CA2064819C - Cpu pipeline having register file bypass on update/access address compare - Google Patents
Cpu pipeline having register file bypass on update/access address compare Download PDFInfo
- Publication number
- CA2064819C CA2064819C CA002064819A CA2064819A CA2064819C CA 2064819 C CA2064819 C CA 2064819C CA 002064819 A CA002064819 A CA 002064819A CA 2064819 A CA2064819 A CA 2064819A CA 2064819 C CA2064819 C CA 2064819C
- Authority
- CA
- Canada
- Prior art keywords
- alu
- operand
- input
- output
- result
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Lifetime
Links
- 238000000034 method Methods 0.000 claims description 8
- 230000008878 coupling Effects 0.000 claims description 6
- 238000010168 coupling process Methods 0.000 claims description 6
- 238000005859 coupling reaction Methods 0.000 claims description 6
- 230000003213 activating effect Effects 0.000 claims 1
- 238000010586 diagram Methods 0.000 description 5
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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, 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; CALCULATING OR 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, look ahead
- G06F9/3824—Operand accessing
Abstract
An A output and a B output of a register file (16) are each provided to an associated multiplexer (18, 20). Each multiplexer has as a further input a bus (CB00:31) that conveys a result from an ALU (22) via an ALU shifter (28). Outputs of the multiplexers are provided to corresponding A or B inputs of the ALU. Each multiplexer is controlled by an associated register file address comparator (24, 26). The address comparators each have as an input corresponding register file A and B update and access addresses. The address comparators compare their associated register file update and access addresses to determine if the register file register selected for access is equal to the register file register selected for update. If these two addresses are found to be equal it is indicated that the result of an ALU operation during an instruction cycle N is to be used as an operand for an ALU operation during a cycle N + 1. When this condition is detected the output of the associated address comparator enables the corresponding multiplexer select input to gate the ALU result directly to the corresponding input of the ALU, thereby effectively bypassing the register file.
Description
VfO 91 /03784 2 Q ~ ~ ~ ~ PCT/US90/01643 IMPROVED CPU PIPELINE HAVING REGISTER FILE BYPASS ON
UPDATE/ACCESS ADDRESS COMPARE
FIELD OF THEINVENTION:
This invention relates generally to data processing apparatus and method and, in particular, to a Central Processor Unit (CPU) arithmetic/logic pipeline of improved speed and efficiency that employs a register file bypass when a register file update address equals a register file access address.
BACKGROUND OF THE INVENTION:
A register file is found in many data processing units and comprises a plurality of registers coupled to an arithmetic/logic unit (ALU), the registers being employed for storing operands and results of arithmetic or logical.
operations such as floating point operations, various control 'operations, etc. The register file may be considered as a local store or cache of high speed, high performance random access memory (RAM). The greater the number of registers within the register file the greater is the amount of data that may be stored within the CPU itself.
Thus, as the register file is made larger fewer accesses are required to be made to typically slower system memory to retrieve operands and to store the results of ALU
operations. As such, it can be appreciated that any WO 91/03784 PCf/US90/01643 improvements in speed and efficiency that are achieved in regard to the operation of the register file and ALU have a direct impact upon the overall speed and processing efficiency of the CPU.
In- some types of systems the register file is accessed (read) and updated (written) during a single CPU instruction cycle. However, as the CPU cycle time is reduced a problem is created in that there is not sufficient time for the register file to be accessed for an ALU operation and the result of the ALU operation written back to the register file during the same cycle. One solution to this problem is to access the register file during a first CPU cycle (cycle N) and to update the register file during a next consecutive CPU cycle (cycle N+1).
However, this solution creates a problem for those types of CPU instructions wherein a result generated during cycle N
is required to written back 'to the register file during cycle N+1 and is also required to be used as an operand during cycle N+1. Such a condition occurs in a pipelined CPU
wherein the execution of instructions are overlapped such that a second instruction is begun before the,execution of a preceding first instruction is completed.
Fig. 1a illustrates in block diagram form a portion of a conventional CPU 1 pipeline having a multiplexer (MUX) 2.
MUX 2 receives a first input from a CPU databus and a second WO 91/03784 PCf/US90/01643 input from a result (R) output of an ALU 4. Interposed between the MUX 2 and the ALU 4 is 'the register file 3. The register file 3 is comprised of a plurality of registers such as 16, 64, 128 or 256 registers. The number of bits (m) of the various data paths and the width of the individual registers varies between implementations and is usually within the range of eight to 128 bits. The register file 3 receives an update register address during a cycle N
and an access register address during a cycle N+1. The update address is an address that selects a register wherein the ALU 4 result is written. The access address is an address that selects a register that is read out to either the A port or the B port and subsequently to the corresponding input of the ALU 4. The result (R) output of the ALU is directed back to the input of the multiplexes for updating a register within the register file. Of course, the R output of the ALU is typically also directed to a number of other circuits that are not shown in the simplified block diagram of figure 1.
In accordance with this conventional system and referring to Fig. lb there is shown a first instruction that has the form A+B=B. That is, the operand stored within a register file location A is added to the operand stored within a register file location B and the result is written back to (updated in) register file location B. A next instruction is of the form C+B=D wherein one of the operands (B) is contained within the updated register from the previous 4 PC?/US90/01643 206~8.~.J, .
