US 4,524,416 AGrant

Stack Mechanism With the Ability to Dynamically Alter the Size of a Stack in a Data Processing System

Issue Date:1985-06-18

•8 Claims

•20 Drawing Sheets

Abstract

In a data processing system, a stack mechanism creates a stack of operands in a series of memory locations. The memory locations are grouped into stack frames corresponding to the operands included within individual procedures executed by a processing unit of the data processing system. The stack has a maximum number of allocatable storage locations with the actual physical size of the stack being equal to the total number of operands stored therein. The size of the stack is dynamically alterable to conserve usable storage locations in the memory and accessing of operands within a stack frame can be relative to the top or bottom of the stack frame.

Metadata

Assignee

Honeywell Information Systems Inc.

Inventors

Philip E. Stanley
Piotr Szorc

Application Information

Application Number:US 4304888

Filing Date:1982-09-30

Priority Date:1980-04-15

Art Unit:237

Classifications

IPC:

G06F 1516

Field of Search:

364200;910;900

Patent Drawings (20 sheets)

Description

FIELD OF THE INVENTION

This invention relates with particularity to an improved stack mechanism for use within a central processing unit to maintain, during processing operations, an orderly set of links between the entry and exit points in successive independent procedures performed by the central processing unit.

BACKGROUND OF THE INVENTION

It has been quite common within a data processing system to reserve a set of memory locations or registers to be used as an area for temporarily storing information. Such memory or registers have been referred to as a stack into which operands are pushed down (stored) and popped off (unloaded) on a last-in, first-out (LIFO) basis. The stack is divided into stack frames usually related to the programming concept of a procedure. A modular program, one having a plurality of procedures, will have multiple stack frames within a particular stack when a given procedure calls another procedure, e.g., returning jump, or calls itself, e.g., reentrant procedure call.

In prior art computer systems, the stack subdivisions or frames were commonly of a predetermined length and were not capable of being dynamically expanded or contracted. Accessing within a stack was accomplished by means of stack pointers, i.e., registers storing selected addresses of elements within the stack. A stack frame was created by storing addresses in the stack beginning with frame base relative address zero and incrementing in a numerically positive direction. One of the stack pointers was continually updated to reflect the address of the most recent entry in the stack frame relative to the base of the stack, i.e., the top of the stack pointer. Addressing of operands within the stack was performed by indexing from the current address of the top of the stack.

A problem has been found to arise when the only way to access operands within a stack frame is by indexing relative to the top of the frame. This is because any significant number of pushes or pops relating to a stack frame results in a continual change in the address of other operands in the stack relative to the top of the stack. As an example, a new stack frame is often created when a call is made from one procedure to a different procedure within a program. A call is accompanied by a number of parameters or operands transferred by the calling procedure to be stored in the stack frame created for the called procedure. These operands may contain the return address in the calling procedure as well as specific values for certain parameters. If, in the course of performing the called procedure, numerous entries are pushed onto the stack or popped off of the stack it may be impossible to determine the proper index relative to the top of the stack for addressing the return address to the calling routine. Such a problem will arise when the called procedure does not pop off all of the entries in its stack frame prior to returning to the calling procedure.

There is a second problem attendant with the prior art stack mechanisms. This arises from the common practice of allocating a fixed number of locations for each stack frame. This poses an artificial constraint on the stack mechanism and results in inefficient use of the memory or registers set aside to store the stack. For example, if a stack frame corresponding to a given procedure only requires a few entries, then the remaining entries in the frame will go unused and result in a significant waste of valuable memory space. In addition, some stack frames may require more space than what is normally allocated and, therefore, a stack overflow will occur. Finally, the selection of the size of the individual stack frames is commonly selected to minimize the number of stack overflows and therefore, has been commonly selected to be larger than necessary. This contributes to the inefficient use of the memory locations allocated for the system stack.

OBJECTS AND SUMMARY OF THE INVENTION

It is accordingly an object of the present invention to provide a stack mechanism within a data processing system which requires a minimum number of storage locations.

A further object of this invention is to improve stack mechanisms whereby operands within a stack frame can be accessed by indexing relative to both the top of the stack frame and the bottom of the stack frame.

Another object of this invention is to increase or decrease in size dynamically a stack frame in a stack mechanism.

Yet another object of this invention is to improve stack mechanisms whereby the amount of storage allocated for storing each operand within a stack frame is dependent upon the length of the operand.

It is yet another object of this invention to assemble stack frames in a manner that efficiently allocates the storage locations to the stack.

Additional objects and advantages of the invention will be set forth in part in the description which follows, and in part will be apparent from the description or may be learned from the practice of the invention. The objects and advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the appended claims.

To achieve the foregoing objects in accordance with the invention, as embodied and described herein, a data processing system having a memory for storing operand stacks comprising sequences of memory locations for addressably storing stack operands and wherein stack macroinstructions are executed to create an operand stack with a specified number of memory locations, to dynamically alter the number of memory locations included in an operand stack, and to delete a created operand stack, includes a stack mechanism for creating, altering, or deleting operand stacks in response to stack macroinstructions, the stack mechanism comprising; a control store for storing a plurality of sequences of microinstructions, each sequence of microinstructions associated with a different one of the stack macroinstructions; stack register means defining the characteristics of an associated operand stack as stack operands are added to and deleted from the operand stack, the register means including a current stack allocated register for storing an indication of the number of memory locations in the associated operand stack, a stack top register for storing the address of the stack operand most recently added to the associated operand stack, and a stack bottom register for storing the address of the memory location preceding the memory location storing the first operand included in said associated operand stack; and a central processing unit coupled to the control store, the stack register means, and the memory, the central processing unit for accessing said microinstructions stored in the control store associated with a stack macroinstruction, for executing the accessed microinstructions, and for dynamically changing the the contents of the registers in the stack register means in accordance with the changes in a operand stack resulting from the execution of the stack microinstructions associated with a stack macroinstruction.

BRIEF DESCRIPTION OF THE DRAWINGS

The manner in which the apparatus of the present invention is constructed and its mode of operation can best be understood in light of the following detailed description, together with the accompanying drawings, in which:

FIG. 1 is a general block diagram of a system in which the stack mechanism of the present invention may be included;

FIG. 2 is a detailed block diagram of an embodiment of a central processing unit capable of employing the stack mechanism of the instant invention;

FIG. 3 illustrates details of connections to the arithmetic logic unit of the central processing unit of FIG. 2;

FIG. 4 illustrates details of the composition of locations of the control store included in the central processing unit of FIG. 2;

FIG. 5 is a detailed block diagram of the control store and associated logic included in the central processing unit of FIG. 2;

FIG. 6 is a table illustrating the signals used for addressing the control store of FIG. 5 and for responding to various branch conditions arising from the performance of the firmware stored in a control store of FIG. 5;

FIG. 7 is a detailed block diagram illustrating test logic used to enable the control store of the central processing unit of FIG. 2;

FIG. 8 illustrates the details of a multiplexer used in addressing the control store included in the central processing unit of FIG. 2;

FIG. 9 illustrates an example of the operation of pairing locations in the control store included in the central processing unit of FIG. 2;

FIG. 10 illustrates the stack mechanism registers resident within the central processing unit of FIG. 2 and employed in implementing the stack operation of the instant invention;

FIGS. 11(a)-(e) illustrate five stack conditions capable of being implemented by the stack registers of FIG. 10;

FIGS. 12(a) and 12(b) illustrate a stack before and after additional entries have been added thereto.

FIGS. 13(a)-13(f) illustrate formats for stack-related instructions executed by the stack mechanism of the invention.

FIGS. 14-23 illustrate in a flowchart form the operation of the stack mechanism of the instant invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

1. System Overview

FIG. 1 is a general block diagram of the configuration of the system in which the CPU to be described may be included. Such system, includes a central processor unit (CPU) 100, and optional processors which may include a scientific instruction processor unit (SIP) 101 and a commercial instruction processor unit (CIP) 102. These optional processor units may be used to extend the range of the basic processor unit 100 for special applications. The system may also include a Cache/MMU (Cache memory/memory management unit) unit 103, an electrical system bus 105, memory modules (main or system memory) 106, input/output (I/O) controllers 107 coupled to I/O devices 108, and a multiline communications controller or processor (MLCP) 109. The system may also include a multiprocessor configuration in which there is a second central subsystem 110 which includes all or a portion of the above described system.

The central subsystem processors are connected to each other by way of a local bus 9 and they are connected to the rest of the system by way of the Cache/MMU unit 103. The function of the Cache/MMU unit in the central subsystem is to provide a buffer storage for the portion of main memory 106 that is currently being used by the processors, and to provide for translation of the main memory addresss. The function of the local bus is to provide an interconnection among the three processors and the Cache/MMU unit. There is one physical local bus 9 which is organized into numerous interfaces. As shown in FIG. 1, there are private interfaces between the CPU 100 and the other two processors. There is also a private interface from each processor unit to the Cache/MMU. The local bus is a time-shared or public interface, that is, shared by all three processor units and by the Cache/MMU. In addition, the Cache/MMU 103 provides an interface via the system bus 105 to the rest of the system, primarily the memory 106 and the I/O devices 108 via controllers 107.

The CPU block diagram of FIG. 2 will now be discussed. More particularly, the various electrical buses and functional units and their interrelationships shall be discussed. The primary elements of the CPU are shown within the dotted lines. A first such element is the local bus interface 10 which includes a data buffer, a round robin procedure storage buffer 36, and various procedure/data word and address multiplexers (muxes) coupled with a source bus 33 as well as other devices to be discussed. The first element 10 is utilized for the purpose of receiving data from the system bus 105.

A second element 16 is the arithmetic element and includes several devices including two sets of register arrays (RAMS) called the DRAM 11 and the BRAM 12, and the RAM or R Bus 13, to which RAMS 11 and 12 connect. It also includes the Input or Inner Bus 14 to which the BRAM, as well as other devices, connects. The second element also includes the DALU 15, i.e., the arithmetic logic unit, which is coupled to drive the DRAM 11.

A third element 22 of the CPU includes the Outer Bus 17 and the Write Bus 84, the shifters 19-1 and 19-2 which connect them, and the selectors which drive them. These selectors include the DALU/R Bus selector 30 and the Y register/P register (YP) selector 21.

A fourth primary element 27 is the address section of the CPU and includes the procedure address register (P) 23 and the data address register (Y) 24. It also includes the address multiplexer 25 and the Prefetch address register 26. Also included in element 27 is the write data register (W) 28, and the write data selector (WS) 29. This portion of the CPU is utilized for the purpose of transferring data to the local bus 9.

Device 30 is the test logic of the CPU, which is shown in FIG. 7, and includes a set of eight-to-one multiplexers which in turn drives a second set of at least two eight-to-one multiplexers which generate a single binary signal called "Test True" or "Test False", which is then used to control branching in the firmware. The Test True and False signals are coupled to the PROMS included in the control store banks 80 and 81 as shown in FIG. 5.

The next address generator 44 is coupled to the F register 38 and the F Bus 37. The next address generator is primarily concerned with generating addresses for use in the CPU.

The function of element 10, which is the local bus data input area, is to receive data which is returning from the Cache/MMU or from other devices on the local bus, to select data from the bootstrap PROMs if a bootstrap procedure execution is requested and to direct such data to the appropriate data buffer. If an instructional fetch is called for, the data is placed in the function register. More particularly, the primary data buffer 31 receives 16 or 32 bit data words from the local bus 9. Its output is connected to the L bus 32. The L bus is used both to drive the source bus 33 by use of a pair of selectors DA 34 and DW 35 and to drive the four word procedural data word buffer 36. Non-procedural data words enter the CPU via the source bus 33 which is driven from the data selectors from the L bus 32. Procedural data words enter the CPU via a different set of selectors, PA 39 and PW 40, coming from the procedure buffer 36. The procedure buffer has the responsibility of containing the next two or three procedural data words to be executed, so that when they are required, time will not have to be spent fetching them. It is automatically reloaded from the Cache/MMU via the local bus as it is emptied by the CPU.

The F bus 37 is a special bus which is used for gating information for the F register 38, i.e., the four areas labeled FOP, FM, FN and Link. The F register is the primary instruction register of the CPU. The function of the F bus 37 is to take data from the F register 38 and provide data to the F register from various sources. There is also a multiplexer 43 which allows the loading of either constants or data from element AS' 42 into either the Link or the FN location of the F register 38.

Also in the element 10, which is the data input area, there is a set of bootstrap PROMs 41 which can be used to provide instructions, in place of instructions from memory if operating in the bootstrap mode.

The storage device 36 having locations A, B, C, and D, is a round robin four-word procedural data word store. Whenever it has room for two procedural data words, i.e., it is at least half empty, then it automatically initiates a memory read of procedural data words via the local bus 9, from the Cache/MMU. The returning procedural data words are dropped into the next two locations which are available in this round robin store. As the CPU uses procedural data words by either emitting it to the source bus 33 for use inside the processing elements of the CPU, or by sending it to the F register 38 because it is part of an instruction, then the two pointers 36-1 and 36-2, which represent in effect cursors in the round robin store, advance to the next location available by use of a conductor included in device 36 which successively enables the four locations to the outputs of device 36. The left-hand cursor or pointer 36-2 indicates the next procedural data words to be fetched; the right-hand cursor 36-1 represents the procedural data word following that. Sometimes one procedural data word is used up in which case the procedural data word indicated by the left-hand cursor would be used to generate the information, passing through the PW multiplexer 40. Sometimes there is a requirement for two procedural data words (as when pulling a large address form (LAF) address). Then the procedural data words indicated by both the left-hand and the right-hand cursors are combined in the PA multiplexer 39 of the procedure selector. This will be described more fully when this logic is described.

