CA2000151C - Parallel data processor - Google Patents
Parallel data processorInfo
- Publication number
- CA2000151C CA2000151C CA002000151A CA2000151A CA2000151C CA 2000151 C CA2000151 C CA 2000151C CA 002000151 A CA002000151 A CA 002000151A CA 2000151 A CA2000151 A CA 2000151A CA 2000151 C CA2000151 C CA 2000151C
- Authority
- CA
- Canada
- Prior art keywords
- control signals
- data
- memory
- cells
- responsive
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F15/00—Digital computers in general; Data processing equipment in general
- G06F15/76—Architectures of general purpose stored program computers
- G06F15/80—Architectures of general purpose stored program computers comprising an array of processing units with common control, e.g. single instruction multiple data processors
- G06F15/8007—Architectures of general purpose stored program computers comprising an array of processing units with common control, e.g. single instruction multiple data processors single instruction multiple data [SIMD] multiprocessors
- G06F15/8023—Two dimensional arrays, e.g. mesh, torus
Abstract
A parallel data processor comprised of an array of identical cells concurrently performing identical operations under the direction of a central controller, and incorporating one or more of a special cell architecture including a segmented memory, conditional logic for preliminary processing, and circuitry for indicating when the cell is active, and programmable cell interconnection including cell bypass and alternate connection of edge cells.
Description
PARALLEL DATA PROCESSOR
BACKGROUND OF THE INVENTION
~ -, -j 1. Field of the Invention The present invention relates to a high speed parallel data processing system and, more particularly, ' to a parallel data processing system comprised of an array of identical interconnected cells operating under the control of a master controller to operate on single bits of data. Even more particularly, the invention pertains to the architecture of the cells and to specific patterns of interconnection.
BACKGROUND OF THE INVENTION
~ -, -j 1. Field of the Invention The present invention relates to a high speed parallel data processing system and, more particularly, ' to a parallel data processing system comprised of an array of identical interconnected cells operating under the control of a master controller to operate on single bits of data. Even more particularly, the invention pertains to the architecture of the cells and to specific patterns of interconnection.
2. Background Art Certain data processing tasks require that substantially identical logical or arithmetic operations be performed on large amounts of data. One approach to carrying out such tasks which is drawing increasing attention is parallel processing. In parallel processing, each element or cell of an array processor made up of such cells processes its own bit of data at the same time as all other cells o~ the array processor ; perform the same process on their own bit of data. Such machines are' referred to by several names, including Single Instruction-Multiple Data (SIMD) machines.
A common arrangement for such a machine is as a rectangular array of cells, with each interior cell :, ~ , -, .. - . ~ , ~ .- .
:
- 2 ~
connected to its four nearest neighboring cells and each - edge cell connected to a data inpuk/output device. Each - cell is connected as well to a master controller whlch coordinates the movement of data through the array by providing appropriate instructlons to the processing elements. Such an array proves useful, for example, in high resolution image processing. The image pixels comprise a data matrix which can be loaded into and processed quickly and efficiently by the processor array.
Although all may be based upon the same generic concept of an array of cells all performing the same function in unison, parallel processors vary in details of cell design. For example, U.S. Patent No. 4,215,401 to Holsztynski et al discloses a cell which includes a random access memory (RAM), a single bit accumulator, and : a simple logical gate. The disclosed cell is extremely simple and, henc~, inexpensive and easily fabricated.
negative consequence of this simplicity, however, is that some computational algorithms are quite cumbersome so that it may require many instructions to perform ai simple and often repeated tasX.
U.S. Patent No. 4,739,474, to Holsztynski et al, represents a higher level of complexi~y, in which ~he logic gate is replaced by a full adder capable o~
performing both arithmetic and logical functions.
Pressing the full adder into dual service creates an . ' .
e~ficiency which more than offsets the added complexity - and cost incurred by including a full adder in each cell.
It is important to note that the substitution of a full adder for a logic gate, while superficially simple, is in reality a change of major consequence. The cell structure cannot be allowed to become too complex. This is because in a typical array the cell will be repeated dozens if not hundreds of times. The cost of each additional element in terms of money and space on a VLSI
chip is therefore multiplied many times. It is therefore no simple matter to identify those functions which are sufficiently useful, to justify their incorporation into the cell. It is similarly no simple matter to implement those functions so that their incorpoxation is not realized at too high a cost.
Parallel processors may also vary in the manner of cell interconnection. As mentioned abovs, cells are typically connected to their nearest physical neighbors.
All cells except those at t~e edge of the entire array are connected to four neighbors. It has not heretofore been completely appreciated, however, that significant benefits may flow from providing for alternate paths of interconnection and, specifically, in providing programmable, flexible interconnection between cells.
.;
- : - : : , -:
~. , .; .
SUMMARY OF THE INVENTION
There is thus a continuing need to identi~y and implement variations in cell architecture which result in a maximal increase in cell efficiency with a minimal increase in cell complexity. This need is addressed in the present invention through the provision in each cell of a segmented random access memory to permit the performance of more than one read/write operation per clock cycle. It is further addressed through the provision of circuitry ahead of the main processing element for conditionally processing data prior to processing in the main processing element. It is further addressed through the provision of circuitry in the cell : by which the master controller can deactivate or activate the cell, or by which the cell can provide indication of whether the cell has activated or deactivated itself.
This permits independent conditional operations for each individual cell.
The~present invention not only exploits variatians in cell architecture, but also in cell interconnection. ;~
In one enhancement, at least one bypass network is intermeshed with the array for bypassing cells as needed or desired. In another embodiment, in which the cells are arranged in an n x m array, cells an at least one edge are arranged to be connectable to either an external device or to another cell in the array, thus creating the .. ~ ' ~, - ,. ~- . - ., , , . ~
. ~ 5 ability to reshape the virtual configuration of the array.
BRIEF DESCRIPTION OF THE DRAWINGS
The foregoing and other features of the present invention will become apparent to one of ordinary skill in the art to which the invention pertains from the - following detailed description when read in conjunction with the drawings, in which:
Figure 1 is a block diagram of a preferred embodiment of a processor array according to the invention;
Figure 2 is a schematic diagram of a second preferred embodiment of a processor array according to the invention:
Figures 3(a), 3(b), and 3(c) are block diagrams showing alternate array interconnection according to a preferred embodiment of the invention;
Figure 4 i5 a schematic diagram of a preferred embodiment of a cell according to the invention;
: 20 Figure 5 is a schematic diagram of a preferred embodiment of a se~mented memory according to the nventlon;
Figure 6 is a schematic diagram of a preferred em~odiment of a conditional logic unit according to the invention; and - . . . : ,. . ~ - .
. ~: . .
~ :
:
., Figure 7 is a block diagram o~ a typical application for a processor array according to the present invention.
.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
General details concerning the arrangement o~
parallel processors as an array of identical cells are available from aforementioned U.S. Pat~nt No. 4,739,474 to Holsztynski et al, the specification of which is hereby incorporated by reference. As shown in that patent and in Figure 1 herein, an array 10 may comprise n rows and m columns of identical processors or cells PO O
through Pn m to form an n x m rectangular network or matrix (the subscripts "n" and "m" representing variables which may assume any positive integral value).
There are thus cells having four nearest neighbors (interior cells) and cells having less than four nearest neighbors (edge cells~. For uniform orientation, the directions north, south, east, and west are used to reference up, down, right, and left, respectively, in the drawing, so that the nearest neighbors will be re~erred to as N, S, E, and N, respectively.
The edge cells of the array 10 are connected to a data inpu /output unit 20 as illustrated. Every cell receives command and address signals co - Cx and Ao - Ay, respectively, from a controller 30. In addition, clock signals may be provided by the controller 30 to each cell, and the data input/output unit 20 may also be -controlled by the controller 30. These and other details are available from the Holsztynski et al '474 patent.
Figure 1 also shows two multiplexer networXs intermeshed with the cell array. The first multiplexer network is made up of multiplexers Ml 1 through Mn m and Ml 1' through Mn m' ("n" and "m" here and elsewhere assuming values not necessarily related to those assumed in connecticn with labelling the cells unless otherwise specified). As will be developed below, this networ~ is used for selective bypass of cells on an individual, row, column, or blocX basis.
The second multiplexer network is made up of multiplexers Ml'' through Mn~l~ and is used for selectively altering the connection of the north edge cells to either the data input~output unit 20 or to another cell in the array. An array may be made up of several interlinked chips, éach supplying its own section or subarray of the main array. The second multiplexer network provides a capability of changing the virtual configuration of the array.
All of the multiplexers are responsive to ~he controller 30 through connections which have been omitted for clarity. It should be noted that the communicatisn paths provided by the multiplexers are bidirectional, so - . . . .. ~. , ::.,: .- .. -.. : ~ . . ... .
. ~ : . : - .: . , . : ....... .. ... . .
~ . . ,.- : : - ~ . : ~ :: . . :, ~ ~ : : - : .: . . . - . : : : :: .::
that the multiplexers should be regarded as multiplexer/demultiplexers. It should also be noted that the term multiplexer is used herein generically to encompass ~ny device for performing the functions ascribed to these elements, i.e., line selection based on a control input. Thus, simple switches may be multiplexers to the extent they are provided to perform such a function.
Figure 2 is a more detailed rendering of an array 10 according to the present invention, in this case an 8 x 16 array of 128 identical cells Pl,l P8,16 P
: equipment such as controller 30 and connections thereto ; have been omitted far clarity, but it will be understood that such connections are present. The embodiment of Figure 2 has a multiplexer bypass network with members arranged between selected cells thus bypassing intervening cells. As can be seen, it is not necessary that there be a multiplexer between every cell and its nearest neighbors in order to bypass entire rows, columns, or blocks and so, in effect, alter the virtual dimensions of the array to suit a particular application.
; In the embodiment shown, there is one set of eight multiplexers per column and one set of four per row.
Thus, in general, it is necessary only to have half as many multiplexers as cells in any row or column. Of course, lf there is an alternate communication path, .
. . ; j ;
(such as the CM path described below) it may be desixable to provide that path with a multiplexer ne~work as well.
When cell bypass is invoked, cell-to-cell communication is based on the value of a latched bypass S control word which continuously drives the row and column multiplexer select lines. If it were desired to bypass the first column, for example, the westernmost I/O pins - become logically connected to the west port of the cells in the second column.
Row and column bypass can be separately controlled so that all possible virtual array sizes, such as 7 x 12, 4 x 10, 1 x 16, or even 1 x 1, can be obtained. The embodiment of Figure 2, however, is designed so that it is not possible to bypass the array lO entirely. The ; 15 cells in the easternmost column are always connected to the east I/0 pins so that data must always pass through the easternmost column of cells. Similarly, the cells in the southernmost row are always connected to the south I/0 pins so that data must always pass through the southernmost row of cells. There is therefore a minimum 1 x l configuration.