.. 4 instruction. In this case, in that the operations are pipelined within the CPU and execute in an overlapping manner with one another, the result of the first operation may not yet be updated within the register file when the second operation is begun. By example, an instruction of the form A+C=C followed by C+D=D presents the same problem in that the register file location to be updated (C) is also required as an operand for the second pipelined instruction.
It is therefore an object of the invention to provide an improved CPU arithmetic/logical pipeline wherein an ALU
result is directly provided as an operand during an immediately subsequent pipelined operation without first being updated within a register file.
It is another object of the invention to provide an improved CPU arithmetic/logical pipeline that includes circuitry for bypassing a local operand store when an ALU result is ' required as an input to the ALU during a cycle wherein the ALU result is also required for updating a location within the local operand store.
SUMMARY OF THE INVENTION
The foregoing problems are overcome and the objects of the invention are realized by apparatus and method that provides a significant speed improvement for those types of operations wherein a storage element such as a register file location or a working register is required to be updated with a result of a previous operation and simultaneously WO 91/03784 PCf/US90/01643 2~~~.~~.~ , accessed to provide an operand for a subsequent operation.
The invention serves to detect such a condition and to provide the output of the previous operation directly to an ALU without incurring an additional delay required to first update and then subsequently access the storage element.
In accordance with a specific embodiment an A output and a B output of a register file are each provided to an associated multiplexes. Each multiplexes has as a further input a bus that conveys a result from an ALU. Outputs of the multiplexers are provided to corresponding A or B inputs of the ALU. Each multiplexes is controlled by an associated register file address comparator. The address comparators each have as an input corresponding register file A and B
access addresses and a register update address. The address comparators compare their associated register file update ' and access addresses to determine if the register file register selected for access is equal to the register file register selected for update. If these two addresses are found to be equal it is indicated that the result of an ALU
operation during an instruction cycle N is to be used as an operand for an ALU operation during a cycle N+1. When this condition is detected the output of the associated address comparator enables the corresponding multiplexes select input to gate the ALU result directly to the corresponding input of the ALU, thereby effectively bypassing the register file.
WO 9i/03784 ~ ' PCT/L'S90/01643 BRIEF DESCRIPTION OF THE DRAWING
The above set forth and other features of the invention are made more apparent in the ensuing Detailed Description_of the Invention when read in conjunction with the attached Drawing, wherein:
Fig. la is block diagram illustrating a conventional register file/ALU interconnection;
Fig. 1b is a graphical depiction of two pipelined instructions having a form wherein during a cycle N+1 a previous ALU result is required for updating a register file location and is also required as an operand for a current ALU operation; and Fig. 2 is detailed block diagram illustrating a register file having bypass circuitry constructed and operated in accordance with the invention.
DETAILED DESCRIPTION OF THE INVENTION
The ensuing description is made the context of a register file having 128 registers each 32 bits wide. It should be realized however that the invention can be practiced with register files of any practical depth and width and is not restricted to operation with_the specific example given below.
Figure 2 is a block diagram that illustrates.a portion of a central processor unit (CPU) 10 constructed and operated in accordance with invention. The CPU 10 may be implemented i7n a number of different types of hardware embodiments including but not limited to microprocessors, minicomputers, and mainframe computers. By, example the CPU 10 may be a VS-type computer, specifically a VS-8000 computer, that is manufactured by Wang Laboratories, Inc. of Lowell, Massachusetts. CPU 10 includes a 32 bit databus DB00:31 coupled through a bidirectional buffer 12 to a first input of a multiplexer (MUX) 14. A second input of the MUX l4 is coupled to a 32 bit C Bus (CB00:31). A choice of DB00:31 or CB00:31 for coupling through the MUX 14 is made by a REG
FILE SELECT input to a MUX 14 select (S) input. The selected 32 bit bus is routed to an output of MUX 14 and is provided to an input of a local operand store implemented as a 128 X 32 register file (REG FILE) 16. REG FILE 16 also has as inputs a seven bit A access address bus and a seven bit B access bus for identifying specific ones of the 128 registers within the REG FILE 16 that are to be accessed (read). The REG FILE 16 also has as an input a seven bit update address kzus for identifying the register to be updated (written). The update address bus is a registered or delayed version of the B access bus. That is, during a cycle N the update address bus updates the register accessed during cycle N-1 by the B access address. REG FILE 16 also includes a 32 bit A output and a 32 bit B output. The data appearing on the A output is addressed and accessed by the a .~
8 .
A access bus and the data appearing on the B output is addressed and accessed by the B access bus.