Thus, element 10 is the area concerned with bringing data into the CPU through the source bus, via either data selectors 34 and 35, or the procedural data word selectors 39 and 40, or bringing data directly into the instruction (F) register 38 via the procedural data word buffer 36 and the F bus 37. Device (F') 76 is used to hold selected bits from auxiliary address information or syllables. The address syllable is actually part of a 16-bit data descriptor. Certain bits in the descriptor have to be retained for subsequent analysis. The operand-size bit in the CIP descriptor and the operand size and signed versus unsigned bits in the K register descriptor have to be retained. These are retained in the five-bit F' register 76.

The second element 16 contains the arithmetic/logic unit (DALU) 15, the BRAM 12 and the DRAM 11, containing all of the programmer visible registers, plus a number of non-programmer visible work registers. These data storage elements are organized as follows: The DRAM 11, which is one of the two sources which can drive the RAM bus 13, contains 32 locations, each of which is 24 bits wide. These locations are used to hold so-called D registers, two halves (K & L) of the so-called K registers which are 32 bits long, a number of internal pointers, and seven so-called M registers. The BRAM 12, which contains 16 words, each 24 bits wide, contains the seven base registers plus a number of programmer visible and programmer invisible pointers of various types.

FIG. 3 shows in more detail the relationship between the RAMs, the buses, and the DALU 15. The DRAM and BRAM can each be sources to the RAM bus 13. The BRAM 12 is actually a dual structure; that is, there are two parallel BRAMs 60 and 61 in the CPU. The two elements of the BRAM are loaded identically. One such element which drives (or transfers data to) the R bus is element 60, and the other which drives the I bus 14 is the BRAM element 61. The important aspect of the two identical BRAMs is that the system speed is increased.

The R bus can be driven either by the BRAM 60, or by the DRAM 11. Other sources, as will be seen, also drive the I bus 14. The DALU 15 uses the I bus and the R bus as its inputs and the output of the DALU is then recycled to the DRAM and/or used elsewhere in the CPO.

Element 16 includes two buses of major interest. The so-called input or Inner bus (I) 14 is the primary source of original input within the processor, that is, data from memory comes into this bus as does the output of the Q register 50 via driver 54, and also the output of the BRAM 12. The second bus, the so-called R or RAM bus is the point at which the outputs of the two different RAMs, the BRAM and the DRAM, are coupled. The outputs of these two buses are the inputs to the DALU 15, which generates data which can either be returned into the DRAM or sent to the outer bus 17 for further processing in the system.

Both the inner and the RAM buses employ a hardware element referred to as the so-called "batlatch" (BL) which is a gate whose input and output are tied together for coupling signals onto the bus. This batlatch is intended to take over control of that signal after the driving source is required to go off the bus for timing and sequencing reasons. When this happens, the batlatch then supports that signal at the same level as it was when the driving sources were present, until the batlatch is turned off at which time the bus signal can relax (go to a predetermined level such as zero volts).

The RAM bus 13 is the interface for the transfer of data between the control panel 90 and the central processing unit, and is connected to it via 16 bidirectional signal lines. Memory data couples to the Inner bus 14 via a set of drivers called the data latches (DL) 51. The data latched are paralleled by a second series of latches called SE or sign extension latches 52. These SE latches are used either to sign or zero extend the data latches when they are only providing a sixteen bit quantity to the 24 bit bus. Similary, the SE logic 52 is used to take either 7 or 8 bits of the current instruction, and sign extend it to 24 bits for use by the branch and value instructions which get their displacement and their operand, respectively, from the right-hand portion of the instruction word.

There is also associated with the BRAM and DRAM logic called the select/modify logic 53, which allows the DRAM and the BRAM to be addressed in blocks of eight registers, using various fields in the instruction to control the individual register selection within that block.

Constants generator (KON) 70 is also included in element 16, the primary ALU area, and is another source to the Inner bus 14, that is, it is the source that can be used to generate 24 bit constants onto the inner bus for use by the arithmetic unit under control of the firmware.

Thus, element 16 is the primary arithmetic operation area in the CPU; that is, it is concerned with operand manipulation and with the generation of operands for insertion into the data RAM, or operands which are intended to be sent to the Outer bus, either to be returned to the BRAM, or to be sent to the various address and data registers for output to the local bus 9.

Element 22 is concerned primarily with the section of the CPU referred to as the Outer bus, and the write bus. The Outer bus is the bus upon which various data paths are collected, i.e., the Y register, the P register, the output of the DALU and the RAM bus. The Write bus is the bus which is used to transfer operands to the BRAM, to the write data registers, and to the BALU 55 which is the ALU primarily responsible for address manipulation.

Between the Outer bus 17 and the Write bus 84, is a set of twenty-four multiplexing shifters 19, which allow the shifts of 1, 2, or 4 bits in either direction on Outer bus 17 operands. In addition, there is also a set of sixteen multiplexing shifters 56 which are used to load operands in the Q register.

Two of the four different types of operands which are sourced to the Outer bus originate at the DALU and at the RAM bus and are selected through a two-to-one multiplexer 20 coupled to the Outer bus 17. Two other sources for the Outer bus are the Y register, which is the data address register, and the procedure address (P) register. The outputs of these two units are coupled via a two-to-one multiplexer (Y/P selector) 21 and sourced onto the Outer bus 17. Also coupled to the Outer bus 17 is the I register 57, the indicator register for the central processing subsystem. The contents of the Outer bus 17 can be transferred either directly or shifted to the Write bus 84 for further processing in the CPU. They can also be loaded into the Q register 50 directly so that the Q register can be used for double precision shift operations.

All shifting which takes place within the CPU takes place by manipulation of operands between the Outer bus and the Write bus. The twenty-four shifters 19 are used to shift 24 bit operands, the size of the operands on the O bus 17 and W bus 84, left or right under the control of the control store word. They can shift by either 1, 2, or 4 positions, transfer the operand straight through, or transfer the operand through with the two right-hand bytes of the operand interchanged. The operation of these shifters is controlled by a special field in the control store word. This section of the CPU is used for 24 bit operands shifts. When 32 bit (double register) operands, are shifted, the right-hand operand is loaded into the Q register via the Outer bus, and then a shift is performed in which not only the W bus shifter 19, but also the sixteen Q register shifters 56 are operated. The Q register contents are thereby treated as a 16 bit extension on the right-hand end of the 24 bit operand which is being shifted onto the W bus 84; the ends of these two shifters are coupled together, thus allowing a 40 bit shift to occur. In such double register shifts, the operands of interest are the 16 bits which are returned to the Q register and the 16 bits which are transferred to the right-hand two bytes of the Outer bus, while the left-hand eight bits of the W bus data are usually discarded.

Element 75 represents the collection of the status bits in the CPU for emission to the W bus via driver 93 under certain conditions. The W bus has two sources, the primary one being the shifter 19 from the Outer bus 17. This shifter 19 can be disabled as an input to bus 88 and instead the status bits (S/Z) 75 can be placed on the Write bus 84 for subsequent analysis by the firmware.

One unit of hardware which is associated with the Q register which has not been previously discussed is the XB register 58 and the decoder 59 coupled thereto. The XB register 58 is used during index shifting to capture those bits of the index which are shifted right for sub-word operands and which are lost because they move to the right of the single word bit in the address. They are captured in the XB register, one bit for half-word shifts, two bits for digit shifts, and four bits for bit-sized operand shifts. The XB register contents are subsequently used to directly control left versus right half selection when dealing with half-word operands, when generating a parameter to be sent to the commercial instruction processor unit for digit operands, and to generate a mask of 15 binary zeros and a single binary one, in conjunction with the decoder 59, when operating on bit operands. That is, for a bit operand, a word is read, a mask is used to set or clear the selected bit in the word, and then the word is written back into the memory. The mask is generated from decoder 59 and is an input to the Q register 50. That is, one of the items that can be loaded into the Q register is the decoder mask generated from the contents of the XB register 58.

The fourth major element in the CPU is area 27, the element which is concerned with the generation of addresses, commands and operands for transmittal to the local bus 9, and via the local bus to either the CIP unit, the SIP unit, or the Cache/MMU and hence, to memory or to Input/Output (I/O) devices. This major element can be broken down into approximately three areas, the first being the write data register (W) 28 and the write selector (WS) 29. Register 28, is a 32 bit register with a partitioned load capability, that is, either the right half or the left half, or both halves of this register can be loaded. The right half is always loaded from 16 bits of the Write bus 84 via drivers 74. The left half is written with data which comes from the write selector 29. This write selector has as its 16 inputs either the 16 bits of the right half of the Write bus or the eight left-hand bits of the Write bus plus either sign or zero extension. The Write bus is the primary path by which data can be sent to the W register 28 in order to subsequently write such data to any other device coupled on the local bus.

The next area in the local bus output interface 27 is the command driver 71. The command driver 71 drives a series of command lines which accompany all transfers by the CPU on the local bus and indicates to the Cache/MMU either a memory reference cycle, an I/O reference cycle, or a local bus reference cycle to the Cache/MMU or one of the local bus processing units. The information which is transferred onto the commonad lines is derived from the F and FK fields of the control store or firmware (FW) word shown in FIG. 4, which also at other times the functioning of the F register 38 of element 10.

The third portion of element 27 includes the two primary address registers, i.e., the Y register 24, for non-procedural data word memory addresses and for addresses to other devices on the local and system buses, and the P register 23 (the program counter), and the prefetch register 26.

The P counter 23 keeps track of the last word which the firmware has used, which word has been taken from the round robin buffer 36 and input to the CPU for operations thereon. The prefetch register 26 keeps trck of which word is to be next fetched from memory; that is, the P and PF registers may differ by any number in memory address from one to four words, depending on how full the round robin buffer is and how recently the central processing system has taken procedural data words from the buffer. If the central processing system has taken all the procedural data words from the buffer, it takes a finite amount of time for the Cache/MMU to respond with new procedural data words in response to a request, and fill buffer 36 again. Under those circumstances, the prefetch register 26 and the address of the P counter or register 23 would be close together or the same in address content. When the buffer 36 is full, and the CPU has not taken any procedural data words recently, then the P register will be two to four words behind the prefetch register because it is indicating the last word taken from the prefetch register indicates the next word to be sent forth from memory as soon as there is room for it in the prefetch register.

The contents of the P register 23 are never admitted to the local bus as a memory address source. The two sources for data which can go to the local bus via the virtual address multiplexer 25 are the Y register 24, which is used for all non-procedural data word fetches, and the prefetch register 26, which is used automatically by the prefetch logic to go to memory and request procedural data words for use in subsequent firmware steps.

The arrows on devices 23, 24 and 26 are indicative of the incrementing capability of those particular registers. That is, the P register can increment by either one or two words depending on the number of procedural data words which are taken out of the prefetch buffer at a time. That is, pulling one procedural data word from the prefetch 36 buffer automatically increments the P register by one; pulling two procedural data words from the prefetch buffer 36 automatically increments the P register by two. The prefetch register always increments by two, because prefetches are always performed on a pair of procedural data words. The Y register can increment by either one or two under the control of the firmware as it is used, or in preparation for future use. There is a bus control field in the firmware word which allows control of such incrementing and the various cycle requests to the local bus.

The inputs to the W register 28 are two 16-bit data paths which are exercised in various ways. If it is desired to write a 24-bit address, then the drivers 74 are turned on enabling the right-hand 16 bits of the Write bus to be transferred to the right-hand 16 bits of the W register. The multiplexer 29 is conditioned such that the left-hand eight bits of the Write bus and the eight binary zeros are emitted to the left-hand half of the W register 28. This loads a two-word, 24-bit, address into the W register for subsequently writing into memory. If it is desired to write a single word into memory, then the right half of the W register is not clocked (i.e., enabled), and does not change; and the multiplexer 29 is conditioned to enable the 16 right-hand bits of the Write bus to the 16 left-hand bits of the W register where they are clocked in. For single words writes, the left half of the W register is the data that is written into memory.

The logic heretofore described is used to request and accept data from other entities on the local bus, such as the Cache/MMU or the CIP unit or the SIP unit to operate on it and store it internally in various registers via the two ALUs which are contained in the system, to create new operands which are subsequently modified and sent out to the local bus to one of the entities thereon with an address of which entity (computed internally in the CPU and used to control the local bus). All of these operations are performed under the control of control firmware which is stored in a 2,048-word by 96 bits-per-word control store 80 and 81 which is included in the CPU.

The control store is partitioned into a number of individual fields, each of which controls some aspect of the internal performance of the CPU. FIG. 4 illustrates the firmware word and the various fields of interest therein. The first of these fields is the local bus (LB) field, bits 0 through 7 of the control store. The LB field is concerned with the requesting of various cycles from the local bus and the responses thereto. It is also responsible for the emission of virtual addresses via the prefetch register or the Y-register, and it controls the various incrementing of the Y and P registers. The local bus field also controls the successive reading of procedural data words from the round robin procedural data buffer 36, and the enabling of the non-procedural data word and procedural data word multiplexers 34, 35, 39 and 40 respectively, onto the source bus 33 for subsequent use by the Inner bus 14.

