JP2011175542A - Image data processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image data processor that efficiently encodes and decodes depth values. <P>SOLUTION: The image data processor includes a floating point separation unit separating floating point data into two parts, i.e. an exponential part serving as a first part and an imaginary part serving as a second part, the floating point data indicating the depth value of one pixel in three-dimensional graphics description; a difference generation unit for computing a first difference value, i.e., the difference between a predicted value of the first part and the value of the first part, as integer data, and computing a second difference value, i.e., the difference between a predicted value of the second part and the value of the second part, as integer data; and an encoder encoding the first difference value and the second difference value into variable-length codes. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present disclosure relates generally to a data processing device, and more particularly to an image data processing device that encodes and decodes image data.

  In recent years, higher drawing performance and drawing quality have been required for three-dimensional graphics drawing LSIs (large scale integrated circuits). This is because the display screen size has been enlarged due to full high-definition or the like, and drawing has been performed at a high frame rate in order to make the movement look smooth. In order to realize this high drawing performance, it is necessary to draw a large number of objects per unit time. In order to realize high drawing quality, it is necessary to use a plurality of textures at the same time or to draw an area larger than the display area for anti-aliasing. In order to realize these things, more memory access is required, but the memory bandwidth becomes tight due to the increase in memory access, which becomes a bottleneck for high performance. In order to avoid this, there is a method of increasing the memory bus width to increase the amount of transfer data per clock, or increasing the memory clock speed to increase the number of memory transfers per hour. However, in order to widen the bus width, it is necessary to increase the number of memories to be connected, and in order to increase the clock speed, it is necessary to use a higher performance memory. Become.

  FIG. 1 is a diagram showing the principle of 3D graphics drawing. In drawing three-dimensional graphics, a three-dimensional object to be drawn is divided into triangles or quadrangles called polygons, and drawing is performed in units of polygons. At that time, the polygon 10 is arranged in the three-dimensional space so that the Z-axis direction of the XYZ coordinates is the depth direction seen from the viewpoint, and the polygon image 11 seen from the viewpoint position is projected onto the XY plane as a display screen. For each polygon, only the pixel closest to the viewpoint is drawn. Portions hidden by other polygons are not drawn. As a result, a scene visible from the viewpoint is drawn, and the drawing result is displayed on the display. When drawing a pixel constituting each polygon, a depth value and a depth buffer are used to determine whether the pixel is closest to another polygon already drawn. The depth value is expressed as a Z coordinate in a three-dimensional space. The depth value of each drawn pixel is stored in the depth buffer. If the depth value of the target pixel to be newly drawn is a value indicating the depth behind the value of the corresponding pixel position in the depth buffer, it is determined that the target pixel is hidden by another polygon, and the pixel color is not drawn. . On the other hand, if the depth value of the pixel of interest is a value indicating the front side of the value of the corresponding pixel position of the depth buffer, the pixel of interest is determined to be visible from the viewpoint, and the pixel color is drawn and the value of the depth buffer Update. Since the determination of the presence / absence of drawing based on the depth value (depth test) is performed for each pixel to be drawn, many memory accesses are required. Reducing the amount of memory access that occurs in this depth buffer access is a major issue.

Japanese Patent Laid-Open No. 2004-212621 JP 2006-186531 A

  In view of the above, an image data processing apparatus that efficiently encodes and decodes depth values is desired.

  Image data processing apparatus separates floating-point data indicating a depth value of one pixel of three-dimensional graphics rendering into two parts: a first part of an exponent part and a second part of a sign and mantissa part And a first difference value that is a difference between the predicted value of the first part and the value of the first part as integer data, and the predicted value of the second part and the value of the second part A difference generation unit that calculates a second difference value that is a difference between the first difference value and the second difference value as an integer data; and an encoder that performs variable-length encoding on the first difference value and the second difference value. To do.

