US20020112142A1

US20020112142A1 - Implementation of a conditional move instruction in an out-of-order processor

Info

Publication number: US20020112142A1
Application number: US09/195,121
Authority: US
Inventors: Joel Springer Emer; Bruce Edwards; Daniel Lawrence Leibholz; Edward J. McLellan; Derrick R. Meyer
Original assignee: Compaq Information Technologies Group LP
Current assignee: Hewlett Packard Enterprise Development LP; Compaq Information Technologies Group LP
Priority date: 1998-11-18
Filing date: 1998-11-18
Publication date: 2002-08-15
Anticipated expiration: 2018-11-18
Also published as: US6449713B1

Abstract

A technique for handling a conditional move instruction in an out-of-order data processor. The technique involves detecting a conditional move instruction within an instruction stream, and generating multiple instructions according to the detected conditional move instruction. The technique further involves replacing the conditional move instruction within the instruction stream with the generated multiple instructions. The generated multiple instructions are generated such that each of the generated multiple instructions executes using no more than two input ports of an execution unit. The generated multiple instructions include a first generated instruction that produces a condition result indicating whether a condition exists, and a second generated instruction that inputs the condition result as a portion of an operand which identifies a register of the out-of-order data processor. The second generated instruction performs a first move operation when the condition is determined to exist, and a second move operation when the condition is determined not to exist.

Description

FIELD OF THE INVENTION

The present invention relates generally to data processing and in particular to techniques for processing a conditional move instruction within a data processor.

BACKGROUND OF THE INVENTION

In general, data processors are capable of executing a variety of instructions. One type of instruction is called a conditional move instruction. From a programmer's perspective, a typical conditional move instruction instructs a processor to test whether a particular condition exists (e.g., whether a particular register stores zero), and to move information into a destination register if the particular condition exists. If the

CMOVXX S_R_A, S_R_B, D_R_C,

where “CMOVXX” indicates that the instruction is a conditional move instruction that tests for a condition “XX”. “S_R _A” and “S_R_B” are source operands that respectively identify registers R_Aand R_B. “D_R_C” is a destination operand that identifies register R_C.

In general, how a processor uses registers depends on whether the processor is capable of executing instructions out of program order. For a processor that cannot execute instructions out of program order (i.e., an in-order processor), instruction source and destination operands typically identify physical registers within the processor. The pseudo-code for executing the CMOVXX instruction in an in-order processor is as follows:

if (XX(R_A)), then R_C=R_B.

According to the pseudo-code, the processor determines whether a condition XX exists involving physical register R _A(e.g., whether physical register R_Astores zero). If the condition XX exists, the processor moves the contents of physical register R_Binto physical register R_C. Otherwise, the processor leaves the original contents of physical register R_Cunaltered.

In a processor that is capable of executing instructions out of program order (i.e., an out-of-order processor), instruction source and destination operands typically identify logical registers instead of the physical registers directly. The out-of-order processor maps these logical registers to physical processor registers just before instruction execution such that the result of each instruction is stored in a new physical register. This approach enables the processor to avoid problems when executing instructions out of program order (e.g., read-after-write data hazards).

The pseudo-code for executing a CMOVXX instruction in an out-of-order processor is therefore somewhat more complex. Suppose that, prior to mapping the CMOVXX instruction, the out-of-order processor maps logical register R _Ato physical register R_A1, logical register R_Bto physical register R_B1, and logical register R_Cto physical register R_C1. Additionally suppose that, after mapping the CMOVXX instruction, the out-of-order processor maps logical register R_Cto physical register R_C2(an new physical register). The pseudo-code for executing the CMOVXX instruction in such a processor is therefore as follows:

if (XX(R_A1)), then R_C2=R_B1else R_C2=R_C3.

According to the pseudo-code, the out-of-order processor determines whether a condition XX exists involving physical register R _A1(logical register R_A). If the condition XX exists, the processor moves the contents of physical register R_B1(logical register R_B) into physical register R_C2(to which logical register R_Cpresently is mapped). As such, the contents of logical register R_Bare stored in logical register R_C. If the condition XX does not exist, the processor moves the contents of physical register R_C1(to which logical register R_Cpreviously was mapped) into physical register R_C2such that a programmer perceives the contents of logical register R_Cas remaining unaltered.

