Saturation detection in floating point to integer conversions

• US 6,195,672 B1
• Filed: 08/31/1998
• Issued: 02/27/2001
• Est. Priority Date: 08/31/1998
• Status: Expired due to Term

• Current Assignee: Mentor Graphics Corporation (Siemens AG)
• Original Assignee: Mentor Graphics Corporation (Siemens AG)
• Inventors: Hossain, Razak, Herbert, Jeffrey Charles, Gouger, Jason F.
• Primary Examiner(s): Donaghue, Larry D.
• Assistant Examiner(s): Chang, Jungwon
• Application Number:
  US09/144,599
• Time in Patent Office: 911 Days
• Field of Search: 708/204, 708/498, 708/552, 708/553, 708/277, 708/495, 708/497, 708/209, 708/551, 712/222, 712/221, 345/523
• US Class Current: 708/204
• CPC Class Codes: G06F 7/49921 Saturation, i.e. clipping the result to a minimum or maximum value
• CPC Class Codes: H03M 7/24 Conversion to or from floating-point codes

Abstract

An improved method and apparatus for saturation detection in floating point to integer conversions is described. A floating point number is tested for saturation conditions based on an integer field size. From testing the saturation conditions on the floating point number, the present invention predicts whether a floating point number can be converted into an integer value having the given integer field size, or whether the integer field would be saturated. In one embodiment, the saturation conditions are tested on the floating point number in parallel with a floating point to integer conversion.

Claims

1. A method comprising:

receiving a floating point number and an integer field size indicating a number of bit positions in an integer field;

comparing an exponent value of the floating point number to the integer field size to produce a set of relational outputs;

determining whether ones or zeros are present in particular bit positions of a mantissa of the floating point number to produce a set of boundary condition outputs; and

predicting whether the floating point number when converted to an integer value would saturate the integer field based on at least one of a sign of the floating point number, the set of relational outputs, and the set of boundary condition outputs.

2. The method of claim 1 wherein the method is performed in parallel with a floating point to integer conversion.

3. The method of claim 1 wherein the set of relational outputs indicates whether a most significant one of the floating point number would be in a bit position more significant than a most significant bit position in the integer field, wherein the integer field would be saturated.

4. The method of claim 3 wherein the most significant one of the floating point number would be in a bit position more significant than a most significant bit position in the integer field if the exponent value is greater than or equal to the integer field size.

5. The method of claim 1 wherein the set of relational outputs indicates whether a most significant one of the floating point number would be in a most significant bit position of the integer field, wherein the integer field would be saturated if the integer value would be rounded up, or the floating point number is positive, or the floating point number is negative and less than all lower order bits of the integer value would be non-zero.

6. The method of claim 5 wherein the most significant one of the floating point number would be in a most significant bit position of the integer field if the exponent value is equal to one less than the integer field size.

7. The method of claim 5 wherein the integer value would be rounded up if a rounding bit of the set of boundary condition outputs is one, wherein the rounding bit is in a bit position in the mantissa of the floating point number equal to a number of bit position shifts needed to convert the floating point number to the integer value.

8. The method of claim 5 wherein the floating point number is positive if a sign bit of the floating point number is zero, and negative if a sign bit of the floating point number is one.

9. The method of claim 5 wherein the set of boundary condition outputs indicates that less than all of the lower order bits of the integer field would be non-zero if less than all bits in the mantissa of the floating point number from one bit position less significant than the most significant one to one bit position more significant than a rounding bit are zero, wherein the rounding bit is in a bit position in the floating point number equal to a number of bit position shifts needed to convert the floating point number to the integer value.

10. The method of claim 1 wherein the set of relational outputs indicates whether a most significant one of the floating point number would be in a second most significant bit position of the integer field, wherein the integer field would be saturated if the integer value would be rounded up, and the floating point number is positive, and all lower order bits of the integer value would be one.

11. The method of claim 10 wherein the most significant one of the floating point number would be in a second most significant bit position of the integer field if the exponent value is equal to two less than the integer field size.