The next field, the clock (CK) field, bits 8 and 9 of the control store, is used to control the CPU clock speed, i.e., the interval between successive clock pulses in the CPU. Each firmware step has assigned to it the clock speed which is appropriate to its complexity. In the system of the present invention, there are four possible intervals between successive clock pulses: 96 nanoseconds, 105 nanoseconds, or 130 nanoseconds, 170 nanoseconds. Which of these is used for a firmware step is controlled by the CK field for that firmware step.

The next field is the TC field, bits 10 through 15 in the control store. This is a 6-bit field and selects one of 64 possible logic functions within the CPU for testing and control of firmware sequences.

The next field is the BR field, bits 17 through 19, and is used to control the type of action which is taken as a result of a test condition, that is, what type of branch occurred, whether the branch simply selects between two firmware steps, or whether it selects between a firmware step and a so-called control store splatter (the generation of one of a multiplicity of control store next addresses based on the state of one or more logic elements in the CPU in which the next address is controlled by some function of the F register 38 contents).

The next field of interest, the TP field, bit 21 is the Test Polarity field. It controls whether the test condition is tested to determine whether it is true or false.

The NA field, bits 22 through 31 is a ten-bit field which is used to generate at least one of the two addresses which are selected by all branches. The other address is either also derived from the same NA field or from the control store splatters previously mentioned.

The next field, bits 32 through 39, is the F register control field, partitioned into a four-bit part called F and a second four-bit part called FK. The F and FK fields are used to control the loading and strobing of various elements in the F register and in the area of the F bus; that is, it is used to control the strobing of F' and AS' devices, the subparts of the F register 38 and the multiplexer 43. Which of these devices are loaded and what sources they have for their loading are all controlled by the F and FK fields.

The next field, bits 40-47, is the K (i.e., constant) field, and is used to drive the I bus 14. It is an 8-bit field which provides the right-hand eight bits for all constants which are admitted to the I bus. The left-hand 16 bits of the constant are controlled by the I bus field which will be discussed hereinafter. The K field is directly related to the KON device 70.

The next field in the control store, bits 48-63, is concerned primarily with the addressing and control of the DRAM 11 and the BRAM 12. The two-bit field, DW, is used to control writing into the DRAM with one bit controlling write operations into the left most (the most significant) 8 bits of the DRAM, and the other bit controlling write operations into the second right most (the least significant) 16 bits. The next bit, bit 50, (field BW), is used to control write operations into both parts of the BRAM, which are always written into together. The next field DS is five bits in length and provides the selection of one of 32 registers in the DRAM 11. The DM field, bits 56 and 57, is the select-modify field associated with the DRAM and allows the selection of either the DRAM directly or any of three different functions of the F register 38 for qualifying the addressing of the DRAM.

The next field of interest is the B select field (BS), bits 60 through 63, which is used for the selection of one of the 16 registers in the BRAM. The two-bit field comprising bits 58 and 59, is the select-mofify field (BM) for BRAM addressing. It performs similarly to that for the BRAM addressing except that it is separate so that independent select and modify functions can be specified for both the BRAM AND DRAM.

The next field of interest is the GP field, which is seven bits in length (bits 65-71), and is used to control a large number of general purpose micro operations within the processor, such as the clocking of various bistable flip-flops and the enabling of various control functions. It is also used to control access to the control panel from the CPU, to input and output data to the control panel, and to control various flip-flops therein.

The next field, the RB field, bits 72 and 73, is used to control the sourcing of data onto the R bus 13 with one bit controlling the BRAM and the other bit controlling the DRAM.

The BA field, bits 74 through 79, is used to control the functions of the BALU 55, the arithmetic/logic unit which has as its inputs the I buas and the W bus and which can be used to drive the Y register 24. These six bits are used as follows: one as a carry inject to the BALU, one for logic/arithmetic conntrol, and the other four to select particular functions.

The next field of interest is the IB field, bits 80 through 83, and is used to control which data source is enabled to the Inner bus 14. This field can enable the data latches, sign extend the F register or select a constant, or the BRAM in various combinations to the I bus. The next field, the SH field, bits 84 through 89, is used to control both the shifters and the W bus sourcing which was discussed earlier. These six bits enable the system to have a large amount of autonomous control over what registers are enabled to the W bus and whether or not the Q register loads and in what direction the shifters shift the operand as it passes through.

The final field, the DA field, bits 90 through 95, is the function control for the DALU 15 with six bits having analogous functions to those in the BA field, i.e., one bit is a carry inject, one bit selects logic versus arithmetic and the other four bits are to select particular functions.

The control store for the CPU is actually two control stores: an upper bank 80 and a lower bank 81. The upper bank is addressed by the next address generator 44 and the lower bank is addressed directly by a portion of the contents of the then current control store word. In the case of the more common types of branches, both use the NA field of the control store word as the basic source of the next address, and modify it in some way so that the two addresses which go to the upper and lower bank are essentially identical. In the case of the other branches, those which select between a next address and a PROM generator splatter, the lower bank receives the next address field unchanged from the current control store word, while the uppper bank receives the splatter address from the current control store word. The control store is partitioned in this way so that the system can be doing a simultaneous access of both possible next address control store words and can postpone the decision of which one is going to be used until almost the very end of the current control store step, just prior to clocking the data in the control register 82. The details of the control store shall be discussed hereinafter.

The manner in which the CPU and firmware interact to perform certain typical central processor unit operations, such as (a) reading a word from memory; (b) computing a next address for reading another word from memory; (c) writing a word into memory and (d) performing some type of iterative algorithm in which the operations are entirely internal to the CPU, will now be discussed. The situations given will be typical and representative operations will be described.

In the case of reading a word from memory, during a control store step, the control store word will enable the virtual memory address of the word from the BRAM 12 onto the RAM bus 13 via the RB, BS, and BM fields. Then the SH field will condition the Outer bus and the W bus sourcing such that the RAM bus is enabled through the selector 20 to the Outer bus 17, and then unaltered through the shifter 19 onto the W bus 84. The W bus is an input to the BALU 55, which is conditioned by the BA field to enable its W bus input directly to its output. At the same time, the LB field will cause the Y register 24 to load, thus transferring the address into the Y register.

In the next firmware step, the LB field then causes a request to be directed to the local bus indicating that a memory request is required and that the address provided to the local bus should be used to make such request. The command lines 71 are conditioned by the F and FK fields (shown as box 72 in FIG. 2) to indicate the memory read request is required. The address multiplexer 25 is conditioned to enable the contents of the Y register onto the local bus via the 24 address lines. The Cache/MMU memory, which monitors and controls all activity on the local bus, acknowledges the request, while the CPU proceeds to the next firmware step. In the next firmware step, the local bus field will then specify a stall, indicating that the CPU should not leave this firmware step until the end of the local bus cycle has been detected returning the data from the Cache/MMU via the local bus into the CPU data buffer. As soon as this return cycle is detected, then the stall ends, and the CPU will then enter the fourth step of this read cycle.

The fourth step will be one in which the firmware will specify an LB field which uses the data that is in the data buffer, and sources it onto the Inner bus for use by the CPU. The LB field will thus enable either the DA driver 34, or the DW driver 35, depending on whether a word, 16 bits in length, or an address, 24 bits in length, is required onto the source bus. In turn, the I bus field will specify that the source bus, either all 24 bits of it, or 16 bits sign extended to 24, will be enabled onto the Inner bus. In the Inner bus, assuming that there is a 16-bit word that is being received back from the data buffer, the DA field would be used to enable the DALU to read the Inner bus, and then the DW, D select and DM fields would be controlled to address and write into one of the 32 registers in the DRAM. This would complete the fourth step of this control sequence, the first of which loaded the Y register, the second of which made the read request, the third of which waited for data to come back, and the fourth of which took the data and enabled it into one of the registers in the CPU.

A second typical sequence of operations which may be performed by the system is the manipulation of operands in an arithmetic operation, such as adding an operand which has been fetched from memory to one of the data registers and then, after the addition, returning to the data register and also placing it into the write data buffer so that it can be subsequently written back into memory. This sequence of operations would begin following the stall referred to in the previous operation. This step would be one which enabled the data, which in the case of a word operand would be a word, from the data buffer 31, via the L bus and the data multiplexers 35, onto the source bus. Then, the I bus field would specify that the source bus, sign extended via the sign extension device 52 and the data latches 51 be enabled onto the inner bus. At the same time, the R bus control field would specify that the particular data register of interest be enabled to the R bus 13 from the DRAM 11. The DA field, i.e., the DALU control field, would then specify that the DALU be conditioned to add the 24 bits on the R bus 13 to the 24 bits on the I bus 14. This logic 15 would have as its output 24 bits representing the sum. The DW bit would indicate that the output of the DALU was to be written back into the DRAM 11.

At the same time, the output of the DALU would be enabled to the W bus via the SH field. This field would indicate that the selector 20 would be enabled to the Outer bus and would be conditioned to select the DALU output instead of the R bus output for that purpose. Also at the same time, the shifter 1 would be conditioned to pass the Outer bus contents through without change to the Write bus 84. The same LB field which specified DW 35, to the source bus 33, would also have specified that the Write bus would be enabled to the left half of the W register 28 via the selector 29. This would all take place in a single firmware step. Since this is an add operation, presumably the overflow from the add would be tested by a test specified by the test logic 30. The next address logic would generate one address which would indicate a particular entry to go to if the operand is to be written back into memory immediately. The other address, which would be to the lower bank, would be generated to take the next step which would then set the overflow indicator in the I register 57. If there was no overflow, the I register would have been automatically cleared.

The third operation would consist of writing an operand into memory. This would take three firmware steps to perform. The first would be a step in which the address in which the operand is to be written is loaded into the Y register 24. The second step would be to put the operand to be written into the W register 28. The third step would be one in which the LB field would specify that a local bus request, specifying a memory write, be addressed to the local bus for the Cache/MMU to receive and execute.

The first step, that which loads the Y register 24, would get the address presumably from one of the 16 locations in the BRAM random access store. This would be performed by conditioning the I bus field so that the I bus would look at the output of the BRAM. The B select field would indicate which of the 16 BRAM registers was addressed for this purpose. The BALU 55, which is controlled by the BA field, would be conditioned to pass the contents (24 bits) of the Inner bus to its output. The LB field would be one which specified load the Y register. Since input to the Y register is the output of the BALU, this would transfer the contents of the selected BRAM into the Y register. The next control store step would take the operand from its origin, wherever that might be, in this case, for example, one of the 32 DRAM locations. The DS field would select which of the 32 DRAMS would be enabled. The RB field would enable the DRAM onto the R bus. The SH field would select the RAM bus to the Outer bus 17 via selector 20, and the Outer bus to the Write bus 84 via shifter 19 with no shifting taking place. The LB field would specify load the W register, left half. This would be performed by conditioning the W selector 29 to enable the right two-thirds of the Write bus to the left half of the W register, and the W register would be enabled to load its left half.

Finally, the third firmware step would be performed. In this firmware step, the only essential operation is that the local bus field, the LB field, should specify a local bus write to memory. This would use the F and FK bits as command lines going to the local bus to indicate to the Cache/MMU that this is a memory write operation. The address would be enabled to the local bus via the virtual address selector 25 from the Y register 24. The data would be enabled to the local bus from the W register 28. The Cache/MMU memory arbitrating all local bus transfers would recognize this as a memory write cycle, and would take the address, map it, send it to memory accompanied by the data, and indicate on the system bus to the memory that this is a memory write operation.

The fourth sequence of a typical operation which the CPU could perform is that in which a double precision operand is shifted some number of bits to the left or right. Presuming that the operands are both in the BRAM, that is, the two operands of interest are in the BRAM, the first firmware step would initiate the transfer of the right-hand of these two operands to the Q register. This would proceed as follows. The BS field would be conditioned to address one of the 16 locations that contains this operand in the BRAM 12. The R bus field, controlling the RAM bus 13, would be conditioned to take the BRAM output instead of the DRAM output. The SH field would be conditioned to transfer the R bus to the Outer bus via the selector 20, by enabling it to the Outer bus and selecting its input from the R bus, and would also specify that the Q register and the W bus both receive the contents of the Outer bus and would clock the Q register to load it. This would cause the operand addressed in the BRAM to be transferred to the Q register. The next step, or steps, would be those in which the shifts are actually performed. In this step, the other of the two registers containing the second operand in the BRAM is addressed by the B select field and the BRAM is enabled onto the RAM bus 13, via the RB field. Then the RAM bus would be enabled to the Outer bus via the selector 20. The SH field would take any of a number of values depending on which direction and for how many bits the shift is to be performed. The SH field is capable of selecting a left or right shift of either one, two or four bits. In any of these cases, the Q register is considered to be connected as an extension of the Outer bus, creating a 32-bit operand. Actually, it is a 40-bit operand, but the left-hand eight bits of the Outer bus are ignored. This 32-bit operand is shifted either left or right as indicated by the particular SH field. The 16 right-hand bits are returned to the Q register and the 16 left-hand bits together with the eight bits which were ignored are transferred onto the W bus. This is controlled by the SH field which has exclusive control over the shift distance.

