Electrical computers and digital processing systems: processing – Processing architecture – Superscalar
Reexamination Certificate
2002-01-03
2004-06-29
Donaghue, Larry D. (Department: 2154)
Electrical computers and digital processing systems: processing
Processing architecture
Superscalar
C712S215000, C712S216000
Reexamination Certificate
active
06757808
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates generally to superscalar computers, and more particularly, a system and method for using tags to control instruction execution in a superscalar reduced instruction set computer (RISC).
2. Related Art
Processors used in conventional computer systems typically execute program instructions one at a time, in sequential order. The process of executing a single instruction involves several sequential steps. The first step generally involves fetching the instruction from a memory device. The second step generally involves decoding the instruction, and assembling any operands.
The third step generally involves executing the instruction, and storing the results. Some processors are designed to perform each step in a single cycle of the processor clock. Alternatively, the processor may be designed so that the number of processor clock cycles per step depends on the particular instruction.
To improve performance, modern computers commonly use a technique known as pipelining. Pipelining involves the overlapping of the sequential steps of the execution process. For example, while the processor is performing the execution step for one instruction, it might simultaneously perform the decode step for a second instruction, and perform a fetch of a third instruction. Pipelining can thus decrease the execution time for a sequence of instructions.
Another class of processors improve performance by overlapping the sub-steps of the three sequential steps discussed above are called superpipelined processors.
Still another technique for improving performance involves executing multiple instructions simultaneously. Processors which utilize this technique are generally referred to as superscalar processors. The ability of a superscalar processor to execute two or more instructions simultaneously depends on the particular instructions being executed. For example, two instructions which both require use of the same, limited processor resource (such as the floating point unit) cannot be executed simultaneously. This type of conflict is known as a resource dependency. Additionally, an instruction which uses the result produced by the execution of another instruction cannot be executed at the same time as the other instruction. An instruction which depends on the result of another instruction is said to have a data dependency on the other instruction. Similarly, an instruction set may specify that particular types of instructions must execute in a certain order relative to each other. These instructions are said to have procedural dependencies.
A third technique for improving performance involves executing instructions out of program order. Processors which utilize this technique are generally referred to as out-of-order processors. Usually, out-of-order processors are also superscalar processors. Data dependencies and procedural dependencies limit out-of-order execution in the same way that they limit superscalar execution.
From here on, the term “superscalar processor” will be used to refer to a processor that is: capable of executing multiple instructions simultaneously, or capable of executing instructions out of program order, or capable of doing both.
For executing instructions either simultaneously or out of order, a superscalar processor must contain a system called an Execution unit. The Execution Unit contains multiple functional units for executing instructions (e.g., floating point multiplier, adder, etc.). Scheduling control is needed to dispatch instructions to the multiple functional units. With in-order issue, the processor stops decoding instructions whenever a decoded instruction creates a resource conflict or has a true dependency or an output dependency on a uncompleted instruction. As a result, the processor is not able to look ahead beyond the instructions with the conflict or dependency, even though one or more subsequent instructions might be executable. To overcome this limitation, processors isolate the decoder from the execution stage, so that it continues to decode instructions regardless of whether they can be executed immediately. This isolation is accomplished by a buffer between the decode and execute stages, called an instruction window.
To take advantage of lookahead, the processor decodes instructions and places them into the window as long as there is room in the window and, at the same time, examines instructions in the window to find instructions that can be executed (that is, instructions that do not have resource conflicts or dependencies). The instruction window serves as a pool of instructions, giving the processor lookahead ability that is constrained only by the size of the window and the capability of the instruction source. Thus, out-of-order issue requires a buffer, called an instruction window between the decoder and functional units; and the instruction window provides a snap-shot of a piece of the program that the computer is executing.
After the instructions have finished executing, instructions must be removed from the window so that new instructions can take their place. Current designs employ an instruction window that utilizes a First In First Out queue (FIFO). In certain designs, the new instructions enter the window and completed instructions leave the window in fixed size groups. For example, an instruction window might contain eight instructions (I
0
-I
7
) and instructions may be changed in groups of four. In this case, after instructions I
0
, I
1
, I
2
and I
3
have executed, they are removed from the window at the same time four new instructions are advanced into the window. Instruction windows where instructions enter and leave in fixed size groups are called “Fixed Advance Instruction Windows.”
In other types of designs, the new instructions enter the window and completed instructions leave the window in groups of various sizes. For example, an instruction window might contain eight instructions (I
0
-I
7
) and may be changed in groups of one, two or three. In this case, after any of instructions I
0
, I
1
or I
2
have executed, they can be removed from the window and new instructions can be advanced into the window. Instruction windows where instructions enter and leave in groups of various sizes are called “Variable Advance Instruction Windows.”
Processors that use Variable Advance Instruction Windows (VAIW) tend to have higher performance than processors that have Fixed Advance Instruction Windows (FAIW). However, fixed advance instruction windows are easier for a processor to manage since a particular instruction can only occupy a fixed number of locations in the window. For example, in an instruction window that contains eight instructions (I
0
-I
7
) and where instructions can be added or removed in groups of four, an instruction can occupy only one of two locations in the window (e.g., I
0
and I
4
). In a variable advance instruction windows, that instruction could occupy all of the locations in the window at different times, thus a processor that has a variable advance instruction window must have more resources to track each instruction's position than a processor that has a fixed advance instruction window.
Current designs use large queues to implement the instruction window. The idea of using queues is disadvantageous, for many reasons including: a large amount of chip area resources are dedicated to a plurality of queues especially when implementing a variable advance instruction window; there is limited flexibility in designing a system with more than one queue; and control logic for directing data in queues is complex and inflexible.
Therefore, what is needed is a technique to “track” or monitor instructions as they move through the window. The system must be flexible and require a small area on a chip.
BRIEF SUMMARY OF THE INVENTION
The present invention is directed to a technique for monitoring instruction execution of multiple instructions in parallel and out of program order using a system that assigns tags to the multiple instruction
Deosaran Trevor A.
Garg Sanjiv
Iadonato Kevin R.
Donaghue Larry D.
Seiko Epson Corporation
Sterne Kessler Goldstein & Fox P.L.L.C.
LandOfFree
System and method for assigning tags to control instruction... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for assigning tags to control instruction..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for assigning tags to control instruction... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3362402