Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F9/00—Arrangements for program control, e.g. control units
  • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
  • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
  • G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
  • G06F9/3867—Concurrent instruction execution, e.g. pipeline or look ahead using instruction pipelines
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F9/00—Arrangements for program control, e.g. control units
  • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
  • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
  • G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
  • G06F9/3824—Operand accessing
  • G06F9/3826—Bypassing or forwarding of data results, e.g. locally between pipeline stages or within a pipeline stage
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06F—ELECTRIC DIGITAL DATA PROCESSING
  • G06F9/00—Arrangements for program control, e.g. control units
  • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
  • G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
  • G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
  • G06F9/3861—Recovery, e.g. branch miss-prediction, exception handling

Definitions

• the present invention pertains generally to the field of digital computation circuits, and in particular, the invention relates to a system and method for an instruction execution device for use with a processor with exception routines.
• ISAs microprocessor Instruction Set Architectures
• pipeline a so-called “pipeline” method to overlap different execution stages of subsequent instructions.
• a conventional four-stage pipeline employs a (1) Fetch, (2) Decode, (3) Execute and (4) a Write-back.
• a load instruction For data transfer type instructions such as a load instruction, one extra instruction pipeline stage is usually required.
• the processor fetches an instruction from memory.
• the address of the instruction to fetch is stored in the internal register, named the program counter, or PC.
• PC program counter
• the processor increments the PC. This means the fetch phase of the next cycle will fetch the instruction in the next sequential location in memory (unless the PC is modified by a later phase of the cycle).
• the processor stores the information returned by the memory in another internal register, known as the instruction register, or IR.
• the IR now holds a single machine instruction encoded as a binary number.
• the processor decodes the value in the IR in order to figure out which operations to perform in the next stage. In the execution stage, the processor actually carries out the instruction.
• the instruction may direct the processor to fetch two operands from memory (for example, storing them in operand registers), add them and store the result in a third location (the destination addresses of the operands and the result are also encoded as part of the instruction).
• the result computed upstream in the pipeline is written ⁇ retired) to a destination register in a register file.
• circuitry that allows operand or result values to bypass the register file.
• the operands or result values are already available to subsequent instructions before the operand-producing instructions are retired (e.g., written-back to register file).
• the large register file typically contributes significantly to the overall power consumption and size of the processor.
• the instruction execution device includes an instruction pipeline for producing a result for an instruction, wherein exception routines may interrupt the instruction pipeline at random intervals; a register file that includes at least one write port for storing the result, a bypass circuit for allowing access to the result, wherein an indication is provided to a register file control as to whether the result is used by only one other instruction, and the register file control prevents the result from being stored in the write port when the result has been accessed via the bypass circuit and is used by only one other instruction, a First-In-First-Out (FIFO) buffer for storing the result and a FIFO control for writing the contents of the FIFO buffer to the register file when an exception occurs.
• FIFO First-In-First-Out
• the indication whether the result is used by only one other instruction includes encoding each instruction.
• a so-called “dead value” field is designated in the "opcode" of each instruction to indicate whether the result will be used by only one other instruction.
• the means for indicating whether the first result is used by only the second instruction includes the instruction pipeline determining whether a result of an instruction in the instruction pipeline and another result of another instruction in the instruction pipeline are designated for storage in the same write port in the register file. Since the write port in the register file is "reused" by a subsequent instruction already in the instruction pipeline, this is used to indicate that the first result will be used by only one other instruction.
• Fig. 1 is a block diagram of one illustrative arrangement of an instruction execution device in accordance with the teachings of the present invention
• Fig. 2 is a block diagram of another illustrative arrangement of an instruction execution device in accordance with the teachings of the present invention.
• FIG. 1 a block diagram is shown illustrating one embodiment of an instruction execution device in accordance with the teachings of the present invention. It will be recognized that Fig. 1 is simplified for explanation purposes and that the full processor environment suitable for use with the invention will comprise, for example, cache memory, RAM and ROM memory, compiler or assembler, I/O devices, etc., all of which need not be shown here.
• instruction execution device 10 uses an n- stage pipeline instruction set register architecture (ISA) 12] through 12 n (hereinafter collectively known as "pipeline 12"), a conventional bypass circuit 14, a register file 16, a register file control 18, a First in First out (FIFO) buffer 24 and a FIFO buffer control 26.
• the pipeline 12 includes a number of pipeline stages
• pipeline 12 also includes a "consumer-id" field in the pipeline stage registers and its use is explained in more detail below. It should be understood, however, that the invention is not limited to particular pipeline architecture.
• the stages in a pipeline may include: instruction fetch, decode, operand fetch, ALU execution, memory access and write back of the operation results.
• the chain of stages in the pipeline can be subdivided more finely or combined. The number of stages in the pipeline is an architectural feature which can be changed according to the intended exploitation of instruction level parallelism.
• Register file 16 inc udes at least one addressable destination write port 20 for storing data.
• the register file can be any conventional database/indexing storage means that can store and allow access to records/data.
• Register file control 8 contains the majority of logic, control, supervisory, translation functions required for controlling the operation of writing-back write data to register file 16.
