APPARATUS AND METHOD FOR BUFFERING CONTEXT ARRAYS REFERENCED FOR PERFORMING ENTROPY DECODING UPON MULTI-TILE ENCODED PICTURE AND RELATED ENTROPY DECODER

Abstract

A buffering apparatus for buffering context arrays of a multi-tile encoded picture having a plurality of tiles includes a first buffer and a second buffer. The first buffer is arranged to buffer a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture. The second buffer is arranged to buffer a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture. When the first tile is currently decoded according to the first context array buffered in the first buffer, the second context array is buffered in the second buffer.

Description

CROSS REFERENCE TO RELATED APPLICATION

This is a continuation of U.S. application Ser. No. 14/343,388 (filed on Mar. 7, 2014), which is National Stage Entry of PCT application No. PCT/CN2012/081288 (filed on Sep. 12, 2012). In addition, the PCT application No. PCT/CN2012/081288 claims the benefit of U.S. provisional application No. 61/553,350 (filed on Oct. 31, 2011) and U.S. provisional application No. 61/566,984 (filed on Dec. 5, 2011). The entire contents of all related applications are incorporated herein by reference.

BACKGROUND

The disclosed embodiments of the present invention relate to decoding a multi-tile video/image bitstream which transmits a plurality of multi-tile encoded pictures/compressed frames each having a plurality of tiles, and more particularly, to an apparatus and a method for buffering context arrays referenced for performing entropy decoding upon a multi-tile encoded picture and a related entropy decoder.

As proposed in High-Efficiency Video Coding (HEVC) specification, one picture can be partitioned into multiple tiles. FIG. 1 is a diagram illustrating tiles adopted in the HEVC specification. FIG. 2 is a diagram illustrating a conventional decoding order of the tiles shown in FIG. 1. As shown in FIG. 1, one picture 10 is partitioned into a plurality of tiles T₁₁′-T₁₃′, T₂₁′-T₂₃′, T₃₁′-T₃₃′ separated by row boundaries (i.e., horizontal boundaries) HB₁′, HB₂′ and column boundaries (i.e., vertical boundaries) VB₁′, VB₂′.

Inside each tile, largest coding units (LCUs)/treeblocks (TBs) are raster scanned, as shown in FIG. 2. For example, LCUs/TBs orderly indexed by the Arabic numbers in the same tile T₁₁ are decoded sequentially. Inside each multi-tile picture, tiles are raster scanned, as shown in FIG. 2. For example, the tiles T₁₁′-T₁₃′, T₂₁′-T₂₃40 and T₃₁′-T₃₃′ are decoded sequentially. Specifically, one picture can be uniformly partitioned by tiles or partitioned into specified LCU-column-row tiles. A tile is a partition which has vertical and horizontal boundaries, and it is always rectangular with an integer number of LCUs/TBs included therein. Hence, tile boundaries must be LCU/TB boundaries.

There are two types of tiles, independent tiles and dependent tiles. As to the independent tiles, they are treated as sub-pictures/sub-streams. Hence, encoding/decoding LCUs/TBs of an independent tile (e.g., motion vector prediction, intra prediction, entropy coding, etc.) does not need data from other tiles. Besides, assume that data of the LCUs/TBs is encoded/decoded using arithmetic coding such as a context-based adaptive binary arithmetic coding (CABAC) algorithm. Regarding each independent tile, the CABAC statistics are initialized/re-initialized at the start of the tile, and the LCUs outside the tile boundaries of the tile are regarded as unavailable. For example, the CABAC statistics at the first LCU/TB indexed by “1” in the tile T₁₁′ would be initialized when decoding of the tile T₁₁′ is started, the CABAC statistics at the first LCU/TB indexed by “13” in the tile T₁₂′ would be re-initialized when decoding of the tile T₁₂′ is started, the CABAC statistics at the first LCU/TB indexed by “31” in the tile T₁₃′ would be re-initialized when decoding of the tile T₁₃′ is started, and the CABAC statistics at the first LCU/TB indexed by “40” in the tile T₂₁′ would be re-initialized when decoding of the tile T₂₁′ is started.