The operand having been shifted from the Outer bus to the W bus and from the Q register back to the Q register, the SH field causes the Q register to reload the shifted operand while at the same time the BW field causes the W bus to be written into the addressed BRAM location. Thus, the contents of the B and Q registers have been shifted and returned back to the B and Q registers. The particular end effects associated with this shift, whether this is an open, circular, or arithmetic shift, are a function of the control flip-flops in the CPU. This type of step, in which the actual shifting is performed, is performed some number of times in various combinations, that is, if it were desired to shift left by five bits, then a step which resulted in a shift left by one bit would be succeeded by a step which shifted left by four bits. To shift right by three would require a shift right by two and then by one, for example.

Then, after the final shift has been performed, i.e., after the operands are now correctly aligned in the addressed BRAM location and the Q register, the final step does not cause a shift, but instead returns the Q register contents back to the BRAM location from which it was originally loaded. This would be performed as follows. The I bus field would specify that the I bus is driven by the Q register (16 bits of Q extended by eight bits of binary zeros). The DALU 15 would be controlled by the DA field so that it passed through the I bus unchanged. The SH field would be selected so that it enabled the DALU, unchanged via the selector 20, to the Outer bus 17, and again unchanged, via the shifter 19 to the W bus 84. The BW bit 50 in the firmware word would then be set to condition the loading of the BRAM from the W bus, and the BS (B select) bits would be conditioned to specify which of the 16 locations in the BRAM were to receive the shifted operand. This would all take place in some number of firmware steps, for example, three or more. One step would be used to load the Q register creating a 40 bit operand, one or more to perform the shifting of that operand as required, and then one to return the Q register contents (the right-hand 16 bits) to the BRAM to complete the operation.

FIG. 5 illustrates the organization of the upper and lower bank control store PROMs 80 and 81 respectively, and more particularly the manner in which they are coupled or paired for maximum speed. Each of the banks of the control store includes one or more PROMs, each of which PROMs include a plurality of electronic chips. By way of example, the storage capacity of each bank may be 1024 (1K) addressable locations, each containing 96 bits, i.e., twenty-four chips of 1K each, each location having four bits of storage. To increase the speed of operation, each bank has at least two PROMs, with one PROM faster (lower address propagation time) than the other in each bank in order to compensate for propagation time in other elements included in the control store. Two banks are utilized primarily for efficiently accommodating branch decisions derived from the current firmware word and the next address generator 44. One bank is organized to be addressed directly by the firmware word whereas the other bank (for example, the upper bank) is organized to be addressed by the firmware word and/or a derivative function of the firmware word and various logic elements included in the CPU 100. Accordingly, the upper bank requires a multiplexed input and as shall be discussed, one of the two PROMs in the upper bank is provided with a faster address propagation time than any of the other PROMs in either bank. This is done as a practical matter by utilizing two PROMs, each with an increased speed of operation, i.e., lower address propagation time.

The purpose for pairing of the control store PROMs is to reduce the amount of time necessary in order to generate the next control store address, to select data appropriate to that address, and to enable it to the input of the main control store data register, the so-called MLR 82, which is the output register of the control store. This has been done in the past by using a single set of control store PROMs and selecting the address supplied to them. In such cases, the time required to propagate data through the control store logic is the address select time, i.e., the time from the change of the address input to the PROM to the time at which the output is stable. For typical control store PROMs, this is much longer than the so-called "enable time". Control store PROMs which have an enable input, can typically be turned on and off much faster (the "enable time") than addressed outputs can be provided. Therefore, the basic approach here is to separate the control store PROMs into two groups or banks so that the respective addresses for each group, for example, one of two branch addresses, can be propagating through the control store chips simultaneously, and so that the decision as to which address to utilize can be postponed until the very last moment when the decision is implemented by enabling the output of one or the other of the two groups or banks.

In the described CPU, as with some other computers, all so-called branch operations have a selection of one of a pair of addresses, one of which addresses is explicitly provided by the control store word in the current step, and the other of which is more or less indirectly generated from data available in the current step. In the instant CPU, this pair of addresses is constrained in such a way that all branches consist of selecting between two addresses, one of which is for the lower bank and the other of which is for the upper bank. The lower bank may utilize addresses in the range from 000 through 3FF (hexadecimal), and the upper bank control store addresses from 400 to 7FF (hexadecimal).

The structure used to implement this method of control store partitioning is shown in the block diagram of FIG. 5. As can be seen, the control store is partitioned into five sets of control store elements or PROMs 251 through 255. Elements 251 and 252, which, as indicated in the drawing in parentheses, include a total of 24 control store chips (19 in element 251 and 5 in element 252) are those that contain the data for the lower bank of the control store. Elements 253, 254 and 255 (which include a total of 29 chips) are those that contain the data for the upper bank. Three types of integrated circuits are used to implement these elements. Element 251 is made up of 19 1K (1024) .times. 4 bit control store chips with a typical address propagation time of 60 nanoseconds. Elements 252 and 253 are made up of specifically selected 1K.times.4 bit control store chips with a typical address propagation time of 50 nanoseconds. Elements 254 and 255 are made up of selected 512.times.4 bit control store chips with a typical address propagation time of 40 nanoseconds.

There are two reasons why chips or circuits with different propagation times are used in this logic. The selected parts in elements 253, 254 and 255 are selected to be faster than their counterparts in elements 251 and 252 in order to compensate for the propagation time through the upper bank address multiplexer 256. In addition, elements 252, 254 and 255 are selected to be faster than elements 251 and 253, respectively, because of the propagation delay required in the preliminary logic 259 that is in series with their output to the MLR 82.

The upper bank address multiplexer 256, which is included in next address generator 44, is used to provide one of the two addresses that are being selected among any firmware branch. The lower bank address (the address used by the lower bank control store PROMs) is more or less directly related to the next address which is in the control store word (the NA field) and is the address which is explicitly selected in that control store word. The other address which is used for a branch, that which is used in the upper bank, is some logical derivative of the NA field or is a logically generated address from PROMs or other logic in the CPU as shown by the three inputs to MUX 256, which are selected in accordance with the branch type. This other logic, i.e., the PROMs, and the other logic, and the upper bank address multiplexer which selects it, take additional propagation time which is not present in the address path in the lower bank. The difference between these two address paths is compensated for by the difference in the speeds of the control store chips in series with it. By utilizing one bank (i.e., the lower bank) which has a direct address path, only some of the PROMs need to have increased speed whereas if a MUX, such as MUX 256, was used for both banks, all PROMs would have to have increased speed in order to provide the same propagation time.

As has been noted the propagation delay up to and through the upper bank address MUX 256, which is generating address bits which are received by the faster PROMs of the upper bank, are uniformly faster than their lower bank counterparts. Thus, the data at the output of each PROM is stable at approximately the same time.

The preliminary logic 259 which is connected to the output of the fastest PROMs in each bank, that is elements 252, 254 and 255, is used to perform some preliminary logic decode of the next control store word prior to latching it in MLR 82. That is, this may include the select modify logic 53 which is used to generate the address inputs for the DRAM 11 and the BRAM 12, and which address inputs have to be available early in the control store cycle. More particularly, the address inputs cannot wait to be generated after the main clock has enabled the data into the MLR, but must be generated at the input to the MLR so as to be available as soon as the clock transition has taken place.

The source for the two addresses which are used for selecting a control store word in the lower bank and the upper bank are as follows. The lower bank address, NA, comes directly from the control store word for the current firmware step. Thus, it is available as soon as that control store word has been clocked into the MLR 82. The address NA is an input to both elements 251 and 252. The upper bank address, at the output of multiplexer 256, i.e., the address inputs to elements 253, 254 and 255, is a logical derivative of a number of logic functions included in the CPU. The address multiplexer may be the present embodiment generate eight different functional addresses which can be used by the upper bank. These are the result of the eight possible branch types which can be specified in the control store word.

These branch types are known as X0 and X1, XA, XB, XR, XE XW and XL as shown in the table of FIG. 6. Branch types X0 and X1, which are the most common types of branches, are basically a direct logical function of the NA field. Such branch types use the NA field, with the low order bit NA10 or NA(A) unchanged NA(A) or complemented NA(A). The other six branches are used for specific purposes. The XA branch is used for selecting the appropriate starting address to initiate the decode of a new instruction's opcode. The XB branch is used to perform the initial decoding of some of the address syllables which are possible in the CPU. The XR branch is used to select the starting address for the execution of the routine to read the operand in most cases, or for the execution of certain instructions which are capable of execution immediately. The XE branch is used to select among the starting addresses of the firmware routines which are used to execute the individual instruction algorithms. The XW branch is used to select one of a number of routines which are used to store the operand. The XL branch is used to allow a simplified firmware splatter under the control of the firmware. This branch uses the contents of the link register, which can be loaded under firmware control to control four bits of the upper bank address.

All of the major branches, XA, XB, XR, XE, XW and XL use the two high order bits (NAS 1, 2) of the NA field as their high order bits, but generate the remaining eight bits for their particular address in a different manner.

The XA splatter uses the output of five (5) XA PROMs to generate bits 3 through 10 of the XA next address.

The XB splatter uses three constant signals, two binary ones (11) and a binary zero (0), then one bit which is the output of one of the bit locations of the F register 38, and four outputs from a PROM which is driven by the F register to generate bits 7 through 10 of its NA field.

The XR splatter jams three bits with three binary zeros (000), generates the next bit with the contents of a control flop (KOP) whose state relates to instruction types, and generates the last four bits, bits 7 through 10, from an XR PROM.

The third bit of the XE splatter is the aforesaid instruction type signal. The next bit indicates whether or not bit 0 of the F register is zero, the next whether bits 1 through 3 of the F register 38 are all zeros or not, and bits 4 through 8 of the F register are utilized as bits 6-10 of the NA field.

The XW splatter gives a combination of signals: a constant zero and three outputs of a PROM which decodes and classifies operand size type fit bits 4 through 6. Bit 7 is a logic one, bit 8 is an output of the PROM which classifies operands as either addresses or non-addresses, and, finally two bits which are the output of an XW PROM, for bits 9 and 10 of the NA which identifies whether the result is to go to memory, to a base or data register or to a so-called K register (not shown).

These address lines for the XL branch use bits 1 through 6 of the NA field for the first six bits, and then use four bits of link register to generate the four last bits (bits 7 through 10).

These various inputs are selected by the set of eight-to-one multiplexers which are the upper bank address MUX 256. The particular address which is selected through this MUX is selected by use of the current control store word. In the current control store word, branching is performed by selecting one test condition and testing it for either being satisfied or not being satisfied. The output of this logic is two signals, one TCTRUE- or NOT TRUE, the other TCTRUE+ or TRUE. TCTRUE- is the signal on line 261 in the block diagram of FIG. 5 and TCTRUE+ is the signal on line 262. These signals are connected to the enable inputs of the respective control store PROMs.

At the beginning of a given firmware step, the NA field becomes stable and that address immediately begins propagating through the control store elements 251 and 252. As fast as the logic operates thereafter, the output of the upper bank address MUX 256 becomes stable, and that address begins propagating through control store elements 253, 254 and 255. Note that one of the address bits at the output of the MUX 256 selects between element 254 and element 255. This is because these elements are implemented with higher speed PROMs having a smaller word storage capacity. Accordingly, what is an address line for a 1K word (4 bits/word) PROM, becomes a selective enable line for two 512 word PROMs, with one enable input of PROM 255 coupled via inverter 264. More particularly, in elements 254 and 255, which are the selected 512.times.4 bit control store PROMs, the address bits are partitioned slightly differently. The reason for this is that a 512 word PROM only requires nine bits of address. The tenth address bit, which is an address bit to all the other control store PROMs, is instead used as a second enable on the 512.times.4 control store PROMs. That is, element 254 would be enabled for addresses 400 through 5FF, and since the inverse of that same address line is coupled to element 255, it would be enabled for addresses 600 through 7FF.

Thus, as fast as the addresses are available, they are propagated through the control store PROMs. Meanwhile, in parallel, the TC (test condition) logic 30, as shown in detail in FIG. 7, is deciding whether the test condition is satisfied or not satisfied, with the outputs TCTRUE- and TCTRUE+ becoming stable thereafter. If the test condition in the polarity indicated was satisfied, then the low state TCTRUE+ will enable PROMs 253, 254 and 255, while TCTRUE-, being in a high state, will disable PROMs 251 and 252. Thus, the outputs of the wired OR connections as shown by symbols 257 and 258 (76 and 20 lines coupled respectively) will be the contents of the addressed upper bank control store location. However, if the condition was not satisfied, then TCTRUE- (not TRUE) will have the other polarity or state and will enable control store elements 251 and 252, while TCTRUE+ will disable elements 253, 254 and 255. In such case, the outputs of the wired OR connections 257 and 258 will be the contents of the addressed lower bank location.

