Computer system for interpolating a value for a pixel

[0002] The present invention relates to the field of computer implemented interpolation methods.

[0003] Interpolation is a well known procedure for determining a value corresponding to a point based on known values corresponding to surrounding points. Methods and apparatus for performing interpolation on computer systems can be critical to the user acceptance of certain computer applications.

[0004] For instance, many of the most advanced computer applications use three dimensional graphics that require texture mapping, i.e., mapping an image, or a texture map, onto a surface, such as an area in a pixel map to be displayed. A prior approach to texture mapping uses nearest neighbor sampling, wherein a color value for a display pixel is chosen to be the color value for the nearest texel, i.e., the nearest point in the texture map. However, this prior approach can result in undesirable jerky animation, because a small change in the texture map can cause dramatic changes in the nearest pixels.

[0005] Smoother animation can be achieved by using an approach involving bilinear interpolation. With bilinear interpolation, a color value for a display pixel is chosen by calculating a weighted average of the four nearest texels. However, using known methods and apparatus for bilinear interpolation on a computer system requires processor intensive iterations that degrade frame rates. Many application developers are unwilling to sacrifice high frame rates for smoother animation. Therefore, a novel approach to bilinear interpolation on a computer system has been developed.

[0006] A computer system comprising an output device, a processor, and a storage device is disclosed. The output device is configured to display an image representing a pixel map. The processor is coupled to the output device, and the storage device is coupled to the processor. A routine for interpolating a value for a pixel in the pixel map is stored in the storage device. The pixel is represented by at least two binary coordinates, each binary coordinate having a fractional portion. The routine, when executed by the processor, causes the processor to multiply together the fractional portions of at least two of the binary coordinates to generate a first sigma value, invert each bit of the first sigma value to generate an inverted first sigma value, and add the fractional portion of at least one of the binary coordinates to the inverted first sigma value to generate a second sigma value.

[0007] FIG. 1 illustrates a point for which a corresponding color value can be calculated using bilinear interpolation.

[0008] FIG. 2 is a flow chart illustrating one embodiment of the method of the present invention.

[0009] FIG. 3 is a block diagram of a computer system, illustrating one embodiment of the apparatus of the present invention.

[0010] A computer system and a computer implemented method for interpolating a value corresponding to a point represented by at least two binary coordinates are described. In the following description, specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, to one skilled in the art that the present invention can be practiced without regard to these specific details. In other instances, well known concepts have not been described in particular detail in order to avoid obscuring the present invention.

[0011] According to one approach, the present invention is embodied in a computer system and a computer implemented method for interpolating a color value for a pixel that is represented by a pair binary coordinates in a texture map. Each binary coordinate has an integer and a fractional portion. First, the fractional portions of the binary coordinates are multiplied together to generate a first sigma value. A known color value for a first of the four nearest texels is multiplied by the first sigma value to determine the first texel's contribution to the weighted average of the four nearest texels. Next, each bit of the first sigma value is inverted to generate an inverted first sigma value. Then, the fractional portion of one of the binary coordinates is added to the inverted first sigma value to generate a second sigma value. A known color value for a second of the four nearest texels is multiplied by the second sigma value to determine the second texel's contribution to the weighted average. The method continues until the weighted average has been calculated.

[0012] To help explain the present invention, the basic principles of bilinear interpolation will be explained with reference to FIG. 1. FIG. 1 illustrates a point 100 for which a corresponding color value can be calculated using bilinear interpolation. Point 100 is represented by coordinates (U.u,V.v). The integer portions of the coordinates of point 100 are U and V, and the fractional portions of the coordinates of point 100 are u and v. The four nearest integer points, i.e. points having both fractional portions of their coordinates equal to zero, are 101, 102, 103, and 104. The coordinates of points 101, 102, 103, and 104 are (U+1,V+1), (U+1,V), (U,V), and (U,V+1), respectively.

[0013] A color value can be assigned to point 100 by calculating the weighted average of the color values of points 101, 102, 103, and 104, which have known color values. The following set of equations, wherein the red color values of points 101, 102, 103, and 104 are RA, RB, RC, and RD, respectively, is used to calculate RP, the red color value for point 100:

[0014] s1=u*v

[0015] s2=u*(1−v)

[0016] s3=(1−u)*(1−v)

[0017] s4=(1−u)*v

[0018] RP=RA*s1+RB*s2+RC*s3+RD*s4

[0019] The variables s1, s2, s3, and s4 are referred to as sigma values. In a typical computer system using an RGB (red,green,blue) color model, the following two equations, wherein the green color values of points 101, 102, 103, and 104 are GA, GB, GC, and GD, respectively, and the blue color values of points 101, 102, 103, and 104 are BA, BB, BC, and BD, respectively, are added to the set the calculate GP, the green color value for point 100, and BP, the blue color value for point 100:

[0020] GP=GA*s1+GB*s2+GC*s3+GD*s4

[0021] BP=BA*s1+BB*s2+BC*s3+BD*s4

[0022] The full set of equations includes sixteen multiply operations, each of which must be executed in a prior approach to bilinear interpolation. Therefore, using this prior approach to calculate RGB values for texture mapping can be so processor intensive that frame rates are degraded.

