Instruction Look Ahead Having Prefetch Concurrency and Pipeline Features

The following patent applications filed in the U.S. Patent Office on an even date with the instant application and assigned to the same assignee as the instant application are hereby incorporated by reference to this application:

A. Processor for Input-Output Processing System, invented by Marion G. Porter, et al., having U.S. Ser. No. 562,317, 3/26/75 and which issued as U.S. Pat. No. 3,976,977 on Aug. 24, 1976.

B. Priority Interrupt Mechanism, invented by Michael Monohan, G. Wesley Patterson and Jamie Calle, having U.S. Ser. No. 562,315, 3/26/75 and which issued as U.S. Pat. No. 4,011,783 on Jan. 4, 1977.

C. Dispatcher Mechanism, invented by Michael Monahan and G. Wesley Patterson, having U.S. Ser. No. 562,314 and filed in the U.S. Patent Office on 3/26/75.

1. Field of the Invention

This invention relates to data processing systems and more particularly to a general purpose computer system having an instruction fetch and execute mechanism having a look ahead feature employing prefetch, concurrency and pipe line computer concepts of instruction fetch and execution.

2. Description of the Prior Art

There are many computer systems employing look ahead features in instruction fetch and execution. They have variously been described in the prior art as instruction prefetch, instruction overlap, instruction concurrency, assembly-line processing and pipe line processing. Some typical examples are listed below.

In U.S. Pat. No. 3,254,329, issued May 31, 1966, an arrangement is disclosed for permitting a plurality of different steps associated with a plurality of different instructions to be carried out simultaneously. In this particular system, there is provided at least two advanced instruction registers, a plurality of addressable accumulator registers and an arithmetic unit. The instruction format contains a signal address of an addressable accumulator register which contains an operand to be manipulated by the arithmetic unit and also the address of the accumulator register which is to store the result of the computation. Additional bits of the instruction indicate the address in main memory of an operand to be processed or alternatively indicates an address to which control is to be transferred. Further additional bits represent the address of another addressable register which stores a value used to modify the main memory address of the operand to be processed by the instruction. In that arrangement, during the extraction of the operand associated with the instruction to be executed, the contents of the program counter is transferred to the arithmetic unit to be incremented by one. This address output is decoded, and in the next instruction is abstracted and deposited in one of the two stage registers. In the same minor cycle but during different time periods, the adder adds the operand address of the instruction to be executed to the contents of the specified addressable accumulator register to obtain the operand from memory to be manipulated in accordance with the operation specified by the instruction to be executed and also the modification of the program counter. When the above operations have been completed, the modified instruction to be executed is stored in the second instruction register and the succeeding instruction is extracted and stored in the first storage register. Subsequently, in the same manner, a new instruction is extracted after the incrementing of the program counter and stored in the first instruction register when the previous instruction has been transferred to the second instruction storage register and the instruction to be executed is almost completely executed. Thus, the extraction of other instructions and the processing thereof may occur during the execution of a previous instruction. This patent also discloses means for comparing the address of the result stored against the addresses of the operands specified in the two instructions stored in the storage registers for determining whether or not the above overlap operation should continue without interruption. If the comparison shows equality between the addresses of the result and either of the operand addresses in the next two instructions, then the overlap operation is modified. If the result is the same as the address of the addressable accumulator register to supply an operand for the instruction to be next executed, the overlap operation is not modified and the results being computed are transferred directly back to the arithmetic unit for further processing.

This reference shows an overlap operation wherein an instruction fetch of a succeeding instruction precedes the completed execution of a previously extracted instruction. The address of the instruction which designates the contents of the addressable accumulator register storing an operand for processing by the arithmetic unit is stored in the register until such time as the arithmetic unit has computed a result based on this instruction for comparison with either of the operand addresses and the next two instructions. Although instructions are immediately available to the processor after completion of the currently executing instruction, there may still be loss of time in instruction execution due to the fact that an arithmetic unit may be performing an arithmetic operation in accordance with an instruction taken from main memory while the next instruction although taken out of main memory and available, remains idle even though it does not involve an arithmetic operation being performed by the currently executing instruction. U.S. Pat. No. 3,202,969 takes note of the above problem in loss of time and provides for the extraction of an instruction from memory and its execution initiated while the arithmetic operation of a currently executing instruction is taking place.

A further prior art reference along these lines is an article entitled, "System Design of a Small Fast Digital Computer", IEEE, December, 1963 which discloses an advanced control or look ahead computer arrangement for processing more than one instruction at a time. The arrangement includes a control unit which comprises an instruction control for fetching instructions from a memory well in advance of their execution, operand control which modifies the addresses of operands and calls for the operands and an arithmetic control which executes the function specified by the instructions on the operands.

Still another prior art reference along these lines is U.S. Pat. No. 3,162,841, issued Dec. 22, 1964 which provides storage registers for storing instruction addresses and operand levels prior to execution and an advanced adder operative to compute the address of the next instruction such that when the instruction contained in one of the look ahead registers is transferred to the arithmetic unit for execution, the instruction address in an associated look-ahead instruction counter, which represents the address of the next instruction, is transferred to an instruction counter buffer wherein the buffer contains the address of the next instruction to be executed during the execution of any given instruction.

Perhaps the best example of a computer system utilizing the pipe line computer principle is to be found in an article entitled, "The IBM System/360 Model 91: Machine Philosophy and Instruction Handling" IMB Journal, January, 1967, pages 8-24. The subject article mentions that the primary organizational objective for high performance is concurrency and to overlay the separate instruction functions to the greatest degree possible. A diagram on page 9 of the reference illustrates the concurrency among successive instructions. It can be seen that a second instruction may be accessed prior to the decoding and generating of the operand address and fetching of the operand associated with a first instruction. This pipe line processing technique sometimes referred to as assembly line processing can increase the internal computational performance of a computer machine by one or two orders of magnitude over conventional processing. However, the techniques suffer mainly in such processing applications as list processing, branching and interrupts and is due primarily to the fact that simple communication between adjacent assembly line stations is inadequate. (See second paragraph of second column of page 12, of the January, 1967, IBM Journal).

Since it has been demonstrated that approximately 33 percent of all executed instructions are branch instructions, improvement in look ahead features which provides better communication from the last step of one instruction to the first step of another instruction is desirable.

It is an object of the invention to provide an improved look ahead feature during instruction fetching and execution.

It is another object of the invention to provide better communications between the last step of a currently executing instruction and the first step of the next instruction to be executed.

It is still another object of the invention to concurrently execute the last step of the currently executing instruction and the first step of the next instruction whereby the last step of the currently executing instruction provides processing for the next instruction to be executed and the first step of the next instruction to be executed provides processing to terminate the currently executing instruction.

