US 4,172,281 AGrant
Microprogrammable Control Processor for a Minicomputer or the Like
Issue Date:1979-10-23
•4 Claims
•64 Drawing Sheets
Abstract
A minicomputer comprises a microprogrammable central processing unit wherein micro-instruction execution speed is optimized through the use of variable micro-instruction timing logic and by grouping micro-instruction according to execution time. Furthermore, data paths are arranged so that micro-routines that implement more complex operations, i.e., memory reference instructions, follow the fastest route possible. When micro-instructions requiring longer data paths are programmed, the computer dynamically varies the length of the microcycle to be a function of both the type of micro-instruction to be executed and the state of the minicomputer when the micro-instruction is to be executed. A microprogrammable processor port is provided to allow coupling of external hardware, e.g., I/O devices, other processors, etc., directly to the microprogrammed control processor. This capability is in addition to the standard input/output system of the minicomputer, thereby providing an alternate interface path for devices requiring very fast transfer rates. These devices interfaced through the microprogrammable processor port are directly coupled to the internal data busses of the minicomputer and addressed under direct microprogram control as if they were internal processor registers. Block transfers of data are provided via this microprogrammable port to allow transfers of large blocks of data without dependence on the I/O system timing of the minicomputer. A remote program load feature is provided whereby an I/O device or data communications interface can initiate a bootstrap operation in a remote computer, i.e., the computer can be halted, a preselected ROM loader program is transfer into memory, all input/output instructions in the loader are automatically configured to the proper select code of the I/O device, and the computer is restarted at the proper loader program starting address.
Metadata
Assignee
- Hewlett-Packard Company
Inventor
- Philip Gordon
Application Information
Application Number:US 8290385
Filing Date:1977-08-30
Priority Date:1977-08-30
Art Unit:237
Classifications
IPC:
G06F 104G06F 900
Field of Search:
364200 MS File;900 MS File
Patent Drawings (64 sheets)
Description
BACKGROUND AND SUMMARY OF THE INVENTION
Some previous minicomputers have utilized a microprocessor in their control section. These machines generally use the microprocessor to execute the assembly language level instructions of the minicomputer by running predetermined series of microprograms. These microprogrammed minicomputers have had the advantages that they are versatile in providing expandable instructions and in providing powerful yet economical minicomputer architectures. They have had the disadvantages that they have generally been slower in execution of programs than computers having sequential gating control sections where the logic had been skillfully minimized. Since they have generally taken longer to execute the same instruction, there have been corresponding decreases in the rate of input/output transfers for throughput of data which could be handled by these machines. Also, since the microprogrammable control section typically operated as a separate computer of its own, many operations required synchronization of the internal timing for their microprocessor and that of other portions of the minicomputer, e.g., the memory and I/O sections.
Similarly, peripheral hardware such as ancillary processors, etc. suffered a great decrease in possible performance because they were interfaced via the standard I/O system with all the interfacing delays and synchronization required between the two units. This has been a critical limitation for many applications where custom arithmetic processing hardware is intended to be added to the computer to augment its performance in a particular area. In contrast, the preferred embodiment provides the ability to directly interface the microprocessor to ancillary peripherals and processors. Extremely fast transfers of large blocks of data as well as communications between the microprocessor and other processors is possible without any additional synchronization or overhead associated with the microcomputer architecture. Since the devices or processors are interfaced directly to the internal data busses of the computer and addressed under direct microprogram control as if they were internal processor registers, the maximum possible throughput is achieved.
Some previous computers had the ability to bootstrap or load predetermined programs into control memory via front panel controls or from an I/O device in response to a bootstrap loader control switch. Other computers also executed some form of bootstrap program when power was first turned on or when power was restored after some non-operating period. However, these previous bootstrap systems had the disadvantage that in the event that power was restored to the computer, but the program stored in memory had been destroyed, there was no way to request a bootstrap operation from a remote site. In contrast to this, the preferred embodiment provides remote program load capability which allows another processor or a data communication device to have control over the run/halt controls of the computer. The remote program load feature allows the user to initiate an automatic system bootstrap operation which may include a complete reloading of system software from a disc, a communication line, or any other user-defined device. The system is then restarted. This technique provides the advantage that the need for experienced operators can be greatly minimized, particularly for those systems with complex start up procedures or for distributive processing systems where automatic cold start may be required. Of course, this feature is essential for computer installations located at unattended, remote or inaccessible sites.
Furthermore, previous computers using a microprocessor have used fixed microcycle times. The time period selected was usually the sum of all the worst case timing delays throughout the processor for the most complex, i.e., lengthy micro-instruction. Therefore, the microprocessor control section was working well below the maximum possible speed, except when executing the worst case micro-instructions. In contrast, the preferred embodiment varies the length of the microcycle as a function of both the type of micro-instruction and the state of the minicomputer at the time that micro-instruction is to be executed. The microcycle timing generator then varies the cycle time in response to these conditions.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 is a block diagram of the preferred embodiment.
FIG. 2 shows the relationship of FIGS. 2A, 2B, 2C and 2D. FIGS. 2A, 2B, 2C and 2D taken together comprise a detailed block diagram of the preferred embodiment.
FIG. 3 is a simplified block diagram of the control processor of the preferred embodiment.
FIG. 4 illustrates the definitions of the basic timing intervals in the preferred embodiment.
FIGS. 5A, 5B, 5C and 5D are timing diagrams illustrating the variable microcycle technique of the preferred embodiment.
FIG. 6 shows the relationship of FIGS. 6A and 6B. FIGS. 6A and 6B taken together are a flow chart for determining the time duration of a microcycle for the preferred embodiment.
FIG. 7 is a flow chart showing the sequence of timing events occuring in any given microcycle.
FIG. 8 is a consolidated microcycle estimating flow chart.
FIG. 9 is a detailed pause time calculation flow chart.
FIG. 10 shows the configuration switch settings for block transfers.
FIG. 11 is a diagram of an interface circuit for use with the remote program load feature of the preferred embodiment.
FIG. 12 is the flow chart of the HALT microprogram feature for use with the remote program load feature of the preferred embodiment.
FIG. 13 shows the timing relationship of the micro-instructions used with the microprogrammable processor port feature of the preferred embodiment.
FIG. 14 illustrates the wave forms used during a synchronous burst data transfer through the microprogrammable processor port of the preferred embodiment.
FIG. 15 shows a block diagram of a first in/first out (FIFO) buffer and control logic circuitry to interface this device to the microprogrammable processor port.
FIG. 16 shows the relationship of FIGS. 16A, 16B, 16C and 16D. FIGS. 16A, 16B, 16C and 16D are a detailed schematic timing diagram of the preferred embodiment.
FIG. 17 shows the relationship of FIGS. 17A, 17B, 17C and 17D. FIGS. 17A, 17B, 17C and 17D taken together comprise a detailed schematic diagram of the ALU circuitry.
FIG. 18 shows the relationship of FIGS. 18A, 18B, 18C and 18D. FIGS. 18A, 18B, 18C and 18D taken together comprise a detailed schematic diagram of the connection register and I/O bus circuitry of the preferred embodiment.
FIG. 19 shows the relationship of FIGS. 19A, 19B, 19C and 19D. FIGS. 19A, 19B, 19C and 19D taken together comprise a detailed schematic diagram of the micro-instruction register and the related decoding circuitry.
FIG. 20 shows the relationship of FIGS. 20A, 20B, 20C and 20D. FIGS. 20A, 20B, 20C and 20D taken together comprise a detailed schematic diagram of the control memory and the control memory address register circuitry.
FIG. 21 shows the relationship of FIGS. 21A, 21B, 21C and 21D. FIGS. 21A, 21B, 21C and 21D taken together comprise a detailed schematic diagram of the macro-instruction register and a portion of the decoding circuitry therefore of the preferred embodiment.
