Dvc delta commands

Cross-Reference To Related Applications

[0001] Tills application claims priority to U.S. provisional application number 60/795,577, the entire contents □f which are incorporated herein by reference.

[0002] 'Phis application is also related to the following co-pending U.S. Patent application which is commonly owned with the present application, contents of which are incorporated herein by reference:

1. . U.S. Application No. 10/260,534 entitled “Video Compression System” filed on

October 1, 2002.

Field Of The Disclosure

[0003] This disclosure relates to a computer video compression system.

Introduction

[0004] A Video Compression Unit is disclosed herein that uses a compression scheme based on the directional algorithm concepts previously disclosed in Application 10/260,534, That algorithm, so called “DVC encoding,” is employed herein with some newly added extensions. The present application reduces the bandwidth used in transmitting a video frame buffer across an extension link. The contents of U.S.

application No. 10/260,534 are assumed to be known to the reader. Products employing the “DVC encoding” of U.S. application No. 10/260,534 have been commercialized and should be considered prior art.

[0005] One of the aspects of the “DVC encoding” algorithm is that each, side of the link always has a complete version of the previous frame to use as a reference. This allows each pixel in subsequent frames to be defined by commands:

1. No change from pixel in previous frame (NO_CHANGE)

2. Same as pixel in line above (COPY_ABOVE)

3. Same as pixel to the left (COPY_LEFT)

4. Series of pixels from a preceding known subset (MAKE_SERIES)

5. New pixel (NEW_P1XEL)

[0006] Only tlie New Pixel option requires that a complete pixel be sent across the link. The first three require only that a short command, message be sent Indicating which type of encoding is used and how many consecutive pixels are encoded according to that encoding type. During encoding, the pixel data for both the current frame being compressed and, if applicable, the previous frame are read from memory. The current pixel is then compared against a reference pixel: PreviousPixel (akin to COPY_LEFT)_־PreviousLine (akin to COPY_ABOVE), and PreviousFrame (akin to NO_CHANGE). For each of the three directional commands, if the command is active and the associated comparison matches, then the command remains active and the prospective set increases by one more pixel. When all directional commands have terminated, due to either failures or end conditions, then the last active command is chosen as the encoding for that set of pixels.

[0007] In the event of a tie, then priority can be assigned in the following order: NO_CHANGE, COPYJLEFT, COPY_ABOVE, for example. This is the order historically used by previous DVC-encoding products, where it was arranged in terms of ease of decoding. However, other orders can be used. With double or triple buffering on each end, all three commands require similar effort by the decoder.

]0008] A single copy of the previous pixel (PreviousPixel) is kept for doing the COPY_LEFT comparison, and a full line of pixel data (PreviousLine) is kept for doing COPY_ABOVE comparisons, PreviousFrame pixels are being supplied by the memory subsystem along with the Currentpixel.

[0009] Because NEWJPIXEL is the least efficient compression method, it is least favored and used only when the other compression types do not apply to a current pixel. Thus, a NEW_PIXEL determination always terminates a pixel encoding stream and sends the preceding command string for transmission and decoding. Then, NEWJPIXEL commands are accomplished on a pixel-by-pixel basis until another encoding type will again apply to a current pixel.

[0010] The MAKEJS ERIES encoding type takes advantage of a sequence of pixels all being from a subset of preceding unique pixel colors. The standard mode is to use a twocolor subset, which is ideal for text windows. This can be expanded up to four-colors or more (powers of two) depending upon the number of series comparators and registers the hardware implementation chooses to incorporate. A series comparator is required in the hardware for each pixel in the subset. As each pixel is processed (read from memory) it is compared against each of the pixels in the current scries subset registers. All the comparisons are done in parallel, with the results being sent to die command process. As long as any one (and it should be no more than one) of the subset comparators is true, then the Series command is valid and can continue.

[0011] These first five command types are what are referred to as the original DVCbased commands which are described in greater detail in Application 10/260,534.

[0012] This disclosure uses the original D VC-based commands in conjunction with more complex encoding commands. The more complex encoding commands are as follows:

6. Delta from the same pixel in the previous frame (DELTA_NC)

7. Delta from the pixel immediately above (DELTA_CA)

8. Delta from the pixel immediately preceding (DELTA_CL)

[0013] Delta commands are an alternative to sending the full precision color, instead sending a much smaller value which is the difference (delta) between the real color and one of the neighboring colors used as a reference. In the case of 24-bit color, the full precision color is 24-bits whereas the delta is either 4 bits or 12 bits. There are several different types of delta commands that could be implemented. Some include a unique delta value for each color channel, while others contain a uniform delta that is applied to all color channels. The size (in terms of bits) of each delta can also vary according to the configuration.

