IE901519A1

IE901519A1 - Multiple instruction preprocessing system with data¹dependency resolution

Info

Publication number: IE901519A1
Application number: IE151990A
Authority: IE
Original assignee: Digital Equipment Corp
Priority date: 1990-04-27
Filing date: 1990-04-27
Publication date: 1991-11-06

Abstract

A method is provided for preprocessing multiple instructions prior to execution of such instructions in a digital computer having an instruction decoder, an instruction execution unit, and multiple general purpose registers which are read to produce memory addresses during the preprocessing. The method comprises: (1) avoiding the preprocessing of a current instruction to read a general purpose register to produce a memory address prior to the modification of the contents of that register by a preceding instruction by (a) generating a composite write mask having a bit set for each general purpose register whose contents are to be modified by at least one of a plurality of decoded but not-yet-executed instructions preceding the current instruction, and (b) stalling the preprocessing of the current instruction when a general purpose register to be read by the current instruction is a,register having a bit set in the write mask, and/or (2) avoiding the preprocessing of a current instruction which modifies the contents of a general purpose register that is to be read by a preceding instruction by (a) generating a composite read mask having a bit set for each general purpose register to be read by at least one of a plurality of decoded but not-yet-executed instructions preceding the current instruction, and (b) stalling the preprocessing of the current instruction when a general purpose register whose contents are to be modified by the current instruction is a register having a bit set in the read mask.

Description

PD88-0257 U.s.: DIGM:011 FOREIGN: DIGM:042 DIGITAL EQUIPMENT CORPORATION. 8 Sheets.

Sheet 1. ixy T- L I PATENTS ACT 1964 COMPLETE SPECIFICATION MULTIPLE INSTRUCTION PREPROCESSING SYSTEM WITH DATA DEPENDENCY RESOLUTION onw »o fSfbuc wrarecTiGw qhMR atOTION 69 AWO ' WVt 11 ? ft ϊ 'PLICATION No_ CiFlCATlON FILED DIGITAL EQUIPMENT CORPORATION, a Corporation organised and existing under the laws of the State of Massachusetts, United States of America, of 111 Powdermill Road, Maynard, Massachusetts 01754-1418, United States of America - 1 6015 PD88-0257 U.S.: DIGM:O11 FOREIGN: DIGM:O42 MULTIPLE INSTRUCTION PREPROCESSING 10 SYSTEM WITH DATA DEPENDENCY RESOLUTION The present application discloses certain aspects of a computing system that is further described in the following U.S. patent applications filed concurrently with the present application: Evans et al., AN INTERFACE BETWEEN A SYSTEM CONTROL UNIT AND A SERVICE PROCESSING UNIT OF A DIGITAL COMPUTER; Arnold et al., METHOD AND APPARATUS FOR INTERFACING A SYSTEM CONTROL UNIT FOR A MULTIPROCESSOR SYSTEM WITH THE CENTRAL PROCESSING UNITS; Gagliardo et al., METHOD AND MEANS FOR INTERFACING A SYSTEM CONTROL UNIT FOR A MULTI-PROCESSOR SYSTEM WITH THE SYSTEM MAIN MEMORY; D. Fite et al., METHOD AND APPARATUS FOR RESOLVING A VARIABLE NUMBER OF POTENTIAL MEMORY ACCESS CONFLICTS IN A PIPELINED COMPUTER SYSTEM; D. Fite et al., DECODING MULTIPLE SPECIFIERS IN A VARIABLE LENGTH INSTRUCTION ARCHITECTURE; D. Fite et al., VIRTUAL INSTRUCTION CACHE REFILL ALGORITHM; Murray et al., PIPELINE PROCESSING OF REGISTER AND REGISTER MODIFYING SPECIFIERS WITHIN THE SAME INSTRUCTION; D. Fite et al., PREPROCESSING IMPLIED SPECIFIERS IN A PIPELINED PROCESSOR; D. Fite et al., BRANCH PREDICTION; Fossum et al., PIPELINED FLOATING POINT ADDER FOR DIGITAL COMPUTER; Grundmann et al., SELF TIMED REGISTER FILE; Beaven et al., METHOD AND APPARATUS FOR DETECTING AND CORRECTING ERRORS IN A PIPELINED COMPUTER SYSTEM; Flynn et al., METHOD AND MEANS FOR ARBITRATING COMMUNICATION REQUESTS USING A SYSTEM CONTROL UNIT IN A MULTI-PROCESSOR SYSTEM; E. Fite et al., CONTROL OF MULTIPLE FUNCTION UNITS WITH PARALLEL OPERATION IN A MICROCODED EXECUTION UNIT; Webb, -2Jr. et al., PROCESSING OF MEMORY ACCESS EXCEPTIONS WITH PRE-FETCHED INSTRUCTIONS WITHIN THE INSTRUCTION PIPELINE OF A VIRTUAL MEMORY SYSTEM-BASED DIGITAL COMPUTER; Hetherington et al., METHOD AND APPARATUS FOR CONTROLLING THE CONVERSION OF VIRTUAL TO PHYSICAL MEMORY ADDRESSES IN A DIGITAL COMPUTER SYSTEM; Hetherington et al., WRITE BACK BUFFER WITH ERROR CORRECTING CAPABILITIES; Flynn et al., METHOD AND MEANS FOR ARBITRATING COMMUNICATION REQUESTS USING A SYSTEM CONTROL UNIT IN A MULTI-PROCESSOR SYSTEM; Chinnaswamy et al., MODULAR CROSSBAR INTERCONNECTION NETWORK FOR DATA TRANSACTIONS BETWEEN SYSTEM UNITS IN A MULTI-PROCESSOR SYSTEM; Polzin et al., METHOD AND APPARATUS FOR INTERFACING A SYSTEM CONTROL UNIT FOR A MULTI-PROCESSOR SYSTEM WITH INPUT/OUTPUT UNITS; Gagliardo et al., MEMORY CONFIGURATION FOR USE WITH MEANS FOR INTERFACING A SYSTEM CONTROL UNIT FOR A MULTI-PROCESSOR SYSTEM WITH THE SYSTEM MAIN MEMORY; and Gagliardo et al., METHOD AND MEANS FOR ERROR CHECKING OF DRAM-CONTROL SIGNALS BETWEEN SYSTEM MODULES.