FIG. 22 shows the relationship of FIGS. 22A, 22B, 22C and 22D. FIGS. 22A, 22B, 22C and 22D taken together comprise a detailed schematic diagram of the extend and overflow register circuitry, the A and B addressable flip-flop and circuitry for the generation of various control signals within the preferred embodiment.
FIG. 23 shows the relationship of FIGS. 23A, 23B, 23C, 23D, 23E and 23F. FIGS. 23A, 23B, 23C, 23D, 23E and 23F taken together comprise a detailed schematic diagram of the I/O circuitry of the preferred embodiment.
FIG. 24 shows the relationship of FIGS. 24A, and 24B. FIGS. 24A and 24B taken together comprise a detailed schematic diagram of the I/O circuitry of the preferred embodiment.
FIGS. 25 and 26 illustrate the operation of the Jump Subroutine (JSB) Save Stack of the preferred embodiment.
DESCRIPTION OF THE PREFERRED EMBODIMENT
Referring now to FIG. 1 there is shown a block diagram of the preferred embodiment. Control processor 10 executes microprograms stored in RAMs 12 (random access memory) and 14 as well as ROMs (read only memory) 16 and 18. While these ROMs and RAMs may be implemented in slightly different ways, they all appear as one large microprogram storage facility to the control processor 10.
The circuit details of the control processor are described in more detail with respect to FIGS. 16-24.
An arithmetic and a logic section 20 includes most of the hardware required to carry out the commands of the micro-instructions. It contains working registers and additional logic to perform arithmetical and logical operations on this stored data. The arithmetic/logic section detailed schematic diagrams are shown with greater detail in FIG. 17. Hereinafter, the collective control circuitry of the control processor 10 and the arithmetic and logic section 20 are collectively referred to sometimes as the CPU (central processing unit).
An input/output section 30 serves as an interface between the computer and external devices. Circuitry is provided to respond to both control processor initiated I/O transfers as well as those initiated by a peripheral device, e.g., peripheral device 35. The detailed schematic diagram of the input/output section is described with reference to FIGS. 23 and 24.
An operator panel 50 is provided to allow manual communication to the central processing unit.
A more detailed block diagram is shown in FIGS. 2A, 2B and 2C.
Referring now to FIGS. 2A, 2B, 2C and 2D, there is shown the detailed block diagram of the preferred embodiment. FIG. 2A contains a detailed block diagram of the I/O section, while FIG. 2B contains a detailed block diagram of the control processor. FIG. 2C contains a detailed block diagram of the arithmetic and logic section, while FIG. 2D contains circuit details of the memory section.
Referring now to FIG. 2B, the instruction register 205 is a 16-bit register that contains the assembly, i.e., machine language instructions for execution. The lower eight bits of this register are also used as an up/down counter. The control memory 210 receives a 14-bit address from the control memory address register 208. The corresponding 24-bit micro-instruction is then loaded into the micro-instruction register 211. The control memory address register 208 is a 14-bit register which will progress sequentially through memory by being incremented at the beginning of every microcyle unless an instruction which specifically changes its contents such as a branch repeat or a jump instruction is executed. Field decoders 220 then provide the appropriate control signals on lines 221, 222, 223, 224 and 225 to execute the micro-instruction in micro-instruction register 211.
The actual coding of the micro-instructions is explained in more detail with reference to Appendix A which shows the binary codes and mnemonic symbols for the various micro-instructions and word types, and with reference to the micro-order definitions of Appendix B.
Also included in the control processor is a three-level micro-subroutine save register 214. The 14-bit control memory address register 208 contents is loaded into the stack, "pushed" at the beginning of every micro-subroutine branch, i.e., a jump subroutine(JSB) instruction. It is retrieved or "popped", i.e., the contents are loaded back into the control memory address register 208 when a micro-subroutine return (RTN) instruction is executed.
Referring now to FIG. 2C, there is shown a detailed block diagram of the arithmetic logic unit (ALU) section of the preferred embodiment. The ALU implements all arithmetic and logic operations in the CPU under the direction of the control processor. The S-bus 350 is the main data bus of the CPU. A register 305 and B register 306 are the main 16-bit accumulators of the preferred embodiment. Additional register storage capability is provided by registers 315. Under microprogram control, data from the S-bus, the A and B registers, registers 315, etc. may have various logical or arithmetic functions performed on them by ALU circuit 330 and rotate shifter circuit 311. The various possible operations are described in more detail with respect to the aforementioned Appendix B.
Included within the CPU of the preferred embodiment is the capability for storing predetermined "loader" programs. In the preferred embodiment, blocks 370, 371 may contain up to four optional loaders. Each loader program contains 64-16 bit instructions. Remote program load switches 372 are associated with these loader ROMs, as more fully described below.
Also shown in FIG. 2C is a connection for the microprogrammable processor port of the preferred embodiment. Control lines 380 carry the signals generated from micro-instruction 211 and associated circuitry necessary to execute the microprogrammable processor port operations. Additional control bits are provided via line 327 which carry the four least significant bits of L register 310. Data may be input to or output from devices coupled to the microprogrammable processor port via S-bus 350 through the latch register. An alternative representation of the control processor is shown in FIG. 3.
Before proceeding with a more detailed description of the circuit diagrams of the preferred embodiment, a more detailed explanation of the timing definition in terms of the control processor will be explained. The control processor executes one micro-instruction during each microcycle. In the preferred embodiment, these microcycles, also designated as T-periods, are the time required to completely execute the micro-instruction, i.e., execute up to five micro-orders which make up the micro-instruction. In order to sequentially order the execution of the micro-orders in the various fields of any particular micro-instruction, it can be seen that another timing interval is needed. These subintervals are designated P1 through P5 as shown in FIG. 4. These T-periods or microcycles can either be short microcycles such as microcycle 410, or long microcycles such as microcycle 415. In a short microcycle of preferred embodiment, each T-period is 35 nanoseconds (ns) long, the total microcycle then being 175 nanoseconds long. In the case of a long microcycle, such as microcycle 415, additional time intervals called E-intervals, E1, E2 and E3 are inserted between P-periods P3 and P4. Each of the P-periods and E-periods is 35 nanoseconds long, but since there are now 5 P-periods and 3 E-periods during the execution of a long microcycle, the total microcycle time period is 280 nanoseconds. An exception to the above described timing occurs when a pause condition exists. Another condition which will cause variations in the above timing cycles is that of a freeze condition. The selection of the short or long microcycles and the generation of timing pauses and timing freezes is discussed in more detail below.
As discussed above, a short microcycle consists of 5-35 nanosecond intervals that run in sequence from P1 through P5. A long microcycle consists of 8-35 nanosecond intervals that always run in the sequence P1, P2, P3, E1, E2, E3, P4 and P5. The arithmetic logic section of the computer is designed to operate with a 175 nanosecond microcycle while other operations such as input/output operations require the longer (8-35 nanoseconds subintervals) microcycle. There are three basic reasons in the preferred embodiment for the control processor timing circuits to switch to the longer, i.e., 8-35 nanosecond subinterval microcycles. First, certain I/O interfaces may require the longer T-period to complete the execution of an operation. Therefore, if an I/O operation is indicated, long microcycles are generated during certain predetermined T-periods, e.g., T3 through T5. Also, when an extended memory package is used, such as described in U.S. Pat. No. 3,970,999 issued to John S. Elward, there is often difficulty in getting the data onto the S-bus fast enough to ensure integrity of the data. Also, the micro-instruction register is clocked at the beginning of each microcycle, approximately P1, and the control memory address register is conditionally loaded at P3 of each microcycle. However, if a microbranch instruction such as a jump or jump subroutine is to be executed, only two subintervals, P4 and P5 or 70 nanoseconds, would be left to access control memory and reload the control memory address register with the address of the new micro-instruction for the subsequent microcycle. Since this is not adequate, time is also extended for this condition. However, if a conditional jump or branch instruction is programmed and the branch condition is not met, the control processor will remain in a short microcycle mode.