The benefits of the cell bypass network are well exemplified in operations such as parallel data summation, also known as an array sum; An array sum is a - 25 summation of the same data type variable in each cell throuyhout the entire array. With programmable bypass, :
this summation may be achieved using the ~ollowing algorithm:
(1) program the cell bypass network to ~ull 8 x 16 configuration;
(2) command each cell to shift its variable to its ~ eastern neighbor;
A common arrangement for such a machine is as a rectangular array of cells, with each interior cell :, ~ , -, .. - . ~ , ~ .- .
:
- 2 ~
connected to its four nearest neighboring cells and each - edge cell connected to a data inpuk/output device. Each - cell is connected as well to a master controller whlch coordinates the movement of data through the array by providing appropriate instructlons to the processing elements. Such an array proves useful, for example, in high resolution image processing. The image pixels comprise a data matrix which can be loaded into and processed quickly and efficiently by the processor array.
Although all may be based upon the same generic concept of an array of cells all performing the same function in unison, parallel processors vary in details of cell design. For example, U.S. Patent No. 4,215,401 to Holsztynski et al discloses a cell which includes a random access memory (RAM), a single bit accumulator, and : a simple logical gate. The disclosed cell is extremely simple and, henc~, inexpensive and easily fabricated.
negative consequence of this simplicity, however, is that some computational algorithms are quite cumbersome so that it may require many instructions to perform ai simple and often repeated tasX.
U.S. Patent No. 4,739,474, to Holsztynski et al, represents a higher level of complexi~y, in which ~he logic gate is replaced by a full adder capable o~
performing both arithmetic and logical functions.
Pressing the full adder into dual service creates an . ' .
e~ficiency which more than offsets the added complexity - and cost incurred by including a full adder in each cell.
It is important to note that the substitution of a full adder for a logic gate, while superficially simple, is in reality a change of major consequence. The cell structure cannot be allowed to become too complex. This is because in a typical array the cell will be repeated dozens if not hundreds of times. The cost of each additional element in terms of money and space on a VLSI
chip is therefore multiplied many times. It is therefore no simple matter to identify those functions which are sufficiently useful, to justify their incorporation into the cell. It is similarly no simple matter to implement those functions so that their incorpoxation is not realized at too high a cost.
Parallel processors may also vary in the manner of cell interconnection. As mentioned abovs, cells are typically connected to their nearest physical neighbors.
All cells except those at t~e edge of the entire array are connected to four neighbors. It has not heretofore been completely appreciated, however, that significant benefits may flow from providing for alternate paths of interconnection and, specifically, in providing programmable, flexible interconnection between cells.
.;
- : - : : , -:
~. , .; .
SUMMARY OF THE INVENTION
There is thus a continuing need to identi~y and implement variations in cell architecture which result in a maximal increase in cell efficiency with a minimal increase in cell complexity. This need is addressed in the present invention through the provision in each cell of a segmented random access memory to permit the performance of more than one read/write operation per clock cycle. It is further addressed through the provision of circuitry ahead of the main processing element for conditionally processing data prior to processing in the main processing element. It is further addressed through the provision of circuitry in the cell : by which the master controller can deactivate or activate the cell, or by which the cell can provide indication of whether the cell has activated or deactivated itself.
This permits independent conditional operations for each individual cell.
The~present invention not only exploits variatians in cell architecture, but also in cell interconnection. ;~
In one enhancement, at least one bypass network is intermeshed with the array for bypassing cells as needed or desired. In another embodiment, in which the cells are arranged in an n x m array, cells an at least one edge are arranged to be connectable to either an external device or to another cell in the array, thus creating the .. ~ ' ~, - ,. ~- . - ., , , . ~
. ~ 5 ability to reshape the virtual configuration of the array.
BRIEF DESCRIPTION OF THE DRAWINGS
The foregoing and other features of the present invention will become apparent to one of ordinary skill in the art to which the invention pertains from the - following detailed description when read in conjunction with the drawings, in which:
Figure 1 is a block diagram of a preferred embodiment of a processor array according to the invention;
Figure 2 is a schematic diagram of a second preferred embodiment of a processor array according to the invention:
Figures 3(a), 3(b), and 3(c) are block diagrams showing alternate array interconnection according to a preferred embodiment of the invention;
Figure 4 i5 a schematic diagram of a preferred embodiment of a cell according to the invention;
: 20 Figure 5 is a schematic diagram of a preferred embodiment of a se~mented memory according to the nventlon;
Figure 6 is a schematic diagram of a preferred em~odiment of a conditional logic unit according to the invention; and - . . . : ,. . ~ - .
. ~: . .
~ :
:
., Figure 7 is a block diagram o~ a typical application for a processor array according to the present invention.
.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
General details concerning the arrangement o~
parallel processors as an array of identical cells are available from aforementioned U.S. Pat~nt No. 4,739,474 to Holsztynski et al, the specification of which is hereby incorporated by reference. As shown in that patent and in Figure 1 herein, an array 10 may comprise n rows and m columns of identical processors or cells PO O
through Pn m to form an n x m rectangular network or matrix (the subscripts "n" and "m" representing variables which may assume any positive integral value).
There are thus cells having four nearest neighbors (interior cells) and cells having less than four nearest neighbors (edge cells~. For uniform orientation, the directions north, south, east, and west are used to reference up, down, right, and left, respectively, in the drawing, so that the nearest neighbors will be re~erred to as N, S, E, and N, respectively.
The edge cells of the array 10 are connected to a data inpu /output unit 20 as illustrated. Every cell receives command and address signals co - Cx and Ao - Ay, respectively, from a controller 30. In addition, clock signals may be provided by the controller 30 to each cell, and the data input/output unit 20 may also be -controlled by the controller 30. These and other details are available from the Holsztynski et al '474 patent.
Figure 1 also shows two multiplexer networXs intermeshed with the cell array. The first multiplexer network is made up of multiplexers Ml 1 through Mn m and Ml 1' through Mn m' ("n" and "m" here and elsewhere assuming values not necessarily related to those assumed in connecticn with labelling the cells unless otherwise specified). As will be developed below, this networ~ is used for selective bypass of cells on an individual, row, column, or blocX basis.
The second multiplexer network is made up of multiplexers Ml'' through Mn~l~ and is used for selectively altering the connection of the north edge cells to either the data input~output unit 20 or to another cell in the array. An array may be made up of several interlinked chips, éach supplying its own section or subarray of the main array. The second multiplexer network provides a capability of changing the virtual configuration of the array.
All of the multiplexers are responsive to ~he controller 30 through connections which have been omitted for clarity. It should be noted that the communicatisn paths provided by the multiplexers are bidirectional, so - . . . .. ~. , ::.,: .- .. -.. : ~ . . ... .
. ~ : . : - .: . , . : ....... .. ... . .
~ . . ,.- : : - ~ . : ~ :: . . :, ~ ~ : : - : .: . . . - . : : : :: .::
that the multiplexers should be regarded as multiplexer/demultiplexers. It should also be noted that the term multiplexer is used herein generically to encompass ~ny device for performing the functions ascribed to these elements, i.e., line selection based on a control input. Thus, simple switches may be multiplexers to the extent they are provided to perform such a function.
Figure 2 is a more detailed rendering of an array 10 according to the present invention, in this case an 8 x 16 array of 128 identical cells Pl,l P8,16 P
: equipment such as controller 30 and connections thereto ; have been omitted far clarity, but it will be understood that such connections are present. The embodiment of Figure 2 has a multiplexer bypass network with members arranged between selected cells thus bypassing intervening cells. As can be seen, it is not necessary that there be a multiplexer between every cell and its nearest neighbors in order to bypass entire rows, columns, or blocks and so, in effect, alter the virtual dimensions of the array to suit a particular application.
; In the embodiment shown, there is one set of eight multiplexers per column and one set of four per row.
Thus, in general, it is necessary only to have half as many multiplexers as cells in any row or column. Of course, lf there is an alternate communication path, .
. . ; j ;
(such as the CM path described below) it may be desixable to provide that path with a multiplexer ne~work as well.
When cell bypass is invoked, cell-to-cell communication is based on the value of a latched bypass S control word which continuously drives the row and column multiplexer select lines. If it were desired to bypass the first column, for example, the westernmost I/O pins - become logically connected to the west port of the cells in the second column.
Row and column bypass can be separately controlled so that all possible virtual array sizes, such as 7 x 12, 4 x 10, 1 x 16, or even 1 x 1, can be obtained. The embodiment of Figure 2, however, is designed so that it is not possible to bypass the array lO entirely. The ; 15 cells in the easternmost column are always connected to the east I/0 pins so that data must always pass through the easternmost column of cells. Similarly, the cells in the southernmost row are always connected to the south I/0 pins so that data must always pass through the southernmost row of cells. There is therefore a minimum 1 x l configuration.
The benefits of the cell bypass network are well exemplified in operations such as parallel data summation, also known as an array sum; An array sum is a - 25 summation of the same data type variable in each cell throuyhout the entire array. With programmable bypass, :
this summation may be achieved using the ~ollowing algorithm:
(1) program the cell bypass network to ~ull 8 x 16 configuration;
(2) command each cell to shift its variable to its ~ eastern neighbor;
(3) command each cell to add the received variable . to its resident value (which results in cells in columns 2, 4, 6, and 8 containing a non-overlapping set of sums);
(4) program the cell bypass network ~o a 4 x 16 configuration logically interconnecting columns 2, 4, 6, and 8;
(5) repeat steps t2) and (3) (which results in cells in columns 4 and 8 containing a non-Qverlapping set of sums);
(6) program the cell bypass network to a 2 x l~
: con~iguration logically interconnecting cells in column 4 to cells in column 8:
: (7) repeat step 5 (which results in cells in column 8 containing the sums for the entire array from west to east); and ; (8) repeat the algorithm north to south to sum ,; cells in column 8, halving the row configuration a~ter each shift and add until the entire array sum resides in the cell at the southeastern corner of the array.
.
Without the cell bypass network, many additional clock cycles would be required to shift the da~a through ; the non-participating cells to the ultimate destination when performing the partial array summations. The cell - 5 bypass network allows direct cell communication for a minimal overhead of 2 cycles for array reconfiguration.
The other multiplexer network comprises bidirectional semiconductor switches or multiplexers Mol' through M7 " on the eight north edge cells. These allow selection of either a "normal" north data path (to data input/output unit 20) or an "alternate" north data path (to another cell on the same or a different chip). The array 10 is thus no longer limited to north, south, east, west connectivity. For example, consider a 32 x 32 array made up of 4 x 4 subarrays of 64 cells each, with the north edge cells of each subarray being provided with a switchable north alternate connection as shown in Figure 3(a). Each subarray may, for example, be on a separate semiconductor chip. By app~opriate interconnection of north and north alternate inputs, the array can be reconfigured as two 16 x 32 arrays (Figure 3(b)) or as one 1024 x 1 array (Figure 3(c)). Thus, the second multiplexer network has great utility in varying connection o~ multiple-chip arrays.