The present invention relates generally to digital computers and, more particularly, to a system for resolving data dependencies during the preprocessing of multiple instructions prior to execution of those instructions in a digital computer. This invention is particularly applicable to the preprocessing of multiple instructions in a pipelined digital computer system using a variable-length complex instruction set (CIS) architecture.

Preprocessing of instructions is a common expedient used in digital computers to speed up the execution of large numbers of instructions. The preprocessing operations are typically carried out by an instruction unit interposed between the memory that stores the PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -3instructions and the execution unit that executes the instructions. The preprocessing operations include, for example, the prefetching of operands identified by operand specifiers in successive instructions so that the operands are readily available when the respective instructions are loaded into the execution unit. The instruction unit carries out the preprocessing operations for subsequent instructions while a current instruction is being executed by the execution unit, thereby reducing the overall processing time for any given sequence of instructions.

The preprocessing of instructions can give rise to data dependency problems when the instruction unit changes the contents of a storage location that is needed by an instruction not yet fully executed by the execution unit. When the only data dependency problem is potential modification of the contents of a register during the execution of an instruction, before that same register is read by the preprocessing unit for preprocessing a subsequent instruction, the problem can be deferred to the execution unit by simply passing register pointers to the execution unit. Thus the register contents are not read during the preprocessing of an instruction, but rather during the execution of that instruction.

When registers are used to produce memory addresses, however, as in indirect or deferred addressing, effective preprocessing of instructions requires that the registers be accessed and sometimes changed during preprocessing. Thus the data dependency problem becomes more severe, particularly when multiple instructions are pre-processed, presenting the potential of changing the contents of several different registers identified by multiple instructions following an instruction that has PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -4not yet been completely executed. A premature change of register contents can be required by the preprocessing of an instruction that reads a register to produce a memory address before the contents of that register have been changed by a preceding instruction that is not yet fully executed. A somewhat different, but equally serious, data dependency problem occurs when the preprocessing read operations include autoincrementing or autodecrementing functions which modify the contents of 10 registers.

’’Autoincrement and autodecrement are addressing modes of a digital computer that includes built-in hardware for automatically stepping a register through the sequential elements of a table or array. Thus, the contents of a register may be stepped through a series of addresses so that it always points to the next sequential element of a table. The register is incremented or decremented by 1, 2, 4 or 8 depending upon the data type, i.e., whether the operand is 1, 2, 4 or 8 bytes in length. In the autoincrement mode, the register contents are incremented after being used as the address of the operand; in the autodecrement mode, the register contents are decremented before being used as the address of the operand.

There is provided a method of preprocessing multiple instructions prior to execution of such instructions in a digital computer having an instruction decoder, an instruction execution unit, and multiple general purpose registers, and in which the general purpose registers are read to produce memory addresses during the preprocessing of said instructions, said method comprising (1) avoiding the preprocessing of a current PD88-0257 U.S.: DIGMiOll FOREIGN: DIGM:042 -5instruction to read a general purpose register to produce a memory address prior to the modification of the contents of that register by a preceding instruction by (a) generating a composite write mask having a bit set for each general purpose register whose contents are to be modified by at least one of a plurality of decoded but not-yet-executed instructions preceding the current instruction, and (b) stalling the preprocessing of the current instruction when a general purpose register to be read by the current instruction is a register having a bit set in the write mask, and/or (2) avoiding the preprocessing of a current instruction to modify the contents of a general purpose register read by a preceding instruction by (a) generating a composite read mask having a bit set for each general purpose register to be read by at least one of a plurality of decoded but not-yet-executed instructions preceding the current instruction, and (b) stalling the preprocessing of the current instruction when a general purpose register whose contents are to be modified by the current instruction is a register having a bit set in the read mask.

Two specific data dependency problems and their resolution by the present invention are illustrated by the following examples: CASE 1: An instruction uses a general purpose register (GPR) to address memory, and the contents of that GPR are to be changed by execution of a previously decoded but not-yet-executed instruction. This problem is exemplified by the following pair of instructions: MOVAL X R0 ADDL3 (R0), Rl, R2 PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 Execution of the first instruction moves the long word address specified by the label X to the register RO. Execution of the second instruction adds the contents at the address in the register RO to the contents of register R1 and stores the sum in register R2. An instance of register conflict occurs when the RO operand of the ADDL3 instruction is fetched before the previous MOVAL instruction has its result stored in the register RO. To detect this case of register conflict, a write mask is generated for each decoded instruction to indicate the registers, if any, that have contents which are changed by execution of that instruction. The write masks for all the decoded but not-yet-executed instructions are queued and used to form a composite write mask for all such instructions. Then during the preprocessing of an instruction that produces memory addresses using data in a GPR, the write masks for the previously decoded but not-yet-executed (i.e., not retired) instructions are inspected to determine whether the GPR containing the address from which an operand is being fetched is a register identified by the write masks as a register whose contents are to be changed by a previous instruction. If the answer is affirmative, the fetching is stalled until the register conflict no longer exists.

