9576results about "Concurrent instruction execution" patented technology

A method is shown for selecting active, or hot, code traces in an executing program for storage in a code cache. A trace is a sequence of dynamic instructions characterized by a start address and a branch history which allows the trace to be dynamically disassembled. Each trace is terminated by execution of a trace terminating condition which is a backward taken branch, an indirect branch, or a branch whose execution causes the branch history for the trace to reach a predetermined limit. As each trace is generated by the executing program, it is loaded into a buffer for processing. When the buffer is full, a counter corresponding to the start address of each trace is incremented. When the count for a start address exceeds a threshold, then the start address is marked as being hot. Each hot trace is then checked to see if the next trace in the buffer shares the same start address, in which case the hot trace is cyclic. If the start address of the next trace is not the same as the hot trace, then the traces in the buffer are checked to see they form a larger cycle of execution. If the traces subsequent to the hot trace are not hot themselves and are followed by a trace having the same start address as the hot trace, then their branch histories are companded with the branch history of the hot trace to form a cyclic trace. The cyclic traces are then disassembled and the instructions executed in the trace are stored in a code cache.

Task Computing computer system by segmenting the system into a plurality of implementation tiers of a presentation layer, a remote procedure call programming interface (API), a middleware layer to which the presentation layer interfaces via the remote procedure call API to real-time, dynamically generate a computer implemented task interface at the presentation layer to a semantically described source of function as a service on a computer system, and a service layer and a function source realization layer providing the semantically described source of function as the service on the computer system to which the middleware layer interfaces. Real-time and dynamically composing an executable task that comprises one or more services using the generated task interface at the presentation layer to one or more services on the computer based upon the semantically described application-, device- and service-rich computer.

The arithmetic processing apparatus of the present invention is an arithmetic processing apparatus that can be reconfigured in accordance with a processing mode and has a plurality of arranged unit arithmetic circuits. Each unit arithmetic circuit includes at least one input terminal, at least one output terminal, a first register which holds data, an adder which calculates a sum of two pieces of data, a second register which holds data, a bit shifter which shifts data left or right, a subtractor which calculates a difference between two pieces of data, an absolute value calculating unit which calculates an absolute value of data, and a path setting unit which sets a path according to the processing mode connecting among these circuit elements.

A single chip protocol converter integrated circuit (IC) capable of receiving packets generating according to a first protocol type and processing said packets to implement protocol conversion and generating converted packets of a second protocol type for output thereof, the process of protocol conversion being performed entirely within the single integrated circuit chip. The single chip protocol converter can be further implemented as a macro core in a system-on-chip (SoC) implementation, wherein the process of protocol conversion is contained within a SoC protocol conversion macro core without requiring the processing resources of a host system. Packet conversion may additionally entail converting packets generated according to a first protocol version level and processing the said packets to implement protocol conversion for generating converted packets according to a second protocol version level, but within the same protocol family type. The single chip protocol converter integrated circuit and SoC protocol conversion macro implementation include multiprocessing capability including processor devices that are configurable to adapt and modify the operating functionality of the chip.

A new general method for building hybrid processors achieves higher performance in applications by allowing more powerful, tightly-coupled instruction set extensions to be implemented in reconfigurable logic. New instructions set configurations can be discovered and designed by automatic and semi-automatic methods. Improved reconfigurable execution units support deep pipelining, addition of additional registers and register files, compound instructions with many source and destination registers and wide data paths. New interface methods allow lower latency, higher bandwidth connections between hybrid processors and other logic.

In a data-processing method, first result data may be obtained using a plurality of configurable coarse-granular elements, the first result data may be written into a memory that includes spatially separate first and second memory areas and that is connected via a bus to the plurality of configurable coarse-granular elements, the first result data may be subsequently read out from the memory, and the first result data may be subsequently processed using the plurality of configurable coarse-granular elements. In a first configuration, the first memory area may be configured as a write memory, and the second memory area may be configured as a read memory. Subsequent to writing to and reading from the memory in accordance with the first configuration, the first memory area may be configured as a read memory, and the second memory area may be configured as a write memory.

