Complex arithmetic vector processor for performing control function, scalar operation, and set-up of vector signal processing instruction

• US 4,760,525 A
• Filed: 06/10/1986
• Issued: 07/26/1988
• Est. Priority Date: 06/10/1986
• Status: Expired due to Fees

• Current Assignee: United States Of America As Represented By The Secretary Of The Air Force
• Original Assignee: United States Of America As Represented By The Secretary Of The Air Force
• Inventors: Webb, Richard F.
• Primary Examiner(s): Zache, Raulfe B.
• Assistant Examiner(s): Mohamed, Ayni
• Application Number:
  US06/872,737
• Time in Patent Office: 777 Days
• Field of Search: 364/200 MS File, 364/900 MS File, 364/736
• US Class Current: 712/2
• CPC Class Codes: G06F 15/8061 Details on data memory access
• CPC Class Codes: G06F 17/10 Complex mathematical operations function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544

Abstract

The processor is optimized for high-speed processing of large vectors, to fully utilize VHSIC technology, and to implement a signal processor having the highest possible throughput per volume while maintaining flexibility. It includes a 25 MHz embedded 1750A computer for performing control function scalar operations and set-up of vector signal processing instructions to be performed by a vector processing unit. The 40 MHz vector processing unit (VPU) performs high speed processing of real and complex vectors. The VPU'"'"'s control portion provides an interface and status to the embedded 1750A computer. It also provides the control necessary for vector processing to occur concurrently with 1750A execution and concurrent with I/O of vector data.

Claims

1. A complex arithmetic vector processor module which is optimized for high speed processing of large vectors, using very high speed integrated circuit (VHSIC) chips, to implement a signal processor having the highest possible throughput per volume while maintaining flexibility, for use in a system having a computer bus and a data bus, said module comprising:

a computer unit operating at a first clock rate and a vector processing unit operating at a second clock rate, the second clock rate being at a substantially higher frequency than the first clock rate, the vector processing unit being comprised of a vector processing control portion and a vector processing element portion;

wherein the computer unit comprises interconnected integrated circuit chip components including a controller, an arithmetic unit, a micro-memory, a program memory, a memory control circuit, and first switching means for selectively interconnecting said components of the computer unit and also for making selective connections to the vector processing unit;

wherein the vector processing unit comprises interconnected integrated circuit chip components including second switching means for selectively interconnecting said components;

the vector processing element portion components being a data memory coupled to memory control circuits and a pipelined arithmetic unit;

the vector processor control portion components being a vector micro memory, vector control circuits, vector timing circuits, and a memory address generator;

said module being integrated in said system via computer bus interface means for coupling said computer bus to the computer unit and to the vector processing unit, and data bus interface means for coupling said data bus via said second switch means to said vector processing element portion;

the computer unit being programmed for performing control function scalar operations and set-up of vector signal processing instructions to be performed by the vector processing unit;

the vector processing unit being programmed for performing high speed processing of real and complex vectors, wherein the vector processing control portion provides an interface and status to the computer unit, and also provides control necessary for vector processing to occur concurrently with execution in said computer unit and concurrent with input/output of vector data;

wherein said pipelined arithmetic unit comprises two pipelined arithmetic integrated circuit chips, each of which is a general purpose, programmable unit, for operation at said second clock rate, for use in advanced signal processors to perform high speed vector-efficient operations, supporting real and complex operations in addition to logical functions, and for use in pairs to perform high speed complex processing applications, and which can also be organized in a variety of pipelined configurations geared for specific signal processing applications;

wherein each of said pipelined arithmetic integrated circuit chips comprises two pipelined multipliers and four arithmetic logic units connected with each other and with input/output buses through multiplexers and registers via crossbar means in said second switch means, with micro-programming means for controlling the crossbar means to make the unit highly reconfigurable, means for equalizing delays insequential pipeline sections, including pipelining the multipliers and placing registers both ahead and after each arithmetic logic unit, which optimizes the speed in vector computations without the need for resorting to multiple clocks;

means for bypassing the multipliers if they are not needed and to bypass one arithmetic logic unit if only one is needed, means for loading microcode for the next instruction concurrent with execution of the present instruction.

