System and method for enabling multiple signed independent data elements per register

• US 6,834,337 B1
• Filed: 09/29/2000
• Issued: 12/21/2004
• Est. Priority Date: 09/29/2000
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Trelewicz, Jennifer Q., Mitchell, Joan Laverne, Brady, Michael Thomas
• Primary Examiner(s): Pan, Daniel H.
• Application Number:
  US09/675,779
• Time in Patent Office: 1,544 Days
• Field of Search: 712/26, 712/210, 712/300, 712/245, 712/22, 712/221, 712/222, 712/223, 712/224, 708/552, 708/707, 708/705, 708/706, 708/672, 708/490, 708/503, 708/236, 708/629, 358/1.16, 710/307, 710/66
• US Class Current: 712/222
• CPC Class Codes: G06F 2207/3816 Accepting numbers of variable word length
• CPC Class Codes: G06F 2207/3828 Multigauge devices, i.e. capable of handling packed numbers without unpacking them
• CPC Class Codes: G06F 7/48 using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
• CPC Class Codes: G06F 7/4991 Overflow or underflow
• CPC Class Codes: G06F 7/49921 Saturation, i.e. clipping the result to a minimum or maximum value
• CPC Class Codes: G06F 7/49994 Sign extension
• CPC Class Codes: G06F 9/30014 with variable precision
• CPC Class Codes: G06F 9/30036 Instructions to perform operations on packed data, e.g. vector, tile or matrix operations
• CPC Class Codes: G06F 9/30109 having multiple operands in a single register

Abstract

A system and method for data processing includes packing multiple signed data elements per register into a processor'"'"'s registers using the rules set forth herein, and simultaneously operating on the elements in a register in a single cycle using the same operand. The elements can be independent of each other, and the sizes of the elements in a register can differ from each other. Moreover, a relatively large element can be split across multiple registers. In an exemplary application, a data stream representing two images can be simultaneously processed using the same number of registers as have been required to process a single image. Or, a single image can be processed approaching N-times faster, where N is the number of elements per register. In any case, the present invention results in a significant increase in processing efficiency.

Claims

1. A computer system, comprising:

a processor, the processor using logic for undertaking method acts to enable the simultaneous processing of multiple multi-bit data elements in a single register, the method acts undertaken by the logic including;

establishing at least first and second signed, multi-bit data elements in at least a first register;

simultaneously processing the elements, wherein the first element is provided from a first data set and the second element is provided from a second data set different than the first;

allocating a respective precision in a register for each data element to be processed in the register during a single cycle such that a maximum negative number that can be represented by a data element is one larger than the maximum negative number that can be represented in the respective precision.

2. The processor of claim 1, wherein the elements are independent of each other.

3. The processor of claim 1, wherein the register sends plural data elements simultaneously to at least one computational subsystem.

4. The processor of claim 1, wherein the operation is a multiplication by a constant or by a variable of known precision, or an addition, or a shift-left logical, or a subtraction, or a bitwise AND, or a bitwise OR.

5. The processor of claim 1, wherein the first element is a first partial element having a related second partial element established in a second register, and the first and second partial elements are combined after processing.

6. The processor of claim 1, wherein the method acts undertaken by the logic include:

determining whether a data element is contained around zero.

7. The processor of claim 1, wherein the method acts undertaken by the logic include:

preserving, in a first register, a sign bit for a data element packed in a second register;

zeroing at least one least significant bit to be shifted out of the data element pursuant to undertaking shift-right arithmetic;

performing an arithmetic shift right; and

restoring sign bits to the data element.

8. The processor of claim 7, wherein the method acts undertaken by the logic include:

if subsequent operations are to be personnel, subtracting the sign bits to repack the register.

9. The processor of claim 1, wherein the method acts undertaken by the logic include:

undertaking at least one equality confirmation on at least two data elements in a single register during a single processing cycle.

10. The processor of claim 1, wherein the method acts undertaken by the logic include:

undertaking at least one arithmetic compare on at least one comparison element in a single register containing at least two elements in a single processing cycle.

11. The processor of claim 10, wherein the arithmetic compare includes using the sign bit of a right-adjacent element to the comparison element.

13. The processor of claim 1, wherein the method acts undertaken by the logic include:

executing a program by;

packing multiple data elements into at least one single register;

simultaneously operating on all data elements in the single register; and

passing the data elements on for further processing, storage, or output after the operating act.