The present invention present methods and architectures for the pipelining of read operation with write operations. In particular, methods are presented for pipelining data relocation operations that allow for the checking and correction of data in the controller prior to its being re-written, but diminish or eliminate the additional time penalty this would normally incur. A number of architectural improve are described to facilitate these methods, including: introducing two registers on the memory where each is independently accessible by the controller; allowing a first memory register to be written from while a second register is written to; introducing two registers on the memory where the contents of the registers can be swapped.

An apparatus and method for performing multithreaded operations includes partitioning the general purpose and/or floating point processor registers into register subsets, including overlapping register subsets, allocating the register subsets to the threads, and managing the register subsets during thread switching. Register overwrite buffers preserve thread resources in overlapping registers during the thread switching process. Thread resources are loaded into the corresponding register subsets or, when overlapping register subsets are employed, into either the corresponding register subset or the corresponding register overwrite buffer. A thread status register is utilized by a thread controller to keep track of READY/NOT-READY threads, the active thread, and whether single-thread or multithread operations are permitted. Furthermore, the registers in the register subsets include a thread identifier field to identify the corresponding thread. Register masks may also be used to identify which registers belong to the various register subsets.

A digital data processor integrated circuit (1) includes a plurality of functionally identical first processor elements (6A) and a second processor element (5). The first processor elements are bidirectionally coupled to a first cache (12) via a crossbar switch matrix (8). The second processor element is coupled to a second cache (11). Each of the first cache and the second cache contain a two-way, set-associative cache memory that uses a least-recently-used (LRU) replacement algorithm and that operates with a use-as-fill mode to minimize a number of wait states said processor elements need experience before continuing execution after a cache-miss. An operation of each of the first processor elements and an operation of the second processor element are locked together during an execution of a single instruction read from the second cache. The instruction specifies, in a first portion that is coupled in common to each of the plurality of first processor elements, the operation of each of the plurality of first processor elements in parallel. A second portion of the instruction specifies the operation of the second processor element. Also included is a motion estimator (7) and an internal data bus coupling together a first parallel port (3A), a second parallel port (3B), a third parallel port (3C), an external memory interface (2), and a data input/output of the first cache and the second cache.

A microprocessor integrated circuit including a processing unit disposed upon an integrated circuit substrate is disclosed herein. The processing unit is designed to operate in accordance with a predefined sequence of program instructions stored within an instruction register. A memory, capable of storing information provided by the processing unit and occupying a larger area of the integrated circuit substrate than the processing unit, is also provided within the microprocessor integrated circuit. The memory may be implemented using, for example dynamic or static random-access memory. A variable output frequency system clock, such as generated by a ring oscillator, is also disposed on the integrated circuit substrate.

A method is disclosed for transforming a portion of a computer program comprising a list of sequential instructions comprising control code and data processing code and a program separation indicator indicating a point where said sequential instructions may be divided to form separate sections that are capable of being separately executed and that each comprise different data processing code. The m method comprises the steps of: (i) analysing said portion of said program to determine if said sequential instructions can be divided at said point indicated by said program separation indicator and in response to determining that it can: (iia) providing data communication between said separate sections indicated by said program separation indicator, such that said separate sections can be decoupled from each other, such that at least one of said sections is capable of being separately executed by an execution mechanism that is separate from an execution mechanism executing another of said separate sections, said at least one of said sections being capable of generating data and communicating said data to at least one other of said separate sections; and in response to determining it can not: (iib) not performing step (iia). If step (iia) is not performed then a warning may be output, or the program may be amended so it can be separated at that point, or the program separation indicator may be removed and the sections that were to be separated merged.