The enable time for the control store PROMs is typically 15 to 20 nanoseconds. This is significantly faster than the address propagation time for a PROM which, as noted on FIG. 5, may be, for example, 60 nanoseconds for PROM 251. Thus, the "delay" from the time when the test condition results are known has been enhanced so that the enable propagation time of the PROMs is controlling rather than the address propagation time through the PROMs. The reason that these particular concerns have been taken with the timing through this chain of logic, i.e., the propagation time through the upper bank address generation and selection, the timing through the control store PROMs, and the timing through the preliminary logic, is that the worst case time at which the data at the input to the MLR 82 must be stable, is the time at which the main clock is triggered. It is one of the major functions controlling the cycle time, and thus the speed of the CPU.

If each of the control store PROMs had a 60 nanosecond propagation time, then in the worst case, if one were selecting an upper bank address, it would take 80 nanoseconds altogether from some preliminary starting point before the input to the MLR 82 became stable. Thus, to increase the speed, the use of PROMs having three different propagation times is desirable. PROM 252 compensates for the reduced speed or slow down caused by the preliminary or select/modify logic 259. PROM 253 compensates for the increased delay in addressing due to the upper bank address MUX 256. PROM 254 and 255 compensates for both the MUX 256 and the logic 259. Thus, the worst case time through this network is almost the same no matter which bank, upper or lower, is being used, and no matter which output is generated, one that goes directly to the MLR 82 or one that goes to the logic 259, before going to the MLR. If elements 253, 254 and 255 were not faster than elements 251 and 252, the clock period would have to be longer for any firmware step in which the upper bank was the one selected. Thus, the cycle time for that control store step would be longer because there would have to be allowance for the extra time for the next address PROMs and the upper bank address MUX to generate a stable address for the control store PROMs.

The basic principle involved in the logic shown in FIG. 5 is summarized as follows. Since the address path through the control store PROMs is so much slower than the enable path, the address path is always left open, that is, the next address is trickling through the lower bank control store and the output of the address MUX 256 is trickling through the upper bank of the control store, while the decision is being made as to which bank is finally going to be used. The timing of that decision becomes critical only when it becomes time to enable the outputs of these control store PROMs, in which case the test condition generated, true or false, will enable either the lower bank or the upper bank PROMs, depending upon which way the branch is going. The enable path is a much shorter time, 60 nanoseconds for the address propagation as compared to 15 nanoseconds for the enable propagation, thus allowing the decision for bank selection to be postponed for as much as 45 nanoseconds in a typical control store step, and thereby allowing each control store step to be significantly faster.

Now referring to FIG. 7, details of the test logic 30 for generating the TCTRUE+ and TCTRUE- signals is shown. The TCTRUE+ signal is generated at the negation output of multiplexer (MUX) 302 on line 302 on line 262, and the TCTRUE+ signal is generated at the negation output of MUX 304 on line 261. The lines 261 and 262 are coupled to the PROMs as shown in FIG. 5. The multiplexers 302 and 304 each have eight inputs (0-7), with each like numbered input coupled to receive the same input. Such inputs are received from the eight outputs of multiplexer 300 which actually includes eight multiplexers MUX 1-MUX 8, each having eight inputs and one output. The sixty-four inputs to multiplexer 300 are each coupled to receive a test condition from various functions included in the central processing unit 100 as indicated by test logic 30. Depending upon the condition of the function tested, either the upper bank or the lower bank of the control store will be enabled and accordingly addressed. The inputs to the multiplexers which are selected and/or enabled are determined by control bits received from the current control word, i.e., firmware word from the control store.

Thus, it is desirable to branch on a variety of different signals, which may, by way of example, be sixty-four in number, such that only one signal represents the selected one of the sixty-four inputs. The logic of FIG. 7 provides this capability with minimal propagation time by use of two levels of multiplexers, one level comprising multiplexer 300 and the other level comprising multiplexers 302 and 304. The logic of FIG. 7 also enables such branching based on either of two polarities of the same signal.

The operation of the logic of FIG. 7 is as follows. One input to each of the eight multiplexers included in multiplexer 300 is selected for propagation to the negation outputs, GPO through GP7 respectively by the three bits received at select inputs 3, 4 and 5 of each of such eight multiplexers. These three bits are received from the TC field (bits 13-15) of the firmware word. The outputs GPO-GP7 are respectively coupled to the 0-7 numbered inputs of each of the multiplexers 302 and 304. The multiplexers 302 and 304 are also coupled to receive at their enable inputs the signals CRTCSP- and CRTCSP+, respectively. The CRTCSP signals (assertion and negation) are also received from the current firmware word, and more particularly, the TP field in the firmware word, transferred via the register 82. For bit 21 (the TP field), the element at the bit 21 location in the register 82 is actually a bistable element (flip-flop) having both a negation and an assertion output. Depending on the logic level required for other bits of the firware word, either or both the assertion and negation outputs are used.

The multiplexers 302 and 304, which may be purchased from a semiconductor vendor such as Texas Instruments Inc. under part number SN74S251, will allow switching of the assertion and negation outputs of the multiplexer if the signal on the enable input is low or a binary zero. If such signal on the enable input is high, the outputs are floating and accordingly, as configured, such multiplexer having a high enable input will be effectively removed from the logic of FIG. 7. Thus, as can be seen, only one of the multiplexers 302 and 304 will be enabled at any given time. The outputs of the multiplexers 302 and 304 are coupled so that the assertion output of each multiplexer is coupled to the negation output of the other multiplexer to produce a so-called wired-OR circuit connection. Thus, with either of the multiplexers 302 and 304 enabled, the TCTRUE signals are obtained with the proper polarity. The TCTRUE signal which is low or a binary zero, will enable the bank to which it is coupled. That is, if TCTRUE- is low, the lower bank will be enabled.

Thus, as can be seen from the operation of the logic of FIG. 7, if the CRTCCSP- signal is high and accordingly the CRTCSP+ signal is low, the MUX 304 is active or enabled and the signal TCTRUE- reflects the level (high or low) of the one of sixty-four conditions being tested and TCTRUE+ reflects the inverse of such level. If the CRTCSP- signal is low and accordingly the CRTCSP+ signal is high, then MUX 302 is active or enabled and the signal TCTRUE+ reflects the level of the condition being tested and TCTRUE- reflects the inverse of such level. Whether the upper or lower bank of the control store is enabled depends upon which of the signals TCTRUE- or TCTRUE+ is low.

It can be seen that the opcode of an instruction read from memory 106 may be utilized to select one of several test conditions, which test conditions may be different from those received by MUX 300. In such case, the opcode bits could be utilized to select the test condition inputs to a first MUX similar to MUX 300 or a subportion thereof, and two multiplexers similar to multiplexers 302 and 304 could also be used in an arrangement similar to that shown in FIG. 7. It can also be seen that the use of the opcode to so select a test condition could be used independently of or in parallel with the logic of FIG. 7. If used in parallel, then the multiplexers 302 and 304 controlled by the firmware bits would have to be disabled when the multiplexers controlled by the opcode are enabled and vice versa.

Details of the MUX 256 and the connections thereto are shown in FIG. 8. More particularly, the MUX 256 is shown to include ten multiplexers each with eight inputs. The inputs to these ten multiplexers correspond to the signals indentified in the table of FIG. 6. These signals are coupled for the eight branch types depicted in fig. 6, such that the first signal for each of the branch types is received at the corresponding eight inputs of MUX 1 included in multiplexer 256. Thus, MUX 1 has coupled to each of its eight inputs the signal NA1. MUX 2 also has the same signal (NA2) coupled to each of its inputs. For each of the other multiplexers of MUX 256, except for the first two inputs, the signals coupled to the input are, for the most part, different. For example, MUX 3 has coupled to its number three input, the signal XA3 which, as shown in FIG. 6, is the third bit of the upper bank address for the XA branch, such third bit being a bit from position 3 of a so-called XA PROM which is not shown, but may simply be a decoder coupled to receive its primary inputs from the F register 38 and whose output is coupled for receipt by MUX 3 of multiplexer 256.

The other inputs of the remaining multiplexers of MUX 256 also receive the inputs as depicted in FIG. 6. The first two inputs of MuX 10, NA(A) and NA (A) are of further interest. More particularly, the use of these two bits, one the complement of the other, allows more flexible and efficient pairing of control store (PROM) locations for branch operations. In the prior art, it has been known to branch on a branch hit condition to an odd PROM location and to branch on a branch no-hit condition to an even location, or vice versa. However, such prior art technique has limitations. For example, assuming four consecutive locations whose addresses are XXX000, XXX01, XXX10 and XXX11 (where X is either a binary 1 or a binary 0), a sequence which is branching can go to address XXX00 or XXX01 depending on a no-hit or hit condition, or another sequence can go to address XXX10 or XXX11. However, even if locations XXX11 and XXX01 contain exactly the same information, they cannot be shared, i.e., two locations have to be used even for the same contents. This is because both addresses are odd and the only pairing allowed is for odd/even addresses. In another prior ar technique, a no-bit condition could be made to cause the addressing of the location specified by the control store word and a hit condition could cause the addressing of a location specified by the higher order address bits with the two lowest order bits, for example, both being binary ones. In this manner, and address whose two lowest order bits were both binary ones, could be paired with addresses whose two lowest ordered bits were either two binary zeros or a binary one and a binary zero (in either order). However this restricted the common address to an address which had its two lowest ordered bits binary ones (or vice versa, i.e., two binary zeros with appropriate change of the other paired addresses to one whose two lowest ordered bits are either both binary ones or a binary one and a binary zero).

In a data processing unit which allows a variety of sources to generate or form the next address for the control store, the use of the arrangement as shown in the Figures, more specifically FIGS. 5 and 8, is desirable. More particularly, the logic shown in such Figures allows a reduction in the total number of firmware or control store locations by allowing all locations to be used as alternates to an increased number of selected locations. In order to accomplish this, the least significant bit position (NA10) is coupled so that for the X0 branch, NA10 is in fact the same as the NA10 bit, NA(a), received from register 82 for the current control store word, whereas, for the X1 branch, the complement of such bit NA(A), is used.

By way of example, FIG. 9 should be reference. If the NA address is as shown on the first line, then the lower bank address is, as shown, the same. The upper bank address is however dependent upon whether there is an X0 or X1 branch. If an X1 branch, then the upper bank address is shown in the third line. The final address depends upon whether the signal TCTRUE- is a binary zero or one. If a binary zero then the lower bank PROMs will be enabled. Thus, as shown in the fourth line, the final address will be for the lower bank if TCTRUE- is a binary zero. If TCTRUE- is a binary one, then the final address will be for the upper bank and depending upon whether the branch is X1 or X0, the final address will be as shown in lines 5 and 6 respectively of FIG. 9. More particularly, it can be seen that the lower bank address on line 4 may be paried with either of the addresses for the upper bank as shown on lines 5 and 6.

In this manner, more efficient pairing of locations may be obtained. Variations of the example shown in FIG. 9 will illustrate that an even address location of the lower bank may be paired with an even or odd address location of the upper bank, and that an odd address location of the lower bank may also be paired with either an even or odd address location in the upper bank.

It can be seen that the above described pairing could be accomplished by setting the binary value of the NA10 bit independent of the binary value of the NA10 bit of the current firmware word. For example, the NA10 bit for the X0 branch may have been set to a binary one instead of NA(A), so long as the NA10 bit for the X1 branch is its complement, in this case, a binary zero.

2. The Stack Mechanism

Having described in detail embodiments for a suitable central processing unit and control store for implementing the stack mechanism of the instant invention, the following will set forth additional details of the CPU, illustrative examples of stacks implemented by the stack mechanism of the instant invention, and a detailed discussion of the operation of the stack mechanism.

The stack of the instant invention is resident in one of memory modules 106 as a reserved memory area for use in system operation. As herein embodied, the stack comprises M memory locations subdividable into a plurality of variable lengths of stack frames. Each stack frame is allocated on an "as needed" basis and corresponds to a procedure executable within a segmented or modular program. At any one time, there may be zero stack frames or one or more stacks frames. The total number of memory locations currently occupied by all of the operands in the existing stack frames comprises the current memory allocated C.

As will hereinafter be discussed, the stack mechanism of the instant invention includes the ability to extend and contract the length of a stack frame after it has been allocated and to address locations within a particular stack frame by addressing relative to either the top of the stack frame or the bottom of the stack frame. The length of a stack frame is dynamically changed by pushing an operand into the frame, popping or operand off the top of the stack frame or by the stack mechanism's modifying the number of memory locations allocated to the stack frame.

Each stack in memory includes a stack header comprising two words of main memory. These words are words N and N-1 of the stack and correspond to the previously mentioned maximum stack area length, M, and current frame allocated, C, respectively.

The stack mechanism of the instant invention will obtain a stack frame from memory responsive to the execution of an ACQUIRE (ACQ) instruction. In the ACQ instruction, the user specifies the length of the stack frame to be created. The stack mechanism will dynamically enlarge or reduce the size of a stack frame in response to a modify frame length (MFL) instruction. The stack mechanism will release a stack frame to memory in response to a RELINQUISH (RLQ) instruction.

The length of a stack frame can also be modified by this stack mechanism in response to push or pop address syllables.

In FIG. 10, the registers with comprised DRAM 11 and BRAM 12 are set forth in detail. The DRAM comprises four sets of registers with eight registers in each set. These sets are DO-D7, K0-K7, E0-E7, L0-L7. Registers E3 and E4 relates specifically to the stack mechanism of the instant invention. Register E3 is the stack array register (STKA) and it stores a designation of a memory address where an array of pointers to stacks existing in memory is stored. Register E4 is the stack pointer register (STKP) and it stores a designation of a memory address for storing the current number of allocated storage positions in the current stack.

