JP4889235B2 - Program control processor - Google Patents

Description

The present invention relates to a program system processor control to use is that the computer for skip processing to invalidate subsequent instructions if the conditions by condition determination is satisfied.

As a method for controlling conditional branching of a program in a CPU (processor),
(I) The instruction set has a conditional branch instruction or the like, and when the condition is satisfied, a method of jumping to a specified address,
(Ii) There is a method in which a skip instruction is included in the instruction set, and the immediately following one or a plurality of consecutive instructions are invalidated. For example, Patent Document 1 discloses a technique for converting a command into a NOP (no operation) when skipping.
JP-A 61-221939

  In both cases (i) and (ii), the number of execution cycles differs between when the condition is satisfied and when the condition is not satisfied. Therefore, in a program and its system in which the instruction execution timing is important, after branch processing (instruction NOP conversion) There is a problem that the instruction (after the skip processing) cannot be executed at the same cycle timing.

In the method (ii), for example, a load / store instruction having a data size exceeding the bus width requires a plurality of cycles to execute the instruction.
The present invention has been made paying attention to the above problems, and an object of the present invention is to provide a program control method and a processor capable of making an execution cycle constant regardless of the validity / invalidity of instructions.

The invention according to claim 1 is a processor comprising skip processing control means for invalidating one or a plurality of consecutive instructions after a condition is established based on a condition determination, wherein the skip processing control The gist of the means is a processor that keeps prohibiting data writing to the inside and outside of the processor regardless of the type of instruction during the skip processing.
The skip processing control means includes a skip state flag for controlling writing of data by the instruction, and the skip state flag is set when a condition is satisfied in a conditional branch instruction permitted to branch forward only. Set to skip state.
Further, the skip processing control means compares the branch destination address register for holding the branch destination address specified by the conditional branch instruction, the branch destination address of the branch destination address register and the current address of the program counter. Address comparison means, flag data saving means for saving flag data of the skip state flag to a register or memory when an interrupt occurs, and flag data return means for returning the saved flag data to a skip state flag by a return instruction This address comparison means
The skip state flag is reset to the normal state with a branch destination address ≦ the current address as a trigger.
The skip processing control means continues to prohibit the writing of data to / from the processor regardless of the type of instruction during the skip processing in which the skip state flag is set to the skip state.

According to the first aspect of the present invention, the skip processing control means keeps prohibiting the writing of data to the inside and outside of the processor regardless of the type of instruction during the skip process, so that the instruction is invalid even when the instruction is invalid. By consuming the number of cycles, the execution cycle can be made constant regardless of whether the instruction is valid or invalid.
Furthermore, the flag data of the skip state flag is registered when an interrupt occurs.
Alternatively, it is possible to continue the skip processing at the time of interruption by providing flag data saving means for saving to the memory and flag data return means for returning the saved flag data to the skip state flag by a return instruction.

As described in claim 2, the processor of claim 1, wherein the address comparison means is constituted by the interior of the arithmetic logic unit of the data path, if the skip state flag is set to skip state Inputs the branch destination address of the branch destination address register and the current address of the program counter to the arithmetic logic unit, and the condition code output from the arithmetic logic unit becomes branch destination address ≦ current address When the skip state flag is reset to the normal state with this as a trigger, an arithmetic logic unit inside the data path is used as a configuration for comparing the branch destination address of the branch destination address register with the current address of the program counter. To reduce the size by utilizing resources It can be.

(First comparative example )
Hereinafter , a first comparative example will be described with reference to the drawings.
FIG. 1 is a block diagram showing an internal configuration of a CPU (processor) in the microcomputer of this comparative example . FIG. 2 shows an example of CPU instruction execution operation, particularly pipeline operation.

  In FIG. 1, the CPU 10 includes a control unit 20 and a data path 30. The data path 30 stores data and performs arithmetic / logical operations. The control unit 20 controls the data path 30 according to the instruction fetched from the memory.