[0023] One advantage of the present invention is that the sigma values can be calculated using a single multiply operation. After multiplying out the factors, the equations for the sigma values are:

[0024] s1=u*v

[0025] s2−u−u*v

[0026] s3=1−u−v+u*v

[0027] s4=v−u*v

[0028] Although each equation still includes a multiply operation, each equation can now use the same multiply operation. This multiply operation, u v, is performed in step 200 in FIG. 2, a flow chart illustrating one embodiment of the method of the present invention. In FIG. 2, u and v are binary fractions having any number of bits. In one embodiment, u and v each have sixteen bits. The first sigma value, s1, is generated by the multiply operation of step 200.

[0029] Although all four equations use the product term that is generated in step 200, two of the equations require the positive value of the product term while the other two require the negative value. To efficiently deal with this situation, the present invention takes advantage of the principle that the sum of a binary fraction and its bitwise inverse is approximately equal to one. The bitwise inverse of a binary fraction is obtained by inverting each bit of the binary fraction, and the operation on a binary fraction f will be referred to herein as inv(f). For example, decimal 0.625 expressed as a four digit binary fraction is 1010, and inv(1010) is 0101. When a binary fraction is added to its bitwise inverse, every bit in the result is a one, so the result is one minus the place value of the least significant bit, or approximately one.

[0030] Applying this principle, an inverted first sigma value can be generated by inverting each bit of the first sigma value in step 210, then the second sigma value, s2, is generated by adding u to the inverted first sigma value in step 220. Together, these two steps are approximately equivalent to subtracting u*v from u. This approximate equivalence exists for two reasons. The first reason is that based on the principle explained above, ((u*v)+inv(u*v)) approximately equals one, so (u+inv(u*v)) approximately equals (u+1−(u*v)). The second reason is that only the fractional portions of the binary numbers that we are dealing with are being used, and the fractional portion of (u+1−(u*v)) is equal to (u−(u*v)). Because we are only interested in the fractional portions of binary numbers, the addition operation in step 220 can be performed in more than one way. In one approach, the carryout from the high order bit can be ignored. In another approach, referred to as wraparound addition, the carryout from the high order bit can be added to the low order bit of the result. Using this latter approach is preferred because wraparound addition eliminates the approximation, i.e. when wraparound addition is used, (u+inv(u*v)) equals (u+1−(u*v)) without approximation.

[0031] The principle explained above can be used again to generate the third sigma value, s3. First, in step 230, v is added to the second sigma value to generate an inverted third sigma value. Then, in step 240, each bit of the inverted third sigma value is inverted to generate s3. To explain, s3 equals (1−u−v+(u*v)), which equals (1−(u+v−(u*v))), which is approximately equal to inv(u+v−(u*v)). If desired, a constant, for example, a constant having a value equal to the place value of the least significant bit of the binary fractions, could be added to the result of step 240 to eliminate the approximation.

[0032] Finally, the fourth sigma value, s4, can be generated by inverting each bit of s3, as in step 250, to regenerate the inverted third sigma value, then subtracting u, as in step 260.

[0033] Many alternatives to the method described above are possible within the scope of the present invention. In one alternative method, s4 is generated prior to s2 by adding v instead of u in step 220. Then, in step 230, u is added instead of v. Finally, in step 260, v is subtracted instead of u to generate s2 instead of s4. Another alternative method begins with step 200, but step 210 is replaced with a step wherein u is subtracted from s1 to generate an inverted second sigma value, and step 220 is replaced with a step wherein each bit of the inverted second sigma value is inverted to generate the second sigma value. A wraparound subtraction, wherein the carryin to the highest order bit of s1 is subtracted from the lowest order bit of the result, is preferred to eliminate the approximation in the same manner that wraparound addition eliminates the approximation in the original step 220. In another alternative method, s4 is generated prior to s3 by deleting steps 240 and 250, performing step 260, then performing a step wherein u is added to s4 to generate an inverted third sigma value, then performing a step wherein each bit of the inverted third sigma value is inverted to generate s3.

[0034] Another advantage of the present invention is that each step involved in calculating the sigma values uses only the result generated by the previous step and perhaps a constant or one or both of the original variables, u and v. In other words, the product term generated by the single multiply operation does not have be saved in a separate register. To extend the benefit of using only one register to calculate the sigma values to the calculation of a color value for point 100, a running sum of the color value can be maintained as described below in conjunction with FIG. 2.

[0035] In step 201, a known color value for point 101, for example, RA, the red color value for point 101, is multiplied by s1 to generate RP_A, a first red partial color value. In step 221, RB, the red color value for point 102, is multiplied by s2 to generate RP_B, a second red partial color value. In step 222, RP_Bis added to RP_Ato generate RP₁, a first red running sum color value. In step 241, RC, the red color value for point 103, is multiplied by s3 to generate RP_C, a third red partial color value. In step 242, RP_Cis added to RP₁to generate RP₂, a second red running sum color value. In step 261, RD, the red color value for point 104, is multiplied by s4 to generate RP_D, a fourth red partial color value. In step 262, RP_Dis added to RP₂to generate RP, the red color value for point P.

