International Business Machines Corporation

Jochen Preiss

Silvia Melitta Mueller

Maarten Jakob Boersma

Tobias Gemmeke

Michael Klaus Kroener

Harry S Barowski

Petra Leber

Kerstin Claudia Schelm

Michael Klein

Jens Leenstra

Tim Niggemeier

Juergen Haess

Son Dao Trong

J Adam Butts

Christian Jacobi

Cedric Lichtenau

Stephen Victor Kosonocky

Markus Kaltenbach

Thomas Pflueger

Klaus Michael Kroener

Nicolas Maeding

Holger Wetter

Son Trong Dao

Jens Kuenzer

Christoph Jaeschke

Maarten Jokob Boersma

K Michael Kroener

A binary logic unit to apply any Boolean operation on two input signals (v, v) is described, wherein any Boolean operation to be applied on the input signals (v, v) is defined by a particular combination of well defined control signals (ctl, ctl, ctl, ctl), wherein the input signals (v, v) are used to select a control signal (ctl, ctl, ctl, ctl) as an output signal (v) of the binary logic unit representing the result of a particular Boolean operation applied on the two input signals (v, v). Furthermore a method to operate such a binary logic unit is described.

Binary logic unit and method to operate a binary logic unit

A method for converting a signed fixed point number into a floating point number that includes reading an input number corresponding to a signed fixed point number to be converted, determining whether the input number is less than zero, setting a sign bit based upon whether the input number is less than zero or greater than or equal to zero, computing a first intermediate result by exclusive-ORing the input number with the sign bit, computing leading zeros of the first intermediate result, padding the first intermediate result based upon the sign bit, computing a second intermediate result by shifting the padded first intermediate result to the left by the leading zeros, computing an exponent portion and a fraction portion, conditionally incrementing the fraction portion based on the sign bit, correcting the exponent portion and the fraction portion if the incremented fraction portion overflows, and returning the floating point number.

Reuse of rounder for fixed conversion of log instructions

A method and system for reducing power consumption when processing mathematical operations. Power may be reduced in processor hardware devices that receive one or more operands from an execution unit that executes instructions. A circuit detects when at least one operand of multiple operands is a zero operand, prior to the operand being forwarded to an execution component for completing a mathematical operation. When at least one operand is a zero operand or at least one operand is 'unordered', a flag is set that triggers a gating of a clock signal. The gating of the clock signal disables one or more processing stages and/or devices, which perform the mathematical operation. Disabling the stages and/or devices enables computing the correct result of the mathematical operation on a reduced data path. When a device(s) is disabled, the device may be powered off until the device is again required by subsequent operations.

Zero indication forwarding for floating point unit power reduction

A distributed residue checking apparatus for a floating point unit having a plurality of functional elements performing floating-point operations on a plurality of operands. The distributed residue checking apparatus includes a plurality of residue generators which generate residue values for the operands and the functional elements, and a plurality of residue checking units distributed throughout the floating point unit. Each residue checking unit receives a first residue value and a second residue value from respective residue generators and compares the first residue value to the second residue value to determine whether an error has occurred in a floating-point operation performed by a respective functional element.

Distributed residue-checking of a floating point unit

A floating point processor unit executes a floating point compare instruction with two operands of the same or different precision by comparing the two operands in integer format, which speeds up the execution of the floating point compare instruction significantly. The floating point processor now executes the floating point compare instruction at least twice as fast or faster (e.g., two clock cycles instead of five clock cycles in the prior art) for nearly most operand cases (e.g., 99% of all cases). Only the rare corner cases require additional operations on one of the operands and thus require additional cycles of execution time because the integer compare operation will not work for these corner cases. This is due to the fact that one operand is a single precision subnormal number in an unnormalized representation (i.e., has two representations) and the other operand is in the SP subnormal range such that the integer compare operation will fail.