In the specific example, the MOVAL” instruction has a write mask of (0000000000000012), which includes a respective bit position for each of fifteen GPRs; the least significant bit position is set to indicate that the MOVAL instruction changes the contents of register RO. During the fetching of the operands for the ADDL3 instruction, the contents of register RO are needed, and therefore the least significant bit position of the write mask for the previous MOVAL instruction is inspected; PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -7since this bit position is set, the fetching is stalled until execution of the MOVAL instruction is completed. The stall is terminated when the least significant bit is cleared by removal of the write mask for the MOVAL instruction from the mask queue upon completion of the execution of that instruction.

CASE 2: The fetching of operands for an instruction changes the contents of a GPR used to produce a memory address, and the contents of that GPR are to be read by execution of a previously decoded but not-yet-executed instruction. This problem is illustrated by the following pair of instructions: ADDL R3, RO, R4 MOVL (R0)+, Rl In this example, the + sign after (RO) causes an autoincrement of the contents of register RO. An instance of register conflict would occur if the autoincrement were to change the contents of the register RO before those contents were read during execution of the previous ADDL instruction. To avoid this case of register conflict, a read mask is generated for each decoded instruction to indicate the registers, if any, that have contents which are read during execution of the instruction. The read masks for all the decoded but not-yet-executed instructions are queued and used to form a composite read mask for all such instructions. Then during the fetching of operands for an instruction in which the contents of a GPR are changed during the fetch (e.g., an autoincrement or auto-decrement of a GPR) the read masks for previously decoded instructions not yet fully executed (i.e., not retired) are inspected to determine whether the GPR having contents to be changed is a register identified by the read masks as a register to be read during execution of the previous instruction.

PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -8If the answer is affirmative, the fetching is stalled until the register conflict no longer exists.

In the above example, the ADDL instruction has a 5 read mask of (0000000000010012) indicating that the execution unit will read the contents of registers R3 and R0. During the fetching of the operands for the MOVL instruction, the contents of register RO are to be autoincremented, and therefore the least significant bit position of the read mask for the previous ADDL instruction is inspected; since this bit position is set, the fetching is stalled until execution of the ADDL instruction is completed. The stall is terminated when the least significant bit is cleared by removal of the read mask for the ADDL instruction from the mask queue upon completion of the execution of that instruction.

As can be seen from the two exemplary cases described above, the read and write masks are used to avoid the reading of a register to produce a memory address either (1) prior to the modification of the contents of that register by a preceding instruction or (2) subsequent to the modification of the contents of that register by a subsequent instruction. In the first case, the reading of the register by the subsequent instruction is stalled until the preceding instruction writes to that register. The write mask is used for this purpose. In the second case, the modification of the register contents by the subsequent instruction is stalled until the preceding instruction reads that register. The read mask is used for this purpose. That is, the adjective describing the mask always refers to the operation (read or write) to be performed by the preceding instruction.

PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -9Other objects and advantages of the invention will become apparent upon reading the following detailed description and upon reference to the accompanying drawings, in which: FIG. 1 is a block diagram of a central processing unit which uses a set of queues, including a mask queue, to fetch operands for subsequent instructions before an execution unit executes a prior instruction; FIG. 2 is a diagram showing various steps performed to process an instruction and which may be performed in parallel for different instructions by a pipelined instruction processor according to FIG. 1; FIG. 3 is a block diagram of the instruction processor of FIG. 1 showing in further detail the instruction decoder and the queues; FIG. 4 is a block diagram of the mask queue and its associated circuits for inserting and removing GPR use masks and for flushing the queue, and for generating composite read and write masks; FIG. 5 is a detailed schematic diagram of the mask queue; FIG. 6 is a schematic diagram of mask-generating circuitry for use in the system of FIGS. 1-4; FIG. 7 is a schematic diagram of register conflict detector circuitry which inspects the read and write masks to determine whether any register conflicts exist; and PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -10FIG. 8 shows how a pair of register conflict detectors are used to obtain a current register conflict signal and a new register conflict signal in a double-buffered pipeline stage including the operand 5 unit.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof have been shown by way of example in the drawings and will be described in detail herein. It should be understood, however, that it is not intended to limit the invention to the particular forms disclosed, but on the contrary, the intention is to cover all modifications, equivalents, and alternatives falling within the spirit and scope of the invention as defined by the appended claims.

Turning now to the drawings and referring first to FIG. 1, there is shown a portion of a digital computer system which includes a main memory 10, a memory-CPU interface unit 11, and at least one CPU comprising an instruction unit 12 and an execution unit 13. It should be understood that additional CPUs could be used in such a system by sharing the main memory 10. Both data and instructions for processing the data are stored in addressable storage locations within the main memory 10. An instruction includes an operation code (opcode) that specifies, in coded form, an operation to be performed by the CPU, and operands identified by specifiers that provide the information needed to locate the operands.