The north alternate connector may also be connected to a memory device to extend logically the memory ~ , ,: " . . - . -: : .
. . .
capability of each cell and to allow data operands to be rolled in and rolled out as necessary. It also provides - a port which can be connected to an I/0 device or other processing cells in the array to allow for corner turn operations.
A preferred embodiment according to the present invention may also include one or mora enhancements in the architecture of the individual cells. As can be seen in Figure 4, the cell may include a group of registers, namely, a north-south (NS) register 40, an east-west (EW) register 50, a carry (C) register 60, and a communication (CM) register 70. It also includes a ~roup of corresponding multiplexers, namely, an NS multiplexer 45, an EW multiplexer 55, a C multiplexer 65, and a CM
multiplexer 75.
The registers are single bit and provide working storage and communications functions for the cell. The NS and CM registers are sources or destinations for data flowing north or south. Thus, the NS register 40 can communicate bidirectionally with the NS registers of its north and south neighboring cells. The CM register provides I/0 communications to each cell bidirectionally along the north-south axis. The difference between these two is that operation of the CM register 70 does not influence oparation of an arithmetic logic unit (ALU) 110 also incorporated into the cell. As a consequence, the .
.
. ~ . - ,, CM register 70 can perform input/output functions without interfering with the ALU 110 which allows I/O to occur without interfering with normal processing. The EW
registers 50 are the sources and destinations of data flowing east or west. Each multiplexer controls data flow from various sources to its respective register under control signals originating ~rom a master controller such as controller 30 of Figure 1.
The cell may incorporate various additional elements, which will be described below in detail. These include a memory 100, which, as will be described below, is segmented to provide flexibility in data storage and retrieval. It also includes a conditional logic unit (CLU) 120 which, as will be developed below, simplifies and accelerates data processing according to certain ;~
algorithms. The cell may also incorporate AF logic circuitry 130, which includes a single-bit register AF
register 140. This circuitry defines two states, activated and deactivated. Also included i5 Global OR
(GLOR) circuitry 150, which is used in ~orming a signal which is a logical "orll of all active cells in the array.
-~ The signal is the product of "anding" the output of the C
register 60 with that of the AF register 140.
The various ~unctions of these circults will now be described in greater detail.
Arithmetic functions within the cell are performed by the ALU 110 which operates on data presented to it under the control o~ the master controller 30. The inputs to the ALU 110 are the C register 60 and the - 5 outputs from a conditional logic circuit 120 described below, which obtains its data from the NS register 40 and the EW register 50. The ALU llo may send its output ; to any of these registers or to the memory 100.
Specifically, when the ALU llo is a single bit full ~ 1o adder, the SUM output can be written to a memory 100 and ~~ the CARRY and BORROW outputs can be used as inputs to the C register 60. Multi-bit operations such as addition are carried out by adding the operands on a bit-by-bit basis, while propagating the resultant carry into the next higher-order addition.
All multi-bit operands must be stored in the memory 100. Thus, performance of almost all operations is highly dependent on the capabilities of memory 100. For : example, in an architecture'where the memory has only a single port, and so can perform only a single read or write per clock cycle, an n-bit operation would require 2n cycles for reading two operands, n cycles for writing each partial sum, and one cycle for ~riting the final carry value, for a total of 3n+1 clock cycles. Thus, a memory which permits only a single read or a single write ' ~ '~ , . . '' '~ , '' :
,:
; - ~5 -per cycle is a hottleneck for computationally in~ensive algorithms.
Due to the great dependence o~ cell e~iciency on thP memory 100, it is prefexred in the present invention to use a segmented RAM as the memory 100. "Segmented"
herein refers to a virtual or logical division of the RAM
into at least semi-independently operable blocks. In a preferred implementation shown in Figure 5, the memory 100 is 128 bits divided into ~our contiguous blocks BLK0 through BLX3, o~ 32 bits each. Each block is preferably a separate, single-ported memory, that can perform either a single read, or a single write, in one clock cycle. It will be appreciated in the embodiment described, however, that there can be no more than one write to the entire RAM per clock cycle. It will also be appreciated that each blocX may be a multi-ported memory, if such would provide advantages for a given application. Servicing the blocks BLK0 through BLK3 are three read ports, ARAM, BRAM, and BRAM*, and one write port ARAM*~ Two address busses service these ports. The first, AADR or A0 through A6, sources the address for operations on the ARAM and ARAM* ports. The second, BADR or B0-B5, sources the address for operations on the BRAM and BR~M* ports.
Memory address decode logic circuits 160 and 170 use selected address bits, for example, the upper address bits, to select which port is to be connected to which .
block. The decode logic is provided to ensure that only one port is allocated to each block for a given address on the busses. The signal WE* is an internally generated write enable signal.
The provision of 2 address busses, 3 read ports, and 1 write port creates a pseudo multi-port capability.
Addressing options allow for either 1, 2, or 3 reads per clock cycle. The inclusion of simultaneous read/write capability means that n-bit operations such as addition - 10 can now be performed in n~2 clock signals, instead of 3n+1. This is a net improvement of 2.5 to 1 in the case of an 8-bit add.
The memory 100 may be loaded from the C register 60, the CM register 70, an AF register 140, or the SUM output of the ALU 110. The outputs of memory 100 may be loaded into the NS register 40, EW register 50, C register 60, CM register 70, a D register 190, or the AF register 140.
The cell is also naturally very dependent on the capabilities o~ ALU 110. Td enhance the capabilities of this element, it is preferred to include in the cell a conditional logic unit ~CLU) 120, as shown in Figure 6.
The CLU 120 includes an input multiplexer 180 with inputs connected to the output of C multiplexer ~5 and to the output of a D register 190. The input multiplexer 180 is controlled by an appropriate control signal. Its output is connected to the input of the D register 190. The :
. ...
~ output of the D register 190 is also connected to a first logic gate 200 and a second logic gate 210. In the embodiment of Figure 6, these are AND gat~s. One input of the first logic gate 200 is connected to the NS
register 40; the other input is connected to the D
register 190 and, in the embodiment of Figure 6, is an inverting input. one input of the second logic 210 is . also connected to the input of D register 190. The other input is connected to the EW register 50.
.The CLU 120 also includes a first output multiplexer 220 and a second output multiplexer 230. The inputs for tha first multiplexer 220 are the outputs of the NS
; register 40 and the first logic gate 200. The inputs for second output multiplexer 230 are the outputs of the EW
register 50 and the second logic gate 2100 :..................... The CLU 120 allows both conditional arithmetic ~;
operation and conditional selection operation on the operands of the NS register 40 and the EW register 50 based on the contents of thé D register 190. This leads . .
to gr~at simplification of operations such as multiplication. For example, without conditional logic, the algorithm for multiplication of two multi-bit numbers involved a reiterative process of "anding'i the first bit - of the multiplier with the multiplicand to form a first partial sum, and then 'landing" the second hit of the multiplier with the multiplicand and adding the result to the first partial sum to form a second partial sum, continuing for each successive bit until the product is formed.
The following example ls indicative, with the usual right justification on bit placement:
101 (Multiplicand) 011 (Multiplier) OOolol (lst Partial Sum) + 101 lo 001111 (2nd Partial Sum) + 000 001111 (Product) The CLU 120 eliminates the requirement for "anding"
by providing a conditional summation capability. This allows the partial sum to be conditionally added to the multiplicand based on the present multiplier-bit value.
If the multiplier bit is o, then the next partial sum is the present partial sum since O*x=0 I~ th~ multiplier bit is 1, the next partial sum is the sum of the present sum and the multiplicand. Right justification in bit placement is accomplished by incrementing the starting address of the partial sum dperand by one after eac~
iteration.
In the context of the device shown in Figure 6, multiplication would be carried out by loading a zero into the NS register 40, the multiplicand into the EW
register 60, and the least significant multiplier bit into the D register 190. The CLU 120 then selects the multiplicand if the contents of the D register 190 is - ~ . . .
~ -- 19 --~ . The partial sum is then loaded into the NS
register 40, the multiplicand into the EW register 60, and the next significant multiplier bit into the D
register 190. The multiplicand is added to the partial sum if the contents of the D register 190 is "1". The - process is repeated as necessary to derive the final product.
The cell also preferably includes "Activity Flag"
(AF) logic or circuitry 130. This circuitry is used to place the cell in one of two states - activated or deactivated. An activated cell participates in algorithm execution. A deactivated cell does not. The state is determined by the contents of the AF register 140. For example, a set register may indicate activated cell, so that a reset register would then indicate a deactivated cell.
The contents of the AF register 140 may be generated either internally, for example, through the C multiplexer 65 as shown, as a result of internal algorithm execution.
This would provide an indication of status of internal algorithm execution. Alternatively, however, the :; contents of the AF register 140 may originate externally and be directed to the cell as a result of decoding by an AF logic unit 240. The AF logic unit may decode an address appearing, for example, on address bus AADR as shown. This permits a row, column, or individual cell to ;~ :
!
".. ' ~'' .
be uniquely selected. The AF logic unit 240 may also be responsive to two select lines X and Y. These lines are used in a multi-chip array to select a particular chip, row of chips, or column of chips.
The AF logic unit 240 generates a value (cell - enable) applied to an AF multiplexer 250 to in turn determine the value of the AF register 140 if that value is to be determined externally.
As an example of how such a system would work in practice, consider the system of Figure 4 where the AF
logic unit 240 is responsive to the AADR address bus, to control signals on two lines of the BADR address bus, and to control signals designated C17 and C18. ~he signals on the AADR address bus select the row, column, or individual cell. The X and Y select lines select chips in a multi-chip array. The control signals BADR1, BADR0, C17, and C18 can then be used to select whether the value ' of the AF register 140 at the address indicated is internally derived, whether~the cell (or the row or column containing it) should be externally activated or ~ deactivated. When a cell is active, its "Globall' output - contributes to the overall GLOR output signal and writes to the cellls memory 100 are enabled. A deactivated cell's "Global" output cannot contribute to the overall GLOR output signal and writes to the cell's memory 100 are inhibited.
.: . . . . .
The AF circuitry 130 address decoding logic thus provides a means to obtain information concerning algorithm execution by a cell through use of the GLOR
output signal. For example, if an array sum as previously described were being carried out, the final sum could be obtained through the GLOR output if all other cells except that at the southeastern corner were - inactive.
As another example of the usefulness of the GLOR
- 10 output in combination with thP AF logic circuitry 130, ; consider the performance of a content addressable memory function. After all cells concurrently search, the combination of GLOR with the AF logic circuitry could be used to determine not only whether any cell uncovered a ; 15 match, but also the location of the cell.