However, encoding/decoding LCUs/TBs of a dependent tile (e.g., motion vector prediction, intra prediction, entropy coding, etc.) has to consider data provided by other tiles. Hence, vertical and horizontal buffers are required for successfully decoding a multi-tile encoded picture/compressed frame having dependent tiles included therein. Specifically, the vertical buffer is used for buffering decoded information of LCUs/TBs of an adjacent tile beside a vertical boundary (e.g., a left vertical boundary) of a currently decoded tile, and the horizontal buffer is used for buffering decoded information of LCUs/TBs of another adjacent tile beside a horizontal boundary (e.g., a top horizontal boundary) of the currently decoded tile. As a result, the buffer size for decoding the multi-tile encoded picture/compressed frame would be large, leading to higher production cost. Besides, assume that data of the LCUs/TBs is encoded/decoded using arithmetic coding such as a CABAC algorithm. Regarding a dependent tile, the CABAC statistics may be initialized at the start of the tile or inherited from another tile. For example, the CABAC statistics at the first LCU/TB indexed by “1” in the tile T₁₁′ would be initialized when decoding of the tile T₁₁′ is started, the CABAC statistics at the first LCU/TB indexed by “13” in the tile T₁₂′ would be inherited from the CABAC statistics at the last LCU/TB indexed by “12” in the tile T₁₁′ when decoding of the tile T₁₂′ is started, the CABAC statistics at the first LCU/TB indexed by “31” in the tile T₁₃′ would be inherited from the CABAC statistics at the last LCU/TB indexed by “30” in the tile T₁₂′ when decoding of the tile T₁₃′ is started, and the CABAC statistics at the first LCU/TB indexed by “40” in the tile T₂₁′ would be inherited from the CABAC statistics at the last LCU/TB indexed by “39” in the tile T₁₃′ when decoding of the tile T₂₁′ is started.

As the conventional decoder design employs a tile scan order for decoding a multi-tile encoded picture, the vertical buffer (column buffer) is necessitated by the tile scan order for buffering decoded information of LCUs/TBs of an adjacent tile beside a vertical boundary (e.g., a left vertical boundary) of a currently decoded dependent tile, which increases the production cost inevitably. Thus, there is a need for an innovative entropy decoder design which is capable of reducing or omitting the vertical buffer (column buffer) when decoding the multi-tile encoded picture/compressed frame.

SUMMARY

In accordance with exemplary embodiments of the present invention, an apparatus and a method for buffering context arrays referenced for performing entropy decoding upon a multi-tile encoded picture and a related entropy decoder, to solve the above-mentioned problems.

According to a first aspect of the present invention, an exemplary buffering apparatus for buffering context arrays of a multi-tile encoded picture having a plurality of tiles is disclosed. The exemplary buffering apparatus includes a first buffer and a second buffer. The first buffer is arranged to buffer a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture. The second buffer is arranged to buffer a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture. When the first tile is currently decoded according to the first context array buffered in the first buffer, the second context array is buffered in the second buffer.

According to a second aspect of the present invention, an exemplary buffering method for buffering context arrays of a multi-tile encoded picture having a plurality of tiles is disclosed. The exemplary buffering method includes: buffering a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture; and buffering a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture when the first tile is currently decoded according to the buffered first context array.

According to a third aspect of the present invention, an exemplary entropy decoder is disclosed. The exemplary entropy decoder includes an entropy decoding core and a buffering apparatus. The entropy decoding core is arranged to perform entropy decoding upon a multi-tile encoded picture, having a plurality of tiles included therein, in a raster scan order, wherein the entropy decoding core starts decoding a portion of a current tile after decoding a portion of a previous tile. The buffering apparatus is coupled to the entropy decoding core, and arranged for buffering context arrays of the multi-tile encoded picture. The buffering apparatus includes a first buffer and a second buffer. The first buffer is arranged to buffer a first context array referenced for entropy decoding a first tile of the multi-tile encoded picture. The second buffer is arranged to buffer a second context array referenced for entropy decoding a second tile of the multi-tile encoded picture. When the first tile is currently decoded according to the first context array buffered in the first buffer, the second context array is buffered in the second buffer.

These and other objectives of the present invention will no doubt become obvious to those of ordinary skill in the art after reading the following detailed description of the preferred embodiment that is illustrated in the various figures and drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a diagram illustrating tiles adopted in the HEVC specification.

FIG. 2 is a diagram illustrating a conventional decoding order of the tiles shown in FIG. 1.

FIG. 3 is a diagram illustrating an entropy decoder according to a first embodiment of the present invention.

FIG. 4 is a diagram illustrating an exemplary entropy decoding operation performed by the entropy decoder shown in FIG. 3.

FIG. 5 is a diagram illustrating a proposed decoding order of the tiles shown in FIG. 4 according to an embodiment of the present invention.

FIG. 6 is a diagram illustrating the entropy decoder which stores a currently used context array into a second buffer and loads a context array needed for decoding a next tile into a first buffer.

FIG. 7 is a diagram illustrating an exemplary buffer maintenance operation of the buffering apparatus shown in FIG. 3.

FIG. 8 is a diagram illustrating an entropy decoder according to a second embodiment of the present invention.

FIG. 9 is a diagram illustrating an entropy decoder according to a third embodiment of the present invention.

FIG. 10 is a diagram illustrating the entropy decoder which switches from one interconnection for accessing a context array used for decoding a current tile to another interconnection for accessing a context array needed for decoding a next tile.

