SIMD four-pixel average instruction for imaging and video applications

• US 20050213842A1
• Filed: 03/26/2004
• Published: 09/29/2005
• Est. Priority Date: 03/26/2004
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Liu, Jianwei, Aldrich, Bradley C., Paver, Nigel C.
• Granted Patent: US 7,529,423 B2
• Time in Patent Office: Days
• US Class Current: 382/264
• CPC Class Codes: G06T 3/4007 based on interpolation, e.g. bilinear interpolation image demosaicing G06T3/4015; edge-driven or edge-based scaling G06T3/403
• CPC Class Codes: G06T 3/4023 based on decimating pixels or lines of pixels; based on inserting pixels or lines of pixels
• CPC Class Codes: H04N 19/42 characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation H04N19/635 takes precedence
• CPC Class Codes: H04N 19/51 Motion estimation or motion compensation

Abstract

According to some embodiments, a Single-Instruction/Multiple-Data (SIMD) averaging instruction is used to process pixels of image data. The averaging instruction generates a set of four-pixel averages, where each average is generated from two pixels in a first source register and two pixels in a second source register. The first source register contains a plurality of pixels from a first row of pixels and the second source register contains a plurality of pixels from a second row. In one embodiment, the first and second rows are adjacent rows in an image and the averaging instruction is used, for example, to down-scale an image, perform color conversion, and the like. In another embodiment, the first and second rows are from different images and the averaging instruction is used, for example, in motion estimation for video encoding, in motion compensation for video decoding, and the like.

Claims

1. A method comprising:

receiving image data; and

executing an averaging instruction to average a plurality of pixels of the image data;

wherein executing the averaging instruction generates a set of four-pixel averages, each four-pixel average generated from two pixels in a first source register and two pixels in a second source register; and

wherein executing the averaging instruction comprises;

adding successive pixels from the first source register to successive pixels of the second source register generating a plurality of intermediate results; and

adding two of the plurality of intermediate results and repeating with different combinations of the plurality of intermediate results generating a plurality of sum results.

2. The method as recited in claim 1, wherein the two pixels in the first source register and the two pixels in the second source register are adjacent pixels in an image.

3. The method as recited in claim 2, wherein the adjacent pixels are adjacent in both horizontal and vertical dimensions.

4. The method as recited in claim 1, wherein the executing the averaging instruction further comprises:

discarding the two least significant bits of each sum result of the plurality of sum results generating the set of four-pixel averages.

5. The method as recited in claim 1, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is adjacent to the second row in an image.

6. The method as recited in claim 5, further comprising:

repeating the executing the averaging instruction for all pixels in the image; and

discarding a portion of generated four-pixel averages, generating a down-scaled image.

7. The method as recited in claim 6, wherein the discarding the portion of the generated four-pixel averages comprises:

masking every other four-pixel average using a AND operation;

executing a pack instruction that compresses two eight element registers into a single eight element register generating the down-scaled image.

8. The method as recited in claim 1, wherein the set of four-pixel averages comprise interpolated pixel values utilized for video coding.

9. The method as recited in claim 8, wherein the video coding is a motion estimation encoding.

10. The method as recited in claim 8, wherein the video coding is a motion compensation decoding.

11. The method as recited in claim 1, wherein the set of four-pixel averages are utilized for image color space conversion.

12. The method as recited in claim 1, wherein the set of four-pixel averages are utilized for image color format conversion.

13. The method as recited in claim 1, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is from a different image than the second row.

14. The method as recited in claim 13, wherein the set of four-pixel averages comprise averaged pixel values utilized for motion estimation video encoding.

15. The method as recited in claim 13, wherein the set of four-pixel averages comprise averaged pixel values utilized for motion compensation video decoding.

16. The method as recited in claim 1, wherein the averaging instruction is a Single-Instruction/Multiple-Data (SIMD) instruction.

17. The method as recited in claim 1, wherein the first source register comprises a plurality of pixels from a row of pixels and the second source register comprises another plurality of pixels from the row of pixels, and wherein each pixel of the plurality of pixels is adjacent in an image to at least one pixel of the another plurality of pixels.

18. The method as recited in claim 17, wherein the executing the averaging instruction comprises a portion of a separable filter implementation.