Figure 7 depicts schematically a concrete ; application for a parallel data processor according to the present invention, as a component of a infrared image processing system. Such a ~ystem typically requires a resolution of 512 x 512 pixels per frame, 8-bits per pixel, and a frame update rate o~ 30Hz. With these characteristics, a real-time pixel processor must maintain an average throughput rate of 262,144 pixels every 33.33 ms. The parallel processor of the present invention is ideally suited to perform a two-dimensional data processing problem of this type. The theoretical .~
ideal would be one processor per pixel, or, in other words, a 512 x 512 array of cells. At a clock rate of 20 MHz, this would permit execution of nearly 666,666 instructions every 33.33 m~. Usually, however, a smaller array is used, with the image being processed as a series -- of overlapping sub-images.
In addition to the array 10 and ~rray controller 30, the system includes an IR sensor 260, a sensor interface/control 270, and a host CPU 280. A data input buffer/formatter 290 and a data output buffer/formatter 300 together perform the functions ascribed to data input/output circuit 20 of Figure l.
The IR sensor 260 and sensor interface 270 are front-end modules common to most applications of this :~ 15 type. The IR energy received by IR sensor 260 is - converted to digital data by sensor interface 270. The digital data must then be transposed from a data word stream to a data bit plane format to enable transfer to the array 10 for processing. A bit plane is a two-dimensional plane (i.e., array or matrix) of data bits of common significance, of the same size as thP array lO.
This transposition, called pixel to plane conversion, is carried out by input buffer/formatter 290. The output buffer/formatter 300 performs the inverse operation, to ' 25 make the output comprehensible to the host CPU 280~ The array controller 30 coordinates program execution with . . . .
the bit-plane I/O. It is similar in design to controllers found in most bit-slice systems.
Other system embellishments are possible. For example, logic can be incorporated to selectively enable, S disable or modi~y functions to make ~he cells or array compatible with other parallel data processors. Such a scheme is implemented in the embodiment of Figure 4 through t~he provision of a series of multiplexers 310 and gates 320 connected as shown to operate in response to, for example, a "mode" command, to disable the segmented RAM and AF logic enhancements, or at ~east to prevent those enhancements ~rom interfering with cell operation.
The invention has been described above in terms of specific embodiments merely for the sake of elucidation.
No statement above is intended to imply that the above embodiments are the only fashion in which the invention may be embodied or practiced~ and no statement above should be so construed. To the contrary, it will be readily apparent to one of ordinary skill in the art that it is possible to conceive of many embodiments not described above which nevertheless embody the principles and teachings of the invention. The invention should therefore not be limited to what is described above, but instead should be regarded as being fully commensurate in scope with the following claims.
: con~iguration logically interconnecting cells in column 4 to cells in column 8:
: (7) repeat step 5 (which results in cells in column 8 containing the sums for the entire array from west to east); and ; (8) repeat the algorithm north to south to sum ,; cells in column 8, halving the row configuration a~ter each shift and add until the entire array sum resides in the cell at the southeastern corner of the array.
.
Without the cell bypass network, many additional clock cycles would be required to shift the da~a through ; the non-participating cells to the ultimate destination when performing the partial array summations. The cell - 5 bypass network allows direct cell communication for a minimal overhead of 2 cycles for array reconfiguration.
The other multiplexer network comprises bidirectional semiconductor switches or multiplexers Mol' through M7 " on the eight north edge cells. These allow selection of either a "normal" north data path (to data input/output unit 20) or an "alternate" north data path (to another cell on the same or a different chip). The array 10 is thus no longer limited to north, south, east, west connectivity. For example, consider a 32 x 32 array made up of 4 x 4 subarrays of 64 cells each, with the north edge cells of each subarray being provided with a switchable north alternate connection as shown in Figure 3(a). Each subarray may, for example, be on a separate semiconductor chip. By app~opriate interconnection of north and north alternate inputs, the array can be reconfigured as two 16 x 32 arrays (Figure 3(b)) or as one 1024 x 1 array (Figure 3(c)). Thus, the second multiplexer network has great utility in varying connection o~ multiple-chip arrays.
The north alternate connector may also be connected to a memory device to extend logically the memory ~ , ,: " . . - . -: : .
. . .
capability of each cell and to allow data operands to be rolled in and rolled out as necessary. It also provides - a port which can be connected to an I/0 device or other processing cells in the array to allow for corner turn operations.
A preferred embodiment according to the present invention may also include one or mora enhancements in the architecture of the individual cells. As can be seen in Figure 4, the cell may include a group of registers, namely, a north-south (NS) register 40, an east-west (EW) register 50, a carry (C) register 60, and a communication (CM) register 70. It also includes a ~roup of corresponding multiplexers, namely, an NS multiplexer 45, an EW multiplexer 55, a C multiplexer 65, and a CM
multiplexer 75.
The registers are single bit and provide working storage and communications functions for the cell. The NS and CM registers are sources or destinations for data flowing north or south. Thus, the NS register 40 can communicate bidirectionally with the NS registers of its north and south neighboring cells. The CM register provides I/0 communications to each cell bidirectionally along the north-south axis. The difference between these two is that operation of the CM register 70 does not influence oparation of an arithmetic logic unit (ALU) 110 also incorporated into the cell. As a consequence, the .
.
. ~ . - ,, CM register 70 can perform input/output functions without interfering with the ALU 110 which allows I/O to occur without interfering with normal processing. The EW
registers 50 are the sources and destinations of data flowing east or west. Each multiplexer controls data flow from various sources to its respective register under control signals originating ~rom a master controller such as controller 30 of Figure 1.
The cell may incorporate various additional elements, which will be described below in detail. These include a memory 100, which, as will be described below, is segmented to provide flexibility in data storage and retrieval. It also includes a conditional logic unit (CLU) 120 which, as will be developed below, simplifies and accelerates data processing according to certain ;~
algorithms. The cell may also incorporate AF logic circuitry 130, which includes a single-bit register AF
register 140. This circuitry defines two states, activated and deactivated. Also included i5 Global OR
(GLOR) circuitry 150, which is used in ~orming a signal which is a logical "orll of all active cells in the array.
-~ The signal is the product of "anding" the output of the C
register 60 with that of the AF register 140.
The various ~unctions of these circults will now be described in greater detail.
Arithmetic functions within the cell are performed by the ALU 110 which operates on data presented to it under the control o~ the master controller 30. The inputs to the ALU 110 are the C register 60 and the - 5 outputs from a conditional logic circuit 120 described below, which obtains its data from the NS register 40 and the EW register 50. The ALU llo may send its output ; to any of these registers or to the memory 100.
Specifically, when the ALU llo is a single bit full ~ 1o adder, the SUM output can be written to a memory 100 and ~~ the CARRY and BORROW outputs can be used as inputs to the C register 60. Multi-bit operations such as addition are carried out by adding the operands on a bit-by-bit basis, while propagating the resultant carry into the next higher-order addition.
All multi-bit operands must be stored in the memory 100. Thus, performance of almost all operations is highly dependent on the capabilities of memory 100. For : example, in an architecture'where the memory has only a single port, and so can perform only a single read or write per clock cycle, an n-bit operation would require 2n cycles for reading two operands, n cycles for writing each partial sum, and one cycle for ~riting the final carry value, for a total of 3n+1 clock cycles. Thus, a memory which permits only a single read or a single write ' ~ '~ , . . '' '~ , '' :
,:
; - ~5 -per cycle is a hottleneck for computationally in~ensive algorithms.
Due to the great dependence o~ cell e~iciency on thP memory 100, it is prefexred in the present invention to use a segmented RAM as the memory 100. "Segmented"
herein refers to a virtual or logical division of the RAM
into at least semi-independently operable blocks. In a preferred implementation shown in Figure 5, the memory 100 is 128 bits divided into ~our contiguous blocks BLK0 through BLX3, o~ 32 bits each. Each block is preferably a separate, single-ported memory, that can perform either a single read, or a single write, in one clock cycle. It will be appreciated in the embodiment described, however, that there can be no more than one write to the entire RAM per clock cycle. It will also be appreciated that each blocX may be a multi-ported memory, if such would provide advantages for a given application. Servicing the blocks BLK0 through BLK3 are three read ports, ARAM, BRAM, and BRAM*, and one write port ARAM*~ Two address busses service these ports. The first, AADR or A0 through A6, sources the address for operations on the ARAM and ARAM* ports. The second, BADR or B0-B5, sources the address for operations on the BRAM and BR~M* ports.
Memory address decode logic circuits 160 and 170 use selected address bits, for example, the upper address bits, to select which port is to be connected to which .
block. The decode logic is provided to ensure that only one port is allocated to each block for a given address on the busses. The signal WE* is an internally generated write enable signal.
The provision of 2 address busses, 3 read ports, and 1 write port creates a pseudo multi-port capability.
Addressing options allow for either 1, 2, or 3 reads per clock cycle. The inclusion of simultaneous read/write capability means that n-bit operations such as addition - 10 can now be performed in n~2 clock signals, instead of 3n+1. This is a net improvement of 2.5 to 1 in the case of an 8-bit add.
The memory 100 may be loaded from the C register 60, the CM register 70, an AF register 140, or the SUM output of the ALU 110. The outputs of memory 100 may be loaded into the NS register 40, EW register 50, C register 60, CM register 70, a D register 190, or the AF register 140.
The cell is also naturally very dependent on the capabilities o~ ALU 110. Td enhance the capabilities of this element, it is preferred to include in the cell a conditional logic unit ~CLU) 120, as shown in Figure 6.
The CLU 120 includes an input multiplexer 180 with inputs connected to the output of C multiplexer ~5 and to the output of a D register 190. The input multiplexer 180 is controlled by an appropriate control signal. Its output is connected to the input of the D register 190. The :
. ...
~ output of the D register 190 is also connected to a first logic gate 200 and a second logic gate 210. In the embodiment of Figure 6, these are AND gat~s. One input of the first logic gate 200 is connected to the NS
register 40; the other input is connected to the D
register 190 and, in the embodiment of Figure 6, is an inverting input. one input of the second logic 210 is . also connected to the input of D register 190. The other input is connected to the EW register 50.
.The CLU 120 also includes a first output multiplexer 220 and a second output multiplexer 230. The inputs for tha first multiplexer 220 are the outputs of the NS
; register 40 and the first logic gate 200. The inputs for second output multiplexer 230 are the outputs of the EW
register 50 and the second logic gate 2100 :..................... The CLU 120 allows both conditional arithmetic ~;
operation and conditional selection operation on the operands of the NS register 40 and the EW register 50 based on the contents of thé D register 190. This leads . .
to gr~at simplification of operations such as multiplication. For example, without conditional logic, the algorithm for multiplication of two multi-bit numbers involved a reiterative process of "anding'i the first bit - of the multiplier with the multiplicand to form a first partial sum, and then 'landing" the second hit of the multiplier with the multiplicand and adding the result to the first partial sum to form a second partial sum, continuing for each successive bit until the product is formed.