  According to at least one embodiment of the present disclosure, since an integer operation is performed instead of a floating point operation, an error caused by the operation does not occur. Further, since the processing is performed separately for the exponent part and the signed mantissa part, efficient encoding can be realized.

It is a figure which shows the principle of three-dimensional graphics drawing. It is a figure which shows a mode that floating point data is isolate | separated into two parts, the 1st part of an exponent part, and the 2nd part of a code | symbol and mantissa part. It is a figure which shows an example of the calculation of a difference value. It is a figure which shows the bit number expansion at the time of difference calculation. It is a figure which shows the division | segmentation of integer data. It is a figure which shows the structure of graphics LSI and memory. It is a flowchart which shows the flow of a depth data process in a polygon drawing process. It is a figure which shows the replacement process of drawing & depth data RAM. It is a flowchart which shows the flow of the block encoding process of step S1 of FIG. It is a flowchart which shows the flow of the block decoding process of FIG.8 S4. It is a flowchart which shows the flow of the block data output process of FIG.8 S2. It is a figure which shows the structure of memory. It is a flowchart which shows the flow of the block data reading process of step S3 of FIG. It is a figure which shows the structure and flow of a process which perform the data encoding process shown to FIG.8 S1. It is a figure which shows the structure and flow of a process which perform the data decoding process shown to FIG.8 S4.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

  In the case of data representing pixel values (pixel colors) instead of the depth values stored in the depth buffer, for example, differential compression can be considered as a compression method. The color of one pixel is expressed by three colors of an R (red) component, a G (green) component, and a B (blue) component, and the gradation of each color component is, for example, an 8-bit integer value (0 to 255). Represented. Since the image has a very high correlation between adjacent pixels, when the difference value for each color component of the adjacent pixel is calculated, the difference value is distributed close to zero. Image data can be compressed by Huffman encoding this difference value.

  On the other hand, the depth value stored in the depth buffer is represented by 32-bit floating point data. Even in the case of depth values, the correlation between adjacent pixels is very high, so it may be possible to apply differential compression. However, if the difference between floating-point data is calculated, the calculation error due to the loss of the lower bits during the calculation Will occur. In the case of differential compression, the difference values are cumulatively added at the time of decompression, so that errors are also accumulated. When an error occurs in the depth value of the drawing pixel, the pixel to be hidden is drawn or the pixel to be drawn is hidden, resulting in a completely different drawing result from that intended by the user. Therefore, an error is not allowed to occur during the compression process of the depth value data. Therefore, it is not realistic to perform the difference calculation of floating point data.

  It is also conceivable to treat the bit pattern of the 32-bit floating point display as a 32-bit integer, calculate the difference between the integers, and perform data compression. However, when a depth value originally represented by a floating-point number is treated as integer data as it is, the correlation between pixels is lowered. Therefore, the encoding efficiency (compression rate) is deteriorated. In particular, when calculating the difference between a positive value and a negative value, the difference may be a large value. This is because the exponent part is expressed in the form of a positive integer biased by 127, and the actual exponent part is obtained by subtracting 127 from the value. For example, -1.0 is expressed as 0xBF800000, and +1.0 is expressed as 0x3F800000, which is a calculation between -127 and +127, and the difference value is a large value. Further, when the depth value is divided into four 8-bit elements in the same manner as the pixel value, the divided portion is in the middle of the exponent part, and a large difference value is generated when the difference between the elements is calculated.

  Therefore, the floating-point data indicating the depth value of one pixel of the three-dimensional graphics drawing is separated into two parts: a first part of the exponent part and a second part of the sign and mantissa part. Next, the first difference value, which is the difference between the predicted value of the first part and the value of the first part, is calculated as integer data, and is the difference between the predicted value of the second part and the value of the second part. The second difference value is calculated as integer data. Further, the first difference value and the second difference value obtained as described above are subjected to variable length coding. Hereinafter, these processes will be described.