FIG. 11 is a diagram illustrating an exemplary buffer maintenance operation of the buffering apparatus shown in FIG. 9.

DETAILED DESCRIPTION

Certain terms are used throughout the description and following claims to refer to particular components. As one skilled in the art will appreciate, manufacturers may refer to a component by different names. This document does not intend to distinguish between components that differ in name but not function. In the following description and in the claims, the terms “include” and “comprise” are used in an open-ended fashion, and thus should be interpreted to mean “include, but not limited to . . . ”. Also, the term “couple” is intended to mean either an indirect or direct electrical connection. Accordingly, if one device is electrically connected to another device, that connection may be through a direct electrical connection, or through an indirect electrical connection via other devices and connections.

FIG. 3 is a diagram illustrating an entropy decoder according to a first embodiment of the present invention. The exemplary entropy decoder 100 includes an entropy decoding core 102 and a buffering apparatus 104, where the buffering apparatus 104 is coupled to the entropy decoding core 102, and has a first buffer 106 and a second buffer 108. It should be noted that only the elements pertinent to the present invention are illustrated in FIG. 3. In practice, the entropy decoder 100 is allowed to have additional elements included therein to support more functions. The entropy decoding core 102 is arranged to perform entropy decoding upon a multi-tile encoded picture PIC_IN, having a plurality of tiles included therein, in a raster scan order. Therefore, the entropy decoding core 102 starts decoding a portion of a current tile after decoding a portion of a previous tile. It should be noted that the proposed decoding method/decoding order may be applied to independent tiles or dependent tiles. In the following, an example of decoding independent tiles is provided for illustrative purposes only, and is not meant to be a limitation of the present invention.

Please refer to FIG. 4, which is a diagram illustrating an exemplary entropy decoding operation performed by the entropy decoder 100 shown in FIG. 3. The multi-tile encoded picture PIC_IN is partitioned into a plurality of tiles (e.g., nine dependent tiles T₁₁-T₃₃ in this embodiment). Each of the tiles T₁₁-T₃₃ is composed of a plurality of LCUs/TBs. If a conventional decoding manner is employed, the LCU/TB index values shown in FIG. 4 indicate the conventional decoding order of the LCUs/TBs included in the multi-tile encoded picture PIC_IN. Specifically, regarding a conventional decoder design, the decoding order in a multi-tile encoded picture with tiles has a raster scan sequence for LCUs/TBs in each tile and a raster scan sequence for the tiles, as shown in FIG. 2. In contrast to the conventional decoder design, the proposed decoder design of the present invention has the entropy decoding core 102 configured to decode all LCUs/TBs of the whole multi-tile encoded picture PIC_IN in a raster scan manner, where the decoding order includes successive decoding sequences S1-S8 as shown in FIG. 4. For example, the LCUs/TBs, located at the first row shown in FIG. 4 and belonging to different tiles T₁₁, T₁₂ and T₁₃, are sequentially decoded from the left-most LCU/TB to the right-most LCU/TB as indicated by the decoding sequence S1; the LCUs/TBs, located at the second row shown in FIG. 4 and belonging to different tiles T₁₁, T₁₂ and T₁₃, are sequentially decoded from the left-most LCU/TB to the right-most LCU/TB as indicated by the decoding sequence S2 following the decoding sequence S1; and the LCUs/TBs, located at the third row shown in FIG. 4 and belonging to different tiles T₁₁, T₁₂ and T₁₃, are sequentially decoded from the left-most LCU/TB to the right-most LCU/TB as indicated by the decoding sequence S3 following the decoding sequence S2. For clarity, please refer to FIG. 5, which is a diagram illustrating a proposed decoding order of the tiles shown in FIG. 4 according to an embodiment of the present invention. By using the entry point of each tile bitstream/partition, a decoder can use raster scan of LCUs/TBs (as shown in FIG. 5) instead of tile scan of LCUs/TBs (as shown in FIG. 2). Hence, due to the proposed raster scan order, no vertical buffer (column buffer) is needed for buffering decoded information of LCUs/TBs of an adjacent tile. In this way, the production cost can be effectively reduced. It should be noted that the proposed raster scan order may be employed for decoding independent tiles or dependent tiles.

