초록 ▼

A fast linked multiprocessor network including a plurality of processing modules implemented on a field programmable gate array and a plurality of configurable uni-directional links coupled among at least two of the plurality processing modules provide a streaming communication channel between at le

A fast linked multiprocessor network including a plurality of processing modules implemented on a field programmable gate array and a plurality of configurable uni-directional links coupled among at least two of the plurality processing modules provide a streaming communication channel between at least two of the plurality of processing modules. Such configuration provides a function accelerator that can feed at least one processor with data values using one custom instruction to put data values on at least one uni-directional serial link and that can extract data values from at least one processor using one custom instruction to get data values from the at least one uni-directional serial link.

대표청구항 ▼

The invention claimed is: 1. A method of accelerating software functions on a processor, comprising the steps of: feeding a plurality of inputs into a processor implemented on a field programmable gate array (FPGA) from an accelerator module via a first configurable uni-directional serial link (fir

The invention claimed is: 1. A method of accelerating software functions on a processor, comprising the steps of: feeding a plurality of inputs into a processor implemented on a field programmable gate array (FPGA) from an accelerator module via a first configurable uni-directional serial link (first link) formed from a first-in-first-out streaming communication network; wherein the accelerator module implements a function in hardware on the FPGA; wherein the first link is a communication channel dedicated to serial communication from the accelerator module to the processor; wherein a data value put by the accelerator module on the first link in one clock cycle is available for the processor to read in the immediate next clock cycle; executing by the processor a single first instruction that reads input data from the first link; executing by the processor a single second instruction that writes output data to a second configurable uni-directional serial link (second link) formed from a first-in-first-out streaming communication network; wherein the second link is a communication channel dedicated to serial communication from the processor to the accelerator module; wherein the first and second links are implemented on the FPGA; wherein a data value written by the processor to the second link in one clock cycle is available for the accelerator module to read in the immediate next clock cycle; extracting a plurality of outputs from the processor via the accelerator module via the second link; wherein the step of feeding comprises a step of feeding at least two or more parallel sequences of input values and the step of extracting comprises computing two or more parallel sequences of outputs; and wherein the feeding and extracting occur within a single clock cycle of the processor. 2. The method of claim 1, wherein the method further comprises a step of starting the computation of a function as soon as a first input is input on the first link. 3. The method of claim 1, wherein the method further comprises: feeding a plurality of inputs into the processor via a plurality of configurable unidirectional serial links (plurality of links) formed from a first-in-first-out streaming communication network using at least one custom instruction to put data values on each of the plurality of serial links; and starting the computation of a function when all inputs are available from the first link and the plurality of links. 4. The method of claim 1, wherein the method further comprises a step of selecting among a blocking and a non-blocking put function using additional custom instructions. 5. The method of claim 1, wherein the method further comprises a step of selecting among a blocking and a non-blocking get function using additional custom instructions.