International Business Machines Corporation

Microsoft Technology Licensing, LLC

Globalfoundries Inc.

Brian R Konigsburg

David S Levitan

Minsik Cho

Hung Qui Le

Jeonghee Shin

Ruchir Puri

Steven Wayne White

John Stephen Muhich

Haoxing Ren

Indira Nair

Richard James Eickemeyer

Rajesh Ramkrishna Bordawekar

Samuel Jonathan Thomas

Daniel Brand

John Edward Derrick

Tejas S Karkhanis

Wolfram M Sauer

Jose Eduardo Moreira

Lee Evan Eisen

Alexander Erik Mericas

Mauricio J Serrano

Venkat Rajeev Indukuru

Larry Edward Thatcher

David Mui

John Wesley Ward, Iii

Benjamin W Stolt

William Robert Reohr

Douglas R G Logan

Dung Q Nguyen

Balaram Sinharoy

James Allan Kahle

David Arnold Luick

George McNeil Lattimore

Paul Joseph Jordan

Kattamuri Ekanadham

Brian William Curran

Brian Robert Mestan

Bruce Joseph Ronchetti

Richard William Doing

Thomas Albert Petersen

Vincent Kulandaisamy

Kevin Neal Magill

Il Park

Kevin Arthur Chiarot

Alexander Edward Okpisz

Robert William Hay

Dave Stephen Levitan

Douglas Robert Gordan Logan

Predicting indirect branch instructions may comprise predicting a target address for a fetched branch instruction. Accuracy of the target address may be tracked. The fetched branch instruction may be flagged as a problematic branch instruction based on the tracking. A pattern cache may be trained for predicting a more accurate target address for the fetched branch instruction, and the next time the fetched branch instruction is again fetched, a target address may be predicted from the pattern cache.

Predicting indirect branches using problem branch filtering and pattern cache

An information processing system, computer readable storage medium, and method for accelerated radix sort processing of data elements in an array in memory. The information processing system stores an array of data elements in a buffer memory in an application specific integrated circuit radix sort accelerator. The array has a head end and a tail end. The system radix sort processing, with a head processor, data elements starting at the head end of the array and progressively advancing radix sort processing data elements toward the tail end of the array. The system radix sort processing, with a tail processor, data elements starting at the tail end of the array and progressively advancing radix sort processing data elements toward the head end of the array, the tail processor radix sort processing data elements in the array contemporaneously with the head processor radix sort processing data elements in the array.

Radix sort acceleration using custom asic

A low level virtual machine (LLVM)-based system C compiler for architecture synthesis is provided. In one aspect, a method for translating a system C model to hardware description language (HDL) is provided. The method includes the steps of: generating a hardware connection model (HCM) from the system C model, wherein the HCM defines modules and interconnects in a hardware system; parsing the system C model into a LLVM intermediate representation (IR); converting the LLVM IR to a system LLVM IR which records correspondence information between the LLVM IR and the HCM; and generating the HDL based on direct mapping of processes from the system LLVM IR and the HCM.

LLVM-based system C compiler for architecture synthesis

Radix sort acceleration using custom ASIC

Systems and methods for sorting a data set stored on an external device. A plurality of smaller radix sizes are determined, based on a first radix size and performance characteristics of an external data storage device, whose sizes add up to a first radix size for an in-place radix sort. The smaller radix sizes reduce a total time to perform the in-place radix sort. Each level of a multiple level in-place radix sort is performed with the smaller radix sizes. Each level of the sort includes dividing the data set into N buckets; dividing the buffer into N buckets; and iteratively loading a respective segment in each bucket of the data set into a respective bucket of the buffer, performing an in-place radix sort on the data in the buffer, and returning sorted buffer data to the data set on the external storage device.

Adaptive radix external in-place radix sort

A graph is constructed, having a plurality of nodes representing a plurality of logical operations and a plurality of edges connecting those of the plurality of nodes which do not conflict. A weight, including a width difference between end nodes of each of the edges, is assigned to each edge. Weighted cliques are enumerated, each including at least two of the nodes. Each of the weighted cliques is replaced with a single one of the logical operations and a multiplexer, to obtain a plurality of multiplexer-operation groups, such that each logical operation in one of the multiplexer-operation groups can be shared within a same clock cycle of a digital electronic integrated circuit.