In this embodiment, data of the LCUs/TBs is encoded using a context-based adaptive binary arithmetic coding (CABAC) algorithm. Hence, the context model, which is a probability model, should be properly selected and updated during the entropy decoding of the multi-tile encoded picture PIC_IN. It should be noted that the entropy decoding core 102 does not necessarily re-initialize the CABAC statistics at the first LCU/TB of each tile. That is, the CABAC statistics at the first LCU/TB of a current tile may be inherited from the CABAC statistics at a specific LCU/TB of a previous tile horizontally adjacent to the current tile, where the first LCU/TB and the specific LCU/TB are horizontally adjacent to each other and located at opposite sides of a tile boundary (i.e., a vertical/column boundary) between the current tile and the previous tile. As can be seen from FIG. 4, the initial CABAC statistics at the first LCU/TB indexed by “13” in the tile T₁₂ is inherited from the CABAC statistics updated at the LCU/TB indexed by “4” in the tile T₁₁; similarly, the initial CABAC statistics at the first LCU/TB indexed by “31” in the tile T₁₃ is inherited from the CABAC statistics updated at the LCU/TB indexed by “18” in the tile T₁₂. The tiles T₁₁-T₁₃ are horizontally adjacent tiles, i.e., horizontal partitions. However, the tiles T₁₁, T₂₁, and T₃₁ are vertically adjacent tiles, i.e., vertical partitions. Regarding the tile T₂₁ which is vertically adjacent to the tile T₁₁, the initial CABAC statistics at the first LCU/TB indexed by “40” in the tile T₂₁ would be inherited from the CABAC statistics updated at the last LCU/TB indexed by “39” in the tile T₁₃. As the initial setting of the CABAC statistics for the rest of the tiles can be easily deduced by analogy, further description is omitted for brevity.

The entropy decoding core 102 employs the decoding order including successive decoding sequences S1-S8. Hence, the LCUs/TBs in the same tile are not decoded continuously due to the fact that the entropy decoding core 102 starts decoding a portion of a current tile after decoding a portion of a previous tile. As can be seen from FIG. 4, after the LCUs/TBs indexed by “1”, “2”, “3” and “4” of the tile T₁₁ are successively decoded, the next LCU/TB to be decoded by the entropy decoding core 102 would be the first LCU/TB indexed by “13” in the next tile T₁₂ rather than the LCU/TB indexed by “5” in the current tile T₁₁; after the LCUs/TBs indexed by “13”, “14”, “15”, “16”, “17” and “18” of the tile T₁₂ are successively decoded, the next LCU/TB to be decoded by the entropy decoding core 102 would be the first LCU/TB indexed by “31” in the next tile T₁₃ rather than the LCU/TB indexed by “19” in the current tile T₁₂; and after the LCUs/TBs indexed by “31”, “32” and “33” of the tile T₁₃ are successively decoded, the next LCU/Tb to be decoded by the entropy decoding core 102 would be the first LCU/TB indexed by “5” in the previously processed tile T₁₁ rather than the LCU/TB indexed by “34” in the current tile T₁₃. Therefore, information of the context/probability model currently obtained by a partially decoded tile should be buffered/maintained since the LCUs/TBs in the same tile are not decoded by the entropy decoding core 102 continuously.

The buffering apparatus 104 is implemented for buffering context arrays of the multi-tile encoded picture PIC_IN. The context arrays include context models each being a probability model for one or more bins of the binarized symbol in the arithmetic coding, such as CABAC in H.264 and HEVC. A context model may be chosen from a selection of available models, depending on the statistics of recently-coded data symbols. The context model stores the probability of each bin being “1” or “0”. Details of the context model can be found in a published paper: Marpe et al., “Context-Based Adaptive Binary Arithmetic Coding in the H.264/AVC Video Compression Standard”, IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY, VOL. 13, NO. 7, July 2003, which is incorporated herein by reference. Further description is therefore omitted here for brevity.

By way of example, but not limitation, the number of buffered context arrays maintained by the buffering apparatus 104 during entropy decoding of the multi-tile encoded picture PIC_IN depends on the partitioning setting of the multi-tile encoded picture PIC_IN. For example, when the multi-tile encoded picture PIC_IN has N horizontally adjacent partitions (i.e., N horizontal partitions/tiles at the same row), the number of buffered context arrays maintained by the buffering apparatus 104 during entropy decoding of the multi-tile encoded picture is equal to N. Regarding the example shown in FIG. 4, N is equal to 3. Hence, there are 3 context arrays concurrently maintained in the buffering apparatus 104, where each of the context arrays records information of the context/probability model needed for performing entropy decoding upon a corresponding tile. Technical features of the buffering apparatus 104 are detailed as below.

