Method and apparatus for moving data

Abstract

A method for moving data from a first register file (4358-1 to 4358-8, 7902) in a computational unit (4308-1 to 4308-M, 7607-1 to 7607-P) to a second register file (5206) in a processor (4322, 7614) is provided. The state of a signal on a data movement lead (risc_is_mtwr) is changed to indicate the data movement instruction from a first register file (4358-1 to 4358-8, 7902) in a computational unit (4308-1 to 4308-M, 7607-1 to 7607-P) to a second register file (5206) in a processor (4322, 7614). A lane address from the processor (4322, 7614) to the computational unit (4308-1 to 4308-M, 7607-1 to 7607-P) is provided over a first address lead (risc_is_ra).; A read address from the processor (4322, 7614) to the computational unit (4308-1 to 4308-M, 7607-1 to 7607-P) is provided over a second address lead (risc_is_ra), and data is transferred from the first register file (4358-1 to 4358-8, 7902) in the computational unit (4308-1 to 4308-M, 7607-1 to 7607-P) to the second register file (5206) in the processor (4322, 7614) over a data interface lead (node_regf_rd).

Description

technical field [0001] The present disclosure relates generally to processors, and more particularly to processing clusters. Background technique [0002] Figure 1 is a graph depicting speedup versus parallel overhead for execution rates for multi-core systems (ranging from 2 to 16 cores), where speedup is single processor execution time divided by parallel processor execution time. It can be seen that the parallel overhead must be close to zero to gain significant benefit from a large number of cores. However, since the overhead tends to be very high if there is any interaction between parallel programs, it is often difficult to efficiently use more than one or two processors for anything but completely separate programs. Therefore, it is necessary to improve processing clusters. Contents of the invention [0003] Accordingly, an embodiment of the invention provides a method. The method is characterized by: changing the signal state on the data movement lead (risc_is_m...

AI Technical Summary

Problems solved by technology

However, since the overhead tends to be very high if there is any interaction between parallel programs, it is often difficult to efficiently use more than one or two processors for anything but completely separate programs

Images