Multi-User Analog/Hybrid System

ABSTRACT OF THE DISCLOSURE

BACKGROUND OF THE INVENTION

I. FIELD OF THE INVENTION

II. PRIOR ART

SUMMARY OF THE INVENTION

BRIEF DESCRIPTION OF THE DRAWINGS

ACRONYMS

DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

I. GENERAL OPERATION

A. System 10 Theory and Organization

B. Organization of DCM 25

II. INTERRUPT TIMER CONTROL (ITC) 26

A. Block Diagram Description

B. ITC Microsequencer 134 Firmware Flowchart

C. ITC Instruction Format for Microsequencer 134

III. BASIC BUS INTERFACE (BBI) 21

A. Block Diagram Description

B. BBI Microsequencer 340 Firmware Flowchart

C. BBI Memory Access Sequencer 341

State Diagram

D. BBI-DDL Signal Descriptions

IV. DATA CONVERSION MULTIPLEXER (DCM) 25

A. General Description

B. Block Diagram Description

C. DCM Microsequencer 860 Firmware Flowchart

D. State Diagram For Loading DAM's 24 & Reading ADC 22 or DAM 24

V. PARALLEL ANALOG MULTIPLEXER (PAM) 20

A. Block Diagram Description

VI. DETAILED EXPLANATION OF CIRCUITS

A. ITC 26

1. RTHC and DCM Timer Controls 121, 126

2. Timer Initial Value Buffers 114, 116 and Registers 118 and 128

3. CPU Bus Interface

4. Microsequencer 134

5. Resource Allocator 104

B. BBI 21

1. Physical Address Recognition & Generator 372

2. Response to CPU System 392

3. Tag Register & Decode 388 and Tag Output Encode 390

4. Memory Access Sequencer (MAS) 341

5. Last Accept Register and Last Retry Register 376, 378

6. Service Interrupt Poll 380 and Bus Poll 394

7. In Register 356 and Address Register and Counter 362

8. Status Block 346

C. DCM 25

1. Run Control 816

2. Gorilla Round Robin 830

3. Active Pending Register 786

4. Active Register 794

D. PAM 20

1. Echo or Wait for Read Block 1108

2. Control Timing 1118

VII. TABLE OF COMPONENTS

VIII. LISTING OF THE FIRMWARE FOR GORILLA ROUND ROBIN 830

IX. HOS AND HIOCS SOFTWARE SYSTEMS

I. Field of the Invention

This invention relates to data processing systems and particularly to analog/hybrid computing systems.

II. Prior Art

Analog/hybrid computing systems are known in the art such as Pacer 700 manufactured by Electronic Associates, Inc. Some of these prior hybrid computing systems have operated with a plurality of analog consoles to run hybrid simulations. However, prior systems have been limited in that they could not effectively run more than one hybrid simulation at a time. During the running of a single hybrid simulation for a user there are gaps of time when no computation or computation at slow speed takes place. For example, a user may be checking his program, he may be debugging his program or he may be examining results before the beginning of computation. Further, he may be waiting for initial conditions to be reached. During these times when the user waited or used slow speeds, the full resources of the hybrid computing system were not being taken advantage of. This resulted in a substantial effective loss in computer time of a costly system. A further drawback of these prior hybrid systems has been that they could not operate in a true multi-task mode. While they could run one hybrid simulation for a user they could not at the same time perform other important tasks for other users to effect maximum cost effectiveness of the hybrid system.

An analog/hybrid system for at least two hybrid users which has a digital computer and a main digital memory with each of the hybrid users having their own analog and digital programs. A plurality of analog computers run on individual analog programs with at least one of the analog programs becoming time critical. A system provides for linking data flow between the analog computers and the digital computer under the control of the main digital memory. Multiplex means control the linking system for substantially allocating all of the data flow to the time critical analog program and as available for allocating data flow to the nontime critical analog program.

FIG. 1 illustrates in basic block diagram form a multi-user analog/hybrid system embodying the invention;

FIGS. 2 and 3 functionally represent in basic block diagram form portions of the organization of FIG. 1;

FIG. 4 illustrates in basic block diagram form the DCM section of FIG. 1;

FIGS. 5A-B taken together illustrate in more detail and in block diagram form the ITC section of FIG. 1;

FIGS. 6A-C, 7A-D, 8 and 9 taken together illustrate flowcharts for the firmware of the ITC microsequencer of FIGS. 5A-B;

FIGS. 10A-C taken together illustrate in more detail and in block diagram form the BBI section of FIG. 1;

FIGS. 11A-C, 12A-B, 13A-B and 14 taken together illustrate flowcharts for the firmware of the BBI microsequencer of FIGS. 10A-C;

FIG. 15 illustrates a state diagram for the BBI memory access sequencer of FIGS. 10A-C;

FIGS. 16A-B, 17A-B and 18A-B taken together illustrate in more detail and in block diagram form the DCM section of FIG. 1;

FIGS. 19, 20, 21 and 22 illustrate flowcharts for the firmware of the DCM microsequencer;

FIG. 23 illustrates a state diagram for reading and ADC/DCM;

FIG. 24 illustrates a state diagram for loading DAM's;

FIGS. 25A-B taken together illustrate in more detail and in block diagram form the PAM section of FIG. 1;

FIGS. 26-37 illustrate in more detail and in block diagram form specific circuits of the ITC section of FIGS. 5A-B;

FIGS. 38-42 illustrate in more detail and in block diagram form specific circuits of the BBI section of FIGS. 10A-C;

FIGS. 43-46 illustrate in more detail and in block diagram form specific circuits of the DCM section of FIGS. 16A-B, 17A-B and 18A-B; and

FIGS. 47-48A-B illustrate in more detail and in block diagram form specific circuits of the PAM section of FIGS. 25A-B.

A. System 10 Theory and Organization

FIGS. 1-3 show the basic concept of sharing hybrid resources among at least two hybrid users in analog/hybrid system 10. System 10 comprises a digital system 18, hybrid controllers 14, linkage 16 and a group of analog computers 12. Digital system 18 includes a central processing unit (CPU) 30 having a main digital memory 32 of, for example, 128 k words, peripherals 38 and related monitors, clocks, interrupts, etc. 36. Digital system 18 includes a system bus 34 which is coupled through hybrid controllers 14 and linkage system 16 to analog computers or consoles 12a-f. Analog consoles 12a-f are programmed to run analog programs. Hybrid users 37a-f control the analog programs as well as their hybrid (digital) programs 39a-f stored in memory 32 by means of terminals 38a-f of peripherals 38. It will be understood that each hybrid user 37a-f may use more than one of consoles 12a-f.

CPU 30 is programmed in memory 32 with time sharing programs such as a SEL real time monitor (RTM) program and a terminal support subsystem (TSS) program. In this manner, digital system 18 is shared in a multi-user, multi-task environment so that analog computers 12a-f may be assigned to one or more of hybrid users 37a-f depending on how many users there are in system 10. The data flow through linkage 16 by way of ADC channels 52 and DAM channels 54 between computers 12 and controllers 14 may be defined as a resource of the system 10 which will be shared amount the hybrid users of the system. After these channels have been allocated, data flows through them without intervention by CPU 30.

There are three types of general users of system 10. The first type is a real time hybrid user which is defined as a user who is going to be time critical (descriptively called the "gorilla"). The second type are hybrid users who run their HYTRAN or HOI simulation programs (Electronic Associates, Inc.) or FORTRAN programs in the debug mode although they may be in a simulation mode which does not require time critical operation. These nontime critical hybrid users are called "monkeys" to distinguish them from the gorilla. The third type of user is a standard digital user such as a user 37g on terminal 38g who may be doing batch programming and other tests that are applicable to digital work in digital system 18.

System 10 operates multi-task with a plurality of hybrid users such as six, for example, and allows at least one of the hybrid users to become time critical with the others being nontime critical. Data flow between the analog and digital computers is substantially allocated to the gorilla and available to the monkeys. At the same time, other digital users 37g may be running with other digital tasks and additionally there may be batch mode 41 running.

FIG. 2 functionally shows an expanded multi-user system 10 comprising six analog computers 12a-f, six terminals 38a-f dedicated to hybrid users and up to ten terminals 38g running digital programs while the remaining mode 41 is batch. This operation occurs substantially all at the same time by means of time sharing. In time sharing, the human reaction of a user is such that he believes that he has all the resources of the system while in fact, the operation is serial in nature. It is understood that it takes about 16-20 milliseconds just for the finger of user 37a to hit a button and in that period of time an entire hybrid simulation may be run.

Six hybrid programs 39a-f are shown in FIG. 2 running in memory 32. A plurality of interrupts 36 are connected to entire system 10 and additionally, there are digital programs 32a as well as batch programs 32b. It will be understood that these programs are not required to be resident at the same time and they may roll in and roll out.

Analog consoles 12a -f may be assigned to hybrid programs 39a -f. In FIG. 2, this assignment is functionally shown as through resource allocation, ADC 52 and DAM 54 and then through reentrant channel programs 60a, b. In this manner, there is an environment in which a six console system runs with six separate hybrid users 37a-f all using the same ADC and DAM subsystem 52,54 through resource allocation later described. Further, some of interrupts 36 may be controlled by allocator 40 depending upon the desires of the time critical user or gorilla desires. The time critical user may choose any one of eight interrupts from each of consoles 12a-f with allocator 40 able to control other hybrid interrupts.

FIG. 3 shows generally from a functional vantage point a time critical user or gorilla and nontime critical users or monkeys. When one of the monkeys decides that he wants his simulation to go time critical, that user then effectively obtains the entire system 10. A user would request to go time critical when he wants his simulation to operate in real time with little or no delay in simulating real time functions. Specifically, a user's simulation would need microsecond responses from system 10 resources. On the other hand, the user who does not need to be time critical may operate with approximately 1000 microsecond responses and essentially time is not important. To go time critical, the user would call for a request real time subroutine (the listings of which are later given) which allows the user's analog programs as well as his digital program to run in real time, viz, become time critical.

A monkey may also become time critical by way of a request signal from the analog program itself. If the request is granted by the digital computer 30, that user then becomes the gorilla and he gets as much of system 10 as he wants, i.e., the resources are substantially allocated to him. After allocation of the data flow channels 52,54, data flow through these channels is automatically controlled by DCM 25 without CPU 30 request or intervention and thus, CPU 30 is able to take care of other tasks. It is to be stressed that as available data flow channels may be allocated to nontime critical analog and digital programs.

As shown in FIG. 3, the gorilla has a resident program 32c with the solid lines from the resident program connected to all of the important hybrid interface resources that he requires to run his hybrid simulation. However, it will be understood that simultaneously in a time sharing mode, the monkeys are allowed to run. These are indicated by the dotted line function to the hybrid interface resources with the exception of interrupts 26. The gorilla has interrupts 26 exclusively with complete control over the respective resource allocation.

Monkey programs 32d are given different priorities from resident programs 32c. These programs may be rolled in and rolled out between core and disc. The program with a higher priority may be rolled in from disc to core and executed and then upon completion of that execution, the checkpointed program that was in core of the lower priority may be rolled back in and executed.

It will be understood that system 10, FIG. 1, may include any number of analog computers 12 and this embodiment provides for a total of six analog computers 12a -f. As shown, hybrid controllers 14 include a parallel analog multiplexer (PAM) 20 which couples analog computers 12a-f to system bus 34. Controllers 14 also include an ADC portion (ADC/DCM) 22 of data conversion multiplexer (DCM) 25 which couples system bus 34 to ADC channels 52 of linkage system 16. DAM/DCM 24 couples bus 34 to DAM channels 54. In this way, the function of ADC/DCM 22 and DAM/DCM 24 is to connect linkage system 16 with system bus 34 where linkage system 16 is coupled to analog computers 12a-f by means of analog interconnections.

The remaining major function of hybrid controllers 14 is the interrupt timer controller (ITC) 26. ITC 26 provides the interrupt resources for system 10 and also provides timer resources for the system. Resource allocator section 40 of ITC 26 controls 16 of the total number of interrupts by way of system bus 34. ITC 26 has a timer resource section 44,46 which provides a total of 15 timers. These 15 timers are distributed through system 10 as a resource.

B. Organization of DCM 25

As shown in FIG. 1, DCM 25 comprises ADC/DCM 22 and DAM/DCM 24. Since the ADC 22 and DAM 24 are identical in structure, only one of them, ADC 22, will have its organization described. In FIG. 4, it will be shown that by the use of channel programming control words (CPCW) and related systems and channel programming subroutines, that ADC/DCM 22 is effective to automatically control the data flow channels 52 so that after allocation data flows automatically without intervention by CPU 30. In this manner, ADC 22 and the CPCW system provides for data transfer without request or intervention by CPU 30 so that the CPU can take care of other tasks.

Specifically, FIG. 4 shows a single ADC for six users shown from a general functional viewpoint. CPCW active block 60 indicates a function which provides that any data which is loaded into that function controls the data coming from analog consoles 12a-f and going into memory 32 as for example at a selected memory address 61. As shown, there is a data path through a conventional floating point converter 67, a buffer 70 and an analog digital converter 68 and multiplexer 68a (ADC 52). When anyone of the users wants to obtain data acquisition from his console 12a-f through the mechanism of CPCW active 60, the user addresses through first control signals (produced by the active CPCW) a particular multiplexer 69a-f assigned to his console. Accordingly, data flows from analog consoles through multiplexers 69a-f, multiplexer 68a, ADC 68, buffer 70, converter 67 and into a memory array of memory 32 under the control of second control signals (produced by the active CPCW). In this manner, there is a transition from a scaled fraction to floating point with an automatic loading of the array starting from a location 804c, for example.

With regard to the CPCW's, six CPCW's in memory 32 may be transferred to respective transfer control registers 62a-f. These CPCW's wait in register 62a-f for some stimuli. The stimuli comes from one of the activate groups 73a-f. Each of groups 73a-f has three basic request functions: (1) patch panel control, (2) timer and (3) digital instruction. Any one of these three request functions can cause a CPCW to become active. Once that CPCW is active, ADC and multiplexer 68 is then automatically controlled. With ADC 68 controlled, the data flow from the analog computers is controlled and is stored in memory under the control of the CPCW by way of memory address 61.

In this manner, there are six CPCW's over a period of time dynamically going in and coming out of active area 60. Further, if ADC 68 is in the process of servicing a monkey and that monkey is swapped out because a gorilla preempts the resources, then the monkey's information is saved in save active CPCW 66 while the gorilla is being placed in the CPCW active area. As soon as the gorilla is finished with system 10, the hardware of ADC/DCM 22 automatically places the particular monkey that was interrupted, back into the system and allows that monkey to continue. In this manner, the resources are independent of any particular user and the hardware is reentrant.

It will now be understood that each of the six analog consoles 12a-f may have all the basic functions required to run a separate hybrid simulation with each simulation having a separate hybrid user. The gorilla or time critical user has the principal allocation of the resources and the gorilla user may in fact use up to six consoles 12a-f if he wishes, though he may need to use only one. As previously described, only the gorilla can use the interrupt resources through the data flow resources may be used by the monkeys as available. Thus, in FIG. 4, CPCW active block 60 defines the ADC function that is active at that time with the other five functions waiting to become active. Once the ADC function is completed, then that function is saved if required and the next one of higher priority is loaded into the active area.