The following example ls indicative, with the usual right justification on bit placement:
101 (Multiplicand) 011 (Multiplier) OOolol (lst Partial Sum) + 101 lo 001111 (2nd Partial Sum) + 000 001111 (Product) The CLU 120 eliminates the requirement for "anding"
by providing a conditional summation capability. This allows the partial sum to be conditionally added to the multiplicand based on the present multiplier-bit value.
If the multiplier bit is o, then the next partial sum is the present partial sum since O*x=0 I~ th~ multiplier bit is 1, the next partial sum is the sum of the present sum and the multiplicand. Right justification in bit placement is accomplished by incrementing the starting address of the partial sum dperand by one after eac~
iteration.
In the context of the device shown in Figure 6, multiplication would be carried out by loading a zero into the NS register 40, the multiplicand into the EW
register 60, and the least significant multiplier bit into the D register 190. The CLU 120 then selects the multiplicand if the contents of the D register 190 is - ~ . . .
~ -- 19 --~ . The partial sum is then loaded into the NS
register 40, the multiplicand into the EW register 60, and the next significant multiplier bit into the D
register 190. The multiplicand is added to the partial sum if the contents of the D register 190 is "1". The - process is repeated as necessary to derive the final product.
The cell also preferably includes "Activity Flag"
(AF) logic or circuitry 130. This circuitry is used to place the cell in one of two states - activated or deactivated. An activated cell participates in algorithm execution. A deactivated cell does not. The state is determined by the contents of the AF register 140. For example, a set register may indicate activated cell, so that a reset register would then indicate a deactivated cell.
The contents of the AF register 140 may be generated either internally, for example, through the C multiplexer 65 as shown, as a result of internal algorithm execution.
This would provide an indication of status of internal algorithm execution. Alternatively, however, the :; contents of the AF register 140 may originate externally and be directed to the cell as a result of decoding by an AF logic unit 240. The AF logic unit may decode an address appearing, for example, on address bus AADR as shown. This permits a row, column, or individual cell to ;~ :
!
".. ' ~'' .
be uniquely selected. The AF logic unit 240 may also be responsive to two select lines X and Y. These lines are used in a multi-chip array to select a particular chip, row of chips, or column of chips.
The AF logic unit 240 generates a value (cell - enable) applied to an AF multiplexer 250 to in turn determine the value of the AF register 140 if that value is to be determined externally.
As an example of how such a system would work in practice, consider the system of Figure 4 where the AF
logic unit 240 is responsive to the AADR address bus, to control signals on two lines of the BADR address bus, and to control signals designated C17 and C18. ~he signals on the AADR address bus select the row, column, or individual cell. The X and Y select lines select chips in a multi-chip array. The control signals BADR1, BADR0, C17, and C18 can then be used to select whether the value ' of the AF register 140 at the address indicated is internally derived, whether~the cell (or the row or column containing it) should be externally activated or ~ deactivated. When a cell is active, its "Globall' output - contributes to the overall GLOR output signal and writes to the cellls memory 100 are enabled. A deactivated cell's "Global" output cannot contribute to the overall GLOR output signal and writes to the cell's memory 100 are inhibited.
.: . . . . .
The AF circuitry 130 address decoding logic thus provides a means to obtain information concerning algorithm execution by a cell through use of the GLOR
output signal. For example, if an array sum as previously described were being carried out, the final sum could be obtained through the GLOR output if all other cells except that at the southeastern corner were - inactive.
As another example of the usefulness of the GLOR
- 10 output in combination with thP AF logic circuitry 130, ; consider the performance of a content addressable memory function. After all cells concurrently search, the combination of GLOR with the AF logic circuitry could be used to determine not only whether any cell uncovered a ; 15 match, but also the location of the cell.
Figure 7 depicts schematically a concrete ; application for a parallel data processor according to the present invention, as a component of a infrared image processing system. Such a ~ystem typically requires a resolution of 512 x 512 pixels per frame, 8-bits per pixel, and a frame update rate o~ 30Hz. With these characteristics, a real-time pixel processor must maintain an average throughput rate of 262,144 pixels every 33.33 ms. The parallel processor of the present invention is ideally suited to perform a two-dimensional data processing problem of this type. The theoretical .~
ideal would be one processor per pixel, or, in other words, a 512 x 512 array of cells. At a clock rate of 20 MHz, this would permit execution of nearly 666,666 instructions every 33.33 m~. Usually, however, a smaller array is used, with the image being processed as a series -- of overlapping sub-images.
In addition to the array 10 and ~rray controller 30, the system includes an IR sensor 260, a sensor interface/control 270, and a host CPU 280. A data input buffer/formatter 290 and a data output buffer/formatter 300 together perform the functions ascribed to data input/output circuit 20 of Figure l.
The IR sensor 260 and sensor interface 270 are front-end modules common to most applications of this :~ 15 type. The IR energy received by IR sensor 260 is - converted to digital data by sensor interface 270. The digital data must then be transposed from a data word stream to a data bit plane format to enable transfer to the array 10 for processing. A bit plane is a two-dimensional plane (i.e., array or matrix) of data bits of common significance, of the same size as thP array lO.
This transposition, called pixel to plane conversion, is carried out by input buffer/formatter 290. The output buffer/formatter 300 performs the inverse operation, to ' 25 make the output comprehensible to the host CPU 280~ The array controller 30 coordinates program execution with . . . .
the bit-plane I/O. It is similar in design to controllers found in most bit-slice systems.
Other system embellishments are possible. For example, logic can be incorporated to selectively enable, S disable or modi~y functions to make ~he cells or array compatible with other parallel data processors. Such a scheme is implemented in the embodiment of Figure 4 through t~he provision of a series of multiplexers 310 and gates 320 connected as shown to operate in response to, for example, a "mode" command, to disable the segmented RAM and AF logic enhancements, or at ~east to prevent those enhancements ~rom interfering with cell operation.
The invention has been described above in terms of specific embodiments merely for the sake of elucidation.
No statement above is intended to imply that the above embodiments are the only fashion in which the invention may be embodied or practiced~ and no statement above should be so construed. To the contrary, it will be readily apparent to one of ordinary skill in the art that it is possible to conceive of many embodiments not described above which nevertheless embody the principles and teachings of the invention. The invention should therefore not be limited to what is described above, but instead should be regarded as being fully commensurate in scope with the following claims.
Claims (42)
1. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
cell logic means for performing logical operations; and a memory segmented into at least two blocks, said memory coupled to said cell logic means such that said cell logic means performs more than one read and a write operation per clock cycle from/to said memory when directed to do so by said control signals.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
cell logic means for performing logical operations; and a memory segmented into at least two blocks, said memory coupled to said cell logic means such that said cell logic means performs more than one read and a write operation per clock cycle from/to said memory when directed to do so by said control signals.
2. The apparatus of claim 1, wherein said memory is a random access memory.
3. The apparatus of claim 2 wherein said random access memory is divided into four blocks and comprises at least two address busses, at least three read ports, and at least one write port.
4. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said cells including:
cell logic means for performing logical operations; and a memory having at least three read ports and at least one write port, said memory coupled to said cell logic means so that said cell logic means performs more than one read and a write operation per clock cycle from/to said memory when directed to do so by said control signals.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said cells including:
cell logic means for performing logical operations; and a memory having at least three read ports and at least one write port, said memory coupled to said cell logic means so that said cell logic means performs more than one read and a write operation per clock cycle from/to said memory when directed to do so by said control signals.
5. The apparatus of claim 4, wherein said memory is a random access memory.
6. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for performing logical and arithmetic operations, and a plurality of memory means, connected to said controller and said first means, for selectively retaining digital data, at least one of said memory means being segmented into at least two blocks, said at least one segmented memory means being coupled to said first means and to a remaining memory means from said plurality of memory means such that said first means and said remaining memory means perform more than one read and a write operation per clock cycle from/to said at least one segmented memory means when directed to do so by the control signals.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for performing logical and arithmetic operations, and a plurality of memory means, connected to said controller and said first means, for selectively retaining digital data, at least one of said memory means being segmented into at least two blocks, said at least one segmented memory means being coupled to said first means and to a remaining memory means from said plurality of memory means such that said first means and said remaining memory means perform more than one read and a write operation per clock cycle from/to said at least one segmented memory means when directed to do so by the control signals.
7. The apparatus of claim 6, wherein said segmented memory means comprises a random access memory.
8. The apparatus of claim 7, wherein said random access memory is divided into four blocks and comprises at least two address busses, at least three read ports, and at least one write port.
9. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for performing logical and arithmetic operations, and a plurality of memory means, connected to said input means and said first means, for selectively retaining digital data, at least one of said memory means having at least three read ports, said at least one memory means being coupled to said first means and to a remaining memory means from said plurality of memory means such that said first means and said remaining memory means perform more than one read and a write operation per clock cycle from/to said at least one memory means when directed to do so by the control signals.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for performing logical and arithmetic operations, and a plurality of memory means, connected to said input means and said first means, for selectively retaining digital data, at least one of said memory means having at least three read ports, said at least one memory means being coupled to said first means and to a remaining memory means from said plurality of memory means such that said first means and said remaining memory means perform more than one read and a write operation per clock cycle from/to said at least one memory means when directed to do so by the control signals.
10. The apparatus of claim 9, wherein said at least one memory means comprises a random access memory.
11. The apparatus of claim 10, wherein said random access memory has at least three read ports and at least one write port.
12. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for carrying out logical algorithms;
a register for receiving and storing data from said at least one neighboring cell; and activity flag means, responsive to said control signals, for selectively generating an indication of whether the said first means is currently engaged in carrying out said algorithms, wherein when said first means is not currently engaged in carrying out said algorithms, said register receives and stores data from said at least one neighboring cell when directed to do so by said control signals.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for carrying out logical algorithms;
a register for receiving and storing data from said at least one neighboring cell; and activity flag means, responsive to said control signals, for selectively generating an indication of whether the said first means is currently engaged in carrying out said algorithms, wherein when said first means is not currently engaged in carrying out said algorithms, said register receives and stores data from said at least one neighboring cell when directed to do so by said control signals.
13. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for carrying out logical algorithms; and second means, responsive to said control signals, for selectively generating an indication of whether the said first means is currently engaged in carrying out said algorithms; and third means for generating a signal to be logically combined with similar signals from other cells to generate a global signal, and wherein said second means selectively disables said third means so that said cell does not contribute to said global signal.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for carrying out logical algorithms; and second means, responsive to said control signals, for selectively generating an indication of whether the said first means is currently engaged in carrying out said algorithms; and third means for generating a signal to be logically combined with similar signals from other cells to generate a global signal, and wherein said second means selectively disables said third means so that said cell does not contribute to said global signal.
14. The apparatus of claim 12, wherein each cell includes memory means for selectively retaining and retrieving digital data, and wherein said second means selectively inhibits storage in said memory means.
15. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for carrying out logical algorithms; and second means, responsive to said control signals, for selectively generating an indication of whether the said first means is currently engaged in carrying out said algorithms, wherein said second means is also responsive to said first means, and generates said indication in response to either said control signals, or a signal generated by said first means.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
first means, including a full adder/subtractor, for carrying out logical algorithms; and second means, responsive to said control signals, for selectively generating an indication of whether the said first means is currently engaged in carrying out said algorithms, wherein said second means is also responsive to said first means, and generates said indication in response to either said control signals, or a signal generated by said first means.
16. The apparatus of claim 12, wherein said second means is individually addressable by said control signals.
17. In a parallel data processor, an apparatus comprising:
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
at least two data registers for storing digital data;
means for storing a control value;
first means, connected to said data registers and to said control value means, for conditionally carrying out preliminary operations on digital data from said data registers in response to an output from said control value storing means; and second means, responsive to said first means and including a full adder/subtractor, for carrying out arithmetic and logical functions on digital data from said first means.
input means for receiving control signals; and a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals, each of said identical cells including:
at least two data registers for storing digital data;
means for storing a control value;
first means, connected to said data registers and to said control value means, for conditionally carrying out preliminary operations on digital data from said data registers in response to an output from said control value storing means; and second means, responsive to said first means and including a full adder/subtractor, for carrying out arithmetic and logical functions on digital data from said first means.
18. A parallel data processor comprising:
a controller for generating control signals; and a plurality of identical cells, each of said cells being connected to at least one neighboring cell and responsive to said controller for processing data in accordance with said control signals, and including:
at least two data registers for storing digital data, and first means, connected to said data registers, for conditionally carrying out preliminary operations on digital data from said data registers, and second means, responsive to said first means and including a full adder/subtractor, for carrying out arithmetic and logical functions on digital data from said first means, wherein said at least two data registers comprise a first data register and a second data register, and wherein said first means includes:
a first logic gate having a first input terminal connected to said first data register;
a second logic gate having a first terminal connected to said second data register;
first selecting means, responsive to said controller, and connected to an output terminal of said first data register and said first logic gate, for selecting one of the output the first data register and the output of the first logic gate;
second selecting means, responsive to said controller, and connected to said second data register and to an output terminal of said second logic gate, for selecting one of the output of the second data register and the output of said second logic gate; and means, connected to a second input terminal of said first logic gate and to a second input terminal of said second logic gate, for generating a control value.
a controller for generating control signals; and a plurality of identical cells, each of said cells being connected to at least one neighboring cell and responsive to said controller for processing data in accordance with said control signals, and including:
at least two data registers for storing digital data, and first means, connected to said data registers, for conditionally carrying out preliminary operations on digital data from said data registers, and second means, responsive to said first means and including a full adder/subtractor, for carrying out arithmetic and logical functions on digital data from said first means, wherein said at least two data registers comprise a first data register and a second data register, and wherein said first means includes:
a first logic gate having a first input terminal connected to said first data register;
a second logic gate having a first terminal connected to said second data register;
first selecting means, responsive to said controller, and connected to an output terminal of said first data register and said first logic gate, for selecting one of the output the first data register and the output of the first logic gate;
second selecting means, responsive to said controller, and connected to said second data register and to an output terminal of said second logic gate, for selecting one of the output of the second data register and the output of said second logic gate; and means, connected to a second input terminal of said first logic gate and to a second input terminal of said second logic gate, for generating a control value.
19. A parallel data processor as claimed in claim 18, wherein said control value generating means comprises:
a third data register connected to said second input terminal of said first logic gate and said second input terminal of said second logic gate; and third selecting means, responsive to said controller for selecting as an output to said third data register an externally-applied signal or the contents of said third data register.
a third data register connected to said second input terminal of said first logic gate and said second input terminal of said second logic gate; and third selecting means, responsive to said controller for selecting as an output to said third data register an externally-applied signal or the contents of said third data register.
20. A parallel data processor as claimed in claim 19, wherein said second means comprises a full adder having first and second data input terminals, and wherein an output of said first selecting means is connected to said first data input terminal, and an output of said second selecting means is connected to said second data input terminal.
21. In a parallel data processor, an apparatus comprising:
input means for receiving control signals;
an n x m array of identical processing cells, n and m being positive integers, each of said identical processing cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals; and n selecting means, respectively connected and responsive to n edge cells of said n x m array, for selectively connecting an associated one of said edge cells to one of at least two external I/O devices in response to said control signals.
input means for receiving control signals;
an n x m array of identical processing cells, n and m being positive integers, each of said identical processing cells being connected to at least one neighboring cell and responsive to said control signals for processing data in accordance with said control signals; and n selecting means, respectively connected and responsive to n edge cells of said n x m array, for selectively connecting an associated one of said edge cells to one of at least two external I/O devices in response to said control signals.
22. In a parallel data processor, an apparatus comprising:
input means for receiving control signals;
a plurality of identical cells each responsive to said control signals for processing data in accordance with said control signals and selectively interconnected to other cells of said plurality to form a cell array; and means for enhancing computational ability of said apparatus, including bypass means interconnected between predetermined cells of said array and responsive to said control signals for selectively establishing data paths bypassing portions of said array.
input means for receiving control signals;
a plurality of identical cells each responsive to said control signals for processing data in accordance with said control signals and selectively interconnected to other cells of said plurality to form a cell array; and means for enhancing computational ability of said apparatus, including bypass means interconnected between predetermined cells of said array and responsive to said control signals for selectively establishing data paths bypassing portions of said array.
23. The apparatus of claim 22, wherein said bypass means comprises a plurality of multiplexer/demultiplexers respectively connected between said portions of said array.
24. The apparatus of claim 22, wherein said array is arranged in rows and columns, and wherein said bypass means is arranged for selectively bypassing selected rows and columns.
25. The apparatus of claim 1, wherein the input means and the plurality of identical cells are fabricated as a very large scale integrated circuit chip.
26. The apparatus of claim 25, wherein said memory is a random access memory.
27. The apparatus of claim 26 wherein said random access memory is divided into four blocks and comprises at least two address busses, at least three read ports, and at least one write port.
28. The apparatus of claim 6, wherein the input means and the plurality of identical cells are fabricated as a very large scale integrated circuit chip.
29. The apparatus of claim 28, wherein said segmented memory means comprises a random access memory.
30. The apparatus of claim 29, wherein said random access memory is divided into four blocks and comprises at least two address busses, at least three read ports, and at least one write port.
31. The apparatus of claim 1, further comprising a controller, coupled to said input means, for generating said control signals.
32. The apparatus of claim 4, further comprising a controller, coupled to said input means, for generating said control signals.
33. The apparatus of claim 6, further comprising a controller, coupled to said input means, for generating said control signals.
34. The apparatus of claim 9, further comprising a controller, coupled to said input means, for generating said control signals.
35. The apparatus of claim 12, further comprising a controller, coupled to said input means, for generating said control signals.
36. The apparatus of claim 17, further comprising a controller, coupled to said input means, for generating said control signals.
37. The apparatus of claim 21, further comprising a controller, coupled to said input means, for generating said control signals.
38. The apparatus of claim 22, further comprising a controller, coupled to said input means, for generating said control signals.
39. The apparatus of claim 1, wherein the memory comprises:
a first block of addressable storage cells for storing data associated with a low part of a memory address space, the first block being capable of alternatively performing either a read or a write operation in one clock cycle;
a second block of addressable storage cells for storing data associated with a high part of the memory address space, the second block being capable of alternatively performing either a read or a write operation in one clock cycle;
a data input port;
a write enable input port for receiving a write enable signal;
a first address input port;
a second address input port;
selective coupling means, responsive to a first received address supplied to the first address input port, for coupling the first address input port, the data input port, and the write enable input port to the first block and coupling the second address input port to the second block when the first received address designates a storage address corresponding to the low part of the memory unit address space, the selective coupling means further responsive to the first received address for coupling the first address input port, the data input port, and the write enable input port to the second block and coupling the second address input port to the first block when the first received address designates a storage address corresponding to the high part of the memory unit address space, whereby the first and second blocks are semi-independently operable.
a first block of addressable storage cells for storing data associated with a low part of a memory address space, the first block being capable of alternatively performing either a read or a write operation in one clock cycle;
a second block of addressable storage cells for storing data associated with a high part of the memory address space, the second block being capable of alternatively performing either a read or a write operation in one clock cycle;
a data input port;
a write enable input port for receiving a write enable signal;
a first address input port;
a second address input port;
selective coupling means, responsive to a first received address supplied to the first address input port, for coupling the first address input port, the data input port, and the write enable input port to the first block and coupling the second address input port to the second block when the first received address designates a storage address corresponding to the low part of the memory unit address space, the selective coupling means further responsive to the first received address for coupling the first address input port, the data input port, and the write enable input port to the second block and coupling the second address input port to the first block when the first received address designates a storage address corresponding to the high part of the memory unit address space, whereby the first and second blocks are semi-independently operable.
40. The apparatus of claim 39, wherein the input means and the plurality of identical cells are fabricated as a very large scale integrated circuit chip.
41. The apparatus of claim 6, wherein the at least one segmented memory means comprises:
a first block of addressable storage cells for storing data associated with a low part of a memory address space, the first block being capable of alternatively performing either a read or a write operation in one clock cycle;
a second block of addressable storage cells for storing data associated with a high part of the memory address space, the second block being capable of alternatively performing either a read or a write operation in one clock cycle;
a data input port;
a write enable input port for receiving a write enable signal;
a first address input port;
a second address input port;
selective coupling means, responsive to a first received address supplied to the first address input port, for coupling the first address input port, the data input port, and the write enable input port to the first block and coupling the second address input port to the second block when the first received address designates a storage address corresponding to the low part of the memory unit address space, the selective coupling means further responsive to the first received address for coupling the first address input port, the data input port, and the write enable input port to the second block and coupling the second address input port to the first block when the first received address designates a storage address corresponding to the high part of the memory unit address space, whereby the first and second blocks are semi-independently operable.
a first block of addressable storage cells for storing data associated with a low part of a memory address space, the first block being capable of alternatively performing either a read or a write operation in one clock cycle;
a second block of addressable storage cells for storing data associated with a high part of the memory address space, the second block being capable of alternatively performing either a read or a write operation in one clock cycle;
a data input port;
a write enable input port for receiving a write enable signal;
a first address input port;
a second address input port;
selective coupling means, responsive to a first received address supplied to the first address input port, for coupling the first address input port, the data input port, and the write enable input port to the first block and coupling the second address input port to the second block when the first received address designates a storage address corresponding to the low part of the memory unit address space, the selective coupling means further responsive to the first received address for coupling the first address input port, the data input port, and the write enable input port to the second block and coupling the second address input port to the first block when the first received address designates a storage address corresponding to the high part of the memory unit address space, whereby the first and second blocks are semi-independently operable.