Fast floating point compare with slower backup for corner cases

A method for operand width reduction is described, wherein two N-bit input operands (A, B) of a bit width of N are processed and two M-bit output operands (A', B&#x2032;) of a reduced bit width of M are generated in a way, that a post-processing comprising an M-bit adder function followed by saturation to M bits performed on said two M-bit output operands (A&#x2032;, B&#x2032;) provides an M-bit result equal to an M-bit result of an N-bit modulo adder function of the two N-bit input operands (A, B), followed by a saturation to M bits. Further an electronic computing circuit () is described performing said method. Additionally a computer system comprising such an electronic computing circuit is described.

Electronic computing circuit for operand width reduction for a modulo adder followed by saturation concurrent message processing

The forcing of the result or output of a rounder portion of a floating point processor occurs only in a fraction non-increment data path within the rounder and not in the fraction increment data path within the rounder. The fraction forcing is active on a corner case such as a disabled overflow exception. A disabled overflow exception may be detected by inspecting the normalized exponent. If a disabled overflow exception is detected, the round mode is selected to execute only in the non-increment data path thereby preventing the fraction increment data path from being selected.

Efficient forcing of corner cases in a floating point rounder

A system and a method for storing numbers in a register file are provided. The system and the method store single precision numbers in double precision format in a register file that is shared between floating point computational units and computational units not supporting floating point numbers.

System and method for storing numbers in first and second formats in a register file

A shifter that includes a plurality of shift stages positioned within the shifter, and receiving and shifting input data to generate a shifted result, and a detection circuit coupled at an input of a final shift stage of the plurality of shifters, in a final stage within the shifter. The detection circuit receives a partially shifted vector at the input of the final shift stage along with a predetermined shift amount, and performing an all-one or all-zero detection operation using a portion of the partially shifted vector and the predetermined shift amount, in parallel, to a shifting operation performed by the final shift stage to generate the shifted result.

Shifter with all-one and all-zero detection using a portion of partially shifted vector and shift amount in parallel to generated shifted result

In a binary floating point processor, the exponents of each of the various types of operands are recoded into an internal format, by biasing the exponents with the minimum exponent value of the result precision ('Emin'), i.e., the recoded value of the exponent is the represented value of the exponent minus Emin. Emin depends only on the result precision of the instruction that is currently being executed in the binary floating point processor. The exponent computations are then performed in this new format. The underflow check for all result precisions is a check against zero and overflow checks are performed against a positive number that depends on the result precision. The exponent values are in a 2's complement representation, so the underflow check simply becomes a check of the sign bit.

Supporting multiple formats in a floating point processor

A method, system and computer program product for reducing power consumption when processing mathematical operations. Power may be reduced in processor hardware devices that receive one or more operands from an execution unit that executes instructions. A circuit detects when at least one operand of multiple operands is a zero operand, prior to the operand being forwarded to an execution component for completing a mathematical operation. When at least one operand is a zero operand or at least one operand is 'unordered', a flag is set that triggers a gating of a clock signal. The gating of the clock signal disables one or more processing stages and/or devices, which perform the mathematical operation. Disabling the stages and/or devices enables computing the correct result of the mathematical operation on a reduced data path. When a device(s) is disabled, the device may be powered off until the device is again required by subsequent operations.

A floating point processor unit includes a shift amount calculation circuit within a normalizer portion of the floating point unit, wherein the shift amount calculation circuit is utilized to compute the normalizer shift amount for a log estimate instruction that runs as a pipelinable instruction.

Normalizer shift prediction for log estimate instructions

A 3-stack floorplan for a floating point unit includes: an aligner located in the center of the floating point unit; a frontend located directly above the aligner; a multiplier located directly below the frontend and next to the aligner; an adder located directly next to the multiplier and directly below the aligner; a normalizer located directly above the adder; and a rounder located directly above the normalizer.

3-stack floorplan for floating point unit

A semiconductor chip subdivided into power domains, at least one of the power domains is separately activated or deactivated and at least a part of the scannable storage elements are interconnected to one or more scan chains. At least one scan chain is serially subdivided into scan chain portions and the scan chain portion is arranged within one of the power domains. For at least one scan chain portion a bypass line is provided for passing by scan data and at least one select unit is provided for selecting between the bypass line and the corresponding scan chain portion in dependence of the activated or deactivated state of the corresponding power domains.

Growing community of inventors

Boeblingen, Germany

Jochen Preiss