordinates (S, T, Q) and the like. Data other than texture coordinates, such as color, is represented by an 8-bit integer, and 0 ≦
In many cases, R, G, and B ≦ 255 are set, and other data can be similarly converted into an n-bit integer.

However, the texture coordinates are -∞≤S, T,
The range of Q ≦ ∞ must be expressed, and cannot be expressed by an integer with a finite bit length.

For this reason, as shown in FIG. 9A, there is a method in which the geometry part 100a transfers the texture coordinates as floating point numbers, and the rendering part 101a performs the calculation of the texture Madupisog by the floating point calculation. Since the floating point arithmetic unit has a larger circuit scale than the integer arithmetic unit and the circuit is complicated, there is a problem that it is difficult to configure a small-scale and high-speed system. As shown in FIG. 9B, there is also a method of limiting the texture coordinate range in the geometry section 100b and converting the texture coordinate range into an integer with a finite bit length, and transferring the integer to the rendering section 101b. There is also a problem in that the accuracy is deteriorated because the effective bit length also becomes small when the texture coordinate value is small, in addition to the limitation.

[0008]

That is, when texture mapping is performed, the texture coordinates calculated by the geometry unit 100 cannot be expressed by an integer having a finite bit length. Therefore, the texture coordinates are transferred as floating point numbers,
Although there is a method of performing floating point calculation in the rendering unit 101, it is difficult to configure a small-scale and high-speed system because the circuit scale is large and the circuit is complicated.

There is also a method in which the range of texture coordinates is limited and an integer having a finite bit length is transferred.
The size of the image data to be pasted is also limited, and the accuracy also deteriorates.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide a numerical conversion device that is small in scale, does not impose restrictions on texture coordinates, and does not deteriorate accuracy. To do.

[0011]

According to the numerical conversion device of the present invention, in a numerical conversion device for converting a floating point number into an integer, floating point registers 10a to 10i as floating point holding means for holding a plurality of floating point numbers. And a maximum exponent detecting device 11 as a maximum exponent detecting means for detecting the maximum exponent from the floating point numbers in the floating point registers 10a to 10i, and the maximum exponent detected by the maximum exponent detecting device 11 and the floating point registers 10a to 10a. Mantissa shift devices 12a to 12i as mantissa shift means for calculating the difference between the exponent of each floating point number in 10i and shifting the mantissa using the difference as the shift amount, and the mantissa shift device 12a.
To 12i to 12i, encoding devices 13a to 13i are provided as encoding means for encoding the mantissas.

[0012]

In the numerical converter of the present invention, the maximum exponent detected by the maximum exponent detector 11 is used as a reference value, and the differences calculated by the mantissa shifters 12a to 12i are used as relative values, and the encoders 13a to 13i are used. By converting each floating point number held in the floating point registers 10a to 10i into an integer while holding the relative size of each floating point number, there is no restriction on the texture coordinates and the accuracy is also deteriorated. Without, it is possible to configure on a small scale.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS As an embodiment, the case of converting texture coordinates given by IEEE 32-bit floating point into a 32-bit integer will be described below.

As shown in FIG. 1, the numerical conversion apparatus of this embodiment has floating point registers 10a to 10i as floating point holding means, maximum exponent detecting apparatus 11 as maximum exponent detecting means, and mantissa as mantissa shift means. Shift devices 12a to 12i, encoding device 13 as encoding means
a to 13i, and a floating point register 10a
To 10i, the maximum exponent detection device 11 detects the maximum exponent of the floating point numbers held in the floating point registers 10a to 10i,
In the mantissa shift devices 12a to 12i, the mantissa is shifted according to the difference between the maximum exponent and each exponent, the encoding devices 13a to 13i perform encoding, and convert the integers to hold the relative size. .

Here, the reason why the relative size of the texture coordinates given by the floating point number is converted into an integer holding the texture coordinate is as follows.

That is, the texture coordinates of the three vertices of the triangle are (S1, T1, Q1), (S2, T2, Q2),
(S3, T3, Q3), the texture coordinates of a point inside the triangle obtained by linearly interpolating this vertex data are (s,
t, q), the texture coordinates of the three vertices are multiplied by α (αS1, αT1, αQ1), and (αS2, αT2,
When αQ2) and (αS3, αT3, αQ3) are set, the texture coordinates of a point inside the same triangle as described above are also multiplied by α to become (αs, αt, αq).

On the other hand, since the position (u, v) on the image data to be obtained is u = s / q and v = t / q, u = αs / αq even when the texture coordinate of the vertex is multiplied by α.
= S / q, v = αt / αq = t / q, and consequently the same position is obtained. This indicates that the texture coordinates given to the vertices have only to retain their relative sizes.

