National Semiconductor Corporation

Via-Cyrix, Inc.

Daniel W Green

Atul Dhablania

Jeffrey A Lohman

Bang Nguyen

Darren Bensley

Douglas R Beard, Sr

There is disclosed an operand queue for use in a floating point unit. The floating point unit comprises floating point processing units for executing floating point instructions that write operands to an external memory and for executing floating point instructions that read operands from the external memory. The floating point also comprises an operand queue for storing a plurality of operands associated with one or more operations being processed in the floating point unit. The operand queue stores a first operand being written to an external memory by a floating point write instruction executed by a first one of the plurality of floating point processing units and supplies the first operand to a floating point read instruction executed by a second one of the plurality of floating point processing units subsequent to the execution of the floating point write instruction.

Operand queue for use in a floating point unit to reduce read-after-write latency and method operation

A floating point unit (FPU) for processing denormal numbers in floating point notation, a method of processing such numbers in an FPU and a computer system employing the FPU or the method. In one embodiment, the FPU includes: (1) a load unit that receives a denormal number having an exponent portion of a standard length from a source without the FPU and transforms the denormal number into a normalized number having an exponent portion of an expanded length greater than the standard length, (2) a floating point execution core, coupled to the load unit, that processes the normalized number at least once to yield a processed normalized number, the expanded length of the exponent portion allowing the processed normalized number to remain normal during processing thereof and (3) a store unit, coupled to the floating point execution core, that receives the processed normalized number and transforms the processed normalized number back into a denormal number having an exponent portion of the standard length.

Formatting denormal numbers for processing in a pipelined floating point unit

There is disclosed a multiplier circuit for use in a data processor. The multiplier circuit comprises a partial products generating circuit that receives a multiplicand value and a multiplier value and generates a group of partial products. The multiplier circuit also comprises a split array for adding the partial products. A first summation array comprises a first group of adders that sum the even partial products to produce an even summation value. A second summation array comprises a second group of adders that sum the odd partial products to produce an odd summation value. The even and odd summation values are then summed to produce the output of the multiplier.

Split multiplier array and method of operation

A floating point unit includes floating point processing units for executing floating point instructions that write operands to an external memory and for executing floating point instructions that read operands from the external memory. The floating point also includes an operand queue for storing a plurality of operands associated with one or more operations being processed in the floating point unit. The operand queue stores a first operand written by a floating point write instruction executed by a first one of the plurality of floating point processing units and supplies the first operand to a floating point read instruction executed by a second one of the plurality of floating point processing units when the first operand is committed or virtually committed.

Operand queue for use in a floating point unit to reduce read-after-write latency and method of operation

For use in a processor having a floating point unit (FPU) capable of managing denormalized numbers in floating point notation, logic circuitry for, and a method of adding or subtracting two floating point numbers. In one embodiment, the logic circuitry includes: (1) an adder that receives the two floating point numbers and, based on a received instruction, adds or subtracts the two floating point numbers to yield a denormal sum or difference thereof, (2) a leading bit predictor that receives the two floating point numbers and performs logic operations thereon to yield predictive shift data denoting an extent to which the denormal sum or difference is required to be shifted to normalize the denormal sum or difference, the predictive shift data subject to being erroneous and (3) predictor corrector logic that receives the two floating point numbers and performs logic operations thereon to yield shift compensation data denoting an extent to which the predictive shift is erroneous. The denormal sum or difference, predictive shift data and shift compensation data are providable to a shifter to allow the denormal sum or difference to be normalized.

Leading bit prediction with in-parallel correction

For use in a processor having a floating point unit (FPU) capable of managing denormalized numbers in floating point notation, logic circuitry for, and a method of, generating least significant (L), round (R) and sticky (S) bits for a denormalized number. In one embodiment, the system includes: (1) a bit mask decoder that produces a bit mask that is a function of a precision of the denormalized number and an extent to which the denormalized number is denormal and (2) combinatorial logic, coupled to the bit mask decoder, that performs logical operations with respect to a fraction portion of the denormalized number, the bit mask and at least one shifted version of the bit mask to yield the L, R and S bits.