SUMMARY OF THE INVENTION

When a processor executes an instruction within an instruction stream, an execution circuit (or unit) of the processor receives instruction data through input ports, and executes the instruction according to the instruction data. For example, an execution unit of an in-order processor may execute the conditional move instruction:

CMOVXX S_R_A, S_R_B, D_R_C

according to the pseudo-code:

if (XX(R_A)), then R_C=R_B

where R _A, R_Band R_Crefer to physical registers within the in-order processor. To receive instruction data used by the CMOVXX instruction, the execution unit requires only two input ports: a first port to receive the contents of physical register R_A, and a second port to receive the contents of physical register R_B.

However, an execution unit of an out-of-order processor executes the CMOVXX instruction according to the following pseudo-code:

if (XX(R_A1)), then R_C2=R_B1, else R_C2=R_C1

where R _A1, R_B1, R_C1and R_C2refer to physical registers within the in-order processor. To implement this instruction, the out of order execution unit requires three input ports: a first port to receive the contents of physical register R_A1, a second port to receive the contents of physical register R_B1, and a third port to receive the contents of physical register R_C1.

There are disadvantages to a processor that uses three input ports to execute instructions. In particular, such a processor would require substantial semiconductor resources (e.g., a disproportionately large area for input port routing). Additionally, processors typically use no more than two input ports to execute non-conditional move instructions. Accordingly, processor designers generally prefer to limit the number of input ports for each instruction to no more than two. Unfortunately, as explained above, a conventional implementation the CMOVXX instruction within an out-of-order processor uses three input ports.

In contrast, an embodiment of the present invention is directed to a technique for handling a conditional move instruction in an out-of-order data processor. The technique involves detecting a conditional move instruction within an instruction stream, and generating multiple instructions according to the detected conditional move instruction. The technique further involves replacing the conditional move instruction within the instruction stream with the generated multiple instructions. Preferably, each of the generated multiple instructions executes using no more than two input ports. As such, it is unnecessary for the processor to use three input ports to execute the instructions.

The generation of multiple instructions preferably involves providing a first generated instruction that determines whether a condition exists, and providing a second generated instruction that performs a move operation based on whether the condition exists. In particular, the second generated instruction performs a first move operation when the condition is determined to exist, and a second move operation when the condition is determined not to exist. When the condition exists, the first move operation loads a new physical register with contents from a specified source register so that, from a programmer's perspective, the processor alters a logical register mapped to the new physical register. When the condition does not exist, the second move operation loads the new physical register with contents of a previously used physical register (to which the logical register was previously mapped) so that, from the programmer's perspective, the processor leaves the logical register unaltered.

Instruction generation may involve providing a first generated instruction that produces a condition result, and providing a second generated instruction that (i) inputs the condition result from a first portion of a register that is separate from a second portion that stores standard contents of the register, and (ii) performs an operation according to the first portion. To this end, the mechanisms for storing the condition result and the standard contents are treated as a single entity (e.g., a register with an extra bit field to store the condition result) rather than as separate registers. As such, the same circuitry for addressing and accessing the standard portion of the registers can be used to address and access the condition field. This feature allows the processor to transfer the condition result through one of two existing input ports alleviating the need for a third input port to carry the condition result. In particular, the processor includes a register file containing instruction registers, each of which has a standard field and a condition field.

Prior to detecting the conditional move instruction, instructions may be loaded from memory in groups (e.g., fetch blocks). In particular, the technique may further involve retrieving a first group of instructions from a memory during a first fetch period, the first group of instructions including the conditional move instruction. Such a retrieval enables instructions to be loaded using less retrieve operations than loading instructions individually.

Other subsequent groups of instructions may be loaded as well. For example, the technique may further include retrieving a second group of instructions from the memory during a second fetch period, the second group following the first group within the instruction stream. The technique may involve retrieving the second group of instructions from the memory again during a third fetch period while the multiple instructions are generated simultaneously. This feature provides an optimization in the sense that, retrieval of the second group of instructions during the third fetch period will make the second group of instructions available at a convenient point in the pipeline to receive one of the generated multiple instructions.

