Processing core data compression and storage system

1. A method for compressing a set of data from a sparse matrix comprising:

evaluating a sequence of data entries from the set of data;

extracting a sequence of sparse data values from the sequence of data entries;

extracting a sequence of non-sparse data value run lengths from the sequence of data entries;

formulating, while extracting the sequence of sparse data values and the sequence of non-sparse data value run lengths, a set of row pointers from the sequence of data entries;

storing the sequence of sparse data values at a first contiguous set of memory addresses;

storing the sequence of non-sparse data value run lengths at a second contiguous set of memory addresses; and

wherein the set of row pointers identify a set of rows of the sparse matrix in both the first and second contiguous sets of memory addresses.

2. The method of claim 1 , wherein:

the sparse data values are non-zero values; and

the non-sparse data value run lengths are zero value run lengths in the sequence of data entries.

3. The method of claim 1 , wherein:

the sequence of sparse data values and the sequence of non-sparse data value run lengths share an equivalent number of elements;

the row pointers provide offsets into the first set of contiguous memory addresses and the second set of contiguous memory addresses; and

the sequence of data entries is a sequence of values from the sparse matrix with a start of each new row placed in sequence with an end of a prior row to that new row.

4. The method of claim 1 , wherein extracting the sequence of sparse data values further comprises:

appending, while extracting the sequence of sparse data values, a non-sparse data value to a current sequence of sparse data values when the non-sparse data value is a first entry in a row of the sparse matrix; and

appending, while extracting the sequence of non-sparse data value run lengths, a zero value to a current sequence of non-sparse data value run lengths in response to the appending of the non-sparse data value to the current sequence of sparse data values.

5. The method of claim 4 , further comprising:

storing the set of row pointers a third contiguous set of memory addresses;

wherein the sequence of data entries includes at least two rows of the sparse matrix; and

wherein the non-sparse data value run lengths are zero value run lengths in the sequence of data entries.

6. The method of claim 1 , further comprising:

storing the set of row pointers at a third contiguous set of memory addresses;

wherein the non-sparse data value run lengths are zero value run lengths in the sequence of data entries; and

wherein the sparse data values are non-zero values.

7. The method of claim 6 , further comprising:

generating a single compressed data tile;

the first contiguous set of memory addresses is mapped to a first row in the single compressed data tile;

the second contiguous set of memory addresses is mapped to a second row in the single compressed data tile; and

the third contiguous set of memory addresses is mapped to a third row in the single compressed data tile.

8. The method of claim 1 , further comprising:

generating a single compressed data tile;

the single compressed data tile is defined by a row length and has a number of rows;

a first portion of the number of rows hold the sequence of sparse data values from the sequence of data entries; and

a second portion of the number of rows hold the sequence of non-sparse data value run lengths.

9. The method of claim 8 , wherein:

the first contiguous set of memory addresses store one of: 8-bit, 16-bit, and 32-bit floating point numbers.

10. The method of claim 9 , wherein the sparse matrix is generated from a sparse tensor,

and further comprises:

generating a mapping from an element of the sparse tensor to an element of the sparse matrix; and

wherein an element of the sparse tensor can be extracted from the single compressed data tile using the mapping.

11. A system for compressing a set of data from a sparse matrix comprising:

a memory; and

a clocked logic that evaluates a sequence of data entries from the set of data;

wherein the clocked logic is configured to execute the following steps in response to the evaluation of the sequence of data entries from the set of data:

extracting a sequence of sparse data values from the sequence of data entries;

extracting a sequence of non-sparse data value run lengths from the sequence of data entries;

formulating a set of row pointers from the sequence of data entries;

storing the sequence of sparse data values at a first set of memory addresses in the memory; and

storing the sequence of non-sparse data value run lengths at a second set of memory addresses in the memory; and

wherein the set of row pointers identify a set of rows of the sparse matrix in both the first set of memory addresses and the second set of memory addresses.

12. The system of claim 11 , wherein:

the sparse data values are non-zero values; and

the non-sparse data value run lengths are zero value run lengths in the sequence of data entries.

13. The system of claim 11 , wherein:

the sequence of sparse data values and the sequence of non-sparse data value run lengths share an equivalent number of elements;

the row pointers provide offsets into the first set of memory addresses and the second set of memory addresses;

the first set of memory addresses are contiguous;

the second set of memory addresses are contiguous; and

the sequence of data entries is a sequence of values from the sparse matrix with a start of each new row placed in sequence with an end of a prior row to that new row.

14. The system of claim 13 , wherein:

the sequence of data entries includes at least two rows of the sparse matrix; and

the non-sparse data value run lengths are zero value run lengths in the sequence of data entries.

15. The system of claim 11 , further comprising:

a non-sparse data value mask generator that evaluates a sequence of data entries and generates a non-sparse data value flag for a set of datums in the sequence of data entries; and

wherein the non-sparse data value flags is used to formulate the set of row pointers.

16. The system of claim 11 , further comprising:

the clocked logic is further configured to store the set of row pointers at a third set of memory addresses;

wherein the non-sparse data value run lengths are zero value run lengths in the sequence of data entries; and

wherein the sparse data values are non-zero values.

17. The system of claim 16 , wherein:

the clocked logic is further configured to store the sequence of sparse data values and the sequence of non-sparse data value run lengths in a single compressed data tile;

the first set of memory addresses is mapped to a first row in the single compressed data tile;

the second set of memory addresses is mapped to a second row in the single compressed data tile; and

the third set of memory addresses is mapped to a third row in the single compressed data tile.

18. The system of claim 11 , wherein:

the clocked logic is further configured to store the sequence of sparse data values and the sequence of non-sparse data value run lengths in a single compressed data tile;

the single compressed data tile is defined by a row length and has a number of rows;

a first portion of the number of rows hold the sequence of sparse data values from the sequence of data entries; and

a second portion of the number of rows hold the sequence of non-sparse data value run lengths.

19. The system of claim 18 , wherein:

the first set of memory addresses store one of: 8-bit, 16-bit, and 32-bit floating point numbers.

20. A method for compressing a set of data from a sparse matrix comprising:

evaluating a sequence of data entries from the set of data;

extracting a sequence of sparse data values from the sequence of data entries;

extracting a sequence of non-sparse data value run lengths from the sequence of data entries;

formulating a set of row pointers from the sequence of data entries;

storing the sequence of sparse data values in a first set of memory addresses;

storing the sequence of non-sparse data value run lengths in a second set of memory addresses; and

wherein the set of row pointers identify a set of rows of the sparse matrix in both the first set of memory addresses and the second set of memory addresses.