In accordance with the above and other objects of the invention, an instruction look ahead mechanism is provided whereby in the last step of the currently executing software instruction the processing of the next software instruction to be executed is commenced. Simultaneously with this last step the first step of the next software instruction to be executed is terminating the currently executing instruction. Hence the currently executing software instruction does not require an additional cycle to terminate itself since that cycle is utilized to initiate processing of the next software instruction. Since the processing of the next software instruction to be executed commences one cycle before the end of the currently executing software instruction, and since during this same cycle the next instruction to be executed is taking the steps to terminate the currently executing software instruction, two clock cycles have been saved. This is particularly advantageous during branching operations where the next instruction to be executed may not already be in the instruction look ahead register.

FIG. 1 is a schematic block diagram of a typical computer system utilizing the invention.

FIGS. 2A-2M are data structure diagrams of various registers utilized by the invention.

FIGS. 2N and 2O show the format of a typical instruction utilized by the invention.

FIG. 3 is a block diagram of a processor in accordance with the invention.

FIG. 4 is a timing diagram of the conventional method of instruction fetch and execution.

FIG. 5 is a timing diagram of instruction fetch and execution of the invention.

Referring now to FIG. 1 there is shown a typical computer system utilizing the invention. The System Interface Unit (SIU) 100, containing the priority interrupt and dispatcher unit mechanisms 101 and 102 respectively, provides connections between the elements of the computer processing system. In addition to providing for access of local or remote memory 121, 122 and 122a, 123 respectively by the active module processors 103-106, configuration panel 124 etc., the SIU 100 provides for direct addressing of the high and low speed multiplexors 107 and 112, 113 respectively and control adapters 108, 109, 134, and 135 by the processors 103-106.

The interrupt unit 101 and the dispatcher unit 102, are more fully described in references B and C above; they control the interrupt discipline and perform processor dispatching as required.

Each processor designated IOP/P 103-106 typically contains 16 registers 130-133 for each level of priority assigned to a process or a total of 128 registers per processor capable of storing 128 words. These registers 130-133 are sometimes referred to in this specification as the scratchpad memory and are more fully described infra. Each processor IOP/P is a general purpose computer processor with a conventional complement of register-register, register-memory, immediate, branch, bit field and shift instructions. These processors may be used as input/output processors (IOP) whose functions include initiation and termination of I/O command sequences, fetching, checking and translating channel programs, and direct control of low speed peripheral devices such as unit record and data communications equipment. Processors 103-106 are attached to and communicate with the SIU 100 via ports H, G, F, and E respectively.

The processor local memory 121, 122 interfaces with the SIU and the processors 103-106 via ports L.sub.M0 and L.sub.M1 and H-E respectively. Each of the local memories 121, 122 is organized as a read-write store 136, 137 respectively, plus read only memory ROM 138, 139 with an optional cache memory 140, 141. Width of the local memory interface LM.sub.0, LM.sub.1 is 36 bits plus 4 bits for parity. Local memory size is typically 256K or 512K bytes wherein each K is equal to 1024 bytes.

The remote memory adapters 122a, 123 are utilized when the SIU with its processors and peripherals is part of a larger general purpose computer system. The Remote Memory Adapters (REMA) 122a, 123 then provide a means of exchanging control and data transfers between the processors 103-106 and remote memories 122b and 123b, of another general purpose computer system designated Med 6 (not shown herein). Up to 2 ports .alpha., .beta. and .gamma., .delta. for each of the REMA units 122a, 123 respectively are provided, each port having a 40-bit data path for input and for output between the computer system shown and described herein and a large computer system not shown herein. The REMA units 122a, 123 are connected and communicate with the SIU 100 via ports RM.sub.0 and RM.sub.1 . Note that primary and secondary control blocks (to be more fully described infra) are stored in the remote memories 122b, 123b, and local memories 121, 122 respectively.

A high speed multiplexor 107 provides direct control of data transfers between high speed peripheral devices (disk drive 110, tape drive 111) and central or local memory 121, 122. The high speed multiplexor 107 interfaces between the SIU and the peripherals via port A.

Each low speed multiplexor 112, 113 permits direct control by a processor 103-106 of a low-speed device such as card punch 118, printer 119 and console 120, via ports J and K and device adapter interface unit DAI 115, 116 and 117 respectively.

Referring now to FIGS. 2A-2L the general visible registers 2A are those processor registers which can be accessed with processor software instructions. The following constitute the visible registers of each processor IOP/P:

The PSR, IC, PTBR and GR's are held in scratchpad 326 of sixteen 36-bit registers and are assigned as shown in FIG. 2A.

Process State Register (PSR) FIG. 2B -- The Process State Register holds information essential to the control of the current process. It has the format of FIG. 2B:

Due to the frequency of access to the PSR, either for modification or reference, the actual value for the current process is held in a special register outside the general register scratchpad (not shown). For performance reasons, changes in the register are not reflected in general register 0 GR.sub.0. This scratchpad location assigned to the PSR is used only to safestore the current PSR value in the event of an interrupt.

All instructions which specify GR.sub.0 as a destination operand will cause the result to be stored in the special register used to hold the PSR, and GR.sub.0 will not be changed. References to GR.sub.0, and not the current PSR. A special instruction copies the PSR to a GR so that it may be used as an operand.

Instruction Counter (IC) FIG. 2C -- The Instruction Counter holds the address of the current instruction. Since instructions must be half-word aligned, the least significant bit is always zero. The IC is held in general register 1 GR.sub.1 , and it has the format of FIG. 2C:

Page Table Base Register (PTBR) FIG. 2D -- The Page Table Base Register points to the page table used to provide paged address relocation for the current process. It may be loaded only in master mode. The PTBR is held in GR15, and it has the format of FIG. 2D:

General Registers (GR) -- The remaining 13 registers GR.sub.2 -GR.sub.14 of visible registers FIG. 2A are general registers. These may be used as source or destination operands, or as first or second-level address modifiers.

Control Block Base Register (CBBR) FIG. 2E -- The Control Block Base (CBB) FIGS. 2F-2G is an absolute address which points to the base address in memory of the Exception Control Block (ECB) FIG. 2J and Interrupt Control Block (ICB) FIG. 2J tables.

The Control Block Base Register is actually held in the scratchpad location assigned to GR.sub.0 for the highest priority process level. Two CBB values, a primary and a secondary are held in the register, which has the format of FIG. 2E.

The Primary CBB FIG. 2F is used for all exceptions, and for all interrupts except those associated with local memory errors. When used, the primary CBB is aligned as shown on FIG. 2F.

This alignment permits the location of bases of the ECB and ICB tables on any 512-word boundary in any memory.