Extended operand management indicators stored during initial program execution enable management and regulation of operand values and streamline their handling. Operand values are stored in new types of stores. Operand location management indicators indicate current operand value locations among various store types for selected operands. Indicated operand-forwarding policies for selected operands streamline forwarding of operand values from source instructions to value receiving target instructions. Indicated loop iterations of operand source instructions enable forwarding of operands over more than one loop iteration. Stride indicators indicate strides of program loop accesses to matrix operands. Inter-loop indicators enable forwarding of operand values from source loop instructions directly to target loop instructions. Constant or nearly constant operands are indicated to enable their storage in special caches. Operands used for cross-CPU serialization are indicated for special handling and storage in spin lock cache. Indicators of farthest back and farthest forward branches since operand last update are used to enhance the prediction of longer-range branch directions. Virtual predicate operand indicators streamline execution of densely branching program code. Stack operand indicators enable nullification of paired stack pointer increment-decrement operations to avoid serious operand serialization bottlenecks in very high issue rate machines.

A method and computer-usable medium including instructions for performing a method for scheduling executable transactions within a multicore processor comprising a plurality of processor elements. The method includes listing, using at least one distribution queue, a portion of the executable transactions in order of eligibility for execution. A plurality of executable transaction schedulers are provided, wherein each executable transaction scheduler includes a scheduling process for determining a most eligible executable transaction for execution from at least one candidate executable transaction ready for execution. The executable transaction schedulers are linked together to provide a multilevel scheduler. The most eligible executable transaction is output from the multilevel scheduler to the at least one distribution queue.

The invention discloses a convolution neural network parallel processing method based on a large-scale high-performance cluster. The method comprises the steps that (1) a plurality of copies are constructed for a network model to be trained, model parameters of all the copies are identical, the number of the copies is identical with the number of nodes of the high-performance cluster, each node is provided with one model copy, one node is selected to serve as a main node, and the main node is responsible for broadcasting and collecting the model parameters; (2) a training set is divided into a plurality of subsets, the training subsets are issued to the rest of sub nodes except the main mode each time to conduct parameter gradient calculation together, gradient values are accumulated, the accumulated value is used for updating the model parameters of the main node, and the updated model parameters are broadcast to all the sub nodes until model training is ended. The convolution neural network parallel processing method has the advantages of being capable of achieving parallelization, improving the efficiency of model training, shortening the training time and the like.

A reconfigurable data path processor comprises a plurality of independent processing elements. Each of the processing elements advantageously comprising an identical architecture. Each processing element comprises a plurality of data processing means for generating a potential output. Each processor is also capable of through-putting an input as a potential output with little or no processing. Each processing element comprises a conditional multiplexer having a first conditional multiplexer input, a second conditional multiplexer input and a conditional multiplexer output. A first potential output value is transmitted to the first conditional multiplexer input, and a second potential output value is transmitted to the second conditional multiplexer output. The conditional multiplexer couples either the first conditional multiplexer input or the second conditional multiplexer input to the conditional multiplexer output, according to an output control command. The output control command is generated by processing a set of arithmetic status-bits through a logical mask. The conditional multiplexer output is coupled to a first processing element output. A first set of arithmetic bits are generated according to the processing of the first processable value. A second set of arithmetic bits may be generated from a second processing operation. The selection of the arithmetic status-bits is performed by an arithmetic-status bit multiplexer selects the desired set of arithmetic status bits from among the first and second set of arithmetic status bits. The conditional multiplexer evaluates the select arithmetic status bits according to logical mask defining an algorithm for evaluating the arithmetic status bits.

A method for prefetching structured data, and more particularly a mechanism for observing address references made by a processor, and learning from those references the patterns of accesses made to structured data. Structured data means aggregates of related data such as arrays, records, and data containing links and pointers. When subsequent accesses are made to data structured in the same way, the mechanism generates in advance the sequence of addresses that will be needed for the new accesses. This sequence is utilized by the memory to obtain the data somewhat earlier than the instructions would normally request it, and thereby eliminate idle time due to memory latency while awaiting the arrival of the data.

An apparatus and method are shown for collecting a branch history value of a program executing in a processor. A current start address register latches a program count value in response to a trace termination condition, such as an indirect branch instruction. A current branch history register is cleared in response to the trace termination condition and shifts in a branch outcome value of the processor in response to a conditional direct branch instruction. A last trace start address latches the content of the current trace start address and a last branch history register latches the content of the current branch history register when a trace termination condition occurs.