The floating point registers 10a-10i are
A 32-bit storage device that holds floating point numbers
As shown in, the stored data is 1-bit code SIGN (i), 8-bit exponent EXP (i), 23
It consists of a mantissa MAN (i) of bits.

The maximum exponent detection device 11 is a device for detecting the maximum exponent of the floating-point numbers of the nine floating-point registers 10a to 10i. As shown in FIG. 3A or 3B, the maximum exponent detection device 11 has eight maximum values. It is composed of the selection devices 30a to 30h. The maximum value selection devices 30a to 30h are devices that select the larger one of the two inputs, and by connecting eight of them to the tournament method, the maximum index MAXEX
Detect P.

The mantissa shift devices 12a to 12i find the difference between the maximum exponent MAXEXP and each exponent EXP (i) and use it as the shift amount SFTM (i) for each mantissa MAN.
A device for generating an unsigned 31-bit integer by right-shifting data STFS (i) based on (i),
As shown in FIG. 4A, the subtractor 40, the most significant bit generation device 41, and the shift device 42 are included.

In the subtractor 40, MAXEXP-EXP is used.
(I) is calculated to obtain the shift amount SFTM (i).
The most significant bit generation device 41 is a device for obtaining the implicit most significant bit MSB (i) of the omitted mantissa. The most significant bit is usually 1, but when the exponent is 0, it becomes 0 as a special case, so it is determined whether EXP (i) is 0, and if 0, MSB (i) is determined. Set to 0, otherwise set MSB (i) to 1. In the shift device 42, the data SFTS (i) composed of MSB (i) and MAN (i) is right-shifted by SFTM (i) to obtain an unsigned 31-bit integer SFT (i). The shifted data SFTS (i) is, for example, as shown in FIG.
As shown in FIG.
The MSB (i) obtained in 1 is an unsigned 31-bit integer in which each mantissa MAN (i) is placed in bits 29 to 7 and 0 is placed in the remaining bits 6 to 0.

As an operation example of the mantissa shift devices 12a to 12i, the case where MAXEXP, EXP (i), and MAN (i) have the following values (numerical values are binary numbers) is shown. MAXEXP = 00110000 EXP (i) = 00101000 MAN (i) = 01001010010100101001010 SFTM (i) is calculated in the subtractor 40, and SFTM (i) = MAXEXP-EXP (i) = 00001000 (8 in decimal)

Next, in the most significant bit generator 41, since EXP (i) is not 0, MSB (i) = 1 shift device 42 shifts the data SFTS.
(I) is SFTS (i) = | MSB (i) | MAN (i) | 0000000 = | 1 | 01001010010100101001010 | 0000000 = 1010010100101001010010100000000, and SFTS (i) is SFTM (i) (8 bits).
Then, SFT (i) is obtained by right shifting. SFT (i) = 0000000010100101001010010100101

If the floating point number to be converted is MAX
If it has EXP, EXP (i) = MAXE
From XP, SFTM (i) = MAXEXP-EXP
Since (i) = 0 and the shift amount becomes 0, SFT (i) = 1010010100101001010010100000000.

As described above, when a floating-point number having the maximum exponent is converted into an integer, 1 is always set at the highest level, and all effective bit lengths are used.

The encoding device 43 uses each code SIGN (i).
The unsigned 31-bit integer SFT (i) generated by the mantissa shift units 12a to 12i is encoded by
This is a device for converting into a signed 32-bit integer FIX (i), and comprises a minus device 50 and a selection device 51, as shown in FIG.

The minus device 50 is a device for converting input positive data into negative data, and generates -SFT (i) for SFT (i) input. The selection device 51 uses S
If IGN (i) is 0, SFT (i) is set to SIGN
If (i) is 1, -SFT (i) is selected, and SIGN (i) is added to the most significant bit (bit 31) as a sign bit to generate a 32-bit signed integer FIX (i). An example of the encoding device 43 is shown below.

FIX (i) = 0000000010100101001010010100101 Input (1) When SIGN (i) = 0 FIX (i) = | SIGN (i) | SFT (i) | = 00000000010100101001010010l00101 (2) SICN (i) = 1 FIX (i) = | SIGN (i) | -SFT (i) | = 11l1111110l01101011010ll01011011

According to the above-described embodiment, the texture coordinates given by the floating point number can be converted into an integer while keeping their relative sizes. Even if you give a floating-point number of any size, it will be converted so that all the effective bit lengths will be used when the floating-point number with the largest exponent is converted to an integer, so it will be converted accurately. It will be.

Further, by using the coordinate value integerizing device described above, it is possible to construct the rendering device 60 as shown in FIG.

