Cray Research, Inc.

Cray Incorporated

Silicon Graphics, Incorporated

William T Moore, Jr

Steven Lee Scott

Gregory J Faanes

George W Leedom

David R Resnick

Nicholas J Krajewski

James Robert Kohn

David J Johnson

Kent T McDaniel

Bricky A Stephenson

Carl D Breske

David R Kiefer

Brick Stephenson

Anthony A Vacca

Michael R Edwards

Seymour R Cray

Eric P Lundberg

Timothy J Johnson

Dennis W Kuba

James R Bedell

A method and apparatus to provide specifiable ordering between and among vector and scalar operations within a single streaming processor (SSP) via a local synchronization (Lsync) instruction that operates within a relaxed memory consistency model. Various aspects of that relaxed memory consistency model are described. Further, a combined memory synchronization and barrier synchronization (Msync) for a multistreaming processor (MSP) system is described. Also, a global synchronization (Gsync) instruction provides synchronization even outside a single MSP system is described. Advantageously, the pipeline or queue of pending memory requests does not need to be drained before the synchronization operation, nor is it required to refrain from determining addresses for and inserting subsequent memory accesses into the pipeline.

Relaxed memory consistency model

Multistream processing memory-and barrier-synchronization method and apparatus

In a computer system having a scalar processing unit and a vector processing unit, wherein the vector processing unit includes a vector dispatch unit, a system and method of decoupling operation of the scalar processing unit from that of the vector processing unit, the method comprising sending a vector instruction from the scalar processing unit to the vector dispatch unit, wherein sending includes marking the vector instruction as complete if the vector instruction is not a vector memory instruction and if the vector instruction does not require scalar operands, reading a scalar operand, wherein reading includes transferring the scalar operand from the scalar processing unit to the vector dispatch unit, predispatching the vector instruction within the vector dispatch unit if the vector instruction is scalar committed, dispatching the predispatched vector instruction if all required operands are ready, and executing the dispatched vector instruction as a function of the scalar operand.

Decoupled scalar/vector computer architecture system and method

An apparatus and a method are disclosed for performing both floating-point operations and integer operations utilizing a single functional unit. The floating-point adder performs logic for comparing exponents, logic for selecting and shifting a co-efficient, and logic for adding coefficients. In operation, the floating-point adder unit performs integer addition, subtraction, and compare operations using substantially the same hardware as used for floating-point operations. The output of the logic for comparing exponents represents the most significant bits of the result of the integer operation. The output of the logic for adding co-efficients represents the least significant bits of the result of the integer operation. If there is a carry from the logic for adding co-efficients, the value of the carry is added to the partial result representing the most significant bits of the integer operation. The floating-point adder permits all integer add, subtract and compare operations be performed by the floating-point adder without adding substantial additional hardware to the arithmetic logic unit.

Floating-point adder performing floating-point and integer operations

An apparatus and a method for extended-precision vector arithmetic capable of extremely long precision (i.e., precision to as many bits as a user desires or is limited to due to memory, disk-storage, or other resource constraints). Vector carry-out bits can be used as vector carry-in bits for successive operations. In performing add or subtract operations on integers that are longer than the word size of the computer, the operands a broken into word-sized parts which are used as operands. A vector of long-integer numbers is thus broken into a series of sub-vectors, each having word-sized elements. Vector add or subtract operations are performed successively on the sub-vectors, starting with the lowest-order sub-vectors. Carry-out (or borrow-out) bits from a first vector operation are used as carry-in (or borrow-in) bits for a successive vector operation. In one embodiment, instructions are added to the instruction set of a vector processor to assist in propagating carry (or borrow) bits between components of long operands, and to assist users in accessing and controlling the carry (or borrow) bits.

Apparatus and method for improved vector processing to support extended-length integer arithmetic

Method and apparatus for a logical address translator which translates a logical address into a physical address in a computer. The computer includes a plurality of address ports. Each address port includes a logical address translator, which includes a plurality of segment-register sets. Each segment-register set holds values which specify address boundaries and translation mapping of a corresponding logical segment. A segment detector is coupled to the plurality of segment-register sets, wherein the segment detector operates to determine whether the logical address is within the specified address boundaries of the logical segment. An address mapper is coupled to the plurality of segment-register sets, wherein the address mapper operates to translate the logical address into a physical address. A translation controller is connected to the segment detector and the address translator, wherein the translation controller operates to output the physical address if the segment detector determines that the logical address is within the specified address boundaries of the logical segment. One embodiment of the segment-register set includes a base address, a limit address, and a physical mapping bias. One embodiment of the computer includes a plurality of address ports, wherein each address port includes a logical address translator.