The execution of an individual instruction is broken down into multiple smaller tasks. These tasks are performed by dedicated, separate, independent functional units that are optimized for that purpose.

PD88-0257.

U.S.: DIGMiOll FOREIGN: DIGM:042 -11Although each instruction ultimately performs a different operation, many of the smaller tasks into which each instruction is broken are common to all instructions. Generally, the following steps are performed during the execution of an instruction: instruction fetch, instruction decode, operand fetch, execution, and result store. Thus, by the use of dedicated hardware stages, the steps can be overlapped in a pipelined operation, thereby increasing the total instruction throughput.

The data path through the pipeline includes a respective set of registers for transferring the results of each pipeline stage to the next pipeline stage. These transfer registers are clocked in response to a common system clock. For example, during a first clock cycle, the first instruction is fetched by hardware dedicated to instruction fetch. During the second clock cycle, the fetched instruction is transferred and decoded by instruction decode hardware, but, at the same time, the next instruction is fetched by the instruction fetch hardware. During the third clock cycle, each instruction is shifted to the next stage of the pipeline and a new instruction is fetched. Thus, after the pipeline is filled, an instruction will be completely executed at the end of each clock cycle.

This process is analogous to an assembly line in a manufacturing environment. Each worker is dedicated to performing a single task on every product that passes through his or her work stage. As each task is performed the product comes closer to completion. At the final stage, each time the worker performs his assigned task a completed product rolls off the assembly line.

PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -12In the particular system illustrated in FIG. 1, the interface unit 11 includes a main cache 14 which on an average basis enables the instruction and execution units 12 and 13 to process data at a faster rate than the access time of the main memory 10. This cache 14 includes means for storing selected predefined blocks of data elements, means for receiving requests from the instruction unit 12 via a translation buffer 15 to access a specified data element, means for checking whether the data element is in a block stored in the cache, and means operative when data for the block including the specified data element is not so stored for reading the specified block of data from the main memory 10 and storing that block of data in the cache 14. In other words, the cache provides a window into the main memory, and contains data likely to be needed by the instruction and execution units.

If a data element needed by the instruction unit 13 or execution unit 13 is not found in the cache 14, then the data element is obtained from the main memory 10, but in the process, an entire block, including additional data, is obtained from the main memory 10 and written into the cache 14. Due to the principle of locality in time and memory space, the next time the instruction and execution units desire a data element, there is a high degree of likelihood that this data element will be found in the block which includes the previously addressed data element. Consequently, there is a high degree of likelihood that the cache 14 will already include the data element required by the instruction and execution units 12 and 13. In general, since the cache 14 will be accessed at a much higher rate than the main memory 10, the main memory can have a proportionally slower access PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -13time than the cache 14 without substantially degrading the average performance of the data processing system. Therefore, the main memory 10 can be comprised of slower and less expensive memory elements.

The translation buffer 15 is a high speed associative memory which stores the most recently used virtual-to-physical address translations. In a virtual memory system, a reference to a single virtual address can cause several memory references before the desired information is made available. However, where the translation buffer 15 is used, translation is reduced to simply finding a hit in the translation buffer 15.

An I/O bus 16 is connected to the main memory 10 and the main cache 14 for transmitting commands and input data to the system and receiving output data from the system.

The instruction unit 12 includes a program counter and an instruction cache 18 and an instruction buffer 19 for fetching instructions from the main cache 14.

From the buffer 19, new instructions are fed to an instruction decoder 20 which decodes both the op-codes and the specifiers. An operand processing unit (OPU) 21 fetches the specified operands and supplies them to the execution unit 13.

The program counter 17 is maintained in the instruction unit 12, so that the proper instructions can be retrieved from the cache 14. The program counter 17 preferably uses virtual memory locations rather than the physical memory locations of the main memory 10 and the cache 14. Thus, the virtual address of the program counter 17 must be translated into the physical address PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 ,E 901519 -14of the main memory 10 before instructions can be retrieved. Accordingly, the contents of the program counter 17 are transferred to the interface unit 11 where the translation buffer 15 performs the address conversion. The instruction is retrieved from its physical memory location in the cache 14 using the converted address. The cache 14 delivers the instruction over data return lines to the instruction decoder 20.

The general organization and operation of the cache 14 and the translation buffer 15 are further described in Chapter 11 of Levy and Eckhouse, Jr., Computer Programming and Architecture, The VAX-11, Digital Equipment Corporation, pp. 351-368 (1980).

The OPU 21 also produces virtual addresses. In particular, the OPU 21 produces virtual addresses for memory source (read) and destination (write) operands.

For at least the memory read operands, the OPU 21 must deliver these virtual addresses to the interface unit 11 where they are translated to physical addresses. The physical memory locations of the cache 14 are then accessed to fetch the operands for the memory source operands.

In each instruction, the first byte contains the opcode, and the following bytes are the operand specifiers to be decoded. The first byte of each specifier indicates the addressing mode for that specifier. This byte is usually broken in halves, with one half specifying the addressing mode and the other half specifying the register to be used for addressing.

The first step in processing the instructions is to decode the opcode portion of the instruction. The first portion of each instruction consists of its opcode PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -15which specifies the operation to be performed in the instruction. The decoding is done using a table-look-up technique in the instruction decoder 20. The instruction decoder finds a microcode starting address for executing the instruction in a look-up table and passes the starting address to the execution unit 13. Later the execution unit performs the specified operation by executing prestored microcode, beginning at the indicated starting address. Also, the decoder determines where source-operand and destination-operand specifiers occur in the instruction and passes these specifiers to the OPU 21 for preprocessing prior to execution of the instruction.