A. Block Diagram Description

In FIGS. 5A-B, there is shown a generalized block diagram of ITC 26 in which the entire resource allocator 40 has been illustrated. For purposes of simplification only one of the three real-time hybrid clock (RTHC) timers 46 has been shown as RTHC timer 110. Only one of the twelve data conversion multiplexer (DCM) timers 44 has been shown as DCM timer (ADC) 112. It will be understood that the remaining two RTHC timers, the five DCM timers (ADC) and the six DCM timers (DAM) are of similar construction and operation. ITC system 100 comprises all of the RTHC and DCM timers as well as the 16 interrupt levels circuit 108, ITC bus interface 134, INREG 136 and timer read buffer 142.

ITC system 100 serves two general functions. The first function is to control 16 interrupt levels, and the other function is to handle 15 timer circuits. The interrupt level function comprises circuits to enable, disable, activate, deactivate and request interrupt on 16 different levels in the system.

More particularly, resource allocator 40 comprises an RA matrix 104. RA matrix 104 is a switching matrix that connects 16 of the 72 lines 105 which come from various interrupt sources (1) ADC/DCM, (2) DAM/DCM and (3) PAM. The 16 outputs from RA matrix 104 are labeled as indicated and provide the inputs to 16 interrupt levels circuit 108, which provides for the 16 interrupt levels. Multiuser analog/hybrid system 10 has approximately 72 sources of interrupt within the entire system. However, circuit 108 only handles 16 interrupts. Under program control which is later described, RA matrix 104 selects which of these 16 interrupts are important at a time for a particular time critical user (gorilla).

It will be understood that some of the sources on lines 105 are general purpose interrupts (GPI) indicated by lines 15a-f, FIG. 1. There are eight of these interrupts for each analog console of an analog computer 12a-f. There is also an amplifier overload from each console. Additionally, from DAM/DCM controller 24 and the ADC/DCM controller 22, FIG. 1, there are six inputs that are used as interrupt sources and applied by way of lines 105. Resource allocator 40 selects 16 of the foregoing sources to be applied to circuit 108. In this way, circuit 108 accepts the signal from an interrupt source that has been selected by RA matrix 104 and requests an interrupt from CPU 30 in accordance with the program in memory 32.

RA matrix 104 is programmable in a manner later to be described so that the user program in memory 32 includes instructions that define the matrix to select the 16 interrupt sources of interest.

Of the 72 lines of interrupt 105, there are also three interrupts originating within ITC system 100. As shown, one of these interrupts is applied from RTHC timer 110 by way of line 111 and then through line 111a to RTHC interrupt output circuit 117 and then to RA matrix 104.

All of the RTHC timers and DCM timers are loaded under control of the program in memory 32. The timers each have a 20-bit binary initial value register or buffer 114, 116. The user program includes an instruction which starts a selected timer and, at the same time, determines the timer mode of operation. Each of the RTHC timers 110 has three control bits included in the instruction with the 20-bit initial count value. The first bit determines whether the counter is running in a single-cycle or multicycle mode. The second bit determines the source used to down-count the timer. The third bit determines whether or not the counter starts operation upon the load of the initial value or stays in hold mode until a subsequent instruction is issued.

The difference between multicycle and single-cycle mode is illustrated as follows. During loading of the initial value register 114, the initial value is automatically transferred to the timer down-counting register 118. When register 118 starts counting, it decrements the number until the count is all zero. Zero is the terminal count, and then a pulse is applied indicating that the end of count interval has occurred. If it is in a single-cycle mode, timer 110 stops at this point since it has gone through a single timing cycle.

If the control bit for multicycle mode has been selected at the end of the interval, buffer 114 is effective to reload the initial value to down-counting register 118 and to allow that register to continue down-counting. Timer 110 cycles continuously until a specific program instruction is issued to either reload it again or to place it in the hold mode.

The source of decrementing pulses will now be discussed. The source of clock pulses is selected by clock source multiplexer 120, which is effective to select either of two inputs. One of the block sources is marked 1 MHz. A 20-bit counter decrementing at a 1 MHz rate therefore yields a maximum time interval of somewhat more than one second.

The other source is from the low external clock, which is generated by an analog computer in conventional manner. Accordingly, a user may select an external clock, and he can apply a pulse or signal to down-count it at a maximum rate of 1/2 MHz all the way down to occasional pulses.

Furthermore, RTHC timer 110 may be used to load precise time intervals from the program in memory 32, which determines the time when a carry pulse 111 may be generated. The foregoing may be used as a source of interrupt if it is selected by RA matrix 104. In this manner, a great amount of flexibility is produced in the hybrid program. In addition to loading RTHC timer 110 from the program, a real-time hybrid clock may also be loaded externally from a timer panel 140 which allows setting of a 20-bit value and also the selection of three control bits. Option 1 establishes whether the timer should start immediately or wait until it is started from a program instruction. Option 2 determines whether it is running multicycle or single-cycle. Option 3 decides whether the clock source is the external clock from the analog computer or is using an internal 1 MHz clock.

As previously described, there are 12 DCM timers in ITC 26, with six assigned to ADC and six assigned to DAM. These DCM timers are only loadable through the user's program and only run in multicycle mode. The clock source to a DCM timer, as for example DCM timer (ADC) 112, is a 1 MHz internal clock. A digital computer instruction starts the timer from the user's program. The only way to turn DCM timer 112 off once it is started is by means of a signal from a predetermined DCM, which resets the operate flip-flop 124. Operate flip-flop 124 has its S input connected to DCM timer control 126. A pulse at this S input provides at the Q output a one state which allows a timer down-counting register 128 to begin counting. It will be understood that user program instructions initially activate the S input by way of DCM timer control 126.

As previously described, input R of operate flip-flop 124 is coupled through a buffer 120 to an ADC, since in this example timer 112 is ADC. The nomenclature of the signal indicates that it is an ADC timer 0, and the signal is clear operate. There would be a total of 12 such signals of this type coming through resource allocator 40.

Register 128 reaches its end-of-interval count and outputs a pulse on line 123 as indicated. That pulse passes through inverter 132 and is then applied to the ADC/DCM 22 and produces the same action as a run signal on an analog computer patch panel in conventional manner. In this manner, this signal indicates thart the ADC/DCM 22 is to process a block of data.

When all the blocks have been run, ADC/DCM 22 FIG. 1 sends a pulse back and operates the operate flip-flop 124 to an off state by way of line 125. Typically, when the programmer desires to generate a block of data, DCM timer 112 is set, and at the end of the timed interval the signal on line 123 is sent out to ADC/DCM 22 as HADCRUNTIMO. In this manner, a block of data is then generated. This pulse generates a new block of data every time it is produced until logic on ADC/DCM 22 indicates all the data blocks are finished. The reset signal on line 125 is produced to turn timer 112 off by way of the R input to operate flip-flop 124.

It will be understood that timer initial value buffer 116 is identical to buffer 114, with register 128 being the same as register 118.

It will be understood that only the RTHC timers may all be turned off at once. DCM timers may all be turned on simultaneously, but not off simultaneously.

ITC bus interface block 134 includes a microsequencer and clocks. This block recognizes when CPU 30 is attempting to make a bus transfer to ITC system 100. Accordingly, it controls the data accepted into register 136 and coordinates communication between CPU system bus 34 and system 100.

Drivers and receivers 138 are drivers and receivers accessed from timer panel 140. RTHC interrupt output circuit 117 generates a 1-microsecond pulse to RA matrix 104 whenever RTHC timer 110 reaches the end of an interval. The 1-microsecond pulse serves as the interrupt request to the interrupt circuit.

B. ITC Microsequencer 134 Firmware Flowchart

The following is a description of the firmware flowchart for microsequencer 134. It will be understood that the microsequencer functions in a manner similar to a small computer, since it has a stored program in memory (PROM).

In the symbology of the flowcharts a diamond represents a branch instruction. A box represents order instructions, and an oval signifies a connector. A two-digit number to the left of a block represents a hexadecimal number and corresponds to the address location in microsequencer or PROM 134 of the specified instruction. The hexadecimal numbers enclosed in parenthesis are for reference and are mainly found on another sheet.

Starting with the flowchart in FIGS. 6A-B, for example, entering from a system reset would start at oval 160 and would then go to memory location 0 at box 162. A NOP instruction is then executed.

If the system reset is pressed on computer 30, a pulse is transmitted to all of the controllers 14 in system 10. This is the general sheet function so that all the parts of the system start at a known state. At reset, microsequencer 134 goes to address memory location 00, pulls out the instruction, decodes the instruction, and finds that it is an NOP (no operation). This is a special case of an order instruction where neither of the pulses are generated and no level is changed.

Microsequencer 134 then goes to the next sequential location, which is 01 as shown in box 162. The next instruction is then pulled out and it is an NOP again. The next location 02 is then incremented in box 164, which is an order instruction where the P represents pulse. The pulse is a signal L micro special clear. This signal will clear certain level orders and it is an initializing type of function. Next are two FOP's followed by an executive loop as shown.

Most of the time, microsequencer 134, if it were examined with a logic analyzer, would be sequencing between these three branch instructions, viz, memory locations 05, 06, 07. From 07, the executive loop goes back to 05.

In location 05, as shown in diamond 166, the decision is based on the query: is the signal L special WDOT here? It is deciding whether or not CPU 30 has made a transfer out of ITC 100 to load a timer. If that signal is low, it is going to branch. For purposes of terminology, the L exits if that signal is low. On the other hand, the H exits if the signal is high. Within diamond 166, the B signifies that microprocessor 134 is doing a branch, and the I indicates that is is just incrementing to the next consecutive memory location.

Assuming that CPU 30 has not transferred in, the signal will be high, which means that there is to be an incrementing down to location 06, diamond 168. At that diamond the question asked at 06 is whether the timer panel 140 is trying to load a timer. If not, it goes down to the next diamond 170 and tests the Signal Low Always Branch. This signal is always held low so that it is an unconditional branch back to location 05.

In the executive loop, the microsequencer 134 is waiting for CPU 30 to transfer some data into system 100 or waiting for panel 140 to actuate a timer load. When either of these events occurs, branch will occur in microsequencer 134. For example, when an ITC instruction in CPU 30 is executed, the first branch at location 05, diamond 166 is made to location OC, diamond 172.

At diamond 172 a decision is made as to whether it is a timer instruction or an instruction to program a resource allocator matrix 104. If it is a timer instruction, the branch exit is taken to location 1A, diamond 174, and the data from CPU 30 is held in ITC system 100 so that the kind of timer instruction may be determined.

Microsequencer 134 basic decision making process is illustrated here. When a condition causes a branch, the branch is made to a subroutine. For example, in loading a timer at address 1A, a load timer instruction is found as the condition. The branch to SR2, oval 176, is taken and that location is 56.

All other instructions are similar. The type of instruction is determined, and then branches to a subroutine are made to increment a given instruction. Diamond 175 indicates start timer, diamond 184 indicates hold timer, diamond 186 designates read timer, and diamond 187 indicates decrement timer where the timer is any one of the RTHC timers 110 or DCM timers 112.

The following are the instructions for loading the timer. The first instruction is oval 202 FIG. 7B. Data is stable in input register 136, and the strobes of the input register are turned off due to the fact that a special WDOT command from CPU 30 has come in or the input-register-busy flip-flop is set. In this case, it is the special WDOT flip-flop that has been set. The listed instructions are executed, and finally the branch diamond 204 cycles back to start oval 206, and the timer is loaded.

In loading the timer from the TCP, oval 200 is entered when loading the timer from the timer panel 140. The switches in this panel that place a 20-bit interval value into an RTHC timer 110 have been set. There are three other control switches. One is for selecting the external clock or the internal clock, one is to specify whether it is to be a multicycle or single-cycle mode of operation, and the third determines whether the timer starts upon loading or whether the timer should just be loaded. This does not apply to a DCM timer 112. The instructions accessed from oval 200 are activated when the load switch on timer panel 140 is pushed. Microsequencer 134 then serially shifts in all the bit information, both initial value and control bits. These bits are brought in serially and loaded into IN register 136. Once register 136 is loaded, then the timer is loaded just as it is under program control. At the end of SR 1, the branch in diamond 208 goes to SR 2 oval block 210 FIG. 7A.

Diamond 212 in SR 2 FIG. 7B decides whether the timer should run immediately. It can go into operate or it can stay in hold. Depending upon this decision, it performs the instructions in box 214 to start. This determines whether it is supposed to start, or bypass box 214, or whether it is supposed to stay in hold.

SR 3 oval 216 accesses a subroutine to allow the program to start all fifteen timers (shown in FIG. 1 as timers 44, 46) or any number of these timers at the same instant. In the subroutine, the timers may be placed into operation if a bit in the program instruction is set.

SR 4 oval 218 FIG. 7D is a hold-timers instruction and is applicable only to the RTHC timers 110. This is a type of inverse instruction of the start-timers SR 3. All three RTHC timers may be placed in hold at the same time or in any combination.

SR 5 oval block 260 FIG. 8 is a read-timer instruction. For example, if a timer was loaded, the programmer might want to read it back and make sure it was loaded. Further, if the timer was down-counting, the programmer might want to read the timer to see what the present count is. The programmer would implement this program instruction in CPU 30 to accomplish this. This is really a two-part instruction. The programmer executes two different instructions in CPU 30 to read a timer. The first instruction goes to system 100 tells which timer is to be read, causes the respective timer to be read, takes the present count, and places it in timer-read buffer 142. Once this has been accomplished, it can be following with a second instruction of the read sequence, which goes out and just transfers the data in that read buffer back into the program. In order to read the timer, the programmer has to provide a series of pulses to get it programmed correctly, that is, to actually get the data out of it and to get it loaded into the timer-read buffer.

The following instructions are all related to resource allocator matrix 104. SR 7 oval 300 FIG. 9 is a program resource allocator instruction, and this subroutine allows the program to select 1 of the 72 interrupt source inputs and connect it to one of the 16 outputs. To fully program the resource allocator, 16 of these instructions are required to be executed to connect all 16 outputs.

The code in box 302, ending in a branch instruction in diamond 304 asking the question: Done?, is taking the data and shifting it serially into an RA matrix 104. Once this data has been shifted out at box 306, there is a signal high load matrix RAM. That signal is asserted and turns off RA 104, which has been operating. Then a strobe is generated to write the data into Ra 104 and then the RA logic is turned back on again at hold load matrix RAM.

SR 0 oval 178 FIG. 6C is a default instruction. If something happens in microsequencer 134, it branches somewhere it should not because of a logic malfunction. It should branch to this subroutine indicated at oval 178 and set an error bit in box 180. A technician can then connect an oscilloscope to a test socket to see if it is set.

The following additional information relates to the instructions listed in the blocks of the ITC microsequencer 134 firmware flowcharts.

C. ITC Instruction Format for Microsequencer 134

Microsequencer 134 has two different instructions. One is the order-type instruction in which it can output one of a number of pulses, or it can output another pulse, or it can change the level of a signal from a high to low or a low to a high. The other type of instruction is a branch instruction which branches on condition true or on condition false to some other part of the microsequencer memory. Each memory location in the microsequencer PROM then contains instructions which are either branch instructions or order instructions.