  The control unit 20 includes an instruction decoder 21 and a skip processing control unit 22. The instruction decoder 21 stores the instruction fetched from the memory and decodes the fetched instruction. The skip processing control unit 22 as the skip processing control means is for performing a skip processing for invalidating one or a plurality of consecutive instructions after the condition is satisfied when the condition is determined in the conditional branch. is there. The skip processing control unit 22 determines from the skip processing start signal and the skip processing end signal output from the instruction decoder 21 whether or not it is in the skip state, and controls data writing to the inside and outside of the CPU 10.

  The data path 30 includes a load / store unit 31, an execution unit 32, a register file 33, a program status register 34, and a program counter 35. The load / store unit 31 controls data reading from the outside of the CPU 10 and data writing to the outside. The execution unit 32 executes predetermined arithmetic / logical operations in accordance with the control signal output from the control unit 20 in accordance with the instruction decoded by the instruction decoder 21. The register file 33 stores data obtained as a result of execution by the execution unit 32 or data taken from outside the CPU 10 such as a memory. The program status register 34 holds the processing status of the CPU 10. The program counter 35 indicates the address of the instruction to be fetched.

  Here, the bus connecting the CPU 10 and peripheral circuits outside the CPU 10 is composed of two types of buses, a data bus 40 and an address bus 41, and the bus architecture of the CPU 10 in FIG. 1 is a Neumann bus configuration. Yes.

As shown in FIG. 2, the CPU 10 executes the program by pipeline processing including five types of stages.
The five types of stages are an IF (instruction fetch) stage, a DEC (decode) stage, an EXE (execution) stage, an MA (memory access) stage, and a WB (write back) stage. The IF (instruction fetch) stage is a stage for reading an instruction from a memory. The DEC (decode) stage is a stage for decoding the instruction read in the IF stage. The EXE (execution) stage is a stage that calculates an address of an instruction to be read in an arithmetic / logical operation or IF stage, an address when accessing a memory, and the like according to the contents of the instruction decoded in the DEC stage. The MA (memory access) stage is a stage for accessing the memory using the operation result in the EXE (execution) stage as an address in accordance with the contents of the instruction decoded in the DEC stage. The WB (write back) stage is a stage in which operation data or data taken from a memory is written to an internal register in accordance with the contents of the instruction decoded in the DEC stage.

  In pipeline processing under the Neumann bus configuration, when a five-stage pipeline is constructed using the five types of stages described above, an IF stage for fetching instructions from the memory, and data writing or reading to the memory are performed. Do the MA stage to compete.

Therefore, there are various measures for avoiding the conflict between the IF stage and the MA stage. The pipeline operation of FIG. 2 as an example of conflict avoidance will be described.
In FIG. 2, the instructions A, B, D and skip instructions that do not require access to the memory are configured by four stages of IF, DEC, EXE, and WB, and after the instruction is executed (EXE stage) In the next stage, the result is written to the register (WB stage). On the other hand, for the instruction C that requires memory access, such as a store instruction, an MA stage is inserted between the EXE stage and the WB stage, and a pipeline is configured with five types of stages. However, the pipeline of other instructions is stalled during execution of the MA stage of instruction C. In this way, only two cycles are required for instructions that require memory access, but conflict between the IF stage and the MA stage can be avoided.

The skip processing procedure using the CPU configuration as described above will be described.
In FIG. 1, the value (address) of the program counter 35 is output from the address bus 41, and the corresponding instruction is output to the data bus 40 from the ROM connected to the bus 41. The CPU 10 takes in data on the data bus 40 via the load / store unit 31 and stores it in the instruction decoder 21. The instruction decoder 21 decodes the content of the fetched instruction, and performs predetermined control on the data path 30 and the outside of the CPU 10 according to the content of the decoded instruction.

  For example, when the fetched instruction is an instruction to add the data of the register R10 and the register R11 and write to the register R11, the control unit 20 inputs the data of the register R10 and the register R11 to the execution unit 32 and adds the data. In addition to instructing the operation, the execution result is input to the register file 33, and write permission is instructed to the register R11.

Here, if it is a conditional branch instruction, the following is performed. That is, the condition determination is performed based on the execution result of the preprocessing, and when jumping to the next instruction or less, the following is performed.
When the fetched instruction is a skip process start instruction, the instruction decoder 21 outputs a skip process start signal to the skip process control unit 22. If the fetched instruction is a skip process end instruction, the instruction decoder 21 outputs a skip process end signal to the skip process control unit 22.