Referring to FIG. 2F:

The secondary CBB FIG. 2G is invoked for interrupts due to local memory errors. When used, the secondary CBB is aligned as shown on FIG. 2G.

This alignment permits the location of the bases of the alternate ECB and ICB tables on a 512-word boundary within the first 64K of any memory.

Referring to FIG. 2G:

The Exception Control Blocks and Interrup-Control Blocks are stored as shown on FIG. 2H with respect to the CBB. Note for the address of Interrupt Control Blocks ICB's relative to the CBB, the relative address of the ICB's is added to the CBB; whereas for the address of exception control blocks ECB's, their relative address is substituted from the CBB.

Process Control Register (PCR) -- There is one Process Control Register (PCR) common to all levels. It has the format of FIG. 21.

Exceptions

Exceptions are processor-detected conditions which cause automatic entry to an exception processing routine. Exception conditions may be created deliberately, or they may be the result of a programming error or a hardware error outside the processor. Exception conditions are defined as shown below. For non-MME exceptions, correspondence is shown between type and bit positions of the PCR FIG. 21.

Exception conditions are identified by a four-bit exception number. For master mode entry exceptions, this number is taken from bits 10-14 of an instruction (not shown). In all other cases, the exception number is zero. The exception number is used as an Exception Control Block Number (ECB #) to identify a four-word Exception Control Block (ECB) which points to the exception processing routine. The byte address of an ECB is given by

Ecg address = Control Block Base - 16 (ECB # + 1).

the format of the ECB is shown on FIG. 2J. Referring to FIG. 2J, a PSR word FIG. 2B is held in the first word; an IC word FIG. 2C is held in the second word; a Saving Area Pointer (SAP) for processor pair 0 is held in the third word of ECB 0, and a SAP for processor pair 1 is held in the third word of ECB 1 FIG. 2H.

Before an exception processing routine can be entered, essential information about the current process must be safe-stored. This is performed as a part of the processor response to an exception. Since occurrences of exceptions may be nested (i.e., a second exception may occur before completion of processing for the first), a stack is used to provide space for process safestore. The stack pointer is called the Saving Area Pointer (SAP), and it is held in the third word of ECB 0. Multiprocessor systems require a second stack, and the SAP for the second processor is held in the third word of ECB 1.

When an exception is detected, the appropriate Saving Area Pointer is retrieved, and information about the current process is safestored in the stack in the order shown on FIG. 2K where ADDRESS, IC, GR2-GR4, PCR, PTBR, PSR are defined supra. The Saving Area Pointer is updated accordingly.

The IC stored in the stack FIG. 2K points to the instruction following the one in process at the time the exception was detected. The address stored in the last stack location is the last address of interest generated before the exception was detected. It is primarily for exceptions involving addresses, including operation not complete, bounds, access and missing page exceptions.

After this information about the current process has been safestored in the stack 2K, the PSR, IC, and PTBR are loaded from the appropriate Exception Control Block, and the address of the Saving Area Pointer used by this processor is loaded into GR.sub.2. This completes the entry to the exception processing routine.

Upon completion, the exception processing routine issues a special instruction (RMM) (not shown) to return to the process in which the exception was encountered. This instruction loads the PSR, IC, GR2, GR3, GR4, and PCR and the PTBR from the stack, and decrements the Saving Area Pointer. If exceptions and RMM instructions do not occur in pairs, the exception processing software must ensure that the stack is properly maintained. There are no checks for errors in software manipulation of the stack pointer, or for stack overflow or underflow.

Interrupts are events detected outside the processor which require a processor response. Interrupts in the IOP/P may be assigned to one of eight priority levels. Level 0 is the highest priority level, and level 7, the lowest. In order to minimize the time required to answer an interrupt request, the IOP/P provides a complete set of registers for each of the eight levels. When an interrupt causes the initiation of a new process, the current process is left intact in the registers assigned to the current level. Control may be returned to the interrupted process simply by reactivating that process level. The need to safestore and restore interrupted processes is eliminated, along with the accompanying overhead.

The 16 registers for each level are held in successive 16-register block in the 128-word IOP/P scratchpad 328. Registers for level 0 are held in scratchpad locations 0-15. Since the PSR for level 0 is never transferred to the scratchpad (level 0 cannot be interrupted), scratchpad location 0 is used to hold the Control Block Base although other registers at level 0 would be used. Communication between registers at different levels is possible only via the master mode copy instructions (not shown) which address the scratchpad.

The IOP System Interface Unit (SIU) constantly monitors both the current process level of the processor and requests for interrupts from I/O system modules. Each interrupt request specifies the number of the processor to be interrupted, the priority (level number) of the request, and steering to identify the interrupt requestor (see Interrupt word FIG. 2). This information is held in each module which may request interrupts, and for most modules it is set using programmable interface commands.

Whenever an interrupt request is present at a level higher than the current processor level, the SIU raises a higher level interrupt present line to the processor by providing an interrupt signal (see reference B supra). If several interrupt requests are present at the same level, the SIU determines which request is passed on to the processor on the basis of priorities established by port number.

If the current process is not interrupt inhibited, an interrupt request causes the IOP/P to suspend the current process and to accept an interrupt word from the SIU. The interrupt word has the format of FIG. 2L. Referring now to FIG. 21:

To initiate the interrupt processing routine, four registers are loaded from the interrupt control block FIG. 2J. When the PSR is loaded, the steering field from the interrupt word is inserted into the steering field of the PSR. The other registers, the IC, GR14, and PTBR, are loaded directly from successive words in the ICB FIG. 2J.

A release instruction (not shown) (REL) is used to exit processes entered as the result of an interrupt. After a REL the SIU selects for execution the highest priority process waiting for the processor.

This process may be one that was previously interrupted, or a new process to be initiated as the result of an interrupt request. At the same priority level, previously interrupted processes have priority over new interrupt requests. Through hardware (see reference B supra) and software loading of the PCR, a processor may present to the SIU an interrupt at any level, 0-7. However, in order to provide a well-defined response to a REL executed at any level, the PCR bit requesting a level-seven interrupt is always set.

If a new process is to be entered as a result of a REL, the processor response is similar to that triggered by a normal interrupt, including acceptance of an interrupt word from the SIU and access to an ICB. If a previously interrupted process is to be re-entered, the SIU supplies only the level number and the fact that an old process is to be re-entered. Since the process state at the time of its interruption is intact in the register scratchpad, this is the only information required to restart the process.

Referring now to FIG. 2N there is shown a typical scratchpad register instruction. This instruction causes the transfer of information between the IOPP scratchpad memories 328, 334, 326 and the general registers. The first 8 bits of the instruction comprise the field for the operation code. GR is a 4 bit field at bit positions 9-12 which specifies one of the general registers in the scratchpad memory 328. AS is an address syllable occupying bits 4, 14-35 of the instruction used to calculate the effective address Y. This address is interpreted as the address of a register outside the IOPP/P.

