Digital Equipment Corporation

Intel Corporation

Hewlett-packard Development Company, L.p.

Ati International Srl

Intel Corproation

Michael C Adler

Joel S Emer

David J Sager

Pratik M Marolia

Francis X McKeen

Paul Geoffrey Lowney

Aamer Jaleel

Nagabhushan Chitlur

Steven O Hobbs

Robert P Nix

Neal C Crago

P Geoffrey Lowney

Utkarsh Y Kakaiya

Joshua David Fender

Henry M Mitchel

Joseph J Grecco

Sundar Nadathur

Aaron J Grier

Angshuman Parashar

Michael I Pellauer

Bushra Ahsan

Simon C Steely, Jr

Stephen C Purcell

John S Yates, Jr

Kent D Glossop

Paul Henry Hohensee

Kermin Elliott Fleming, Jr

Stephen S Chang

David Reese

Samantika S Sury

Robert S Cohn

Chiachen Chou

P Geoffrey Lowny

A system comprising an accelerator circuit comprising an accelerator function unit to implement a first function, and one or more device feature header (DFH) circuits to provide attributes associated with the accelerator function unit, and a processor to retrieve the attributes of the accelerator function unit by traversing a device feature list (DFL) referencing the one or more DFH circuits, execute, based on the attributes, an application encoding the first function to cause the accelerator function unit to perform the first function.

Modular accelerator function unit (AFU) design, discovery, and reuse

There is disclosed in one example an accelerator apparatus, including: a programmable region capable of being programmed to provide an accelerator function unit (AFU); and a platform interface layer (PIL) to communicatively couple to the AFU via an intra-accelerator protocol, and to provide multiplexed communication with a processor via a plurality of platform interconnect interfaces, wherein the PIL is to provide abstracted communication services for the AFU to communicate with the processor.

Platform interface layer and protocol for accelerators

A technology for implementing a method for distributed memory operations. A method of the disclosure includes obtaining distributed channel information for an algorithm to be executed by a plurality of spatially distributed processing elements. For each distributed channel in the distributed channel information, the method further associates one or more of the plurality of spatially distributed processing elements with the distributed channel based on the algorithm.

Executing distributed memory operations using processing elements connected by distributed channels

Systems, methods, and apparatuses relating to a configurable spatial accelerator are described. In one embodiment, a processor includes a plurality of processing elements; and an interconnect network between the plurality of processing elements to receive an input of a dataflow graph comprising a plurality of nodes, wherein the dataflow graph is to be overlaid into the interconnect network and the plurality of processing elements with each node represented as a dataflow operator in the plurality of processing elements, and the plurality of processing elements is to perform an operation when an incoming operand set arrives at the plurality of processing elements. The processor also includes a streamer element to prefetch the incoming operand set from two or more levels of a memory system.

Processors, methods, and systems for a configurable spatial accelerator with memory system performance, power reduction, and atomics support features

A processing device for executing distributed memory operations using spatial processing units (SPU) connected by distributed channels is disclosed. A distributed channel may or may not be associated with memory operations, such as load operations or store operations. Distributed channel information is obtained for an algorithm to be executed by a group of spatially distributed processing elements. The group of spatially distributed processing elements can be connected to a shared memory controller. For each distributed channel in the distributed channel information, one or more of the group of spatially distributed processing elements may be associated with the distributed channel based on the algorithm. By associating the spatially distributed processing elements to a distributed channel, the functionality of the processing element can vary depending on the algorithm mapped onto the SPU.

A computer having an instruction pipeline and profile circuitry. The profile circuitry detects and records, without compiler assistance for execution profiling, profile information describing a sequence of events occurring in the instruction pipeline. The sequence includes every event occurring during a profiled execution interval that matches time-independent selection criteria of events to be profiled. The recording continues until a predetermined stop condition is reached. The profile circuitry detects the occurrence of a predetermined condition, after a non-profiled interval of execution, and then commences the profiled execution interval.

Profiling execution of a sequence of events occuring during a profiled execution interval that matches time-independent selection criteria of events to be profiled

A mechanism for executing computer instructions in parallel includes a compiler for generating and grouping instructions into a plurality of sets of instructions to be executed in parallel, each set having a unique identification. A computer system having a real state and a speculative state executes the sets in parallel, the computer system executing a particular set of instructions in the speculative state if the instructions of the particular set have dependencies which can not be resolved until the instructions are actually executed. The computer system generates speculative data while executing instructions in the speculative state. Logic circuits are provided to detect any exception conditions which occur while executing the particular set in the speculative state. If the particular set is subject to an exception condition, the instructions of the set are re-executed to resolve the exception condition, and to incorporate the speculative data in the real state of the computer system.