  The skip processing control unit 22 includes a skip state flag 22a indicating whether it is a skip state or a normal state, and the skip state flag 22a is for controlling data writing by an instruction. The skip state flag 22a is set to the skip state upon receipt of the skip process start signal, and is reset to the normal state upon receipt of the skip process end signal.

  When the skip state flag 22a is set to indicate a skip state, an external write control signal that controls writing to the peripheral circuit including the memory regardless of the state of the write control signal output from the instruction decoder 21; In addition, write prohibition is output as an internal write control signal for controlling writing to an internal register such as the register file 33. As a result, when the subsequent instructions are invalidated in the skip processing, the instruction sequence control is continuously executed, and only data writing is prohibited.

  The skip processing control unit 22 as the skip processing control means described above will be described in more detail using the pipeline operation during the skip processing (FIGS. 2, 3, and 4) while comparing with a general example.

FIG. 2 shows an example in which the instruction A to the instruction D between the skip process start instruction and the skip process end instruction are skipped.
FIG. 3 shows a pipeline operation during general skip processing, and FIG. 4 shows a pipeline operation during skip processing in this comparative example .

First, the operation diagram of the general example of FIG. 3 will be described.
In a general skip process control method, after a skip process start instruction is executed, subsequent instructions are converted into NOP (no operation instructions). At this time, if the instruction C in FIG. 2 is a store instruction that requires two cycles to execute, there is no MA stage due to NOP conversion. Therefore, compared with the case where skip processing is not performed, the cycle is shortened, and the number of cycles (the number of execution cycles) related to the instructions A to D is “4”.

On the other hand, according to this comparative example of FIG. 4, even after the skip processing start instruction is executed, the instruction is not changed to NOP, and only data writing is prohibited.
The instruction C (store instruction) in FIG. 4 will be described as an example. The skip state flag is set to the skip state by executing the skip process start instruction, and is reset to the normal state by executing the skip process end instruction. The Even when the skip state flag is in the skip state, the skip processing control unit 22 continues to execute the sequence control of the instruction as it is, and disables writing of the write control signal. That is, during the skip process, data writing to and from the CPU continues to be prohibited regardless of the type of instruction. Therefore, data writing is prohibited at the MA stage. Therefore, the store instruction is substantially invalid, but the consumption cycle is the same two cycles as when the instruction is valid. Accordingly, the number of cycles (the number of execution cycles) related to the instructions A to D, which is the skip processing section, is “5”, which is the same as when the instruction is valid, regardless of the type of instruction. That is, the instruction execution cycle can be made constant regardless of the validity of the instruction, and the number of cycles for processing can be made the same when the instruction group of the instructions A to D is skipped or not. it can.

  As described above, in this method, even when the condition is satisfied as a result of the condition determination and the subsequent instructions are invalidated by the skip processing, the instruction sequence control is continued and only data writing is prohibited. Thus, a constant number of instruction cycles is consumed regardless of whether the condition determination is satisfied or not (the same instruction cycle can always be consumed regardless of the validity / invalidity of the instruction by skip processing control). That is, the execution timing of the instruction after completion of the skip processing can be made the same in both cases where the skip processing is executed and when it is not executed. For example, the skip processing section is between instruction 1 and instruction 2 (between instruction 1 and instruction 2 is a conditional branch section), which is very effective in a program in which the execution timing of instruction 2 with respect to instruction 1 is always constant. (When the instruction execution timing is important, it is possible to prevent deviations in the execution cycle due to conditional branching of the program).

As described above, this comparative example has the following features.
As a program control method, when performing a skip process that invalidates one or more consecutive instructions after the condition is satisfied, even if the subsequent instructions are invalidated by the skip process, instruction sequence control Was continuously executed, and only data writing was prohibited. As a result, even when the instruction is invalid, the same number of cycles as when the instruction is valid is consumed, and the execution cycle can be made constant regardless of the validity of the instruction.