Referring now to FIG. 20 there are shown two formats that the address syllable AS of GRX instruction shown on FIG. 2N can take. The top format shows an address syllable having a long displacement D whereas the bottom format of FIG. 20 shows an address syllable having a short displcement D. Referring now to the long displacement format, XR1 at bit positions 14-17 is the address of an index register XR.sub.1. Bit position 18 specifies whether a long or short displacement is to be utilized. When this bit is set to zero, it indicates a long displacement D whereas when it is set to one, it indicates a short displacement D. At bit positions 19 there is shown a sign bit S whereas bit positions 20-35 are reserved for the long displacement address B. Referring to the bottom format of address syllable AS in FIG. 20 XR.sub.1 at bit positions 14-17 again is the address of an index register 1. As previously noted, bit 18 indicates whether a long or short displacement D is utilized. XR2 occupying bit positions 19-22 is the address of a second index register R2. S occuying bit position 23 is the address syllable AS in a sign bit; whereas D occupying bit positions 24-35 is the short displacement address.

FIG. 3 is a block diagram of a processor in accordance with the present invention. Data and instructions from the system interface unit (SIU) are provided at register 310 and processed data is provided to the SIU at data out register 312. As data and instructions are clocked into register 310 a parity check is made and parity errors are noted.

Instructions are placed in a "look ahead" dual read-out register 314 which provides four words of buffering for instructions. An 8-level control store address switch 316 (which may typically be a multiplexor) provides an address to control store 318. (Data Selectors, typically are multiplexors and are well known in the art and commercially available. See The Integrated Circuit Catalog for Design Engineers published by Texas Instruments, pages 9-339 to 9-364.) One level of the CSA switch 316 is provided by the instruction register 314 via pathfinder unit 317. The control store 318 contains the microinstructions which control data manipulations, and an addressed microinstruction is stored in control store register 319.

Data from input register 310 is loaded into one level of the eight-level "B" switch 320 (also commercially available) which, along with two-level "A" switch 322 (commercially available), provides operands for the adder/shifter network 324. The "B" switch 320 is controlled by a field in the control store register 319. "A" switch 322 provides inputs from the dual read-out working registers 326 or from general registers scratch pad 328 via SPB register 330. The dual read-out register 326 contains 3 working registers and an image of the instruction count-(IC) contained in the general register scratchpad 328. The WRR output from working register 326 is supplied to "A" switch 322, and the WRP output from general register 326 is supplied to the general register scratch pad 328. A seven-bit address for the general register scratchpad is generated in a one of eight switch (SPA) 322 (commercially available).

The adder/shifter network 324 which receives operands from "B" switch 320 and "A" switch 322 performs all of the arithmetic, logical, and shift operations required for address development and instruction execution.

A page table word (PTW) scratchpad 334 provides storage of 16 page table words for each of the 8 levels of "B" switch 320. The four-level address switch 336 (commercially available) concatenates addresses for either programmable interface-commands or read/write memory cycles (either paged or absolute).

Output from the adder/shifter network 329 may be supplied through a four-level output switch 338 (commercially available) to result crossbar (R-XBAR) 340 and to data output register 312. The R-XBAR provides simultaneous transfer of its input to both the data output register 312 and the working registers 326. Switch 338 also receives inputs from a process state register (PSR) 342, from a process control register (PCR) 344 through switch 346, and from the general register scratchpad 328.

The General Register (GR) scratchpad 328 contains 128 forty bit registers. Each register consists of 4 nine bit bytes with a parity bit per byte. Data written into the scratchpad comes from one of the four working registers on the WRP output of the dual readout register bank used to implement the working registers. Registers included in the scratchpad are a process state register (PSR), an instruction counter (IC), a page table base register (PTBR), thirteen general registers (GR), and a process control register (PCR). The seven bit address for the scratchpad is generated in one of eight switch (SPA) 332. The switch control inputs are wired directly to the Control Store Output register (CSR) 319. The most significant three bits of the address define one of eight levels and the least significant four bits define one of sixteen registers within the level. The level is supplied by Active Interrupt Level (AIL) lines (not shown) from the SIU for six of the eight positions. The eight address sources are as follows:

(0) Seven bits of the constant field (2-8) of the CSR which allows addressing any register in any level.

(1) The AIL lines and four bits of the CSR constant field (5-8) which allows addressing any register in the current level.

(2) The WRR output of the dual readout working registers bits 29-35. This allows a working register to provide the address for either initialization or software addressing (CSPG or CGSP).

(3) the AIL lines and bits 19-22 of the current instruction. This provides the XR2 read address for second level indexing.

(4) The AIL lines and bits 14-17 of the current instruction. This provides the XR1 read address for first level indexing or the GR/GR2 read address for operand access.

(5) The AIl lines and bits 9-12 of the current instruction. This provides the GR1/GR read address for operand-access.

(6) the AIl lines and bits 0-2 of the Write Address (WA) register 48 with the least significant bit wired to logical one. This provides the odd address of an even odd pair read or write instruction.

(7) The AIl lines and bits 0-3 of the WA register. This provides the address for all softwre writes into a GR at the current level. This includes GR loads and returning execution results to the destination GR.

the output of the scratchpad goes to the one of two switch 330A into the SPB register 330 which is also an input to the Result Crossbar (R-XBAR) 340 through adder shifter network. The switch into SPB allows operations on a GR and a working register or on two working registers by loading one into SPB. The switch 330A into SPB 330 is controlled by the SP control field in the CSR 319.

The Write Address (WA) register 348 can be loaded from either bits 9-12 or 14-17 of the current instruction (see reference A supr). This provides the address for loading a General Register (GR) or returning a result to a GR. This is necessary since the GR address in the instruction being executed is no longer available out of the dual readout Instruction register once the IC is updated. The GR address is therefore saved in WA and used for the write operation by setting a Write (W) flip/flop (see ref. A supra) associated with WA. W resets on the first clock after it is set unless the WA control field in the CSR once again sets it (two word load of GR). A GR scratchpad write clock pulse is generated on all clocks occurring while W is set unless WA=0 in any mode or WA=15 is slave mode.

SPB register 330 is a 40 bit register (4 bytes with parity per byte). It provides buffering for words read out of the scratchpad and eliminates the scratchpad access time from the time required for an Adder/Shifter operation. Parity is checked on the data in the SPB register 330. The SPB load clock is controlled by the SCR SP control field 319.

