Advanced Micro Devices Corporation

Vlsi Technology, Inc.

Fujitsu Corporation

Hal Computer Systems, Inc.

Creigton S Asato

Christoph Ditzen

Vydhyanathan Kalyanasundharam

William Alexander Hughes

Kevin John McGrath

John J Zasio

James Allely Rowson

Robert K Montoye

Suresh K Dholakia

Tarang Patil

In one embodiment, a processing node includes a plurality of processor cores and a reconfigurable interconnect. The processing node also includes a controller configured to schedule transactions received from each processor core. The interconnect may be coupled to convey between a first processor core and the controller, transactions that each include a first corresponding indicator that indicates the source of the transaction. The interconnect may also be coupled to convey transactions between a second processor core and the controller, transactions that each include a second corresponding indicator that indicates the source of the transaction. When operating in a first mode, the interconnect is configurable to cause the first indicator to indicate that the corresponding transactions were conveyed from the second processor core and to cause the second indicator to indicate that the corresponding transactions were conveyed from the first processor core.

Processing node including a plurality of processor cores and an interconnect configurable in a test-mode to cause first and second transaction source indicators to be interchanged

A method and circuit is disclosed for tagging and invalidating speculatively executed instructions. The method includes fetching a first plurality of instructions which includes a conditional branch instruction which identifies a target address if the branch is taken. The conditional branch instruction is detected and in response thereto first and second instruction tags are generated. At least a first instruction is tagged with the first instruction tag wherein the first instruction is included in the first plurality of instructions and wherein the first instruction sequentially follows the first conditional branch instruction in program order. Thereafter, a second plurality of instructions are fetched wherein the second plurality of instructions corresponds to the target address of the conditional branch instruction. At least one of these second plurality of instructions is tagged with the second instruction tag. Thereafter, the conditional branch instruction is executed and resolved. A branch status operand is generated in response to executing the conditional branch instruction wherein the branch status operand indicates whether the conditional branch instruction is resolved as taken or not taken. The generated branch status operand is compared against the first and second instruction tags. In response to comparing the branch status operand, one of the first and second instructions is invalidated indicating that it was within the wrong program path of execution.

Circuit and method for tagging and invalidating speculatively executed instructions

A circuit and method is disclosed which reduces data dependencies between instructions within an application program thereby reducing time delays associated therewith. In one embodiment, a data dependent instruction is translated into at least first and second speculative data independent instructions wherein the data dependent instruction, if executed, produces results which are dependent on a data result provided by execution of a data independent instruction. The first and second speculative data independent instructions are executed to generate first and second speculative results. It is noted that these results are generated independent of the results produced by executing the data independent instruction. Once the data independent instruction is executed and its data result is generated, one of the first and second speculative results is invalidated depending on the value of the data results.

Circuit and method for reducing data dependencies between instructions

A circuit combines the functions of a logic gate and a latch to lower steady state power dissipation during gate operation. The circuit operates in two modes: a flow-through mode and a latched mode. In the flow-through mode, a gate portion which receives one or more digital input signals implements the complement of a desired Boolean logic function on the input signals and provides an internal signal. The gate portion may have a steady-state power dissipation while providing the internal signal. An inverter in a latch portion of the circuit inverts the internal signal to generate an output signal which represents the desired logical combination of the input signals. The inverter provides the output signal with a full-range CMOS voltage. In latched mode, the gate portion is disabled to stop the steady-state power dissipation while the latch portion of the circuit preserves the desired output signal.

Combined logic gate and latch

A buffer circuit which exhibits increased speed in transitions between binary states. A control transistor is coupled between a pull-up transistor and an input terminal. During low-to-high input signal transitions, the control transistor limits the signal swing at the input terminal such that small variations in the input terminal voltage result in larger voltage variations at the output terminal. During such transitions, the control transistor simultaneously decouples the input terminal from the pull-up transistor, thereby decoupling the input capacitance from the pull-up transistor. As a result, the speed with which the pull-up transistor can pull the output high is increased. As the number of input signal desired to be processed increases, the reduction in logic transition time becomes more significant. Some versions further include a pull-down transistor having a control terminal coupled to the gate of the pull-up transistor and to a power-down terminal. When an appropriate voltage is applied to the power-down terminal, the pull-up transistor is turned off and the pull-down transistor is turned on such that no DC current flows in the circuit.

CMOS buffer circuit having increased speed

Output data points of a digital FIR filter are calculated by storing input data points in an addressable memory and accessing the addressable memory to supply a new input data point exactly once for each output data point after a first output data point and storing each input data point in a first recirculating memory for so long as that input data point is needed to calculate a next output data point. The input data points stored in the first recirculating memory are used to calculate output data points. Furthermore, coefficients are stored in a second recirculating memory and are used to calculate the output data points. As a result, only one memory access is required per output data point.

Finite impulse response filter

A sequentially accessible, non-volatile data storage circuit for generating constants includes a logic array for non-volatile storage of programmed data words and a recirculating shift register for causing the first one of the data words to appear at a data output of the data storage circuit in response to a reset signal and to cause the next data word to appear at the data output in response to a clock signal.

Sequentially accessible non-volatile circuit for storing data

According to a technique for determining delays through multi-stage datapath elements, estimates of the delays through each stage of the datapath element are computed in accordance with an the equation such as:

Growing community of inventors

San Jose, CA, United States of America

Creigton S Asato