In a conventional system, such as the system shown in Figure la, the 32 bit A and B outputs would be coupled directly to an ALU. However, and in accordance with the invention, the 32 bit A output and the 32 bit B output of REG FILE 16 are instead each provided to an associated MUX 18 and MUX 20, respectively. MUX 18 and MUX 20 each have as a further , input CB00:31. An output of MUX 18 is provided to an A
input of an ALU 22 and an output of MUX 20 is provided to a B input of ALU 22. MUX 18 and MUX 20 are each controlled by an associated address comparator (ADDR COMP) 24 and 26, respectively. Address comparators 24 and 26 each have as an input corresponding REG FILE 16 access addresses and the update address. The address comparators 24 and 26 compare these REG FILE 16 addresses to determine if the REG FILE 16 register selected for access during a current cycle is equal to the REG FILE 16 register selected for update during the cycle. Tf these two addresses are found to be equal it is indicated that the result of an ALU operation during an instruction cycle N is to be used as an operand for an ALU
operation during the subsequent cycle N+1. When this condition is detected the output of the associated address comparator 24 and 26 enables the corresponding multiplexes MUX 18 or MUX 20 select (S) input to gate the ALU 22 result on CB00:31 directly to the corresponding input of the ALU
22, thereby effectively bypassing the REG FILE 16.
WO 91/03784 PCT/L'S90/01643 If the update address does not equal the corresponding A or B access address the S input to MUX 18 and MUX 20 is not asserted and these multiplexers select instead the corresponding output of REG FILE 16 as the source of data for the ALU 22. For example, the operation A+B=C followed by the operation A+D=E would not result in the REG FILE 16 being bypassed in that neither operand (A or D) of the second instruction is updated as a result of the first instruction.
Continuing with the description of the portion of CPU 10 shown in Fig. 2 the output of the ALU 22 is provided to a latch 28a and from the latch 28a to an ALU shifter 28.
Shifter 28. is under the control of a shift control logic block 30. ALU shifter 28 operates in a conventional manner to position if required the latched ALU result before the ALU result is provided to the CB00:31 bus.
Further in accordance with the invention there is also provided a bypass for an A working register (WR-A) 32 arid for a B working register (WR-B) 34. In this case the working register is a temporary register used for certain arithmetic/logic operations. In accordance with the invention the output of the working register is provided to an input of an associated multiplexes AMUX 36 or BMUX 38.
A second input to both the AMUX~36 and the BMUX 38 is the CB00:31 bus that conveys the ALU 22 result from shifter 28.
Each of the multiplexers 36 and 38 has a select input coupled to an output of an associated comparator 40 and 42, respectively. Comparator 40 generates an output to cause , the AMUX 36 to select the CB00:31 input instead of the WR-A
32 output when the state of logical signals indicate that the source of an operation is WR-A 32 (ACCESS WRA) and also ;
that the WRA 32 is being updated from the ALU 22 during the same cycle. The comparator 42 operates in an identical fashion for providing data to the B port of the ALU 22. The operation of AMUX 36 and comparator 40 thus serves to provide a fast bypass around the WR-A 32 when it is determined that the WR-A 32 is required to be updated and accessed in the same cycle, in a manner similar to that previously described for REG FILE 16. Similarly the operation of BMUX 38 and comparator 42 serves to provide a fast bypass around the WR-B 34 when it is determined that the WR-B 34 is required to be updated and accessed in the same cycle. That is, this aspect of the invention detects if an update signal for one of the working registers is asserted in conjunction with an access signal for the working register and, if so, the working register is bypassed and the ALU 22 output is coupled back to the ALU 22 input.
For the cases described above wherein the REG FILE 16 or one of the working registers is bypassed the update of the register with the ALU 22 result preferably still occurs during the cycle (N+1). However, the ALU 22 is 11 ~064v~.9 simultaneously provided with the update data and is thus not required to wait for the REG FILE 16 or the working register to, be updated.
The CPU 10 further includes a control logic block 44 operable for decoding microinstructions and for generating, in conjunction with a CPU instruction cycle clock (CLK), various control and address signals required for the operation of the CPU 10.
In accordance with the foregoing description it can be appreciated that the invention provides a significant speed improvement for those types of operations wherein a storage element such as a register file location or a working register is required to be updated with a result of a previous operation and simultaneously accessed to provide an operand for a subsequent operation. The invention serves to provide the output of the subsequent operation directly to the ALU 22 without incurring the additional delay required to update and then subsequently access the storage element.
While the invention has been particularly shown and described with respect to a preferred embodiment thereof, it will be understood by those skilled in the art that changes in form and details may be made therein without departing from the scope and spirit of the invention.
UPDATE/ACCESS ADDRESS COMPARE
FIELD OF THEINVENTION:
This invention relates generally to data processing apparatus and method and, in particular, to a Central Processor Unit (CPU) arithmetic/logic pipeline of improved speed and efficiency that employs a register file bypass when a register file update address equals a register file access address.
BACKGROUND OF THE INVENTION:
A register file is found in many data processing units and comprises a plurality of registers coupled to an arithmetic/logic unit (ALU), the registers being employed for storing operands and results of arithmetic or logical.
operations such as floating point operations, various control 'operations, etc. The register file may be considered as a local store or cache of high speed, high performance random access memory (RAM). The greater the number of registers within the register file the greater is the amount of data that may be stored within the CPU itself.
Thus, as the register file is made larger fewer accesses are required to be made to typically slower system memory to retrieve operands and to store the results of ALU
operations. As such, it can be appreciated that any WO 91/03784 PCf/US90/01643 improvements in speed and efficiency that are achieved in regard to the operation of the register file and ALU have a direct impact upon the overall speed and processing efficiency of the CPU.