The following are the two different types of format:

A. Block Diagram Description

Basic bis interface (BBI) 21 is used in the parallel analog multiplexer 20. In addition, a modified form of BBI 21 is used in ADC/DCM 22 and DAM/DCM 24. As shown in FIGS. 10A-C, BBI 21 interfaces with CPU system bus 34, and the signal nomenclature is that used with SEL 3200 series computer. BBI 21 also interfaces, for example, with the devices of PAM 20, and these devices are defined as device-dependent logic (DDL). The details of this nomenclature will be described later in detail.

The purpose of BBI 21 is to simplify the interfacing with CPU system bus 34 to take care of the protocol. In addition, the interfacing of the bus itself provides the user with signals to allow transfers between memory 32 and also allow other inputs/outputs without having to operate within the structure of the timing of bus 34. Microsequencer 340 is very similar to microsequencer 134 of ITC 26 except that the flowchart is different since it has been programmed for a different purpose. However, the same concepts are used; for example there are two instructions, viz, a branch instruction and an order instruction. The flowchart for the BBI also has an executive loop. The executive loop of microsequencer 134 checks to see whether or not a CPU transfer has occurred.

Simpler than microsequencer 340 is the memory access sequencer (MAS) 341, which handles the control signals to perform memory read and memory write transfers. MAS 341 is normally inactive until microsequencer 340 recognizes the DDL logic request as, for example, a memory write. The DDL logic places the address on the destination bus lines and puts the data on the data lines. The DDL logic then asserts a requesting memory write service, and microsequencer 340 recognizes that signal and invokes MAS 341. Microsequencer 340 then accepts the memory address and the memory data and activates some of the lines of CPU bus 34, thereby giving access to the bus. This transfers information in the address to the memory bus controller of CPU 30, which then takes over and does the writing into memory 32.

Memory read works in a similar fashion. The DDL user, in order to access data from memory, has to supply memory address called a destination bus, LDEST 364. Then a signal is asserted requesting some memory read service from microsequencer 340. Microsequencer 340 recognizes that and, in each case, sends back to the DDL logic a signal recognizing the request. It then invokes the MAS 341, which gives the address to the memory bus controller of CPU 30. A few cycles later the CPU memory bus controller sends the required data back to BBI 21, and the BBI then asserts a signal to the DDL saying that the data is present.

For memory read and memory write, there are three modes of operation, viz, fast mode, delay mode and slow mode. In the fast mode, for memory write the DDL has a block of data, and it makes this data available to BBI 21 fast enough so that the BBI operates at its maximum rate in transferring data to memory 32. In the slow mode, the DDL sends one word at a time, and the BBI 21 runs at something less than full rate. The delay mode is somewhere in between where the DDL logic can have its data available faster than the slow mode but not quite as fast as the fast mode.

One of the other services provided by the BBI is the interrupt service in which the DDL can require a service interrupt. The DDL senses a signal which is an interrupt request, and microsequencer 340 recognizes it and returns a signal which is a request acknowledge. BBI 21 then goes through the procedure of generating an interrupt request to CPU 30.

Another service of BBI 21 allows the programmer to send information out of the BBI. In this case, CPU 30 executes a command device instruction or a test device instruction. The CPU interprets and codes these instructions and outputs data. There are several different command device instructions and several different test device instructions. Whenever one of these transfers is made to BBI 21, the BBI (depending on which type of transfer it is) asserts a signal to the DDL, notifying the DDL that there is some kind of data from CPU 30. The DDL accepts it, and microsequencer 340 is then done.

Lines 342 are 32-bit lines designated LDLFD, and the user puts data on this bus primarily when requested to by BBI 21. An OR gate 344 ORs this data with data from a status block 346 which provides status bits. The output of gate 344 is ORed by gate 348 with physical address data on lines 350. The output of gate 348 is applied to one input of an AND gate 352, the output of which is applied to bus 34. In this manner, information may be driven on to bus 34, or 32 bits may be taken off the bus and loaded into a data bus input register 356. Register 356 is always loaded.

Every 150 nanoseconds, the input register 356 is loaded from bus 34 whether the data on the bus is for BBI 21 or not. When the data is for BBI 21, coincident with that transfer another signal to BBI 21 indicates the data should be saved. The signal is high in lines 358 at that time. The output of register 356 becomes the signal LTLTD on lines 360, and that is the DDL source of data from the CPU.

At different times there is different information on bus 34. At times there is an address destined for counter 362. BBI 21 can save data which is an address in register 362, and then the output lines 363 may be driven to the next bus 364a by way of a gate 365. Another bus driver or AND gate 366 drives bus 34 by way of bus LDT 368. Another register 370 is fed from the output lines 360 of register 356. Register 370 is called a holding register, which holds several bits specifying the type of instruction on bus 34. The destination bus 368 from bus 34 is also a bidirectional bus. Bus 368 has 19 lines, some of which are coupled to a physical address recognition and generator circuit 372. Other lines go to interrupt level save register 374. The remaining three lines from bus 368 drive the last accepted subaddress register 376 and last retried self-addressed register 378. Selection of which of these registers is used is determined by decoded control bits of the instruction, which places the address on bus 368.

A service interrupt poll 380 has applied to it inputs from an interrupt level save register 374 and from microsequencer 340. Poll 380 has two output lines 382 to bus 34 called LIPOL and LINTR, as well as two input lines 384. Block 380 is primarily concerned with DDL request for service interrupt. This logic at the request of microsequencer 340 generates service interrupt requests to CPU 30.

A tag bit receiver 386 drives a tag register input and decoder 388. Receivers 386 and 388 are effective to decode bits to determine what kind of CPU transfers are being made to BBI 21. Output encode 390 operates so that when BBI 21 is performing a transfer to either memory 32 or back to CPU 30, it configures tag bits correctly.

Response to CPU 392 is the logic that recognizes that the physical address of BBI 21 is on destination bus 368. It says that CPU 30 is trying to transfer to BBI 21, and the decision must be made as to whether the transfer may be accepted, because BBI 21 is not busy, or whether to reject this CPU transfer, because BBI is at a state in which it cannot accept the input transfer.

Bus poll 394 is used every time BBI 21 is required to transfer data onto bus 34, to make sure that no other controller is transferring at the same time. This circuit determines whether a higher priority controller wishes to use the bus and if not, it seizes the bus, and no controller of a lower priority can use the bus. Memory-busy 396 attempts to ensure that memory 32 is not going to be busy the next time it is accessed. This block uses the principle that the last memory module accessed is going to be the next memory module accessed; that is, the same one is accessed. AND gate 398 provides an output LDTF 398a, which supplies a bit that is used to determined what kind of transfer is being made. Namely, whether it is a word transfer, a half-word transfer, or a byte transfer. Clock 400 accepts the master clock signal from bus 34 and generates four or five different clocking signals for use on BBI 21. Initialize 402 accepts the reset signal from CPU 30 which occurs when the system reset button is pressed on the front panel of CPU 30. Initialize 402 generates a general reset pulse to the logic on BBI 21 including microsequencer 340 which forces the microsequencer back to address location 0.

Lines 347 from microsequencer 340 have the following designations:

HCDPNDG

HTCWA

HCDBITS

HIOADR

HALTREQ

LUSTOP

HTCWDATA

DHATAREQ

HDINTACK

HACTIV

HHENBLD

HQUED

HRQPNG

Lines 346a to status block 346 have the following designations

LPROGVIO

LCNTLRACTIVE

LTDBCC1

LTD4CC4

LTD2CC1

LTD2CC3

LTD2CC4

Also from status block 346, lines 346b have the following designations

LDLFD00

LDLFD02

LDLFD03

LDLFD16

LDLFD(28:31)

Further, response to CPU 392, lines 392b have the following designations

LUCLRCDPF

LUCLRHALTIO

LUACK

LURDY

HSELO

HSEL1

B. BBI Microsequencer 340 Firmware Flowchart

An executive loop FIG. 11A includes diamonds 428-434 labeled respectively MWT, MRT, DINT and CDH. Initial oval 420 indicates the path that microsequencer 340 would take when a system reset button is pushed on the console. It initializes the microsequencer to location 00 in box 421, and the microsequencer begins to sequence. Several NOP's are executed and then there is a branch to start block 423. Intializing is then performed with checks to see if the half IOF flip-flop is set as in diamond 424. If this flip-flop is not set, there is a check to see if CPU data is in CDH diamond 426 with a jump to diamond 428. These four diamonds: 428, 430, 432 and 434 constitute what is called the executive loop.

When BBI 21 is inactive, microsequencer 340 continues sequencing through the executive loop, and the BBI serves the device-dependent logic (DDL) such as PAM logic or DCM logic. The first diamond MWT 428 is memory write transfer, and it is deciding whether or not the DDL user is requesting that the data be written to memory. MRT diamond 430 is deciding whether the DDL wants to read some data from memory 32. The next diamond DINT 432 is deciding whether the DDL wants interrupt service. Finally CDH diamond 434 is deciding whether or not CPU 30 needs some kind of data transfer. The executive loop is continuously scanning these four items.

For example, CPU 30 has decided to transfer some data. When microsequencer 340 goes to diamond 434, the CDH signal would not be 0, but would be a 1, so that a branch would occur down. The code that it would execute would be the ARSTX diamond 436. CPU 30 has labeled some of the bus transfers with various names such ARSTX transfer, RSTX but transfer, WDOT bus transfer, AICT bus transfer and ICT bus transfer. All of these are SEL identification. A transfer to BBI 21 is indicated because of a branch down. It is now required to determine what type of transfer is coming from bus 34.

In the event that it was an ARSTX transfer, the sequencer does a branch out to location OE shown as oval 438. There are four different reasons why CPU 30 should make an ARSTX or advanced read status transfer. One is to gain status information on what is called the 8,000/4,000 level. Another is to obtain status information on the 2,000 level, where these levels are SEL identification. A third reason is that there is a CPU instruction called transfer current word address TCWA in diamond 472. A fourth reason is that there is an RSTX transfer called acknowledge interrupt which occurs as a result of BBI 21 performing a system interrupt.

From oval 440, FIGS. 11B-C information is sent to CPU 30, by way of boxes 442, 444, indicating that the transfer was accepted and that an RSTX transfer is now expected. Accordingly, the box indicating pulse LDTO 446 starts a timer. Diamond CDH 448 is executed to check whether the CPU is now transferring the RSTX transfer. If not, the branch to the right to diamond time out 450 is taken to check whether the timer is timed out. If not, the program waits between blocks 448 and 450. It will do that twelve times to give CPU 30 some time to send the RSTX transfer.

At the end of the time out, if CPU 30 has not transferred a new transfer, the CPU has probably been interrupted by some other system on bus 34, and this entire sequence is repeated at some later time. To abort this loop, the branch to the right to oval 452 jumps back to start, and the executive loop is entered again to wait for the next transfer. In the event the next transfer occurs, the branch out to CDH diamond 448 to RSTX diamond 454 is taken which asks the question: is it the executed RSTX transfer? If not, the program jumps to block 456.

If it was not an RSTX transfer, then following block 458, the decision diamonds starting with diamond 436 are executed to find the nature of the transfer. If it was assumed there was an RSTX. then there is a branch out of RSTX 454 to the right to four decision diamonds which test to find out which one of the RSTX transfers it is. If it was an acknowledge interrupt tested by diamond 460, the branch down is taken, and pulses from the microsequencer are sent to the interrupt circuit on BBI 21. The last box executed by diamond 462 is an acknowledge back with a transfer to CPU 30 saying that everything is taken care of. The final jump in jump 464 returns execution to start, which enters the executive loop and waits for the next transfer.

If the RSTX was not an acknowledge interrupt, the next question is whether is was a status check at the 8,000/4,000 level in diamond 466. If it was the branch down, a signal is sent to DDL at box 468 which says the CPU 30 wants status information at 8,000/4,000. A fixed amount of time is given to the DDL to collect that information and give it to the BBI. The flow then continues to do a bus transfer of that information back to CPU 30. After this is finished, the program goes back to start through jump 464.

If it was not that status level check the program goes to the right to the next diamond 470 for a 2,000 level status decision. The 2,000 level status is almost identical to the 8,000/4,000 level status code except that a different signal is sent to the DDL, indicating that it is a 2,000 level status that is to be sent back. If such is in the CPU, then there is a jump back to start through jump 464.

The last type of RSTX check is called transfer current word address block 472. It will be assumed for purposes of explanation that it has been found out that there was a CPU transfer. The manner in which the CPU activates the ARSTX/RSTX type transfer has been described. On the other hand, it may be assumed that the transfer was not of that type and that it was a WDOT transfer. This transfer is checked in the executive loop at WDOT diamong 474. From there there is a branch out to the WDOT code as shown in oval 476.

In the WDOT transfer, FIG. 12A there are several different ones that can come through bus 34. These transfers usually indicate that the programmer has placed in a dedicated memory location in memory 32 certain information that is to be transferred out to BBI 21. This information is not directly transferred to BBI 21, but CPU 30 does send a WDOT out. Part of the information in WDOT is the address where this information is in memory 32. It is then the responsibility of BBI 21 to actually go into memory 32 with this address, read out the data, and find out what it should do with the data. The only other WDOT is called HALTIO or terminate transfer. This is used to inform the card that if it is in the process of writing to memory it is to stop writing to memory.

From WDOT block 478 there is a branch diamond 480. If there is not a start IO, the branch is to the right, and it must be an HALTIO instruction that is tested at diamond 482. If it is not an HALTIO it would branch to the right again and jump back to start at block 484.

A start IO instruction branches out the bottom exit of start IO diamond 480, FIGS. 12A-B. The BBI goes through a procedure, starting at box 486, reading from memory the data which actually comprises two consecutive memory locations called input/output command double word (IOCD). In the second part of the IOCD is the address which points to TCW memory location. BBI 21 is going to do three memory reads, pull out information, and then pass it on to DDL as shown in box 488.

Starting at box 486 and continuing through box 488, BBI 21 goes out to read a single memory location. Continuing through MER diamond 490, the TCW diamond 492 and diamond 494 are in a first pass. Three passes are to be made. In the first pass, the program continues to box 496, with an arrow pointing to DDL. BBI 21 has thus gone out to memory and read in IOCD word 0. In that word, there are 12 bits of information, placed in the command device instruction by the programmer, that he executed and that resulted in WDOT being sent out by CPU 30. BBI 21 sets signal CDBITS, which tells the DDL that this information is available and that the DDL should pick it up. The program then goes to diamond 498 DDL response and waits for the DDL to confirm that it has the data and that the program may proceed. Continuing down in the program, it is set to read the second word from memory 32. Accordingly, the program goes through diamond 498, box 500, diamond 502 and box 504. From box 504 there is a branch all the way back to box 488 again. The steps required to read the next piece of information in memory 32 are from box 488 down to diamond 490. BBI 21 picks up that second word which has an address to the tCW. The TCW has been loaded by the programmer before he actually executed the command device instruction. The programmer inserts some specific information that he wishes to pass to the BBI.

Diamond 494 indicates that the program is now on the second pass, or that IOCD 1 has just been read, and that the answer is yes. The program branches to the left down to box 506 and reloads the address counter for this new address, which is the address of the TCW. There is a branch up to box 488 again to perform a third memory read.

In this manner, all the characters of the TCW are being read out, and the program is down to diamond 492. In a branch to the right, diamond 510 indicates that there are several different types of transfers. Diamond 510 shows a single word transfer, with a branch to the right to box 512 to set a signal for the DDL which says the TCW is available and that the program is to go to diamond 514.

