US4866651A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Method and circuit arrangement for adding floating point numbers
Assignee: International Business Machines Corp
•
Filing: 1987-08-26
•
Granted: 1989-09-12
•
Expires: 2007-08-26
Abstract
Abstract For successively adding a series of floating point numbers, a floating point adder stage (FIG. 2) is used which, in addition to the sum of two floating point operands, emits the remainder, truncated from the smaller operand, as a floating point number. For obtaining an exact sum of the operands, these remainders are summed in the form of intermediate sums. A circuit arrangement for parallel operation comprises series-connected floating point adder stages (FIG. 6), the intermediate sum occurring at the output of each stage and the intermediate remainder being buffered. Remainders are in each case passed on to the next stage, their value decreasing until they are zero. A serially operating arrangement (FIG. 8) comprises a single adder stage (30) and a register stack (34) for buffering the intermediate sums and the final result. A remainder occurring is stored in a remainder register (32) at the output of the adder stage and added to the intermediate sums until the remainder is zero. Subsequently, a fresh operand is applied to the input of the adder stage.
Inventors
- 👤 J. Hartmut Bleher
- 👤 Axel T. Gerlicher
- 👤 Siegfried M. Rump
- 👤 Dieter K. Unkauf
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-29 16:45
CPC Classifications
G; G06; G06F; G06F7/00; G06F7/38; G06F7/48; G06F7/483; G06F7/485
Claims Analysis
Total Claims: 17
Claim 1
Independent Claim
129 words
A method for summing an input set of fixed word length summands consisting of sign, exponent and mantissa, wherein the smaller summand of two operand input summands of the input set is right-shifted for obtaining identical exponents, and mantissa addition is carried out in a floating point adder stage, said method comprising the following steps for obtaining an exact end sum: (a) generating a sum signal (sum) at the output of the adder stage and a smaller remainder signal (remainder) truncated from the smaller operand by right shifting, both in floating point notation, (b) adding a newly entered summand or a remainder of the sum, (c) repeatedly adding remainders to sums during and after the input of the various summands until the remainders have been cancelled and sums generated.
Claim 2
Independent Claim
129 words
A method for summing an input set of fixed word length summands consisting of sign, exponent and mantissa, wherein the smaller summand of two operand input summands of the input set is right-shifted for obtaining identical exponents, and mantissa addition is carried out in a floating point adder stage, said method comprising the following steps for obtaining an exact end sum: (a) generating a sum signal (sum) at the output of the adder stage and a smaller remainder signal (remainder) truncated from the smaller operand by right shifting, both in floating point notation, (b) adding a newly entered summand or a remainder of the sum, (c) repeatedly adding remainders to sums during and after the input of the various summands until the remainders have been cancelled and sums generated.
Claim 3
Independent Claim
268 words
A method for summing an input set of fixed word length summands consisting of sign, exponent and mantissa (floating point operand), wherein the smaller summand of two operand input summands of the input set is right-shifted for obtaining identical exponents, and mantissa addition is carried out in a floating point adder stage, said method comprising the following steps for obtaining an exact end sum: (a) applying one of the input set summands at successive computation clocks to a first input (B input) of a first adder stage, (b) generating a sum signal (sum) at the output of the adder stage and a smaller remainder signal (remainder) truncated from the smaller operand by right shifting, both in floating point notation, (c) adding another one of the input set summands to the sum generated in that stage at the preceding clock and which sum is applied to a second input (A input), (d) transferring a remainder resulting from that addition to the first input of a subsequent adder stage and adding said remainder to a sum generated in that subsequent adder stage at the preceding clock, (e) transferring a remainder generated in the subsequent adder stage to the first input of a further adder stage, (f) retruncating a generated remainder in subsequent stages and retransferring said remainder, (g) adding the remainders to the sums in the individual adder stages after application of all summands until all remainders are zero, performing a floating point addition in steps (a) to (g) in all adder stages during a computation clock, the totality of the sums of the individual stages representing the final result.