[0014] The video compression unit can also use “Color reduction” to reduce the bandwidth usage. Color reduction removes some number of least significant bits from each color channel, thus increasing the likelihood that neighboring pixels will appear “identical” to the comparators, and also reducing the size of data required to send an uncompressed pixel.

Brief Description Of The Presently Preferred Exemplary Embodiments [0015] The following description, given with respect to the attached drawings, may be better understood with reference to the non-limiting examples of the drawing, wherein the drawings show:

WO 20(17/127452

[0016] Figure 1: An exemplary Video Compression System;

[(1017] Figure 2: An exemplary Input Message Cell Format;

[0018] Figure 3: An Exemplary Read Command Cell Format;

]0019] Figure 4: An exemplary encoding Header Cell Format;

[0020] Figure 5: Exemplary 8-bit Cells for the original DVC commands;

[0021] Figure 6: Exemplary 8*bit Cells for the Delta Commands;

[0022] Figure 7: Exemplary Delta Modes;

[0023] Figure 8: Cells incorporating exemplary Delta Modes;

[0024] Figure 9: An exemplary Comparison block; .

[0025] Figure 10: Exemplary Command Cell Formats and Table of Commands;

[0026] Figure 11: Exemplary Color Depth Command Cell and Color Depth Mode Table;

[0027] Figure 12: Exemplary Command Cells;

[0028] Figure 13: Exemplary multi-byte Command Cells;

[0029] Figure 14: An exemplary Command Data Cell; and

[0030] Figure 15: Exemplary table of Core Clock Rates.

The Presently Preferred Exemplary Embodiments

[0031] Figure 1 shows an exemplary Video Compression System 100 of a video system.

[0032] The Video Compression System 100 includes, in part, Command Process 102 and an Output Process 103. The Command Process 102 is the main process that responds to input from the Digital Video Input (DVI) Unit 104 and then starts processing a frame by making a request to the Memory Controller 106. Output Process 103 generates the output for the rest of the system. The Command Process 102 comprises a Message Interface 105 for receiving messages from the DVI Unit 104 and a Memory Interface 107 for communicating ־with Memory Controller 106.

[0033] The Message Interface 105 of the Command Process 102 initially receives an Input Message 101 from the DVI Unit 104. The message is a FIFO 8-bit message written by the DVI Unit 104. The most common messages fit into a single byte, while the rather infrequent. timing values take multiple bytes and therefore multiple clocks.

[0034] ^,The basic message header (the first byte of any message) of the input message 101 is shown in Fig 2. The message header has a four-bit Type field (7:4) and a four bit Data field (3:0). The Type field specifics one of the following message types: Start Frame (0001), End Frame (0010), Horizontal Timing (0100), Vertical Tuning (0101), and Pixel Clock Rate (0110). It should be noted that although five message types are currently defined more message types can be defined using the “Reserved” bit values. [0035] The Data field of the header provides message information in accordance with, the type of message. If the message type is a Start Frame or End Frame message, the message is contained within the Data field of the header.

[0036] Video Timing messages (Horizontal Timing, Vertical Timing, or Pixel Clock Rate) require 5 or 6 bytes of data after the message header. For these messages, the Data field of the header specifies the number of bytes to follow. The data structures for each type of Video Timing messages are shown in Figure 2. The bytes are transmitted least significant byte and bit first, so using the Horizontal Timing as a reference bit 0 of ActiveData maps to bit 0 of the byte 0. ActiveData bit 7 maps to bit 7 of byte 0. ActiveData bit 8 maps to bit 0 of byte 1.

[0037] The Memory Interface 107 handles two types of data: a read command 108 sent to memory and the subsequently returned data (pixel data). The Command Process 102 uses a read command 108 to send a request to start reading pixel data from the current and previous frames and specify the size of the frame via width and height in pixels. The read command 108 is shown in Fig. 3 and contains Command type, Frame Width, FrameHeight and two frame ID fields (one for the identification of the oldest previous frame to access and the other for the identification of the most current frame to access). These fields specify which pixels should be sent to the Command Process 102.

[0038] Once the Memory Controller 106 has received a read command 108 identifying pixels, the Memory Controller 106 should start filling the current data FIFO 110 and previous data FIFO 112 with pixel data from the two source frames. The data returned is assumed to be full 24-bit pixel data, and will be subsequently reduced by the Video Compression System 100 if necessary for lower color depths. The Memory Controller 106 writes the returned data directly to the current data FIFO 110 and previous data FIFO 112. The Command Process 102 monitors the empty flags of both current data FIFO 110 and previous data FIFO 112 to determine when each has pixel data to process. The flags may be combined using a Boolean OR operation to form one signal.