  The CPU (processor) includes a skip processing control unit 22 that invalidates one or a plurality of successive instructions when a condition is established based on the condition determination, and skip processing control. During the skip process, the unit 22 keeps prohibiting data writing to the inside and outside of the processor (inside and outside the processor) regardless of the type of instruction such as a 1-cycle instruction or a 2-cycle instruction. Therefore, even when the instruction is invalid, the same number of cycles as when the instruction is valid is consumed, and the execution cycle can be made constant regardless of the validity of the instruction. In this way, a basic configuration for prohibiting data writing during skipping can be constructed.

Here, the skip processing control unit 22 includes a skip state flag (22a) for controlling writing of data by an instruction, and the skip state flag (22a) is set to a skip state by executing a skip processing start instruction. Since the normal state is reset by executing the skip processing end instruction, the skip state flag can be set / reset.
(Second comparative example )
Next, the second comparative example will be described focusing on the differences from the first comparative example .

FIG. 5 shows a pipeline operation diagram during the skip processing in this comparative example instead of FIG.
Compared to the first comparative example , in this comparative example , a conditional branch instruction is used as means for setting the skip state flag in FIG. 4 to the skip state.

  In FIG. 5, the instruction set has conditional branch instructions that are permitted to branch forward only. When the condition of the conditional branch instruction permitted to branch forward only is satisfied, a skip process start signal is output from the instruction decoder 21 to the skip process control unit 22. As a result, the skip state flag is set to the skip state, and subsequent instructions are invalidated. On the other hand, if the condition is not satisfied, the skip state flag remains in the normal state, and the subsequent instructions are executed. As described above, as a program control method, when a conditional branch instruction is executed, if the condition is satisfied, skip processing for invalidating subsequent instructions is started.

The control method of the skip processing control unit 22 as the skip processing control means is the same as the procedure in the first comparative example . When the branch does not branch (execute the instruction) and when the branch branches (invalidate the instruction) The number of cycles is the same.

Further, according to the program control method in this comparative example , an instruction for starting the skip process is not necessary, and the number of instructions can be reduced. That is, by starting skip processing with a conditional branch instruction, the number of instructions can be reduced compared to skip instructions. Also, skip processing instructions need not be implemented.

On the other hand, as a CPU (processor), the skip processing control unit 22 includes a skip state flag (22a) for controlling writing of data by an instruction, and the skip state flag (22a) is used when a condition in a conditional branch instruction is satisfied. Set to the skip state. This is preferable as another configuration for setting the skip state flag.
(First Embodiment)
Next, the first embodiment of the present invention will be described with a focus on differences from the second comparative example .

  FIG. 6 is a block diagram showing the internal configuration of the CPU in the present embodiment that replaces FIG. FIG. 7 shows a pipeline operation diagram during skip processing in this embodiment instead of FIG.

In the second comparative example , as a means for resetting the skip state flag to the normal state, in the present embodiment, the following is performed.
In FIG. 6, the CPU 10 includes a skip processing control unit 22 that invalidates one or a plurality of subsequent commands when a condition is established based on the condition determination, like the CPU 10 described in the first comparative example. ing.

  The CPU 10 executes a program by pipeline processing including five types of stages of IF, DEC, EXE, MA, and WB. The CPU 10 includes a data path 30 that executes data storage and arithmetic / logical operations, and a control unit 20 that controls the data path 30 according to an instruction fetched from the memory.

  The control unit 20 includes an instruction decoder 21 and a skip processing control unit 22. The instruction decoder 21 stores the instruction fetched from the memory and decodes the fetched instruction. The skip processing control unit 22 determines whether or not it is in the skip state from the skip processing start signal output from the instruction decoder 21 and the skip processing end signal (second skip processing end signal II) output from the data path 30. Judgment is made and data writing is controlled inside and outside the CPU 10.