Rounding denormalized numbers in a pipelined floating point unit without pipeline stalls

A multiplier circuit for use in a data processor. The multiplier circuit contains a partial products generating circuit that receives a multiplicand value and a multiplier value and generates a group of partial products. The multiplier circuit also contains a split array for adding the partial products. A first summation array has a first group of adders that sum the even partial products to produce an even summation value. A second summation array has a second group of adders that sum the odd partial products to produce an odd summation value. The even and odd summation values are then summed to produce the output of the multiplier.

There is disclosed, for use in a processing device having an N-way set associative data array (such as an L1 cache), a built-in self-test (BIST) circuit for testing the validity of storage locations in the data array. The BIST circuit comprises 1) a memory capable of storing a test program executable by the processing device, wherein the test program is capable of testing the validity of the storage locations in the data array; and 2) a controller capable of copying the test program from the memory into first selected storage locations in a first way in the data array, wherein the processing device executes the copied test program stored in the first selected storage locations subsequent to the copying to thereby test the validity of second selected storage locations in at least one of the N ways.

Built-in self-test circuit and method for validating an associative data array

There is disclosed, for use in an x86-compatible processor, a translation look-aside buffer (TLB) that stores region configuration bits (or attribute bits) associated with each physical address stored in the TLB and that makes the region configuration bits available at the same time that the physical address is generated/translated by the TLB. The TLB comprises: 1) a tag array capable of storing an untranslated address in one of N tag entries in the tag array; 2) a data array capable of storing a translated physical address corresponding to the untranslated address in one of N data entries in the data array; and 3) a region configuration array capable of storing region configuration bits associated with the translated physical address in one of N region configuration entries in the region configuration array.

Translation look-aside buffer for storing region configuration bits and method of operation

There is disclosed, for use in an x86-compatible processor capable of operating in real mode and paging mode and having a physically-addressable cache, an address translation device for providing physical addresses to the cache. The address translation device comprises: 1) a tag array for storing received untranslated addresses in selected ones of N tag entries in the tag array during real mode operations and paging mode operations; and 2) a data array for storing translated physical addresses corresponding to the untranslated addresses in selected ones of N data entries in the data array, wherein the untranslated addresses stored in the tag array during real mode operations are physical addresses equal to the corresponding translated physical addresses stored in the data array.

Real mode translation look-aside buffer and method of operation

For use in an x-86 processor having a physically-addressable cache and an associated translation look-aside buffer (primary TLB) that stores corresponding logical and physical addresses for addressing the cache, a circuit for increasing a retrieval speed of a line from the cache. In one embodiment, the circuit comprises: 1) a TLB slice, located proximate the cache, that contains copies of portions of the physical addresses stored in the primary TLB and returns one of the portions as a function of a portion of a logical address supplied thereto; and 2) a decoder, coupled to the TLB slice, that decodes the one of the portions to yield multiplexer selection signals, the TLB slice and the decoder cooperating to increase the retrieval speed by avoiding logic circuitry interposed between the primary TLB and the cache.

Translation look-aside buffer slice circuit and method of operation

A cache has programmable, finely ganular, locked-down regions within a way or way(s) so that the contents of the locked-down regions are not evicted. The finely granular locked-down regions need not be contiguous and are programmed as either 'locked-valid' or 'locked-invalid' to provide general purpose memory that is local and private to the processor or for masking defected cache lines or portions thereof. Finely granular, programmable spatial regions of the cache that are locked-down are preferably, although not exclusively, programmed through two additional states to the standard MESI (Modified, Exclusive, Shared, Invalid) protocol for multipurpose cache coherency.

Growing community of inventors

McKinney, TX, United States of America

Daniel W Green