In- some types of systems the register file is accessed (read) and updated (written) during a single CPU instruction cycle. However, as the CPU cycle time is reduced a problem is created in that there is not sufficient time for the register file to be accessed for an ALU operation and the result of the ALU operation written back to the register file during the same cycle. One solution to this problem is to access the register file during a first CPU cycle (cycle N) and to update the register file during a next consecutive CPU cycle (cycle N+1).
However, this solution creates a problem for those types of CPU instructions wherein a result generated during cycle N
is required to written back 'to the register file during cycle N+1 and is also required to be used as an operand during cycle N+1. Such a condition occurs in a pipelined CPU
wherein the execution of instructions are overlapped such that a second instruction is begun before the,execution of a preceding first instruction is completed.
Fig. 1a illustrates in block diagram form a portion of a conventional CPU 1 pipeline having a multiplexer (MUX) 2.
MUX 2 receives a first input from a CPU databus and a second WO 91/03784 PCf/US90/01643 input from a result (R) output of an ALU 4. Interposed between the MUX 2 and the ALU 4 is 'the register file 3. The register file 3 is comprised of a plurality of registers such as 16, 64, 128 or 256 registers. The number of bits (m) of the various data paths and the width of the individual registers varies between implementations and is usually within the range of eight to 128 bits. The register file 3 receives an update register address during a cycle N
and an access register address during a cycle N+1. The update address is an address that selects a register wherein the ALU 4 result is written. The access address is an address that selects a register that is read out to either the A port or the B port and subsequently to the corresponding input of the ALU 4. The result (R) output of the ALU is directed back to the input of the multiplexes for updating a register within the register file. Of course, the R output of the ALU is typically also directed to a number of other circuits that are not shown in the simplified block diagram of figure 1.
In accordance with this conventional system and referring to Fig. lb there is shown a first instruction that has the form A+B=B. That is, the operand stored within a register file location A is added to the operand stored within a register file location B and the result is written back to (updated in) register file location B. A next instruction is of the form C+B=D wherein one of the operands (B) is contained within the updated register from the previous 4 PC?/US90/01643 206~8.~.J, .
.. 4 instruction. In this case, in that the operations are pipelined within the CPU and execute in an overlapping manner with one another, the result of the first operation may not yet be updated within the register file when the second operation is begun. By example, an instruction of the form A+C=C followed by C+D=D presents the same problem in that the register file location to be updated (C) is also required as an operand for the second pipelined instruction.
It is therefore an object of the invention to provide an improved CPU arithmetic/logical pipeline wherein an ALU
result is directly provided as an operand during an immediately subsequent pipelined operation without first being updated within a register file.
It is another object of the invention to provide an improved CPU arithmetic/logical pipeline that includes circuitry for bypassing a local operand store when an ALU result is ' required as an input to the ALU during a cycle wherein the ALU result is also required for updating a location within the local operand store.
SUMMARY OF THE INVENTION
The foregoing problems are overcome and the objects of the invention are realized by apparatus and method that provides a significant speed improvement for those types of operations wherein a storage element such as a register file location or a working register is required to be updated with a result of a previous operation and simultaneously WO 91/03784 PCf/US90/01643 2~~~.~~.~ , accessed to provide an operand for a subsequent operation.
The invention serves to detect such a condition and to provide the output of the previous operation directly to an ALU without incurring an additional delay required to first update and then subsequently access the storage element.
In accordance with a specific embodiment an A output and a B output of a register file are each provided to an associated multiplexes. Each multiplexes has as a further input a bus that conveys a result from an ALU. Outputs of the multiplexers are provided to corresponding A or B inputs of the ALU. Each multiplexes is controlled by an associated register file address comparator. The address comparators each have as an input corresponding register file A and B
access addresses and a register update address. The address comparators compare their associated register file update ' and access addresses to determine if the register file register selected for access is equal to the register file register selected for update. If these two addresses are found to be equal it is indicated that the result of an ALU
operation during an instruction cycle N is to be used as an operand for an ALU operation during a cycle N+1. When this condition is detected the output of the associated address comparator enables the corresponding multiplexes select input to gate the ALU result directly to the corresponding input of the ALU, thereby effectively bypassing the register file.
WO 9i/03784 ~ ' PCT/L'S90/01643 BRIEF DESCRIPTION OF THE DRAWING
The above set forth and other features of the invention are made more apparent in the ensuing Detailed Description_of the Invention when read in conjunction with the attached Drawing, wherein:
Fig. la is block diagram illustrating a conventional register file/ALU interconnection;
Fig. 1b is a graphical depiction of two pipelined instructions having a form wherein during a cycle N+1 a previous ALU result is required for updating a register file location and is also required as an operand for a current ALU operation; and Fig. 2 is detailed block diagram illustrating a register file having bypass circuitry constructed and operated in accordance with the invention.
DETAILED DESCRIPTION OF THE INVENTION
The ensuing description is made the context of a register file having 128 registers each 32 bits wide. It should be realized however that the invention can be practiced with register files of any practical depth and width and is not restricted to operation with_the specific example given below.