Mechanism for executing computer instructions in parallel

Methods for handling exceptions caused by speculatively scheduled instructions or predicated instructions executed within a computer program are described. The method for speculatively scheduled instructions includes checking at a commit point of a speculatively scheduled instruction, a semaphore associated with the speculatively scheduled instruction and branching to an error handling routine if the semaphore is set. A set semaphore indicates that an exception occurred when the speculatively scheduled instruction was executed. For a predicated instruction the method includes checking a predicate of an eliminated branch and a semaphore associated with the speculative instruction at a commit point of the speculative instruction and branching to an error handling routine if the semaphore indicates that an exception occurred when said speculative instruction was executed, and the predicate is true, which indicates that said speculative instruction was properly executed.

Software mechanism for accurately handling exceptions generated by

A method of compiling an application to reduce the occurrence of speculative exceptions is described. The method includes the steps of compiling the application to provide a speculation table and an executable file, and obtaining profile information about said compiled application using representative data sets. The compiler includes a scheduler unit for rearranging the order of the instructions in the application to provide optimal performance. The speculation table comprises a number of entries corresponding to the instructions of the application, each entry including a tag identifying the instruction and a semaphore indicating whether or not the instruction is likely to cause an exception. The executable file is run using a number of representative data sets to profile information identifying those instructions that result in exceptions, and the tag of the instruction is stored in a log file. After the profiling has completed, the tags of the instructions causing exceptions are used to set the semaphores in the speculation table corresponding to the tag. The application is then re-compiled. During the recompilation, those instructions with their semaphores set; i.e. those instructions causing exceptions, will not be speculatively scheduled by the compiler.

Software mechanism for reducing exceptions generated by speculatively

Methods for handling exceptions caused by speculatively scheduled instructions or predicated instructions executed within a computer program are described. The method for speculatively scheduled instructions includes checking at a commit point of a speculatively scheduled instruction, a semaphore associated with the speculatively scheduled instruction and branching to an error handling routine in the semaphore is set. A set semaphore indicates that an exception occurred when the speculatively scheduled instruction was executed. For a predicated instruction the method includes checking a predicate of a eliminated branch and a semaphore associated with the speculative instruction at a commit point of the speculative instruction and branching to an error handing routine if the semaphore indicates that an exception occurred when the speculative instruction was executed, and the predicate is true, which indicates that the speculative instruction was properly executed.

There is provided a mechanism for propagating exception conditions in a computer system when instructions are subject to exception conditions. The apparatus includes a set of data registers for storing data manipulated by the instructions of the computer system, and a set of state registers for storing speculative states of data manipulated by the instructions, there being one state register associated with each data register. Furthermore, the apparatus includes a logic circuit, coupled to the set of state registers, for propagating the states from a source one of the state registers to a destination one of the state registers, if data stored in an associated source one of the data registers are used as a source for an associated destination one of data registers, and if data stored in the source data register were manipulated by a particular instruction subject to an exception condition.

Method and apparatus for propagating exception conditions of a computer

A compiler groups instructions into sets. The sets of instructions are related by data and control dependencies which are unresolvable by the compiler. Sets of instructions having unresolved dependencies are executed in a speculative state of the computer system under the assumption that an exception condition will not occur. However, if an exception condition does occur while executing a set of instructions in the speculative state, that exception condition is detected and the set of instructions is re-executed in a real state of the computer system to resolve the exception condition.

Apparatus and method for speculatively executing instructions in a

An apparatus for enforcing that selected instructions are executed in a correct order, comprising a first content addressable memory for storing load addresses of data read from the memory by the selected instructions. The first content addressable memory comparing the store addresses with the load addresses of data to be written to the memory. The first content addressable memory generating a first signal, if one of the load addresses is identical to a subsequently compared one of the store addresses. The apparatus further including a second content addressable memory for storing and comparing states of the data read and written by the selected instructions. The second content addressable memory generating a second signal, if one of the stored states is identical to one of said compared states. The stored states including a program counter to repeat the execution of the selected instructions upon detecting the first and second signals.

Growing community of inventors

Lexington, MA, United States of America

Michael C Adler