Most microcycles will be of the shorter type to maximize throughput performance. However, a change to a long microcycle timing can occur based on the monitored conditions described more fully below during peak subinterval P3 of any microcycle. In other words, the conditions that determine a switch to long microcycles are monitored at every P3.
The control processor and main memory operate asynchronously until they must communicate to accomplish read or write operations. This communication is accomplished in a "handshake" manner. A "pause" in microcycle timing is used to interact with the asynchronous memory interface. This feature permits greater performance with existing systems and compatibility with various speed memories.
A pause operates in the following way: a read or write operation can be started with the appropriate micro-order in any microcycle. Memory then independently completes the operation under its own timing, asynchronously to the CPU. If the control processor should through another micro-instruction request another memory operation while memory has not completed the first task or some other operation, a conflict in timing occurs. This conflict is monitored at subinterval P3 of every microcycle before the control processor actually makes the request for the use of memory. If a conflict is detected, i.e., there is an attempt to use memory while it is busy, a pause state, i.e., the suspension of all timing clocks in the control processor will be initiated until memory is available.
In the preferred embodiment, a pause is accomplished by having the timing circuits "latch-back" into P3 over and over again so that P3 is repeated until the pending memory request can be processed. Pause time, therefore, is always an integer multiple of 35 nanoseconds in the preferred embodiment.
At the completion of a pause, the CPU timing will progress either to subinterval P4 or subinterval E1, depending upon whether the conditions for a short or a long microcycle exist.
Referring now to FIGS. 5A, 5B, 5C and 5D, there are shown four examples of microcycles with a pause. FIGS. 5A and 5B are both short microcycles, while FIGS. 5C and 5D are examples of long microcycles with a pause. FIG. 5A illustrates a short microcycle wherein subinterval P3 is repeated over and over again because there is a read operation underway, e.g., a read memory instruction was encoded in the previous micro-instruction with a TAB micro-instruction (read memory data onto the S-bus) coded in the S-bus field in this micro-instruction. FIG. 5B, a write into memory operation is underway, i.e., a WRTE micro-instruction was encoded in the previous micro-instruction and another write WRTE (write memory) instruction was attempted immediately following that micro-instruction.
FIG. 5C shows a long microcycle with a pause due to a read operation underway, similar to that shown in FIG. 5A. However, in this case, there is a TAB micro-order in the S-bus field and an RTN (return from subroutine) in the special field of this micro-instruction. Therefore, there is both a pause as well as the addition of subintervals E1, E2 and E3. FIG. 5D is similar to FIG. 5B except that once again an RTN instruction is encoded in the special field of the micro-instruction.
Freeze Operations
The CPU and I/O section operate asynchronously until an I/O instruction begins execution and communication between the two is needed. Although T-periods run sequentially from T2 through T6, and each T-period is initiated by subinterval P1 of a microcycle, I/O micro-instructions must begin at the appropriate part of the I/O cycle. Therefore, the freeze condition suspends micro-instruction execution, but continues T-period generation, until the appropriate T-period starts, that is the T-period which is required for execution of the I/O micro-instruction currently in micro-instruction register 211. In the preferred embodiment, the freeze synchronizes micro-instruction execution with either time period T2 or time period T6. The timing freeze can begin at the end of any microcycle and when the I/O instructions are to be executed, long microcycles will automatically occur in correspondence with time periods T3, T4 and T5.
In summary, the freeze or pause conditions occur whcn communication is required between the control processor and either the I/O section or the memory section. Generally, the freeze occurs to synchronize the control processor with the I/O section while a pause occurs to suspend control processor operations and wait for the completion of a memory operation, if it is necessary to use the memory before proceeding. Also, long or short microcycles may be selected depending on whether additional time is required to complete a task in a microcycle period.
FIGS. 6A and 6B can be used to calculate the execution time for individual microcycles in the preferred embodiment. The flow chart formed by FIGS. 6A and 6B is to be read from left to right, once per each microcycle. All the conditions that change timing in the preferred embodiment during the microcycle are shown in FIGS. 6A and 6B along with their respective execution times in nanoseconds. These times are summed while proceeding through the microcycle. All instructions referred to are amply described in the coding and definitions present in Appendixes A and B respectively. FIGS. 7 and 8 are additional flow charts showing the microcycle timing.
The following examples indicate how the technique used in the preferred embodiment provides the fastest possible micro-instruction execution time consistent with the input/output and memory systems selected.
Arithmetic/Logic Micro-instructions
The fastest microcycle timing is found when microprogrammed operations deal with the Arithmetic/Logic section registers. For example:
To determine the execution for the micro-instruction in Table I, read FIGS. 6A and 6B from left to right with the first micro-instruction in mind. The total time for the first two intervals (P1+P2) is 70 nanoseconds. The special field in the first micro-instruction does not contain an RJ30 and the OP field does not contain a READ or WRTE. Also, the S-bus field does not contain TAB. Thus, in following the timing line into P3, no pause condition exists.
Continuing in subinterval P3, since an I/O operation is not being performed, no synchronization with the T-periods is required. Also, a microbranch will not occur since there is no MEU in the S-bus field of this micro-instruction and no JSB, JMP or RTN micro-orders coded. Following the timing line in FIGS. 6A and 6B through the end of P3, time in this microcycle thus far is 105 nanoseconds. Intervals P4 through P5 are executed immediately, making the total time for execution of the micro-instruction labeled FIRST=175 nanoseconds. Assuming that no freeze conditions are in effect, the timing line can be followed back to the beginning of P1.
Control Memory Branch Micro-Instructions
The switch to long microcycles is made in P3 when any of the three conditions shown in FIGS. 6A and 6B can be answered affirmatively. For example, consider a control memory branch condition shown in the following portion of a microroutine. In the example of Table II, the microcycle times are included in the right-hand column.
By using FIGS. 6A and 6B and checking the microroutine, it can be seen that the JSB and RTN micro-orders in the micro-instructions labeled ONE, THREE and CLEAR can cause long microcycles when the branch condition is met. By not extending the microcycle when the branch condition is not met, unnecessary time is present in the micro-instruction cycle.
For this example, assume the T-period is T4 and the Control Processor has just placed the first micro-instruction of your microroutine in MIR. Suppose further that part of the microroutine is as shown in 211.
The micro-instruction at label XXX includes micro-orders in the S-bus and store fields as well as the IOG micro-order in the special field. As subintervals P1 and P2 occur, the S-bus and store field micro-orders will be executed but the effect of the IOG in the special field is not felt until the end of the microcycle (in following the timing line in FIGS. 6A and 6B, note that the freeze condition is not in effect until the micro-instruction labeled XXX completes execution). At the end of the microcycle, the IOG micro-order causes all micro-instruction execution to be suspended until T2 completes. The total waiting time in the freeze condition in this case is 525 nanoseconds. Note that with a freeze condition present, T-periods will be short microcycles until synchronization occurs. Time T3 starts the I/O cycle and each micro-instruction is executed in the appropriate long microcycle (T-periods). If T6 is short (as shown in the example), the total time for the I/O cycle will be 1.120 microseconds. If T6 had been long (e.g., a RTN coded), the total time for the I/O cycle would be 1.225 microseconds. This example microroutine is used only to illustrate the freeze until T2 starts.
Read From Memory Example
First consider a read from main memory with a TAB micro-order in the S-bus field two micro-instructions after the micro-instruction containing the READ micro-order. In the example microroutine in Table IV, assume no memory operation is in progress as the microroutine begins at label START.
using FIGS. 6A and 6B, note that START executes in 175 nanoseconds. In FIRST, note that although there is a READ in the OP field in this micro-instruction (which begins a memory operation) there is not a memory operation already in progress; thus, FIRST also executes in 175 nanoseconds. Point A shows where the main memory read cycle timing starts (the request for memory is made at the end of the microcycle). No delays occur for execution of the micro-instructions labeled SECOND and THIRD; they each execute in 175 nanoseconds as shown at point B, while main memory is still busy executing the read request. (Note that these two micro-instructions do not contain micro-orders that would cause a freeze.)