Alternatively, the technique may involve overwriting the conditional move instruction in the retrieved first group of instructions with one of the generated multiple instructions, and overwriting an instruction following the conditional move instruction in the retrieved first group of instructions with another of the generated multiple instructions. In this situation, the instruction following the conditional move instruction is preferably a blank instruction that performs no operation when executed. Accordingly, the processor simply modifies the fetch block containing the conditional move instruction without affecting a subsequent fetch block.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention. [0021]
FIG. 1 is a block diagram of an instruction pipeline for a data processor in which the present invention may be used. [0022]
FIG. 2 is a block diagram of a portion of an instruction fetch stage of FIG. 1 that detects a conditional move instruction within an instruction stream. [0023]
FIG. 3 is a block diagram of a portion of the instruction fetch stage of FIG. 1 that generates multiple instructions according to the detected conditional move instruction, and replaces the conditional move instruction within the instruction stream with the generated multiple instructions. [0024]
FIG. 4A is a block diagram of instructions before and after being handled in a first manner by the instruction fetch stage of FIG. 1. [0025]
FIG. 4B is a block diagram of instructions before and after being handled in a second manner by the instruction fetch stage of FIG. 1. [0026]
FIG. 5 is a flow diagram of a procedure performed by the instruction fetch stage of FIG. 1. [0027]
FIG. 6 is a block diagram of execution circuitry within an instruction execution stage of FIG. 1 that executes the generated multiple instructions.[0028]