  FIG. 2 is a diagram showing how floating point data is separated into two parts, a first part of the exponent part and a second part of the sign and mantissa part. As shown in FIG. 2, the 32-bit floating point data 20 includes a 1-bit code 21, an 8-bit exponent part 22, and a 23-bit mantissa part 23. Of these, the exponent part 22 is the first part 24 of an 8-bit integer. The 1-bit code 21 and the 23-bit mantissa part 23 are combined to form a second part 25 of a 24-bit signed integer. In this way, the 32-bit floating point data is decomposed into two integers.

  FIG. 3 is a diagram illustrating an example of the calculation of the difference value. FIG. 3 shows first portions E0 to E3 and second portions F0 to F3 of four pixels that are continuous in a predetermined direction (for example, the scanning direction). The first-order differences between the first portions E0 to E3 of the four pixels are dE0 to dE2, and the first-order differences between the second portions F0 to F3 of the four pixels are dE0 to dE2. For example, dE1 = E2-E1. Further, the difference between dE0 to dE2, that is, the second difference between the first portions E0 to E3 is d2E0 to d2E1, and the difference between dF0 to dF2, that is, the second difference between the second portions F0 to F3 is d2F0 to d2F1. It is. For example, d2E1 = dE2-dE1.

  For example, it is assumed that the value of the second part of the depth value of the pixel of interest is F2, and the value of the second part of the depth value of the previous adjacent pixel is F1. At this time, the predicted value of the target pixel obtained by the prediction determination that the same value will continue is F1, and the difference F2-F1 between the predicted value F1 and the value F2 of the second portion of the target pixel is calculated. . This is a first-order difference, and the predicted value in the case of the first-order difference is a depth value of an adjacent pixel. The adjacent pixel is not limited to the previous pixel (one pixel left if the scanning direction is left to right), but may be, for example, one pixel above.

  Further, the value of the second portion of the depth value of the pixel of interest is F2, the value of the second portion of the depth value of the previous adjacent pixel is F1, and the second of the depth values of the next adjacent pixel. Assume that the value of the part is F3. At this time, the predicted value of the target pixel obtained by the prediction determination that the value changes with a constant increase or decrease is (F3 + F1) / 2, and the predicted value and the value F2 of the second portion of the target pixel are The difference (F3 + F1) / 2−F2 is calculated. If this difference is doubled, F3 + F1-2F2 is obtained, which is equal to (F3-F2)-(F2-F1). This is a second-order difference, and the predicted value in the case of the second-order difference is an average value of the values F1 and F3 of the peripheral pixels of the target pixel. The peripheral pixels are not limited to the immediately preceding pixel and the immediately following pixel, but may be, for example, 8 surrounding pixels.

  The depth value often shows a change (that is, a uniform slope) having a constant increase or decrease, and if a second-order difference is taken as described above for the second part corresponding to the mantissa part, the majority The values will be distributed around zero. Therefore, for example, if a Huffman coding table assuming a normal distribution centered on zero is prepared, a code with a short code length is assigned to a value near zero, so that highly efficient coding can be realized. it can. The exponent part is often a constant value, and if a second-order difference is taken, the majority of the values are similarly distributed around zero. Therefore, if the second-order differential value is similarly variable-length encoded, highly efficient encoding can be realized. As for the exponent part, even if only the first-order difference is taken, the majority of the values are distributed in the vicinity of zero. Therefore, the result of the first-order difference may be encoded. In such prediction processing and encoding processing, integer arithmetic is performed instead of floating point arithmetic, so that an error caused by the arithmetic does not occur. Further, since the processing is performed separately for the exponent part and the signed mantissa part, efficient encoding can be realized.

