Samsung Electronics Co., Ltd.

Advanced Micro Devices Corporation

Other

University of Texas System

Eric C Quinnell

Earl E Swartzlander, Jr

Carl E Lemonds

Fuzhou Zou

Monika Tkaczyk

Daniel William Bailey

Vikas Sinha

Tarun Nakra

James David Dundas

Atchyuth Gorti

Madhu Saravana Sibi Govindan

Akhil Arunkumar

Ryan J Hensley

Aaron S Rogers

Christopher S Thomas

Kevin C Heuer

Aditya Jagirdar

Jyotsna Kartha

Bikash Kumar Agarwal

Reshma C Jumani

According to one general aspect, an apparatus may include an instruction fetch unit circuit configured to retrieve instructions from a memory. The apparatus may include an instruction decode unit configured to convert instructions into one or more micro-operations that are provided to an execution unit circuit. The apparatus may also include a micro-operation cache configured to store micro-operations. The apparatus may further include a branch prediction circuit configured to: determine when a kernel of instructions is repeating, store at least a portion of the kernel within the micro-operation cache, and provide the stored portion of the kernel to the execution unit circuit without the further aid of the instruction decode unit circuit.

Micro-operation cache using predictive allocation

The present disclosure includes a mispredict recovery apparatus, which may comprise an instruction execution unit, a branch predictor, and a misprediction recovery unit (MRU). The MRU may provide discrete cycle predictions after a misprediction redirect from the instruction execution unit. The MRU may include a branch confidence filter to generate prediction confidence information for predicted branches. The MRU may include a tag content-addressable memory (CAM). The tag CAM may store frequently mispredicting low-confidence branches, probe the misprediction redirect, and obtain the prediction confidence information from the branch confidence filter. The MRU may include a mispredict recovery buffer (MRB) to store an alternate path for frequently mispredicting low-confidence branches present in the tag CAM without storing the instructions themselves. Also disclosed is a method for recovering from mispredicts associated with the instruction fetch pipeline.

Mispredict recovery apparatus and method for branch and fetch pipelines

A method and apparatus are provided. The apparatus includes a plurality of central processing units, a plurality of core input/output units, a plurality of last level cache memory banks, an interconnect network comprising multiple instantiations of dedicated data channels, wherein each dedicated data channel is dedicated to a memory transaction type, each instantiation of dedicated data channels includes arbitration multiplexors, and each dedicated data channel operates independently of other data channels.

Non-uniform bus (NUB) interconnect protocol for tiled last level caches

According to one general aspect, an apparatus may include a monolithic shifter configured to receive a plurality of bytes of data, and, for each byte of data, a number of bits to shift the respective byte of data, wherein the number of bits for each byte of data need not be the same as for any other byte of data. The monolithic shifter may be configured to shift each byte of data by the respective number of bits. The apparatus may include a mask generator configured to compute a mask for each byte of data, wherein each mask indicates which bits, if any, are to be prevented from being polluted by a neighboring shifted byte of data. The apparatus may include a masking circuit configured to combine the shifted byte of data with a respective mask to create an unpolluted shifted byte of data.

Bit-masked variable-precision barrel shifter

Embodiments of the present system and method provide cache replacement in a victim exclusive cache using a snoop filter where replacement information is not lost during a re-reference back to the CPU. Replacement information is stored in a snoop filter, meaning that historical access data may be fully preserved and allows for more flexibility in the LLC re-insertion points, without additional bits stored in a L2 cache. The present system and method further include snoop filter replacement technique. The present system and method passes replacement information between a snoop filter and a victim exclusive cache (e.g., LLC) when transactions move cachelines to and from a master CPU. This maintains and advances existing replacement information for a cacheline that is removed from the victim exclusive cache on a read, as well as intelligently replaces and ages cachelines in the snoop filter.