42. The apparatus of claim 41, wherein the input means and the plurality of identical cells are fabricated as a very large scale integrated circuit chip.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US25471888A | 1988-10-07 | 1988-10-07 | |
| US07/254,718 | 1988-10-07 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2000151A1 CA2000151A1 (en) | 1990-04-07 |
| CA2000151C true CA2000151C (en) | 1997-12-09 |
Family
ID=22965319
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002000151A Expired - Lifetime CA2000151C (en) | 1988-10-07 | 1989-10-04 | Parallel data processor |
Country Status (6)
| Country | Link |
|---|---|
| US (1) | US5421019A (en) |
| EP (1) | EP0390907B1 (en) |
| JP (1) | JP2930341B2 (en) |
| CA (1) | CA2000151C (en) |
| DE (1) | DE68926783T2 (en) |
| WO (1) | WO1990004235A1 (en) |
Families Citing this family (82)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| IE920032A1 (en) * | 1991-01-11 | 1992-07-15 | Marconi Gec Ltd | Parallel processing apparatus |
| KR0125623B1 (en) * | 1991-04-09 | 1998-07-01 | 세끼자와 다다시 | Data processor and data processing method |
| US5689719A (en) * | 1991-06-28 | 1997-11-18 | Sanyo Electric O., Ltd. | Parallel computer system including processing elements |
| US5717947A (en) * | 1993-03-31 | 1998-02-10 | Motorola, Inc. | Data processing system and method thereof |
| JPH076146A (en) * | 1993-06-18 | 1995-01-10 | Fujitsu Ltd | Parallel data processing system |
| US6073185A (en) * | 1993-08-27 | 2000-06-06 | Teranex, Inc. | Parallel data processor |
| US5603046A (en) * | 1993-11-02 | 1997-02-11 | Motorola Inc. | Method for complex data movement in a multi-processor data processing system |
| US5742848A (en) * | 1993-11-16 | 1998-04-21 | Microsoft Corp. | System for passing messages between source object and target object utilizing generic code in source object to invoke any member function of target object by executing the same instructions |
| EP0728337B1 (en) * | 1994-09-13 | 2000-05-03 | Teranex, Inc. | Parallel data processor |
| GB2293468B (en) * | 1994-09-21 | 1999-09-29 | Sony Uk Ltd | Data processing systems |
| US5943242A (en) | 1995-11-17 | 1999-08-24 | Pact Gmbh | Dynamically reconfigurable data processing system |
| WO2002029600A2 (en) | 2000-10-06 | 2002-04-11 | Pact Informationstechnologie Gmbh | Cell system with segmented intermediate cell structure |
| US7266725B2 (en) | 2001-09-03 | 2007-09-04 | Pact Xpp Technologies Ag | Method for debugging reconfigurable architectures |
| DE19651075A1 (en) | 1996-12-09 | 1998-06-10 | Pact Inf Tech Gmbh | Unit for processing numerical and logical operations, for use in processors (CPU's), multi-computer systems, data flow processors (DFP's), digital signal processors (DSP's) or the like |
| DE19654595A1 (en) | 1996-12-20 | 1998-07-02 | Pact Inf Tech Gmbh | I0 and memory bus system for DFPs as well as building blocks with two- or multi-dimensional programmable cell structures |
| DE19654593A1 (en) | 1996-12-20 | 1998-07-02 | Pact Inf Tech Gmbh | Reconfiguration procedure for programmable blocks at runtime |
| US6338106B1 (en) | 1996-12-20 | 2002-01-08 | Pact Gmbh | I/O and memory bus system for DFPS and units with two or multi-dimensional programmable cell architectures |
| ATE243390T1 (en) | 1996-12-27 | 2003-07-15 | Pact Inf Tech Gmbh | METHOD FOR INDEPENDENT DYNAMIC LOADING OF DATA FLOW PROCESSORS (DFPS) AND COMPONENTS WITH TWO- OR MULTI-DIMENSIONAL PROGRAMMABLE CELL STRUCTURES (FPGAS, DPGAS, O.L.) |
| DE19654846A1 (en) * | 1996-12-27 | 1998-07-09 | Pact Inf Tech Gmbh | Process for the independent dynamic reloading of data flow processors (DFPs) as well as modules with two- or multi-dimensional programmable cell structures (FPGAs, DPGAs, etc.) |
| DE19704044A1 (en) * | 1997-02-04 | 1998-08-13 | Pact Inf Tech Gmbh | Address generation with systems having programmable modules |
| US6542998B1 (en) | 1997-02-08 | 2003-04-01 | Pact Gmbh | Method of self-synchronization of configurable elements of a programmable module |
| DE19704728A1 (en) | 1997-02-08 | 1998-08-13 | Pact Inf Tech Gmbh | Method for self-synchronization of configurable elements of a programmable module |
| DE19704742A1 (en) | 1997-02-11 | 1998-09-24 | Pact Inf Tech Gmbh | Internal bus system for DFPs, as well as modules with two- or multi-dimensional programmable cell structures, for coping with large amounts of data with high networking effort |
| US8686549B2 (en) | 2001-09-03 | 2014-04-01 | Martin Vorbach | Reconfigurable elements |
| US9092595B2 (en) | 1997-10-08 | 2015-07-28 | Pact Xpp Technologies Ag | Multiprocessor having associated RAM units |
| DE19861088A1 (en) | 1997-12-22 | 2000-02-10 | Pact Inf Tech Gmbh | Repairing integrated circuits by replacing subassemblies with substitutes |
| DE19807872A1 (en) | 1998-02-25 | 1999-08-26 | Pact Inf Tech Gmbh | Method of managing configuration data in data flow processors |
| US6173388B1 (en) | 1998-04-09 | 2001-01-09 | Teranex Inc. | Directly accessing local memories of array processors for improved real-time corner turning processing |
| US6067609A (en) * | 1998-04-09 | 2000-05-23 | Teranex, Inc. | Pattern generation and shift plane operations for a mesh connected computer |
| US6212628B1 (en) | 1998-04-09 | 2001-04-03 | Teranex, Inc. | Mesh connected computer |
| US6185667B1 (en) | 1998-04-09 | 2001-02-06 | Teranex, Inc. | Input/output support for processing in a mesh connected computer |
| FR2778764B1 (en) * | 1998-05-15 | 2001-01-05 | France Etat | METHOD FOR CONTROLLING A PROCESSOR NETWORK |
| US8230411B1 (en) | 1999-06-10 | 2012-07-24 | Martin Vorbach | Method for interleaving a program over a plurality of cells |
| US6532317B2 (en) | 2000-04-17 | 2003-03-11 | Polyoptic Technologies, Inc. | Optical circuit board |
| EP2226732A3 (en) | 2000-06-13 | 2016-04-06 | PACT XPP Technologies AG | Cache hierarchy for a multicore processor |
| US8058899B2 (en) | 2000-10-06 | 2011-11-15 | Martin Vorbach | Logic cell array and bus system |
| US20040015899A1 (en) * | 2000-10-06 | 2004-01-22 | Frank May | Method for processing data |
| US6854117B1 (en) | 2000-10-31 | 2005-02-08 | Caspian Networks, Inc. | Parallel network processor array |
| GB2370381B (en) | 2000-12-19 | 2003-12-24 | Picochip Designs Ltd | Processor architecture |
| GB2370380B (en) | 2000-12-19 | 2003-12-31 | Picochip Designs Ltd | Processor architecture |
| US6990555B2 (en) * | 2001-01-09 | 2006-01-24 | Pact Xpp Technologies Ag | Method of hierarchical caching of configuration data having dataflow processors and modules having two- or multidimensional programmable cell structure (FPGAs, DPGAs, etc.) |
| US9436631B2 (en) | 2001-03-05 | 2016-09-06 | Pact Xpp Technologies Ag | Chip including memory element storing higher level memory data on a page by page basis |
| US9037807B2 (en) | 2001-03-05 | 2015-05-19 | Pact Xpp Technologies Ag | Processor arrangement on a chip including data processing, memory, and interface elements |
| US7444531B2 (en) | 2001-03-05 | 2008-10-28 | Pact Xpp Technologies Ag | Methods and devices for treating and processing data |
| US9552047B2 (en) | 2001-03-05 | 2017-01-24 | Pact Xpp Technologies Ag | Multiprocessor having runtime adjustable clock and clock dependent power supply |
| US9250908B2 (en) | 2001-03-05 | 2016-02-02 | Pact Xpp Technologies Ag | Multi-processor bus and cache interconnection system |
| US9141390B2 (en) | 2001-03-05 | 2015-09-22 | Pact Xpp Technologies Ag | Method of processing data with an array of data processors according to application ID |
| US7210129B2 (en) | 2001-08-16 | 2007-04-24 | Pact Xpp Technologies Ag | Method for translating programs for reconfigurable architectures |
| US7844796B2 (en) | 2001-03-05 | 2010-11-30 | Martin Vorbach | Data processing device and method |
| US7581076B2 (en) | 2001-03-05 | 2009-08-25 | Pact Xpp Technologies Ag | Methods and devices for treating and/or processing data |
| RU2202123C2 (en) * | 2001-06-06 | 2003-04-10 | Бачериков Геннадий Иванович | Programmable-architecture parallel computer system |
| US10031733B2 (en) | 2001-06-20 | 2018-07-24 | Scientia Sol Mentis Ag | Method for processing data |
| EP1402382B1 (en) | 2001-06-20 | 2010-08-18 | Richter, Thomas | Data processing method |
| US7996827B2 (en) | 2001-08-16 | 2011-08-09 | Martin Vorbach | Method for the translation of programs for reconfigurable architectures |
| US7434191B2 (en) | 2001-09-03 | 2008-10-07 | Pact Xpp Technologies Ag | Router |
| US8686475B2 (en) | 2001-09-19 | 2014-04-01 | Pact Xpp Technologies Ag | Reconfigurable elements |
| US7577822B2 (en) | 2001-12-14 | 2009-08-18 | Pact Xpp Technologies Ag | Parallel task operation in processor and reconfigurable coprocessor configured based on information in link list including termination information for synchronization |
| AU2003208266A1 (en) | 2002-01-19 | 2003-07-30 | Pact Xpp Technologies Ag | Reconfigurable processor |
| JP3902741B2 (en) * | 2002-01-25 | 2007-04-11 | 株式会社半導体理工学研究センター | Semiconductor integrated circuit device |
| WO2003071432A2 (en) | 2002-02-18 | 2003-08-28 | Pact Xpp Technologies Ag | Bus systems and method for reconfiguration |
| US9170812B2 (en) | 2002-03-21 | 2015-10-27 | Pact Xpp Technologies Ag | Data processing system having integrated pipelined array data processor |
| US8914590B2 (en) | 2002-08-07 | 2014-12-16 | Pact Xpp Technologies Ag | Data processing method and device |
| US7657861B2 (en) | 2002-08-07 | 2010-02-02 | Pact Xpp Technologies Ag | Method and device for processing data |
| AU2003286131A1 (en) | 2002-08-07 | 2004-03-19 | Pact Xpp Technologies Ag | Method and device for processing data |
| US7394284B2 (en) | 2002-09-06 | 2008-07-01 | Pact Xpp Technologies Ag | Reconfigurable sequencer structure |
| US7237041B2 (en) * | 2002-12-02 | 2007-06-26 | Adc Telecommunications, Inc. | Systems and methods for automatic assignment of identification codes to devices |