The first buffer 106 and the second buffer 512 may be allocated in the same storage device or implemented using separate storage devices, depending upon actual design consideration. For example, the first buffer 106 maybe implemented using a register of the entropy decoder 100, and the second buffer 104 may be implemented using an internal buffer (e.g., a static random access memory (SRAM)) of the entropy decoder 100. The first buffer 106 is arranged to buffer a context array (e.g., CA₁) referenced for performing entropy decoding upon a specific tile of the multi-tile encoded picture PIC_IN, and the second buffer 108 is arranged to buffer context arrays (e.g., CA₂-CA_N) referenced for performing entropy decoding upon other tiles of the multi-tile encoded picture PIC_IN, where N is equal to the number of horizontally adjacent partitions (i.e., horizontal partitions/tiles at the same row). When the specific tile is currently decoded according to the context array CA₁ buffered in the first buffer 106, the context arrays CA₂-CA_N are buffered/maintained in the second buffer 108. That is, the first buffer 106 stores a context array of a currently decoded tile, and the second buffer 108 stores context arrays of other tiles which are not currently decoded. When the entropy decoding of the specific tile encounters a tile boundary (e.g., a right vertical/column boundary), the currently used context array CA₁ is stored into the second buffer 108 to update the original context array CA₁ stored in the second buffer 108, and the context array CA₂ needed for decoding the next tile is loaded into the first buffer 106, as shown in FIG. 6. Thus, during the entropy decoding of the multi-tile encoded picture PIC_IN, context arrays for different tiles are concurrently maintained by the buffering apparatus 104, thereby facilitating the discontinuous entropy decoding of LCUs/TBs in each tile.

An exemplary buffer maintenance operation of the buffering apparatus 104 is described with reference to FIG. 7. Supposing that the multi-tile encoded picture PIC_IN has the partition setting shown in FIG. 4, the number of maintained context arrays is equal to 3 (i.e., N=3). In the beginning, the context array CA₁ at the LCU/TB indexed by “1” (i.e., the initial context array CA₁ for the tile T₁₁) is initialized and loaded into the first buffer 106 by the entropy decoding core 102. When the entropy decoding of the tile T₁₁ encounters a tile boundary (e.g., a vertical/column boundary VB₁), the context array CA₁ updated at the LCU/TB indexed by “4” is stored into the second buffer 108, and the context array CA₂ at the LCU/TB indexed by “13” (i.e., the initial context array CA₂ for the tile T₁₂) is inherited from the context array CA₁ updated at the LCU/TB indexed by “4” and loaded into the first buffer 106. When the entropy decoding of the tile T₁₂ encounters a tile boundary (e.g., a vertical/column boundary VB₂), the context array CA₂ updated at the LCU/TB indexed by “18” is stored into the second buffer 108, and the context array CA₃ at the LCU/TB indexed by “31” (i.e., the initial context array CA₃ for the tile T₁₃) is inherited from the context array CA₂ updated at the LCU/TB indexed by “18” and loaded into the first buffer 106. When the entropy decoding of the tile T₁₃ encounters a tile boundary (e.g., a vertical/column boundary VB₃), the context array CA₃ updated at the LCU/TB indexed by “33” is stored into the second buffer 108, and the context array CA₁ needed to resume decoding the LCU/TB indexed by “5” is read from the second buffer 108 and then loaded into the first buffer 106. As a person skilled in the art can readily understand loading and storing of the context arrays required for decoding the following LCUs/TBs by referring to FIG. 7, further description is omitted here for brevity.

As mentioned above, the entropy decoding core 102 decodes all LCUs/TBs of the whole multi-tile encoded picture PIC_IN in a raster scan manner, where the decoding order includes successive decoding sequences S1-S8 as shown in FIG. 4. Therefore, only the horizontal buffer 101 is needed to buffer decoded information of LCUs/TBs of an adjacent tile beside a horizontal/row boundary (e.g., a top horizontal/row boundary) of a currently decoded tile. Since there is no vertical buffer needed for buffering decoded information of LCUs/TBs of an adjacent tile beside a vertical/column boundary (e.g., a left vertical/column boundary) of the currently decoded tile, the overall buffer size required by a video decoder is reduced, thus decreasing the production cost accordingly.

As the first buffer 106 is used to maintain a context array of one currently decoded tile and the second buffer 108 is used to maintain context arrays of other tiles that are not currently decoded, the context arrays are loaded and stored between the first buffer 106 and the second buffer 108. If the first buffer 106 is a register or an internal buffer (e.g., SRAM) of the entropy decoder and the second buffer 108 is an external buffer such as a dynamic random access memory (DRAM), the decoding performance of the entropy decoder may be degraded due to read/write latency of the external buffer. The present invention therefore proposes a modified entropy decoder with enhanced buffer read/write efficiency.

Please refer to FIG. 8, which is a diagram illustrating an entropy decoder according to a second embodiment of the present invention. The entropy decoder 500 has a buffer access enhancement circuit 512 coupled between the first buffer 106 and the second buffer 108, where the buffer access enhancement circuit 512 includes a third buffer 514, a post-store mechanism 516, and a pre-fetch mechanism 518. By way of example, but not limitation, the first buffer 106, the buffer access enhancement circuit 512 and the entropy decoding core 102 may be disposed in the same chip, and the second buffer 108 may be an off-chip storage device. For example, the first buffer 106 may be a register, the third buffer 514 may be an internal buffer (e.g., SRAM), and the second buffer 108 may be an external buffer (e.g., DRAM). However, this is for illustrative purposes, and is not meant to be a limitation of the present invention.