DETAILED DESCRIPTION OF A PREFERRED EMBODIMENT

The present invention involves detecting a conditional move instruction within an instruction stream, and replacing it with multiple replacement instructions such that a data processor processing the instruction stream executes the multiple replacement instructions rather than the original conditional move instruction. Preferably, the data processor uses no more than two input ports when executing each of the multiple instructions so that additional processor resources (e.g. a third input port for each instruction) are unnecessary. [0029]
The invention is preferably used in an instruction pipeline of a speculative execution out-of-order data processor such as the pipeline [0030] 10 shown in FIG. 1. The pipeline 10 has a series of stages including an instruction fetch stage 12, an instruction slot stage 14, an instruction map stage 16, an instruction issue/queue stage 18, an instruction read stage 20, an instruction execution stage 22, an instruction write stage 24, and an instruction retire stage 26.
The pipeline [0031] 10 processes a stream of instructions 28. First, the instruction fetch stage 12 retrieves the instructions from memory. Second, the instruction slot stage 14 determines to which execution unit the instructions should be sent, e.g., a floating point unit or an integer unit (not shown). Third, the instruction map stage 16 maps the instructions such that the instructions refer to physical registers rather than logical registers. Fourth, the instruction issue/queue stage 18 queues the instructions for execution. Fifth, the instruction read stage 20 reads data used by the instructions from the physical registers. Next, the instruction execution stage 22 executes the instructions. Then, the instruction write stage 24 stores results of the executed instructions in the physical registers. Finally, the instruction retire stage 26 retires the instructions by committing the processor state to the results of the instructions.
FIG. 2 shows a [0032] circuit portion 30 of the instruction fetch stage 12 that retrieves the instructions 28 from a memory 32 (e.g., main memory or a second-level cache), and temporarily stores the retrieved instructions 28 in an instruction cache (or ICACHE) 34. The circuit portion 30 includes a program counter circuit 36 and a detect circuit 38. The program counter circuit 36 provides program counter information (e.g., a FILL PC pointer) identifying locations within the memory 32 that store instructions to be retrieved. The detect circuit 38 reads instructions from the memory 32 based on the program counter information, scans the retrieved instructions for any conditional move instructions (e.g., CMOVXX), and stores the instructions and scan results in the instruction cache 34. In particular, the detect circuit 38 groups the instructions into fetch blocks (e.g., fetch block 40), generates a conditional move code for each fetch block (e.g., conditional move code 42) indicating the locations of any conditional move instructions within that fetch block, and stores each fetch block and its corresponding conditional move code as an entry of the instruction cache 34 (e.g., entry 44). By way of example, the conditional move code 42 has the binary value “0100” to indicate that the second instruction of fetch block 40 is a conditional move instruction, as shown in FIG. 2.
Each entry within the [0033] instruction cache 34 further includes cache tag information (e.g., TAG) indicating whether that entry is valid or invalid. When the tag information indicates that the entry is valid, a read attempt of that entry results in a cache hit. If the tag information indicates that the entry is invalid, a read attempt of that entry results in a cache miss.
FIG. 3 shows a circuit portion [0034] 50 of the instruction fetch stage 12 that provides instructions from the instruction cache 34 to other circuits in the pipeline 10. If a conditional move instruction exists within the instructions, the circuit portion 50 generates multiple instructions according to the conditional move instruction, and replaces the conditional move instruction with the generated multiple instructions.
The circuit portion [0035] 50 includes a PC latch 52, a PC multiplexer 54, a PC silo 56, an instruction latch 58, an instruction sequencer 60, an instruction counter 62, and conditional move logic (or CMOVXX logic) 64. The PC latch 52, the PC multiplexer 54 and PC silo 56 (hereinafter generally referred to as PC circuitry) operate to provide program counter information identifying instruction cache entries (e.g., entry 44) to be transferred out of the instruction cache 34. The instruction latch 58 holds the fetch blocks from the identified entries, and provides them to the CMOVXX logic 64. Simultaneously, the instruction sequencer 60 retrieves the corresponding conditional move codes from the identified entries, and controls the operation of the PC circuitry and the CMOVXX logic 64 based on the retrieved conditional move codes. In particular, when a conditional move code indicates that its corresponding fetch block does not include a conditional move instruction, the instruction sequencer 60 signals the CMOVXX logic 64 simply to pass the fetch block from the instruction latch 58 to circuits further down the pipeline 10 (e.g., a register mapper 68). Additionally, the instruction sequencer 60 signals the PC circuitry to continue providing a program counter signal (NEXT PC) received on an input 72 of the PC multiplexer 54 so that another entry of the instruction cache 34 can be identified for transfer.
