Sparse matrix optimization mechanism
An apparatus to facilitate matrix processing is disclosed. The apparatus comprises a matrix accelerator to receive input matrix data, transform the input matrix data into a plurality of sub-blocks, examine a first block of the sub-blocks to determine whether the first block comprises sparse data, select a first tile size upon a determination that the first block comprises sparse data and generate output matrix data based on the first tile size.
1. An apparatus comprising:
a matrix accelerator to receive input matrix data, transform the input matrix data into a plurality of sub-blocks, retrieve bitmap data having a plurality sub-block bits, each associated with one of the plurality of sub-blocks, examine sub-block bits associated with adjacent sub-blocks in a first block to determine a tile size that is to be processed, select a first tile size for processing of the sub-blocks upon a determination that each of the sub-block bits within the bitmap data comprise a first bit value and process the sub-blocks to generate output matrix data.
2. The apparatus of claim 1 , wherein the first bit value indicates that a sub-block comprises sparse data.
3. The apparatus of claim 2 , wherein the first tile size comprises a size of the first block.
4. The apparatus of claim 3 , wherein the matrix accelerator selects a second tile size for processing of the sub-blocks upon a determination that at least one of the sub-block bits comprise a second bit value.
5. The apparatus of claim 4 , wherein the second bit value indicates that a sub-block does not comprise sparse data.
6. The apparatus of claim 5 , wherein the second tile size comprises a size of a sub-block.
7. The apparatus of claim 6 , further comprising a systolic pipeline that comprises a first configuration to process each sub-block.
8. The apparatus of claim 7 , wherein the systolic pipeline is transformed to a second configuration to process the first block of sub-blocks upon a determination that each of the sub-block bits comprise the first bit value.
9. A method comprising:
receiving input matrix data;
transforming the input matrix data into a plurality of sub-blocks;
retrieving bitmap data having a plurality sub-block bits, each associated with one of the plurality of sub-blocks;
examining sub-block bits associated with adjacent sub-blocks in a first block to determine a tile size that is to be processed block comprises sparse data;
selecting a first tile size for processing of the sub-blocks upon a determination that each of the sub-block bits within the bitmap data comprise a first bit value; and
processing the sub-blocks to generate output matrix data.
10. The method of claim 9 , wherein the first bit value indicates that a sub-block comprises sparse data.
11. The method of claim 10 , wherein the first tile size comprises a size of the first block.
12. The method of claim 11 , further comprising selecting a second tile size for processing of the sub-blocks upon a determination that at least one of the sub-block bits comprise a second bit value.
13. The method of claim 12 , wherein the second bit value indicates that a sub-block does not comprise sparse data.
14. The method of claim 13 , wherein the second tile size comprises a size of a sub-block.
15. At least one non-transitory computer readable medium having instructions stored thereon, which when executed by one or more processors, cause the processors to:
receive input matrix data;
transform the input matrix data into a plurality of sub-blocks;
retrieve bitmap data having a plurality sub-block bits, each associated with one of the plurality of sub-blocks;
examine sub-block bits associated with adjacent sub-blocks in a first block to determine a tile size that is to be processed block comprises sparse data;
select a first tile size for processing of the sub-blocks upon a determination that each of the sub-block bits within the bitmap data comprise a first bit value; and
process the sub-blocks to generate output matrix data.
16. The computer readable medium of claim 15 , wherein the first bit value indicates that a sub-block comprises sparse data.
17. The computer readable medium of claim 16 , wherein the first tile size comprises a size of the first block.
18. The computer readable medium of claim 17 , having instructions stored thereon, which when executed by one or more processors, further cause the processors to select a second tile size for processing of the sub-blocks upon a determination that at least one of the sub-block bits comprise a second bit value.
19. The computer readable medium of claim 18 , wherein the second bit value indicates that a sub-block does not comprise sparse data.
20. The computer readable medium of claim 19 , wherein the second tile size comprises a size of a sub-block.