[0039] Once the Command Process 102 determines that there is pixel data to process, it then begins processing a frame of data (a set number of pixels) as it is available. The Command Process 102 is the core that makes the decisions as to which encoding to perform and sends the resulting decision, per active clock, to the Output Process 103.

This decision may be any of the following:

1. Non-decision in the case where the pixel has been encoded into an actively-running command (nothing is sent to the Output Process 103 at this point)

2. A request to store a value in the output buffer (pixel, delta, or series)

3. A request to generate command output directly (typically completion of a directional command)

4. A request to generate a command, copying data from one of tire output buffers (completion of a pixel, delta, or series-based command).

[0040] Two types of encoding the Command Process 102 can perform are shown in Figs. 5-6. As in DVC-based encoding, the video packets are based on 8-bit cells, but other cell numbers can be substituted. There are several types of header cells defined in accordance with the commands shown in Figs. 5-6, but each has the following basic format for the first byte (the header) which is shown in Fig. 4.

[0041] The first three bits (7:5) of the header are Command bits that specify the command. The fourth bit (4) is the Ext bit. Whep a header has the Ext bit set, then it is guaranteed that the command has at least one more byte to follow. The main purpose of the Ext bit is to extend the size of the count field for commands that contain a count i.e applying a command to be applied for a greater number of pixels than can be specified with four bits. The basic format of an Extension byte is shown in Fig. 4. The byte will have an additional Ext bit and 7 count bits. For these commands, subsequent cell (s) contain additional most significant bits of the count.

[0042] Thus, a single-byte command, which is limited to 16 (2⁴) pixels, can be extended with a two-byte command to 2048 (2¹¹) pixels, with a three-byte command to 262144 (2¹⁸) pixels, and four bytes (the most that should be required) to over 33 million pixels. Since four bytes is the most that should be required with current video coloration and video resolutions the command is typically limited to a maximum number of four bytes. Another reason for limiting the command to a maximum of 4 bytes is for the convenience of being able to write the command to a 32-bit wide FIFO in one cycle. It should be noted that although the command is typically limited to four bytes, the command can be extended to more bytes if necessary without departing from the scope of the present invention.

[0043] Fig. 5 shows the packet format of the original-DVC commands.

[0044] NO_CHANGE (NC) (Command = 000) specifies that the Count number of consecutive pixels have not changed since the previous frame, and thus may bo copied from the previous (reference) frame,

[0045] COPY_LEFT (CL) (Command = 001) specifies that the Count number of consecutive pixels are identical to the pixel immediately to their left (the previous pixel). [0046] COPY_ABOVE (CA) (Command =010) specifies that the Count number of consecutive pixels are identical to the pixel immediately above them (the previous line). [0047] MAKE_SERIES (MS) (Command = 011) represents a series of pixels of Count length that are each one of two possible colors. This is typically very useful for a window of single color text on a solid background. One data bit for each pixel represented specifies whether that pixel is the first or the second of the designated colors. In one embodiment, the designated colors are the two most recent and different colors encountered in a line. This can be the color of the immediately preceding pixel to the series (selected when a data bit is clear) and the color of the last pixel before it that was a different color (selected when a data bit is set). This can be easily extended to a larger depth by using multiple data bits to represent each pixel. The depth is defined by the implementation and is communicated downstream via the appropriate INFO command (Figure 10).

[0048] Due to the size of the output holding buffer, the maximum series depth is typically 384 bits, or 12 four-byte words. This permits support for up to 384 pixels at a depth of 2 colors, 192 at 4 colors, 96 at 8 colors, etc. The maximum series length is limited to 256 in this implementation. It should be noted that the maximum series depth can be extended if the size of the output holding buffer is increased.

[0049] Since MakeSeries does not offer the most efficient compression, (that is the number of pixels that can be represented per byte) compared to the directional run-length encodings (NC, CL and CA), there are times when it is better to terminate a MakeSeries and switch to a directional encoding instead. It is not always easy to determine ahead of time whether it is better to terminate the MakeSeries or to stay in it. The present embodiment keeps a count of the consecutive NC, CL, and CA pixel comparisons during all operations, including MakeSeries. The counts are reset at the start of a MakeSeries, and then potentially after every 8 pixels processed by the MakeSeries command.

[0050] A directional command of 8 or more pixels is guaranteed to be no worse than another 8 pixels tacked on to the end of a MakeSeries. But, if the next pixels after those 8 can only be encoded with the same MakeSeries that was interrupted, then the interruption has actually required an extra MakeSeries header of at least one byte. For an interruption and subsequent resumption of MakeSeries, the set needs to be more than 32 pixels in length to assure that it is better to interrupt the MakeSeries. This allows for a potential restart using a 2-byte MakeSeries header for a count greater than 16.