Now the micro-instruction labeled DATA begins to execute. FIGS. 6A and 6B show that if there is a TAB in the S-bus field while memory is busy, there will be a pause time added to the microcycle. FIG. 9 can be used to calculate the time as follows. At the first decision point in the flow chart, no READ, or WRTE, or RJ30 micro-order is encoded in this micro-instruction. Entry is made at Step I (FIG. 9) because there is a TAB micro-order encoded in the S-bus of the micro-instruction under consideration.
In Step I, add the execution times for micro-instructions labeled SECOND and THIRD which =350 nanoseconds (point B). In step II the result =525 nanoseconds. Since the last operation (in the micro-instruction labeled FIRST) was a READ, the flow chart in FIG. 9 directs you to step III which when completed, provides pause time =35 nanoseconds in this case. Returning to FIGS. 6A and 6B, the result through P3 =4.times.35 nanoseconds =140 nanoseconds. Since micro-instruction DATA will be short, P4 and P5 are entered immediately with a resulting total time for this micro-instruction =210 nanoseconds (point D). Micro-instruction END will be long (point E) because of the CM branch. One may look at the partial microroutine just illustrated and consider that one can simply subtract the time for all micro-instructions executed (before the micro-instruction labeled DATA but after the one labeled FIRST) from the memory cycle time and in this case obtain 210 nanoseconds; however, this procedure will not always yield correct results. The next microprogram example illustrates why this is so.
Write Into Memory Example
Consider a write operation to main memory using the following microroutine. For this example, assume the DMS is installed. Also, consider conditions for the microbranch (in micro-instruction CHECK) not met and no memory operation in progress as entry is made. Again, note that the microroutines in the examples shown are used only to show timing relationships and the use and advantages of the present invention.
micro-instructions labeled ENTER and WRITE (point A) both execute in 175 nanoseconds each and the main memory write cycle timing begins at point B. Micro-instruction CHECK executes in 175 ns (point C) since branch conditions are not met, then a read from main memory is next attempted. Using the flowcharts in FIG. 9, it can be seen that the calculation for the time shown at point E is made for micro-instruction GO as shown below. (The write time at point D is 630 nanoseconds because of the DMS factor.)
Remote Program Load
The remote program load feature allows users of the preferred embodiment to initiate automatic system bootstrap operations. These may include a complete reloading of system software from disc, communication line, or other user defined device and a restart of the particular software operating system selected. This feature is particularly useful for systems where experienced operators are not available or the computer is located at a remote or inaccessible location. The feature allows a central computer to issue commands to reboot or reload a remote or unattended satellite system.
The remote program load allows a reloading procedure to occur automatically and unattended. Process can be triggered in a number of ways, but most commonly under one of the following three conditions; a special interface is connected to the hardware "run" line, a cold-start of the computer is initiated by the operator, and upon the execution of a HALT instruction. All of the above can be conditioned upon the happening of certain other preselected conditions. In the preferred embodiment some of these conditions may be set using the "configuration switch block" located in the CPU circuitry. The configuration codes for the preferred embodiment are shown in FIG. 10.
The run line in the preferred embodiment is a two-way communication link with the processor. It is not only an indication of the RUN flip-flop status, high if the computer is running, low if it is not running, but through an appropriate interface as shown in FIG. 11, it can be used to control the RUN state of the processor remotely. When the base of transistor 510 is driven, the collector must sink up the 250 milliamps from the driver which is driving the run line. (In the preferred embodiment, the driver is assumed to be an 8T13 or the like, manufactured by Signetics, Raytheon, and others.) The processor can detect the transition in typically less than one microsecond. The tolerance for duration of the hold-down of this line is very broad, but the line should generally not be held low for longer than a few milliseconds to avoid retriggering the HALT process period. The computer is then switched into the HALT microprogram where, in the preferred embodiment, it performs a sequence of operations as shown in FIG. 12. In the preferred embodiment it is to be noted that the HALT microprogram includes full memory and CPU micro-diagnostics as well as a reconfiguration of the computer to the present settings. If the remote program load feature has been selected, the computer is not halted, but rather control is transferred to a predetermined user microprogram at location 33000.sub.8.
Basically, to initiate externally triggered remote program load, the I/O device must drive the communication line 512 high to reset the run flip-flop as described above to begin the microprogrammed RPL procedure. For example, a CRT terminal could have this line hardwired to a special load button on its keyboard for initiating this sequence. Alternatively, a pair of data communication boards linking two minicomputer based systems together could have specific command decode logic on each board, which upon the detection of a certain predetermined command sequence, would raise line 512, clear the run flip-flop, thereby initiating the RPL procedure.
Microprogrammable Processor Port
The microprogrammable processor port (MPP) permits external hardware to be directly connected to the central processing unit and interfaced under fast and direct microprogram control. The physical connection to the microprogrammable processor port consists of two 50-pin connectors which carry the timing control and addressing signals as well as the internal S-bus lines from the computer to the external device. In the preferred embodiment, this high-speed direct link between the control processor and the external devices allows transfer rates up to 5.7 megawords per second in short bursts and up to 1.5 megawords per second for longer sustained synchronous transfers.
The processor port is a controlled processor programmed I/O method, different from direct memory accessed channels and the typical I/O bus schemes used for standard I/O. The processor port is particularly suited for data transfers to and from buffered interfaces, external high-speed random access memory devices, special external processors such as fast Fourier transform or floating point processors, as well as video displays and other high-speed synchronous devices where it is advantageous to couple them directly to the central processing unit.
Because the basic control of the processor port is provided through microprogramming, any hardware interface can be accommodated by simply writing a program to control it.
The microprogrammable processor port interface signals are listed below in Table VII. Also, the microprogrammable processor instructions and their relative timing waveforms are shown in FIG. 13.
The following sample applications below will illustrate the operation of the microprogrammable processor port. These examples are presented in order of increasing complexity.
Synchronous Data Burst Using The Microprogrammable Processor Port
Assuming a device will transfer 12 words of data (192 bits) at very high speed, it will use the P5 subinterval signal as a clock, passing one data word per cycle, thus running synchronously with the central processor. The signal PLRO is used to address the device. The data transfer is triggered by a handshaking sequence which uses signal PPlSP to indicate "processor-ready" and MPP to indicate "device-ready". Thus, when PLRO and PPlSP are both true, the device performs the required operation, i.e., obtains the data, presets the counters, etc. Then, the device indicates readiness by setting MPP true and clocking out 12 words of data to the S-bus connection. See FIG. 14 where the waveforms for this transfer are shown. Since this transfer is under microprogram control, the words can be directly loaded into scratch pad registers for further processing.
Device Selectable Synchronous Transfer Using MPP
For this example, it is assumed that the device allows the implementation of an addressing scheme and synchronous data transfers of varying lengths up to 255 words are possible. The external device's circuitry is modified so that the PP2SP-True signal can be used with PLRO-True to indicate "address mode" operation. In this mode, the lowest four bits of the S-bus output will be latched into an address register and compared against the specific address of the device. If the device-select code does not match, the device will clear itself and wait for another address mode pulse. If there is a match, the handshake sequence described in the synchronous burst example above will initiate and synchronous the device with the central processor.
Data will now be clocked out using the subinterval P5 signal. A microprogram using this technique is shown in Table VIII below.