Snoop filter with stored replacement information, method for same, and system including victim exclusive cache and snoop filter shared replacement policies

According to one general aspect, an apparatus may include a floating-point addition unit that includes a far path circuit, a close path circuit, and a final result selector circuit. The far path circuit may be configured to compute a far path result based upon either the addition or the subtraction of the two floating-point numbers regardless of whether the operands or the result include normal or denormal numbers. The close path circuit may be configured to compute a close path result based upon the subtraction of the two floating-point operands regardless of whether the operands or the result include normal or denormal numbers. The final result selector circuit may be configured to select between the far path result and the close path result based, at least in part, upon an amount of difference in the exponent portions of the two floating-point operands.

High performance floating-point adder with full in-line denormal/subnormal support

According to one general aspect, an apparatus may include a floating-point addition unit that includes a far path circuit, a close path circuit, and a final result selector circuit. The far path circuit may be configured to compute a far path result based upon either the addition or the subtraction of the two floating point numbers regardless of whether the operands or the result include normal or denormal numbers. The close path circuit may be configured to compute a close path result based upon the subtraction of the two floating point operands regardless of whether the operands or the result include normal or denormal numbers. The final result selector circuit may be configured to select between the far path result and the close path result based, at least in part, upon an amount of difference in the exponent portions of the two floating point operands.

According to one general aspect, an apparatus may include a memory, a normalization engine, a lookup table, and an adder. The memory may be configured to store a floating-point number formatted in a floating-point format. The normalization engine may be configured to normalize at least a portion of the floating-point number to create a normalized number. The lookup table may be configured to generate an injection constant based upon a predefined set of rounding values specifically for converting a floating-point number to an integer number. The adder may be configured to create an integer result by adding the normalized number and the injection constant.

Rounding injection scheme for floating-point to integer conversion

We report methods relating to scan warmup of integrated circuit devices. One such method may comprise loading a scan test stimulus to and unloading a scan test response from a first set of logic elements of an integrated circuit device at a scan clock first frequency equal to a test clock frequency; adjusting the scan clock from the first frequency to a second frequency by a scan warmup unit, wherein the scan clock second frequency is equal to a system clock frequency; and capturing the scan test response by a shift logic at the scan clock second frequency. We also report processors containing components configured to implement the method, and fabrication of such processors. The methods and their implementation may reduce di/dt events otherwise commonly occurring when testing logic elements of integrated circuit devices.

Scan warmup scheme for mitigating di/dt during scan test

Methods relating to distribution of a clock signal to logic devices of an integrated circuit. The method includes controlling, by a logic element, the distribution of a clock signal by a clock gater and distributing the clock signal by the clock gater to at least one first logic device, wherein the logic element allows the first clock gater to distribute the clock signal only when at least one first logic device requires the clock signal. An integrated circuit configured to perform such a method. Fabrication of such an integrated circuit.

Coarse gating of clock tree elements

A semiconductor device includes a primary voltage rail, a secondary voltage rail, a plurality of transistors coupled between the primary and secondary voltage rails, and control logic operable to enable a first subset of the plurality of transistors to couple the primary voltage rail to the secondary voltage rail. During a steady state condition, the first subset comprises less than all of the plurality of transistors.

Variable-width power gating module

A bridge fused multiply-adder is disclosed. The fused multiply-adder is for the single instruction execution of (A&#xd7;B)+C. The bridge fused multiply-add unit adds this functionality to existing floating-point co-processor units by including a fused multiply-add hardware 'bridge' between an existing floating-point adder and a floating-point multiplier unit. This fused multiply-add functionality is added to existing two-operand architecture designs without degrading the performance or parallel pipe execution of floating-point adder and floating-point multiplier instructions.

Bridge fused multiply-adder circuit

A three-path floating-point fused multiply-adder is disclosed. The fused multiply-adder is for the single instruction execution of (A&#xd7;B)+C. The three-path fused multiply-adder is based on a dual-path adder and reduces latency significantly by operating on case data in parallel and by reducing component bit size. The fused multiply-adder is a common serial fused multiply-adder that reuses floating-point adder (FPA) and floating-point multiplier (FPM) hardware, allowing single adds, single multiplies, and fused multiply-adds to execute at maximum speed.

Growing community of inventors

Austin, TX, United States of America

Eric C Quinnell