The look-up table is organized as an array of multiple blocks, each having multiple entries. Each entry can be addressed by its block and entry index. The opcode byte addresses the block, and a pointer from an execution point counter (indicating the position of the current specifier in the instruction) selects a particular entry in the block. The entry specifies the data context (byte, word, etc.), data type (address, integer, etc.) and accessing mode (read, write, modify, etc.) for each specifier.

After an instruction has been decoded, the OPU 21 parses the operand specifiers and computes their effective addresses; this process involves reading GPRs and possibly modifying the GPR contents by autoincrementing or autodecrementing. The OPU passes these effective addresses to the interface unit 11 which either returns the operand data (in the case of a read) to the source list 24 in the execution unit 13 or translates the effective address to a physical write address and awaits the data from the execution unit (in PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -16the case of a write) or does both (in the case of a modify).

Each time an instruction is passed to the execution 5 unit, the instruction unit sends a microcode dispatch address and a set of pointers for (1) the locations in the execution-unit register file where the source operands can be found, and (2) the location where the results are to be stored. Within the execution unit, a set of queues 23 includes a fork queue for storing the microcode dispatch address, a source pointer queue for storing the source-operand locations, and a destination pointer queue for storing the destination location. Each of these queues is a FIFO buffer capable of holding the data for multiple instructions.

The execution unit 13 also includes a source list 24, which is a multi-ported register file containing a copy of the GPRs and a list of source operands. Thus entries in the source pointer queue will either point to GPR locations for register operands, or point to the source list for memory and literal operands. Both the interface unit 11 and the instruction unit 12 write entries in the source list 24, and the execution unit 13 reads operands out of the source list as needed to execute the instructions. For executing instructions, the execution unit 13 includes an instruction issue unit , a microcode execution unit 26, an arithmetic and logic unit (ALU) 27, and a retire unit 28.

The present invention is particularly useful with pipelined processors. As discussed above, in a pipelined processor the processor’s instruction fetch hardware may be fetching one instruction while other hardware is decoding the operation code of a second instruction, PD88-0257 U.S.: DIGMiOll FOREIGN: DIGM:042 fetching the operands of a third instruction, executing a fourth instruction, and storing the processed data of a fifth instruction. FIG. 2 illustrates a pipeline for a typical instruction such as: ADDL3 RO, B*12 (Rl), R2.

This is a long-word addition using the displacement mode of addressing.

In the first stage of the pipelined execution of 10 this instruction, the program count (PC) of the instruction is created; this is usually accomplished either by incrementing the program counter from the previous instruction, or by using the target address of a branch instruction. The PC is then used to access the instruction cache 18 in the second stage of the pipeline.

In the third stage of the pipeline, the instruction data is available from the cache 18 for use by the instruction decoder 20, or to be loaded into the instruction buffer 19. The instruction decoder 20 decodes the opcode and the three specifiers in a single cycle, as will be described in more detail below. The Rl number along with the byte displacement is sent to the OPU 21 at the end of the decode cycle.

In stage 4, the RO and R2 numbers are passed as pointers to the queue unit 23. Also, the operand unit 21 reads the contents of its GPR register file at location Rl, adds that value to the specified displacement (12), and sends the resulting address to the translation buffer 15 in the interface unit 11, along with an OP READ request, at the end of the address generation stage. The OPU also sends a source list pointer to the queue unit 23 to reserve an entry in a source list queue (24 in FIG. 3) for the second operand.

PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -18In stage 5, the interface unit 11 selects the address generated in stage 4 for execution. Using the translation buffer 15, the interface unit 11 translates the virtual address to a physical address during the address translation stage. The physical address is then used to address the cache 14, which is read in stage 6 of the pipeline, and data is returned to the source list 24.

In stage 7 of the pipeline, the instruction is issued to the ALU 27 which adds the two operands and sends the result to the retire unit 28. During stage 4, the register numbers for R1 and R2, and a pointer to the source list location for the memory data, was sent to the execution unit and stored in the pointer queues. Then during the cache read stage, the execution unit started to look for the two source operands in the source list.

In this particular example it finds only the register data in RO, but at the end of this stage the memory data arrives and is substituted for the invalidated read-out of the register file. Thus both operands are available in the instruction execution stage.

In the retire stage 8 of the pipeline, the result data is paired with the next entry in a retire queue in the retire unit 28. Although several functional execution units can be busy at the same time, only one instruction can be retired in a single cycle.

In the last stage 9 of the illustrative pipeline, the data is written into the GPR portion of the register files in both the execution unit 13 and the instruction unit 12. The instruction decoder 20 in the instruction unit 12, and the queues 23 in the execution unit 13, are shown in more detail in FIG. 3. It can be seen that the PD88-0257.

U.S.: DIGMzOll FOREIGN: DIGM:042 -19decoder 20 includes a decoder 20a for the program counter, a fork table RAM 20b, two source-operand specifier decoders 20c and 20d, a destination-operand specifier decoder 20e, and a register-operation decoder 20f which will be described in detail below. In a preferred embodiment, the decoders are intimately interlinked and integrated into a large complex decode unit, as further described in the above referenced Fite et al. U.S. patent application Ser. No. , filed , entitled Decoding Multiple Specifiers in a Variable Length Instruction Architecture, incorporated herein by reference.