There are two single-type transfers. In one of these, the programmer wants to output 32 bits of data to the DDL, and in the other the end result is that the DDL is to take 32 bits of data and have the BBI transfer it down to TCW. If it is assumed that there is an output of 32 bits to the DDL, when diamond 516 (write diamond) is reached, there is a branch out of the bottom indicating the completion of 32 bits of information. If it is desired that 32 bits of data should be read from the DDL and written down to the TCW, the instruction would branch to the right from diamond 516. A signal would go to the DDL saying that 32 bits of data are wanted and would indicate what data was available.

The other variation of this instruction is shown in diamond 510. If it was not a single-type transfer, the branch would go out the bottom. A signal would be sent to the DDL called IO address in box 520. This indicates the address of a block of information in memory, and the DDL would then read that information out of memory.

The other WDOT is HALTIO. This is used in a situation where, for example, an address is given and the DDL is reading or writing some information in memory that the programmer decides he wants to halt. There is a command device instruction called terminate IO (or halt IO) that tells the DDL to stop the IO. In that case, starting at WDOT, diamond 478, the program goes to start IO branch at diamond 480. It is not a start IO so there is a branch to the right over to diamond 482, which is a HALTIO instruction. BBI 21 sends a signal to the DDL called "halt request for stopping transfers" and then waits to receive a response. A transfer is sent back to CPU 30 saying that the BBI is set to stop but hasn't stopped yet. In a state between diamonds 522 and 524 the time-out feature started, and there is a wait for 12 time cycles for the CPU to send another WDOT transfer. A pulse is then sent to the DDL stating that it is the time to stop, and then there is a jump back to start.

Interrupt flow sequencers are shown on FIG. 13B as a result of the DDL requesting interrupt service by asserting a signal to BBI 21 called DINT. Another manner in which to obtain the flow is to use certain programmed instructions available that will activate, deactivate, enable, disable, and request particular service interrupts by a program instruction. None of the interrupts will operate until they are enabled. One of the program interrupt instructions is to enable interrupts. The program interrupt instructions always result in an AICT/ICT bus transfer as defined by SEL.

Returning to the executive loop, CDH would be set because CPU 30 would make a transfer. If it is determined that an AICT has occurred, there is a branch down to send a signal back called "ready" to CPU 30. This signal states that there is a transfer, and the CPU is set for the next transfer, which is probably the ICT. In box 562, the timeout timer is again started and in diamond 564, there is a toggle between this diamond and diamond 566, waiting for the CPU to transfer ICT in.

When ICT does come in, there is a branch to location DO in oval 568, and then there is a determination as to which one of the ICT's it is. There are five different types of ICT's and the one in question is to enable interrupts. Further, there is a branch out of diamond 570 over to box 572 to set the bit called "enabled". Some checking is performed to determine the state of the previous interrupts, and then there is flow through circle 574, to oval 576 FIG. 13A, to diamond 578 to send the transfer back to CPU 30, indicating that the instructions have been executed. There are four other interrupt control instructions. These are decoded in similar fashion by way of ICT diamond 580.

The other way to get into the interrupt circuit is at the request of DDL logic which wishes to generate a service interrupt request to CPU 30. In the executive loop, diamond 432 shows that when the DINT signal is asserted by the DDL, it is going to cause a branch to the interrupt circuit at location 06. The branch goes down to box 582 FIG. 13B, which acknowledges to the DDL. BBI 21 is thereby telling the DDL that it has the request for interrupt and it will be taken care of. Thus, the DDL can go about its operation.

Once the BBI 21 acknowledges the DINT signal, it does a jump at circle 584 to oval 586. This flow is doing a request for interrupt out to CPU 30. Diamond 588 FIG. 13A indicates that if this level has been enabled, everything is fine and there is a branch down. If it has not been enabled, there is a branch to the right to box 590, and the interrupt would be pulled.

In diamond 588, it is determined whether the level has been enabled. If it has, an interrupt can be pulled if the level is nonactive. If the level is active, Q flip-flop output is set so that there is a branch around. If it is not active, there is a branch to box 592 to actually request an interrupt to CPU 30. Diamond 594 shows a DINT request from the DDL, which is considered an external interrupt request. There is a branch back to start, in contrast with the program interrupt request instruction, which would cause a branch down to send a transfer back to CPU 30 indicating that the instruction was completed, with a jump back to start.

The following is a description of when the DDL wants to write something or read something from memory. This flow is gotten into from the executive loop. There are two signals from the DDL, viz, MWT and MRT. FIG. 14.

If a memory write transfer is required, there is a branch through box 620, and at box 622 the BBI pulses the signal .mu. MWT. At that point, memory access sequencer 341 logic is activated, and it assumes it has been loaded with an address. This address accesses the bus to memory and either requests a read or a write memory. In the meantime, microsequencer 340 goes to diamond 624 and waits for MAS 341 to go back into idle. As long as MAS 341 is turning away reading and writing to memory, it is not in the idle state, so there is a wait in the loop for it to go back to idle. Once MAS 341 goes back to idle, it branches down to diamond 626. If nothing unusual occurs during the memory write it would just branch back to start. In the event of an unusual occurrence, there is a branch or jump from block 628 to block 630. Unusual conditions are defined as those that arise when the write is progressing and the CPU 30 tries to make a transfer to BBI 21. Another unusual condition would be where the CPU tries to halt the BBI so that HALTIO flip-flop is set.

If it was a memory transfer, the convention is to request an interrupt. If not, then was it the HALTIO flip-flop? If it was, there is a branch into the HALTIO flow. If it was neither of these, it must have been the fact that CPU 30 was trying to do a transfer to BBI 21. Accordingly, a signal is set to the DDL at box 632. At diamond 634, there is a wait for the DDL to acknowledge the message and to signify its readiness to proceed. At box 636, the time-out timer is again started, and there is a toggle between diamond 638 and diamond 640, waiting for CPU 30 to retry the transfer that it had tried before. When it comes, it will branch back and decode to see what kind of CPU transfer it was.

The following additional instruction relates to the instructions listed in the blocks of the BBI microsequencer 340 firmwave flowcharts.

C. BBI Memory Access Sequencer 341--State Diagram

In FIG. 15, the circles represent states in the logic, and the directed line having an arrowhead indicates a transition from one state to another.

Starting at state 6, circle 660, a poll bus is shown. In this state, MAS 341 is polling bus 34 to determine whether or not any user has a high priority. At the bottom of the circle, an arrow goes to the top of this same circle, and the signal associated with that is a lost poll. Any time that signal is true, MAS 341 stays in the same state. It is also to be noted that lost poll is underlined and that there are some signals listed below. None of these signals occur when it branches back and it stays in that same location.

When a lost signal becomes false, the state is changed to state 4, circle 662, which is the transfer state. It will be understood that the sequencer is in the idle state of circle 664. The only way the sequencer 341 is activated is when microsequencer 340 pulses the .mu.MRT signal or .mu.MWT signal. Once this is accomplished, there is a transition out of the idle state down to the memory-busy state 7, circle 666. There is a check to determine whether the memory bus controller last contacted is busy at the moment. If it is busy, the sequencer remains in state 7 until a signal inhibit becomes false. Once the signal inhibit becomes false, the MAS drops into state 6, circle 660, in which the bus is checked to see that no user with a higher priority is going to transfer. Once this bus priority is one, there is a drop to state 4, circle 662, to actually transfer the address and data to bus 34 in the case of memory write. The address in memory will be picked up by the memory bus controller.

From this point, there may be a branch in two directions. In the case of the memory write, there is a branch to the right to state 0, circle 668. There are four ways to transition out of this state. If another consecutive write is desired, there is a branch out the D exit to state 7, circle 666, to do another memory write. On the other hand, it may be found that the memory bus controller was actually busy, so it sent a signal to try again, so that there is now a branch out the B exit to circle 666. Alternatively, it may be that the address put on bus 34 did not exist.

In the event of completion, there is a transition out of Exit C to go back to state 2, circle 670, which is the delay state and transition back to the idle state. In the event that there is a memory read, as opposed to memory write, the action is the same down to state 4, circle 662. However, there is a transition to the left to state 3, circle 672. In this state, there is a check to determine whether or not the transfer to memory was successful. If it was a successful transfer, there is a transition out exit C to state 5, circle 674. In the event that there was a nonpresent memory error, there is an exit to idle circle 664 to set a flip-flop that can be read later on. In the event that memory 32 was busy, there is a transition out of exit B to circle 666 to try again.

In the event of a successful transfer, there is a transition to state 5, circle 674, to wait for the memory bus controller to look up information and send it back. The MAS remains in state 5 until memory data is transferred back in, and then there is a transition to state 1, circle 676. In the event that a sequencer read is being performed and there is not difficulty, there is a branch out exit B to go back to state 7, circle 666, to continue reading another piece of information. In a memory read, parity error may be invoked, which is a memory error. In that event, there is a transition out exit A back to the idle state, circle 664, setting a parity error flip-flop. This would cause an interrupt to occur, and the program would be able to come back and find out what had gone wrong.

In the event that the CPU 30 is trying to make transfers, there is a transition out exit B going back to the idle state in circle 664. In the event of completion, there is a transition out of exit C to the delay state, circle 670 and then down to the idle state.

The following is a description of the terms used in the state diagram of FIG. 15 of memory access sequencer 341.

DELAY--A DDL controlled signal that will allow the MAS to enter a delay state at the completion of an access rather than return to the idle state should the DDL have to drop MRT or MWT due to slow transfer capability. The MAS will wait in the delay state for another MRT or MWT. This feature is used when the next MRT or MWT is expected shortly since micro seq. is tied up waiting for the MAS to return to the idle state.

.mu.MWT--Pulse command from .mu. sequencer to initiate a memory write.

TA--Transfer acknowledge, from memory

US--Unsuccessful--from memory. Says memory was busy and didn't accept your transfer

.mu.MRT--Pulse command from .mu.sequencer to begin a memory read transfer

INH--Inhibit--High if the last MBC accessed by the card is currently busy.

LOST POLL--says you tried but failed to become bus master.

READ--Tells which command pulse (.mu.MRT or .mu.MWT) activated this sequence.

MEMTRANSIN--Active near the end of bus cycle when a memory DRT is directed to this card.

MRT--DDL request for mem. read transfer

MWT--DDL request for mem. write transfer.

PE--Parity error--sent by MBC

SAVECHO--Save MBC echo bits. These bits identify the MBC that contains the mem. loc. just accessed.

CDP--CPU data transfer pending--says that a CPU transfer attempt has been made to this card. CDP=HALTIOF+CDPF

MDH--Memory data here--to DDL--signifies requested data is available and that mem. ADR may be changed

REQACK--to DDL--says that the MWT or MRT request has been recognized. DDL may lower the request.

CWT--Completed write transfer, to DDL--says that data write has been completed successfully.

TRANSOUT--Pulse that drives info to SEL bus during the transfer cycle

XFR--The state where this controller is the source of the info on the SEL bus

POLL--A pulse that causes the controller to actively compete for bus mastership by polling.

NPM--Nonpresent memory

RESET--Systemwide reset

MASK--(Not shown) A DDL controlled signal that can be used to disable the CDP signal.

In the state diagram FIG. 15, the following are notes to the drawing.

1. Read FF is set by .mu.MWT.

2. The asterisk denotes the logical negation of the variable and not to the signal level.

3. The signals between the MAS and the DDL are normally disabled. They will be enabled by the micro sequencer when it recognizes that the memory access was requested by the DDL. The signals are again disabled when the MAS returns to the idle state. These signals are:

The truth table of the ROM sequencer in memory access sequencer 341 is as follows:

D. BBI--DDL Signal Descriptions

The following is a listing and explanation of BBI--DDL signal nomenclature used herein and shown in the drawings. The signals are grouped by function.

Memory Access

LMRT--Request for memory read

LMWT--Request for memory write

LMASK2--Maintain block integrity

LDELAY--Maintain MAS control for med. speed transfers

LMDH--Memory data here

LCWT--Completed write transfer

LREQACK--Request acknowledged

LCOMPARE--Control rejection of CPU transfers

LFBIT--Define mode of transfer (BYTE, 1/2 word, etc.)

HMER--NPM or PE memory error occurred

Handshaking Sequences

HCDBITS--16 bits from CD instruction here

HIOADR--1O address and some data here

LWRITE--Direction of memory transfers

HTCWDATA--Data from TCW here

HDATAREQ--DDL supplies data for TCW location

HCDPNDG--DDL must get ready for a CPU input transfer

HTCWA--DDL supplies data for TCW location

HALTREQ--DDL should prepare to halt

LUSTOP--DDL should halt

HRSTX8/4--DDL supplies status

HRSTX2--DDL supplies status

LDDLRSP--DDL's response line

Service Interrupts

LDINT--Request for interrupt

HDINTACK--Acknowledgement for LDINT

HACTIV--Interrupt level active

HENBLD--Interrupt level enabled

HRQPNDG--Interrupt request to CPU pending

HQUED--Interrupt request queued

HRQACK--Leading edge indicates intr. svc. routine about to begin

Status Transfers

LTD8CC1

LPROGVIO (LTD8CC2)

LTD4CC4

LTD2CC1

LTD2CC3

LTD2CC4

LCNTLRACTIVE--Default status control bit

Buses

LDLFD (00:31)--32 bit data bus from DDL

LDLTD (00:31)--32 bit data bus to DDL

LDEST (00:23)--32 bit addr. bus from DDL

HDT (20:23)--Bus level device sub-addr

HSAR (0H:3H)--Device sub-addr of last accepted CPU transfer

HSAR (0P:3P)--Device sub-addr of last retried CPU transfer

HCDBITS--A "Here Is" handshaking signal, says that the rightmost 16 bits of the CD instruction are on the data bus to the DDL, LDLTD (16:31). DDL responds with LDDLRSP signal.

HCDPNDG--A"Get Ready" handshaking signal. Tells DDL to prepare itself for an input transfer. When ready the DDL responds with LDDLRSP.

LCOMPARE--The compare signal is used by the DDL during DDL initiated memory access sequences. The signal controls the rejection of CPU input transfers to the controller once the BBI has accepted the request for memory accesses.

LCWT--"Completed Write Transfer" is generated by the memory access sequencer (MAS) as a result of a DDL initiated memory write sequence.

LDEST (00:23)--"Destination Bus", a 24 bit tri-state address bus from DDL. DDL places address of memory location on this bus during MWT sequences. DDL should use the LREQACK to enable its tri-state bus drivers.

LDINT--"DDL Interrupt" is a "Will You" handshaking signal. Used to request a service interrupt (requests execution of an interrupt service routine).

HDINTACK--"DDL Interrupt Acknowledge" is generated by the BBI when it has recognized the DDL's request for interrupt (LDINT).

LDLFD (00:31)--"Data Lines from DDL". The 32 bit tri-state data bus from the DDL. DDL passes all information other than addresses on this bus.

LDLTD (00:31)--"Data Lines to DDL" 32 bit data bus to device. All info (data or address) from the BBI will pass over this bus during "Here Is" sequences. The DDL may use the LDDLRSP line to store the info from the bus.

HDT (20:23) (To DDL) "Data Lines" These four lines come directly from the SEL bus (through receivers). These lines can be used with the LCOMPARE signal and the last accept sub-address lines [HSAR(0H:3H)] to dynamically reject CPU input transfers.