[0051] One embodiment makes the decision based on a set size of 16. This leaves open the possibility of switching into and out of MakeSeries, but since that is expected to be a very infrequent occurrence, the simplification should be worth it.

[0052] After every 8 pixels processed by MakeSeries, the directional lengths will be checked. If any length is 16 or more, then a directional set could have processed the last 16 pixels. The MakeSeries count is reduced by 16, and the following directional command is seeded with a starting count of 16. This needs to be based on the 8-pixel boundaries within a MakeSeries. Clearing the directional check at the start of each byte if it is less than 8 takes care of this by eliminating any partial bytes.

[0053] The NEW_PIXEL command specifies the complete color values for a series of pixels that could not be otherwise compressed. The amount of data is equal to the number of bytes of color per pixel times the number of pixels encoded in this command. No attempt is made to extend this command beyond support for 16 pixels due to the amount of buffering required for the algorithm to buffer up pixels and then go back and fill in the header’s count at a later time. Thus, NEWP1XEL with the Ext set is currently unused.

[0054] In one example implementation, when the Video Compression System 100 is operating in 7־bit color mode, the NewPixel command has a set count of 1 and a NewPixel header cell is required for each pixel of data. ^,This simple header cell is shown as “Single Plxel-7” in Fig. 5. This can result in up to a 2X bandwidth impact for 8-bit data, or 33% additional overhead for 24-bit data. The protocol therefore allows for a count to be used so that an implementation can choose to buffer up or otherwise delay sending the pixel data until a count can be determined. Again, imposing a limit of 16 pixels of data alleviates the buffering requirements and latency.

[0055] It should be noted that the PreviousLine buffer 130 may provide this capability if it is designed with two independent read ports. Otherwise a separate 16x24-bit FIFO will be required.

[0056] Due to the header format, 7־bit mode docs not.support NewPixel runs or the additional Delta Commands described below.

[0057J Fig. 6 shows the additional Delta commands. These command formats shown in Fig. 6 are tailored to 24-bit and 15-bit color modes.

[0058] The Delta commands attempt to limit the number of bits transmitted when the directional-based commands fail and an inefficient NEW_PIXEL has to be sent, The Delta commands operate under the assumption that the pixel may be close in color to one of its neighbors. Rather than sending the NEW_PIXEL, a simple reference to that neighbor (via the command field) and a signed color delta may reduce the data to as little as 25% of what would otherwise have been required for the NEW_PIXEL command, if the pixel is close in color to one of its neighbors.

[0059] The Delta commands as described below can either specify the difference in color between two pixels as the absolute difference (the four-bit signed difference . between two 24-bit numbers) or the difference for each color channel (three four-bit differences for each color). As shown in Fig. 6 when the Ext bit is not set (a “0” in bit 4), a single delta value is used as the signed difference between the 24-bit string of the current pixel and the previous pixel. When the Ext is set (a “1” in bit 4), a delta value for each color channel is provided. Delta׳values are added to respective values from the pixel which is compared to the cunent pixel to generate the color of the current pixel.

PC׳I7US2OO7/010376

The Delta commands can use signed 4-bit values supporting a range of [-8,7] or another range depending on the Delta Mode.

[0060] DELTA-NC specifies that this pixel is only slightly different from the same pixel in the previous frame, that is, it is different only by tlie specified Delta.

[0061] DELTA_CL specifies that this pixel is only slightly different from the previous pixel, again by the specified Delta;

[0062] DELTA_CA specifies that this pixel is only slightly different from the pixel immediately above, again by the specified Delta.

[0063] The Delta commands shown in Fig, 6 are the standard Delta Commands. As described below, there are many Delta modes. The Delta Commands are enabled and configured according to the DeltaMode. The Delta modes that are defined are shown in Fig. 7.

[0064] The Delta commands shown in Fig. 6 arc Mode 0, the standard format as shown above with a single pixel represented per command. Modes 4,5,8, and 9 are embodiments of delta commands employing 3-bit delta values. Modes 4,5, 8 and 9 are distinguishable by the range of delta values (i.e. [-4,3] or [0,7]) and the type of packaging scheme employed (i.e. Type 1 or Type 2). Type 1 and Type 2 are shown in Fig. 8 and are additional formats that support packing multiple pixel deltas into a single command, thus reducing overhead. Modes 4 and 5 are Type 1 and Modes 8 and 9 are Type 2.