14. The processor of claim 13, wherein the method acts undertaken by the logic include:

for at least first and second data elements in a single register that are to be independent of each other, adding a sign bit in the first element to a cast significant bit in the second element.

12. A computer system, comprising:

a processor, the processor using logic for undertaking method acts to enable the simultaneous processing of multiple multi-bit data elements in a single register, the method acts undertaken by the logic including;

establishing at least first and second signed, multi-bit data elements in at least a first register;

simultaneously processing the elements;

determining a net number of bits required for any multiplicative constants to achieve a desired precision;

using the net number of bits and multiplicative constant, determining a net number of bits of precision; and

allocating space in the register in accordance with the using act.

15. A computer system comprising:

a processor, the processor using logic for undertaking method acts to enable the simultaneous processing of multiple multi-bit data elements in a single register, the method acts undertaken by the logic including;

establishing at least first and second signed, multi-bit data elements in at least a first register;

simultaneously processing the elements, wherein the first element is provided from a first data set and the second element is provided from a second data set different than the first;

splitting, into two or more sub-elements, at least one relatively large data element and packing the sub-elements into separate respective registers; and

recombining the sub-elements after processing.

16. A computer system, comprising:

a processor, the processor using logic for undertaking method acts to enable the simultaneous processing of multiple multi-bit data elements in a single register, the method acts undertaken by the logic including;

establishing at least first and second signed, multi-bit data elements in at least a first register;

simultaneously processing the elements; and

determining a precision bound for a data element based on [−

2<sup>N−

1</sup>+1] to [2<sup>N−

1</sup>−

1], wherein N is the number of bits in the data element.

17. A computer program device comprising:

a computer program storage device readable by a digital processing apparatus; and

a program on the program storage device and including instructions executable by the digital processing apparatus for performing method acts for processing multi-bit, signed data elements, the program comprising;

computer readable code means for packing at least first and second data elements into a single register;

computer readable code means for processing the elements simultaneously;

computer readable code means for determining a net number of bits required for any multiplicative constants to achieve a desired precision;

computer readable code means for using the net number of bits and multiplicative constant to determine a net number of bits of precision; and

computer readable code means for allocating space in the register in accordance with the using act.

18. The computer program device of claim 17, further comprising:

computer readable code means for allocating a respective precision in a register for each data element to be processed in the register during a single cycle such that the maximum negative number that can be represented by a data element is one larger than the maximum negative number that can be represented in the respective precision.

19. The computer program device of claim 17, further comprising:

computer readable code means for executing a program by;

packing multiple data elements into at least one single register according to the net number of bits of precision;

simultaneously operating on all data elements in the single register; and

passing the data elements on for further processing, storage, or output after the operating act.

20. The computer program device of claim 19, further comprising:

computer readable code means for, for at least first and second data elements in a single register that are to be independent of each other, adding a sign bit in the first element to a least significant bit in the second element.

21. The computer program device of claim 19, further comprising:

computer readable code means for splitting, into two or more sub-elements, at least one relatively large data element and packing the sub-elements into separate respective registers;

computer readable code means for recombining the sub-elements after processing.

22. The computer program device of claim 21, further comprising:

computer readable code means for determining whether a data element is contained around zero.

23. The computer program device of claim 22, further comprising:

computer readable code means for preserving, in a first register, a sign bit for a data element packed in a second register;

computer readable code means for zeroing at least one least significant bit to be shifted out of the data element pursuant to undertaking shift-right arithmetic;

computer readable code means for performing an arithmetic shift right; and

computer readable code means for restoring the sign bits and guard bits to the data element.

24. The computer program device of claim 23, further comprising:

computer readable code means for undertaking at least one equality confirmation on at least two data elements in a single register during a single processing cycle.

25. A method, comprising:

determining first and second precisions to be allocated in a single register to hold respective first and second signed data elements;

packing the elements into the register;

operating on the elements;

determining third and fourth precisions to be allocated in the single register to hold respective third and fourth signed data elements, at least one of;

the first and third precision being different from each other, and the second and fourth precisions being different from each other;

packing the third and fourth elements into the register; and

operating on the third and fourth elements.

26. The method of claim 25, wherein the register sends plural data elements simultaneously to at least one computational subsystem.

