US 4,777,587 AGranted

System for processing single-cycle branch instruction in a pipeline having relative, absolute, indirect and trap addresses

AssigneeAdvanced Micro Devices, Inc.

InventorsBrian W. Case, Rod G. Fleck, Cheng-Gang Kong, Ole Moller

Life of this patent

Estimating the term…

Loading patent…

Abstract

An instruction processor suitable for use in a reduced instruction-set computer employs an instruction pipeline which performs conditional branching in a single processor cycle. The processor treats a branch condition as a normal instruction operand rather than a special case within a separate condition code register. The condition bit and the branch target address determine which instruction is to be fetched, the branch not taking effect until the next-following instruction is executed. In this manner, no replacement of the instruction which physically follows the branch instruction in the pipeline need be made, and the branch occurs within the single cycle of the pipeline allocated to it. A simple circuit implements this delayed-branch method. A computer incorporating the processor readily executes special-handling techniques for calls on subroutine, interrupts and traps.

Claims8

Dependent claims

A digital instruction processor control which cyclically executes, in a single cycle, instructions from a set, including a plurality of plural-bit branch instructions, stored in an instruction cache having a plurality of locations each with a designator, said processor control comprising:

means for generating signals indicative of a "continue with instruction" address;
means responsive to continue address signals and to predetermined bit portions of said branch instructions for generating signals indicative of a "branch target" address;
and
first multiplexer means having an output terminal connected to said instruction cache responsive to a control signal indicative of contents of a predetermined "condition" bit portion of said branch instruction, responsive to branch target address signals applied to a first input terminal of said first multiplexer means and responsive to said continue address signals applied to a second input terminal of said first multiplexer means for selectively conducting to said output terminal one of said address signals indicative of a location within said instruction cache from which to fetch a next instruction to be processed by said instruction processor;
wherein said branch target address generating means comprises:
second multiplexer/register means having an output terminal connected to said first input terminal of said first multiplexer means responsive to a control signal indicative of which of said plurality of branch instructions is being executed, responsive to signals applied to a first input terminal indicative of an "indirect address" determined by said branch instructions, and responsive to signals applied to a second input terminal indicative of a "relative or absolute branch address" for selectively conducting to said output terminal one of said address signals indicative of said branch target address;
and
adder means having an output terminal connected to said second input terminal of said second mutliplexer/register means responsive to said control signal indicative of which of said plurality of branch instructions is being executed, responsive to signals applied to a first input terminal indicative of a "branch displacement" portion of said branch instructions, and responsive to said continue address signals applied to a second input terminal for selectively generating at said output terminal said relative or absolute branch address. PG,15

5 more elements

A method of performing branches in one cycle of a digital instruction processor control having an instruction pipeline and a program counter, which cyclically executes instructions from a set, including a "branch delay" instruction and a plurality of branch instructions each determining a "branch condition", stored in an instruction cache having a plurality of locations each with a designator, comprising steps of:

(a) fetching from said cache at a location designator specified by contents of said program counter a branch instruction placing said instruction in said instruction pipeline and placing said branch delay instruction on said instruction pipeline during a next-following cycle;
(b) decoding said branch instruction stored in said instruction pipeline;
(c) saving said branch condition determined by said branch instruction;
(d) selecting a branch target address from an indirect, relative and absolute branch address based on information generated at decoding step (b);
(e) fetching an instruction from a location in said cache selected from said branch target address determined at step (d) and the contents of said program counter, based on said branch condition saved at step (c) and placing a fetched instruction in said instruction pipeline during a cycle next-following a cycle during which said branch delay was placed in said pipeline;
and
(f) replacing the contents of the program counter with the address used to fetch said instruction at step (e) incremented by one.

4 more elements

A method of processing an interrupt routine by a digital instruction processor control having an instruction pipeline, which cyclically executes instructions from a set, including an interrupt procedure call instruction, a "branch delay" instruction and a plurality of branch instructions each determining a "branch condition", stored in an instruction cache having a plurality of locations each with a designator, wherein said interrupt routine is to be processed between a branch instruction placed in said pipeline one cycle prior to occurrence of said interrupt, and having a branch target address and a branch delay instruction placed in said pipeline to two cycles prior to the occurence of said interrupt, comprising steps of:

(a) saving a location designator of a branch target instruction;
(b) saving said branch target address;
(c) processing said interrupt routine;
(d) prior to returning from said interrupt routine fetching for said pipeline an instruction located at the location designator saved at step (a);
and
(e) prior to returning from said interrupt routine fetching for said pipeline the instruction located at the location designator saved at step (b).

3 more elements

Dependencies are inferred from the claim text.