Claim 4
Independent Claim
268 words
A method for summing an input set of fixed word length summands consisting of sign, exponent and mantissa (floating point operand), wherein the smaller summand of two operand input summands of the input set is right-shifted for obtaining identical exponents, and mantissa addition is carried out in a floating point adder stage, said method comprising the following steps for obtaining an exact end sum: (a) applying one of the input set summands at successive computation clocks to a first input (B input) of a first adder stage, (b) generating a sum signal (sum) at the output of the adder stage and a smaller remainder signal (remainder) truncated from the smaller operand by right shifting, both in floating point notation, (c) adding another one of the input set summands to the sum generated in that stage at the preceding clock and which sum is applied to a second input (A input), (d) transferring a remainder resulting from that addition to the first input of a subsequent adder stage and adding said remainder to a sum generated in that subsequent adder stage at the preceding clock, (e) transferring a remainder generated in the subsequent adder stage to the first input of a further adder stage, (f) retruncating a generated remainder in subsequent stages and retransferring said remainder, (g) adding the remainders to the sums in the individual adder stages after application of all summands until all remainders are zero, performing a floating point addition in steps (a) to (g) in all adder stages during a computation clock, the totality of the sums of the individual stages representing the final result.
Claim 5
Independent Claim
245 words
A method for summing an input set of fixed work length summands consisting of sign, exponent and mantissa (floating point operand), wherein a plurality of sums and a remainder are buffered, and the smaller summand of two operand input summands of the input set is right-shifted for obtaining identical exponents, and mantissa addition is carried out in a floating point adder stage comprising the steps of (a) feeding a summand to a first input (B input) of an adder stage and adding said summand to a first sum applied to a second input (A input), (b) generating a sum signal (sum) at the output of the adder stage and a smaller remainder signal (remainder) truncated from the smaller operand by right shifting, both in floating point notation, (c) storing the generated remainder and storing the generated sum in place of the first sum, (d) adding the remainder to a second sum and storing the generating remainder in place of the previously stored remainder and storing the resulting sum in place of said second sum, (e) adding in further arithmetic operations the respective remainder stored to the next sum and replacing the stored values by newly generated values, (f) applying a new summand and executing step (a) after the remainder to be stored has become zero, (g) terminating the arithmetic operation after all summands have been applied and the remainder to be stored has become zero, the totality of the stored sums representing the final result.
Claim 6
Independent Claim
245 words
A method for summing an input set of fixed work length summands consisting of sign, exponent and mantissa (floating point operand), wherein a plurality of sums and a remainder are buffered, and the smaller summand of two operand input summands of the input set is right-shifted for obtaining identical exponents, and mantissa addition is carried out in a floating point adder stage comprising the steps of (a) feeding a summand to a first input (B input) of an adder stage and adding said summand to a first sum applied to a second input (A input), (b) generating a sum signal (sum) at the output of the adder stage and a smaller remainder signal (remainder) truncated from the smaller operand by right shifting, both in floating point notation, (c) storing the generated remainder and storing the generated sum in place of the first sum, (d) adding the remainder to a second sum and storing the generating remainder in place of the previously stored remainder and storing the resulting sum in place of said second sum, (e) adding in further arithmetic operations the respective remainder stored to the next sum and replacing the stored values by newly generated values, (f) applying a new summand and executing step (a) after the remainder to be stored has become zero, (g) terminating the arithmetic operation after all summands have been applied and the remainder to be stored has become zero, the totality of the stored sums representing the final result.
Claim 7
Dependent Claim
50 words
The method of claim 3, wherein for storing the sums, a stack of storage registers is used, the number of storage registers which is kept variable by requesting a further register only if the remainder is other than zero and the last register of the stack has already been used.
Claim 8
Dependent Claim
32 words
The method of claim 4, wherein the storage registers are successively addressed using an address incremented by 1 after each computing clock, and said address is reset if the remainder is zero.