• Register file control 18 also includes programs for the operations functionally described in Fig. 3. As described in detail below, execution of these program implements the functionality necessary to reduce the number of registers file write operations in the pipeline. Instructions can be classified as one of three major types: arithmetic/logic, data transfer, and control. Arithmetic and logic instructions apply primitive functions of one or two arguments, for example, addition, multiplication, or logical AND.
• the timing of each stage depends on the internal construction of the processor and the complexity of the instructions it executes.
• the quantum time unit for measuring operations is known as a clock cycle.
• the logic that directs operations within a processor is controlled by an external clock, which, for example, may be a circuit that generates a square wave with a fixed period. The number of clock cycles required to carry out an operation determines the amount of time it will take.
• the consumer enters the pipeline architecture before the producer retires. This holds even stronger for specific types of processors, such as superscalar and VLIW processors. Accordingly, the consumer obtains the result value through a method other than the register file, for example, the bypass circuit. However, the result is nevertheless written-back to the register file.
• the register file control 16 determines whether a particular instruction will be used by more than one consumer and, if so, the result is written-back to register file 16 in stage-n, for example, the Write-Back stage. In either case the result (which includes the write data, destination address and consumer-id) is temporarily stored in FIFO buffer 24.
• explicit encoding in each instruction is used to indicate whether the result value of a particular instruction will be used by only one consumer (or by only other consumers in the pipeline).
• a dedicated "dead value” bit in the instruction encoding (the so-called “opcode” of an instruction) is used, which is set or cleared by a compiler or assembler (not shown), depending on the degree of consumption. If the dead value bit is set, then the result value is not written-back (for example, via a write- enable signal) to register file 16 in stage-n of pipeline 12, but is stored in FIFO buffer 24.
• the dead value bit as well as the instruction-id of the associated instruction, is provided to register file control 18, which in turn controls register file 16 via a write-enable signal to write-back the result value or not.
• register file control 18 controls register file 16 via a write-enable signal to write-back the result value or not.
• a dedicated bit is unavailable, then a few commonly used instructions can be selected (e.g., ADD and LOAD), which are assigned an alternative opcode to indicate the degree of consumption.
• the invention By storing the result in FIFO buffer 24, the invention accounts for the effects of exceptions that are allowed to interrupt the instruction flow through the pipeline.
• Exceptions can disrupt the instruction flow in such a way that the producer does not re-enter the pipeline after the exception is handled while the consumer does.
• the consumer When the consumer re- enters the pipeline after the exception routine finishes, it will read its operands from register file 16 because now the producer is not downstream of the consumer in the pipeline anymore. But since the producer had not updated register file 16 before the exception occurred, the consumer reads stale data from register file 16, and as a consequence the program behaves incorrectly.
• Trimedia ISA Certain pipeline architectures, such as the Trimedia ISA, do not allow interrupts to be taken at random points in the instruction flow. Instead, the ISA provides special instructions that cause the CPU to synchronize with pending exceptions. As long as the producer and the consumer are not separated by such instructions, no harm can be done and it is safe to discard dead values.
• MIPS Microprocessor without Interlocking Pipeline Stages
• most other conventional ISAs do not allow application programs to decide when the processor should synchronize with pending exceptions.
• exceptions can disrupt the instruction flow in such a way that the producer does not re-enter the pipeline after the exception is handled while the consumer does.
• FIFO buffer 24 stores the results of each instruction at the tail of the FIFO buffer 24. Again, the result includes the write data, destination address of register file 16 and consumer-id from the pipeline stage register that uniquely identifies a consumer.
• FIFO control 26 continuously monitors the flow of retiring instructions and compares their identity to the "consumer-id" at the head of FIFO 24. Whenever it detects a match, it so-called “pops" the head from FIFO buffer 24 and discards its value. However, during an exception, FIFO control 26 writes the content of FIFO buffer 24 to register file 16, starting with the head and progressing to the tail, until FIFO 24 is empty. As will be understood by persons skilled in the art, the FIFO buffer write operation can be performed in a variety of ways, for example, via the register file control or a direct connection to the register file. Thereafter a consumer will read the correct data from register file 16 after the exception routine finishes.
• pipeline 12 determines whether another instruction in the pipeline, will use the same destination address (e.g., write port 20) in register file 16, as a subsequent instruction in the pipeline. This can be determined in a number of ways. For example, if the result value of the instruction in stage-3 12 3 is designated for the destination address of write port 20 in register file 16 and the result value of the instruction in stage- 1 1 ] is also designated for the same destination address in register file 16, then the result value of the instruction in stage-3 is "dead" because its destination address will be reused by another instruction in the pipeline.
• a reduction in power consumption is dependent on the actual properties of the implementation technology. For instance, the data in the FIFO buffer is only stored there for a few clock cycles before it gets discarded. With this in mind, it may be possible to store the data in the FIFO buffer as dynamic charges with no refresh facilities. This could reduce the area complexity of the FIFO buffer significantly, which would be particularly advantageous for processors that exploit a high degree of instruction-level parallelism and which consequently must be able to retire many instructions per clock cycle.
• any element expressed as a means for performing a specified function is intended to encompass any way of performing that function including, for example, a) a combination of circuit elements which performs that function or b) software in any form, including, therefore, firmware, microcode or the like, combined with appropriate circuitry for executing that software to perform the function.
• the invention as defined by such claims resides in the fact that the functionalities provided by the various recited means are combined and brought together in the manner which the claims call for. Applicant thus regards any means which can provide those functionalities as equivalent as those shown herein.