The A and B operand switches provide the two operands for the Adder/Shifter network. A switch 322 selects either SPB or the WRR output of the dual readout working registers. The selection is controlled by a bit in CSR 319. However, the control is forced to select WRR if W is set and the address in WA is equal to XR1. This causes the new value of XR1 or GR2 to be used if the previous instruction modified them. The switch output is forced to logical zeros if the DL position is selected in B switch 20 and no indexing is called for (XR 1=0).

B switch 320 selection is controlled by a three bit field in CSR 319. However, the least significant bit is forced to logical one if the DL position is selected and second level indexing is required (bit 18 of the Instruction = 1. See reference A). The eight B switch positions are formatted as follows:

(0) Bits 0-19 are equal to IRSW 319a bit 19. Bits 20-35 are wired to IRSW bits 20-35. This is the displacement field for either first level or no indexing.

(1) Bits 0-23 are equal to IRSW bit 23. Bits 24-35 are wired to IRSW bits 24-35. This is the displacement field for second level indexing.

(2) Bits 0-30 are equal to IRSW bit 8. Bits 31-35 are wired to IRSW bits 9-13. This is the short immediate value.

(3) bits 0-17 are equal to IRSW bit 8. Bits 18-35 are wired to IRSW bits 18-35. This is the long immediate value.

(4) This position selects the WRR output of the dual readout working registers 322.

(5) Bits 0-31 are equal to logical zero. Bit 32 is equal to the most significant bit of the CSR constant field. This provides the number 8 for incrementing the instruction counter IC to point to the next even/odd Instruction pair (8 bytes) in memory. Bits 33 and 34 are equal to the word length in bytes of the current instruction if the two most significant bits of the CSR constant field are zero (10 for word and 01 for half word). Bit 35 is equal to the carry bit in the process state register PSR if the next to most significant bit of the control store register CSR constant field is one.

(6) Bits 0-26 are equal to zero. Bits 27-35 are wired to the CSR constant field.

(7) This position selects the SIU Data In (DI) register.

A detailed block diagram of the Adder/Shifter network is shown in FIG. 4 in the application of reference A supra. The Adder-Logical Unit (ALU) executes 36 bit arithmetic and logical operations. It also provides the transfer path for either the A or B operands to the R-XBAR at 340/ The ALU operations are controlled by the ALU/Shift input bits in the CSR 319. The AlU mode is controlled by the least significant bit of the PSR/PCR control bits in the CSR.

The Page Table Word (PTW) scratchpad 334 provides storage for 16 PTW's for each of the eight levels (128 PTW's). The output of the B operand switch 320 provides the write data into the scratchpad and the write clock is controlled by a bit in the CSR.

The PTW scratchpad address is generated from either the least significant 7 bits of the WRP output of the working register 326, or the level and bits 21-24 of the WRP output. The first position is for initialization and generl register GR to base table and PTW transfer. The second position is for reading/loading PTW's while paging addresses and loading missing PTW's. The address selection is controlled by a bit in the CSR.

Each byte of the PTW output is parity checked. The PTW scratchpad output provides input data to two of the four address switch positions. If the PTW output is selected by the Paged position of the Address switch the following checks are made to determine if the PTW is valid (the priority of the checks for declaring faults is in the order shown):

(1) Bits 30-35 of the PTW are compared to a Key register 372. (The Key register identifies the process with which the PTW is associated and is loaded with bits 30-35 of WRP each time GR 15 is loaded.)

(2) Bits 27-29 of the PTW are compared with bits 18-20 of WRP. This is to verify that the correct PTW is resident in this PTW Scratchpad location. (0, 16, 32, etc. all reside in the same scratchpad location.)

(3) The next check is to see if the page is resident in read/write R/W memory 122b, 123b. A zero in PTW bit 6 indicates that the page is not resident in memory.

(4) If the first three checks pass, bits 4-5 of the PTW are compared with the type of operation being initiated. A data read is always legal. An Instruction fetch requires bit 4 while a write requires bit 5.

If the PTW in the scratchpad fails any of the above checks, it will be accessed from the Page Table in R/W memory and checked again prior to causing an exception.

The address word for either R/W memory or the Programmable Interface is generated in the Address switch 336. The switch is controlled by the Address switch control bits in the CSR. If the paged position is selected and the PSR reflects the absolute address mode, the absolute position of the switch will be forced so that paging is bypassed. The four positions are as follows:

(0) This position generates the paged address to R/W memory. Bit 0 equal to zero defines a R/W address. Bits 1-3 are provided by the ZAC bits in CSR. Bit-4 is equal to zero. Bits 5-8 are the zone bits and are generated as a function of the R/W memory operation. Reads cause zeros, word or double word writes cause ones, and byte writes cause a one in the byte position to be written. Bits 9-24 are equal to PTW scratchpad 9-24 which is the page base address. Bits 25-35 are equal to WRP 25-35 which is the page relative address. When this position is selected, the WRP output of the working registers must refluect the unpaged address.

(1) This position generates the R/W memory address when no paging is required. It can be selected by the CSR or will be forced if position 0 is selected and the PSR reflects the absolute address mode. Bits 0-8 are the same as position 0. Bits 9-35 are equal to WRP 9-35 which must be equal to the absolute memory address when this position is selected.

(2) This position generates a Programmable Interface (PI) command word. Bit 0 equal to one defines a PI command word. Bit 1 is supplied by the CSR ZAC field. Bit- 2 is equal to bit 9 of the PSR and defines whether the current program can alter certain external registers. Bit 3 is equal to the processor number supplied by the SIU. Bit 4 is equal to zero. Bits 5-8 are equal to PSR bits 4-7 and define the port within the multiplex. Bits 9-35 are equal to WRP 9-35 and must be equal to the absolute address generated for either RDEX or WREX.

(3) this position provides the path for reading a PTW from the scratchpad.

Bits 0-2 of the address switch 336 are modified to reflect the R/W memory steering during loading of absolute addresses into GR. This requires bits 0-2 to reflect PTW scratchpad 0-2 if paged and WRP 0-2 if absolute address mode. This would be enabled due to position 0 of the address switch being selected and no R/W memory cycle being initiated by the CSR SIU request type control bits.

The steering switch provides the SIU steering for either a R/W memory cycle or a Programmable Interface command. It is controlled by the Address switch control bits in the CSR. The steering is generated for R/W memory as follows:

Bit 0 -- This bit equals 0 for R/W memory

Bit 1 -- This bit defines local or remote memory. It is equal to PTW bit 0 if paged or WRP bit 0 if absolute.

Bits 2-4 -- These bits are the memory steering bits. The initial value is equal to PTW bits 1-3 if paged or WRP bits 1-3 if absolute. This is also the final value if bit 1 defines remote memory. When bit 1 defines local memory, bits 2 and 3 define the local memory port and steer addresses to the ROM in the local memory controller. Bit 2 is equal to the Exclusive OR of the initial value and the Local Memory Port Specifier (LMPS) line from the SIU. Bit 3 is equal to the Exclusive Or of the PCR ROM bit if the initial value is zero.