However, when a conditional move code indicates that its corresponding fetch block includes a conditional move instruction, the [0036] instruction sequencer 60 signals the CMOVXX logic 64 (i) to generate multiple instructions (i.e., CMOV1XX and CMOV2XX), and (ii) to replace the conditional move instruction with the generated multiple instructions. In response, as shown in FIG. 4A, the CMOVXX logic 64 forms two copies of the fetch block (e.g., fetch block 40) containing the conditional move instruction. In the first copy (fetch block 74A), the CMOVXX logic 64 overwrites the conditional move instruction with one of the multiple instructions (CMOV1XX), and invalidates any instructions in the first copy that follow the conditional move instruction. In the second copy (fetch block 74B), the CMOVXX logic 64 overwrites the conditional move instruction with another of the multiple instructions (CMOV2XX), and invalidates any instructions preceding the conditional move instruction in the second copy. As a result, the CMOVXX logic creates two fetch blocks that preserve the fetch block positions of the non-conditional move instructions, and that have the conditional move instruction replaced with the multiple generated instructions (CMOV1XX and CMOV2XX).
With reference again directed to the circuit portion [0037] 50 and FIG. 3, when the instruction sequencer 60 signals the CMOVXX logic 64 to convert a fetch block containing a conditional move instruction into two fetch blocks, the instruction sequencer 60 signals other circuits of the event so that they may adjust their operation accordingly. In particular, the instruction sequencer 60 signals the PC circuitry to provide extra time for the CMOVXX logic 64 to convert one fetch block into the two conditional move fetch blocks 74. The PC circuitry responds by repeating previously provided program counter information enabling the instruction latch 58 to read twice a fetch block of the entry following the entry having the conditional move instruction. When the instruction latch 58 reads this fetch block the first time, the CMOVXX logic 64 ignores it since this read coincides with formation of the second copy of the two conditional move fetch blocks 74. However, when the instruction latch 58 reads this fetch block the second time, the CMOVXX logic 64 processes it in a normal fashion.
It should be understood that the pipeline [0038] 10 is preferably capable of speculative execution of instructions since the processor is an out-of-order data processor. The pipeline 10 includes silos for storing prior processor states so that the pipeline can return to a previous state when instruction execution occurs down an incorrect instruction branch. In particular, the pipeline 10 includes a PC silo 56 that stores prior ICACHE addresses (e.g, FILL PC), and a register silo 70 that stores prior logical register to physical register mappings (or assignments).
When the [0039] instruction sequencer 60 signals the CMOVXX logic 64 to convert fetch block containing a conditional move instruction into two fetch block with replaced instructions, the instruction sequencer 60 signals the silos (e.g., the PC silo 56 and the register silo 70) of the event. In particular, the instruction sequencer 60 signals the instruction counter 62 which, in turn, updates the PC silo 56 and the register silo 70. Accordingly, if the pipeline 10 executes down an incorrect instruction branch and attempts to recover, the pipeline 10 will have accounted for the conversion of the one fetch block containing a conditional move instruction into two fetch blocks.
As an alternative to creating two fetch blocks, one fetch can be used, as shown in FIG. 4B. In particular, when one or more programs are compiled to form an executable, the compiler can append a blank instruction (e.g., a NO-OP instruction) after the conditional move instruction within the executable. When the processor executes the executable, the [0040] CMOVXX logic 64 can simply modify the instruction stream (e.g., fetch block 92) by replacing the CMOVXX instruction with the CMOV1XX instruction, and replacing the subsequent blank instruction with the CMOV2XX instruction (e.g., fetch block 94). In this situation, it is unnecessary to signal other circuits (e.g., the PC circuitry and silos) to account for a change in the number of fetch blocks in the pipeline 10.
FIG. 5 shows a flow diagram of a [0041] procedure 80 performed by the circuit portions 30,50 of the instruction fetch stage 12. In step 82, the detect circuit 38 reads a group of instructions (e.g., a fetch block) from the memory 32 (see FIG. 2). In step 84, the detect circuit 38 determines whether the group includes any conditional move instructions. If the group does not include any conditional move instructions, step 84 proceeds to step 86, which involves providing the group of instructions to other circuits (e.g., to the register mapper 68) further down the pipeline 10 (see FIG. 3). However, in step 88, if the group includes a conditional move instruction, the CMOVXX logic 64, under control of the instruction sequencer 60 which reads the conditional move code provided by the detect circuit 38, generates multiple instructions according to the conditional move instruction (i.e., the multiple instructions preserve the “XX” operation of the CMOVXX instruction), and replaces the conditional move instruction within the instruction stream with the generated multiple instructions. In particular, the CMOVXX logic 64 performs the replacement in a manner that preserves the instruction positions of the non-conditional move instructions within the fetch blocks. Step 90, which follows steps 86 and 88, loops back to step 82 to handle more instructions within the instruction stream, unless the procedure 80 is terminated (e.g., due to a reset or power down of the processor).