The output of the program-counter decoder 20a is stored in a program counter queue 23a in the execution unit 13. The RAM 20b receives only the opcode byte of each instruction, and uses that data to select a fork (microcode) dispatch address from a table. This dispatch address identifies the start of the microcode appropriate to the execution of the instruction, and is stored in a fork queue 23b in the execution unit 13.

Each of the four decoders 20c-20f receives both the opcode byte and the operand specifier data from the instruction buffer 19. The decoders 20c and 20d decode two source-operand specifiers to generate source-operand pointers which can be used by the execution unit to locate the two source operands. These two pointers are stored in a source-pointer queue 23c in the execution unit. The destination-operand specifier is decoded by the decoder 20e to generate a destination-operand pointer which is stored in a destination-pointer queue 23e in the execution unit.

In order to check for the register conflicts PD88-0257 U.S.: DIGM-.011 FOREIGN: DIGM:042 -20discussed above, a pair of masks are generated each time a new instruction is decoded, to identify all GPRs that the execution unit will read or write during the execution of that instruction. These masks are generated in the register-operation decoder 2Of (described below in connection with FIG. 4) and are stored in a mask queue 23f in the instruction unit. Each mask comprises a number of bit positions equal to the number of GPRs. In the read mask, a bit is set for each GPR to be read during execution of the new instruction, and in the write mask, a bit is set for each GPR to be written during execution of that instruction.

Both the read and write masks for a given instruction are stored as a single entry in the mask queue 23f. When there are fifteen GPRs, each entry in the mask queue consists of thirty bits (fifteen bits in each read mask to identify GPRs to be read and fifteen bits in each write mask to identify GPRs to be written).

The composite of all the valid masks in the mask queue 23f is used to check each register to be used to produce a memory address during the preprocessing of instructions in the instruction unit 12 to determine whether the preprocessing of that instruction should be stalled, as will be described in more detail below.

Circuitry for generating the read and write mask for each newly decoded instruction, in the register-operation decoder 20f, is illustrated in FIG. 4. This circuitry includes three decoders 30, 31 and 32; the first two decoders 30 and 31 receive two 4-bit register numbers from the source-operand specifiers, and the third decoder 32 receives a 4-bit register number from the destination-operand specifier. (The numbers of bits carried by respective data lines shown in FIG. 4 are PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -21indicated by respective numerals adjacent to double slash symbols intersecting the respective lines.) In addition to the register number, each of the decoders 30, 31 and 32 receives a 2-bit data-type number which indicates the number of registers to be used (1, 2, 3 or 4) for the particular format being used at any given time. Some formats use 8 bits, 16 bits, or 32 bits (requiring one register), others use 64 bits (requiring two registers), and others use 128 bits (requiring four registers).

The two decoders 30 and 31 set corresponding bits in a 15-bit output; the set bits identify the particular GPRs to be read by the execution unit for one particular specifier.

From the decoder 30, the fifteen outputs are supplied to an array of fifteen AND gates 33 whose outputs are connected to an array of fifteen OR gates 34.

The second input to each of the AND gates 33 is the output of an AND gate 35 whose inputs are asserted when the first source-operand specifier and the register number from that specifier are valid.

The fifteen outputs from the decoder 31 are similarly supplied to an array of fifteen AND gates 36 whose outputs are connected to the array of OR gates 34. The second input to each of the AND gates 36 is the output of an AND gate 37 whose inputs are asserted when the second source-operand specifier and the register number from that specifier are valid.

The fifteen outputs of the OR-gate array 34 produces a single 15-bit read mask, comprising bits RM0 through RM14, to be stored in the mask queue 23f. A PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -22recirculating register 38 provides a buffer between the OR gates 34 and the mask queue 23f. In the preferred embodiment of the instruction decoder referred to above, it is possible that source specifiers for the same instruction will be decoded in different cycles of the system clock, and therefore the output of the register 38 is recirculated through a set of AND gates 38b feeding back to the OR gates 34 until a new instruction is decoded. The AND gates 38b are therefore enabled by the inversion of the HOLD signal, as provided by an inverter 38a.

The third decoder 32 determines which of the fifteen GPRs are to be used as a destination of the result produced by execution of the newly decoded instruction. This decoder 32 sets from one to four of fifteen bits corresponding to the fifteen GPRs, with the set bits identifying the particular GPRs to be used as the destination. From the decoder 32, the fifteen outputs are supplied to an array of fifteen AND gates 39. The output of the register is the 15-bit write mask, comprising bits WMO through WM14, which are stored in the mask queue 23f. The second input to each of the AND gates 39 is the output of an AND gate 41 whose inputs are asserted when the destination-operand specifier and the register number from that specifier are valid.

As illustrated in FIGS. 5 and 6, the mask queue 23f is a first-in, first-out (FIFO) circular buffer including six registers 40-45 to accommodate the entry of read and write masks for six different instructions. The thirty bits in each register are connected to thirty parallel AND gates such as the gates A1-A30 and B1-B30 for registers 40 and 41, respectively. The thirty AND gates in each group all receive a common signal from PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -23valid-queue-entry logic 46. This logic keeps track of the number of valid masks in the mask queue 23f at all times, and the particular registers in which those masks are located. The two inputs to the valid-queue-entry logic 46 are (1) an insert pointer signal from an insert pointer counter 47 and (2) a size signal from a size counter 48. The logic 46 responds to these two inputs to produce control signals at six different outputs according to the logic in Table I.