[0065] In Type 1, the Ext (extension) field is used to identify whether another pixel follows. When a cleared Ext field is encountered (Ext = 0), then the command has ended, and the unnecessary bits in the current byte (if any) are ignored. As shown in Fig. 8, using Type 1 allows two pixels to be represented by three bytes. It would require four bytes to represent two pixels if the delta command had to be sent for each pixel.

Likewise, five pixels can be represented by seven bytes instead of ten.

[0066] Type 2, takes advantage of the unused Command = 111 and Ext - 1 shown in Fig. 5. In Type 2, when Command =111 and Ext = 1, one of Delta NC, Delta CL, or Delta CA applies. A two-bit command (Dcmd) embedded within the stream for each pixel determines which one of the Delta NC, Delta CL, or Delta CA applies and replaces the single-bit extension field of the previous format. It is particularly useful for long runs

PCT/US2007/0I0376 of deltas that are a mix of multiple directions. Dcmd specifies otic of: Delta NC (00), Delta CL (01), Delta CA (10), or termination of command (11).

[0067] Type 2 Packed 3־bit Delta becomes a better compression option oyer a string of

4-bit non-uniform Deltas as soon as two consecutive Deltas are encountered. With Delta data being written to the DeltaBuffer, the only additional information necessary is whether any of the current consecutive Deltas has exceeded 3-bit format. The decision to switch to packed deltas is made when the second consecutive Delta is encountered. At this time, the type of all subsequent Deltas will need to be written, along with the 3-bit data fields, to the 12-bit DeltaBuffer. The Output Process 103 can then handle using the current DeltaType and ConsecutiveDeltas count to parse and output the DeltaBuffer accordingly.

[0068] Fig. 7 also shows Mode 1 which is similar to Mode 0 except that the range is [0,15] instead of [-8,7]. Mode 1 is useful for adding a delta value to previously truncated colors.

[0069] The Command Process 102 determines which of the DVC or Delta commands are applicable through the three directional comparison blocks 114,116 and 118, which send information to the Command Process 102. The comparison blocks 114,116, and 118 operate in parallel, each comparing the current pixel (cPixel) with their respective reference pixel and sending a TRUE signal to the Command Process 102 if the pixels are determined to be equivalent. The comparison blocks also compute deltas of each channel and send that information packed into a 12-bit value, for example, for use by the Command Process 102. A separate Delta flag is passed in addition to the 12-bit value in order to indicate whether (lie observed deltas were within the current range or not. This keeps the Command Process 102 from having to decode a delta value of all zeros.

[0070] Fig. 9 shows an exemplary Delta comparison block. Each Delta comparison block receives two 24-bit strings (DataA[24] and DataB[24]), one representing a present pixel and another representing a previous pixel. The bit strings are compared using a 24bit comparator 902. The 24-bit comparator 902 outputs a true value if the bit strings are equal. The exemplary Delta comparison block also includes tliree 8-bit subtraction blocks 904,906, and 908 for subtracting 8-bit color values of a 24-bit number representing the current pixel from lire respective 8-bit color values from a 24-bit number representing the previous pixel. The four most significant bits from the subtraction blocks 904,906, and 908 arc sent to 12-Bit NOR gate 910 which determines whether a delta condition exists, i.e. the four most significant bits of each color channel are equal, i.e. the result of subtraction is 0000. ^,The subtraction blocks 904, 906, and 908 also transmit the delta value (i.e. the difference between the four least significant bits) for each color channel to the Command Process 102. It should be noted that although the exemplary Delta comparison block are configured for a 24-bit color implementation and a delta mode where a four-bit delta value is generated for each color channel, this is simply for explanatory purposes and not intended to be limiting. One ordinary skill in the art would appreciate other hardware configurations could be used for the current implementation and that other hardware configurations may be necessary for other implementations. Block 912 is optional logic that is used for make series determination. [0071] The Video Compression System 100 also has some number of series comparison blocks 120 and 122 (with no delta computation) operating in parallel to determine whether a make series condition occurs. Comparison blocks 120 and 122 compare the current pixel with the most recent unique pixels and send a TRUE signal to the Command Process 102 if the pixels are equivalent. It should be noted that although the present embodiment is using only two series comparators for the two most recent pixels, more comparators could be used.

[0072] The Command Process 102 tracks which commands (signals) arc permissible (based on position within a frame, etc), which ones still have active runs, and how many pixels are in the current run. When the Command Process 102 determines that a run ends, the end of the line is reached, or the end of a frame is reached, then the appropriate command 124 and data 126 are generated and sent to Output Process 103. The Command Process 102 also updates PreviousPixel 128 and the current location in the PreviousLine buffer 130.