Note that the data transfer rate will typically be much lower than that which is possible because normally some microprogram steps are added to process and store the data. Also, the end of transfer is signalled by placing all devices in "address-mode" and sending device select code zero which corresponds to a nonexistent unit. This will clear and reset all devices. This program is initiated by a microinstruction which maps to the microcode shown in Table VIII. The four least significant bits of the microinstruction are used to specify the device-select code. The words in the two memory locations immediately after the microinstruction specify the address of the destination buffer and the address of the word count for the transfer. Also, for improved transfer speed, the microprogram illustrated in Table VIII uses an 8-bit dynamic word counter (CNTR) an approach that accounts for the 255 word transfer limitation in this preferred embodiment. Alternatively, a scratch pad register could be used as a counter with a range from 0 to 32,767. Also, by saving the P register and scratch pad 2 and using CNTR as a counter, the microprogram achieves the maximum data throughput to memory that is possible (point A in Table VIII). Also, the routine minimizes loop time by incrementing the buffer address (kept in the P register) and decrementing the word count (in CNTR) in one instruction. (See point B in Table VIII.
Interruptable, Asynchronous Transfers Using The MPP
In this example, we use a scratch pad register 2 as a word counter to allow greater than 255 word transfers and permit asynchronous operation. Since the total transfer time may be very long depending on the word count and transfer rate, the microprogram must be interruptable by other I/O devices. FIG. 15 shows a block diagram of a circuit using a counter similar to the Fairchild 9403 or the like to form a FIFO (first-in-first-out) buffer and having control logic circuitry to interface this device to the microprogrammable processor port. A microprogram designed to communicate with this device shown in FIG. 15 in an interruptable asnychronous fashion is shown in Table IX.
As in the previous example, an address latch and comparator are used for device-select with PP2SP-True. The device/ microprogram uses PP1SP-True and MPP-True to "handshake" on the transfer of each data word. At each loop of the handshake, the microprogram checks for pending interrupts. If interrupt processing is required, intermediate results (word count and current buffer pointer) are saved, and the microprogram jumps to an interrupt process routine.
Note that with the microprogrammable processor port design an important element is to have a data buffer large enough to hold the entire burst of data. Thus, the elimination of "handshaking" during the data transfer is a prime advantage of the microprogrammable processor port of the preferred embodiment.
Detailed Description of the Schematics
Referring now to FIG. 16, there is shown the timing circuitry of the preferred embodiment. Timing is generated from a third overtone 28.5 MHz crystal control master oscillator circuit 1610. The oscillator consists of two high-speed inverters 1611 and 1612 with negative feedback provided by two resistors 1614 and 1615. Selection of the resistors and the blocking capacitor 1617 determine where the crystal will "lock in" (i.e., the fundamental or desired overtone frequency).
The 28.5 MHz buffered output signal on line 1620 is passed through two NAND gates 1625 and 1630 to an eight-state counter 1640. The purpose of the two NAND gates 1625 and 1630 is to allow an external oscillator to be connected to the central processing unit.
Counter 1640 includes three flip-flops labeled PA, PB and PC. These three flip-flops define clock signals that are used in the central processing unit. They are sequenced by the steering logic 1638 which feeds the D inputs of these flip-flops. Assuming that both signals SHORT and PAUSE are high, an analysis of the circuit would show that flip-flops PA, PB and PC will sequence through the states as illustrated by the timing diagram 1641. The eight subintervals P1, P2, P3, E1, E2, E3, P4 and P5 are thus generated.
The individual P-period or P-periods are decoded from the flip-flops PA, PB and PC by combinational logic throughout the central processing unit. Also, many of these signals are decoded as shown in FIG. 16B for use throughout the machine. Note that the signal FRZFF is used as a qualifier for generation of these timing signals. Signals whose generation are not stopped during a freeze as hereinabove described include a suffix NF to indicate that they are "not freezable".
The timing logic for the I/O system of the preferred embodiment is shown in FIG. 23A. A buffered image of the PC flip-flop signal clocks a mod 5 counter 2410, shown in FIGS. 23A and 23D, at the end of every microcycle. The counter is implemented using JK flip-flops individually labeled TA, TB and TC and counts as the timing diagram 2402 in FIG. 24A indicates. These timing signals generated are subsequently decoded into 5 distinct periods, referred to as T-periods, numbered T2, T3, T4, T5 and T6, that control I/O system operation. Five such T-periods are defined as one complete I/O cycle. Various circuitry associated with the counter is designed so that no lock up in any state can occur.
In general, the I/O system performs a specific task no more frequently than in one complete I/O cycle. Thus, the 5 T-periods are used for precise implementation of specific tasks.
The purpose of the CPU freeze as hereinabove described is to selectably disable control processor micro-instruction execution for certain time periods without disabling other computer activities. With the exception of I/O related activities and memory refreshing operation, if a refreshable memory is used, all computer operations are suspended when the freeze is invoked. Therefore, destinations of the nonfreezable signals are to be preselected to include usually the I/O accessory and memory systems. A freeze flip-flop 1650 is updated at the end of every microcycle and determined if the control processor is to be frozen during the next microcycle. Gate 1648 is fed by circuitry which monitors the four conditions in the preferred embodiment that may freeze the CPU.
One condition that will freeze the CPU is if the control processor is preparing to initiate a memory cycle which is indicated by the signal RDORWT on a line 1671 and the existence of either a memory refresh or a memory accessory data transfer, then a control processor freeze will occur. The signal REFRESH is generated by the memory controller to indicate to the CPU that the memory resources will be occupied during some preselected time. In the preferred embodiment, the signal REFRESH on line 1642 is generated during T5 and T6 to indicate that the refresh will occur during time periods T2 and T3.
A second condition which will freeze the central processor occurs when an interrupting I/O device requires processing by the CPU. The interrupt acknowledge signal (IAK) is sent to the I/O system, when the device interrupt address is loaded into the central interrupt register of the computer.
A third condition that will "freeze" the CPU is when the signal IOGSP on line 1645 indicates that an I/O instruction is to be executed. Since as described above, the control processor is normally executing micro-instructions asynchronously of the T-period interval designations, a freeze is required until the two timing systems are synchronized. The control processor therefore freezes until the signal ENF on line 1643 is true.
The fourth condition for a freeze is when a direct memory access transfer is being used to directly transfer data from an I/O device into memory without transferring it through the central processing unit. The signal DMACYC on a line 1644 is interrogated each T5 time period to determine whether the CPU should be frozen during the duration of the next T2, T3, T4, T5 I/O time periods to allow the I/O device to use both the data bus and memory system.
Long/Short Cycles
In the preferred embodiment, three circumstances will cause the control processor to generate a long microcycle rather than a short microcycle. These are an unconditional branching instruction or a conditional branching instruction where the branch condition is met, and I/O instruction where it is required to extend the microcycles corresponding to I/O time intervals T3, T4 and T5, or some memory operation such as the Dynamic Mapping System used in the preferred embodiment, will be unable to complete the interfacing tasks required in the 175 nanoseconds which is normally available after a micro-instruction is loaded in the micro-instruction register. This extension of the microcycle is performed by gates 1631 and 1632 as shown in FIG. 16A. These gates detect these special conditions early in every microcycle before the signal PB goes high at subinterval P3. If no special conditions are present at subinterval P3, the SHORT signal goes low and forces all inputs to the P flip-flops to go high. Therefore, at the next clock period, i.e., the end of subinterval P3, all three flip-flops will be clocked to the P4 state. Thereafter, counting proceeds to subinterval P5 since the signal PB is still low. In this case, the state counter would have counted P1, P2, P3, P4 and P5 for a total of five P-periods or a total of 175 nanoseconds.
The P subinterval flip-flops utilize an eight-state Gray code which is a binary representation characterized by only a single flip-flop changing at any common clock pulse.
Pause
The processor may request an operation of main memory while the memory is completing another task. The memory system will inform the central processing unit that it is busy by causing a signal MBUSY on line 1651 (FIG. 16C) to go low. In the preferred embodiment there are four micro-orders that affect memory; they are READ, WRTE, RJ30 and TAB (see appendix B for instruction definitions of these micro-orders). These conditions are detected by gate 1655. After the signal PB goes high at subinterval P3, the PAUSE signal will go low and force the D inputs to the P interval flip-flops to the binary equivalent of P3. This P3 state is then cycled back into the P flip-flops repeatedly until the signal NOT MBUSY goes high to allow the microcycle to complete the normal sequence of subintervals. The resulting micro-instruction is thus extended 35 nanoseconds each time the P3 subinterval is repeated.