Thus, the logic 46 asserts the outputs QO and Ql, i.e., the six outputs from the logic 46 are 000011. This enables the AND gates B1-B30 and C1-C30 (not shown) to pass the masks from registers 41 and 42.

Both the insert pointer counter 47 and the size counter 48 are incremented each time a new instruction is decoded, and the size counter 48 is decremented each time the execution of an instruction is completed. The Ql and Q2 outputs of the size counter 48 are supplied to an AND gate 49 to produce a control signal indicating that the mask queue is full when both these bits are set (bits Ql and Q2 represent a decimal 6 when both bits are set, indicating that the mask queue contains six valid mask entries). This queue full signal stalls the decoding of any further instructions until space becomes available in the mask queue 23f (see FIG. 8 discussed below).

The insert pointer counter 47 and the size counter 48 have preset command inputs PS receiving a flush signal which flushes or clears the preprocessing system.

However, an important feature of the invention is that a specified number of the oldest entries are retained in the mask queue 23f in the event of a flush. This feature is useful in connection with flushing the most recent portion of the queue which stores the masks for instructions after a predicted branch when the branch PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -25prediction is found to be incorrect. The number of entries to be retained is specified by Table II.

TABLE II No. Complex Branch, Memory Entries & Preceding In- Access COMMENTS To Keep_struction Stalled _ 1 0 0 The branch 10 prediction is found to be incor- rect before the current [branch] instruction is 15 retired; keep the branch only. 2 0 1 Keep the branch instruction and the 20 stalled memory access instruction. 0 1 0 The branch will have 25 been completed in this case so nothing to save. DON'T 1 1 This case can't happen; 30 CARE a stalled memory access instruction cannot be retired. 35 During a flush, the preset inputs PS of the insert PD88-0257 U.S.: DIGMzOll FOREIGN: DIGM:042 -26pointer counter 47 and the size counter 48 are activated. In order to determine the present value of the insert pointer counter, the number of entries to be retained is subtracted from the number of valid entries in the queue, to determine the number of entries to be removed. The number of entries to be removed is then subtracted from the current value of the insert pointer. The subtractions and the insert pointer counting are performed according to modulo six arithmetic. The truth table of a modulo six subtractor is shown in Table III.

TABLE III register_insert c 000 000 001 010 Oil 100 101 o v 001 101 000 001 010 Oil 100 1 010 100 101 000 001 010 Oil I Λ011 Oil 100 101 000 001 010 e 100, 010. Oil, 100, 101, 000, U p pr pp p t 1Q111_0.0.111_OlOj | Ollj 1 100|| 101| 110 1 000 1 001 I 010 1 011 1 100 001| —ϋίώ. —IPX 1 The preset value for the size counter 48 is either the number of entries to be retained, if an instruction is not being retired during the flush, or one less than the number of entries to be retained, if an instruction is being retired. This selective presetting of the size counter 48 is necessary because the number of entries to be retained, as specified by Table II, includes any instruction being retired. As illustrated in FIG. 6, a pair of modulo six subtractors 50 and 51 receive a signal representing the number of entries to be retained. The subtractor 50 receives this signal at its negative input, and its positive input receives the output of the size counter 48. The output of the subtractor 50 is supplied PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 IE 9010’* -27to the negative input of a third modulo six subtractor 52, whose positive input receives the output of the insert pointer counter 47. Thus, the output of the subtractor 52 represents the difference between the number of valid entries in the queue and the number of entries to be removed. This signal is the preset value input to the counter 47.

The positive input of the subtractor 51 receives 10 the signal representing the number of entries to be retained, and this value becomes the preset value for the size counter if an instruction is not being retired during the flush. If an instruction is being retired, the preset value for the size counter is reduced by one in the subtractor 51.

It should be noted that when the increment and decrement inputs of the size counter 48 are both active, the value of the size count should not change. As is conventional, it is assumed that when a flush occurs to preset the size counter, the signals on the increment and decrement inputs are ignored; this is of course presumed since the preset output is selectively subtracted by one based on the state of the instruction retired or removal signal. Therefore, the size counter is constructed in the conventional fashion by a register, a multiplexer having the preset, increment and decrement signals as select signals, and various feedback paths including add-by-one and subtract-by-one circuits from the output of the register to the various inputs of the multiplexer.

Each of the six output signals Q0-Q5 from the valid-queue-entry logic 46 enables or disables the thirty AND gates associated with one of the six registers 40-45 in the mask queue. Whenever one of these control signals PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -28is asserted, it signifies that a valid mask is present in the corresponding register of the mask queue 23f; consequently, the thirty AND gates associated with that register are enabled to allow the mask in that register to become a part of the composite mask that is used to determine whether a register conflict exists. For example, the assertion of output signal Q0 from the logic 46 enables the thirty AND gates A1-A30 associated with the corresponding register 40 in the mask queue 23f. It will be appreciated that any number of the six output signals Q0-Q5 may be asserted at the same time, depending upon the number of valid entries in the mask queue.