2. A complex arithmetic vector processor module according to claim 1, wherein each of said pipelined arithmetic integrated circuit chips includes function modification means, which is the changing of the arithmetic logic unit function or crossbar select code as a function of the processed data, thereby omitting frequent micro program jumps to making multiple passes on a vector under reconfigured arithmetic unit states, whereby the chip has controlled flexibility which allows the chip to be used for implementing a very wide range of data processing algorithms, which includes both real and complex functions, and whereby it can be used to perform in both single and double precision modes;

some functions that the pipelined arithmetic unit can perform being fact Fourier transform, recursive filtering, transversal filtering, matched filtering, convolutional filtering, spectrum, shifting, weighting and band limited interpolation, each of these being performed under micro-program control.

3. A complex arithmetic vector processor module according to claim 2, wherein said memory address generator is a gate array with means to initialize it, and which once initialized, comprises means including storage of a current address value to provide an automatic sequence of bit addresses to said data memory of the vector processing element, with sequence types including straight, increment, bit reverse, bit twist, 1/4, 1/2 arithmetic unit bit twist, FFT Constant, segment, first and second type corner turns, fractional DAG, integer DAG, table look up, and I/O Counter, said sequence types being operative as follows:

(a) straight - an increment of one is added to the current address value on each successive clock cycle, with an offset value (initial value of the address) specified in the signal processing instruction;

(b) increment - an increment value is added to the current address value on each successive clock, an offset value and the increment value being specified in the signal processing instruction;

(c) bit reverse - the least significant bit of a straight sequence is reversed with K^th bit, the second least significant bit and the (K-1)^th bit are reversed, all bits remaining between the second least significant bit and the (K-1)^th bit are also reversed in a similar manner, the bit reverse number specifies the K^th bit and is specified in the signal processing as is the offset value, the bit referxe sequence being used to order FFT results in frequency ascending order;

(d) bit twist - the K^th least significant bits of a straight sequence are inverted, the bit twist number which specifies K is specified in the signal processing instruction as is the offset value, the bit twist sequence being used in performing the FFT instruction;

(e) 1/4, 1/2 arithmetic unit bit twist - similar to bit twist, required to perform the FFT in processors with a 1/4 AU or a 1/2 arithmetic unit architecture;

(f) FFT constant - an increment of one is added to the current address value at a rate that is dependent on what stage of the FFT is being performed, the current address will be incremented in such a way as to allow one FFT constant to be accessed during the first stage, the number of FFT constraints to be accessed is increased by a factor of two for each successive stage, the offset value being specified in the signal processing;

(g) segment - the segment sequence is used in associated with a four segment memory where each memory segment can only be accessed at 1/4 the system rate, the segment sequence stores the data in such a way that it can be accessed from each memory segment sequentially and thus achieves the maximum throughput equal to the system rate;

(h) first type cornerturn - used to access data which has been stored using either the straight or segment sequence, the cornerturn being a composite of increment sequences with each increment sequence starting at an offset of one greater than the previous increment sequence offset;

(i) second type cornerturn - used to access data which has been stored using either the striaght or segment sequence, being a composite of increment sequences with each increment sequence starting at an offset of one less than the previous increment sequence offset;

(j) fractional DAG - an increment value (0) is added to the current address value (0) on each successive clock, in addition, a second increment value (0) is added to the first increment value (0) on each successive clock, the initial value of 0, and value 0 are pointed by SP instruction arguments, this sequence type being used to generate non-linear address sequence;

(k) integer DAG - one of the four possible increment values is added to current address value on each successive clock, based on prioritized conditions, the offset value and four increment values are pointed to by SP instruction arguments, this sequence being used to generate address seqeunces which allow a matrix to be accessed in such a way that interpolation operations are performed;

(l) table look up - memory address generator is used as an order, the data output of the arithmetic unit is added to an offset, the resulting sequence being an address to one of the data memories, a form of indirect addressing;

(m) I/O counter - memory address generator is used as a counter with reset and hold control, used in this mode to provide addresses for module input and output buffers;

the memory address generator gate array being capable to be used at said first or second clock rate;

and wherein for some of the sequence types, two memory address generator gate arrays can operate in parallel in order to achieve a wider bit length output.