| US20040252547A1 (en) * | 2003-06-06 | 2004-12-16 | Chengpu Wang | Concurrent Processing Memory |
| EP1676208A2 (en) | 2003-08-28 | 2006-07-05 | PACT XPP Technologies AG | Data processing device and method |
| US7707387B2 (en) | 2005-06-01 | 2010-04-27 | Microsoft Corporation | Conditional execution via content addressable memory and parallel computing execution model |
| US7451297B2 (en) * | 2005-06-01 | 2008-11-11 | Microsoft Corporation | Computing system and method that determines current configuration dependent on operand input from another configuration |
| US7793040B2 (en) * | 2005-06-01 | 2010-09-07 | Microsoft Corporation | Content addressable memory architecture |
| EP1974265A1 (en) | 2006-01-18 | 2008-10-01 | PACT XPP Technologies AG | Hardware definition method |
| WO2007143278A2 (en) * | 2006-04-12 | 2007-12-13 | Soft Machines, Inc. | Apparatus and method for processing an instruction matrix specifying parallel and dependent operations |
| WO2010044033A1 (en) * | 2008-10-16 | 2010-04-22 | Nxp B.V. | System and method for processing data using a matrix of processing units |
| GB2470037B (en) | 2009-05-07 | 2013-07-10 | Picochip Designs Ltd | Methods and devices for reducing interference in an uplink |
| GB2470771B (en) | 2009-06-05 | 2012-07-18 | Picochip Designs Ltd | A method and device in a communication network |
| GB2470891B (en) | 2009-06-05 | 2013-11-27 | Picochip Designs Ltd | A method and device in a communication network |
| GB2482869B (en) | 2010-08-16 | 2013-11-06 | Picochip Designs Ltd | Femtocell access control |
| US9274793B2 (en) | 2011-03-25 | 2016-03-01 | Soft Machines, Inc. | Memory fragments for supporting code block execution by using virtual cores instantiated by partitionable engines |
| GB2489919B (en) | 2011-04-05 | 2018-02-14 | Intel Corp | Filter |
| GB2489716B (en) | 2011-04-05 | 2015-06-24 | Intel Corp | Multimode base system |
| KR20150130510A (en) | 2013-03-15 | 2015-11-23 | 소프트 머신즈, 인크. | A method for emulating a guest centralized flag architecture by using a native distributed flag architecture |
Family Cites Families (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3287703A (en) * | 1962-12-04 | 1966-11-22 | Westinghouse Electric Corp | Computer |
| US3638204A (en) * | 1969-12-19 | 1972-01-25 | Ibm | Semiconductive cell for a storage having a plurality of simultaneously accessible locations |
| US3815095A (en) * | 1972-08-29 | 1974-06-04 | Texas Instruments Inc | General-purpose array processor |
| US4187551A (en) * | 1975-11-21 | 1980-02-05 | Ferranti Limited | Apparatus for writing data in unique order into and retrieving same from memory |
| JPS5352029A (en) * | 1976-10-22 | 1978-05-12 | Fujitsu Ltd | Arithmetic circuit unit |
| US4215401A (en) * | 1978-09-28 | 1980-07-29 | Environmental Research Institute Of Michigan | Cellular digital array processor |
| US4309755A (en) * | 1979-08-22 | 1982-01-05 | Bell Telephone Laboratories, Incorporated | Computer input/output arrangement for enabling a simultaneous read/write data transfer |
| US4314349A (en) * | 1979-12-31 | 1982-02-02 | Goodyear Aerospace Corporation | Processing element for parallel array processors |
| US4384273A (en) * | 1981-03-20 | 1983-05-17 | Bell Telephone Laboratories, Incorporated | Time warp signal recognition processor for matching signal patterns |
| US4524455A (en) * | 1981-06-01 | 1985-06-18 | Environmental Research Inst. Of Michigan | Pipeline processor |
| US4574394A (en) * | 1981-06-01 | 1986-03-04 | Environmental Research Institute Of Mi | Pipeline processor |
| US4533993A (en) * | 1981-08-18 | 1985-08-06 | National Research Development Corp. | Multiple processing cell digital data processor |
| DE3279328D1 (en) * | 1981-12-08 | 1989-02-09 | Unisys Corp | Constant-distance structure polycellular very large scale integrated circuit |
| US4507726A (en) * | 1982-01-26 | 1985-03-26 | Hughes Aircraft Company | Array processor architecture utilizing modular elemental processors |
| EP0114852B1 (en) * | 1982-07-21 | 1987-11-11 | Gec-Marconi Limited | Multi-dimensional-access memory system |
| US4489381A (en) * | 1982-08-06 | 1984-12-18 | International Business Machines Corporation | Hierarchical memories having two ports at each subordinate memory level |
| US4511967A (en) * | 1983-02-15 | 1985-04-16 | Sperry Corporation | Simultaneous load and verify of a device control store from a support processor via a scan loop |
| US4739474A (en) * | 1983-03-10 | 1988-04-19 | Martin Marietta Corporation | Geometric-arithmetic parallel processor |
| US4630230A (en) * | 1983-04-25 | 1986-12-16 | Cray Research, Inc. | Solid state storage device |
| US4635292A (en) * | 1983-12-19 | 1987-01-06 | Matsushita Electric Industrial Co., Ltd. | Image processor |
| US4573116A (en) * | 1983-12-20 | 1986-02-25 | Honeywell Information Systems Inc. | Multiword data register array having simultaneous read-write capability |
| GB8401805D0 (en) * | 1984-01-24 | 1984-02-29 | Int Computers Ltd | Data processing apparatus |
| US4660155A (en) * | 1984-07-23 | 1987-04-21 | Texas Instruments Incorported | Single chip video system with separate clocks for memory controller, CRT controller |
| US4663742A (en) * | 1984-10-30 | 1987-05-05 | International Business Machines Corporation | Directory memory system having simultaneous write, compare and bypass capabilites |
| US4623990A (en) * | 1984-10-31 | 1986-11-18 | Advanced Micro Devices, Inc. | Dual-port read/write RAM with single array |
| DE3576229D1 (en) * | 1984-11-05 | 1990-04-05 | Hughes Aircraft Co | COMMAND FLOW CALCULATOR. |
| CA1233260A (en) * | 1985-03-13 | 1988-02-23 | Chuck H. Ngai | High performance parallel vector processor having a modified vector register/element processor configuration |
| US4739476A (en) * | 1985-08-01 | 1988-04-19 | General Electric Company | Local interconnection scheme for parallel processing architectures |
| US4720780A (en) * | 1985-09-17 | 1988-01-19 | The Johns Hopkins University | Memory-linked wavefront array processor |
| CN1012297B (en) * | 1985-11-13 | 1991-04-03 | 奥尔凯托N·V公司 | Array recognization with internal cellular control and processing |
| US4769779A (en) * | 1985-12-16 | 1988-09-06 | Texas Instruments Incorporated | Systolic complex multiplier |
| US4773038A (en) * | 1986-02-24 | 1988-09-20 | Thinking Machines Corporation | Method of simulating additional processors in a SIMD parallel processor array |
| US4933846A (en) * | 1987-04-24 | 1990-06-12 | Network Systems Corporation | Network communications adapter with dual interleaved memory banks servicing multiple processors |
-
1989
- 1989-09-29 WO PCT/US1989/004281 patent/WO1990004235A1/en active IP Right Grant
- 1989-09-29 EP EP89911722A patent/EP0390907B1/en not_active Expired - Lifetime
- 1989-09-29 JP JP1510895A patent/JP2930341B2/en not_active Expired - Lifetime
- 1989-09-29 DE DE68926783T patent/DE68926783T2/en not_active Expired - Fee Related
- 1989-10-04 CA CA002000151A patent/CA2000151C/en not_active Expired - Lifetime
-
1992
- 1992-09-23 US US07/948,617 patent/US5421019A/en not_active Expired - Lifetime
Also Published As
| Publication number | Publication date |
|---|---|
| EP0390907B1 (en) | 1996-07-03 |
| DE68926783T2 (en) | 1996-11-28 |
| JPH03501787A (en) | 1991-04-18 |
| US5421019A (en) | 1995-05-30 |
| JP2930341B2 (en) | 1999-08-03 |
| WO1990004235A1 (en) | 1990-04-19 |
| DE68926783D1 (en) | 1996-08-08 |
| EP0390907A1 (en) | 1990-10-10 |
| CA2000151A1 (en) | 1990-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2000151C (en) | Parallel data processor | |
| US4739474A (en) | Geometric-arithmetic parallel processor | |
| US4507726A (en) | Array processor architecture utilizing modular elemental processors | |
| US4215401A (en) | Cellular digital array processor | |
| EP0086052B1 (en) | Segregator functional plane for use in a modular array processor | |
| US5179714A (en) | Parallel bit serial data processor | |
| US4745546A (en) | Column shorted and full array shorted functional plane for use in a modular array processor and method for using same | |
| JP2756257B2 (en) | Parallel processing system and method | |
| US5247613A (en) | Massively parallel processor including transpose arrangement for serially transmitting bits of data words stored in parallel | |
| US6032246A (en) | Bit-slice processing unit having M CPU's reading an N-bit width data element stored bit-sliced across M memories | |
| US7073039B2 (en) | Providing a register file memory with local addressing in a SIMD parallel processor | |
| US7386689B2 (en) | Method and apparatus for connecting a massively parallel processor array to a memory array in a bit serial manner | |
| US4939642A (en) | Virtual bit map processor | |
| KR100280854B1 (en) | Data processing device and data processing method | |
| US4524428A (en) | Modular input-programmable logic circuits for use in a modular array processor | |
| US4543642A (en) | Data Exchange Subsystem for use in a modular array processor | |
| EP0557997A2 (en) | Information processing apparatus and information processing system | |
| EP1792258B1 (en) | Interconnections in simd processor architectures | |
| Kondo et al. | Pseudo MIMD array processor—AAP2 | |
| GB2370139A (en) | Parallel loaded shift register in parallel processor element | |
| US5162796A (en) | Digital signal inversion employing cross-over switch | |
| JPS6074053A (en) | Array memory | |
| JPH0553898A (en) | Function memory | |
| JPH05204744A (en) | Two-dimensionally arranged data access device | |
| GB2393288A (en) | method of generating an interleave pattern for work loads in an array of processing elements. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed | ||
| MKEC | Expiry (correction) |
Effective date: 20121202 |