Further details of how the multiple instructions execute within the pipeline [0042] 10 will now be provided. The pipeline 10 includes a register file that includes physical processor registers (see physical registers R_A1, R_B1, R_C1, R_C2and R_C3in FIG. 6). Each of the registers includes a standard field for storing a standard register value (e.g., a 64-bit value), and a predicate (or condition) field (e.g., a single bit). The standard field corresponds to what programmers commonly refer to as the contents of the register. The predicate field is a special field that is preferably used only by the multiple instructions replacing the CMOVXX instruction. That is, the predicate field is preferably not readable directly by the programmers.
The conditional move instruction, which is replaced by the multiple instructions generated by the [0043] CMOVXX logic 64, has the following format:
CMOVXX S_R_A, S_R_B, D_R_C
where S_R[0044] _Aand S_R_Bidentify logical source registers R_Aand R_B, respectively, and D_R_Cidentifies a logical destination register R_Cwithin the processor. When the CMOVXX logic 64 encounters such an instruction, the CMOVXX logic 64, under control of the instruction sequencer 60, generates two instructions, the first of which has the following format:
CMOV1XX S_R_A, S_R_C, D_R_C
where CMOV[0045] 1XX indicates that the instruction is a first instruction generated from the CMOVXX instruction, S_R_Aand S_R_Cidentify logical source registers R_Aand R_C, respectively, and D_R_Cidentifies a logical destination register R_Cwithin the processor. “XX” within “CMOV1XX” indicates that the CMOV1XX instruction performs the same type of operation (or function) as that of the CMOVXX instruction (e.g., checking whether the contents of a particular register equal zero). The pseudo-code for the CMOV1XX instruction is as follows:
R_C3.P=xx(R_A1); R_C2=R_C1
where R[0046] _A1and R_C1are physical registers respectively mapped to logical registers R_Aand R_Cprior to mapping the CMOV1XX instruction, R_C2is a physical register mapped to logical register R_Cafter mapping the CMOV1XX instruction, and R_C2.P is a predicate field of the physical register R_C2.
The second instruction has the following format:[0047]
CMOV2XX S_R_B, S_R_C, D_R_C
where CMOV[0048] 2XX indicates that the instruction is a second instruction generated from the CMOVXX instruction, S_R_Band S_R_Cidentify logical source registers R_Band R_C, respectively, and D_R_Cidentifies a logical destination register R_Cwithin the processor. The pseudo-code for the CMOV2XX instruction is as follows:
if (R_C2.P) R_C3=R_C2else R_C3=R_B1
where R[0049] _B1and R_C2are physical registers respectively mapped to logical registers R_Band R_Cafter mapping the CMOV1XX instruction and prior to mapping the CMOV2XX instruction, R_C3is a physical register mapped to logical register R_Cafter mapping the CMOV2XX instruction, and R_C2.P is the predicate field of the physical register R_C2.
FIG. 6 shows an [0050] execution circuit 100 of the instruction execution stage 22 that operates during execution of the CMOV1XX and CMOV2XX instructions. The execution circuit 100 includes a CMOV1XX circuit 102 and a CMOV2XX circuit 104 that connect with physical registers of the register file. The CMOV1XX circuit 102 has a first input port 106 that receives the contents of the standard field of physical register R_A1, and a second input port 108 that receives the contents of the standard field of physical register R_C1. The CMOV2XX circuit 104 has a first input port 110 that receives the contents of the standard field of physical register R_C3, and a second input port 112 that receives the contents of the standard field of physical register R_B1.
The input port [0051] 110 further receives the predicate field of physical register R_C2. In particular, the predicate field of physical register R_C3(i.e., R_C2.P) passes through a connection 110 _A, and the standard field of physical register R_C2passes through a set of connections 110 _B.
The [0052] CMOV1XX circuit 102 includes an evaluation block 114 and a pass-thru block 116. The evaluation block 114 evaluates the contents of physical register R_A1using the function XX (e.g., equal to, greater than, less than, not equal to, etc.). The pass-thru block 116 transfers the standard field of physical register R_C1to the standard field of physical register R_C2.
The [0053] CMOV2XX circuit 104 includes a multiplexer 118 that selects between the standard field of physical register R_C2(the connection 110 _Bof input port 110) and the standard field of physical register R_B1(input port 112) according to the predicate field of physical register R_C2(the connection 110 _Aof input port 110). The multiplexer 118 outputs the contents of the selected standard field to the standard field of physical register R_C3. Accordingly, if the predicate field of physical register R_C2indicates that condition XX exists at physical register R_A1, the multiplexer 118 transfers the standard field of physical register R_C2to the standard field of physical register R_C3. On the other hand, if the predicate field of physical register R_C2indicates that the condition XX does not exist at physical register R_A1, the multiplexer 118 transfers the standard field of physical register R_B1to the standard field of physical register R_C3.
As is shown in FIG. 6, each of the CMOV[0054] 1XX instruction and the CMOV2XX instruction uses no more than two input ports. An extra connection (e.g., a bit line) is used rather than an entire third input port (multiple bit lines). Each register of the register file preferably has such an extra connection to provide access to the predicate field of that register. Such an arrangement provides substantial savings in semiconductor resources relative to providing each instruction with a third input port.