  FIG. 4 is a diagram illustrating the bit number expansion during the difference calculation. When performing the difference calculation as described above, the first part, which is the exponent part, requires a 1-bit extension to express a negative calculation result when the first-order difference is used, and the calculation result is displayed as a digit when the second-order difference is used. In order to express without overflow, it is necessary to extend one bit further. Further, for the second part, which is a signed mantissa part, 1-bit extension is required to express the calculation result without overflowing in each of the first-order difference and the second-order difference. Therefore, as shown in FIG. 4, for the first part which is the exponent part, the original 8-bit integer data becomes 9-bit integer data after the first-order difference and becomes 10-bit integer data after the second-order difference. For the second part, which is a signed mantissa, the original 24-bit integer data becomes 25-bit integer data after the first-order difference and becomes 26-bit integer data after the second-order difference.

When performing variable length encoding, it is not efficient to encode data with a long number of bits. In variable-length coding such as Huffman coding, generally, a variable-length coding table in which data to be coded and coded data after coding correspond one-to-one is used. When data to be encoded is given, encoding is performed by referring to the encoding table to obtain corresponding encoded data. In this case, an attempt to directly encode the 26-bit integer, the number of entries coding table becomes 2 ^26, thereby using a large memory space. Therefore, it is preferable to divide the 26-bit integer data into a plurality of data having a smaller number of bits and separately perform the variable length coding on the plurality of divided data. The exponent data may also be finely divided as necessary.

  FIG. 5 is a diagram showing division of integer data. As shown in FIG. 5, for example, 26-bit integer data, which is second-order difference data between the sign part and the mantissa part, is divided into three data lengths of 9 bits, 9 bits, and 8 bits from the upper order that are short. . The combination of the number of bits is not limited to this division example, and other combinations of the number of bits may be used. For example, you may divide into four like 7 bits, 7 bits, 6 bits, and 6 bits.

  FIG. 6 is a diagram illustrating a configuration of the graphics LSI and the memory. In FIG. 6, the part related to the processing of the depth data is mainly shown, and the whole processing part for the drawing data (pixel color data and the like) is not shown. The graphics LSI 60 in FIG. 6 is connected to the external memory 66 via a bus. The graphics LSI 60 includes a graphics drawing core 61, a drawing & depth data RAM 62, an expansion circuit unit 63, a compression circuit unit 64, and a memory controller unit 65. The external memory 66 includes a depth buffer 67 and a block length buffer 68.

  The encoded data of the depth value of each pixel (data separated into the first part and the second part) is stored in the depth buffer 67 of the external memory 66. The most recently accessed depth value is stored in the drawing & depth data RAM 62 in the graphics LSI 60. The drawing & depth data RAM 62 plays a role similar to that of a cache memory in which a part of data is kept at hand. The drawing & depth data RAM 62 stores data of some small areas of the entire image corresponding to the display screen. For example, the depth value and drawing color data for a block of 16 × 16 pixels are stored. The Depth value encoded data is stored in the depth buffer 67 of the external memory 66 in units of blocks. As a result of the encoding, the number of bits of the encoded data of the depth value for the block of 16 × 16 pixels is not constant, so that data indicating the length of effective data for each block is stored in the block length buffer 68. By a series of drawing processes by the graphics LSI 60, drawing color data is written in a predetermined area of the external memory 66 for all pixels of the image corresponding to the display screen.

  FIG. 7 is a flowchart showing the flow of depth data processing in polygon rendering processing. When drawing processing of one polygon is started, the polygon figure is decomposed into drawing pixels in step S1. That is, the pixels that enter the polygon are specified, and these pixels are scanned in a predetermined order. In this manner, a predetermined drawing process is executed for each pixel that is sequentially scanned. In step S2, a drawing position, a drawing color, and the like are generated for the pixel of interest. For example, by scanning one pixel at a time in the horizontal direction from a pixel located on the side connecting the vertices of the target polygon, the pixel position of the target pixel to be processed is determined. Further, based on the image parameter obtained for each vertex of the target polygon, the image parameter of each pixel included in the polygon is obtained by interpolation. For example, scanning is performed pixel by pixel in the horizontal direction as described above, and image parameter such as depth value and drawing color in each pixel is obtained by adding the increment of the parameter value in the horizontal direction. The image parameters include the luminance of each RGB color, a depth value representing a distance in the depth direction, a texture coordinate value for texture display, and the like. The processing in steps S1 and S2 is executed by the graphics drawing core 61.

