Tile level interconnect design for central processing unit image access patterns
Aspects of the disclosure are directed to providing access to multimedia data in a tile format. In accordance with one aspect, the disclosure includes detecting one or more dependencies of a first tile read request with a second tile read request by classifying a head node and a dependent node; sending the first tile read request to a main memory; storing the dependent node in a buffer; and outputting a first cache line and a second cache line using the first tile read request.
1 . An apparatus for implementing access to a multimedia data in a tile format, the apparatus comprising:
means for storing a first cache line and a second cache line in the tile format;
means for detecting one or more dependencies of a first tile read request with a second tile read request by classifying a head node and a dependent node;
means for sending the first tile read request to a main memory;
means for storing the dependent node;
means for outputting the first cache line and the second cache line using the first tile read request; and
means for clearing the dependent node after outputting the first cache line and the second cache line.
2 . A method comprising:
detecting one or more dependencies of a first tile read request with a second tile read request by classifying a head node and a dependent node;
sending the first tile read request to a main memory;
storing the dependent node in a buffer; and
outputting a first cache line and a second cache line using the first tile read request.
3 . The method of claim 2 , wherein the first cache line or the second cache line is a multimedia data.
4 . The method of claim 2 , further comprising storing the first cache line and the second cache line in a tile format.
5 . The method of claim 2 , further comprising clearing the dependent node after outputting the first cache line and the second cache line.
6 . The method of claim 5 , further comprising converting a plurality of cache line read requests into the first tile read request.
7 . The method of claim 6 , wherein the first tile read request corresponds to a first of the plurality of cache line read requests.
8 . The method of claim 7 , wherein the second tile read request corresponds to a second of the plurality of cache line read requests.
9 . The method of claim 6 , further comprising receiving the plurality of cache line read requests from a processor.
10 . The method of claim 2 , further comprising classifying the first tile read request as the head node and classifying the second tile read request as the dependent node.
11 . The method of claim 10 , wherein the head node and dependent node are associated with a tile.
12 . The method of claim 11 , wherein the tile is a compression unit of an image and wherein the tile has a size which depends on a compression ratio.
13 . The method of claim 2 , further comprising detecting one or more dependencies of the first tile write request with a tile write request.
14 . The method of claim 13 , wherein a detection of the one or more dependencies of the first tile read request with the second tile read request or of the one or more dependencies of the first tile read request with the tile write request is facilitated by a tree-based data structure based on ordering of incoming read requests or write requests.
15 . The method of claim 14 , wherein the detection provides a dynamic switching between one or more read and write access paths of a main memory if there are dependencies among the first tile read request, the second tile read request, and the tile write request.