  The data path 30 includes a load / store unit 31, an execution unit 32, a register file 33, a program status register 34, a program counter 35, a branch destination address register 36, and a pipeline register 37. The load / store unit 31 controls data reading from the outside of the CPU 10 and data writing to the outside. The execution unit 32 executes predetermined arithmetic / logical operations in accordance with the control signal output from the control unit 20 according to the instruction decoded by the instruction decoder 21. Here, the execution unit 32 has an ALU (arithmetic logic unit) 50, and this ALU 50 is also used during skip processing. The register file 33 stores data obtained as a result of execution by the execution unit 32 or data fetched from outside the CPU 10 such as a memory. The program status register 34 holds the processing status of the CPU 10. The program counter 35 indicates the address of the instruction to be fetched. The branch destination address register 36 holds a branch destination address in a conditional branch instruction. The pipeline register 37 uses the condition code signal, which is the output of the execution unit 32 (ALU 50), as the first skip processing end signal I, delays the signal by one cycle, and uses this signal as the second skip processing end signal. Output to the skip processing control unit 22 as II.

Further, the bus connection is the same as that of the CPU of the first comparative example , and has a Neumann bus configuration.
In the present embodiment, the skip processing control unit 22 and the data path 30 constitute a skip processing control means.

The skip processing procedure in the CPU configuration as described above will be described with reference to FIGS.
Similar to the second comparative example , the skip processing control unit 22 includes a skip state flag 22a. When the skip state flag 22a is in the skip state, the skip processing control unit 22 outputs write prohibition as an internal / external write control signal.

Similarly to the second comparative example , the skip state flag 22a decodes the conditional branch instruction by the instruction decoder 21, receives the skip processing start signal output when the condition is satisfied, and enters the skip state. Set.

  When the skip processing start signal is input, the skip processing control unit 22 in this embodiment stores the branch destination address calculated by the execution unit 32 in the branch destination address register 36 at the same time.

  Further, the skip processing control unit 22 of this embodiment controls the execution unit 32 while the skip state flag is in the skip state, and compares the branch destination address of the branch destination address register 36 with the current address of the program counter 35. To do. Specifically, the branch destination address of the branch destination address register 36 and the current address of the program counter 35 are input to an ALU (arithmetic logic unit) 50 in the execution unit 32, and a comparison operation is instructed to the execution unit 32. To do. Since the skip processing section by a conditional branch instruction (conditional branch instruction that is permitted to branch only forward) is from the next instruction of the conditional branch instruction to one before the branch destination address, When the branch destination address of the address register 36] = [current address of the program counter 35], the skip state flag may be reset. In the case of this configuration, the condition code signal that is the output of the execution unit 32 (ALU 50) is the first skip processing end signal I.

  However, in the case of a CPU that performs pipeline processing using a plurality of stages, there is a difference between the timing at which the operation result is written and the value (address) of the program counter 35, so that the pipeline register 37 for timing adjustment (in this case, 1 A signal delayed by using a delay of a cycle) is set as a second skip processing end signal II. In this way, only data writing can be prohibited up to the instruction immediately before the branch destination address (the same number of cycles as when the instruction is valid can be consumed).

  Here, the configuration in which the execution unit 32 processes the comparison of the address of the program counter 35 and the address of the branch destination address register 36 is intended to reduce the CPU size by sharing the arithmetic unit. Since the instruction is invalid while the skip state flag is in the skip state, the execution unit 32 can be used effectively for comparison between the branch destination address of the branch destination address register 36 and the current address of the program counter 35. is there.

The operation in which the skip state flag is set and reset by the conditional branch instruction will be described in detail with reference to FIG.
A conditional branch instruction (conditional branch instruction permitted to branch only forward) at the address 100 in the memory is fetched at cycle number # 1 and executed at cycle number # 3. When the condition is determined in cycle number # 3 and the condition is satisfied, a skip processing start signal is output from the instruction decoder 21, and the skip state flag is set at the time of cycle number # 4. At the same time, the branch destination address (in this case 105) is set in the branch destination address register 36.

  When the skip state flag is set to the skip state, data writing is prohibited from the instruction currently in the EXE stage, and in FIG. While the skip state flag is in the skip state, the execution unit 32 compares the address of the program counter 35 with the address of the branch destination address register 36. At the time of cycle number # 6 where [the current address of the program counter] ≧ [address of the branch destination address register], the first skip processing end signal I is output. The skip state flag is reset with a delay of the cycle. That is, the skip state flag is set up to cycle number # 7 and reset at cycle number # 8.