Landscapes

• Engineering & Computer Science (AREA)
• Software Systems (AREA)
• Theoretical Computer Science (AREA)
• Physics & Mathematics (AREA)
• General Engineering & Computer Science (AREA)
• General Physics & Mathematics (AREA)
• Advance Control (AREA)
• Executing Machine-Instructions (AREA)
• Computer And Data Communications (AREA)

Priority Applications (4)

Applications Claiming Priority (2)

Publications (2)

Family

ID=24012699

Family Applications (1)

Country Status (6)

Cited By (5)

Families Citing this family (1)

Family Cites Families (13)

• 2000
  • 2000-02-16 US US09/505,986 patent/US6851044B1/en not_active Expired - Lifetime
• 2001
  • 2001-01-24 JP JP2001560791A patent/JP2004508607A/ja not_active Withdrawn
  • 2001-01-24 AT AT01953031T patent/ATE264520T1/de not_active IP Right Cessation
  • 2001-01-24 EP EP01953031A patent/EP1208424B1/en not_active Expired - Lifetime
  • 2001-01-24 WO PCT/EP2001/000775 patent/WO2001061469A2/en not_active Ceased
  • 2001-01-24 DE DE2001602777 patent/DE60102777T2/de not_active Expired - Lifetime

Also Published As

Similar Documents

Legal Events