27. The method of claim 25, wherein the operation is a multiplication by a constant or by a variable of known precision, or an addition, or a shift-left logical, or a subtraction, or a bitwise AND, or a bitwise OR.

28. The method of claim 25, wherein the elements are independent of each other.

29. The method of claim 25, wherein the first element is provided from a first data set and the second element is provided from a second data set different than the first.

30. The method of claim 25, wherein the first element is a first partial element having a related second partial element established in a second register, and the first and second partial elements are combined after processing.

31. The method of claim 25, wherein the first element is provided from a first data set and the second element is provided from the first data set.

32. The method of claim 25, wherein the step of determining first and second precisions includes determining the precisions such that the maximum negative number that can be represented in an element is one larger than the maximum negative number that can be represented in the respective precision.

33. The method of claim 25, wherein the elements are processed such that in a single processing cycle a carry propagating left can only occur in the event of a previous borrow.

34. The method of claim 25, wherein at most one borrow per element can occur in a single processing cycle.

35. The method of claim 25, further comprising:

allocating a respective precision in a register for each data element to be processed in the register during a single cycle such that the maximum negative number that can be represented by a data element is one larger than the maximum negative number that can be represented in the respective precision.

36. The method of claim 35, further comprising:

determining a net number of bits required for any multiplicative constants to achieve a desired precision;

using the net number of bits and multiplicative constant, determining a net number of bits of precision; and

allocating space in the register in accordance with the using act.

37. The method of claim 36, further comprising:

executing a program by;

packing multiple data elements into at least one single register according to the net number of bits of precision;

simultaneously operating on all data elements in the single register; and

passing the data elements on for further processing, storage, or output after the operating act.

38. The method of claim 37, further comprising;

for at least first and second data elements in a single register that are to be independent of each other, adding a sign bit in the first element to a least significant bit in the second element.

39. The method of claim 37, further comprising:

splitting, into two or more sub-elements, at least one relatively large data element and packing the sub-elements into separate respective registers; and

recombining the sub-elements after processing.

40. The method of claim 35, further comprising determining a precision bound for a data element based on [−

2<sup>N−

1</sup>+1] to [2<sup>N−

1</sup>−

1], wherein N is the number of bits in the data element.

41. The method of claim 25, further comprising:

determining whether a data element is contained around zero.

42. The method of claim 25, further comprising:

preserving, in a first register, sign bits for a data element packed in a second register;

zeroing at least one least significant bit to be shifted out of the data element pursuant to undertaking shift-right arithmetic; and

restoring the sign bits to the data element.

43. The method of claim 25, further comprising:

undertaking at least one equality confirmation on at least two data elements in a single register during a single processing cycle.

44. A computer system, comprising:

a processor, the processor using logic for undertaking method acts to enable the simultaneous processing of multiple multi-bit data elements in a single register, the method acts undertaken by the logic including;

establishing at least first and second signed, multi-bit data elements in at least a first register;

simultaneously processing the elements;

for at least first and second data elements in a single register, masking sign bits in the first and second elements;

adding the sign bits back in to the register; and

discarding a value in a position of the sign bits in each element.

45. A computer program device comprising:

a computer program storage device readable by a digital processing apparatus; and

a program on the program storage device and including instructions executable by the digital processing apparatus for performing method acts for processing multi-bit, signed data elements, the program comprising;

computer readable code means for packing at least first and second data elements into a single register;

computer readable code means for processing the elements simultaneously;

computer readable code means for, for at least first and second data elements in a single register that are to be independent of each other, masking sign bits in the first and second elements;

computer readable code means for adding the sign bits back in to the register; and

computer readable code means for discarding a position of the sign bits in each element, with the possible exception of the sign bit for a left-most element in the register.

46. A method, comprising:

determining first and second precisions to be allocated in a single register to hold respective first and second signed data elements;

packing the elements into the register;

operating on the elements;

determining third and fourth precisions to be allocated in the single register to bold respective third and fourth signed data elements, at least one of;

the first and third precision being different from each other, and the second and fourth precisions being different from each other;

packing the third and fourth elements into the register;

operating on the third and fourth elements;

for at least first and second data elements in a single register that are to be independent of each other masking sign bits in the first and second elements;

adding the sign bits back in to the register; and

discarding a position of the sign bits in each element.