The signal NOT TST on a line 1622 is synchronized with the fourth flip-flop in counter 1640. The NOT MBUSY signal on line 1651 is synchronized by a separate dedicated flip-flop 1657.
CONTROL PROCESSOR OPERATION
The Run Flip-Flop
The Run Flip-Flop 1670 (RUNFF) is shown in FIG. 16B. The RUNFF commands the control processor to either process micro-instructions that fetch and execute main memory instructions or execute micro-instructions that perform operator panel scanning tasks.
The RUNFF is set in one of three ways. Pressing the RUN button (RUNB) on the operator panel will set the RUNFF if the run enable (RUNE) signal on a line 1677 is high. The RUNFF may also be set if the SRUN micro-order is executed in the SP field or if the power fail logic detects an auto restart condition (PWUST). This will be discussed in more detail later.
The RUNFF is reset when a micro-instruction with SHLT in the SP field is executed, when PONB is low, when a HLT instruction is executed (HLTIO), when the HALT button on the operator panel is pressed and the RUNE signal is high, and when a memory parity error is indicated and switch AlSl is in the HALT position. The RUNFF is also reset when an I/O interface wishes to call the Remote Program Load (RPL) facility of the computer. This requires directing the control processor to the operator panel microroutines. The interface pulls down on the RUN line in the I/O backplane when the RUNFF is set. BRUNFF is high, bu the RUN line will be low. The conflict situation is sensed by a gate 1679 which causes the RUNFF to be reset.
The internal logic in the computer is initialized by the Preset (PRST) line. PRST is primarily created by the PRESET button on the operator panel.
RPL Circuitry
The remote program load feature is a dual purpose capability in the preferred embodiment that permits (1) automatic system bootstrapping on cold power-ups and (2) automatic down line system loading from a second computer system via the I/O system of the first system.
Several previous minicomputers had hard-wired or ROM imprinted instruction routines that execute a fixed set of instructions when the attendant presses an appropriate switch on the operator panel or that automatically communicate and wait for load instructions from a terminal device when power is applied to the minicomputer. The remote program load capability of the preferred embodiment is a hardware-firmware combination that begins with firmware reading a set of hardware configuration switches to obtain, inter alia, a ROM number and I/O select code information, unpacking the selected ROM contents into 64 16-bit words, loading them into main memory, modifying all I/O instructions with the select code indicated by the configuration switches, setting the program counter to the starting address of the routine, and placing the preferred embodiment minicomputer in the running state. The ROM contains an appropriate bootstrap program that accesses the selected peripheral device to obtain the operating programs desired.
This feature thus allows the single application of power to the preferred embodiment to automatically bootstrap the system up entirely without further operator interaction. This is especially advantageous when no terminal is connected to the system, there is no operator panel installed, or the computer itself is buried in an inaccessible location in the system. The remote program load feature also allows automatic system bootstrap operations to be initiated without having to gain access to the system directly (down line loading). Any two computers that are running properly are capable of communicating with one another and pass programs or log data between them. However, if the operating system in one computer is destroyed for any reason and that system is unattended, it is impossible to reload the destroyed system and return to normal operation. Because the operating system is not functioning properly, the communication link between the two systems could be down. An attendant would be required to halt the ailing system and execute a load routine via its operator panel. The remote program load approach in the preferred embodiment is to permit a master computer system the capability to pull a hard-wire line in the slave (down) system's I/O backplane that calls immediate attention to the down computer. Like cold power-up discussed above, the remote program load firmware/hardware feature calls for the appropriate ROM loader to be loaded, all I/O instructions configure with the proper select code, and the computer placed in the running mode, thus executing the loader program. The loader, of course, brings in a valid copy of the desired operating system and the slave system is thus again running properly.
The remote program load hardware consists of the run flip-flop 1670, the buffered run flip-flop 2110 (FIG. 21) and various gates. When the run flip-flop is on, the computer is executing machine language instructions. The buffered run flip-flop signal on a line 2111 is also high so that the run line which travels up the backplane is also high. The output of gate 1680 is low so that the output of gate 1679 is high which is the normal running state. Should the backplane run line be pulled low as explained above, the output of U389 would become high, allowing a low pulse during P3 at the output of gate 1679 to clear the run flip-flop thus sending the control processor from the run mode microcode to the halt code micro-instructions. The tests that it makes are outlined in FIG. 8. A listing of this RPL microprogram is in Appendix D.
The configuration switches 1710 are shown on FIG. 17. They can be examined by reading them onto the S-bus of the preferred embodiment by passing them through gate 1715 when the signal DESEN on line 1716 is low. But .phi. is the first bit tested and if it is set, a check is made for the inclusion of the operator panel. The line SFP (standard front panel) is low if the operator panel is installed, high if it is not. It is tested via the control branch multiplexes 2120, 2121, 2122, 2123, 2124 (FIGS. 21A and 21C). If the operator panel is not installed, or if the minicomputer power supply switch is in the lock position (tested similarly by the line RUNE, low if it is in lock, high if it is not), the remote program load firmware is entered. The firmware looks again at the data in the configuration switches 1710 to select one of ROM loaders 1805, 1810, 1815, 1820 (see FIG. 18) and load the desired loader into main memory. The select code of the I/O device is read from the configuration switches per the format shown in FIG. 10 and merged into all I/O instructions in the loader program. The program counter is set to the starting address of the loader, the computer is set into the run mode, and the loader is again executed.
Control Memory Address Register (CMAR)
The rising edge of the CMAR clock signal occurs at CP12+E1E2. CP12 is responsible for incrementing the CMAR at the beginning of the microcycle. If a branch is detected, LCMAR (Load CMAR) comes high requesting a long microcycle (from FIG. 16) and permitting E1E2 to occur. At E1E2 (end of P3) the CMAR again may be clocked. (Registers 2030, 2031, 2032, and 2033 in FIG. 20 comprise the CMAR).
If the enable input to the CMAR is high it may count; if low, it will not. Since the enable input is high only during P5 and P1, it will easily bracket the beginning of a microcycle and CP12 will increment the CMAR. The load control is low during the middle of every microcycle if a branch is to occur so the parallel load operation happends synchronous to the clock at E1E2.
The signal LCMAR on line 2010 is the result of straight-forward combinational logic that looks at all of the ways the control processor may branch (FIG. 20). The three OP fields RTN, JSB and JMP can force a branch if no CNDX is in the SP field (word type 4), or if the condition interrogated by the COND field is met (word type 3). The MET signal on line 2012 is low if the condition specified by the COND field is met.
JTAB will force a branch if the indirect flip-flop (INDFF) is not set (i.e., no INCI has preceded the JTAB in the SP field), or if the condition specified is met. RTN in the SP field always forces a branch. A special return (SPRTN) is requested if the micro-orders ASG or SRG2 require branches. The PONB term will be explained later.
The inputs to the CMAR are the outputs of the nine multiplexers 2039-2047 that are switched by eight basic modifiers that determine how the CMAR is to be loaded. These can be listed from the first eight SP field micro-orders: RTN, JTAB, CNDX, ION, RJ30, J74, IOG and NOP. The actual logic implementation is simplified by using the least significant three bits of the SP field of the MIR as the selector inputs to the multiplexers. The multiplexer data inputs can be connected directly to the appropriate source that matches the function called by the SP modifier. A special case is RTN in the OP field which must force the selector inputs to match the code of RTN in the special field. Using this associative bit pattern technique, three other SP field micro-orders (STFL, RPT and IOFF) may be used in word type 4 micro-instructions without modifying the jump target because their least significant three SP bit patterns are the same as NOP. Table X reviews the bit patterns at the selector inputs of the multiplexers.