Figure 2 is a block diagram that illustrates.a portion of a central processor unit (CPU) 10 constructed and operated in accordance with invention. The CPU 10 may be implemented i7n a number of different types of hardware embodiments including but not limited to microprocessors, minicomputers, and mainframe computers. By, example the CPU 10 may be a VS-type computer, specifically a VS-8000 computer, that is manufactured by Wang Laboratories, Inc. of Lowell, Massachusetts. CPU 10 includes a 32 bit databus DB00:31 coupled through a bidirectional buffer 12 to a first input of a multiplexer (MUX) 14. A second input of the MUX l4 is coupled to a 32 bit C Bus (CB00:31). A choice of DB00:31 or CB00:31 for coupling through the MUX 14 is made by a REG
FILE SELECT input to a MUX 14 select (S) input. The selected 32 bit bus is routed to an output of MUX 14 and is provided to an input of a local operand store implemented as a 128 X 32 register file (REG FILE) 16. REG FILE 16 also has as inputs a seven bit A access address bus and a seven bit B access bus for identifying specific ones of the 128 registers within the REG FILE 16 that are to be accessed (read). The REG FILE 16 also has as an input a seven bit update address kzus for identifying the register to be updated (written). The update address bus is a registered or delayed version of the B access bus. That is, during a cycle N the update address bus updates the register accessed during cycle N-1 by the B access address. REG FILE 16 also includes a 32 bit A output and a 32 bit B output. The data appearing on the A output is addressed and accessed by the a .~
8 .
A access bus and the data appearing on the B output is addressed and accessed by the B access bus.
In a conventional system, such as the system shown in Figure la, the 32 bit A and B outputs would be coupled directly to an ALU. However, and in accordance with the invention, the 32 bit A output and the 32 bit B output of REG FILE 16 are instead each provided to an associated MUX 18 and MUX 20, respectively. MUX 18 and MUX 20 each have as a further , input CB00:31. An output of MUX 18 is provided to an A
input of an ALU 22 and an output of MUX 20 is provided to a B input of ALU 22. MUX 18 and MUX 20 are each controlled by an associated address comparator (ADDR COMP) 24 and 26, respectively. Address comparators 24 and 26 each have as an input corresponding REG FILE 16 access addresses and the update address. The address comparators 24 and 26 compare these REG FILE 16 addresses to determine if the REG FILE 16 register selected for access during a current cycle is equal to the REG FILE 16 register selected for update during the cycle. Tf these two addresses are found to be equal it is indicated that the result of an ALU operation during an instruction cycle N is to be used as an operand for an ALU
operation during the subsequent cycle N+1. When this condition is detected the output of the associated address comparator 24 and 26 enables the corresponding multiplexes MUX 18 or MUX 20 select (S) input to gate the ALU 22 result on CB00:31 directly to the corresponding input of the ALU
22, thereby effectively bypassing the REG FILE 16.
WO 91/03784 PCT/L'S90/01643 If the update address does not equal the corresponding A or B access address the S input to MUX 18 and MUX 20 is not asserted and these multiplexers select instead the corresponding output of REG FILE 16 as the source of data for the ALU 22. For example, the operation A+B=C followed by the operation A+D=E would not result in the REG FILE 16 being bypassed in that neither operand (A or D) of the second instruction is updated as a result of the first instruction.
Continuing with the description of the portion of CPU 10 shown in Fig. 2 the output of the ALU 22 is provided to a latch 28a and from the latch 28a to an ALU shifter 28.
Shifter 28. is under the control of a shift control logic block 30. ALU shifter 28 operates in a conventional manner to position if required the latched ALU result before the ALU result is provided to the CB00:31 bus.
Further in accordance with the invention there is also provided a bypass for an A working register (WR-A) 32 arid for a B working register (WR-B) 34. In this case the working register is a temporary register used for certain arithmetic/logic operations. In accordance with the invention the output of the working register is provided to an input of an associated multiplexes AMUX 36 or BMUX 38.
A second input to both the AMUX~36 and the BMUX 38 is the CB00:31 bus that conveys the ALU 22 result from shifter 28.
Each of the multiplexers 36 and 38 has a select input coupled to an output of an associated comparator 40 and 42, respectively. Comparator 40 generates an output to cause , the AMUX 36 to select the CB00:31 input instead of the WR-A
32 output when the state of logical signals indicate that the source of an operation is WR-A 32 (ACCESS WRA) and also ;
that the WRA 32 is being updated from the ALU 22 during the same cycle. The comparator 42 operates in an identical fashion for providing data to the B port of the ALU 22. The operation of AMUX 36 and comparator 40 thus serves to provide a fast bypass around the WR-A 32 when it is determined that the WR-A 32 is required to be updated and accessed in the same cycle, in a manner similar to that previously described for REG FILE 16. Similarly the operation of BMUX 38 and comparator 42 serves to provide a fast bypass around the WR-B 34 when it is determined that the WR-B 34 is required to be updated and accessed in the same cycle. That is, this aspect of the invention detects if an update signal for one of the working registers is asserted in conjunction with an access signal for the working register and, if so, the working register is bypassed and the ALU 22 output is coupled back to the ALU 22 input.