19. A system comprising:

an addressable memory to store an instruction for an averaging operation;

a processing core coupled to the addressable memory, the processor core comprising;

an execution core to access and execute the instruction;

a first source register to store a first plurality of pixels;

a second source register to store a second plurality of pixels; and

a destination register to store a plurality of results of the averaging operation;

a wireless interface to receive a digital signal comprising a third plurality of pixels; and

an I/O system to provide the first and second plurality of pixels to the first and second source registers from the third plurality of pixels;

wherein to execute the instruction, the execution core is operative to;

generate a set of four-pixel averages, each four-pixel average generated from two pixels in the first source register and two pixels in the second source register.

20. The system as recited in claim 19, wherein to generate the set of four-pixel averages the execution core is operative to:

add successive ones of the first plurality of pixels to successive ones of the second plurality of pixels generating a plurality of intermediate results;

add two of the plurality of intermediate results and repeating with different combinations of the plurality of intermediate results generating a plurality of sum results; and

discard the two least significant bits of each sum result of the plurality of sum results generating the set of four-pixel averages.

21. The system as recited in claim 19, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is adjacent to the second row in an image.

22. The system as recited in claim 19, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is from a different image than the second row.

23. The system as recited in claim 19, wherein the first source register comprises a plurality of pixels from a row of pixels and the second source register comprises another plurality of pixels from the row of pixels, and wherein each pixel of the plurality of pixels is adjacent in an image to at least one pixel of the another plurality of pixels.

24. The system as recited in claim 19, wherein the averaging instruction is a Single-Instruction/Multiple-Data (SIMD) instruction.

25. A system comprising:

an addressable memory to store an instruction for an averaging operation;

a processing core coupled to the addressable memory, the processor core comprising;

an execution core to access and execute the instruction;

a first source register to store a first plurality of pixels;

a second source register to store a second plurality of pixels; and

a destination register to store a plurality of results of the averaging operation;

a video capture interface to receive a digital signal comprising a third plurality of pixels; and

an I/O system to provide the first and second plurality of pixels to the first and second source registers from the third plurality of pixels;

wherein to execute the instruction, the execution core is operative to;

generate a set of four-pixel averages, each four-pixel average generated from two pixels in the first source register and two pixels in the second source register.

26. The system as recited in claim 25, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is adjacent to the second row in an image.

27. The system as recited in claim 25, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is from a different image than the second row.

28. The system as recited in claim 25, wherein the first source register comprises a plurality of pixels from a row of pixels and the second source register comprises another plurality of pixels from the row of pixels, and wherein each pixel of the plurality of pixels is adjacent in an image to at least one pixel of the another plurality of pixels.

29. The system as recited in claim 25, wherein the instruction is a Single-Instruction/Multiple-Data (SIMD) instruction.

30. An article comprising a machine-readable medium that includes machine readable instructions, the instructions operative to cause a machine to:

receive image data; and

execute an averaging instruction to average a plurality of pixels of the image data, wherein executing the averaging instruction generates a set of four-pixel averages, each four-pixel average generated from two pixels in a first source register and two pixels in a second source register.

31. The article as recited in claim 30, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is adjacent to the second row in an image.

32. The article as recited in claim 31, wherein the instructions are further operative to cause the machine to:

repeat the executing the averaging instruction for all pixels in the image; and

discard a portion of generated four-pixel averages, generating a down-scaled image.

33. The article as recited in claim 32, wherein the instructions operative to cause the machine to discard the portion of generated four-pixel averages comprise:

an AND instruction to mask alternating four-pixel averages; and

a PACK instruction to compress two eight element registers into a single eight element register to generate the down-scaled image.

34. The article as recited in claim 30, wherein the first source register comprises a plurality of pixels from a first row of pixels and the second source register comprises a plurality of pixels from a second row of pixels, and wherein the first row is from a different image than the second row.

35. The article as recited in claim 30, wherein the averaging instruction is a Single-Instruction/Multiple-Data (SIMD) instruction.

36. The article as recited in claim 30, wherein the averaging instruction is operative to cause the machine to:

add successive pixels from the first source register to successive pixels of the second source register generating a plurality of intermediate results;

add two of the plurality of intermediate results and repeating with different combinations of the plurality of intermediate results generating a plurality of averaging results; and

discard the two least significant bits of each averaging result of the plurality of averaging results generating the set of four-pixel averages.

37. The article as recited in claim 30, wherein the first source register comprises a plurality of pixels from a row of pixels and the second source register comprises another plurality of pixels from the row of pixels, and wherein each pixel of the plurality of pixels is adjacent in an image to at least one pixel of the another plurality of pixels.

38. The article as recited in claim 37, wherein the executing the averaging instruction comprises a portion of a separable filter implementation.