Although there are eight different ways to branch through CM, there is substantial similarity among five of them. ION, RJ30, J74, IOG, and NOP are all unconditional jumps that modify the target address in the micro-instruction in the lower four bits. The other three modifiers are unique and call their own selected pattern. RTN requests an address from the SAVE stack, JTAB requests an address from a PROM jump table based upon the IR, and CNDX calls for a nine-bit (512-word) address displacement around the current CMAR address.
Halt or Interrupt Detection
When either a halt state or a pending I/O interrupt condition prevails in the computer, an interruption of the normal fetch-to-execute flow is made. The CMAR is then loaded with the constant 6 when JTAB is executed without a preceding 1NC1 at CM location 1. The special conditions are detected by a gate 2003 and a gate 2004 on FIG. 20. Gate 2003 disables the lower jump table 2007 thus forcing its output to all ones. The signal HOIFF on a line 2013 disables all of the multiplexers, except multiplexers 2040 and 2041 to zeros. Since the multiplexers are switched to the jump tables, multiplexers 2040 and 2041 pass their ones to their outputs, consequently loading the constant 6 into the CMAR. The terms PONFF and PONFF will be discussed later.
The Save Stack
The save stack holds microprogramming subroutine return addresses. Three levels of microsubroutine calls are permitted by the control processor, but the stack is actually implemented by a four-word dual-port RAM. Control logic makes the RAM appear as a stack. The fourth register normally holds an all-zeros address that is used when more microsubroutine returns are executed than branches.
Three levels of microsubroutine calls are available for executing an instruction after the fetch micro-instructions. Initially a read (or "RTN" pointer) addresses an all zeros word and a write (or "JSB" pointer) addresses the first available save word. (See FIG. 25)
Any executed return will load the contents of the CMAR at the address pointed to by the RTN pointer. In this case, a return to CM location 0 will occur. An executed JSB will write the return address into the location pointed to by the JSB pointer and then advance the pointers. If a JSB occurs from location 200, the pointers will be advanced to the state shown in FIG. 26.
An executed RTN will place the contents of the save at the address pointed to by the RTN pointer (201 in this case) into the CMAR and the pointers will move back to the previous state after the fetch. Deeper levels of subroutines operate in a similar manner with the pointers moving back and forth always one away from each other. When one more RTN is executed than a JSB, a branch to zero is made. FIG. 20 contains the circuitry for the save register.
The RAM includes a write enable, a read enable, two write address and two read address lines. Simultaneous reading and writing of two different locations is permitted, and this is the key to the selection of the control logic used.
Register 2080 is a left-right four-bit shift register set up in a circular rotate configuration which actually holds the read address and write address pointers. The two mode controls manipulate the pointers in response to subroutine branches and returns. S0 is high when a RTN is executed, S1 is high when a JSB is executed, and JTAB forces both high. Synchronous to P3, a RTN causes a right shift, JSB causes a left shift, and JTAB forces a load.
Three of the four outputs of the shift register become the read and write address pointers to the save stack which comprises registers 2081, 2082, 2083 and 2084. The read enable is always low since reading of the return address always happens, and, assuming that the output signal from gate 2086 is high, a write occurs every P12. Therefore, the saved word pointed to by the WRTE pointer is actually written every microcycle with the contents of the CMAR. The CMAR points to one word beyond the address of the currently executing micro-instruction. When a JSB is executed, both pointers are advanced beyond their present positions. The last write is left behind and is the valid future return address.
FTCH and JTAB are specific hardware-oriented micro-orders that are never used by user-microprogrammers. They are infrequently used in the base set. In regard to the save stack, the most important usage is at CM locations 0 and 1. First the FTCH is executed clearing the 74LS191 shift register and setting both pointers to 00. Next, the micro-instruction at CM location 1 is executed. During P1 and P2, a save stack write occurs storing the present value of the CMAR into the location pointed to by the JSB pointer (00). Recall that the CMAR is equal to 2 at this time. Since the micro-instruction located at CM location 2 includes a JTAB, the CMAR 1 and 2 inputs to the save stack are masked by gate 2003, thus loading all zeros. After P3, JTAB causes a load of constant 1001 into the register 2080, completing the initialization of the control logic.
The shift register simulates the effect of two pointers displaced by one position from each other. See Table XI below:
Note that the write address (JSB) is the read address (RTN) for the next subroutine level. RAM 00 holds all zeros, RAM 01 holds the first subroutine return address, RAM 11 the second, and RAM 10 the third. See Table XII below:
jtab also occurs at locations 3 and 306 in the base set. The JTAB at 3 in the MRGIND routine again forces zeros into the save stack. While JTAB is being executed, the CMAR is equal to 4 (advanced by one) but gate 2003 masks this bit before loading into the stack. The JTAB at 306 in the operator panel routines is not required to load zeros onto the stack, but rather to initialize Save control logic.
The purpose of gate 2089 is to detect the third JSB state and disable the write pulse which would modify RAM 00 which is the all zeros word and must remain valid.
Control Memory
The CMAR drives the base set control memory PROMs 2090, 2091, 2092, 2093, 2094, 2095. Gates 2060-2063 and the related circuitry decode the module addresses.
Micro-Instruction Register (MIR)
The 24-bit micro-instruction from the control memory PROMs appears on the ROM bus and is clocked into the MIR Registers 2072, 2071, and 2096 at the beginning of the microcycle by clock signal CP12. The outputs of the MIR travel throughout the CPU. Simultaneous with loading of the MIR, the CMAR is incremented to point to the next micro-instruction.
Decoders
The computer described herein employs a vertical format for its micro-instruction. Vertical format micro-instructions are often characterized by micro-orders encoded into several fields of operation. Each field is usually decoded to a unique function it is responsible for. The advantages to using a vertical format include cost (a relatively narrow micro-instruction width is required) and ease of microprogramming. In a horizontal format a micro-instruction is substantially wider (more than one hundred bits is not uncommon) and little or no decoding is required because bits in the micro-instruction may themselves be responsible for a single function. The advantage is speed (no decoding circuits may be required decreasing microcycle time). Therefore, the preferred embodiment uses a 24-bit vertical format micro-instruction.
FIG. 19 is a detailed schematic diagram of the decoding circuits for the micro-instruction fields. The fields not entirely decoded are the SB and ST fields that reference the sixteen registers in the RAM, the ALU field, and the COND field. Understanding the operation of a decoded micro-order requires a parallel review of the definition of each micro-order.
Operation Field
Decoders 1910 and 1911 are high-speed three-to-eight decoders that decode the OP field micro-orders. IMMOP is combined with MIR bit 18 to determine whether the lower byte (LBE) or upper byte (UBE) is to be enabled onto the S-bus. The READOP and WRTEOP micro-orders logically set their respective flip-flops at the end of the microcycle. The falling edge of the READ or WRTE pulse actually initiates the memory cycle. The signal DMAFRZ on a line 1901 is low when an extended memory such as the Dynamic Mapping System referenced above requires the S-bus during a data transfer. Its purpose is to disable the decoding of the OP field primarily to inhibit the IMMOP signal on line 1903 to prevent data from being enabled onto the S-bus.
Special Field
All thirty-two special field micro-orders are decoded by decoders 1920, 1923, 1924, 1925. P5 (at the end at the microcycle) is the basic timing pulse used to actually trigger these operations.
Store Field
The sixteen SB and ST field micro-orders that are not associated with the sixteen registers contained in the RAM are decoded. The RAM addressing logic is discussed later. All registers of the CPU are loaded at the end of P5 (end of the microcycle) except the Instruction Register (IR) and the Memory Address Register M which are loaded from the S-bus at P4 (due to signal skewing and delays, it may be marginally unsafe to clock data off the S-bus at the end of P5).