EQUIVALENTS

While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims. [0055]
For example, the circuitry shown in FIG. 6 can be optimized to bypass physical register R[0056] _C2such that the output of the evaluation block 114 goes directly to the multiplexer 122 through input port 112 ₂. Such an optimization removes the steps of storing a value in the predicate field of physical register R_C2, and subsequently reading the predicate field of physical register R_C2.
Furthermore, it should be understood that the invalidated instructions of the fetch [0057] blocks 74A and 74B in FIG. 4A may be blank instructions (e.g., NO-OP instructions). Alternatively, the invalidated instructions may be the original instructions (e.g., INSTA, INSTB and INSTC) with corresponding flags (not shown) set to indicated to the processor that these instructions are invalid.

Claims

What is claimed is:

1. A method for handling a conditional move instruction in an out-of-order data processor, comprising the steps of:

detecting a conditional move instruction within an instruction stream;

generating multiple instructions according to the detected conditional move instruction; and

replacing the conditional move instruction within the instruction stream with the generated multiple instructions.

2. The method of claim 1, wherein the step of generating includes the step of:

forming the multiple instructions such that each of the multiple instructions executes using no more than two input ports of an execution unit.

3. The method of claim 1, wherein the step of generating includes the step of:

providing a first generated instruction that determines whether a condition exists; and

providing a second generated instruction that performs a first move operation when the condition is determined to exist, and a second move operation when the condition is determined not to exist.

4. The method of claim 1, wherein the step of generating includes the step of:

providing a condition result in a first portion of a register, the first portion of the register being separate from a second portion of the register that stores standard register contents; and

providing a second generated instruction that (i) inputs the condition result from the first portion of the register, and (ii) performs an operation according to the condition result.

5. The method of claim 1, further comprising the steps of:

retrieving a first group of instructions from a memory during a first fetch period, the first group of instructions including the conditional move instruction; and

retrieving a second group of instructions from the memory during a second fetch period, the second group following the first group within the instruction stream.

6. The method of claim 5, further comprising the step of:

retrieving the second group of instructions from the memory again during a third fetch period while the multiple instructions are generated simultaneously.

7. The method of claim 1, wherein the step of replacing includes the step of:

overwriting the conditional move instruction with one of the generated multiple instructions; and

overwriting an instruction following the conditional move instruction in the instruction stream with another of the generated multiple instructions.

8. The method of claim 7, wherein the instruction following the conditional move instruction in the retrieved first group of instructions is a blank instruction that performs no operation when executed.

9. The method of claim 1, wherein the step of replacing includes the step of:

retrieving, from the memory, a first group of instructions that includes the conditional move instruction;

generating a second group of instructions having the conditional move instruction from the first group; and

revising the first and second groups of instructions such that (i) the conditional move instruction of the first group is replaced with one of the generated multiple instructions and any instructions following the conditional move instruction of the first group are invalidated, and (ii) the conditional move instruction of the second group is replaced with another of the generated multiple instructions and any instructions ahead of the conditional move instruction of the second group are invalidated.

10. The method of claim 1, further comprising the step of:

associating a same program counter value with each of the generated multiple instructions such that the generated multiple instructions are identifiable when speculative execution occurs down an incorrect instruction branch.

11. The method of claim 1, further comprising the step of:

generating a code that identifies a position of the conditional move instruction within a group of instructions.

12. A pipeline circuit for handling a conditional move instruction in an out-of-order data processor, comprising:

a detect circuit that detects a conditional move instruction within an instruction stream; and

a control circuit, coupled to the detect circuit, that generates multiple instructions according to the detected conditional move instruction, and replaces the conditional move instruction within the instruction stream with the generated multiple instructions.

13. The pipeline circuit of claim 12, wherein the control circuit includes:

an instruction forming circuit that forms the multiple instructions such that each of the multiple instructions executes using no more than two input ports of an execution unit.

