Replication of data
AssigneeSGS-Thomson Microelectronics S.A.
InventorNathan Mackenzie Sidwell
Life of this patent
Estimating the term…
Abstract
A computer instruction is provided which replicates a bit sequence to generate a data string consisting only of a plurality of the replicated bit sequences. The computer instruction allows this to be done in a register store having a predetermined bit capacity addressable by a single address. The computer instruction is useful in the context of packed arithmetic instructions, where it is often desirable to combine each of a set of objects arithmetically or logically with a common object. A computer and a method of operating a computer using the instruction are described.
Claims23
A method of generating data string in a register store having a predetermnined bit capacity addressable by a single address, the method comprising:
- identifying a bit string having a plurality of bit sequences of an identical bit length less than said predetermined bit capacity;
- selecting one of said plurality bit sequences for replication;
- and
- replicating the selected bit sequence in successive locations in said register store to generate said data string having a bit length matching said bit capacity and consisting only of said plurality of bit sequences.
A method of operating a computer which method comprises holding in data storage a plurality of bit sequences representing discrete data and forming one of a plurality sub-strings of a data string, said method further comprising:
- executing an instruction sequence including at least one replicate instruction, the replicate instruction includes selecting one of said sub-strings for replication and replicating said selected sub-string at at least one successive location in a register store having a predetermined bit capacity addressable by a single address to generate a new data string having a bit length matching said bit capacity and consisting only of said plurality of bit sequences.
A method of operating a computer system to effect a matrix multiplication operation which method comprises forming a first data string from a plurality of sub-strings each representing respective data value at locations of at least one row or column of a first matrix, forming a second data string from a plurality of subs-strings each representing a respective data value at locations of at least one row or column in a second matrix, and generating a new data string from one of said first or second data strings by selecting one of said sub-strings and replicating said selected sub-string in successive locations in a register store to generate said new data string prior to multiplication of said new data string by the other of said first and second data strings.
A computer comprising a processor, memory and data storage circuitry for holding bit sequences each forming a sub-string of discrete data within a data string, said computer having stored in said memory an instruction sequence comprising a first instruction which selects one of said sub-strings within the data string and replicates the selected sub-string at a plurality of successive locations to generate a new data string, and a second instruction which designates a common operation on at least some sub-strings of the new data string without identification of said sub-strings within the data string.
A method of operating a computer which method comprises holding in data storage a plurality of bit sequences each representing discrete data and forming one of a plurality of sub-strings of a data string, executing a first instruction which selects at least one of said sub-strings within the data string and replicates the selected sub-string at a plurality of successive locations to generate a new data string, and executing a second instruction which effects a common operation on at least some sub-strings of the restructured data string without identification of said sub-strings within the data string.
Dependencies are inferred from the claim text.