The ST field is decoded by decoders 1921 and 1922. The decoding is disabled by the rather complex combinational logic which provides the signal on line 1919. The first term JMPOP+JSBOP+CNDXSP is a detection of word type 3 or 4 micro-instructions. Such a micro-instruction does not operate on data, and hence does not require usage of the S-bus. However, MIR bits 10 through 14 will enable a register onto the S-bus and MIR bits 5 through 9 will decode a register to be stored if not disabled. The second set of qualifier conditions is input through gate 1960 and is essentially a logical definition of the micro-order JTAB (JTAB is to suppress the operation of the ST field if an indirect MRG instruction is in the IR or if a JMP, JSB, ISZ, STA or STB instruction is in Instruction Registers 1802, 1803, 1804 and 1805.
A third qualifier inhibits the ST field if a halt or interrupt condition is pending as JTAB is decoded. JTAB will force the CMAR to CM location 6 as explained previously.
The A and B addressable flip-flops 2205 and 2206, (ABFF), shown in FIG. 22A, can be clocked at the end of every microcycle. Addresses 0 and 1 are detected on the M-bus and the flip-flops are set accordingly.
S-bus Field
The SB field is decoded by decoders 1945 and 1951 (FIG. 19C). Decoding is suppressed if DMAFRZ is low or if IMMOP has been decoded in the OP field. IMMOP implies a word type 2 wherein an eight bit byte is to be enabled onto the S-bus rather than a register.
IOI is typically used to execute I/O input instructions. The literal definition of IOI is to gate the I/O bus onto the S-bus. This is performed by signal EIOBO on line 1965. There are, however, certain cases in this context when the I/O bus must not be enabled onto the S-bus. All I/O interfaces reside on the I/O bus, but accessories with select codes less than ten reside on the S-bus. To enable the I/O bus onto the S-bus when an I/O input instruction is requesting data from, e.g., a memory channel, would be disastrous. Therefore, the gating of the I/O bus is inhibited if select codes one through seven (SC1-7) are detected. Select code 0 does enable the I/O bus but this is just to maintain compatibility with earlier computers when executing LIA 0 to LIB 0 instructions. Therefore, IOIEN is qualified with SC1-7 by gate 1960 and enables both the lower byte of the I/O bus by signal ELIOB on line 1966 and the upper byte of the I/O-bus by signal EUIOB on line 1968.
Conditional Field
The control processor jump conditional logic is illustrated in FIG. 21. MIR bits 15 through 19 compose the COND field and are selectors to a high speed multiplexing tree comprising decoders 2120, 2121, 2122, 2123 and 2124 which selects one of thirty-two possible jump conditions. MIR 14 is the reverse jump sense (RJS) bit and is used in the last stage to complement the condition sensed. The signal output of the tree is the signal MET which goes into the CMAR load logic previously discussed with respect to FIG. 20.
Arithmetic/Logic
The arithmetic/logic section is shown in FIGS. 17, 18, 21 and 22. FIGS. 17 and 18 are the most important and contain all of the data registers and the four 16-bit data busses in the CPU. The tri-state bidirectional positive-true S-bus is the main data bus upon which the SB field of the micro-instruction enables a register for the duration of the microcycle. The S-bus drives the ALU which in turn drives the RS and ground true T-bus. The CTL-compatible bidirectional positive true I/O bus can be gated from or onto the S-bus. The tri-state positive true M-bus is shared by the CPU's M-register and DCPC's memory address registers. It may also be enabled back onto the S-bus.
The RAM
There are sixteen registers that are physically located in the 16-word by 16-bit RAMs 1780, 1781, 1782 and 1783 as shown in FIG. 17. The registers contained in the RAM are the P (program counter), S, X, Y, SP and eleven scratch registers. The 74S189 RAM or the like, manufactured by Texas Instruments and others is suitable for use in the preferred embodiment. Writing and reading at different locations are not permitted simultaneously. The control logic illustrated above and below the RAMs determines how they are to operate. The microinstruction control bits of the SB and ST fields are multiplexed by multiplexer 1794 into the address inputs of the RAM. The Write Enable (WE) lines is low during P5 of any microcycle that the ST field calls for a RAM register to be loaded. The Chip Enable (CE) input is low when a RAM register is to be read onto the S-bus or stored from the T-bus.
There are four possible permutations of RAM registers between the S-bus and Store fields.
First consider the case of a micro-instruction that has a RAM register neither in the S-bus nor Store field. In this case, both MIR14 and RAMST are low, so both CE and WE are high and the RAM remains completely inactive during the microcycle.
Arithmetic/Logic Unit, Rotate Shifter, and L Register
The ALU is made of four high-speed ALU circuits 1730, 1731, 1732 and 1733 and one carry look-ahead generator 1729. The ALU is driven directly by MIR bits 15 through 19.
The Rotate Shift logic is made of eight dual four-to-one multiplexers 1741, 1742, 1743, 1744, 1745, 1746, 1747 and 1748 respectively. They function as a high speed shifter, capable of shifting left one bit, left four bits, right one bit, or straight pass through in 9ns.
Instruction Register
The IR is located in FIG. 18 and comprises registers 1802, 1803, 1804 and 1805. The lower eights bits are also called the CNTR. The CNTR can be read onto the S-bus if CNTR is in the SB field. ##SPC1## ##SPC2## ##SPC3## ##SPC4## ##SPC5## ##SPC6##
Claims
I claim:
1. Computer apparatus comprising: memory means having instructions stored in addressable locations therein for outputting said instructions in response to address and timing signals; control means coupled to said memory means for providing said address signals thereto and for causing instructions to be read therefrom in response to timing signals; instruction register means coupled to said memory means for storing electrical representations of instructions read from said addressable locations of said memory means; instruction decoding means coupled to said instruction register means for generating control signals in response to the contents of said instruction register means and said timing signals; data means coupled to said instruction decoding means for performing preselected logical and arithmetic operations in response to said control signals and for providing state signals indicating the result of said logical and arithmetic operations during the execution of the instruction contained in the instruction register; timing means coupled to said control means, said instruction decoding means and said data means for providing a plurality of selectable instruction execution timing intervals and for providing timing signals corresponding to one of said execution timing intervals to said control means, said instruction decoding means, and said data means in response to an execution control signal; timing control means coupled to said instruction register means, said data means and said timing means for causing variations in the basic instruction timing interval; said timing control means further including means for selecting one of said instruction execution timing intervals and means for providing said execution control signal in response to both the contents of said instruction register means and said state signals.
2. The computer apparatus as in claim 1 wherein said timing means includes means for providing a basic instruction timing interval comprising a plurality of sub-intervals and said timing control means comprises means for causing the number of timing sub-intervals to vary in response to said state signals, thereby varying the basic instruction timing interval.
3. The computer apparatus as in claim 2 wherein said timing control means further comprises means for suspending the providing of said basic instruction timing interval in response to said state signals.
4. The computer apparatus as in claim 3 and further comprising a master memory means coupled to said data means and having a plurality of different memory types having different operating parameters; said timing control means further comprising means for detecting these different memory types and for varying the number of timing sub-intervals in response thereto.
Patent Citations (10)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US3656123 | 1972-04-01 | Carnevale et al. | |
| US3764992 | 1973-10-01 | Milne | |
| US3771144 | 1973-11-01 | Belady et al. | |
| US3809884 | 1974-05-01 | Nibby et al. | |
| US3868647 | 1975-02-01 | Zandveld | |
| US3972023 | 1976-07-01 | Bodner et al. | |
| US3999169 | 1976-12-01 | Perschy | |
| US4037090 | 1977-07-01 | Raymond | |
| US4040021 | 1977-08-01 | Birchall et al. | |
| US4058851 | 1977-11-01 | Scheuneman |
Non-Patent Literature (1)
- Fangmeier et al., "Processor Control by Two Independent Phase-Locked Clocks", Apr., 1972, pp. 3394, 3395, IBM Tech. Disc. vol. 14, No. 11.