HALTREQ--"Halt Request" is a "Get Ready" signal that warns the DDL that it should get ready to halt its I/O operation or some similar system-defined function. The DDL respond when ready with LDDLRSP.

LFBIT--Used in conjunction with the two least significant bits of the destination bus (LDEST22 & LDEST23) to specify the data format mode of memory accesses.

HIOADR--"I/O Address" is a "Here Is" signal and signifies that an address is right-hand justified on the data lines to DDL. The DDL should take the address and initiate a memory access in the direction specified by the LWRITE signal.

LMASK--"Mask" is a control line that the DDL may assert during memory access sequences to allow the sequence to go to completion without being interrupted by any event except a memory error.

LMDH--"Memory Data Here" is generated by the MAS during memory read sequence. It tells the DDL that the current read has been completed successfully and that the memory data is available.

LMRT--"Memory Read Transfer Request" Asserted by DDL when it requires a memory read sequence. This is a "Will You" handshaking signal.

LMWT--"Memory Write Transfer Request" A "Will You" handshaking signal used to request memory write transfer sequence. DDL must supply data.

HMER--"Memory Error". Signal set when the BBI detects a memory error as the result of the BBI accessing memory.

LPROGVIO--"Program Violation" is a status bit that is under DDL control and should be supplied to the BBI.

LREQACK--"Request Acknowledged" is generated by the MAS during MRT and MWT memory access sequences.

HRSTX8/4--"Read Status Transfer, 8000/4000 Level". This is a "Give Me" signal requesting 8000 and 4000 level status from the controller. The DDL may place information in these user defined bits:

LTD8CC1

LTD4CC3

LTD4CC4

HRSTX2--"Read Status Transfer, 2000 Level" This is a "Give Me" signal requesting 2000 level status from the controller. The DDL may place user defined status information in the following bits:

TD2CC1

TD2CC3

TD2CC4

HTCWDATA--(To DDL) "TCW Data" is a "Here Is" handshaking signal telling the DDL that 32 bits of data (from the TCW location) are available.

LWRITE--(To DDL) The "Write" line may be checked by DDL when the "Here Is" HIOADR sequence is in progress.

A. General Description

As previously described with respect to FIG. 4, the sequence of events to transfer words is to first load one of six transfer control registers 62a-f from memory locations in memory 32. TCR 62a-f hold six double words which describe the proper transfer to take place. For example, the first and last channel of the ADC 52 is described. These words also describe the data array address of CPU 30 where these words are to be transferred and the type of transfer (whether loop, chain, jam, transfer on end of block, hold, scaled fraction or floating point) and whether it is to or from memory.

After loading TCRs, these words wait in the locations until an activate request is received. This activate request can come from CPU 30, from the run patch panel hole of analog computers 12a-f, or the carry output from timers 64a-f. All of these activate requests are brought together, and any one of these requests can activate TCR 62a-f and actually start the transfers by way of CPCW active block 60.

The main function of ADC/DCM 22 is to decide which of the six TCRs 62a-f is next to get access to ADC 52. A transfer control register (TCR) may be labeled time-critical or non-time-critical. Assume that the priorities are worked out so that number five TCR 62e is presently active. Accordingly, when a transfer is completed, the next priority to come up would then be TCR 62f followed by 62a, 62b, 62c and 62d.

If any gorilla wants to come in at any time while the monkeys are active, then the gorilla takes priority and gains control. When one of the TCRs 62a-f is preempted by either a CD pending from CPU 30 or another transfer request, then that updated TCR is stored back into the save active CPCW register 66. TCR 62a-f and register 66 comprise a single active pending RAM, described below, which is a 16-word 32-bit wide-programmable RAM.

It will be understood that FIG. 4 represents ADC/DCM DAM/DCM 24 is similar in circuitry except that the data flow is in reverse. Accordingly, only the ADC/DCM has been described in detail.

A portion of ADC channel 52, FIG. 1 is shown in FIG. 4 and comprises a floating point conversion block 67. Accordingly, information flows from ADC 68 as a scaled fraction, and if the TCR desires the information to be changed into floating point format, then it goes through a floating point conversion before going into memory 32.

B. Block Diagram Description

In FIGS. 16A-B, the active pending register is referred to in its entirety as register 750, and a portion of the register 751 stores loop, float, jam, chain, and hold. Register portion 752 stores memory address, portion 754 stores word count, portion 756 stores upper limit address, and portion 758 stores lower limit address. RAM 750 may be loaded from CPU 30 by way of multiplexer blocks 760, 762, and 764. Specifically, data from CPU 30 is transferred from a modified BBI 21 by way of line 766 and then loaded into the active pending register during a CPCW load sequence.

When a particular TCR 62a-f wins the priority poll of the round robin later to be described, then that particular TCR is transferred to the active register 60, which comprises three counters 768, 770 and 772. The memory address from active pending register 752 is transferred into memory address counter 768 of active register 60. In turn, all of the counters of active register 60 are loaded. As soon as this information is in the active register, the ADC/DCM 22 recognizes that it has to start the transfer.

Accordingly, the first word will be transferred from ADC 68 into memory 32; word counter 770 will decrement by 1, and memory address counter 768 will increment by 1. In addition, the current address counter 772 will also increment by 1. In this manner, the block of transfers continues until word counter 770 reaches 0. At that time, the active TCR is completed, and there will be no further transfers to be performed on that TCR.

However, before the end of the foregoing transfers, another transfer request may come from either a gorilla or a monkey. During the rest period an updated memory address from counter 768, word counter 770, and current address counter 772 will be loaded back into the active pending registers 752, 754, and 758 respectively. Thus, the word that was active and is now saved has to wait in the pending registers until it again wins priority and comes out to the active register.

The two methods of priority allow the gorilla to always have priority over the monkey. The gorilla is transferred into active register 60, and unless another gorilla requests a transfer, the gorilla will go to the end of its block (block transfer) without being interrupted. At the end of the block, during the time a gorilla is running, a window is provided for another computer command to come in or another transfer to take place.

TCW register 774 holds address 1-6 of a particular command which is trying to talk to DCM 22 from CPU 30. For instance, if CPU 30 wants to hold the transfer, it indicates which of the six TCR's 62a-f it wishes to hold. This information is loaded into a TCW register 774 and is decoded to hold the particular TCR. CPCW address register 776 has the same memory address counter used for the transfer, and it is also used for bringing out six CPCW double words to load TCR 62a-f. Specifically, register 776 comprises a RAM that is loaded with memory address where the CPCW's are resident and that allows DCM 22 to access CPU memory 32 and bring out the CPCW's. In block 778, bits on line 778a are used for the CPU 30 to read the status of DCM 22. In this way, all of the TCR's may be read, as well as the current word count, current first and last address, whether the DCM is busy and whether a particular TCR is busy or not. Block 778 shows the manner in which status is read back to CPU 30. In DCM 22 there is a signal called equal or end of block. On line 782 block 780 compares the upper limit address to the current address counter so that if channel 0-6 is required to be transferred, then when a current address counter reaches 6, it recognizes that the end of the block has been reached, and an equal signal on line 782 is produced. This is used throughout DCM 22 to signify an end of block.

System 784 (included in TCR activate control 72, FIG. 4) indicates a gorilla round robin and a monkey round robin which are effective to decide the priority of TCR's 62a-f. System 784 FIGS. 17A-B is programmed with information which allows the gorilla to have top priority and the monkey second priority within TCR's 62a-f, if the first TCR 62a is active. Then the circuit allows TCR 62b to have the next priority, 62c the next priority, and so on. If TCR 62f is presently active, then the next priority is TCR 62a, etc. The foregoing is shown as feeding into address select box 786. This block takes information from the priority decoder in system 784 and instructs the rest of the circuit that the presently active address lines 788 indicate whether a monkey is active or a gorilla is next on line 790 and shows the address of APR RAM address line 792.

It will be understood that one gorilla may have more than one TCR allocated to it. With a gorilla having more than one TCR, then its respective TCR's become time critical when the gorilla becomes time critical. It is these TCR's which would operate in a round robin fashion in gorilla round robin 830.

Active register 794 takes the state from ADC channels 51, FIG. 1, and assigns the timing of loading the active registers. Further, it decides the timing of unloading the active register 60 back to the active pending register 750. Active to APR line 796 indicates that its TCR has been preempted, that it is going back into APR 750, and that it again has to wait to win priority. APR to active line 798 indicates that a new TCR is now going into the active register 60 and when it will start transferring to or from the ADC 22 or DAM 24.

The inputs to register 794 are state 7, line 800, which comes from the channels 52. Active register loaded line 802 indicates that there is presently a TCR in the active register 60. Line 804 actually does the transfer, and it is an output from the modified BBI gorilla/monkey priority hardware. Enable DDL sequencer line 810 is from BBI firmware in microsequencer 340, with gorilla active line 812 coming from the same circuitry as the gorilla next line 808.

TCR loaded flip-flops 814 indicate that until a TCR is loaded, then even though there are signals from the patch panel of consoles 12a-f or from CPU 30 or from timers 44, 46, the TCR still cannot be activated. Wnen CPCW is loaded into active pending register 750, this is effective to set a TCR loaded flip-flop 814. Similarly when a transfer is completed, th word count is 0. There are no more transfers depending on that TCR, and that TCR loaded flip-flop 814 is reset.

Run control 816 shows the "OR" tie of six signals which can activate either a gorilla or a monkey, once signals are obtained from patch panel run holes of consoles 12a-f. Three of these six signals consist of the six inputs from the patch panel on line 818, six inputs from timer carry out signals on line 820, and six digital run signals from the CPU on line 822. The remaining three are described as follows: Six timer-allocate signals on line 824, which determine whether timer 64a-f is to start the activation or whether the patch panel is to start activation; six signals on line 826 which are fed into block 816 to inhibit run signals once the TCR is finished, and finally, the six-timer critical signals on line 828, which are fed through block 816 into gorilla round robin 830 and which allow the gorilla to take priority over the monkey.

Six-timer critical flip-flops 850 FIGS. 18A-B are set by command of CPU 30. There is a select timer run 852, which also allocates timer blocks 1-6. These are set by CPU 330 to select the timer to activate the TCR's rather than permitting the patch panel to activate the TCR's. Block 854 indicates the six patch panel run holes which are fed in to set the six JK flip-flops within block 854. These flip-flops are set when a pulse arrives from the patch panel, and they are reset when a transfer is complete. Block 856 is set when there is an APR to active signal and is reset when there is an active to APR signal. Block 858 indicates when a word count reaches 0. When the word count is finally decremented to 0, the TCR is finished and is used to indicate that the end of the transfer is for that particular TCR.

MRT, MWT and delay blocks 864, 866 and 870 are used for signals from DCM 22 to tell BBI 21 when it wants to read or write or delay.

Microsequencer 860 is part of microsequencer 340 and is used for DCM functions. The firmware for these functions is a modification and addition to standard BBI firmware shown in the BBI firmware flowchart. The signals on the left of block 860 indicate the outputs from the programmable ROM's of microsequencer 340. The additional inputs are shown on the right-hand side.

Time-out 862 provides for a set up of priorities so that if a gorilla is running from the patch panel, it has priority over all of the other TCR's 62a-f. Unless the signals in the patch panel come frequently, it could hold up all of the other transfers until a gorilla reaches its end of block. Time-out 862 makes sure that unless the pulse signals of the patch panel when a gorilla is running come in less than 100 microseconds apart priority is dropped and another CD pending, or monkey transfer, is allowed to come in.

The following setting instructions are provided for the respective circuits.

C. DCM Microsequencer 860 Firmware Flowchart

The following is a firmware flowchart for microsequencer 860. This firmware is in addition to the BBI microsequencer 340 firmware previously described.

Blocks 870-873 and 880 FIG. 19 are similar to blocks of the BBI firmware flowchart which arrive at write diamond 880, which provides for a transfer from CPU 30 to DCM 22. TCW data in jump 882 indicates that there is information on bus 34 which describes an instruction address to the DCM. Diamonds 884, 886, and 888 decode one of six instructions to the DCM. For example, if IOCD 29 is not set as indicated in diamond 884 (location 146) and IOCD 30 is not set as indicated in diamond 886 (location 147), then the program goes to diamond 888. A pulse is provided by the firmware to set the timer to allocate block. Thus, in block 852, the set timer allocate signal is applied as the clock pulse to set one of the six D flops. A clear timer allocate sends one pulse over to the D flops and sets the correct timer allocate as determined from the information in the TCW register 774.

Similarly, if IOCD 31 is set in block 888, then a set run pulse is provided by block 894 to another six blocks which allow the TCR to become activated.

The third command is to halt the TCR. All of the jumps 879, 881, 883, 885 provide a single pulse which is used to carry out the particular command. In the case of the set halt pulse in block 896, the TCR loaded flip-flops 814 are reset, and that essentially stops any more transfers on that particular TCR.

In locations 160-163, as shown in blocks 876-879, the time-critical flops are reset and that carries to block 850. These previous instructions are the six functions that the DCM performs under control of CPU 30, viz set timer allocate flops, set run flops, set halt flops, set time-critical flops. These are all single-word transfers from CPU memory 32 to DCM 25 to perform each of those six functions.

FIG. 20 shows a sequence which allows the DCM to access CPU memory 32 and store the CPCW double words into active pending registers 750. CPCW read is shown in circle 920, which comes from the BBI firmware. In reading from CPU to memory, it is first required that the CPU be told at which location this information is being stored. When the informaton comes from the CPU and is available in the BBI, it is loaded into TCR 62a-f.

Starting at block 922, the memory address counter 768 is loaded with address of the two CPCW's. In box 924, the destination bus is enabled, to allow the address to be sent to the CPU 30, and pulse acknowledge and MRT indicate that a read is requested.

There is a wait in idle diamond 926 to allow the memory access sequencer in CPU 30 to actually perform that transfer. When out of idle, the data of the first CPCW is available at the DCM. At that point, the address going to the CPU is a clear enable destination bus and check in block 928 for memory error. If there is no memory error, then the first bit of data is considered, which is the data line to device 00 to see whether this is a loop instruction or not.

Assuming that it is a loop, then CPCW 1 is loaded into the appropriate TCR register or active pending register 60. The next CPCW is then brought up. Since that word is a double word, the first word in CPCW 1 is loaded into active pending register 750. In block 930, the end memory address counter is incremented to send a new address to CPU memory 32. The same sequence is continued for block 930 and 932, waiting for the memory access sequencer in block 934 to reset the enable destination bus. Memory error is then checked in block 936, and CPCW 2 is loaded into the active pending register. In blocks 938-940, the memory access counter is incremented and stored in the CPCW RAM 776. In the event that the TCR is activated, a chain bit is set, and two more CPCW's are brought out for the next incremental address.

From diamond 952, a loop is taken, and to the left of this block a chain bit is shown. The purpose of the loop and chain is so that the DCM can continue doing an entire sequence of transfers without requiring to be loaded from CPU 30. If a chain bit is set when the TCR is initially loaded and if the word count reaches 0 in that TCR, then the chain bit is interrogated to see whether the DCM should bring another two CPCW's or whether it should just finish the TCR. Thus, without using any commands, the DCM may go on continually chaining through the entire memory, bringing in new CPCW's. Specifically, blocks 940, 944, and 946 show that the CPCW RAM 776 address which holds the address of the next pair of CPCW's is always incrementing, whether the chain bit is set or not. When the TCR is activated and reaches word count 0, a block 970 FIG. 21 indicated as disable DDL is provided. In block 972, chain active is interrogated, and if the chain bit is set when word count 0 has been reached, then block 948 is entered. Assuming that the chain bit is set, then blocks 948 and 950 FIG. 20 are reached and the program continues up to block 924, which again starts the accessing of two more CPCW's from the CPU memory 32.