[0073] Position is tracked by XPos variable 132 and YPos variable 134 to count horizontal and vertical positions within a frame in units of pixels. These positions are used to address the current location within the PreviousLine buffer 130 and to determine line and frame boundaries.

[0074j The Output Process 103 generates the output 136 to the rest of the system. If the Output Process 103 can not write, or is busy writing multiple pieces of data, and can not accept input, then everything else in the unit must stall waiting. The reason for this separate process is because of the buffered nature of the outgoing commands. The commands are buffered so that the header can be filled in once the size (count) is known. Without this separate process, the main Command Process 102 would take as much as twice as long to process pixels, With this dual process implementation, both processes operate in parallel in a pipelined fashion.

[0075] The Output Process 103 tracks the compression performance during each frame in terms of the number of bytes required to compress a number of pixels. This ratio can be tracked after each command, and when it exceeds a threshold (defined via an applet by the user), the Output Process 103 will inform the Command Process 102 that it needs to reduce the color depth in an effort to conserve bandwidth. The Output Process 103 will define and track BytesPcrFrame and PixelsPerFrame, both of which are reset at the end of each frame. To avoid erroneous decisions, the values are not actually used until the PixelsPerFrame indicates that the process is at least one line into the frame.

[0076] The most likely threshold is a limit on the number of Bytes/second. This can be translated to a ratio of bytes/pixel (BP) as follows:

X bytes/second >=־ (width *height )* BP * fps

BP = X/(width*height*fps)

[0077] For a maximum bandwidth allocation of, say, 8MBps, using 1024x768 at 30fps, this would equate to a Byte/Pixel ratio of:

[0078] BP ratio = 8M7(1024*768*30) = 0.34

[0079] The straightforward implementation would multiply the current pixel count by this value and then sec if the result is still greater than the current byte count. If it is not, then the bandwidth is being exceeded and the Command Process 102 needs to throttle back if possible.

]0080] Including a floating-point multiplier, however, is likely unnecessary. Reversing the ratio and defining it as pixels per byte allows for integer multiplication with a little less accuracy on the desired MBps rate.

[0081] PB ratio = 1024*768*30/8M = 2.94 3 ־־־ (is actually 7.86MBps)

PC17US2OO7/010376

[0082] Rounding this ratio up to the nearest integer is the conservative approach, but it could potentially give up a significant amount of bandwidth.

[0083] The Video Compression System 100 can also use “Color reduction” to reduce the bandwidth usage. This is separate from the operating system based color depth setting that the user will have configured for the computer. Regardless of the operating system setting, the Video Compression System 100 is storing 24-bit color data. The Video Compression System 100 will likewise read 24-bit color data from memory, but color reduction will remove some number of least significant bits from each color channel (e.g setting them to zero), thus increasing the likelihood that neighboring pixels will appear “identical” to the comparators 114,116, and 118, and also reducing the size of data required to send an uncompressed pixel.

[0084] The Video Compression System 100 could be designed to operate on 8 or 16-bit color throughout, but since it needs to handle 24-bit color at high resolution and frame rates, there is nothing to be gained by trying to handle the lower color depths natively. This would simply complicate the implementation for something that is expected to be rare. If it were the only way to support the higher resolutions, it may be desirable.

[0085] It also may be desirable for an embodiment which has no intention of supporting 24-bit color, but such implementation would need to have global changes made throughout the Video Compression System 100 to best optimize the performance and minimize gate count and cost

[0086] Besides the operating system and native color depths, there are a couple of other terms related to color depth. “Comparison depth” is the term used to describe the number of most significant color bits used (per color channel) when comparing two colors. There is also the current color depth (not the native depth) of the subsystem, which is what is referred to when the Video Compression System 100 references “Color Depth.” The comparison depth does not necessarily have to be the same as the ColorDepth. For instance, comparisons can be done with 15-bit color (5 bit per channel), and yet 24-bit pixels can be sent via the Delta and NEW_P1XEL commands, Tliis will result in somewhat more than 2¹⁵ colors potentially displayed on the screen, but will reduce tire color fidelity by slightly changing some neighboring colors. The advantage is that compression is potentially improved via the relaxed equality requirement.

l’CT/US2007/010376

[0087] The valid ColorDepth values are limited by the byte nature of the video protocol to the following: 24-bit, 15-bit, and 7-bit. The comparison depths that can be used to improve comparisons are: 3,6,9,12,15,18,21, and 24. The depth could be varied per channel.

[0088] The comparison depth to be used by the comparison logic is held in the ComparisonDepth register. This specifies the depth of a single channel and is used to create a mask that is applied to each pixel being fed into the comparator source registers.

