摘要:
A timing and control circuit and method for a synchronous vector processor (SVP) device having a plurality of one-bit processor elements organized in a linear array. This circuit includes a master controller circuit, a vertical timing generator circuit, a constant generator circuit, a horizontal timing generator circuit and an instruction generator circuit.
摘要:
A timing and control circuit and method for a synchronous vector processor (SVP) device having a plurality of one-bit processor elements organized in a linear array. This circuit includes a master controller circuit, a vertical timing generator circuit, a constant generator circuit, a horizontal timing generator circuit and an instruction generator circuit.
摘要:
A synchronous vector processor (SVP) device having a plurality of one-bit processor elements organized in a linear array. The processor elements are all controlled in common by a sequencer, a state machine or a control circuit (controller) to enable operation as a parallel processing device. Each processor element includes a set of input registers, two sets of register files, a set of working registers, an arithmetic logic unit including a one-bit full adder/subtractor, and a set of output registers. In video applications each processor element operates on one pixel of a horizontal scan line and is capable of real-time digital processing of video signals. In video applications a data input control circuit including a master controller circuit, a vertical timing generator circuit, a constant generator circuit, a horizontal timing generator circuit and an instruction generator circuit is provided.
摘要:
An apparatus and method for performing rotation of data in a register file memory. The apparatus utilizes a rotation address generator including rotation value, modulus, and offset registers, a comparator, a data selector, logic circuitry, and a subtractor. A predetermined area (P.times.Q) of the register file memory and a rotation value corresponding to the number of bits to be rotated in the rotation area is designated by an instruction program memory. An instruction decoder signals the register file, modulus register, rotation value register, and offset register of an impending rotation of data, thereby enabling loading of the modulus and rotation value registers and resetting of the offset register. A counter provides a relative address to the comparator and data selector. The comparator compares the relative address with the output of the modulus register, determining whether selected ones of the addressed register file locations fall inside or outside of the rotation area, and send an appropriate signal to the logic circuitry, an OR gate. This OR gate also receives a rotate or not-rotate signal. Consequently, either an absolute address equal to (the value of the relative address-2 * the offset value) mod (8 * the value in the modulus register) or equal to the relative address, based on predetermined conditions, is utilized to access rotationally data from the register file.
摘要:
A data processing apparatus includes a dual port data input register, first and second sequential ring counters, first and second register files, first and second data transfer circuits, a dual port data output register and N single bit processing elements. The dual port data input register has an M bit wide input port and an N bit wide output port. The first sequential ring counter cyclically selects one column of the data input register for input. The first data transfer circuit has a plurality of input segments, which are subsets of consecutive columns of the data input register. The first data transfer circuit transfers data from a selected row of the data input register to a selected row of the first register file for all columns of each input segment in a repetitive sequence of consecutive input segments in synchronism with said first sequential ring counter. The dual port data output register, the second register file, the second sequential ring counter and the second data transfer circuit are similarly organized to output data. Each of the N single bit processing elements is connected to a predetermined column of the first and second register files and capable of data processing operations under program control including data transfer to and from selected rows of said predetermined column of said first and second register files.
摘要:
A synchronous vector processor SVP device having a plurality of one-bit processor elements organized in a linear array. The processor elements are all controlled in common by a sequencer, a state machine or a control circuit (controller) to enable operation as a parallel processing device. Each processor element includes a set of input registers, two sets of register files, a set of working registers, an arithmetic logic unit including a one-bit full adder/subtractor, and a set of output registers. In video applications each processor element operates on one pixel of a horizontal scan line and the SVP is capable of real-time digital processing of video signals. In video applications a data input control circuit including a master controller circuit, a vertical timing generator circuit, a constant generator circuit, a horizontal timing generator circuit and an instruction generator circuit is provided. In order to distribute variables to each processor element simultaneously the data input control circuit is provided with a set of auxiliary registers and an addressing structure to modulate one of the processor elements' working registers. In this manner variables are provided to the SVP device in lieu of a designated control instruction bit.
摘要:
A synchronous vector processor SVP device having a plurality of one-bit processor elements organized in a linear array. The processor elements are all controlled in common by a sequencer, a state machine or a control circuit (controller) to enable operation as a parallel processing device. Each processor element includes a set of input registers, two sets of register files, a set of working registers, an arithmetic logic unit including a one-bit full adder/subtractor, and a set of output registers. In video applications each processor element operates on one pixel of a horizontal scan line and is capable of real-time digital processing of video signals. The SVP includes interconnecting circuitry enabling the individual processor elements to retrieve data from and transmit data to their first and second nearest neighbors on either side. At the chip level external connections are provided to enable cascading of several SVP devices.
摘要:
A constant generator is described which provides a sequence of digital constants in a synchronous vector processor. The constant generator includes a constant loop memory for storing data words organized into a plurality of data constant patterns and an end of loop bit, a constant loop counter for supplying sequential addresses to the constant loop memory, and a constant loop counter controller for loading the counter with one of a set of predetermined starting addresses associated with a desired constant pattern stored in the constant loop memory. Additionally, a method of supplying a sequence of digital constants in said constant generator is disclosed and includes the steps of storing a plurality of data words in a plurality of constant patterns, where each constant pattern includes an end of loop bit, supplying an address to the constant loop memory and supplying sequential addresses to the constant loop memory.
摘要:
A synchronous vector processor (SVP) (30) is provided to realize a horizontal decimation filter by processing in input value through a plurality of parallel processing elements (40). A plurality of input pixel values (80) representing a horizontal line of information in a video display are input to a data input register (DIR) (31) of the SVP (30). Each of the processing elements (40) is associated with a filter output and is operable to perform all calculations necessary to realize a multi-tap filter structure for the associated output. This is achieved by first increasing the frequency of the input signal by inserting zeros therein and then performing a number of multiplications and additions to generate an output value for that processing element, this realizing an interpolation FIR filter algorithm. The finite impulse response (FIR) filter algorithm is defined by predetermined filter coefficients stored in a constant generator (71d). Each of the processing elements are utilized to multiply a plurality of near-neighbor input values with FIR filter coefficients that are obtained from a constant generator (71d). The resulting sum for each of the processing elements is then input to the a data output register (DOR) (16) as the filter output. The output of the SVP (30) is then input to line memory (90) that is operable to decimate the output of select ones of the processing elements of the SVP (30). This rearranges the outputs to decrease the number of output pixels for each line relative to the number of input pixels for each line.
摘要:
A single-instruction multiple-data processor (10) has an input layer especially designed for high data input and output rates. The processor (10) has a number of processing elements (20), each corresponding to incoming data samples. The processing elements (20) are interleaved so that a set of samples can be input in parallel. The processor (10) is programmable, which makes it especially useful for digital filtering. Near-neighbor communications (41) among processing elements (20) realize the delays required for horizontal filtering.