For the cases described above wherein the REG FILE 16 or one of the working registers is bypassed the update of the register with the ALU 22 result preferably still occurs during the cycle (N+1). However, the ALU 22 is 11 ~064v~.9 simultaneously provided with the update data and is thus not required to wait for the REG FILE 16 or the working register to, be updated.
The CPU 10 further includes a control logic block 44 operable for decoding microinstructions and for generating, in conjunction with a CPU instruction cycle clock (CLK), various control and address signals required for the operation of the CPU 10.
In accordance with the foregoing description it can be appreciated that the invention provides a significant speed improvement for those types of operations wherein a storage element such as a register file location or a working register is required to be updated with a result of a previous operation and simultaneously accessed to provide an operand for a subsequent operation. The invention serves to provide the output of the subsequent operation directly to the ALU 22 without incurring the additional delay required to update and then subsequently access the storage element.
While the invention has been particularly shown and described with respect to a preferred embodiment thereof, it will be understood by those skilled in the art that changes in form and details may be made therein without departing from the scope and spirit of the invention.
Claims (10)
1. A method for operating an arithmetic pipeline of the type that includes a local addressable store means having a plurality of locations for storing operands, the store means including working register means accessed separately from the store means for storing temporary results of operations on the operands, the pipeline further comprising ALU means having inputs for receiving a first operand and a second operand from the store means, the ALU means having an output for expressing the result of an operation performed upon the first operand and the second operand, the method comprising the steps of:
during a first time interval addressing the store means to retrieve the first operand and a second operand therefrom;
applying the first operand and the second operand to an A input and to a B input, respectively, of the ALU means to produce at an ALU means output a result or a temporary result;
during a second time interval determining if a next operation requires for either a first operand or a second operand the result or the temporary result produced during the first time interval; and if the result or the temporary result produced during the first time interval is determined to be required for the next operation for either the first operand or the second operand the method includes the steps of uncoupling either the ALU means A input or B input from the store means and coupling the ALU means output to either the A input or the B input of the ALU means.
during a first time interval addressing the store means to retrieve the first operand and a second operand therefrom;
applying the first operand and the second operand to an A input and to a B input, respectively, of the ALU means to produce at an ALU means output a result or a temporary result;
during a second time interval determining if a next operation requires for either a first operand or a second operand the result or the temporary result produced during the first time interval; and if the result or the temporary result produced during the first time interval is determined to be required for the next operation for either the first operand or the second operand the method includes the steps of uncoupling either the ALU means A input or B input from the store means and coupling the ALU means output to either the A input or the B input of the ALU means.
2. A method as set forth in Claim 1 wherein the step of determining during a second time interval includes a step of comparing an update address associated with the store means to an access address associated with the store means to detect if the two addresses are equal.
3. A method as set forth in Claim 1 wherein the steps of coupling and uncoupling include a step of activating a multiplexer means to select an input having the ALU means result coupled thereto for coupling the input to an output of the multiplexer means, the output of the multiplexer means being coupled to an input of the ALU means.
4. A method as set forth in Claim 1 wherein the step of determining during a second time interval includes a step of detecting if an update signal for the working register means is asserted in conjunction with an access signal for the working register means.
5. Data processing apparatus comprising an arithmetic pipeline of the type having a local addressable store means having a plurality of locations for storing operands, the pipeline further comprising ALU means having inputs for receiving a first operand and a second operand from the store means, the ALU means having an output for expressing the result of an operation performed upon the first operand and the second operand, the apparatus further comprising:
working register means accessed separately from the store means for storing temporary results of operations on the operands, the working register means having an input coupled to the output of the ALU means and an output coupled to a first input and to a second input of the ALU means;
means for addressing the store means during a first time interval to retrieve a first operand and a second operand therefrom;
means for applying the first operand and the second operand to the first input and to, the second input, respectively, of the ALU means to produce at an ALU means output a result or a temporary result;
means for determining during a second time interval if the ALU means requires for either the first operand or the second operand the result or the temporary result produced during the first time interval; and means coupled to and responsive to the operation of the determining means for coupling the ALU means output to either the A
input or the B input of the ALU means if a next operation requires for either the first operand or the second operand the result or the temporary result produced during the first time interval.
working register means accessed separately from the store means for storing temporary results of operations on the operands, the working register means having an input coupled to the output of the ALU means and an output coupled to a first input and to a second input of the ALU means;
means for addressing the store means during a first time interval to retrieve a first operand and a second operand therefrom;
means for applying the first operand and the second operand to the first input and to, the second input, respectively, of the ALU means to produce at an ALU means output a result or a temporary result;
means for determining during a second time interval if the ALU means requires for either the first operand or the second operand the result or the temporary result produced during the first time interval; and means coupled to and responsive to the operation of the determining means for coupling the ALU means output to either the A
input or the B input of the ALU means if a next operation requires for either the first operand or the second operand the result or the temporary result produced during the first time interval.
6. Data processing apparatus as set forth in Claim 5 wherein the determining means comprises comparator means for comparing an update address associated with the store means to an access address associated with the store means to detect if the two addresses are equal.
7. Data processing apparatus as set forth in Claim 5 wherein the coupling means comprises multiplexer means having the ALU means output coupled to an input thereof, an output of the multiplexer means being coupled to one of the inputs of the ALU means.