  The instruction E, which is an instruction at the address 105, becomes valid again from the data write after the instruction E because the EXE stage is at the time of cycle number # 8 when the skip state flag is reset to the normal state.

  As described above, the CPU 10 according to the present embodiment can realize the skip process control without using the instruction dedicated to the skip process, and can have a small configuration by effectively using the hardware resources. .

  In addition, according to the program control method of the present embodiment, an instruction for ending the skip process is unnecessary, and the number of instructions can be reduced. Furthermore, by setting a flag when a condition is satisfied in a conditional branch instruction and performing a flag reset by comparing addresses, an instruction for starting and ending skip processing becomes unnecessary, and the number of instructions can be reduced. In addition, there is an effect that the programmer is aware of the position where the instruction is validated / invalidated in the program, and the work of embedding a necessary instruction is unnecessary.

As described above, the present embodiment has the following features.
As a program control method, it is triggered that the branch destination address specified by the conditional branch instruction permitted to branch only forward and the current address of the program counter 35 satisfy the branch destination address ≦ the current address. The skip process ends. Thus, the number of instructions can be reduced as compared with the case where a skip invalid instruction is inserted. Also, skip processing instructions need not be implemented.

  As the CPU (processor), the skip processing control means (22, 30) includes a branch destination address register 36 for holding a branch destination address specified by the conditional branch instruction, and a branch destination address of the branch destination address register 36. An ALU (arithmetic logic unit) 50 as an address comparison means for comparing the current address of the program counter 35. The ALU 50 is triggered by the fact that the branch destination address ≦ the current address. To the normal state. This is preferable as another configuration for resetting the skip state flag.

The address comparison means is composed of an ALU (arithmetic logic unit) 50 in the data path 30. When the skip state flag is set to the skip state, the branch of the branch destination address register 36 is sent to the ALU 50. The destination address and the current address of the program counter 35 are input, and the condition code output from the ALU 50, that is, the comparison result is a branch destination address ≦ the current address as a trigger, and the skip state flag is set to the normal state. Reset to. Thus, as a configuration for comparing the branch destination address of the branch destination address register 36 and the current address of the program counter 35, the ALU 50 inside the data path 30 can be used to reduce the size by utilizing resources. .
(Second Embodiment)
Next, a second embodiment of the present invention will be described focusing on differences from the first embodiment.

  FIG. 8 is a block diagram showing the internal configuration of the CPU in this embodiment instead of FIG. FIG. 9 shows a pipeline operation diagram during skip processing in this embodiment instead of FIG.

  In FIG. 8, the skip processing control unit 22 includes a skip state flag 22a, a flag data saving unit 22b as flag data saving means, and a flag data restoration unit 22c as flag data restoration means. The data path 30 is provided with a flag data saving register 38.

In the present embodiment, the skip processing control unit 22 and the data path 30 constitute a skip processing control means.
This configuration operates as follows.

  In FIG. 9, after a conditional branch instruction with an address of 100 (a conditional branch instruction permitted to branch only forward), an instruction A with an address of 101, an instruction B with an address of 102, and an address of Instruction C at address 103 and instruction D at address 104 are skipped, and an interrupt occurs between instruction B at address 102 and instruction C at address 103, so that the instruction at address 106 is addressed. Return after executing F.

  A skip state flag is set by a conditional branch instruction (a conditional branch instruction that is permitted to branch forward only), and the flag data saving unit 22b of the skip processing control unit 22 is set to the skip state flag when a subsequent interrupt occurs. The flag data is saved in the register 38. When an interrupt occurs, “106” is set as an address (address) in the program counter 35 and the branch destination address register 36 (branch destination address ≦ current address is established).

  The flag data return unit 22c of the skip processing control unit 22 returns the flag data saved in the register 38 to the skip state flag in response to the return instruction after the interruption is completed. Further, “103” is set as the address (address) in the program counter 35 to process the instruction C at the address 103 at the end of the interruption (“105” as the address (address) in the branch destination address register 36. Is set, branch destination address> current address).

  Thereafter, in cycle number # 7, the branch destination address of the branch destination address register 36 is “105” and the current address of the program counter 35 is “105”, that is, the branch destination address = current address. Reset the skip state flag (end the skip process).