That is, in FIG. 6, the coordinate value integer conversion device 61 is a device for converting the above floating point number into an integer, and converts the floating point texture coordinates (Sf, Tf, Qf) into integer texture coordinates (Si, Ti, It is a device for converting into Qi). The interpolating device 62 converts the integer texture coordinates (S of the vertex of the triangle converted by the coordinate integer converting device 61 (S
It is a device that interpolates integer texture coordinates (s, t, q) inside the triangle from i, Ti, Qi). The divider 63 calculates u = s / q, v = t / from the integer texture coordinates (s, t, q) inside the triangle obtained by the interpolator 62.
This is a device for dividing q to obtain a position on image data.

Since the rendering device 60 converts the floating-point texture coordinates into the optimum integer, it does not deteriorate the accuracy of the calculation after the conversion, and the floating-point calculation device does not have a large circuit scale, but a small-scale integer calculation. It can be configured with a device.

[0033]

As described above, according to the numerical conversion apparatus of the present invention, the maximum exponent detected by the maximum exponent detecting means is used as a reference value, and the difference calculated by the mantissa shifting means is used as a relative value for encoding. Since the means converts into an integer while holding the relative size of each floating point number held in the floating point holding means, there is no restriction on texture coordinates and there is no deterioration in accuracy. The effect is that it can be configured on a small scale.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a configuration of an embodiment of a numerical conversion device of the present invention.

FIG. 2 is a configuration diagram showing a configuration of data stored in a floating point register in FIG.

FIG. 3 is a configuration diagram showing a configuration of the maximum index detection device of FIG.

FIG. 4 is a configuration diagram showing the configuration of the mantissa shift device of FIG.

5 is a configuration diagram showing a configuration of the encoding device in FIG. 1. FIG.

6 is a configuration diagram showing a configuration of a rendering device using the integer conversion device of FIG. 1. FIG.

FIG. 7 is a diagram illustrating the concept of texture mapping.

FIG. 8 is a diagram illustrating a general example of a graphics system.

FIG. 9 is a configuration diagram showing a configuration of a conventional numerical value conversion device.

[Explanation of symbols]

 10a to 10i Floating point register 11 Maximum exponent detection device 12a to 12i Mantissa shift device 13a to 13i Encoding device 30a to 30h Maximum value selection device 40 Subtractor 41 Most significant bit generation device 42 Shift device 50 Minus device 51 Selection device 61 Coordinate integer conversion device 62 Interpolation device 63 Division device

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location 9365-5H G06F 15/72 A 9365-5H 450 A

Claims (4)

[Claims] 1. A numerical conversion device for converting a floating point number into an integer, a floating point holding means for holding a plurality of floating point numbers, and a maximum for detecting a maximum exponent from the floating point number in the floating point holding means. An exponent detecting unit, and a mantissa for shifting the mantissa using the difference as a shift amount by obtaining a difference between the maximum exponent detected by the maximum exponent detecting unit and the exponent of each floating point number in the floating point holding unit. Shift means, and encoding means for encoding the mantissa shifted by the mantissa shift means, the maximum exponent as a reference value, the difference as a relative value, relative to each of the floating point number A numerical conversion device characterized by converting to an integer while maintaining the size. 2. A floating point holding means for holding a plurality of floating point numbers, a maximum exponent detecting means for detecting a maximum exponent from the floating point numbers in the floating point holding means, and a maximum exponent detecting means for detecting the maximum exponent. A mantissa shift unit that shifts the mantissa using the difference as a shift amount by obtaining a difference between the maximum exponent and the exponent of each floating point number in the floating point holding unit; and a mantissa shifted by the mantissa shift unit. Encoding means for encoding, the maximum exponent as a reference value, the difference as a relative value, while the relative size of each of the floating point number is converted into an integer while holding, It is characterized by inputting the floating-point homogeneous coordinates of, converting it to the number of boards while keeping the relative size of all coordinate values, and outputting the integer homogeneous coordinates whose normalized values are equivalent. seat Integer apparatus. 3. Floating point holding means for holding a plurality of floating point numbers, maximum exponent detecting means for detecting a maximum exponent from the floating point numbers in the floating point holding means, and maximum exponent detecting means A mantissa shift unit that shifts the mantissa using the difference as a shift amount by obtaining a difference between the maximum exponent and the exponent of each floating point number in the floating point holding unit; and a mantissa shifted by the mantissa shift unit. Encoding means for encoding, the maximum exponent as a reference value, the difference as a relative value, while converting the floating point number relative magnitude of each of the floating point is converted into an integer, Floating-point homogeneous coordinates are input and converted into integers while maintaining the relative size of all coordinate values of multiple homogeneous coordinates, and the normalized values are equivalent and interpolable multiple integers Coordinates integer unit and outputs the coordinates. 4. The coordinate according to claim 3, wherein the plurality of floating-point homogeneous coordinates are texture coordinates of vertices of a polygon, and the interpolation means interpolates the texture coordinates by integer calculation. Value integerizer.