In the loop function, the loop bit is interrogated when the first CPCW is brought in and before it is loaded into the TCR. The reason for this is that if the loop bit is set, the CPCW 1 has a different meaning than normal CPCW 1. The normal CPCW 1 bits 12-31 describe the data array address of the transfer. If the loop bit is set, bits 12-31 describe the new CPCW address.

The reason for the looping feature is that the first CPCW 1 and 2 are brought in and completed to the end of the word count and then the chain bit is set. Another CPCW pair is brought out, and if this pair has the loop bit set, then instead of working on the next pair, the program goes back to the first CPCW 1 and 2 and brings in that original pair. In this way, without another CPU command, the first CPCW 1 and 2 can be looped and looped repeatedly until halted by the CPU.

In diamond 952, if the data line to device 00 is set when CPCW 1 is available to the DCM, then there is a loop to block 906, 908 and 910. In this manner, CPCW address is loaded from the data line to devices 12-31, which are presently available to the DCM. This brings in the address of the next CPCW pair, and that goes through blocks 906, 908 and 910 and then back to oval 920, which is the normal address of the CPCW pair.

Block 970 FIG. 21 is entered when a transfer has been completed by either a gorilla or a monkey. This allows the firmware to check whether there are any commands CD pending or whether another monkey or gorilla has requested a transfer. Diamonds 974 and 976 check whether the ADC or DAM I/O card is in its rest state 5. At that point, the word count could be checked to see whether it has reached 0 or whether it is in an end of block, whether the chain is set, whether the loop is set, or whether there is a CD pending. Diamond 978 checks whether the count has equaled 0. If it has not, then there is no further checking for memory error or chain active, and the program goes straight to diamond 980. If there is an unusual signal, this means there is a CD pending; if there is a monkey running, the BBI firmware will service that CP pending. If there is a gorilla running and it is not the end of block, then the signal unusual will be inhibited until the end of block. A gorilla will not look to see whether the CD is pending.

Assuming that there is a CD pending and that either this is a monkey running or the gorilla has reached the end of block, then the program ges to circle 982, which is in the BBI flowchart. If there is not a CD pending, the program goes to oval 984. Diamond 1010 FIG. 22 checks whether there is a TCR requesting to run. If there is not, unusual is again checked, and the program stays in loop diamonds 1010, 1012. The reason this occurs is that if there is no CD pending, then there is no point in wasting time. The program just waits until a valid run signal occurs. If there is a CD pending, then unusual will branch from diamond 1012 and into the normal servicing of CD pending.

A valid run means that the TCR is loaded and activate signal high means that a particular TCR is requesting a transfer sequence. Accordingly, in diamond 1010, when a valid run signal occurs, there is a branch out of the left-hand to box 1016. Accordingly, the DDL sequencer is an enable, and there is a search through to see whether the DCM is requesting a memory write, memory read, or if it is a valid run or not. In this manner, an idle loop waiting for a memory write or memory read sequence is provided. If there is a transfer from the ADC, it would be a memory write. If there is a load of the DAM, it would be a memory read. MWT and MRT oval blocks, 1018 and 1020 respectively, go to the BBI flow chart.

Box 1022 FIG. 22 is reached from the BBI flowchart, and valid means that a TCR has been loaded and a run request has been received from CPU 30 or a timer or the patch panel. Diamond 1024 checks whether the CD pending has not sneaked through while reject mode 3 was being set. Assuming no CD pending is waiting, then the program goes to box 1016 to enable the DDL sequencer and wait for memory write or memory read request from DCM 25.

The BBI flowchart FIGS. 11A-C has been modified for DCM use as follows. Taking the start loop at oval 422 through blocks 423, 425, 424, 426, there is a jump to location 2D. Instead of jumping to this BBI location, DCM jumps from location 2D to 06, which is a device interrupt interrogator of the DCM. From location 06 to 07, there is a check to see whether there is a CD pending in diamond 434. If there is no CD pending, the program comes back to diamond 428, which checks for a valid run. The idle sequence of the BBI when it is used as a DCM therefore uses blocks 428, 432, 434 and then back to blocks 428, 432, 434. If there is a valid run, the program jumps to decision diamond 430, which is within the DCM flowchart.

The following additional information relates to the instructions listed in the blocks of the DCM control flowcharts FIGS. 19, 20, 21, 22.

D. State Diagram for Loading DAM's 24 Reading ADC 22 or DAM 24

The state diagram for loading DAM's is shown in FIG. 24, with state 5 being the idle state in oval 1040. I/O controller 16 waits in this idle state for a valid run signal. Valid run means that a TCR 62a-f has been loaded and the run flop has been activated in one of three methods. When a valid run has been received, the change to state 4, oval 1042, sets the delay which enables the transfers to progress as fast as they can. A mask is set if this is a gorilla, and this inhibits the CD pending from coming in until the end of the block. When count equals 0 the flip-flop is reset, and active-register-loaded flip-flop 856 is set. From state 4 in oval 1042 to zero, oval 1044 there is always a transfer, as there is no signal to wait for to go to state 0. This will decrement the word count by one, and update the last equals-current flip-flop.

From state 0, oval 1044, to state 1, oval 1046 there is always a transfer and this is the time when MRT 864 is activated to indicate to the BBI that memory read has been requested. State 1 waits for request acknowledge from the BBI, and as soon as the request acknowledge is received, the diagram goes to state 3, oval 1048. During that period, the MRT is disabled, and the destination bus is enabled with the address of the data array. There is a wait for memory data in oval 1048, with the memory data allowing the diagram to go from state 3 in oval 1048 to state 2 in oval 1050. During that time, the data buffer is loaded and the destination bus disabled.

In oval 1050, there is a wait for the load pulse to be completed. During that waiting time, a load pulse is applied to DAM 24, and there is a transfer pulse if the jam flop is set. As soon as the load pulse is completed, the diagram goes to state 6 in oval 1052, which activates the TCR if the counter equals zero and the chain is not set. The time-critical flop is reset under the same conditions.

During transition from state 6, oval 1052, to state 7, oval 1054 there is always a transfer, and during this time there is an increment in the current address counter if it is not the end of block. If it is the end of block, the run flop is reset in the normal mode. During this time, the patch panel run flop is also reset and the memory address is incremented. During transition from state 7, oval 1054, back to state 5, oval 1040, there is always a transfer. During this time, the active to APR signal is produced, and the mask is cleared if it is the end of block to thereby reset the delay . Back in state 5, the system waits in the idle state for another valid run.

In the state diagram for loading DAM's 24, FIG. 24 the following are notes to the ovals and between the ovals.

Referring now to FIG. 23, the state diagram for reading ADC 22 or DAM's 24 is shown. State 5, oval 1070, is in the idle condition waiting for a valid run, which means that a TCR is loaded and the run flop is activated. During this valid run, the delay is to be set and the mask is to be set if this is a gorila or time-critical request. In addition, the read-back signal to DAM file is activated, the count-equals-0 flop is reset and the active-register-loaded flop is set. Accordingly, state 4, oval 1072, is reached. During transition from oval 1072 to state 0, oval 1074, there is always a transfer, and during that time the word count is decremented and the last equals-current flip-flop is updated.

During transition from state 0, oval 1074, to state 1, oval 1076, there is always a transfer, and in ADC system a convert pulse is sent to ADC 22. This is a wait in state 1, oval 1076, during the time the ADC 22 is busy. At the end of the convert pulse, the ADC goes not busy, and there is a transition from state 1, oval 1076, to state 3, oval 1078. During this time, ADC data is loaded to the buffer or DAM information is loaded to the buffer.

During transition from state 3, oval 1078, to state 2, oval 1080, there is always a transfer. During this time, the MWT 866 is set. The system remains in state 2 waiting for a request acknowledge from the BBI. When that acknowledge arrives, there is a transition from state 2, oval 1080, to state 6, oval 1082. During that period, the destination and data buses are enabled and the TCR ended is activated. In addition, the time-critical flop is reset if the count equals 0 and the chain bit is not set.

In state 6, there is a wait for CWT. When CWT arrives, there is a transition from state 6 to state 7, oval 1084. During that period, the destination and data buses are disabled, the current address counter is incremented if not the end of block, the run flop is reset if end of block (normal mode), and the run flop is reset if the count equals 0 and the chain bit is not set. In addition, the patch panel run flop is reset, and the memory address incremented. During transition from state 7, oval 1084, back to state 5, oval 1070, there is always a transfer. During that period the mask is cleared, if end of block, and the delay is reset.

In the state diagram for reading ADC 22 or DAM's 22, FIG. 23, the following are notes to the ovals and between the ovals.

A. Block Diagram Description

Parallel analog multiplexer (PAM) 20, FIG. 25A, B is an interface system and operates to talk to one analog computer 12a-f at a time. PAM 20 is coupled between BBI 21 and the analog computer.

Data transfer from BBI 21 is in 32-bit words which are stored in registers in PAM 20. Specifically, these words are stored in instruction register 1112, console select register 1114, and data register 1116. The CD word in a signal called CD bits is applied by way of line 1100, which indicates that BBI 21 has CD data available. The least significant 12 bits of the CD word contain the address of the device in the analog computer 12a-f which is being communicated with.

The TCW word is also stored in PAM 20, along with a TCW data signal on line 1102 that informs PAM 20 that the TCW word is available. Bits 1-6 of this word are the console select bits for analog computers 12a-f. Bits 16-31 are data bits shown together as line 1104. Each command, such as CD bits or TC data or data request coming from BBI 21, is required to be answered with a SP DDLR signal on line 1106, which is a device dependent logic response. This answering is performed by an echo command circuit 1108.

CD data is decoded by instruction decoder 1110. There are four types of instructions, and they are analogous to instructions of the Pacer computer of Electronics Associates, Inc. These four types of instructions are data out, device function, status in, and data in. These instructions start the control timing circuit 1118. Circuit 1118 operates the signals going to the analog computers and also operates the echo circuit to wait for signals coming back from the analog computer before issuing a response. The signals leaving control timing 1118 are address enable 1120, strobe enable 1122, and data enable 1124. The first signal to become active is data enable 1124, which allows data lines to be activated going out to a selected analog console 12a-f. The next signal is address enable 1120, which allows the address to go out. The last signal is strobe enable 1122, which allows the PAM 20 to issue a strobe to a selected analog console. The strobe is the first to go inactive, followed by the address and data lines going inactive. The information of which console 12a-f is selected is stored in console select register 1114.

Address decode 1126 has eight lines coming from the eight least significant bits of the CD word. These least significant bits relate to eight lines coupled to ADR decode 1126 from instruction register 1112, and also relate to the selection of control lines or mode control in the analog computer.

In conventional manner, an analog computer develops the signal called "device ready" on slow instructions. Therefore, there is a signal path from address decode line 1128 to a time-out block 1130 for function relay, time-scale changes and mode changes. Accordingly, there is a time-out for the length of time that it would take the analog computer to perform this function and then give a device-ready signal to interrupt storage 1132. Device-ready signals come directly from the analog console by way of line 1134 and are applied to device-ready interrupt storage 1132.

The console select register also operates a multiplexer 1136, 1138 to receive signals from the analog computers to select which of the analog computers is going to be talking back to BBI 21. Multiplexer 1138 has 16 lines applied to it from the three analog consoles 12a-c. For consoles 11d-f, an expansion multiplexer 1136 is provided, and the connections are made in similar member.

The device-ready and the overload interrupt also come from an analog console and are multiplexed together in block 1140, with the data lines feeding the BBI 21 on lines DLFD (data lines from device) 1142. The device-ready and the overload interrupt are also stored in flip-flops and are sent back to BBI 21. The interrupts are maskable by an interrupt mask register 1144, which masks the overload interrupt. There is one bit in register 1144 for each console.

From consoles 12a-c, it may be seen that there are signals DLO-15 on lines 1146, 1148 and 1150, these data lines being bidirectional. If a control line is to be set, a data line is assigned to a specific control line, and that data line would be in a 1 state to set that particular control line. Similarly, if sense lines are being read, each data line would be assigned a sense line and, if that line should be in a 1 state, the sense line would be set. These incoming data lines go to multiplexer 1138.

There are further functions that the analog computer will perform for data out, device function, data in, status in, as indicated on lines 1152, 1154 and 1156. Additional steering-register-bit and link-bit-1 lines 1158, 1160 and 1162 relate to address information. When the steering register bit is through, this is a signal to the analog console itself.

Lines 1164, 1166 and 1168 are binary encoded steering lines to select one of 16 devices. Further, there is a device-ready signal and overload signal on lines 1170, 1172 and 1174. These perform interrupt functions to BBI 21.

Control timer 1118 has 16 states. In the normal operation there is a short instruction which uses four of these states to transmit information to the analog console. When data is to be read back from the analog console it must also be synchronized with BBI 21, so there is a wait state. The control timer 1118 counts from 0-4 and then waits for BBI 21 to request the data by raising line 1176. The control timer then continues to advance.

CD bit signal on line 1100, TCW data signal on line 1102 and data request signal on line 1176 each require a response back to BBI 21, to inform the BBI that PAM 20 has acknowledged that command. PAM 20 receives these signals and stores them as long as the PAM is not busy performing another function. Then these signals are applied to a shift register in echo or wait-for-read circuit 1108. Activating any one of these signals generates a strobe pulse which applies the data into a data register in block 1108. The ORing of any of these circuits sets a flip-flop that sends a DDLRSP signal on line 1106 back to BBI 21, acknowledging command. Thereafter,, BBI 21 lowers the active signal. A DDLRSP is not issued until the wait period is over, and then such a signal is issued and BBI 21 reads the data from PAM 20.

A. ITC 26

1. RTHC and DCM Timer Controls 121, 126

The circuitry of both timer controls 121, 126 is similar, thus only one of them, RTHC timer control 121 will be described in detail. The basic circuitry is a timer pulse distributor 1466, FIG. 26, in which four signals are produced by any one of the 15 timers of timer system 1460, FIG. 29, formed by an 8253 LSI circuit (Intel), with each of the timers being 16 bits wide. Three of the four signals are used in loading timer 1460, and one is used in reading the timer. Accordingly, the timer pulse distributor generates four pulses as required as a result of signals coming from microsequencer 134 at the proper time. In addition to these four pulses, there is an address specifying which one of the timers of system 1460 is the correct pulse to go to. The distributor then transmits the pulse to the correct address and the correct timer, with the addresses having the designations LDLTD 00, 01, 02, 04. It will be understood that in distributor 1466, only one set of four circuits, 1461-1464, has been shown and that two more sets of four are used with similar connections to the first set.

FIG. 27 shows a timer address comparator 1470 and a chip select decoder 1472. These circuits take the address as it comes from the program in memory 32. By the address it is meant that the programmer is attempting to specifically address one of the timers of system 1460 to do something to the timer such as reading and loading it, etc. Comparator 1470 converts from one format of code into another compatible with timer 1460.