4. A complex arithmetic vector processor module according to claim 3, wherein said vector timing circuits comprise a gate array for performing a variety of control functions required in the vector processing unit, including

vector timing logic means and data memory write control logic means for providing read and write control to memory address generators and data memories, based on lengths of vectors being processed and the length of the pipeline in the vector processing unit;

start control logic means for providing an interface on status to said computer unit with regard to the activity in the vector processing unit, in order that vector instruction set up can be overlapped with execution of the previous vector instruction;

special FFT logic means for providing timing signals that control the memory address generator and data memory crossbars in a way that is unique to the FFT instruction.

logic control status register means for providing the control required to single step the vector processing unit at several different rates; and

arithmetic unit flags logic means for providing control logic that defines a window during which time the AU flags have a valid effect on the flag registers.

5. A comples arithmetic vector processor module according to claim 4, wherein said vector micro memory comprises a configuration micro memory and an arithmetic unit micro memory, wherein said vector control circuits comprise a gate array for providing address and write enable control to both the configuration micromemory and the arithmetic unit micro memory at said second clock rate, wherein address and write enables to the configuration micro memory are generated based on the computer bus interface means and configuration micro memory sequencer logic;

wherein computer bus interface logic means allows the configuration micro memory to be written from the computer bus interface for initial download of microcode, and also allows the configuration micro memory to be read from the computer bus interface in order to verify what was written; and

the configuration micro memory sequencer logic means provides the sequencing types given below, in order that the configuration micro memory can provide the control required in the module;

configuration micro sequencing types;

(a) increment the address,

(b) jump to a specified address,

(c) loop on a segment of microcode,

(d) loop on a segment of microcode and repeat within the loop.

6. A complex arithmetic vector processor module according to claim 5, wherein the arithmetic unit micro memory consists of two parts, one part being a bulk arithmetic unit memory which contains all arithmetic unit micro instructions, and the other part being a pipelined arithmetic unit micro memory which comprises four memory locations resident of each of the pipelined arithmetic unit chips, which comprise the complex arithmetic vector processor vector arithmetic unit;

the pipelined arithmetic unit micro memory being 192 bits wide and being capable of being loaded 32 bits at a time, whereby this feature alleviates chip pin constraints and allows half of the complex arithmetic vector processor vector arithmetic unit to be implemented on a single VHSIC custom chip;

wherein control of the arithmetic unit micro memory comprises an interface to the computer interface means, to control loading micro instructions from the bulk arithmetic unit micro memory into the pipelined arithmetic unit micromemories and the arithmetic unit micro memory sequencing logic, wherein the computer interface logic means permits the bulk arithmetic unit memory to be written from the computer interface means for initial download of microcode to be read back for verification, wherein in an operational mode, micro insturctions must be loaded from the bulk arithmetic unit micro into the pipelined arithmetic unit micros prior to execution of a signal processing instruction, and wherein control is provided to allow simultaneous signal processing instruction execution and loading of the pipelined arithmetic unit micros in order to keep overhead to a minimum, wherein loading of the pipelined arithmetic unit micro memories from the bulk arithmetic unit micro memory is transparent to the user, wherein the arithmetic unit micro sequencing logic provides the sequencing types given below in order that the arithmetic unit micro memory can provide the control required in the module;

(a) jump to a specified address,

(b) loop on a segment of microcode.