Bit 5 -- This bit defines a single or double word memory cycle. It is equal to bit 1 of the CSR ZAC field.

Bit 6 -- This bit defines a read or write cycle. It is equal to bit 0 of the CSR ZAC field.

The steering is generated for a PI command as follows:

Bit 0 -- This bit equals 1 for a PI command

Bits 1-4 -- These bits define the SIU port to which the PI command is directed and equal bits 0-3 of PSR.

Bits 5-6-- These bits are the same as for a R/W memory cycle and are generated in the same way.

The outputs of the steering switch are clocked into the steering register at the SIU interface each time a memory cycle or PI command is initiated.

The Result Crossbar (R-XBAR) 340 provides simultaneous transfer of its inputs to both the Data Out and Working registers 312 and 326 respectively. A third output is wired to a display panel (not shown) and provides the path to display most of the IOPP registers. The output to the working registers is controlled by the WR Write Address bits in the CSR and can select any of the four inputs. The output to the DO register 312 is controlled by the DO Write Address bit in the CSR and can select either the ALU/Shifter Output switch 338 or the Address switch 336. However, this position is forced to select the PSR/PCR input if the DPCR line (not shown) from the SIU is activated.

(0) ALU/Shifter Output switch

(1) Address Switch

(2) PSR/PCR Switch

(3) SPB Input Switch

The four working registers are contained in the dual readout register bank 326. Register 0 contains the current Instruction Counter (IC). (The IC is also maintained in the current level's GR1 of the GR scratchpad). Registers 1, 2 and 3 are working registers for instruction execution. They are labeled R1, R2 and R3.

The two Working register outputs are labeled WRP and WRR. WRP is used to access PTW's from the PTW scratchpad, R/W memory address generation and supplies the Working register input to both the GR scratchpad 328 and the SPB register 330. The register enabled out of WRP is controlled by the WRP bits in the CSR 314. WRR is used to provide operands to the A and B operand switches 322 and 320 respectively and the input to both the PSR and PCR registers. The register enabled out of WRR is controlled by the WRR bits in the CSR.

The Working registers can be loaded from any of the XBAR inputs. The register to be loaded and the write clock is controlled by the WR write address and Write WR bits in the CSR.

There is no restriction on the reqisters selected for the read and write operations. It can be three different registers or they can all be the same one.

The Process State register (PSR) 342 is kept outside the GR scratchpad since it is continuously monitored and updated. It is loaded from the WRR output of the Working regiters 326. A write clock is generated for the PSR each time a master mode program loads GR0 (GR0 written using the WA address) or the PSR/PCR control bits in the CSR define a write PSR operation.

The entire PSR is loaded during a master mode load of GR0, execution of an Exception fro the ECB, or the execution of a DSIP, MME, RMM or REL instruction (see Appliction of Ref. A). When an interrupt is executed, the steering from the Interrupt data word is inserted into the PSR data from the ICB prior to loading.

A condition code (cc), carry (c) and process timer are continuously updated. The cc is loaded each time an instruction is executed requiring a cc update. C is loaded with the carry out of the arithmetic and logic unit ALU each time the cc is loaded and the ALu is in the arithmetic mode. The process timer is decremented each time a Timer Ticker (not shown) rolls over. The Timer Ticker is an eight bit counter which counts on all system clocks (controlled clocks in step mode). The Timer Ticker is also used to detect an operation not complete or lock up exception as described in the section on exceptions.

The Process Control register (PCR) 344 is common to all levels. It is loaded from the WRR output of the working registers (not all bits are loadable). A write clock is generated for the loadable bits when the PSR/PCR control bits in the CSR define a write PCR operation.

Bits 18-19 and 28-34 are loadable. Bits 0-16 set when the defined condition occurs and are reset by the set/reset bit control in the CSR. Bits 23-26 are provided for software to read.

The PSR/PCR switch into the R-XBAR selects the corresponding register to be loaded into one of the working registers. This switch is controlled by the PSR/PCR control bits in the CSR but is forced to select PCR if the DPCR line (not shown) from the SIU is activated.

The dual readout register bank 314 provides four words of buffering for instructions. The current instruction read output (CIR) 380 and next instruction read output (NIR) 380a provide access to the entire instruction independent of the instruction length and address. This is provided through the Instruction register switch (IRSW) 319a. The CIR address is equal to the current Instruction Counter (IC) bits 32 and 33 which points to one of the four words. The NIR address is generated to point to the following word. IRSW 319a is controlled by the current bit 34 of the instruction counter IOC which defines whether the instruction starts on a word or half word address. The two IRSW positions are therefore (0) CIR bits 0-35 and (1) CIR its 18-35, NIR bits 0-17. IRSW bits 0-17 will reflect a half word instruction and IRSW bits 0-35 will reflect a full word instruction. The CIR and NIR addresses are updated each time the working register instruction counter IC is updated. All fields of the Instruction word must therefore be used prior to updating the IC.

The IR 314 is loaded each time a new value is loaded into the IC due to an interrupt, exception, branch, etc. or each time CIR address crosses over a two word boundary when the IC is updated by the current instruction length. The instruction access control is described below for the two conditions 1) enter new procedure and 2) incrementing through current procedure. In both cases the instruction fetches are double precision memory cycles and the addresses are paged unless the PSR defines absolute mode.

(1) The double word instruction fetch is initiated and the IR write address loaded on the clock that pages (if requied) the new value of IC. The IR write address is loaded with 00 if IC bit 32=0 or 10 if IC bit 32=1. (The CIR and NIR addresses are loaded when the new IC value is loaded). When the first word is available from memory, it is written into IR and the least significant bit of the write address is set. This causes the next memory word to be written into the second word of the pair (01 or 11). The IC value plus eight (bytes) is then used to initiate another double precision memory read using the paged (if required) address. The IR write address is updated to the next two words (10 if IC bit 32=0 or 00 if IC bit 32=1 and a test is made to see if instruction execution can begin or if execution must wait for the memory cycle to complete. The test is on bit 33 of the IC. If the test indicates the new procedure is being entered at the last half word of a two word pair (33, 34=1, 1), the instruction execution must be delayed until the data is available from the second double precision cycle to guarantee IR contains a full instruction word.

(2) The execution of each instruction includes an update of the IC by that instruction's length. If this update causes the IC to pass over a two word boundary (old IC 32 = new IC 32), the two word area of the IR that was just finished (old IC 32 value) can be loaded with new instructions. The new IC value plus eight (bytes) is then used to initiate a double presicion memory read using the paged (if required) address. The IR write address is updated to point to the IR area available. When the two words are received, they are written into the two word area as described above.