The BRAM contains registers B0-B7 and A0-A7. Register A4 is the stack top register (STKT) and it stores a designation of the address of the top data word in the current stack frame. Similarly, register A5 is the stack bottom register (STKB) and it stores a designation of the address one below the first operand in the current stack frame.

FIGS. 11(a)-(e) illustrate five representative possible combinations of stack frames within a stack. The first two locations in an existing stack contain the M word for storing an indication of the maximum size of the stack and the C word for storing a designation of the number of stack locations currently allocated. An attempt to increase the number of currently allocated words to an amount greater than the maximum number of words permitted in the stack by either increasing the number of stack frames or by increasing the size of existing stack frames will result in a trap or error condition designated stack overflow.

The stack of FIG. 11(a) contains three frames, F.sub.0, F.sub.1, and F.sub.c. Frame F.sub.0 contains i entries and includes a b-word, b.sub.o, for storing a designation of the number of entries in frame F.sub.0.

Similarly, F.sub.1 contains j entries and has associated with it a b-word, b.sub.1 for storing a designation of the number of entries in F.sub.1.

Frame F.sub.c is the current stack frame which means that it is the most recently established stack frame. As illustrated herein, F.sub.c contains K entries and the b-word associated with f.sub.c stores the amount K.

Frame F.sub.c, being the current frame, dictates the values to be stored in the STKT register and the STKB register. Entries are made into the current stack frame F.sub.c such that the first entry in time is stored in location K (K=number of entries in the stack) the second entry in location K-1, and the third entry in location K-2, and the most recent entry in location O. The STKT register contains the address of location 0 of stack frame F.sub.c and the STKB register contains the address of location one below or greater than location K in F.sub.c, i.e. the address of b-word, b.sub.1. As illustrated herein, the total number allocated entries in the stack is less than the maximum stack size (M) and, therefore, no stack overflow exists and either stack F.sub.c can be dynamically increased or an additional stack frame can be added to the stack.

FIG. 11 (b) illustrates a stack containing two stack frames wherein stack frame F.sub.0 contains i entries and the current stack frame F.sub.c contains zero stack entries. Thus, the b-word associated with stack frame F.sub.0 is equal to i and the b-word associated with the current stack frame is equal to zero (b.sub.c equal to zero). Since there are no entries in the current stack frame, the addresses stored in STKT and STKB will be the same and equal to an address one less than the address of the b-word b.sub.c associated with the current stack frame F.sub.c. Thus, STKT and STKB will contain the address of the b-word, b.sub.0, associated with frame F.sub.0.

FIG. 11(c) illustrates a stack containing one non-empty stack frame which is the current stack frame F.sub.c. F.sub.c has K entries and, therefore the b-word associated with F.sub.c will store the value K. The STKT register of the stack mechanism stores the addresses of location 0 of the stack frame F.sub.c and STKB stores an address one greater than the location b.sub.c in stack, frame F.sub.c. Thus, STKB stores the address of the C word and, therefore, will duplicate the contents of the STKP register.

FIG. 11(d) illustrates a stack wherein there is one stack frame allocated and that stack frame has the zero entries.

Finally, FIG. 11(e) illustrates a stack wherein there is no current stack frame and no stack frames an allocated. In this case, the contents of the c-word will be equal to zero, the STKP register will store a designation of the address of the c-word. The STKT and STKB registers will contain identical contents equal to the address of the M-word of the stack.

FIG. 11(f) illustrates a situation wherein there is no stack. In this case, the STKP register will contain a null address (=000000) and the STKT register an address of 1 (=000001).

FIGS. 12(a) and (b) illustrate a stack before and after additional elements have been added to it, i.e., before pushdowns. The stack of FIG. 12(a) includes two stack frames, frame F.sub.o and the current stack frame. Stack frame F.sub.o includes b.sub.o entries and the data stored in the b-word associated with frame F.sub.o will be equal to i. The current stack frame includes four entries E1-E4. E1 was the first entry in the current stack frame, E2 the second entry, etc. As indicated, entry E4 is stored in location zero of the current stack frame, E3 in location 1, E2 is location 2, and E1 in location 3. The STKT register stores the address of location zero in the current stack frame and STKB stores the address one below the highest location (location 3) in the current stack frame and, therefore, STKB will contain the address of the b-word of stack frame F.sub.o. The b-word associated with stack frame F.sub.c contains the quantity 4 indicating that there are four entries in the current stack frame. The STKP register contains the address of the c-word for the stack. At that address, the quantity (i=6) will be stored reflecting the total number of entries in the stack.

FIG. 12(b) illustrates the stack of FIG. 12(a) with the addition of five entries into the current stack frame. The contents of STKP and STKB remain the same as does stack frame F.sub.o and the b-word associated therewith. In the current stack frame, however, five additional entries have been made, i.e., E5-E9. The STKT register stores the address of location zero in the current stack frame and the b-word associated with the current stack frame will store the value nine as an indication of the number of entries in the frame. Thus, FIG. 12(b) illustrates the dynamic expansion of the current stack frame from the state illustrated in FIG. 12(a).

The elements in the current stack frame of FIG. 12(b) are accessible by addressing relative to the address stored in the STKT register or the address stored in the STKB register. This addressing relative to the top or bottom of the stack frame is controlled by the sign of the displacement. As an example, entry E7 stored in location two of the current stack frame can be accessed from the top of the stack frame by displacing from the address stored in STKT by a positive 2. Similarly, entry E7 can be accessed relative to the bottom of the stack frame by displacing relative to the address stored in the STKB register, i.e., the address of the b-word by a negative 7. As will hereinafter be explained, the stack mechanism of the instant invention executes displacements within the current stack frame relative to the content of the STKT register of the STKB register depending upon whether the sign of the index quantity is positive or negative, respectively.

Considering FIGS. 11(a)-11(e) and FIGS. 12(a) and (b), it is evident that adding to or subtracting from the number of entries in the current stack frame and creating and deleting stack frames requires the constant updating of the addresses stored in the STKT register and the STKB register. Also, the quantities stored in the C word as addressed by the STKP register and the value stored in the b-word associated with the current top stack frame must be continually updated. Such updates are performed in the following manner.

3. Description of Stack-related Macroinstructions

The stack mechanism of the instant invention operates in response to five different stack macroistructions. These instructions are set forth as FIGS. 13(a)-13(e). Each of these instructions is two words in length with each word comprising 16 bits, i.e., bits 0-15. The first word for each instruction is identical, having hexadecimal value 0010. The second word in each of the instructions is different and provides means for distinguishing among the macroinstructions.

FIG. 13(a) illustrates the format of the LDT macroinstruction. The function of LDT is to update the contents of the c-word and b-word of the current frame, reload the STKT register from the base register B.sub.m, and the reload the STKT and STKB registers to point to the new top frame.

The store stack pointer macroinstruction (STT), illustrated in FIG. 13(b), causes the contents of STKP register to be transferred to register B7 in the BRAM.

FIG. 13(c) illustrates the instruction format for the acquire (ACQ) macroinstruction which creates a new stack frame having a certain number of entries. The M field specifies which B register in the BRAM stores an indication of the number of entries to be included in the acquired stack frame and the field N specifies which B register is to store an indication of the top of the newly-created stack. This B register, as specified in the N field, functions as a temporary STKT register. Together the M field and N field comprise what is designated an address syllable.

FIG. 13(d) illustrates the instruction format for the relinquish (RLQ) macroinstruction wherein the N field designates which B register within the BRAM stores the address of the top of the stack to be relinquished.

FIG. 13(e) illustrates the instruction format for the modify field length (MFL) macroinstruction which, when executed, causes the size of the current stack frame to be increased or decreased. The M field specifies the D register in DRAM which stores an indication of the number of words by which the frame size is to be modified. If the quantity in the D register is negative, the size of the current stack will be reduced, whereas if the quantity is positive the size of the current stack will be increased.

FIG. 13(f) relates to stack operations performed in response to different address syllables and the meanings of the format of FIG. 13(f) will be discussed further in this specification.

4. Detailed Description of System Operation

The following will describe in detail the operation of the stack mechanism to perform both indexing within a current stack frame and to execute the above-described macroinstructions. In the preferred embodiment, the control store 80 & 81 (FIG. 2) includes a sequence of firmware instrustion which are executed by the stack mechanism to perform stack indexing and macroinstruction execution. A listing of exemplar firmware for use by the stack mechanism is attached as an Appendix to this specification. In explaining the operation of the stack mechanism as illustrated by the flowcharts of FIGS. 14-23, the line numbers of the firmware instructions pertinent to the functional operations illustrated in the diagrammed steps of the flowcharts are written next to the flowchart boxes. This will enable one of ordinary skill in the art to practice the invention having been apprised of the structure of the stack mechanism by means of the foregoing description relevant to FIGS. 1-13.

As discussed above, displacement within a current stack frame can be relative to either the top of the stack frame or the bottom of the stack frame. When indexing relative to the top of the stack frame, a positive displacement is added to the values stored in the STKT register and when indexing from the bottom of the stack frame, a negative value is added to the content of the STKB register. As illustrated in FIG. 14, displacement addressing within a stack frame is initiated by doing a one-word procedure fetch to obtain from memory the value of the displacement. The procedure word is transferred into the procedure data multiplexer 40, signed extended and admitted to the inner (I) bus. From the I bus, it is transferred via the DALU into register E0 (DRAM) where it is stored. During the transfer, bit 16 of the displacement, which is the sign bit, is tested for ONE or ZERO. A ZERO indicates a positive or zero displacement and the left branch of a flow diagram of FIG. 14 will be taken. If the displacement is negative, as indicated by the sign bit being equal to ONE, the right branch of FIG. 14 will be taken.

If the displacment is positive, then it must be added to the content of the STKT register. This is accomplished by adding the contents of E0 to the contents of STKT and storing the result in E0. An address overflow test is performed to see whether a carry resulted from addition (which would indicate that the resulting address had overflowed or lost significance). This is accomplished by the contents of the E0 register being enabled to the R bus and the contents of the STKT register being enabled to the I bus. An addition is performed in the DALU and the result is stored back into E0. A trap or firmware error routine will be executed if a carry occurred due to that addition.

If no carry occurred, then a test is performed to see whether the resulting address is stored within the current frame. This is accomplished by transferring the contents of the STKB register to the I bus, transferring the contents of register E0 to the R bus, and conditioning the DALU to subtract the I bus from the R bus. Once again, the carry bit is tested and if a carry occurred, then the address stored in E0 is larger than the address stored in the STKB register and an attempt has been made to reference out of the current stack frame. A trap will be performed under this condition.

If no carry occurred, then the contents of register E0 is transferred, via the R bus, W bus, and BALU, into the Y register and the next operand is read. The Y register serves as the read/write address register for the CPU of FIG. 2. It is transferred to the local bus 9 by means of the address multiplexer 25. If a carry has resulted, then the address in E0 must be compared to the address in the STKT register to make sure that reference is not being attempted outside of the current stack frame. This test is performed as follows. The contents of the STKT register is admitted to the I bus and the contents of the E0 is admitted to the R bus. The DALU is enabled to subtract the I bus from the R bus and the carry output is tested. If no carry has occurred, then the address stored in E0 is smaller than the address stored in the STKT register, and addressing will occur outside of the current stack frame. This will again cause the transfer to a firmware error routine. If a carry does occur, then the address in E0 is within the bounds of the current stack frame and will be admitted to the Y register by means of the R bus, W bus and BALU to provide the address for an operand read.

The following will be a description of the operation of the stack mechanism in response to the five macroinstructions set forth in FIGS. 13(a)-13(e). As illustrated in these figures, the generic instruction, i.e., the first word of each of the macroinstructions is identical and the instructions are distinguished from each other by means of the contents of the second instruction word. This process is accomplished by first initiating a procedure word fetch to acquire this second data word. A word of procedure is thereby retrieved from memory and transferred into register E1 of DRAM 11. Similtaneously, the contents of register E1 are enabled to R-bus 13, shifted right four positions, and transferred into the AS' register 42 by means of selector 20, O-bus 17, shifter 19, and Write (W) bus 84.

The contents of the AS' register are transferred to select-modify (SEL) unit 53 through register 38. The SEL register 53, therefore, contains bits 8-11 of the second word of the instruction. The contents of E1 are again transferred to the AS' register so that the AS' register contains bits 8-15 of the second word of the instruction. Next, the m field of the instruction as stored in the SEL register selects one of the B registers and the contents stored in the selected B register are transferred to the Y register. Simultaneously, the M field as stored in the SEL register is employed to select a D register and the DRAM. The contents of this register are transferred to Q register 50. In the case of an LDT macroinstruction SEL will be equal to 1-7 and, therefore, registers B1-B7 will be selected. Similarly, in the case of an ACQ instruction one of the registers D1-D7 will be selected. Registers D1-D7 are also employed by the MFL instruction.

Following the selection of the B and D registers indicated by the m field, the same procedure is performed using the n field to select BRAM and DRAM registers.

