Sign generation bypass path to aligner for reducing signed data load latency
AssigneeSun Mirosystems, Inc.
InventorsDavid M. Pini, Yuefei Ge, Anup S. Tirumala
Life of this patent
Estimating the term…
Abstract
A method for reducing signed load latency in a microprocessor has been developed. The method includes transferring a part of data to an aligner via a bypass, and generating a sign bit from the part of the data. The sign bit is transferred to the aligner along the bypass, and the data is separately transferred to the aligner along a data path.
Claims9
A method for reducing signed load latency in transfer of data from a cache memory to another element in a microprocessor comprising:
- transferring data from the cache memory to a bypass path and a data path;
- transferring the data to an aligner from the data path;
- generating a sign bit for the data in the bypass path, wherein the generating comprises:
- selecting a plurality of candidate bits from the data, and
- generating the sign bit from the plurality of candidate bits dependent on a select signal, wherein the sign bit is generated independent of propagation of the data through the data path;
- transferring the sign bit to the aligner from the bypass path;
- arranging the data and the sign bit in the aligner into signed data;
- and
- transferring the signed data to the another element.
An apparatus for reducing signed load latency in transfer of data from a cache memory to another element in a microprocessor, comprising:
- a data path connecting the cache memory to an aligner, wherein data is transferred from the cache memory to the aligner via the data path;
- and
- a bypass path connecting the cache memory to the aligner, comprising:
- a sign multiplexer for selecting a plurality of candidate bits from the data in the cache memory;
- and
- a real-sign multiplexer for selecting a sign bit from the plurality of candidate bits dependent on a select signal, wherein the sign bit is selected independent of propagation of data through the data path,
- wherein the sign bit is transferred to the aligner from the bypass path, and
- wherein the aligner arranges the data and the signed bit into signed data and transfers the signed data to the another element.
An apparatus comprising:
- means for transferring data from a cache memory to a bypass path and a data path;
- means for generating a sign bit for the data in the bypass path, wherein the means for generating a sign bit for the data comprises:
- means for selecting a plurality of candidate bits from the data, and
- means for generating the sign bit from the plurality of candidate bits dependent on a select signal,
- wherein the sign bit is generated independent of propagation of data through the data path;
- means for transferring the sign bit to the aligner via the bypass path;
- means for adjusting the data during transfer to the aligner via the data path;
- means for adjusting the sign bit during transfer to the aligner via the bypass path;
- and
- means for selectively processing the part of the data selected for use in generating the sign bit based on an instruction from a CPU;
- means for arranging the data and the sign bit into signed data;
- and
- means for transferring the signed data to another element in a microprocessor.
An apparatus comprising:
- a data path connecting a cache memory to an aligner;
- wherein data is transferred from the cache memory to the aligner along the data path;
- a bypass path connecting the cache memory to the aligner;
- wherein the data is transferred from the cache memory to the bypass path, and
- wherein a sign bit is transferred to the aligner from the bypass path;
- a select component in the bypass path for providing signals to select the sign bit for the data, wherein the select component comprises:
- a sign multiplexer for selecting a plurality of candidate bits from the data;
- and
- a real-sign multiplexer for selecting the sign bit from the plurality of candidate bits,
- wherein the select component provides a signal for selecting a part of the data to generate the sign bit for the data based on an instruction from a CPU;
- wherein the aligner comprises a plurality of sub-aligners;
- and
- wherein the aligner arranges the data and the sign bit into signed data.
Dependencies are inferred from the claim text.