The Control Store Address is generated in the CSA switch 316. The first four positions of the CSA switch are controlled by the CSA switch control field in the CSR 319. The CSA switch control 316 can select the Next address register (NA) 382, the Return address register (RA) 384, the Execution address register (XA) 386, or the output of the Standard Sequence decode network (SS) 388. The Exception/Interrupt position is forced when either of these two conditions exist. The Exception address is reflected unless an Interrupt is being executed. The two PTW miss is detected. The constant position is selected when the Branch control field in the CSR calls for a branch to the constant address.

NA 382 is loaded on each execution clock by the sum of the CSA switch 316 output pluss one plus a condition skip constant 390. If no skip is called for by the CSR skip control field, NA is loaded with the address of the microinstruction immediately following the one being accessed (i.e., the clock that loads the microinstruction at address M into the CSR loads the address M+1 into NA). If a number of microinstructions are to be conditionally skipped, the CSR skip field can specify that a skip be executed with the CSR constant field defining the condition to be tested and the number (1 through 7) of microinstructions to be skipped. The sequence for a skip is as follows: microinstruction at M calls for a conditional skip, the execution of this microinstruction loads M+1 into the CSR and load the address of M+1+1+SKP into NA.

SKP=0 if the skip is not satisfied and equals the skip count defined in the least significant three bits of the CSR constant field if satisfied. The skip is inhibited if any of the last four positions are selected in the CSA switch.

The conditions that can be tested for skip execution are defined by bits 3-5 of the CSR constant field. WRR 35, WRR 0, WRR 33 and the carry bit in PSR need to be tested for zero or one. The PSR cc field will be tested for zero, one, two or three. Bits 1-2 of the constant field are used to define the test. The conditions to be tested are as follows:

The RA register 384 is loaded from the NA register 382 whenever the Load RA bit is on in the CSR.

The XA register 386 is loaded with the Pathfinder 317 output each time the SS position is selected in the CSA switch. The use of the Pathfinder will be described below. Its output is one control bit and eight adress bits. The address is used to address the upper 256 words of Control Store (address bit zero is forced to 1 in the XA position of the CSA switch).

The execution of a software instruction is in two phases. The first phase is a microinstruction sequence common to a group of instructions. The second phase is a microinstruction sequence (which is only one microinstruction in most cases) unique to the specific software instruction being executed. After completing the second phase, the common phase of the next instruction would be entered (in some cases, the second phase may return to the first phase via RA register 384 for a few additional common steps prior to entering the next instructions common phase).

The operation code of IRSW 219a provides the Pathfinder address and an input to the Standard Sequence Decode network 388. The Standard Sequence Decode network generates the Control Store address of the start of the microinstruction sequence common to the group of instructions containing this one. (This sequence is referred to as a standard sequence). The location in the Pathfinder addressed by the operation code contains the address in Control Store where the unique sequence for this instruction starts. The instruction is then executed by branching to the Standard Sequence address, executing the common steps, branching to the unique sequence address in XA register 386, executing the unique steps, updating the Instruction Count (IC) so that the next instructions operation code is enabled out of IRSW and repeating the above sequence by branching to the new Standard Sequence.

The interrupt answering, exception processing the PTW missing sequences are entered by forcing the corresponding position to be selected in the CSA switch 316. Interrupts are executed at the completion of software instructions. If the Higher Level Interrupt Present (not inhibited) or the Level Zero Present lines from the SIU are active when the SS position of the CSA switch is selected by the SCA switch control bits in CSR, the CSA switch control logic is forced to select the Exception/Interrupt position. This causes the interrupt answering sequence to be entered rather than the next instructions standard sequence. (The Exception/Interrupt position reflects the address of the interrupt answering sequence at this time).

Missing Page Table Words (PTW) cause immediate entry into the PTW missing sequences. Either the operand or instruction missing position is forced by the CSA switch 316 control logic during the clock period immediately following the paging step. The return from either sequence is to the standard sequence decoded from IRSW 319a. This causes the instruction that was being executed to be started over again. Therefore, the microprogram will not do anything prior to the PTW missing detection that can't be done again. A flip/flop (not shown) is set when the PTW miss is detected that stays set until the address is once again paged. A miss the second time through causes an exception as defined below.

The exceptions fall into two categories. The first type causes an immediate entry into the exception processing sequence. The second type does not affect the CSA switch 316 control logic until the next instructions standard sequence is entered. Both types cause the Exception/Interrupt position in the CSA switch 316 to be selected and set the corresponding bit in the PCR register 344.

The first category of exceptions are operations not complete, Page faults, Page not resident, and illegal instructions. They all must be serviced as soon as they are detected since continuation of instruction execution is impossible. The second category is process timer run out, overflow, lockup fault and address misalignment. Divide check is handled by a test and branch if divisor is zero. They all indicate faults but do not need to be immediately serviced and can wait until the start of the next instruction.

The Control Store Output register (CSR) 319 contains the microinstruction being executed. Provision is made for a remote CSR register, as indicated.

There is a one of four position switch 394 supplying the input to CSR. The four positions on the CSR input switch are as follows:

(0) This position is the input from the ROM chips on the Control Store substrates.

(1) This position is not used.

(2) This position is the input from the maintenance panel.

(3) This position reflects the local CSR. It is used to reload the remote CSR bits when the maintenance panel switches are used to display data.

Data is displayed in the I/O System by simulating CSR with maintenance panel switches. When the switches are enabled out of the CSR input switch 394, a signal is generated causing the remote CSR bits to be loaded with the microinstruction simulated by the switches. The old contents must be reloaded when the display of the registers is completed. This is accomplished by selecting position 3 for one clock prior to switching back to position 0 and reloading the remote CSR bits during the one clock period. (A block diagram of the input to the local and remote CSR is shown in Application of Reference A supra in FIG. 3 with the timing diagram for reloading the remote CSR after using the maintenance panel switches for display).

The format of CSR is as follows:

The control store 318 is a typical prior art read only memory ROM well known in the art. The theory, construction and use of ROM's is detailed in a book by Samir S. Husson entitled, "Microprogramming Principles and Practices" published by Prentice Hall in 1970. The microprogram which resides in the control store 318 and controls the operations of the invention is included in the Appendix of this specification. When the control store 318 is microprogrammed as shown in the Appendix, it is termed the firmware of the invention. The firmware differs from a software program in that an actual physical operation is performed on the ROM chip, and removal from the computer or loss of power does not alter the microprogram; this is akin to hardwiring the program in the computer. However, with software programming there is no physical alteration of memory -- merely electrical, hence loss of power or removal of the memory from the computer destroys the program; there is no kinship to hardware logic.