  In step S <b> 3, it is determined whether or not a block including the target pixel position is stored in the drawing & depth data RAM 62. When the block including the target pixel position is stored in the drawing & depth data RAM 62, the process proceeds to step S5, where depth value comparison and pixel writing are performed. If the block including the pixel position of interest is not stored in the drawing & depth data RAM 62, a drawing & depth data RAM replacement process is executed in step S4. As a result, the block data (depth value and drawing color data) currently stored in the drawing & depth data RAM 62 is stored in the external memory 66, and the block data including the target pixel position is read out from the external memory 66. The drawing & depth data is stored in the RAM 62. Here, the process of storing the data currently stored in the drawing & depth data RAM 62 in the external memory 66 is executed in order to reflect the update process performed on the local data in the data on the external memory 66. After this replacement process, the process proceeds to step S5, where depth value comparison and pixel writing are performed.

  The depth value comparison and the pixel writing process are executed by the graphics drawing core 61. In this depth value comparison and pixel writing process, the depth value of the target polygon calculated for the target pixel is compared with the depth value of the already drawn data (the depth value stored in the drawing & depth data RAM 62). . If the target polygon is positioned in front of the drawn data, the drawing process and the depth value update process are performed. That is, the drawing color generated in step S2 is written as the drawing color of the pixel of interest in the block stored in the drawing & depth data RAM 62 (overwriting existing drawing color data). Furthermore, the depth value generated in step S2 is written as the depth value of the pixel of interest of the block stored in the drawing & depth data RAM 62 (overwriting existing depth data). If the target polygon is located behind the already drawn data, the drawing process and the depth value update process are not performed.

  In step S6, it is determined whether or not there is an undrawn pixel in the polygon graphic. If there are undrawn pixels, the process returns to step S2 and the subsequent processing is repeated. If there is no undrawn pixel, the drawing process for the target polygon is terminated.

  FIG. 8 is a diagram illustrating a process of replacing the drawing & depth data RAM. In step S1, the compression circuit unit 64 in FIG. 6 executes block coding processing. That is, the above-described processing of floating point data separation, differential data generation, and variable length encoding is executed on the block of depth values stored in the drawing & depth data RAM 62. In step S2, the memory controller unit 65 of FIG. 6 executes block data output processing. That is, the memory controller unit 65 stores the encoded data generated by the compression circuit unit 64 in a predetermined block storage position in the depth buffer 67 and stores the block length data in the block length buffer 68.

  In step S3, the memory controller unit 65 in FIG. 6 executes block data read processing. That is, the memory controller unit 65 reads data having a length indicated by the block length of the block length buffer 68 from the head of a predetermined block storage position in the depth buffer 67 for the block including the target pixel. In step S4, the decompression circuit unit 63 in FIG. 6 executes block decoding processing. That is, the encoded data read in step S3 is decoded to generate difference value data, the first part value and the second part value are calculated based on the difference value data, and the first part (exponent) Part) and the second part (signed mantissa part) are combined to restore the floating point data. The restored floating-point format depth value block is stored in the drawing & depth data RAM 62.

  As for the drawing color data, a predetermined encoding process is performed on the drawing color data in the drawing & depth data RAM 62 as necessary, and the drawing color data is written in a predetermined area of the external memory 66. Further, for the block including the pixel of interest, the drawing color data is read from the predetermined area of the external memory 66 into the drawing & depth data RAM 62. At this time, a decoding process is executed as necessary.