The next task is to determine what type of macroinstruction has just been read. First, the contents of register E1 are copied into register A0 and a logical AND is performed with the constant FF88 in the DALU. The result is stored in E1 and enabled through the write bus where it is zero tested. This zero test separates the MFL macroinstruction from the LDT, STT, ACQ, and RLQ because only the MFL instruction will have bit 12 equal to ONE.

If the identified instruction is an LDT, STT, ACQ, or RLQ, the constant FFF0 will be enabled to the D bus, the contents of register A0 will be enabled to the R bus, and the DALU will add the R bus and I bus together and store the contents in the A0 and Y registers. If no carry is generated, the instruction is either an RLQ or an STT. Finally, another check will be made of the SEL register for zero or non-zero. Since the n field of the STT instruction is always zero and the n field of the RLQ macroinstruction is always non-zero this test will identify whether an RLQ or an STT instruction has been read.

If a carry was generated, then the instruction just read is either an LDT or ACQ. The n field of an LDT macroinstruction is always equal to zero and the n field of an ACQ macroinstruction is always non-zero. Thus, a test of the SEL register will identify the instruction as either an LDT or an ACQ.

The top of FIG. 16 illustrates that the only step performed in response to an STT macroinstruction is to copy the contents of the STKP register from the DRAM into the BRAM register B7. After this operation, the next address will be calculated and a fetch of the next instruction will be performed.

The contents of the STKP register are transferred to the BRAM by enabling the contents of the STKP register to the R bus and the R bus to the Write bus 84. From the Write bus, the contents are transferred into register B7 of BRAM 12.

The functionality for performing an LDT macroinstruction is set forth in the procedure illustrated at the bottom of FIG. 16 and in FIG. 17. As stated above, the LDT macroinstruction necessitates an update of the c-word of the current stack, the b-word of the top frame of the current stack, the reloading of the STKP register, and the loading of new values, pointing to the new stack, into the STKT and STKB registers. The firmware instructions performed in response to an LDT macroinstruction begin at line number 11,900 and are common to three different macroinstructions. The LDT macroinstruction will enter this firmware with the n field and link field equal to zero, the LEV instruction, which does not comprise a part of this invention but is employed to monitor the loading of stacks, will have the n field equal to zero and link field equal to six, and the ACQ macroinstruction will have n equal to 1-7 and link equal to zero. The different values for the n field and link field cause each of these three macroinstructions to generate different results from the same firmware instructions.

In response to the LDT macroinstruction, the contents of the STKT register are enabled to the R bus and decremented and shifted, as indicated in FIG. 16, and if there is a stack, the operations in FIG. 16, and particularly in lines 11903-907 of the Appendix, are executed.

Next, the contents of the STKP register are copied into the Y register and the difference between the contents of the STKP register and the STKB register is calculated in the DALU and transferred to the write register through the R bus, O bus, and write bus. The contents of the write register then reflect the new value for the c-word of the old stack and the Y register contains the address of the c-word in memory. A memory write is performed and at this point the b-word and the c-word have been updated in the stack as stored in main memory.

Next, the contents of register A0, which contains the new stack pointer, are copied into the STKP register. Simultaneously, this value is transferred via the R bus to the O bus where it is tested for ZERO. This test is to determine whether or not there is a new stack as will be indicated by the new contents of the STKP register being zero or non-zero, respectively.

If no stack exists the contents of the STKP register are incremented by one and stored in the STKT register. The values of these two registers then reflect the no stack condition as shown in FIG. 11(f) and a branch is then made to the XA procedure for calculating the address of the next instruction.

If there is a stack, STKP will not be zero and the contents of the STKP register are copied into the Y register. In addition, the contents of the STKP register are incremented by one and stored in the STKT register. The Y register 24 now points to the c-word for the new stack.

Next, the stack mechanism, in executing an LDT macroinstruction, reads the new c-word. Simultaneously, the contents of the Y register are transferred into the STKB register and the value of the link field in register 38 is decremented by one. As previously explained, the link value for an LDT macroinstruction is initially zero and, therefore, decrementing it by one causes the link field in register 38 to be equal to a hexadecimal F (minus 1 modulo 16). A stall is then executed until the requested data is transferred from memory. The returning data is subtracted from the contents of the STKT register and stored as the new value for the STKT register. Since the STKT register contained the value of the stack pointer plus one, the difference between the contents of the STKT register and the just-read c-word is the correct value for addressing the new top frame of the new stack and, therefore, it is stored in the STKT register. At this point the STKP register and STKT register contain the correct contents for the new frame in the stack.

The new contents in the STKT register are decremented by ONE and stored in the Y register so that the Y register now contains the address of the b-word for the new stack. A read of the b-word is performed and the returning data is sign extended and enabled to the I bus. The contents of the STKB register is enabled through the R bus and is incremented by the sum of ONE and the value of the b-word on the I bus. The result is stored in the STKB register. Since the previous contents of the STKB register were equal to the contents of the STKT register, adding the returning b-word and ONE results in the new value stored in the STKB register being the address of the first location below the current frame in the stack. At this point, the contents of the STKP, STKT, and STKB registers have all been updated and the LDT macroinstruction fully performed.

Referring to FIG. 18 and 19 the following firmware instructions are performed by the stack mechanism in response to an ACQ macroinstruction. First, the size of the frame to be acquired, as stored in register AO, is incremented by 1 and stored in the AO register. Next, bit 16 of this incremented quantity is tested for 0. If it is not 0, then the quantity is negative and an attempt is being made to acquire a new stack frame with a negative number of locations. This is a detectable programming error and a forced branch is made to a firmware routine for reporting the error.

IF the size of the acquire was positive, i.e., a positive number of words has been designated for the size of the new current stack frame, a copy of the contents of the STKP register is stored in the BO register and in the Y register. At the same time, the transferred value is tested for ZERO, and if it is ZERO, an acquire is being attempted for a non-existent stack. In such a case, a branch is made to the firmware routine for reporting a detectable programming error.

If the stack does exist, a memory read is initiated to obtain the C word and the M word for the current stack. Since the address register, Y, contains a copy of the contents of the STKP register, this read will return the C and M words.

It must be determined, when performing an acquire, whether the adding of the acquired frame to the stack will result in a stack overflow. This is accomplished by transferring the contents of the AO register to the EO register in the DRAM, storing the contents of the Y register into the AO register of the BRAM, and incrementing the value of Y. Register EO now conatins a value which is one more than the size of the acquire, register AO contains a copy of the contents of the STKP register, and the Y register stores the address one less than the address stored in the STKP register.

Next, the contents of the STKT register are enabled to the I bus and the contents of the EO register are enabled to the R bus. A subtraction of the I bus from the R bus is performed in the DALU and the contents are stored in the EO register. A stall is executed until the requested data is returned. Simultaneously, the contents of the AO and EO registers are added together and stored in the AO register. This places the new value of the C word into BRAM register AO.

The returning data is sign extended and admitted to the I bus. The contents of the AO register are admitted to the R bus and a substration of the I bus from the R bus is performed by the DALU. This is a comparison of the C word for the stack with the M word of the stack and, if the new c word is larger in value than the M word, a carry will occur in the substraction and a branch will be forced to a firmware trap routine to report a stack overflow. If no stack overflow occurs, then the C word is less than or equal to the M word and this acquire can be satisfactorily performed. The contents of the BO register are complemented and a branch is made to the previously explained firmware associated with the LDT macroinstruction.

As explained above, the performance of the firmware associated with a LDT macroinstruction will be different when it is being performed in response to an ACQ macroinstruction. When it is being performed in response to an ACQ macroinstruction, it is required to update the b-location for the current frame and the C word for the stack. The new b-word associated with the acquired stack frame must be written into memory. At the point of entry of the LDT firmware from the ACQ firmware, register EO contains the new value of the stack pointer to be stored in the STKT register.

Upon entering the LDT firmware, the contents of the STKT register are complemented and stored in the E2 register of the DRAM. This value is also decremented by 1 and copied into the Y register with the result that the contents of the Y register comprise the address of the b-word for the previous top frame. The value for the b-word of the previous top frame is then computed by subtracting from the contents of the STKB register the contents of the STKT register and storing the difference in the write data register 28. This is accomplished by enabling the contents of the STKB register to the I bus, enabling the contents of the E2 register (the complement of the contents of the STKT register) to the R bus, adding the R bus and the I bus together plus 1. This sum is enabled to the output data register 28 through the DALU and Write bus. A memory write is performed and the b-word of the previous top stack frame is updated.

Next, the C word of the stack must be updated. The C word takes on the value of the contents of the STKP register minus the contents of the BO register minus 1. This calculation is made in the following manner. After writing the b-word, the contents of the Y register are stored in the STKB register. The contents of the EO register are enabled to the R bus and decremented by 1 and stored back in EO. Next, assuming that the stack exists, the contents of the BO register are enabled to the I bus and the contents of the STKP register are enabled to the R bus. The I bus is subtracted from the R bus by the DALU and enabled through the Write bus to the output data register 28. Again, a memory write is performed and the address register (the Y register) contains the address of the C word in memory. Following the update of the b-word associated with the previous top stack frame and the C word for the stack, the b-word for the current stack frame just acquired must be written into memory. The b-word associated with this current stack frame is equal to the sum of the contents of the STKB register, the EO register and 1. Since EO contains the complement of the new contents of the STKT register, this calculation is in effect equal to one less than the difference between the contents of the STKB register and the STKT register.

The calculation is performed as follows. The contents of the STKB register are enabled to the I bus and the contents of EO are enabled to the R bus. The R bus and I bus are added together by the DALU and enabled to the Write bus. Next, the contents of the EO register are enabled to the R bus and complemented by the DALU. The output of the DALU is enabled to the Write bus and stored in the STKT register. Next, the BALU performs a subtraction of 1 from the value on the Write bus and the output of the BALU is transferred to the Y register and a memory write is performed. The contents of the EO register are enabled to the R bus and complemented by the DALU. The output of the DALU is enabled to the Write bus and stored in register B1-B7, as controlled by the contents of the FN register 38.

This ends the firmware instructions performed in response to an ACQ macroinstruction. The STKT register contains the address of the most recently stored operand in the current stack frame and the Y register contains the address of the b-word for the newly-acquired current stack frame.

The RLQ macroinstruction when performed by the stack mechanism of the instant invention, removes the current stack frame from the stack. This is performed as follows. First, with reference to FIG. 20, a memory read is performed to access the b-word of the frame below the current stack frame. At the same time, a test is conducted to see whether a frame exists in the stack. This is accomplished by calculating the difference between the contents of the stack pointer register STKP and the stack top pointer register, STKT. If no carry occurs from this subtraction, that means that the C value for the stack is 0 and no frames exist therein. Under such conditions a branch is made to a firmware trap routine. If a carry has occurred, then another test for 0 conducted to determine whether the stack exists. If the contents of the STKP register are 0, then the firmware instructions for updating the C word and the stack registers will not be executed.

Assuming that a stack does exist, the new value of the C word must be calculated and stored in the STKB register and the write register. This is accomplished by enabling the contents of the STKB register to the I bus and the contents of the STKB register to the R bus. A subtraction of the I bus from the R bus is conducted in the DALU and the output is enabled to the write bus and the Q register. The output of the write bus is transferred to the STKB register and to the data output register. Next, the data returned in response to the read request is sign extended and transferred to the I bus. This value is the b-word for the frame which was next to the current stack frame.

The received data word is transmitted to the I bus wherein it is added to the contents of the EO register by the DALU. The EO register contains the previous contents of the STKB register incremented by 1 so that the resultant sum is the new value to be stored in the STKB register. This is accomplished by enabling the output of the DALU to the STKB register in the BRAM by means of the O bus and the write bus.

Next, the address of the C word is copied from the STKP register into the Y register 24 and the newly calculated value of the C word is written into memory by means of a write instruction. Simultaneously, the contents of EO are transferred into the STKT register in the BRAM through the R bus, 0 bus, and Write bus. The same value is transferred to one of the B registers in the BRAM as selected by the n field of the RLQ macroinstruction.

The previously stored contents of the Q register are emitted to the Write bus through the I bus, DALU, and O bus and the value thereof is checked for 0. Since the contents of the Q register reflect the new value for the C word of the stack, if it is 0, then the last frame in the stack has been relinquised and a forced transfer to a firmware routine is executed to report the transferring of the last frame in the stack. If the last frame has not just been relinquished, then the address of the next operand is generated and a fetch of the next instruction is performed.

The operation of the stack mechanism of the instant invention is response to a MFL macroinstruction is illustrated in FIGS. 21 and 22. As stated above, the MFL macroinstruction can be employed to either increase or decrease the number of entries in the current stack frame. To implement the MFL macroinstruction, it must be first determined whether the change in frame size is positive, negative, or 0. If it is 0, then the instruction is treated as a no-op. If it is positive or negative, processing of the MFL macroinstruction continues.

Initially, the Q register specifies the number of words to be added or subtracted to the current stack frame. The contents of the Q register are enabled to the I bus and from the I bus to the DALU which implements their storing in the KO register. The contents of the KO register are then enabled to the R bus, W bus, and stored in the AO register. If the value transferred to the AO register is not equal to 0, then a positive or negative change in the frame size of the current stack frame is indicated.