[0089] The ColorDepth register controls how the Output Process 103 constructs NEW_P1XEL commands, and how logic in the rest of the system operates. If the ColorDepth is set to 24-bit pr 15־bit modes, then the NEW_PIXEL command is constructed with an 8-bit header (supporting up to 16 pixels) and 24-bits (or 15־bits) of color data per pixel.

[0090] For a ColorDepth of 7, DVC algorithms have typically used a special mapping of the color ranges. That is not done in this current implementation that favors color masking. Instead, 7 bits are used with 2-bits for each of red and blue, and 3 bits for green. Green is commonly favored in digital imaging color formats due to the fact that the human eye more readily detects green and the subtle shading variations than it does either red or blue. This results in a 7-bit protocol with a NEW_P1XEL that is constructed with a single-bit header and 7-bits of color data per pixel.

[0091] Both ColorDepth and ComparisonDepth registers are part of the state that is configurable by the user. The defaults can be controlled via jumpers or hard-coded into the hardware. The user should be able to adjust these either directly or indirectly via an applet that lets them adjust for bandwidth, etc. This applet communicates with the hardware through the Control Unit 102. When these values are being set, it must be assured that the video subsystem is either idle, or the values must be staged so that the changes are not made until the video subsystem is at a point where it can deal with the change.

[0092] The Video Compression System 100 needs to know the pixel depth (number of bits) and frame size (height and width in terms of pixels) to account for the line width and any padding that may be necessary at the end of a line or frame. It also must know the

WO 21)07/127452

PCI7US20l)7/010376 start of each frame since COPY_ABOVE may be turned off for the first line of a frame, and COPY_LEFT may be turned off for the first pixel of each line.

[0093] Note that this particular implementation does not suppress COPY_ABOVE or COPY_LEFT, but instead compares to the last row of the previous frame or the last pixel of the previous line respectively. In the event of a timing change which results in starting with a clean slate after an EndFrametClear message, left and above are compared against black pixel values of 0x000000.

[0094] Fig. 10 shows control commands that are generated at the Command Process 102. Control commands need to be distinguished from the encoding commands. Control commands are recognized by the fact that the least significant 5 bits arc all 0, indicating a zero count, and the most significant bit is clear indicating that it is a command that requires a count (versus a Della or New Pixel command with a color).

[0095] The group of control commands shown in Fig. 10 is grouped under the 1NFO_HEADEI< with sub-header byte that specifies Type and a Value for the Type. Types are defined in the table shown in Fig. 10. Value is specific to each Type, but in general is the number of bytes following this two-byte header (not including the two-byte header) for commands longer than this header. For commands consisting only of the two bytes of this header, value is a 4-bit data value.

[0096] Fig. 11 shows the INFO_HEADER for each Type.

[0097] The ColorDcpth command is used to specify the color mode. Currently, 24-Ηζ 15־bit, and 7-bit color modes are supported in this implementation. ColorDepth modes are defined in Fig. 11. Color Depth is provided to the Video Compression System 100 via a configuration register, and does not come in through the message stream. It should be noted that although the present embodiment only defines three color modes, more color modes could be defined in the future without departing from the scope of the present invention.

[0098[ The SERIES_DEPTH command shown in Fig. 11 is used to specify the width of tlie bit-field used to identify the color of each pixel in a Series command. The default depth is 1, which allows for series composed of 2 colors. The maximum depth supported varies by implementation, but in the preferred embodiment, a depth of 4 bits (16 colors) is the maximum desirable.

[0099] DELTA_MODE specifies the delta mode to be used for all subsequent commands. There are no data bytes following.

[00100] The CLEAR_FRAME command is used to indicate that a previous frame is to be cleared to zeros, or at least that the next frame should be compared to all zeros rather than the contents, of the previous frame.

[00101] The COMP ARI SON_DEPTH command is used to control the number of bits that are used when comparing pixels to each other. This is specified in terms of bits/channel. Comparison Depth is provided to the Video Compression System 100 via a configuration register,.and does not come in through the message stream.

[00102] The FRAME_STATUS command is used to convey the status of a frame, signifying cither successful completion or the need to abort further processing of the current frame. A Status field of 0 indicates success, while 1 indicates the need to abort. [00103] Fig. 12 shows the FRAME_SIZE command. FrameWidth and FrameHeight specify the dimensions of die subsequent frames in terms of pixels, ranging from 0 to 65,535. Implementations will generally have a smaller maximum' supported value on the order of 4,000 or less depending upon the application. BitsPerPixel specifies the number of bits used to define the color per pixel. ׳This typically is in the range of 24 bits or less, although larger depdis are possible in the future. This is fixed to 24״bits for this implementation.