RTHC jam carry latch 1473 is used when loading a real time hybrid clock, and the upper 16 bits at the interval value are 0, so that the bit is latched.

FIGS. 28A,B shows an operate/hold flip-flop 1491 and a subcircuit defined as an operate and decrement sync register 1490. These circuits operate primarily on the inputs to timer 110 in which the clock signals are decremented. The function of sync register 1490 is to take the synchronous clock signals and synchronize them up to the clock used in the timer circuits. Specifically, flip-flops 1491-1494 operate as on and off switches with respect to the timer 1460. On the other hand, flip-flops 1495-7 are called operate/hold flip-flops. Each one of these flip-flops corresponds to one of the timers in the real time hybrid clock timer. If the flip-flop is clear, then the circuit is thought of as being in the hold mode in which the timer is not running to turn off. The associated gating allows the programmer to come with an instruction that will either start the timer or stop the timer. In effect, the programmer has the ability to throw a switch and to turn the timer on or off.

There is a further mode of operation in which once the timer is started and its down-count reaches a 0 count, it will reload itself with the initial value and continue counting. This is called multicycle mode. This is in contrast to the single-cycle mode, in which the programmer would load a value up to the timer, start the timer or put in an operate. Once the timer down-counts and reaches 0, the timer would turn itself off. The operate flip-flops 1492-1494 operate the sync register 1490.

2. Timer Initial Value Buffers 114, 116 and Registers 118 and 128

Fifteen timer system 1460, FIG. 29, operates as a clock source multiplexer. This circuit selects whether the decrementing clock or the real-time hybrid timer comes from the 1 MHz. internal clock or whether it comes externally from a patch panel hold on one of the analog computer consoles 12a-f. As previously described, 15 timers of system 1460 are provided by an 8253 LSI, and that circuit has initial value buffering in addition to control circuitry which provides the timing for timer 110.

FIG. 30 shows one of the 15 RTHC timer registers 1445. Since all of the timer registers are identical, only one of them is shown in detail. The ITC timer length is 20 bits. Since 15 timer system 1460 constitutes a 16-bit timer, register 1445 has a four-bit initial value buffer and a four-bit counter. In this manner, timer 1460 provides an effective 20-bit timer. Specifically, circuit 1440 is an initial value buffer, and circuit 1430 is a down-counting register for the least significant four bits of this timer 1445. When down-counting register 1430 reaches 0, it generates a carry to timer 1452. When that timer 1452 reaches 0, it carries over to timer 1460. When both timers have carried, this signifies that the entire 20-bit count value has reached 0 and it is time to generate a carry pulse. The remaining gates 1444, 1448 and 1450 are used to generate various carry pulses out of timer 1445. In this manner, the drivers allow the reading of the current value of the time down-counting register on program control.

It will be understood that, if you desire to read the current value of one of the 15 timers of system 1460, there is a read construction the programmer can execute that will cause microsequencer 134 to go with the specified timer, read out the current 20-bit value, and cause it to be loaded into a predetermined register. Once this is accomplished, the programmer can execute another instruction that will cause the predetermined register to be emptied back to the program.

3. CPU Bus Interface

Further, CPU bus interface circuits are provided for bus 34. These circuits function as a result of CPU 30 attempting to transfer information into ITC 26. Such interface circuits respond back to CPU 30 when an appropriate signal says that the ITC is busy at a particular moment and it cannot accept an input transfer from CPU 30. In the event that the ITC is not busy and it does accept the input transfer, the data is clocked off bus 34 into register 136. At the same time, a signal is sent back to CPU 30 indicating that the transfer has been accepted and microsequencer 134 should be signaled that there is a new instruction in input register 136 that should be processed.

A subfunction of this circuit is to recognize the physical address of the ITC so that a particular CPU transfer may be recognized and targeted. This circuit decodes that type of transfer and provides signals to microsequencer 134, so that it knows what kind of instruction it is and how it should go about processing it. These circuits are identical to SEL RTOM circuits as defined with respect to circuit 108.

4. Microsequencer 134

Microsequencer 134 is the heart of the congtrol of ITC 26. The microsequencer operates as a small computer having two instructions: a branch instruction and an order instruction. An order instruction generates a change of level signal and at the same time generates at most two pulses as selected by the instruction.

A branch instruction causes microsequencer 134 to branch away from its sequential flow to the firmware to some other point in the firmware. A particular branch instruction specifies which one of 24 test inputs should be tested. The input is tested and then, if the input is true, a branch occurs to another part of the firmware. If the input is false, then the microsequencer simply increments to the next sequential instruction in the firmware.

FIG. 31 shows a control ROM and program counter having a counter circuit 1571. It will be understood that two other identical counter circuits are provided and identically connected to the lines between counter circuit 1571 and PROM chip 1570. Similarly, three other identical PROM chips are provided and similarly connected. The microsequencer firmware is stored in the four PROM chips.

FIG. 32 shows a test structure comprising four select circuits 1590-3 which select which one of 24 inputs is to be tested for a branch instruction. Each of circuits 1590-3 is a multiplexer that selects one of its eight input lines and connects the selected line to an output of the respective circuit.

FIG. 33 is part of the pulse order structure of microsequencer 134. If a particular instruction is an order instruction specifying a pulse, one of the decoders 1604-6 produces an output pulse at one of its eight output lines. The outputs of decoder 1604 are defined as pulse group "Y", while the outputs of decoder 1604-6 are defined as pulse group "X". Given an order instruction, one pulse may be applied in "Y" group and one pulse may come out of the 16 lines of the "X" group.

FIG. 34 shows level order structure for microsequencer 134. This structure provides 24 output lines from latches 1584-6. From an order instruction previously described, it is possible to specify which one of the foregoing 24 outputs should access or change. In addition, these latches specify in which way the output should be changed, either in a set state or a cleared state.

5. Resource Allocator 104

There are two modes of operation: the running mode and the loading mode. When resource allocator 104 is in the running mode, 20 MHz clock 1674 drives binary counter 1678, which cycles continuously. When in the loading mode, the output of counter 1678 continuously cycles RAM's 1670-1. The output of these RAM's, which is a function of the particular address input, provides the addressing for the 1 of 72 multiplexers 1654.

One of 72 multiplexers 1654 comprises a group of multiplexers and obtains its inputs from analog computer consoles 12a-f. As shown on FIG. 35, these inputs are applied to a bank of eight low-order secondary multiplexers 1650, the outputs of which are applied to a primary multiplexer 1652. Multiplexer 1652 is connected to high/low-order multiplexer 1655, which is also connected to high-order multiplexer 1656. The combination of these multiplexers provides the 1 of 72 multiplexers 1654.

When 1 of 72 multiplexers 1654 has selected its particular input from this addressing data, the output that corresponds to that address will appear at its output. This signal will be clocked into the serial shift register 1690 by way of line LDECIN.

Shift register 1690, FIG. 37, is effective to shift in the data from the multiplexer by way of the output of the same binary counter 1678 that addressed RAM's 1670-1. In this respect, the foregoing is synchronous. It is synchronized with the RAM addressing of the multiplexer. The foregoing continues throughout the entire 15 cycles of binary counter 1678. At the end of this cycle, shift register 1690 will be loaded with 16 data bits derived from 16 different addressings of the 1 of 72 multiplexers 1654. At this point, binary counter 1678 is decoded, so that the particular point in time when shift register 1690 is full is known. Accordingly, at this time, the decoded output creates a pulse which transfers the contents of shift register 1690 through an interrupt transfer blocking gate 1692 into an interrupt holding register 1694.

The purpose of blocking gates 1692 is to correct a problem involved in the resource allocator 104. Specifically, if one of the inputs into multiplexer 1654 was active for a period of time that would allow two cycles of the binary counter 1678, the counter would cycle twice so that this active input would have been looked at twice. Thus, the output to interrupt holding register 1694 would show an active pulse that would be twice as long as it should be. Blocking gates 1692 obtains feedback from holding register 1694. If it determines that one of its output lines is already active, then it will not allow as input the next active data. Due to timing, the divide-by-16 binary counter 1678 addresses RAM's 1670-1 at approximately two data locations ahead of the actual data that is shifting into shift register 1690. Solving this timing problem requires decoding of the output of binary counter 1678, to recognize when the 16-bit shift register 1690 is correctly full.

The following is the operation of the loading of RAM's 1670-1 (FIGS. 36A-B). When it is desired to load the RAM's, an active level is brought in on line high load matrix RAM. This is effective to turn off clock 1675, which has been driving binary counter 1678. It also clears this binary counter. This in effect shuts down the cycling of the entire resource allocator 104. This loading active level that turned off the clock also drives tristate gates 1674 to redirect the addressing of RAM's 1670-1 from counter 1678 to the RAM's loading register 1660-1. These registers have been loaded at some previous time independent of the predescribed operation. With registers 1660-1 loaded and ready, the data therein is effective to address locations in the RAM's. Another part of the data in the loading register 1660-1 will actually be loaded into that location in the RAM's. In this manner, a loading of 1 of 16 data locations of the RAM is provided. When the high load matrix RAM line drops back to an inactive level, the clock is again enabled, and binary counter 1678 starts cycling again. Accordingly, resource allocator 104 restarts its operation, as previously described, with new data loaded into the particular location as provided by the RAM floating register 1660-1.

The following table shows the instruction for programming RAM's 1670-1 from CPU 30 together with the codes that select which one of the 72 inputs will be connected to 16 output lines.

B. BBI 21

1. Physical Address Recognition and Generator 372

Jumpers 372a, FIG. 41, are set up by the user to reflect a unique address of BBI 21. Comparators 1800-1 are effective to compare the jumper configuration with the address bits coming from CPU bus 34. When a comparison shows that the address on bus 34 is the same as the address on jumper 372a, a signal called HIGH TRANS IN is generated. This signal is the first indication that BBI 21 has some kind of bus transfer existing.

Tri-state gates 1805 take the configuration of jumper 372a and drive it back into bus 34 during memory read transfer, as compared with memory write transfer. When BBI 21 is requesting data from memory 32, the BBI has to supply the memory with a physical address so that at some later time memory 32 may come back and readdress the BBI and send the data back. This is the generation part of the circuit. A further portion of the circuit is a memory-busy having an S151 device (not shown) that determines which memory module has accessed last. The assumption is that there will be the same module access the next time to select an inhibit line from the memory module that allows monitoring of the busy state of the memory. The next time memory is accessed, the memory inhibit line is checked, and if the memory is inhibited, there is a delay in the request to memory until the line becomes noninhibited when the memory is not busy. This memory-busy 396 may be the busy circuit of the IOM card of System Engineering Laboratories, SEL Publication 325-32-9000-000.

2. Response to CPU System 392

There are three responses that BBI 21 can return to CPU 30 or to the device originating the transfer to BBI 21. The BBI may not be busy, in which case a transfer may be accepted and, in so doing, data on bus 34 would be captured. Accordingly, a signal is then returned called "transfer acknowledge" which indicates that BBI 21 has accepted the transfer from bus 34. BBI 21 may be temporarily busy, in which case an indication is sent back to the originator of the transfer by means of a signal called "retry". The retry signal informs the transmitting system that the BBI is going to be busy for at least a short while and that the transmitting system should come back and try the transfer again shortly. A third response is that BBI is going to be busy for an extended period of time and the transmitting system should cease transmitting and not retransmit for an extended period of time. In that case, BBI 21 would send back a signal called "channel busy". The foregoing is accomplished by circuit 392, FIG. 38.

Only in the case of the acceptance of a transfer does circuit 392 generate a required signal to clock into input register 1712 that data is out on bus 34. Also, at this time, a signal would be generated called "CPU data here," which tells miscrosequencer 340 that this input transfer in the input register has begun to process the instruction. There are three responses: transfer acknowledge, retry and channel-busy, which are required to be sent out one clock time after the attempt to transfer. Register 1712 is a type of circuit that delays the signals one clock time.

Input transfers come from two sources. Unsolicited transfers come from CPU 30 by way of the bus 34. The other type of transfer is a response from the memory-busy controller when BBI 21 has asked the controller to read some information from memory 32 and return it. This circuit also recognizes the memory transfer in type of transfer.

3. Tag Register and Decode 388 and Tag Output Encode 390

The tag bits are SEL terminology for a set of bus lines that are used and are active during bus transfers. When BBI 21 receives a bus input transfer, it decodes the tag bit line to determine what type of transfer it is. When BBI 21 has to make a memory transfer, it generates the tag bits so that other circuitry can determine what type of transfer is being made. Accordingly, register 388 constitutes an S138 decoder. Encoder 390 comprises a series of OR gates to provide the encoding. The tag bit receivers 386 are shown in FIG. 38.

4. Memory Access Sequencer (MAS) 341

As previously described, BBI 21 is used as a foundation or building block for PAM 20 and for DCM 25 as used in System 10. These controllers 14 support direct memory access, which means that a controller on its own can initiate memory read and memory write transfers. MAS 341 is the means by which BBI 21 accomplishes these memory transfers.

The heart of MAS 341 is PROM's 1784-5 (FIG. 40) similar to microsequencer 340 except that the circuitry is simpler. As previously described, the DDL logic signals BBI 21 that it would like to read something to memory or read something from memory. Microsequencer 340 recognizes this signal and acknowledges back to the DDL that it has accepted the request. Microsequencer 340, which is identical to the ITC microsequencer 134, signals to MAS 341 to start its operation by sequencing through a series of five states if it is a memory write transfer, and seven states if it is a memory read transfer. As MAS 341 sequences through the various states, the associated control lines are activated to accomplish the memory transfer out to bus 34.

Two signals called LUMRT and LUMWT are coupled to flip-flop 1795. The MRT signal is the pulse that would come in from microsequencer 340, commanding MAS 341 to start a memory read transfer. The other signal would be generated by microsequencer 340 if this was a memory write transfer. Either one of these signals sets one of the flip-flops 1794-5 that enables PROM memory 1784-5. The outputs of PROM 1784-5 are dedicated to setting or resetting level signals that are command signals to accomplish the bus interfacing for the transfer. Other output lines are dedicated to forming the next address for the firmware within the PROM's. The next address is fed back through the state flip-flops 788 and then becomes part of the next address at the input of PROM 1784-5. Two of the next address bits of the PROM are coming through a multiplexer 1780-1. Depending on the current state of flip-flop 1788, it selects various inputs to pass. At any instant, flip-flop 1788 would be in a given state, and the next state it would go to is dependent upon the state it is in as well as the next state of information being transmitted by multiplexer 1780.

Flip-flops 1792-3, as well as register 1790, comprise the circuitry that is set or reset by the control output of PROM's 1784-5. This circuitry generates action signals to bus 34 to gate the bus at the proper time after memory transfer.

Further, MAS 341 has gating FIG. 39 (including gates 1740, 1742, 1746, etc.) that aids in forming a branch signal identity throughout BBI 21. The signals produced by these gates are applied to a multiplexer 1780-1 which provides the next state information.

5. Last Accept Register and Last Retry Register 376, 378

Subaddressing allows devices in BBI 21 to be accessed. When an input transfer arrives, it always contains subaddressing information. The four bits of the subaddressing information point to which one of the subcontrollers is coupled to BBI 21. The last acceptance of the transfer register 376 holds the subaddress of the last bus transfer that has been accepted by BBI 21. On the other hand, the last retry subaddress register 378 holds the subaddress of the last transfer that was rejected, either with a retry or channel-busy signal, by BBI 21.