  FIG. 9 is a flowchart showing the flow of the block encoding process in step S1 of FIG. When the block encoding process is started, the data encoding process is executed in step S1. This executes the above-described floating point data separation, difference data generation, and variable length encoding processing on the depth value of each pixel of the block stored in the drawing & depth data RAM 62. In step S2, it is determined whether uncoded data remains in the block. If there is unencoded data, the process returns to step S1 and the process is repeated. If there is no unencoded data, the block encoding process is terminated.

  FIG. 10 is a flowchart showing the flow of the block decoding process in step S4 of FIG. When the block decoding process is started, the data decoding process is executed in step S1. This executes the above-described decoding, restoration of each data from the difference value data, and restoration of the floating-point data on the coded depth value of each pixel of the block read from the external memory 66. Is. In step S2, it is determined whether or not undecoded data remains in the read data. If there is undecrypted data, the process returns to step S1 and the process is repeated. If there is no undecoded data, the block decoding process is terminated.

  FIG. 11 is a flowchart showing the flow of block data output processing in step S2 of FIG. In step S1, the block data length is calculated. That is, the length (bit length, byte length, etc.) of the encoded data generated by the variable length encoding by the compression circuit unit 64 is calculated. In step S2, block data is written in a predetermined block position of the depth buffer 67 of the external memory 66. In step S3, the block data length is written. That is, data indicating the block data length calculated in step S1 is written into the block length buffer 68 of the external memory 66.

  FIG. 12 is a diagram illustrating the structure of the memory. A depth buffer 67 area and a block length buffer 68 area are provided in the memory space of the external memory 66. An area for writing drawing color data is also provided in the memory space of the external memory 66. The depth buffer 67 is divided into a plurality of block areas 121. Each block area has a predetermined fixed length, and the encoded data 125 is stored from the beginning of each block area. The block length buffer 68 is divided into a plurality of block data length storage areas 122. Each block data length storage area stores data indicating the data length of the encoded data 125 stored in the corresponding block. For example, data (block 2 data length) indicating the data length of the encoded data 125 (block 2 data) stored in the third block is stored in the third block data length storage area.

  FIG. 13 is a flowchart showing the flow of the block data reading process in step S3 of FIG. In step S1, the block data length is read. That is, for the desired block, data indicating the data length stored in the block length buffer 68 is read. In step S2, block data is read. That is, the data corresponding to the data length indicated by the data read in step S 1 is read from the corresponding block of the depth buffer 67.

  FIG. 14 is a diagram showing a configuration for executing the data encoding process shown in step S1 of FIG. 8 and a flow of the process. The configuration shown in FIG. 14 corresponds to the configuration of the compression circuit unit 64 in FIG. 6, and includes a Z value input unit 141, a floating-point separation unit 142, a first-order difference generation unit 143, a second-order difference generation unit 144, and a difference data division unit. 145, an encoding unit 146, and an encoding table 147. When the data encoding process is started, the Z value input unit 141 reads the depth value of the pixel of interest from the drawing & depth data RAM 62. The floating point separation unit 142 separates the floating point data of the depth value into two parts, a first part of the exponent part and a second part of the sign and mantissa part. The first floor difference generation unit 143 obtains a first floor difference separately for each of the first part and the second part. That is, the difference between the first portion of the pixel adjacent to the target pixel (previously read pixel) and the first portion of the target pixel is obtained, and the second portion of the pixel adjacent to the target pixel (previously read pixel) and the target A difference from the second part of the pixel is obtained. Data of the first part and the second part of adjacent pixels (pixels read last time) may be held in an internal register or the like. The second floor difference generation unit 144 obtains the second floor difference based on the first floor difference value separately for each of the first part and the second part. For example, the difference between the first-order difference value obtained for the pixel adjacent to the target pixel (the pixel read last time) and the first-order difference value obtained for the target pixel is obtained. Data of the first-order difference value of adjacent pixels (pixels read last time) may be held in an internal register or the like.