[0036] When more than one color value is to be calculated, for example, in a typical computer system using an RGB color model, each of the red partial color values can calculated in parallel with the corresponding green and blue partial color values. Therefore, in step 201, GA, the green color value for point 101, is multiplied by s1 to generate GP_A, a first green partial color value, and likewise for blue. In step 221, GB, the green color value for point 102, is multiplied by s2 to generate GP_B, a second green partial color value, and likewise for blue. In step 222, GP_Bis added GP_Ato generate GP₁, a first green running sum color value, and likewise for blue. In step 241, GC, the green color value for point 103, is multiplied by s3 to generate a GP₃, a third green partial color value, and likewise for blue. In step 242, GP₃is added to GP₁, to generate GP₂, a second green running sum color value, and likewise for blue. In step 261, GD, the green color value for point 104, is multiplied by s4 to generate GP_D, a fourth green partial color value, and likewise for blue. In step 262, GP_Dis added to GP₂to generate the GP, the green color value for point P, and likewise for blue. Parallel calculation of the partial color values and of the running sum color values is preferred to reduce the total latency of the interpolation method.

[0037] To facilitate these parallel calculations of the partial color values, three copies of the sigma values can be calculated in parallel. In one embodiment, an identical sequence of operations is performed in parallel on the contents of three separate registers in order to obtain three identical copies of each sigma value. In another embodiment, a sequence of operations is performed on the contents of one register that is divided into at least three groups of bits, where an identical copy of each sigma value is obtained in at least three of the groups of bits. However, the method of present invention is not limited to calculating three color values for a point in parallel, but can be used to calculate any number of color, or other values for a point, in parallel or in series. Therefore, any number of copies of the sigma values can be calculated, in parallel or in series.

[0038] FIG. 3 is a block diagram of a computer system for according to one embodiment of the present invention. Computer system 300 includes output device 310, storage device 320, and processor 330. Processor 330 is coupled to both output device 310 and storage device 320.

[0039] Output device 310 can be any device for outputting data from a computer system, such as, but not limited to, a cathode ray tube display, a liquid crystal display, any type of display, or a printer. In embodiment of FIG. 3, output device 310 is a device configured to display image 311 representing a pixel map. In the pixel map, each pixel is represented by at least two binary coordinates as described above.

[0040] Storage device 320 can be any device for storing data in a computer system, such as, but not limited to, read only memory, random access memory, flash memory, or a magnetic or optical storage device. In one embodiment, storage device 320 has stored therein routine 321 for interpolating a value for a pixel in the pixel map. In one embodiment, routine 321 is data representing a sequence of instructions.

[0041] Processor 330 can be any type of computer processing unit, such as, but not limited to, a processor supporting a complex or a reduced instruction set computer architecture or a very long instruction word architecture. Processor 330 can be implemented in any form, including one or more integrated circuits. In the embodiment of FIG. 3, processor 330 includes instruction decode unit 331, register unit 332, and execution unit 333. Processor 330 can also include additional circuitry. Decode unit 331 is used for decoding instructions received by processor 330 and into control signals and/or microcode entry points. Register unit 332 is used for storing data, such as the u, v, sigma, and color values described above. Execution unit 333 is used for executing instructions, such as instructions corresponding to the bitwise inverse and the wraparound add operations described above, in response to control signals and/or microcode entry points generated by decode unit 331. Register unit 332 can be the source and/or the destination for the data upon which these instructions are executed. Decode unit 310, register unit 320, execution unit 330, and any other circuitry within processor 300 can be implemented using any well known technique.

[0042] Routine 321, when executed by processor 330, causes processor 330 to perform the steps of any method for interpolating a value described above, or any other method equivalent to any method described above. Accordingly, processor 330 includes circuitry for executing the steps of at least any one of the methods for interpolating a value described above, or at least one of any other method equivalent to any method described above.

[0043] The present invention can also be embodied in a machine-readable medium, such as, but not limited to, any type of magnetic or optical data storage media including a removable disk, where the medium is storing data representing a sequence of instructions, which when executed by a processor, causes the processor to perform the steps of any method for interpolating a value described above, or any other method equivalent to any method described above.

[0044] Thus, the exemplary embodiments of the present invention illustrated by FIGS. 2 and 3 have been described. However, the invention is not limited to these embodiments or any of the details described. For example, the interpolation is not limited to two dimensions, for the invention can be practiced in an embodiment wherein a value for point 100 is interpolated from values corresponding to points surrounding point 100 in three or more dimensions. Also, the value to be interpolated for point 100 is not limited to an RGB color value, for the invention can be practiced in an embodiment wherein the value to be interpolated can be any other color value or any other value corresponding to point 100. Also, the interpolation is not limited to a use wherein the value corresponds to a pixel in a pixel map to be displayed, for the invention can be practiced in an embodiment wherein the value can correspond to any point, such as a point in a bitmap to be printed. The specification and drawings must be regarded in an illustrative rather than a restrictive sense. The scope of the invention is defined by the following claims.