Interrupt level save register 374 receives transfers of seven bits of information which represent the system interrupt level assigned to BBI 21. Register 374 holds this information.

6. Service Interrupt Poll 380 and Bus Poll 394

These circuits are shown in detail in the SEL IOM system previously identified. Within SEL 3200 series digital computer there are 128 different interrupt levels identified by a seven-bit counter. Assigning any one of the interrupt levels depends upon the contents of an interrupt level save register. If BBI 21 is required to generate an interrupt request on this level, microsequencer 340 asserts various signals that come into poll 380 and generates the appropriate signals onto SEL bus 34 to accomplish the interrupt request. Thus, poll 380 makes the interrupt request and performs the interrupt poll.

Before BBI 21 can make a bus transfer, bus poll 394 determines whether or not any other controller 14 of a higher priority also wishes to make that transfer. If this is not the case, BBI 21 waits until the device for the higher priority has completed the transfer, and BBI 21 then goes out and makes the transfer.

7. In Register 356 and Address Register and Counter 362

System bus 34 is made up of two major buses, one of which is a 32-bit data bus and the other a 24-bit destination bus. The data bus is bidirectional. In register 356 captures the information and holds it locked in the register for the length of the time needed to process the information. Most of the time, register 356 is continually clocking in or "snapshotting" the information from the data bus every system clock cycle. Only when it is determined that the information is actually needed on the bus is the information in register 356 frozen. A bus transfer is made in a period of 140 nanoseconds, at which time there is an enabling signal called LANYXOUT that turns a driver on for the 150-nanosecond time period.

Address register and counter 362 is coupled to the destination bus of bus 34; the information on that bus is usually a memory address. The address counter of counter chips later defined provides a parallel loading to take an address from the destination bus and load it and then redrive it back onto the bus. A transfer would come from CPU 30 as an address and indicate that there is information in memory 32, for example, that should be read out. The address given is the memory address information. BBI 21 then assembles a memory read transfer. One of the pieces of information that must be supplied in a memory read transfer is the address in memory of the data to be accessed. That address is now located in address register 362. After accessing that location, memory 32 sends back the data.

8. Status Block 346

As previously described, tag bits have been used to define different types of bus 34 transfers that come to BBI 21. Three types of status transfers are used by CPU 30 to obtain information from BBI 21, such as whether it is busy, etc. During these transfers, BBI 21 gates the various status bits, as shown on FIG. 42, back to CPU 30. There are 12 status bits dedicated to predetermined functions. The majority of these functions are left open for defining the type of information represented by these bits for a particular application.

C. DCM 25

1. Run Control 816

FIG. 43 shows one of six identical circuits that comprise run control 816. All of the run control circuits are identical, and therefore only one of them need be described. A run request LTCRUN or LRUN may be activated in three different ways. A command labeled LDIGRUN may come from computer 30, a signal may be provided from an external timer designated HTIMRUN or there may be a patch panel request LPPRUN. Any one of these signals sets a flip-flop 1702 which, through a series of gates 1706-8, provides a signal LTCRUN if it has been previously selected as time-critical, or LRUN if it is non-time-critical.

The six LRUN signals are applied to monkey round robin circuit 832. The six timer-critical circuits LTCRUN are applied to the gorilla round robin 830.

2. Gorilla Round Robin 830

Gorilla round robin 830 (FIG. 44) is identical to monkey round robin 832 with the exception that if a gorilla is active, then a signal feeds into round robin 832 from circuit 830, preventing any monkeys from become active. Accordingly, only circuit 830, in which six inputs are identified as LTCRUN 1-6, will be described in detail. These inputs are applied to PROM 1710, which selects which gorilla is to be the next one acted upon. Assuming that gorilla 4 is presently active, and if there are simultaneous run requests from all of the remaining LTCRUN 1-6, for example, then the next active gorilla will be No. 5, etc. The remainder of circuit 830 is employed in synchronizing and storage of the signals. The lines defining the presently active gorilla are labeled HACTIVEGORILLA 4, 2, 1, and the lines defining the next active gorilla are HNEXTGOR 4, 2, 1. The lines to flip-flop 1716 are used to inhibit a monkey from becoming active while a gorilla is active. The firmware program for PROM 1710 is described later in detail, and this program is also used for the PROM for the monkey round robin 832.

3. Active Pending Register 786

The six addresses of active pending register 786 may come from one of five sources, viz, HNEXTMON, HACTMONK, HNEXTGOR, HACTGOR and HTCRSEL. These addresses enable the ADC/DCM 22 to address a particular transfer control register in the active pending RAM 750. Accordingly, the output of APR 786 is an HRAM address which provides three addressing lines to active pending RAM 750. The foregoing five address sources are applied to multiplexers 1730-2 (FIG. 45) to provide inputs to multiplexer 1733.

4. Active Register 794

The gates show details of active register 794 (FIG. 46). There are two output signals: HAPRTOACTIVE, which moves the active pending register to the active register, and HACTIVETOAPR, which moves the active register back to the active pending RAM 750.

It will be understood that interface circuits of analog computers 12a-f, including ADC and DAM driver circuits such as those in Pacer 100 of Electronic Associates, Inc., are coupled to the float, jam, halt, lines of APR 751. Activate-deactivate gorilla and monkey circuit 817 selects one of six outputs, while TCR loaded flops 814 indicate which TCR is active and when it goes inactive.

D. PAM 20

1. Echo or Wait for Read Block 1108

A CD bits signal on line 1200 is applied from BBI 21 to a flip-flop 1202 (FIG. 47). The flip-flop clocks the signal with a 1 MHz clock, used by the analog consoles 12a-f by way of line 1206. The resultant signal is clocked into a second flip-flop 1204, and in the time between the first and second flip-flop a strobe is generated to load the data from the CD word.

At the trailing edge of this period, a signal is sent back to BBI 21 indicating to the BBI that PAM 20 has acknowledged the command and that it has been stored. The same operation is accomplished with respect to TCW data line 1208. That line is coupled through flip-flops 1210 and 1212 to generate a 1-microsecond period for a strobe to load the data and, at the trailing edge, to generate a DDLRSP on line 1106.

Data request line 1216 is also coupled through flip-flops 1218 and 1220. When the controlled timing counter gets beyond state 4 and data request is high, there is a response with a DDLRSP on line 1106. Accordingly, the signals coming from BBI 21 are CD bits, TCW data and data request. These signals all require a DDLRSP to unlock the communications and allow BBI 21 to go on and capture another signal. In this manner, the DDLRSP is effectively an echo of the command coming from the BBI in which there is a two-microsecond wait to send it back.

2. Control Timing 1118

An oscillator 1241 (FIG. 48A,B) provides an 8 MHz signal to a binary counter 1240, which divides the signal down to 1 MHz. It will be understood that the 8 MHz signal is used throughout PAM 20 to clock the flip-flops. The control timing circuitry is initiated by either a read or a write command coming from BBI 21.

Initially, a control counter 1242 is first cleared, and this counter feeds a 3 to 8 line decoder 1244, which divides the output of counter 1242 into eight timing steps, viz, LT0-LT7. These timing steps are used to enable the setting of a group of flip-flops 1246, 1248 and 1250. Flip-flop 1246 is used for address enable, flip-flop 1248 is used for data enable, and flip-flop 1250 is used for strobe enable. Additionally, there is a busy flip-flop 1252, to indicate when the control timer is activated. Flip-flop 1252 is set by a read or write command and reset by either state 4 or state D of control counter 1242.

Further, gate 1254 is effective to AND the 1 MHz signal with a signal hold count. The signal hold count is activated for read from an analog computer, since it requires more time to get the data back from the analog computer than to just transfer it in one direction. The address enable flip-flop 1246, data enable 1248, and strobe enable 1250 provide outputs to the PAM 20 to activate by way of lines 1120, 1122 and 1126 the output buffers (FIGS. 25A,B) in the proper timing sequence. This strobe is inside both the data and the address. At the completion of an instruction, counter 1242 is again cleared and held in a zero state waiting for the next instruction.

The following is a listing for the firmware for PROM 1710 of gorilla round robin 830 as shown in FIG. 44. As previously described, the aforegoing listing is the same as that used in monkey round robin 832. ##SPC1## ##SPC2##

As previously described, the software required for operation of system 10 must support real time multi-task hybrid operations. The following software integrates with the SEL real time monitor (RTM) and SEL terminal support subsystem (TSS) software. Specifically, the hybrid input/output control system (HIOCS) and hybrid operations subroutines (HOS) are set forth although other hybrid programs and utilities may also be used. The HIOCS package is integrated into RTM to mate with the hybrid hardware controllers. This allows one resident copy of these input/output routines to service several users in a multi-task environment.

HOS is the user interface to the hybrid software system for real time use. HOS allos the allocation of various resources such as TCR's 62a-f, analog consoles 12a-f, channels 52,54 and interrupt sources to a proper user's application program. The following are HOS subroutines divided into four major categories or sets of subroutines. The HOS listings are later set forth in detail.

System Initialization (QJOB)

Allocate An Analog Console 12a-f (QSACI)

Select Analog Console 12a-f (QSC)

Allocate an ADC/DCM 22 Transfer Control Register 62a-f (QSADCTCR)

Construct ADC/DCM 22 Channel Program (QRBADCPR)

Chain ADC/DCM 22 Control Words (QSADCHN)

Loop ADC/DCM 22 Channel Program (QSADLOOP)

Initiate an ADC/DCM 22 Channel Program (QIADCP)

Digitally Activate a Block of ADC's (QSADCTRA)

Halt an ADC/DCM 22 Channel Program (QHADCP)

Read ADC/DCM 22 Channel Program Status (QRADSI)

Load an ADC/DCM 22 Timer (QWADDCMT)

Read an ADC/DCM 22 Timer (QRADDCMT)

Start ADC/DCM 22 Timer(s) (QSADTMR)

Allocate a DAM/DCM 24 Transfer Control Register (QSDMICR)

Construct DAM/DCM 24 Channel Program (QWBDACPR)

Chain DAM/DCM 24 Control Words (QSDACHN)

Loop DAM/DCM 24 Channel Program (QSDALOOP)

Initiate a DAM/DCM 24 Channel Program (QIDACP)

Digitally Activate a Block of DAM's (QSDAMIRA)

Halt a DAM/DCM 24 Channel Program (QHDACP)

Read DAM/DCM 24 Channel Program Status (QRADSI)

Load a DAM/DCM 24 Timer (QWDADCMI)

Read a DAM/DCM 24 Timer (QRDADCMI)

Start DAM/DCM 24 Timer(s) (QSDATMR)

Connect Subroutine to Hybrid Interrupt (QCONCT)

Real Time Hybrid Usage Request (QSGORT)

In the above first category of initialization, the first subroutine that a programmer must call is QJOB. This subroutine initializes the user's program to the hybrid interface. It also initializes the software determining which method the user wishes to run. For example, the user may wish to have run time errors that are sensed by the subroutine logged on a CRT line printer 38. In this subroutine, there are arguments that he can use to operate the method of his further execution.

The next initialization subroutine is QSACI which is a subroutine the user invokes to allocate to an analog console 12a-f. If the user calls this subroutine to allocate a console and the console is not being used by another task in the system, the request will be okayed by the subrouting. If this console is being used by another task, the user request will be denied and he will be aborted. The user may allocate as many consoles 12a-f as there are within system 10. One program may be running on two consoles. As a result, subroutine QSC selects a previously allocated console. All subsequent calls to the HOS subroutines will be directed to the console that was last selected.

The second set of subroutines are channel programming subroutines which are called prior to a real time run in the static test mode of operation. Only the ADC/DCM subroutines will be discussed as the DAM/DCM subroutines are similar. The first subroutine to be called is QSADCTCR which attempts to allocate one of six TCR's 62a-f. If there are no TCR's available to this user, as other users are using them, the request will be denied. If there is a free TCR, the user would be granted the TCR. This is the hardware component that the next call will be directed to when constructing a channel program. Once the user has successfully allocated a TCR, he calls a subroutine to construct a channel program. In this subroutine, the user tells the software the channels that are to convert and read the ADC, where in the FORTRAN program this data is to be transferred and other arguments,

QSADCHN subroutine enables the user to chain and loop channel programs. The maximum allowed is four chain channel programs or three chain channel programs which can be looped at the top of the chained stack. The user may also desire to do ADC transfer synchronously using the built in timers of the TCR's 62a-f. To use these timers, the user loads an initial value into these timers by calling channel program subroutine QWADDCMT. Also there is a call to start the timers in channel programming subroutine (QSADTMR).

After the user has constructed the CPCW, the last subroutine to be called is QIADCP which initiates the ADC channel program. This subroutine instructs the currently selected TCR to fetch the first CPCW pair in core. The TCR would then attempt to perform the I/O but only if the TCR is activated will the I/O be completed. The activation is provided by one of: patch panel control, timer timing out of a previously loaded value or a digital instruction. If the TCR is not active, no I/O will occur until later in the real time phase of the program.

The next subroutine set is the subroutine to connect the user written FORTRAN subroutine directly to a hybrid interrupt. This subroutine is also called in the nonreal time portion of this program. The actual physical connection in hardware instruction occurs when the user at a later call is granted real time use. In general, this subroutine recognizes the interrupts that the user wishes to use and the subroutines the user wishes to drive directly with those interrupts.

The last subroutine set is QSGORT which defines that the user desired to go real time. There are three arguments in this subroutine. A first argument tells the software which of the users prior allocated TCR's should be time critical. This means that this TCR will be able to preempt other users TCR's data transfers. Thus, if a user has requested real time usage and one of his TCR's has become active, he will get full control of ADC 52 at that time. Another argument relates to a similar situation for the DAM TCR's. The last argument is the ordering of resource allocator matrix 104 to indicate the interupts from the hybrid interface to go to priority levels in digital system 10. If the user passes the calling of this subroutine, that indicates no one else is using the real time subroutine and he will be granted full usage of system 10. Normally, the user suspends his task and the RTM has no knowledge of this task being resident in core. Upon occurrence of external stimuli of interrupts, the directly connected FORTRAN subroutines will do his simulation work in real time.

HIOCS is executed on a very high priority interrupt level in system 18. It is invoked through the usage of the illegal instruction trap so therefore it is higher than any software and most other hardware interrupts. HIOCS serves two broad purposes. The first purpose is to allocate hybrid resources amongst users. This guarantees system integrity of linkage 16 and controllers 14. Another purposes is to keep system integrity in digital computer 30. Mistakes of one user will not destroy another user's program or change I/O state of the system or destroy anything within monitor 36. The user is enabled to run in the unprivileged mode of operation. HIOCS also performs post hybrid task clean up in case a user forgets to perform post execution clean up operation. For example, if the user would give me a loop to channel program at the end of his task, HIOCS will issue a halt command to that channel program so that when RTM realizes that this task has ended, it loads another program into the same core memory, that memory will not be overlaid with ADC values.

HOS and HIOCS listings follow. ##SPC3## ##SPC4## ##SPC5## ##SPC6## ##SPC7## ##SPC8## ##SPC9## ##SPC10## ##SPC11## ##SPC12## ##SPC13## ##SPC14## ##SPC15## ##SPC16## ##SPC17##