In-cycle resource sharing for high-level synthesis of microprocessors

A memory system including a content addressable memory having an array of content addressable memory elements including a plurality of rows of content addressable memory elements and a plurality of columns of content addressable memory elements is provided. Each of the content addressable memory elements further includes a first superconducting quantum interference device (SQUID) and a second superconducting quantum interference device (SQUID), where an input bit to each of the content addressable memory elements is compared with: (1) a first state of the first SQUID and (2) a second state of the second SQUID to generate an output signal. The memory system further includes a Josephson magnetic random access memory (JMRAM), coupled to the content addressable memory.

Memory system with a content addressable superconducting memory

A technique for mapping logical threads to physical threads of a simultaneous multithreading (SMT) data processing system includes mapping one or more logical threads to one or more physical threads based on a selected SMT mode for a processor. In this case, respective resources for each of the one or more physical threads are predefined based on the SMT mode and an identifier of the one or more physical threads. The one or more physical threads are then executed on the processor utilizing the respective resources.

Techniques for mapping logical threads to physical threads in a simultaneous multithreading data processing system

In a computing system running an environment for designing operation of circuitry, at least the following are performed for providing simulations and evaluations of one or more user-defined modules of circuitry including one or more pipeline stages in a pipeline. A model of the pipeline is automatically generated by using a pipeline block diagram, where the model is generated in a high-level modeling language able to perform simulations of circuitry with the pipeline. An interface is automatically generated between the one or more user-defined modules and the generated model of the pipeline, the interface including access to the pipeline, wherein the pipeline access allows the one or more user-defined modules to interact indirectly. Evaluation is performed of the one or more user-defined modules using the automatically generated model of the pipeline and the automatically generated interface. Methods, apparatus, and computer program products are disclosed.

Automating a microarchitecture design exploration environment

Tracking global history vector in high performance out of order superscalar processors, in one aspect, may comprise providing a shift register storing global history vector that stores branch predictions and outcomes. A counter is maintained to determine a number of bits to shift the shift register to recover branch history. In another aspect, the global history vector may be implemented with a circular buffer structure. Youngest and oldest pointers to the circular buffer are maintained and used in recovery.

Tracking long GHV in high performance out-of-order superscalar processors

In a computing system running an environment for designing operation of circuity, at least the following are performed for providing simulations and evaluations of one or more user-defined modules of circuitry including one or more pipeline stages in a pipeline. A model of the pipeline is automatically generated by using a pipeline block diagram, where the model is generated in a high-level modeling language able to perform simulations of circuitry with the pipeline. An interface is automatically generated between the one or more user-defined modules and the generated model of the pipeline, the interface including a set of access methods to the pipeline. Evaluation is performed of the one or more user-defined modules using the automatically generated model of the pipeline and the automatically generated interface. Methods, apparatus, and computer program products are disclosed.

During a pipeline stall in a processor, until a next to complete instruction group completes, a monitoring unit receives, from a completion unit of a processor, a next to finish indicator indicating the finish of an oldest previously unfinished instruction from among a plurality of instructions of a next to complete instruction group. The monitoring unit receives, from functional units of the processor, finish reports including completion reasons for separate instructions. The monitoring unit determines at least one stall reason from among multiple stall reasons for the oldest instruction from a selection of completion reasons from a selection of finish reports aligned with the next to finish indicator from among the finish reports. Once the monitoring unit receives a complete indicator from the completion unit, indicating the completion of the next to complete instruction group, the monitoring unit stores each determined stall reason aligned with each next to finish indicator in memory.

Determining each stall reason for each stalled instruction within a group of instructions during a pipeline stall

A technique for branch target prediction includes storing, based on an instruction fetch address for a group of fetched instructions, first predicted targets for first indirect branch instructions in respective entries of a local count cache. Second predicted targets for second indirect branch instructions are stored in respective entries of a global count cache, based on the instruction fetch address and a global history vector for the instruction fetch address. One of the local count cache and the global count cache is selected to provide a selected predicted target for an indirect branch instruction in the group of fetched instructions.

Techniques for selecting a predicted indirect branch address from global and local caches

Growing community of inventors

Austin, TX, United States of America

Brian R Konigsburg