Computer having multiple address ports, each having logical address

Method and apparatus for a scalar data cache in a scalar/vector supercomputer. The scalar data cache comprises a cache array and a cache controller. The cache array comprises a plurality of cache frames; each cache frame comprises a plurality of cache lines; and each cache line comprises a plurality of data words. The cache controller performs a broadside compare of the reference address against all addresses held in the cache, and translates the reference address into a cache-array address. For each cache line, there is a corresponding cache-line validity indication which is set 'valid' only when every data word in the cache line contains valid data. A cache-line validity comparator operates to provide a cache-line-hit indication if a data word requested is in a valid cache line. A cache-load controller is described for loading data from a common memory into every data word of a cache line and for marking the cache line 'valid'. For at least one of the plurality of scalar registers, a cache accessor is described for providing fetch access to the data words in the cache array, and for providing write-through-cache capability to the data words in the cache array.

Associative scalar data cache with write-through capabilities for a

Method and apparatus for vector processing on a computer system. As the last element of a group of elements (called a 'chunk') in a vector register is loaded from memory, the entire chunk is marked valid and thus made available for use by subsequent or pending operations. The vector processing apparatus comprises a plurality of vector registers, wherein each vector register holds a plurality of elements. For each of the vector registers, a validity indicator is provided wherein each validity indicator indicates a subset of the elements in the corresponding vector register which are valid. A chunk-validation controller is coupled to the validity indicators operable to adjust a value of the validity indicator in response to a plurality of elements becoming valid. An arithmetic logical functional unit (ALFU) is coupled to the vector registers to execute functions specified by program instructions. A vector register controller is connected to control the vector registers in response to program instructions in order to cause valid elements of a selected vector register to be successively transmitted to said ALFU, so that elements are streamed through said ALFU at a speed that is determined by the availability of valid elements from the vector registers. The ALFU optionally comprises a processor pipeline to hold operand data for operations not yet completed while receiving operands for successive operations. The ALFU also optionally comprises an address pipeline to hold element addresses corresponding to the operands for operations not yet completed while receiving element addresses corresponding to the operands for successive operations.

Vector register validity indication to handle out-of-order element

Improved performance is obtained in computers of the type having vector registers which communicate with one or more functional units and common memory. As elements of a vector are read from a vector register for transmission to common memory or as operands to a functional unit, the vector register immediately becomes available to receive and store elements of a vector from common memory or a functional unit. The element-by-element storing takes place simultaneously with the element-by-element reading, and trails the reading by at least one element so as to not overwrite elements yet to be read. Through the use of this technique a vector register can be loaded with a vector for a subsequent operation without having to wait for the completion of the previous operation which uses the same vector register.

Apparatus for calculating delay when executing vector tailgating

The invention comprises a plurality of stacked planar processing circuit boards surrounded on at least one side by a plurality of memory boards located substantially perpendicular to the planar processing boards, the processing and memory boards connected by orthogonal interconnect modules. The orthogonal interconnect modules allow closely-spaced orthogonal connection of the processing boards to the memory boards. The memory boards are of a densely packed design having a plurality of removeable memory chip stacks located on the memory boards.

High density interconnect apparatus

A system for scheduling instruction issuance in a vector register computer achieves increased efficiency of operation by performing pre-issuance checks to determine if resources requested by the instruction will be available when the instruction issues. A decoding apparatus first determines particular resources requested by the instruction, which includes apparatus for decoding vector register requests, functional unit requests, and address and scalar register path requests. Following decoding, a conflict resolution apparatus checks resource reservation flags to determine if the requested resources will be available when the instruction issues. If any requested resources will be busy, the system issues a primary conflict signal. At the scheduled instruction issuance time, the system again checks the resource reservation flags in response to the primary conflict signal. If the requested resources are still busy, the system generates a secondary conflict signal for as long as the requested resources remain busy. Finally, when the instruction issues, the system sets the resource reservation flags to reserve the requested resources for the instruction.

Growing community of inventors

Elk Mound, WI, United States of America

William T Moore, Jr