Circuit and method for generating fixed point vector dot product and matrix vector values
View Patent ↗An integrated circuit ( 102 ) and method computes fixed point vector dot products ( 424 ) and/or matrix vector products using a type of distributed architecture that loads bit planes (add 00 -add 30 ) and uses the loaded bit planes to generate a plurality of partial products ( 416 - 422 ) directly, such as without a lookup table, and the plurality of partial products are computed in real time and are not read out of addressable memory. In one example, pixel coefficients and corresponding data are loaded such that, for example, a bit plane is loaded to generate partial product results on a per bit plane basis. The plurality of partial products are then summed ( 414 ) or accumulated to produce fixed point vector dot product data ( 424 ).
1. An integrated circuit for computing fixed point vector dot products comprising:
circuitry operative to loading bit planes of data and for using the loaded bit planes to generate a plurality of partial products directly wherein the plurality of partial products are computed in real time and are not read out of addressable memory; and
circuitry operative to generate decoded pixel information from transform coefficients generated from encoded video data using the circuitry operative to load bit planes of data wherein the data comprises transform coefficients and weighting values and for using the loaded bit planes and weighting values to generate the plurality of partial products.
2. The integrated circuit of claim 1 wherein the circuitry operative to load bit planes of data and for using the loaded bit planes to generate a plurality of partial products directly comprises summing logic operative to sum the plurality of partial products to produce fixed point vector dot product data.
3. The integrated circuit of claim 1 wherein the circuitry operative to load bit planes and for using the loaded bit planes to generate a plurality of partial products directly comprises:
a plurality of partial product generation circuits;
a coefficient register array containing at least a plurality of coefficients or weighting values;
a data register array containing at least a portion of a plurality of bit planes; and
control logic operatively coupled to the register array and the data register array and operative to sequence loading of corresponding coefficients or weighting values and data bits for each of the plurality of partial product generation circuits.
4. The integrated circuit of claim 3 wherein the plurality of partial product generation circuits generate a respective partial product result on a per bit plane basis.
5. The integrated circuit of claim 3 wherein each of the plurality of partial product generation circuits comprises:
at least a first multiplier circuit comprising a switching structure having inputs operatively coupled to receive coefficients from the coefficient register array and operatively controlled by data of a bit plane; and
a plurality of adder stages operatively coupled to outputs of the switching structure.
6. The integrated circuit of claim 5 wherein the plurality of partial product generation circuits comprises bit shifting logic operatively coupled to an output of a last adder stage and further comprising a sign bit processing circuit comprising a switching structure having inputs operatively coupled to receive coefficients from the coefficient register array and operatively controlled by data of a bit plane; and
a plurality of adder stages operatively coupled to outputs of the switching structure; an inverter operatively coupled to an output of a last adder stage of the plurality of adder stages and an adder coupled to receive an output of the inverter.
7. The integrated circuit of claim 1 comprising circuitry operative to generate the transform coefficients based on encoded video data and wherein the circuitry operative to load bit planes of transform coefficients and weighting values and for using the loaded bit planes and transform coefficients to generate a plurality of partial products directly comprises summing logic operative to sum the plurality of partial products to produce decoded pixel information based on fixed point vector dot product data.
8. The integrated circuit of claim 1 wherein the circuitry operative to load bit planes and for using the loaded bit planes to generate a plurality of partial products directly comprises:
a plurality of partial product generation circuits;
a coefficient register array containing at least a plurality of weighting values;
a data register array containing at least a portion of a plurality of bit planes of transform coefficients; and
control logic operatively coupled to the register array and the data register array and operative to sequence loading of corresponding coefficients and data bits for each of the plurality of partial product generation circuits;
and wherein the plurality of partial product generation circuits generate a respective partial product result on a per bit plane basis to produce a portion of luma inverse transform coefficient data.
9. The integrated circuit of claim 1 comprising chroma transform coefficient logic operative to produce pixel data for display or further processing based on the luma inverse transform coefficient data.
10. A video decoding circuit comprising:
discrete cosine transform logic operative to generate discrete transform coefficient matrix data from encoded video data;
circuitry operative to generate decoded pixel information in the form of inverse transform matrix data for luma discrete cosine transformation coefficients from the discrete transform coefficient matrix data comprising:
vector multipliers that include circuitry operative to load bit planes of data transform coefficients and weighting values and for using the loaded bit planes of transform coefficients to generate a plurality of partial products directly wherein the plurality of partial products are computed in real time; and
circuitry operative to generate blocks of decoded pixels using the decoded pixel information in the form of inverse transform matrix data for luma discrete cosine transformation coefficients.
11. The video decoding circuit of claim 10 wherein the circuitry operative to load bit planes and for using the loaded bit planes to generate a plurality of partial products directly comprises:
a plurality of partial product generation circuits;
a coefficient register array containing at least a plurality of weighting values;
a data register array containing at least a portion of a plurality of bit planes of transform coefficients; and
control logic operatively coupled to the register array and the data register array and operative to sequence loading of corresponding coefficients and data bits for each of the plurality of partial product generation circuits.
12. The video decoding circuit of claim 11 wherein the plurality of partial product generation circuits generate a respective partial product result on a per bit plane basis to produce a portion of luma inverse transform coefficient data.
13. A method comprising:
generating discrete cosine transformation coefficients associated with a block of encoded video data;
generating decoded pixel information using circuitry operative to load bit planes of data and for using the loaded bit planes to generate a plurality of partial products directly wherein the plurality of partial products are computed in real time and are not read out of addressable memory;
further processing the decoded pixel information to produce decoded pixels for output to a display; and
displaying the decoded pixels on a display.
14. The method of claim 13 comprising sequencing loading of corresponding discrete cosine coefficients and data bits for each of a plurality of partial product generation circuits.
15. The method of claim 13 wherein generating decoded pixel information comprises generating a respective partial product result on a per bit plane basis to produce a portion of luma inverse transform coefficient data.
16. The method of claim 13 comprising loading discrete cosine transform coefficients into a coefficient register array prior to providing the bit planes to partial product generation circuits.
17. The method of claim 16 comprising, in subsequent clock cycles, enabling a data register array to provide bit planes for each of a respective partial product generation circuit.
18. The method of claim 17 comprising summing a plurality of partial products produced by the respective partial product generation circuits to produce fixed point vector dot product data.
19. The method of claim 18 comprising using the fixed point vector dot product data to produce pixel information.