The pre-fetch mechanism 518 is used for pre-fetching a context array (e.g., CA₂ shown in FIG. 6) from the second buffer 108 and temporarily storing the pre-fetched context array in the third buffer 514. Therefore, when the pre-fetched context array is needed by the entropy decoding core 102, the pre-fetched context array in the third buffer 514 may be quickly loaded into the first buffer 106. The post-store mechanism 516 is used for post-storing a context array (e.g., CA₁ shown in FIG. 6), which is read from the first buffer 106 and temporarily stored in the third buffer 514, into the second buffer 108. Therefore, the context array read from the first buffer 106 is not required to be immediately transferred to the second buffer 108. In this way, the read latency and write latency of the second buffer 108 can be concealed to effectively reduce the time consumed on loading and storing context arrays.

In the embodiment shown in FIG. 8, the buffer access enhancement circuit 512 is configured to support both of the post-store function and the pre-fetch function. However, the buffer access enhancement circuit 512 may be modified to support only one of the post-store function and the pre-fetch function. Specifically, in a case where the buffer access enhancement circuit 512 is modified to omit the post-store mechanism 516, enhanced buffer reading efficiency of the second buffer 108 is achieved via the pre-fetch mechanism 518. In another case where the buffer access enhancement circuit 512 is modified to omit the pre-fetch mechanism 518, enhanced buffer writing efficiency of the second buffer 108 is achieved via the post-store mechanism 516. These alternative designs all fall within the scope of the present invention.

The entropy decoder 500 in FIG. 8 employs the buffer access enhancement circuit 512 to mitigate the decoding performance degradation caused by the buffer read/write latency. However, using a different decoder configure to achieve the same objective is feasible. Please refer to FIG. 9, which is a diagram illustrating an entropy decoder according to a third embodiment of the present invention. The exemplary entropy decoder 600 includes the aforementioned entropy decoding core 102 and a buffering apparatus 604, where the buffering apparatus 604 is coupled to the entropy decoding core 102, and has a multiplexer 606 and a plurality of buffers 608_1, 608_2 . . . 608_N. It should be noted that only the elements pertinent to the present invention are illustrated in FIG. 9. In practice, the entropy decoder 600 is allowed to have additional elements included therein to support more functions. As mentioned above, the entropy decoding core 102 is arranged to perform entropy decoding upon the multi-tile encoded picture PIC_IN, having a plurality of tiles included therein, in a raster scan order as shown in FIG. 4. Since the entropy decoding core 102 employs the decoding order including successive decoding sequences S1-S8, the LCUs/TBs in the same tile are not decoded continuously. In other words, the entropy decoding core 102 starts decoding a portion of a current tile after decoding a portion of a previous tile. Hence, information of the context/probability model currently obtained by a partially decoded tile should be buffered/maintained due to the fact that the LCUs/TBs in the same tile are not decoded continuously. Thus, the buffering apparatus 604 is implemented for buffering context arrays of the multi-tile encoded picture PIC_IN. By way of example, but not limitation, the number of buffered context arrays maintained by the buffering apparatus 604 during entropy decoding of the multi-tile encoded picture PIC_IN depends on the partitioning setting of the multi-tile encoded picture PIC_IN. For example, when the multi-tile encoded picture PIC_IN has N horizontally adjacent partitions (i.e., N horizontal partitions/tiles at the same row), the number of buffered context arrays maintained by the buffering apparatus 604 during entropy decoding of the multi-tile encoded picture is equal to N. Regarding the example shown in FIG. 4, N is set by 3. Therefore, there are 3 context arrays concurrently maintained in the buffering apparatus 604, where each of the context arrays records information of the context/probability model needed for performing entropy decoding upon a corresponding tile.