12. The method of claim 10 wherein the integer value would be rounded up if a rounding bit of the set of boundary condition outputs is one, wherein the rounding bit is in a bit position in the mantissa of the floating point number equal to a number of bit position shifts needed to convert the floating point number to the integer value.

13. The method of claim 10 wherein the floating point number is positive if a sign bit of the floating point number is zero, and negative if a sign bit of the floating point number is one.

14. The method of claim 10 wherein the set of boundary condition outputs indicates that all of the lower order bits of the integer field would be one if all bits in the floating point number from one bit position less significant than the most significant one to one bit position more significant than a rounding bit are one, wherein the rounding bit is in a bit position in the mantissa of the floating point number equal to a number of bit position shifts needed to convert the floating point number to the integer value.

15. A machine readable storage medium having stored thereon machine executable instructions, the execution of said machine executable instructions to implement a method comprising:

receiving a floating point number and an integer field size indicating a number of bit positions in an integer field;

comparing an exponent value of the floating point number to the integer field size to produce a set of relational outputs;

determining whether ones or zeros are present in particular bit positions of a mantissa of the floating point number to produce a set of boundary condition outputs; and

predicting whether the floating point number when converted to an integer value would saturate the integer field based on at least one of a sign of the floating point number, the set of relational outputs, and the set of boundary condition outputs.

16. The machine readable storage medium of claim 15 wherein the method is performed in parallel with a floating point to integer conversion.

17. The machine readable storage medium of claim 15 wherein the

the set of relational outputs indicates whether a most significant one of the floating point number would be in a bit position more significant than a most significant bit position in the integer field, wherein the integer field would be saturated.

18. The machine readable storage medium of claim 15 wherein the

the set of relational outputs indicates whether a most significant one of the floating point number would be in a most significant bit position of the integer field, wherein the integer field would be saturated if the integer value would be rounded up, or the floating point number is positive, or the floating point number is negative and less than all lower order bits of the integer value would be non-zero.

19. The machine readable storage medium of claim 15 wherein the

the set of relational outputs indicates whether a most significant one of the floating point number would be in a second most significant bit position of the integer field, wherein the integer field would be saturated if the integer value would be rounded up, and the floating point number is positive, and all lower order bits of the integer value would be one.

20. An apparatus comprising:

an input port to receive a floating point number and an integer field size indicating a number of bit positions in an integer field;

a comparator to compare an exponent value of the floating point number to the integer field size to produce a set of relational outputs;

a plurality of logic elements to determine whether ones or zeros are present in particular bit positions of a mantissa of the floating point number to produce a set of boundary condition outputs; and

a combinational logic unit to predict whether the floating point number when converted to an integer value would saturate the integer field based on at least one of a sign of the floating point number, the set of relational outputs and the set of boundary condition outputs.

21. The apparatus of claim 20 wherein the apparatus is to operate in parallel with a floating point to integer conversion unit.

22. The apparatus of claim 20 wherein the

set of relational outputs indicates whether a most significant one of the floating point number would be in a bit position more significant than a most significant bit position in the integer field, wherein the integer field would be saturated.

23. The apparatus of claim 20 wherein the

set of relational outputs indicates whether a most significant one of the floating point number would be in a most significant bit position of the integer field, wherein the integer field would be saturated if the integer value would be rounded up, or the floating point number is positive, or the floating point number is negative and less than all lower order bits of the integer value would be non-zero.

24. The apparatus of claim 20 wherein the

set of relational outputs indicates whether a most significant one of the floating point number would be in a second most significant bit position of the integer field, wherein the integer field would be saturated if the integer value would be rounded up, and the floating point number is positive, and all lower order bits of the integer value would be one.

25. The apparatus of claim 20 wherein the plurality of logic elements comprise:

a rounding detector to predict if the integer value would be rounded up;

a sign detector to detect a sign bit of the floating point number; and

a zeros/ones detector to predict if less than all lower order bits of the integer field would be non-zero.