14. The pipeline circuit of claim 12, wherein the control circuit includes an output that provides, as the generated multiple instructions:

a first generated instruction that determines whether a condition exists; and

a second generated instruction that performs a first move operation when the condition is determined to exist, and a second move operation when the condition is determined not to exist.

15. The pipeline circuit of claim 12, wherein the control circuit includes an output that provides, as the generated multiple instructions:

a first generated instruction that produces a condition result in a first portion of a register, the first portion of the register being separate from a second portion of the register that stores standard register contents; and

a second generated instruction that (i) inputs the condition result from the first portion of the register, and (ii) performs an operation according to the condition result.

16. The pipeline circuit of claim 15, further comprising:

a register file that includes multiple registers, each of the multiple registers having a standard field and a condition field, a particular one of the multiple registers being the register having the first and second portions, the first portion being a condition field and the second portion being a standard field.

17. The pipeline circuit of claim 12, wherein the control circuit includes:

a sequencing circuit that (i) retrieves a first group of instructions from a memory during a first fetch period, the first group of instructions including the conditional move instruction, and (ii) retrieves a second group of instructions from the memory during a second fetch period, the second group following the first group within the instruction stream.

18. The pipeline circuit of claim 17, wherein the sequencing circuit is further adapted to retrieve the second group of instructions from the memory again during a third fetch period while the multiple instructions are generated simultaneously.

19. The pipeline circuit of claim 12, wherein the control circuit further includes:

instruction logic that (i) overwrites the conditional move instruction with one of the generated multiple instructions, and (ii) overwrites an instruction following the conditional move instruction in the instruction stream with another of the generated multiple instructions.

20. The pipeline circuit of claim 19, wherein the instruction following the conditional move instruction in the retrieved first group of instructions is a blank instruction that performs no operation when executed.

21. The pipeline circuit of claim 12, wherein the control circuit includes circuitry that:

generates a second group of instructions having a conditional move instruction from a first group of instructions having the conditional move instruction; and

revises the first and second groups of instructions such that (i) the conditional move instruction of the first group is replaced with one of the generated multiple instructions and any instructions following the conditional move instruction of the first group are invalidated, and (ii) the conditional move instruction of the second group is replaced with another of the generated multiple instructions and any instructions ahead of the conditional move instruction of the second group are invalidated.

22. The pipeline circuit of claim 12, wherein the control circuit includes:

an instruction sequencer that associates a same program counter value with each of the generated multiple instructions such that the generated multiple instructions are identifiable when speculative execution occurs down an incorrect instruction branch.

23. The pipeline circuit of claim 12, wherein the detect circuit is adapted to generate a code that identifies a position of the conditional move instruction within a group of instructions.

24. A method for handling a conditional move instruction in an out-of-order data processor, comprising the steps of:

detecting a conditional move instruction within an instruction stream;

replacing the conditional move instruction within the instruction stream with the generated multiple instructions, the generated multiple instructions being generated such that each of the generated multiple instructions executes using no more than two input ports of an execution unit, the generated multiple instructions including a first generated instruction that produces a condition result indicating whether a condition exists, and a second generated instruction that inputs the condition result as a portion of an operand which identifies a register of the out-of-order data processor, the second generated instruction performing a first move operation when the condition is determined to exist, and a second move operation when the condition is determined not to exist.

25. A pipeline circuit for handling a conditional move instruction in an out-of-order data processor, comprising:

a control circuit, coupled to the detect circuit, that (i) generates multiple instructions according to the detected conditional move instruction, and (ii) replaces the conditional move instruction within the instruction stream with the generated multiple instructions, the generated multiple instructions being generated such that each of the generated multiple instructions executes using no more than two input ports of an execution unit, the generated multiple instructions including a first generated instruction that produces a condition result indicating whether a condition exists, and a second generated instruction that inputs the condition result as a portion of an operand which identifies a register of the out-of-order data processor, the second generated instruction performing a first move operation when the condition is determined to exist, and a second move operation when the condition is determined not to exist.