8. Data processing apparatus as set forth in Claim 5 wherein the determining means comprises means for detecting if an update signal for the working register means is asserted in conjunction with an access signal for the working register means.
9. Data processing apparatus as set forth in Claim 5 wherein the store means is comprised of register file means organized as n locations each comprised of m bits, a particular one of the n locations being specified by a register file update address and/or by a register file access address.
10. Data processing apparatus as set forth in Claim 5 wherein the determining means is responsive to a content of a microinstruction associated with an instruction.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US07/405,794 US5123108A (en) | 1989-09-11 | 1989-09-11 | Improved cpu pipeline having register file bypass and working register bypass on update/access address compare |
| US405,794 | 1989-09-11 | ||
| PCT/US1990/001643 WO1991003784A1 (en) | 1989-09-11 | 1990-03-28 | Improved cpu pipeline having register file bypass on update/access address compare |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CA2064819A1 CA2064819A1 (en) | 1991-03-12 |
| CA2064819C true CA2064819C (en) | 2000-02-08 |
Family
ID=23605267
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002064819A Expired - Lifetime CA2064819C (en) | 1989-09-11 | 1990-03-28 | Cpu pipeline having register file bypass on update/access address compare |
Country Status (7)
| Country | Link |
|---|---|
| US (1) | US5123108A (en) |
| EP (1) | EP0491693B1 (en) |
| JP (1) | JP2840444B2 (en) |
| AU (1) | AU643432B2 (en) |
| CA (1) | CA2064819C (en) |
| DE (1) | DE69027932T2 (en) |
| WO (1) | WO1991003784A1 (en) |
Families Citing this family (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0594546A (en) * | 1991-02-05 | 1993-04-16 | American Teleph & Telegr Co <Att> | Digital processor |
| JPH04275628A (en) * | 1991-03-01 | 1992-10-01 | Mitsubishi Electric Corp | Arithmetic processor |
| JP2693651B2 (en) * | 1991-04-30 | 1997-12-24 | 株式会社東芝 | Parallel processor |
| US5488729A (en) * | 1991-05-15 | 1996-01-30 | Ross Technology, Inc. | Central processing unit architecture with symmetric instruction scheduling to achieve multiple instruction launch and execution |
| JPH0520066A (en) * | 1991-07-16 | 1993-01-29 | Mitsubishi Electric Corp | Parallel computer |
| JP2539974B2 (en) * | 1991-11-20 | 1996-10-02 | 富士通株式会社 | Register read control method in information processing apparatus |
| DE69330875T2 (en) * | 1992-02-28 | 2002-04-11 | Oki Electric Ind Co Ltd | CACHE DEVICE. |
| JPH0612107A (en) * | 1992-06-02 | 1994-01-21 | Mitsubishi Electric Corp | Sequence arithmetic processor and sequence arithmetic processing unit |
| AU4219693A (en) * | 1992-09-30 | 1994-04-14 | Apple Computer, Inc. | Inter-task buffer and connections |
| EP0650116B1 (en) * | 1993-10-21 | 1998-12-09 | Sun Microsystems, Inc. | Counterflow pipeline processor |
| US5704052A (en) * | 1994-11-06 | 1997-12-30 | Unisys Corporation | Bit processing unit for performing complex logical operations within a single clock cycle |
| US5870581A (en) * | 1996-12-20 | 1999-02-09 | Oak Technology, Inc. | Method and apparatus for performing concurrent write operations to a single-write-input register file and an accumulator register |
| US5799163A (en) * | 1997-03-04 | 1998-08-25 | Samsung Electronics Co., Ltd. | Opportunistic operand forwarding to minimize register file read ports |
| US5996065A (en) * | 1997-03-31 | 1999-11-30 | Intel Corporation | Apparatus for bypassing intermediate results from a pipelined floating point unit to multiple successive instructions |
| US5872986A (en) * | 1997-09-30 | 1999-02-16 | Intel Corporation | Pre-arbitrated bypassing in a speculative execution microprocessor |
| US6131108A (en) * | 1998-03-31 | 2000-10-10 | Lsi Logic Corporation | Apparatus, and associated method, for generating multi-bit length sequences |
| US6088784A (en) * | 1999-03-30 | 2000-07-11 | Sandcraft, Inc. | Processor with multiple execution units and local and global register bypasses |
| EP1124181B8 (en) * | 2000-02-09 | 2012-03-21 | Texas Instruments Incorporated | Data processing apparatus |
| US7206927B2 (en) * | 2002-11-19 | 2007-04-17 | Analog Devices, Inc. | Pipelined processor method and circuit with interleaving of iterative operations |
| US20040249782A1 (en) * | 2003-06-04 | 2004-12-09 | International Business Machines Corporation | Method and system for highly efficient database bitmap index processing |
| US20190377580A1 (en) * | 2008-10-15 | 2019-12-12 | Hyperion Core Inc. | Execution of instructions based on processor and data availability |
| US9152427B2 (en) | 2008-10-15 | 2015-10-06 | Hyperion Core, Inc. | Instruction issue to array of arithmetic cells coupled to load/store cells with associated registers as extended register file |
| TWI607375B (en) * | 2012-11-05 | 2017-12-01 | 義隆電子股份有限公司 | Numerical comparing method of a processor and the processor applied to an electronic device |