  The difference data dividing unit 145 divides the second-order difference value data obtained for at least the second portion into a plurality of data having a smaller number of bits. The encoding unit 146 refers to the encoding table 147 and performs variable-length encoding on the second-order difference value obtained for each of the first part and the second part. Note that the data of the first part and the second part of the first pixel of the block are used for the first-order difference calculation and output as encoded data as they are. This data may be output after being encoded. Further, the first-order difference value between the first pixel and the second part of the first pixel and the next pixel of the block is used for the second-order difference calculation and is directly output as encoded data. The first-order difference value may be output after being encoded.

  FIG. 15 is a diagram showing a configuration for executing the data decoding process shown in step S4 of FIG. 8 and a flow of the process. The configuration shown in FIG. 15 corresponds to the configuration of the decompression circuit unit 63 of FIG. , A Z value output unit 156, and a decoding table 157. When the data encoding process is started, the decoding unit 151 receives the encoded depth value (data separated into the first part and the second part) and refers to the decoding table 157 to perform decoding. Find the data. At this time, since the depth value of the first pixel and the first-order difference value of the first pixel and the next pixel are not encoded, the values as they are are sent to the floating-point combination unit 155 and the second-order difference addition unit 153, respectively. It is done. The difference data combining unit 152 generates second-order difference data of 26-bit integer data by combining data corresponding to the second portion of the data variable-length decoded by the decoding unit 151. Second floor difference addition section 153 generates the first floor difference value by adding the second floor difference value to the immediately preceding second floor difference value. The first floor difference adding unit 154 generates a first floor difference value by adding the first floor difference value to the immediately preceding first floor difference value. Thereby, exponent part data as the first part and signed mantissa part data as the second part are generated. These data are combined by the floating-point coupling unit 155 and converted into data in the floating-point format. The Z value output unit 156 outputs floating point data as a depth value. Note that the encoding table and the decoding table used for encoding and decoding can be optimized by preparing tables having different values for each processing target. Alternatively, the table capacity may be reduced by sharing the table capacity.

  By the above processing, the depth value expressed by a 32-bit floating point number can be compressed without error. As a compression ratio of the compressed data with respect to the uncompressed original data, 20% to 30% can be realized.

  As mentioned above, although this invention was demonstrated based on the Example, this invention is not limited to the said Example, A various deformation | transformation is possible within the range as described in a claim.

60 Graphics LSI
61 Graphics Drawing Core 62 Drawing & Depth Data RAM
63 Decompression circuit unit 64 Compression circuit unit 65 Memory controller unit 66 External memory 67 Depth buffer 68 Block length buffer

Claims (5)

A floating-point separator that separates floating-point data indicating the depth value of one pixel of three-dimensional graphics rendering into two parts, a first part of the exponent part and a second part of the sign and mantissa part;
The first difference value, which is the difference between the predicted value of the first part and the value of the first part, is calculated as integer data, and the difference between the predicted value of the second part and the value of the second part A difference generation unit that calculates a second difference value as integer data;
An image data processing apparatus comprising: an encoder that performs variable length coding on the first difference value and the second difference value.   The image data processing apparatus according to claim 1, wherein the predicted value for the target pixel is an average value of peripheral pixels of the target pixel.   The encoder includes a difference data dividing unit that divides at least the data of the second difference value into a plurality of data having a smaller number of bits, and the plurality of pieces of data divided by the difference data dividing unit are individually variable. 3. The image data processing apparatus according to claim 1, wherein the image data processing apparatus performs long encoding. A decoding unit that decodes data encoded by the encoder to generate the first difference value and the second difference value;
A difference adding unit that calculates the value of the first part and the value of the second part based on the first difference value and the second difference value decoded by the decoder;
2. The method according to claim 1, further comprising: a floating point coupling unit that restores the floating point data by coupling the value of the first part and the value of the second part calculated by the difference adding unit. 4. The image data processing device according to any one of items 3 to 3.   The image data processing apparatus according to claim 3 or 4, wherein the difference data dividing unit divides the data of the second difference value into three data.

Images