Next, it is determined whether a stack containing at least one stack frame exists. This is performed by enabling the contents of the STKT register to the I bus and the STKP register to the R bus. The I bus is subtracted from the R bus by the DALU and the output of the DALU is enabled to the Write bus. From the Write bus it is enabled to the BALU and stored in the address register Y. If no carry occurs, then the contents of the STKT register have been determined to be greater than the contents of the STKP register which can only occur when either no stack or a stack with no frames exists. Such a situation results in a forced jump to a firmware trap routine.

If a carry did occur, then a stack with a frame exists and operation continues by setting the contents of the EO register equal to the contents of the AO register. Similarly, the contents of the Y register are enabled to the write bus and transferred to the AO register. The Y register is incremented and a memory read is performed. This memory read accesses the M word associated with the stack to enable a check to be made to determine whether a MFL macroinstruction increasing the size of the current stack has resulted in a stack overflow.

Meanwhile, the contents of the STKT register are enabled to the I bus and the contents of the EO register are enabled to the R bus. The DALU performs a subtraction of the I bus from the R bus and stores the result in the EO register. The effect of such an operation is to put the complement of the new contents of the STKT register into the EO register. A test is then performed to see whether the delta (change in size of the current stack frame) is positive (makes the stack frame layer) or negative (makes the stack frame smaller).

If the delta is negative, the contents of the STKB register are enabled to the I bus and the contents of the EO register are enabled to the R bus. The DALU adds the R bus and the I bus together and a test is conducted to see whether a carry has occurred. If a carry has occurred, then an attempt has been made to decrease the size of the current stack frame by an amount greater than its original size. This is an error and a forced branch is taken to a firmware error routine.

If a carry has occurred, the contents of EO are complemented by the DALU and stored back in the EO register the contents of the STKT register are copied into Y register 24, and the contents of the EO register are incremented by 1 and stored in the STKT register. The processing of the MFL macroinstruction to reduce the size of the current stack frame is then completed.

If the delta was positive, the contents of the AO register are enabled to the I bus and the contents of the EO register are enabled to the R bus. The DALU adds the I bus and the R bus together plus 1 and stores the result in the AO register by means of the Write bus. The returning data word is zero extended and transferred to the I bus and AO is enabled to the R bus. A subtraction of the I bus from the R bus minus 1 is performed in the DALU and a check is conducted for a carry. If a carry has occurred, then the MFL macroinstruction has caused a stack overflow and a forced branch is made to the firmware error routines. If no carry occurs, the contents of the EO register is enabled to the R bus and complemented by the DALU. The output of the DALU is enabled to the EO register and to the Write bus. The Write bus is then enabled to the BO register and the newly stored contents of the BO register are incremented by 1 by the BALU and stored in the Y register.

In sum, the MFL macroinstruction, when executed with a positive or negative delta that does not result in stack overflow or frame underflow, operates to change the value of the STKT register to address a different entry in the current stack frame.

Referring back to FIGS. 13(a)-13(g), the least significant seven bits of the first data word for each operation comprise the address syllable (ASI). Further, as illustrated in FIG. 13(g), if the seven bits of ASI 5C, then, bits 9-15 of the next data word contain a different address syllable, (AS2/3). Bit eight of the second data word selects between the two address syllables, AS2 and AS3.

Of particular interest to the instant invention is an address syllable format which is interpreted to control the operation of the stack mechanism. If the most significant seven bits of the first data words specify an m field of five and an n field of eight. If m equals five and n equals eight then a frame top push-pop operation is specified. If bits 9-15 of the first data word are the hexadecimal value 5C, and bits 8-15 of the next data word are CC.sub.16, then an indirect frame top pop operation is specified.

A frame-top (FT) push-pop causes the stack mechanism to modify the content of the STKT register by the size of the current operand. If the operation specifies a read operation such as a Load or an Add, then a pop or removal of the operand from the current stack frame will take place after the operand has been read. If the operation specifies a write operation such as a Store or a Clear then a push or store of the operand will be performed by increasing the size of the current stack frame and adding the operand to the stack.

The indirect FT pop means that the top operand in the current stack frame is the address of a location containing the instruction operand in the current stack frame and that after accessing that operand, that address will be popped or removed from the stack frame.

FIG. 23 illustrates the operation of the stack mechanism when the address syllable specifies an FT-push pop operation. In this operation, a three-bit code specifies the number of words required by the current operand. For example, the code is equal to 1 whenever the operands's size is less than or equal to one word, a double word operation is associated with the code value of 2 and a quad-word operation is associated with a code value of 4. The operand size code is one of the quantities emitted to the I bus as a result of the previously discussed I bus field in the control word.

The functionality of the operations performed by the stack mechanism of the instant invention is response to a FT push-pop address syllable is illustrated in FIG. 23. First, a complement of the operand's size is stored in the A0 register. This is performed by enabling the operand size, KWOS, to the I bus and complementing the I bus contents by the DALU. The DALU output is transferred to the W bus and stored in register AO. Next, it is determined whether the operand is a read or write by testing one of the two bits of the operation type outputs in the CPU. If it is a write operation the contents of AO are incremented by one and stored in AO. If a read operation is specified then the complement of AO is transmitted to AO. The effect of these operations is to make the operand size for a read operation negative and the sign of the operand size for a non-read operand positive.

If the operand was determined not to be a read, it is tested to see if it is a write operation. If it is a write operation then the stack mechanism performs the operations associated with the MFL macroinstructions. If it is not a read or write operation then a forced branch is made to a firmware error routine.

It if was a read operation then it is checked to see if it was a read/write operation. If it is, then it is a special class of operation which will push or pop 23 words on or off the stack, and the complement of 23 is stored in A0, after which the bit of the upcode distinguishing push from pop and the contents of AO are either complemented or incremented before performing an MFL operation. If it is not a read/write operation then a branch is taken to perform the firmware operations associated with the MFL macroinstructions.

If a branch is made to the firmware associated with the MFL macroinstruction, then processing will be performed by the stack mechanism in the above-described manner. The stack mechanism will calculate a new value for the STKT register, a new value for the C word, increase or decrease the current stack frame size, and the occurrence of any of the previously explained error conditions such as trying to increase the size of the current stack frame beyond the maximum size for the stack will be recognized.

The only difference in the execution of the previously explained firmware corresponding to the MFL macroinstruction will be in the situation that the operand is not a read operand. Thus, at line 1189g of the attached firmware listing, the new value of the stack top pointer as stored in register EO will be incremented in the DALU and stored in the STKT register. The subsequent test for SEL bein equal to zero will be true and the XR branch will be taken to read the next operand.

If a read is specified by the operand then in line SEL will be found to be equal to zero and a branch will be taken to read the next operand. Upon exiting the firmware associated with the MFL macroinstructions when the FT push-pop is indicated, all of the stack registers will be updated and the Y register will contain the address of the next operand location to be read.

The frame size of the current stack frame is altered in response to the indirect FT pop in the following manner. The contents of the STKT register are copied into the Y register such that the Y register contains the address of the location that stores the indirect address, i.e., the address to be used to access the operand. Simultaneously, contents of the STKP register are transferred to the read bus and tested for zero. If zero then no stack exists and a forced jump is performed to the firmware error routine.

If not zero, the contents of STKT is incremented by one and copied to the EO register by means of the DALU and the I bus.

A test is then conducted to determine whether the currect stack frame is capable of containing an address (two data words). This is accomplished by computing the difference between the contents of the EO register and the STKB register in the DALU. If a carry occurs then the current stack frame does not contain entries sufficient to hold two words and a firmware error routine is executed.

If no carry occurs then the frame is large enough to contain a two-word address and both words of data coming through the data multiplexer 35 to the source bus 33 are transmitted to the I bus, the DALU, and the Y register. Again, placing the returning data in the Y or address register implements the indirect addressing mode. Simultaneously, the contents of EO+1 are enabled into the STKT register. Since the contents of EO contains the previous value of the STKT register incremented by one, this operation effectively puts the previous contents of the STKT register incremented by two back into the STKT register. The result is the removal of two words from the current stack frame. An operand fetch is then performed.

In sum, the preceding disclosure sets forth a stack mechanism capable of addressing entries in a current stack frame by indexing relative to the top or bottom entries in the stack frame. Also, the stack mechanism, as herein embodied, is capable of creating or deleting stack frames within a stack, creating or deleting stacks, and dynamically changing the size of a stack frame. Further, the stack mechanism is capable of increasing or decreasing the frame size of the current stack frame by amounts reflecting the size of operands being executed by the central processing unit.

It will be apparent, to those skilled in the art, that modifications and variations can be made in the preferred embodiment disclosed herein without departing from the scope or sphere of the invention. Thus, it is intended that the present invention include those modifications and variations which come within the scope of the appended claims and their equivalents. ##SPC1## ##SPC2## ##SPC3## ##SPC4## ##SPC5## ##SPC6## ##SPC7## ##SPC8## ##SPC9##

Claims

What is claimed is:

1. In a data processing system having a memory comprised of a plurality of addressable storage locations and a processing unit for executing procedures, each of the procedures including a series of operands, a stack control mechanism comprising: stack forming means for establishing in said memory a stack of said addressable storage locations for storing said operands; frame forming means for dynamically altering the number of storage locations in said stack by (1) acquiring a stack frame associated with each procedure to be executed by said processing unit, each of said stack frames comprising one or more storage locations for storing operands and data relating to said associated procedures, (2) modifying the number of storage locations in said acquired stack frame responsive to execution of said associated procedure; and (3) relinquishing each of said storage location in a stack frame after the execution of said associated procedure to make said relinquished storage location available to said processing unit, the most recently acquired stack frame comprising the top stack frame; a bottom of frame register associated with each of said acquired stack frame for storing the number of storage locations included in said associated stack frames; stack frame control means for dynamically modifying the size of said top stack frame by adding storage locations thereto or deleting storage locations therefrom in response to said processing unit executing said procedure associated with said top stack frame; a stack top register for storing the address of the storage location in said top stack frame storing the operand most recently stored in said top stack frame; a stack base register for storing the address of said storage location immediately preceding a storage location storing the first operand stored in said top stack frame; means for updating the contents of said stack top register and said stack bottom register whenever a said operand is added to or deleted from said top stack frame; and operand accessing means for selectively accessing an operand stored in said top stack frame by indexing relative to the address stored in said stack top register or to the address stored in said stack base register.

2. A stack control mechanism according to claim 1 wherein said operand accessing means comprises an offset register for storing a positive or negative index value representing the relative address offset of an address storing a selected operand from said address stored in said stack top register and said stack bottom register, respectively, and means for adding a said positive index value to said address stored in said stack top register and a said negative index value to said address stored in said stack bottom register.

3. In a data processing system having a memory comprised of a plurality of addressable storage locations and a processing unit for executing procedures, each of the procedures including a series of operands, a method for creating a dynamically alterable operand stack in said memory comprising the steps of: establishing in said memory a stack for storing said operands, said stack including a storage location storing an indication of the maximum number of storage locations allocatable to said stack, a storage location storing an indication of the number of storage locations currently allocated in said stack, and one or more storage locations for storing said operands; acquiring different stack frames in said stack, each acquired stack frame being associated with a different one of said procedures to be executed by said processing unit and comprising one or more storage locations storing the operands and data relating to said associated procedure, the most recently established stack frame comprising the top stack frame; dynamically modifying the size of said top stack frame in response to said processing unit executing a said procedure by allocating additional storage locations to said top stack frame to store additional operands and data or by deleting storage locations from said top stack frame as operands and data are removed from said top stack frame; relinquishing the storage locations in a stack frame following the execution of said procedure associated with said stack frame, said relinquished storage locations being available for use by said processing unit; storing in a stack top register the address of the storage location in said top stack frame storing the operand most recently stored in said top stack frame; storing in a stack base register the address of said storage location immediately preceding the storage location storing the first operand stored in said top stack frame; updating the contents of said stack top register and said stack bottom register whenever a said operand is added to or deleted from said top stack frame, respectively; and selectively accessing operands in said top stack frame by addressably indexing with respect to the address stored in said stack top register or the address stored in said stack base register.

4. A method according to claim 3 wherein said operand accessing step comprises the steps of: generating an index offset having a positive value or a negative value; and generating the address of a selected operand in said top stack frame by adding said index offset to said address stored in said stack top register or said stack bottom register if said value of said index offset is positive or negative, respectively.

5. A stack control mechanism as in claim 1 wherein said stack forming means comprises means for indicating the maximum number of addressable storage locations allocatable to said established stack and means for indicating the total number of storage locations actually allocated to said stack.

6. A stack control mechanism as in claim 5 wherein said stack forming means further includes means associated with each said established stack frame for storing an indication of the number of said storage locations included within said associated stack frame.

7. A stack control mechanism according to claim 6 wherein said means for indicating the number of storage locations currently allocated to said stack comprises a storage location in said stack for storing a value corresponding to the number of said allocated storage locations.

8. A stack control mechanism according to claim 7 further including a stack pointer register for storing the address of said storage location of said means indicating the total number of storage locations allocated in said stack.

Patent Citations (3)

Patent	Date	Inventor
US3686641	1972-08-01	Logan et al.
US4130867	1978-12-01	Bachman et al.
US4325120	1982-04-01	Colley et al.

Abstract

Description