| US9569214B2 (en) * | 2012-12-27 | 2017-02-14 | Nvidia Corporation | Execution pipeline data forwarding |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4594682A (en) * | 1982-12-22 | 1986-06-10 | Ibm Corporation | Vector processing |
| GB8401807D0 (en) * | 1984-01-24 | 1984-02-29 | Int Computers Ltd | Pipelined data processing apparatus |
| AU553416B2 (en) * | 1984-02-24 | 1986-07-17 | Fujitsu Limited | Pipeline processing |
| JPS6386033A (en) * | 1986-09-30 | 1988-04-16 | Fujitsu Ltd | Pipeline processing system |
| JPH0810430B2 (en) * | 1986-11-28 | 1996-01-31 | 株式会社日立製作所 | Information processing device |
| US4901267A (en) * | 1988-03-14 | 1990-02-13 | Weitek Corporation | Floating point circuit with configurable number of multiplier cycles and variable divide cycle ratio |
-
1989
- 1989-09-11 US US07/405,794 patent/US5123108A/en not_active Expired - Lifetime
-
1990
- 1990-03-28 WO PCT/US1990/001643 patent/WO1991003784A1/en active IP Right Grant
- 1990-03-28 DE DE69027932T patent/DE69027932T2/en not_active Expired - Fee Related
- 1990-03-28 AU AU53539/90A patent/AU643432B2/en not_active Ceased
- 1990-03-28 EP EP90905845A patent/EP0491693B1/en not_active Expired - Lifetime
- 1990-03-28 CA CA002064819A patent/CA2064819C/en not_active Expired - Lifetime
- 1990-03-28 JP JP2505555A patent/JP2840444B2/en not_active Expired - Lifetime
Also Published As
| Publication number | Publication date |
|---|---|
| AU5353990A (en) | 1991-04-08 |
| JPH05503381A (en) | 1993-06-03 |
| DE69027932D1 (en) | 1996-08-29 |
| CA2064819A1 (en) | 1991-03-12 |
| EP0491693B1 (en) | 1996-07-24 |
| WO1991003784A1 (en) | 1991-03-21 |
| EP0491693A1 (en) | 1992-07-01 |
| JP2840444B2 (en) | 1998-12-24 |
| DE69027932T2 (en) | 1997-02-27 |
| US5123108A (en) | 1992-06-16 |
| AU643432B2 (en) | 1993-11-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2064819C (en) | Cpu pipeline having register file bypass on update/access address compare | |
| US4734852A (en) | Mechanism for performing data references to storage in parallel with instruction execution on a reduced instruction-set processor | |
| US6233670B1 (en) | Superscalar processor with direct result bypass between execution units having comparators in execution units for comparing operand and result addresses and activating result bypassing | |
| US5408626A (en) | One clock address pipelining in segmentation unit | |
| US4896259A (en) | Apparatus for storing modifying data prior to selectively storing data to be modified into a register | |
| US5961637A (en) | Split branch system utilizing separate set branch, condition and branch instructions and including dual instruction fetchers | |
| US4187539A (en) | Pipelined data processing system with centralized microprogram control | |
| US6112297A (en) | Apparatus and method for processing misaligned load instructions in a processor supporting out of order execution | |
| US4679140A (en) | Data processor with control of the significant bit lengths of general purpose registers | |
| EP0199946A2 (en) | Instruction prefetch buffer control and method of controlling an instruction prefetch buffer | |
| US6343337B1 (en) | Wide shifting in the vector permute unit | |
| US4251860A (en) | Virtual addressing apparatus employing separate data paths for segment and offset portions of a virtual address and utilizing only the offset portion to calculate virtual address | |
| CA1181865A (en) | Microprogrammed control of extended integer instructions through use of a data type field in a central processor unit | |
| CA2002895C (en) | Arithmetic unit | |
| US4639866A (en) | Pipelined data processing apparatus | |
| US6178500B1 (en) | Vector packing and saturation detection in the vector permute unit | |
| US5420997A (en) | Memory having concurrent read and writing from different addresses | |
| US5678016A (en) | Processor and method for managing execution of an instruction which determine subsequent to dispatch if an instruction is subject to serialization | |
| US5526500A (en) | System for operand bypassing to allow a one and one-half cycle cache memory access time for sequential load and branch instructions | |
| US5276822A (en) | System with enhanced execution of address-conflicting instructions using immediate data latch for holding immediate data of a preceding instruction | |
| US6725355B1 (en) | Arithmetic processing architecture having a portion of general-purpose registers directly coupled to a plurality of memory banks | |
| US5504870A (en) | Branch prediction device enabling simultaneous access to a content-addressed memory for retrieval and registration | |
| US7134000B2 (en) | Methods and apparatus for instruction alignment including current instruction pointer logic responsive to instruction length information | |
| US5732005A (en) | Single-precision, floating-point register array for floating-point units performing double-precision operations by emulation | |
| EP0061586A2 (en) | Data processing apparatus with extended general purpose registers |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| MKLA | Lapsed | ||
| MKEC | Expiry (correction) |
Effective date: 20121202 |