The buffers 608_1-608_N may be allocated in the same storage device or implemented using separate storage devices, depending upon actual design consideration. For example, the buffers 608_1-608_N may be implemented using registers, internal buffers (e.g., SRAMs), external buffers (e.g., DRAMs), or a combination thereof. The major difference between the buffering apparatuses 104 and 604 is that each of the buffers 608_1-608_N is dedicated to maintaining one context array. As can be seen from FIG. 9, context arrays CA₁, CA₂, . . . , CA_N referenced for performing entropy decoding upon different tiles are stored in the buffers 608_1-608_N, respectively. For example, the first buffer 608_1 is arranged to buffer a context array CA₁ referenced for performing entropy decoding upon the first tile of the multi-tile encoded picture PIC_IN, the second buffer 608_2 is arranged to buffer a context array CA₂ referenced for performing entropy decoding upon the second tile of the multi-tile encoded picture PIC_IN, the third buffer 608_3 is arranged to buffer a context array CA₃ referenced for performing entropy decoding upon the third tile of the multi-tile encoded picture PIC_IN, and the N^th buffer 608 N is arranged to buffer a context array CA_N referenced for performing entropy decoding upon the n^th tile of the multi-tile encoded picture PIC_IN, where the1^st-n^th tiles are horizontally adjacent tiles at the same row. In other words, N is equal to the number of horizontally adjacent partitions (i.e., horizontal partitions/tiles at the same row). The multiplexer (MUX) 606 is used to control which one of the buffers 608_1-608_N is allowed to be accessed by the entropy decoding core 102. The MUX 606 has a plurality of first connection ports P₁, P₂, P₃, . . . , P_N and a second connection port N, where the buffers 608_1-608_N are coupled to the first connection ports P₁-P_N, respectively, and the entropy decoding core 102 is coupled to the second connection port N. When the first tile is a current tile selected to be decoded by the entropy decoding core 102, the MUX 606 enables an interconnection 607_1 between the first connection port P₁ and the second connection port N, thus allowing the entropy decoding core 102 to access the context array CA₁ stored in the first buffer 608_1 and perform entropy decoding upon the first tile according to the context array CA₁. It should be noted that when the first tile is currently decoded according to the context array CA₁ buffered in the buffer 608_1, the context arrays CA₂-CA_N are buffered/maintained in other buffers 608_1-608_N, respectively.

When the entropy decoding of the first tile encounters a tile boundary (e.g., a right vertical/column boundary), the MUX 606 switches the interconnection 607_1 between the first connection port P₁ and the second connection port N to another interconnection 607_2 between the first connection port P₂ and the second connection port N, as shown in FIG. 10. Hence, the entropy decoding core 102 is allowed to access the context array CA₂ stored in the second buffer 608_2 and perform entropy decoding upon the second tile according to the context array CA₂.

To put it simply, during the entropy decoding of the multi-tile encoded picture PIC_IN, multiple context arrays for different tiles may be concurrently maintained by the buffering apparatus 604, thereby facilitating the discontinuous decoding of the LCUs/TBs in each tile. Besides, as the context arrays are buffered in respective designated buffers and selected under the control of a multiplexer, there is no loading and storing of context arrays between two buffers. In this way, the decoding performance of the entropy decoder may be improved by using a switchable hard-wired interconnection between the entropy decoding core and the buffers.

An exemplary buffer maintenance operation of the buffering apparatus 604 is described with reference to FIG. 11. Supposing that the multi-tile encoded picture PIC_IN has the partition setting shown in FIG. 4, the number of maintained context arrays is equal to 3 (i.e., N=3). Therefore, the buffering apparatus 604 would be configured to have three buffers 608_1-608_3 only. In the beginning, the first buffer 608_1 is selected by the MUX 606, and the context array CA₁ at the LCU/TB indexed by “1” (i.e., the initial context array CA₁ for the tile T₁₁) is initialized by the entropy decoding core 102. When the entropy decoding of the tile T₁₁ encounters a tile boundary (e.g., a vertical/column boundary VB₁), the first buffer 608_1 maintains the context array CA₁ updated at the LCU/TB indexed by “4”, the MUX 606 selects the second buffer 608_2, and the context array CA₂ at the LCU/TB indexed by “13” (i.e., the initial context array CA₂ for the tile T₁₂) is inherited from the context array CA₁ updated at the LCU/TB indexed by “4”. When the entropy decoding of the tile T₁₂ encounters a tile boundary (e.g., a vertical/column boundary VB₂), the second buffer 608_2 maintains the context array CA₂ updated at the LCU/TB indexed by “18”, the MUX 606 selects the third buffer 608_3, and the context array CA₃ at the LCU/TB indexed by “31” (i.e., the initial context array CA₃ for the tile T₁₃) is inherited from the context array CA₂ updated at the LCU/TB indexed by “18”. When the entropy decoding of the tile T₁₃ encounters a tile boundary (e.g., a vertical/column boundary VB₃), the context array CA₃ updated at the LCU/TB indexed by “33” is maintained in the third buffer 608_3, and the MUX 606 selects the first buffer 608_1 again such that the decoding operation of the tile T₁₁ is resumed. As a person skilled in the art can readily understand loading and storing of the context arrays required for decoding the following LCUs/TBs by referring to FIG. 11, further description is omitted here for brevity.

In conclusion, a buffering method employed by any of the aforementioned entropy decoders for buffering context arrays of a multi-tile encoded picture having a plurality of tiles may include at least the following steps: buffering a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture; and buffering a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture when the first tile is currently decoded according to the buffered first context array.

Those skilled in the art will readily observe that numerous modifications and alterations of the device and method may be made while retaining the teachings of the invention. Accordingly, the above disclosure should be construed as limited only by the metes and bounds of the appended claims.