[00104[ The only intent for this Frame Size message is to more neatly convey the Frame Size information to the Video Decompression Unit (VDU) (not shown) on the other side of the link. This unit typically gathers the separate pieces of data from the timing messages, which are destined for the DVI Output Unit on the far side of the link. Packaging all die information in this message (which is bound for the VDU) allows the VDU to ignore die timing messages. The fields of the FRAME_SIZE command are defined as follows:

[00105] Pixel Clock: This command is used to transmit the values necessary to regenerate a pixel clock for the display on the other side of the link.

[00106] Horizontal timing: This command is used to transmit the horizontal display timing values to the odier side of the link.

WO 2007/127-452

[00107] Vertical timing: This command is used to transmit the vertical display timing values to the other side of the link.

[00108! Fig. 13 shows a general data packet. This general purpose data packet is used to send data across the link. It is assumed that the other side □f the link can parse the data.

[00109] It should be noted that maximum compression is achieved when an entire frame can be encoded as one NoChange message. However, in the double-buffered architecture of the present embodiment such encoding would result in a full frame-time of latency (on the order of 16ms) between the times that the Video Compression System 100 starts processing a frame and when the VDU starts to receive the results of that processing.

Instead of letting a run-length command build throughput the entire frame, the commands are limited in size and sent periodically throughout the frame. For simplicity, this is done at the end of each line, if absolute performance becomes more important than latency, this could be fine tuned in numerous ways.

[00110] It should be noted that due to the architecture of the Digital Input Unit (DIU) (not shown), the Video Compression System 100 may already be processing a frame when it receives information indicating that the resolution is changing. This information will be in the form of an EndFrame message with the Clear bit set.

[00111] When the Video Compression System 100 finishes a frame, it should see an EndFrame message with a 0 Clear bit in its incoming FIFO. If it instead sees an EndFrame with Clear bit or any Timing message at the front of its FIFO, then it knows that the previous frame (the one it just finished processing) was invalid. The Video Compression System 100 then sends a FrameAbort message to the VDU to indicate that the preceding Same should not be displayed. If there is no message in the incoming FIFO, the Video Compression System 100 waits until one is present.

[00112] If the resolution has decreased, then the Video Compression System 100 will have attempted a much larger read to memory than what memory will actually be able to supply. The Memory Interface Unit (M1U) 107 assures that the Video Compression System 100 read pointer trails behind the DIU write pointer so that the Video Compression System 100 does not overtake the data being written. The early termination of DVI104 data going to memory 106 will mean that the DIU write pointer will stall for the timeout period, causing the Video Compression System 100 to stall as well waiting on data. When the Memory Interface Unit 107 decides to terminate the request, it will simply abort the processing of the Video Compression System’s 100 read request. [00113] The Video Compression System 100 must continually monitor its incoming command FIFO while processing pixels coming from memory 106. If the Video Compression System 100 sees an EndFrame, Clear or Vertical Timing message from the DILI, it immediately terminates processing of the current frame and sends the FrameAbort message to the VDU. The Video Compression System 100 also immediately sends a Purge Frame message to the MIU107. This message tells the MIU 107 to stop sending further data to the Video Compression System 100 for the frame currently in progress. The MIU 107 subsequently will flush any data remaining to be read from the DIU. The Video Compression System 100 is responsible for flushing the Current and Previous FIFOs after notifying the MIU 107 to stop adding to them.

[00114J The Video Compression System 100 needs to process a pixel per clock with per-frame latency on the order of only a few pixels and no latency between pixels. This depends heavily on the memory controller 106 being able to provide input data at the necessary rate, and also on the output not backing up and forcing a stall.

[00115] Assuming that the above criteria is met, and that a pixel can be processed per clock, then the clock rate is based on the number of pixels per frame and the desired frame rate. The table in Fig. 15 shows the rates required for various combinations. [00116] Allowing for some overhead, the table shows that an 80MHz core clock would be sufficient for 30 fps HDTV. A more likely target would be something on the order of 100-125 MHz, which would handle even QXGA at 30 fps, and would support HD'IV at over 40 ips. While very high resolutions at very high frame rates mayrequire too high'a clock rate to be commercially/economically feasible, advances in processing rates and costs will eventually enable even those very high resolutions and frame rates to be implemented with economically feasible off-the-shelf products. Likewise, clock rates that are border-line for commercial/economic reasonableness today will be economically feasible with off-the-shelf products even sooner.

[00117] Being able to compress a-single frame at a 60fps rate improves the latency perceived by the end user, and is therefore a worthwhile goal even if frames are intentionally dropped or are otherwise not transmitted 60 times per second.

PCT/US2(I()7A) 11)376

[00118] While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.