In this manner, the skip process can be continued even when the process is switched by the interrupt and the process is resumed by the return instruction.
Here, a case where the flag data saving unit 22b, the flag data restoration unit 22c, and the flag data saving register 38 are not provided will be described with reference to FIG.

  In FIG. 10, the flag data of the skip state flag is saved even if an interrupt occurs after the skip state flag is set by a conditional branch instruction (conditional branch instruction permitted to branch only forward). I will not. In this state, when an interrupt occurs, “106” is set as the address (address) in the program counter 35 and the branch destination address register 36, and the skip state flag is set when this branch destination address ≦ current address is established. It will be reset.

  When the interruption ends, “103” is set as the address (address) in the program counter 35 and “105” is set as the address (address) in the branch destination address register 36 to process the instruction C at the address 103. Although the branch destination address> the current address, the skip processing cannot be continued because the operation for resetting the skip state flag has already been completed.

On the other hand, in this embodiment, the skip state flag is not reset by an interrupt.
The skip state flag data may be saved in a memory outside the CPU instead of the flag data saving register 38 of FIG.

  In FIG. 8, the skip state flag 22 a may be provided as the skip state flag 34 a in the program state register 34 of the data path 30 instead of the control unit 20.

As described above, the present embodiment has the following features.
As the CPU (processor), the skip processing control means (22, 30) saves the flag data save unit 22b that saves the flag data of the skip state flag (22a) to the register 38 or the memory when an interrupt occurs, and the return instruction. A flag data return unit 22c for returning the flag data to the skip state flag (22a). Thereby, the skip process at the time of interruption can be continued. This configuration may be applied to the first comparative example .

  In the above description, the case where the sequence control is performed in the pipeline has been described. However, the present invention is not limited to this.

The block diagram which shows the internal structure of CPU of a 1st comparative example . The pipeline operation | movement figure of CPU. Pipeline operation diagram during skip processing in the general example. The pipeline operation | movement figure in the skip process of the 1st comparative example . The pipeline operation | movement figure in the skip process of the 2nd comparative example . The block diagram which shows the internal structure of CPU of 1st Embodiment. The pipeline operation | movement figure in the skip process of 1st Embodiment. The block diagram which shows the internal structure of CPU of 2nd Embodiment. The pipeline operation | movement figure in the skip process of 2nd Embodiment. The pipeline operation | movement figure in the skip process for a comparison.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... CPU, 20 ... Control unit, 21 ... Instruction decoder, 22 ... Skip processing control unit, 22a ... Skip state flag, 22b ... Flag data saving unit, 22c ... Flag data restoration unit, 30 ... Data path, 31 ... Load store Unit: 32 ... Execution unit, 33 ... Register file, 34 ... Program status register, 35 ... Program counter, 36 ... Branch destination address register, 37 ... Pipeline register, 38 ... Flag data saving register, 50 ... ALU.

Claims (2)

When a condition is established based on the condition determination, the processor includes skip processing control means for invalidating one or more consecutive instructions thereafter,
The skip processing control means includes:
A flag that controls writing of data by the instruction, a skip state flag that is set to a skip state when a condition in a conditional branch instruction that is permitted to branch forward only is satisfied, and
A branch destination address register for holding a branch destination address specified by the conditional branch instruction;
Compare the branch destination address of the branch destination address register with the current address of the program counter,
Address comparison means for resetting the skip state flag to a normal state, triggered by the fact that the branch destination address ≦ the current address ,
Flag data saving means for saving flag data of the skip state flag to a register or a memory when an interrupt occurs;
Flag data return means for returning the saved flag data to a skip state flag by a return instruction ;
A processor characterized in that during the skip process in which the skip state flag is set to the skip state, data writing to the inside and outside of the processor is continuously prohibited regardless of the type of instruction. The address comparing means is composed of an arithmetic logic unit inside the data path, and when the skip state flag is set to the skip state, the arithmetic logic unit has the branch destination address of the branch destination address register. The current address of the program counter is input, and when the condition code output from the arithmetic logic unit becomes a branch destination address ≦ the current address, the skip state flag is reset to the normal state. The processor according to claim 1.

Images