Claims

1. A buffering apparatus for buffering context arrays of a multi-tile encoded picture having a plurality of tiles, the buffering apparatus comprising:

a first buffer, arranged to buffer a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture;

a second buffer, arranged to buffer a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture; and

a multiplexer, coupled to one of the first buffer and the second buffer;

wherein when the first tile is currently decoded according to the first context array buffered in the first buffer, the second context array is buffered in the second buffer; entropy decoding of the second tile is started before the first tile is fully entropy decoded; and when entropy decoding of the first tile encounters a tile boundary, the multiplexer switches between the first buffer and the second buffer.

2. The buffering apparatus of claim 1, wherein when the entropy decoding of the first tile encounters the tile boundary, the first context array is stored into the second buffer, and the second context array is loaded into the first buffer.

3. The buffering apparatus of claim 2, further comprising:

a buffer access enhancement circuit, coupled between the first buffer and the second buffer, for pre-fetching the second context array from the second buffer or post-storing the first context array into the second buffer.

4. The buffering apparatus of claim 1, wherein when the second tile is currently decoded according to the second context array buffered in the second buffer, the first context array is buffered in the first buffer.

5. The buffering apparatus of claim 1, wherein the multiplexer has a plurality of first connection ports and a second connection port; wherein the first buffer and the second buffer are coupled to a first specific port and a second specific port included in the first connection ports, respectively.

6. The buffering apparatus of claim 5, wherein when the entropy decoding of the first tile encounters the tile boundary, the multiplexer switches an interconnection between the second connection port and the first specific port to an interconnection between the second connection port and the second specific port.

7. The buffering apparatus of claim 1, wherein the first tile and the second tile are dependent tiles.

8. The buffering apparatus of claim 1, wherein at least one of the first buffer and the second buffer is a register, an internal buffer or an external buffer of an entropy decoder.

9. The buffering apparatus of claim 1, wherein the multi-tile encoded picture has N horizontally adjacent partitions, and a number of buffered context arrays maintained by the buffering apparatus during entropy decoding of the multi-tile encoded picture is equal to N.

10. A buffering method for buffering context arrays of a multi-tile encoded picture having a plurality of tiles, the buffering method comprising:

buffering a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture;

buffering a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture when the first tile is currently decoded according to the buffered first context array; and

when entropy decoding of the first tile encounters a tile boundary, performing a multiplexing operation to switch between the buffered first context array and the buffered second context array;

wherein entropy decoding of the second tile is started before the first tile is fully entropy decoded.

11. The buffering method of claim 10, wherein the first context array is buffered in a first buffer, the second context array is buffered in a second buffer, and the buffering method further comprises:

when the entropy decoding of the first tile encounters the tile boundary, storing the first context array into the second buffer, and loading the second context array into the first buffer.

12. The buffering method of claim 11, further comprising:

pre-fetching the second context array from the second buffer; or

post-storing the first context array into the second buffer.

13. The buffering method of claim 10, wherein the step of buffering the first context array comprises:

buffering the first context array when the second tile is currently decoded according to the second context array.

14. The buffering method of claim 10, wherein the step of performing the multiplexing operation to switch between the buffered first context array and the buffered second context array comprises:

outputting the buffered second context array to substitute for the buffered first context array.

15. The buffering method of claim 10, wherein the first tile and the second tile are dependent tiles.

16. The buffering method of claim 10, wherein at least one of the first context array and the second context array is stored in a register, an internal buffer or an external buffer of an entropy decoder.

17. The buffering method of claim 10, wherein the multi-tile encoded picture has N horizontally adjacent partitions, and a number of buffered context arrays maintained during entropy decoding of the multi-tile encoded picture is equal to N.

18. An entropy decoder, comprising:

an entropy decoding core, arranged to perform entropy decoding upon a multi-tile encoded picture, having a plurality of tiles included therein, in a raster scan order, wherein the entropy decoding core starts decoding a portion of a current tile after decoding a portion of a previous tile; and

a buffering apparatus, coupled to the entropy decoding core, for buffering context arrays of the multi-tile encoded picture, the buffering apparatus comprising: a first buffer, arranged to buffer a first context array referenced for performing entropy decoding upon a first tile of the multi-tile encoded picture; a second buffer, arranged to buffer a second context array referenced for performing entropy decoding upon a second tile of the multi-tile encoded picture; and a multiplexer, coupled to one of the first buffer and the second buffer;

wherein when the first tile is currently decoded according to the first context array buffered in the first buffer, the second context array is buffered in the second buffer; entropy decoding of the second tile is started before the first tile is fully entropy decoded; and when entropy decoding of the first tile encounters a tile boundary, the multiplexer switches between the first buffer and the second buffer.

19. The entropy decoder of claim 18, wherein the multi-tile encoded picture has N horizontally adjacent partitions, and a number of buffered context arrays maintained by the buffering apparatus during entropy decoding of the multi-tile encoded picture is equal to N.