Claim 9
Dependent Claim
155 words
A circuit arrangement (floating point adder stage) comprising a first input (B input) and a second input (A input) for the floating point addition of two summands (A, B) consisting of sign, exponent and mantissa; a shift means (8) and a mantissa adder (10) for aligning and adding the two mantissas, and a post normalization means (12, 14, 16) for eliminating leading zeros of the mantissa sum, said adder stage also comprising a further shift means (18) for truncating the right-shifted remainder of the smaller summand (B"), which has not been added to the larger summand (A"), a post normalization means (20, 22, 24, 26) for normalizing the truncated remainder and for emitting a "remainder zero" signal, an output (S) for the sum of the two summands and an output (R) for the remainder with the same floating point notation at the output (S, R) of the circuit arrangement and at its input (A, B).
Claim 10
Dependent Claim
155 words
A circuit arrangement (floating point adder stage) comprising a first input (B input) and a second input (A input) for the floating point addition of two summands (A, B) consisting of sign, exponent and mantissa; a shift means (8) and a mantissa adder (10) for aligning and adding the two mantissas, and a post normalization means (12, 14, 16) for eliminating leading zeros of the mantissa sum, said adder stage also comprising a further shift means (18) for truncating the right-shifted remainder of the smaller summand (B"), which has not been added to the larger summand (A"), a post normalization means (20, 22, 24, 26) for normalizing the truncated remainder and for emitting a "remainder zero" signal, an output (S) for the sum of the two summands and an output (R) for the remainder with the same floating point notation at the output (S, R) of the circuit arrangement and at its input (A, B).
Claim 11
Dependent Claim
98 words
The circuit arrangement of claim 6, wherein the post normalization means comprises a detector (20) for determining the number of leading zeros of the remainder truncated by the further shift means (18) and for generating a shift value signal corresponding to the number of leading zeros, a third shift means (22) controlled by the detector for eliminating leading zeros, an adder (26) controlled by the detector for generating the exponent of the remainder, said detector also receiving the difference between the exponents of the two summand (A, B) as well as the exponent of the smaller summand (B").
Claim 12
Dependent Claim
63 words
The circuit arrangement of claim 7, further comprising a gate circuit (24) for applying the least-significant digit of the sum of mantissas of the two summands emitted by the mantissa adder (10), the gate circuit being controlled by a carry signal of the mantissa adder and its output being connected to an input of the detector (20) and the third shift means (22).
Claim 13
Dependent Claim
19 words
The circuit arrangement of claim 7, wherein the detector (20) has an output for generating the "remainder zero" signal.
Claim 14
Dependent Claim
72 words
A circuit arrangement comprising a plurality of adder stages of claim 6 arranged in pipeline form, wherein the output R for the remainder of a stage is connected to the first input (B) of a subsequent stage, and the output (S) for the sum is connected to the second input of the stage in which it was computed such that newly resulting remainders are in each case transmitted to the next stage.
Claim 15
Dependent Claim
28 words
The circuit arrangement of claim 10, wherein, for buffering, the output of each stage is provided with one register for the sum and one register for the remainder.
Claim 16
Dependent Claim
124 words
The circuit arrangement with an adder stage according to claim 6, further comprising the following features: the output (R) for the remainder is connected to a remainder register (32) and its output to a first input of a switch (36), a second input of the switch is connected to an input (SMD) for applying summands to be successively summed, the output of the switch is connected to the first input (B) of the adder stage, the input of a register stack (34) for buffering sum signals is connected to the output (S) for the sum, and the output of the stack is connected to the second input (A) of the adder stage, the stack is addressed by an address means (38) with successive addresses.
Claim 17
Dependent Claim
28 words
The circuit arrangement of claim 12, wherein said address means (38) comprises a reset input (RST) which is connected to the "remainder zero" output of the adder stage.
References Cited (Backward Citations)
US4644490A
Cited
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 1992-09-11 | 1993-03-13 | 1993-09-12 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 1996-09-11 | 1997-03-13 | 1997-09-12 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2000-09-11 | 2001-03-13 | 2001-09-12 | $7700.00 | Window Closed / Lapsed |