Techniques for performing a forward transformation by a video encoder using a forward transform matrix

1. An apparatus, comprising:

processing circuitry;

a memory coupled with the processing circuitry, the memory to store at least one master forward transform matrix comprising signed constants having a defined number of precision bits and a sign bit, and

the processing circuitry to perform the transformation on residuals of pixel values of a frame using one of the at least one master forward transform matrix or a forward transform matrix derived from one of the at least master forward transform matrix, and determine which forward transform matrix to use to perform the transformation based on at least a transform unit size, the processing circuitry to defer a rounding operation until a final forward transform matrix is generated.

2. The apparatus of claim 1 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix and a 4×4 asymmetric discrete sine transform (ADST) forward transform matrix.

3. The apparatus of claim 1 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

the processing circuitry to derive a 16×16 DCT forward transform matrix to perform the transformation by extracting the signed constants from every other row and the first 16 columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

4. The apparatus of claim 3 , the processing circuitry to derive an 8×8 DCT forward transform matrix to perform the transformation by extracting the signed constants from every other row and the first eight columns of the 16×16 DCT forward transform matrix starting with the first row and the first column.

5. The apparatus of claim 1 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

the processing circuitry to derive an 8×8 DCT forward transform matrix to perform the transformation by extracting the signed constants from every fourth row and the first eight columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

6. The apparatus of claim 5 , the processing circuitry to derive a 4×4 DCT forward transform matrix to perform the transformation by extracting the signed constants from every other row and the first four columns of the 8×8 DCT forward transform matrix starting with the first row and the first column.

7. The apparatus of claim 1 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

the processing circuitry to derive a 4×4 DCT forward transform matrix to perform the transformation by extracting the signed constants from every eighth row and the first four columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

8. The apparatus of claim 1 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

the processing circuitry to derive a 16×16 asymmetric discrete sine transform (ADST) forward transform matrix to perform the transformation by extracting the signed constants from every other row in reverse order and the first 16 columns of the 32×32 DCT forward transform matrix starting with a last row and a first column, wherein every other column is negated starting with a second column.

9. The apparatus of claim 1 , the at least one master forward transform matrix comprising a 32×32 DCT forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

the processing circuitry to derive an 8×8 asymmetric discrete sine transform (ADST) forward transform matrix to perform the transformation by extracting the signed constants from every fourth row in reverse order and the first eight columns of the 32×32 DCT forward transform matrix starting with a second to last row and a first column, wherein every other column is negated starting with a second column.

10. The apparatus of claim 1 , the memory comprising a read-only memory.

11. The apparatus of claim 1 , the processing circuitry to perform the transformation based on a VP9 video compression standard.

12. A computer-implemented method, comprising:

storing, in memory, at least one master forward transform matrix comprising signed constants having a defined number of precision bits and a sign bit;

determining, by processing circuitry, which forward transform matrix to use to perform a transformation based on at least a transform unit size;

performing, by the processing circuitry, the transformation on residuals of pixel values of a frame using one of the at least one master forward transform matrix or a forward transform matrix derived from one of the master forward transform matrix at least partially based on the determination; and

deferring, by the processing circuitry, a rounding operation until a final forward transform matrix is generated.

13. The computer-implemented method of claim 12 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix and a 4×4 asymmetric discrete sine transform (ADST) forward transform matrix.

14. The computer-implemented method of claim 12 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants; and

deriving a 16×16 DCT forward transform matrix to perform the transformation by extracting the signed constants from every other row and the first 16 columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

15. The computer-implemented method of claim 12 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants; and

deriving an 8×8 DCT forward transform matrix to perform the transformation by extracting the signed constants from every fourth row and the first eight columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

16. The computer-implemented method of claim 12 , the at least one master forward transform matrix comprising a 32×32 DCT forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

deriving a 4×4 DCT forward transform matrix to perform the transformation by extracting the signed constants from every eighth row and the first four columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

17. The computer-implemented method of claim 12 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

deriving a 16×16 asymmetric discrete sine transform (ADST) forward transform matrix to perform the transformation by extracting the signed constants from every other row in reverse order and the first 16 columns of the 32×32 DCT forward transform matrix starting with a last row and a first column, wherein every other column is negated starting with a second column.

18. The computer-implemented method of claim 12 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and

deriving an 8×8 asymmetric discrete sine transform (ADST) forward transform matrix to perform the transformation by extracting the signed constants from every fourth row in reverse order and the first eight columns of the 32×32 DCT forward transform matrix starting with a second to last row and a first column, wherein every other column is negated starting with a second column.

19. An article comprising a computer-readable storage medium comprising a plurality of instructions that when executed enable a system to:

store, in memory, at least one master forward transform matrix comprising signed constants having a defined number of precision bits and a sign bit;

determine, by processing circuitry, which forward transform matrix to use to perform a transformation based on at least a transform unit size;

perform, by processing circuitry, the transformation on residuals of pixel values of a frame using one of the at least one master forward transform matrices or a forward transform matrix derived from one of the master forward transform matrix at least partially based on the determination; and

defer, by processing circuitry, a rounding operation until a final forward transform matrix is generated.

20. The storage medium of claim 19 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transformation (DCT) forward transform matrix and a 4×4 asymmetric discrete sine transformation (ADST) forward transform matrix.

21. The storage medium of claim 19 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and the storage medium comprising instructions that when executed enable the system to derive a 16×16 DCT forward transform matrix to perform the transformation by extracting the signed constants from every other row and the first 16 columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

22. The storage medium of claim 19 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and the storage medium comprising instructions that when executed enable the system to derive an 8×8 DCT forward transform matrix to perform the transformation by extracting the signed constants from every fourth row and the first eight columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

23. The storage medium of claim 19 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and the storage medium comprising instructions that when executed enable the system to derive a 4×4 DCT forward transform matrix to perform the transformation by extracting the signed constants from every eighth row and the first four columns of the 32×32 DCT forward transform matrix starting with a first row and a first column.

24. The storage medium of claim 19 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and the storage medium comprising instructions that when executed enable the system to derive a 16×16 asymmetric discrete sine transform (ADST) forward transform matrix to perform the transformation by extracting the signed constants from every other row in reverse order and the first 16 columns of the 32×32 DCT forward transform matrix starting with a last row and a first column, wherein every other column is negated starting with a second column.

25. The storage medium of claim 19 , the at least one master forward transform matrix comprising a 32×32 discrete cosine transform (DCT) forward transform matrix further comprising 32 columns and 32 rows of signed constants, and the storage medium comprising instructions that when executed enable the system to derive an 8×8 asymmetric discrete sine transform (ADST) forward transform matrix to perform the transformation by extracting the signed constants from every fourth row in reverse order and the first eight columns of the 32×32 DCT forward transform matrix starting with a second to last row and a first column, wherein every other column is negated starting with a second column.