The instant invention utilizes many prior art look-ahead features in prior art computers wherein an instruction fetch of a succeeding instruction precedes the completed execution of a previously extracted instruction such as disclosed in previously referenced U.S. Pat. No. 3,254,329 issued May 31, 1966. It also utilizes features similar to prior mentioned U.S. Pat. No. 3,202,969 issued Aug. 24, 1965 which suggests the fetching of this instruction from memory and the initiation of its execution while the arithmetic operation is going on. Moreover, some of the principles of a pipe-line or assembly-line concept of instruction fetch and execution as previously reference herein in conjunction with the IBM 360 model 91 machine are also utilized. Accordingly the instruction register 314 holds up to 4 instructions to be executed which have been fetched from main memory during the execution of prior instructions. Moreover, execution of the next instruction in instruction register 314 awaiting execution is commenced during the execution of a current instruction by placing bits 0-17 (see FIGS. 2M-2N) into bit 0-17 of the instruction register switch 319a. The index bits 14-17 of FIG. 2N of the next instruction are then used to address a general register in scratchpad memory 328 via path 319a multiplexor switch 332 (LEV, XR1). This operation is performed as part of the DONE statement which is the last microinstruction in the currently executing software instruction. The DONE statement is defined on page 7 of the Appendix which constitutes a part of this specification and is a listing of the microinstructions of the control store. Referring to page 7, line 20, the DONE statement is defined. The contents of the scratchpad SP 328 are read out to the scratchpad buffer SPB 330; the addrress of the index register XR1 is supplied also to the scratchpad 328 via switch 332 and path SPA. Bit 18 of the address syllable of the instruction (FIG. 2N) is set to a one if second level indexing is required. The DONE microinstruction also calls for a skip test to determine whether or not second level indexing is required. Finally, it requires the standard sequence SS to be utilized for the control store address CSA. The standard sequence SS is the common routine for executing the next software instruction. An example of the use of the DONE statement is on page 65 of the Appendix. Referring to page 65 of the Appendix on the second paragraph, there is shown the firmware of the last step of an LDBG (Load Byte Into a General Register) instruction.

READ-BYTE, PAGED-AD=R2 -- Starts a word read memory cycle. The address is enabled out of R2 in the working registers 326; addressing the PTW scratchpad 334 and building the paged (or absolute) address in the address switch 336. The address is transferred through the X-BAR 338 into Data Out 312. CALL, TO-ICMPW1 -- Save NA in RA and branch to 136.

R SHIFT-DI, SHCNT=BL, BYTE-LOC=R2, SHFTR=R1 -- Data In 310 transfers through B switch 320; through shifter 324, shifts right using BL position of shift count switch and R2 as byte address to define shift count, through X-BAR 338 and into working register R2 326.

R1 & 511-OR1, LOGIC.fwdarw.CC COND-I-FETCH -- R1 & 511 (111111111).fwdarw. R1 zeroes all of R1 except bits 27-35 & CC is loaded with logical test (no overflow). COND-I-FETCH -- Start Instruction Fetch if room in IR 314.

R1-R1 -- This loads R1 on last step which causes W F/F to be set and will write R-1 .fwdarw. SP on next clock. DONE -- as explained on -1-.

With the registers and buffers thus conditioned, the conventional method of executing a software instruction is for the new software instruction to read out the index register from a general register XR1 in scratchpad memory 328 and place the contents of the index register in the scratchpad buffer SPB 330. The second step of the software instruction then is to obtain the address of the operand by adding the contents of the index register in SPB 330 to the displacement DL of the instruction (FIGS. 2M--2N) contained in IRSW 319a. Accordingly, the displacement DL or DS is placed in adder shifter 324 via position 0 or position 1 of B switch 320. The contents of SPB register 330 are likewise routed through A switch 322 into adder shifter 324 and added to the displacement DL or DS, depending on whether it was a long displacement or a short displacement. The next step is to place the modified address into register R2 of working register 326 via path 338 and 338a. This address is used to address a R/W memory location via PTW scratchpad 334 (if paged) through address switch 336, through the X-BAR into Data Out register. The memory location is read into Data In 310, through the B switch 320, through the adder shifter 324, through the X-BAR 338, into the working register R1 336. Finally, the contents of the working register R1 326 are written into a general register of scratchpad memory 328 via path WRP.

With the look ahead feature of the current invention, it is not necessary for any step of the new software instruction to cause a reading out of the scratchpad memory 328 into scratchpad buffer 330 since as previously described this is performed as the last microinstruction of the currently executing software instruction via the DONE statement. Also it is not necessary for the new instruction to write the contents of the working register 326 into the scratchpad as it will be executed by the first step of the next instruction. When the new software instruction is accessed through the standard sequence SS used for the control store address CSA (which as previously described is also a part of the DONE statement), the first step of the new software instruction is the second step of the conventional method of fetching and executing a software instruction i.e. the first step of the new software instruction modifies the address by adding the displacement DL of the instruction to the index register contained in the scratchpad buffer SPD 330 as previously described in step 2 of the conventional method of executing a software instruction. Note also that in the conventional method of executing a software instruction, the data loaded into the working register 326 from R/W memory has to be transferred into the scratchpad 328 on the last step. With the look ahead (or overlap) feature, the transfer of the working register 326 into the scratchpad 328 is executed by the first step of the new instruction. Since the old software instruction and the new software instruction perform their execution concurrently, two clock cycles have been saved. Also, since it is not necessary for the old software instruction to complete itself, leaving this operation for the first portion of the new software instruction, the old software instruction can utilize this time not only in accessing the old instruction but accessing information that is required for the new instruction so that the new instruction can immediately begin to concurrently execute as its first step, in addition to completing the old instruction, what would have normally been the second step of the new instruction. Accordingly, there is a greater efficiency in fetching and executing instructions, particulaly in branch and interrupt, since the concepts of prefetch, overlap and pipe line execution are combined herein in this novel way.

A better appreciation may be obtained of the invention by referring to FIGS. 3, 4 and 5 and comparing the conventional method of instruction fetch and execution with that of the invention. FIGS. 4 and 5 show minor cycles during which microoperations of an instruction are executed and a major cycle comprised of the minor cycles for executing macro-operations.

Referring to FIG. 4:

Having shown and described a preferred embodiment of the invention, those skilled in the art will realize that many variations and modifications may be made to affect the described invention and still be within the scope of the claimed invention. Thus, many of the elements indicated above may be altered or replaced by different elements which will provide the same result and fall within the spirit of the claimed invention. It is the intention, therefore, to limit the invention only as indicated by the scope of the claims.