US 5,333,284 AGranted

Repeated ALU in pipelined processor design

AssigneeHoneywell Inc.

InventorPatrick R. Nugent

Life of this patent

Estimating the term…

Loading patent…

Abstract

A six-stage pipeline processor comprised of the sequential stages: instruction fetch; instruction decode; first full ALU; second full ALU; Fill Register; and Write Back Register. Memory addresses are calculated in the first ALU stage and this stage presents the address along with a read command to memory at the end of the cycle. Two cycles are allowed for a data response from memory but only one intervening instruction is required to occupy the pipeline for most instruction sequences because bypass logic makes available data operands from memory at the second ALU stage with an apparent load latency of one cycle. Subtraction and other arithmetic operations are performed in the first ALU if the result is to be used by a subsequent LOAD instruction to calculate a memory address so that the result can be used by the first ALU to calculate the memory address without any intervening instruction and potential loss of overall performance.

Claims3

Dependent claims

A pipeline processor for executing a sequence of instructions, including main memory LOAD/STORE instructions, the combination comprising:

a main memory with a load latency of two clock cycles;
a first ALU for executing add instructions, subtract instructions;
and conditional branch instructions;
a first ALU operand register for coupling operands to said first ALU for execution;
said first ALU executing a first instruction during a first clock cycle on a first operand coupled to said first ALU from said first ALU operand register;
means for coupling an output of said first ALU to said main memory;
a second ALU for executing add instructions, subtract instructions, and conditional branch instructions;
a second ALU operand register for coupling operands to said second ALU for execution;
said second ALU executing said first instruction, during a second clock cycle, on said first operand coupled to said second ALU from said second ALU operand register;
and
means for coupling an output of said second ALU to said first ALU operand register in a way that bypasses said main memory and provides the result of an execution of said first instruction on said first operand by said second ALU as an operand to said first ALU after a latency of one clock period.

8 more elements

Dependencies are inferred from the claim text.