From the AND gates A1-A30, B1-B30, etc., the read and write masks are supplied to thirty parallel OR gates OR1-OR30 whose outputs are connected to the multiplexers 60, 61 and 62 shown in FIG. 7 (to be described below). Each OR gate has six inputs so that masks from any of the six registers 40-45 in the mask queue can be passed on to the multiplexers 60-62. Of course, if the bit corresponding to a given GPR is set in any one of the six registers in the mask queue, that bit will result in the assertion of the output signal from the OR gate corresponding to that particular GPR. That is, each of the thirty OR gates OR1-OR30 corresponds to one of the thirty bit positions in each of the registers 40-45, and each OR gate receives inputs from each of the six registers 40-45 in the mask queue. Thus, the outputs of the OR gates OR1-OR30 represent composite read and write masks for all the instructions represented by entries in the mask queue. Specifically, the fifteen OR gates 0R1-0R15 produce a composite write mask, and the fifteen OR gates OR16-OR30 produce a composite read mask.

The six registers 40-45 in the mask queue are PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -29controlled by clock enable (CE) signals from a decoder 53, as illustrated in FIG. 5. The 30-bit composite read/write mask from the decoder 20f is delivered to the six registers 40-45 over a bus 54. The decoder 53 receives the 3-bit insert pointer signal from the counter 47, and thus the decoder 53 enables only one of the registers 40-45 to receive the composite read/write mask present on the bus 54 at any given time. The decoder 53 is sequenced each time a new instruction is decoded, by applying to the output enable (OE) line of the decoder 53 the same new-instruction-decoded signal that increments the insert pointer counter 47 and the size counter 48.

From the mask queue 23f, the composite read and write masks in the six registers 40-45 are supplied via the thirty OR gates OR1-OR30 to the three multiplexers 60-62 (FIG. 7) so that the composite masks can be used in three simultaneous comparisons to determine whether any register conflicts exist. The first multiplexer 60 receives the composite read mask from the OR gates OR16-OR30, and compares that mask with a base-register number from the most recently decoded specifier. This read mask identifies GPRs to be read by the execution unit 13 during execution of all previously decoded instructions which have not yet been fully executed; if any of these GPRs were to be prematurely modified by the preprocessing of an instruction in the instruction unit 12, the wrong data would be read during the subsequent execution of preceding instructions in the execution unit. Such premature modifications could be effected by autoincrementing or autodecrementing of the GPRs in question. Normally there can be only one register per operand that is autoincremented or autodecremented, and this register is identified by the base register PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -30number, which is the number supplied to the multiplexer 60 for comparison with the composite read mask.

When the multiplexer 60 detects that the register 5 identified by the base-register number is a register having a set bit in the composite read mask, the Q output of that multiplexer is asserted. This output is one of the inputs to an AND gate 63, which also receives as inputs a specifier valid signal indicating whether the base register specifier is valid, and a specifier write signal indicating whether that specifier is auto increment or decrement. When all three inputs are asserted, a register read conflict exists, and the output of the AND gate 63 is asserted to indicate that fact.

The second multiplexer 61 receives the composite write mask from the OR gates OR1-OR15, and compares that mask with the same base-register number supplied to the multiplexer 60, i.e., the base-register number from the most recently decoded specifier. This write mask identifies GPRs to be written by the execution unit 13 during execution of all previously decoded instructions which have not yet been fully executed; if any of these GPRs were to be prematurely read during the preprocessing of an instruction in the instruction unit 12, the wrong register data would be read and used as an operand address for that instruction, and consequently the wrong operand data would be fetched from memory and used during the subsequent execution of that instruction in the execution unit.

When the multiplexer 61 detects that the register identified by the base-register number is a register having a set bit in the composite write mask, the Q output of that multiplexer is asserted. This output is PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -31one of the inputs to an AND gate 64, which also receives as inputs the same specifier valid signal supplied to the AND gate 63, and a specifier read signal indicating whether that specifier uses the base register for address generation. When all three inputs are asserted, a register write conflict exists, and the output of the AND gate 64 is asserted to indicate that fact.

The third multiplexer 62 receives the same 10 composite write mask supplied to the multiplexer 61, and compares that mask with an index-register from the most recently decoded specifier. As mentioned above, this write mask identifies GPRs to be written by the execution unit 13 during execution of all previously decoded instructions which have not yet been fully executed, and if any of these GPRs were to be prematurely read by the instruction unit 12 before being written by the execution unit, the wrong register data would be read.

When the multiplexer 62 detects that the register identified by the index-register number is a register having a set bit in the composite write mask, the Q output of that multiplexer is asserted. This output is one of the inputs to an AND gate 65, which also receives as inputs an index register specifier valid signal indicating whether the index register specifier is valid. When both inputs are asserted, the output of the AND gate 65 is asserted to indicate that a register write conflict exists. This register conflict signal is used to stall the preprocessing operation currently in progress until the register conflict is cleared.

The outputs of the three AND gates 63-65 are connected to a single OR gate 66 which passes an asserted output from any of the three AND gates 63-65 to a fourth PD88-0257 U.S.: DIGMzOll FOREIGN: DIGM:042 -32AND gate 67 which controls the setting of an output latch 68. More specifically, the fourth AND gate 67 receives as its second input a control signal which is asserted only when the detector first receives new specifiers.

This prevents the detection of an intra-instruction conflict which could occur if the specifiers for a newly decoded instruction were compared with a composite read or write mask which includes a mask generated for that same instruction.

In order to hold specifier information during a stall, double buffering is provided between the pipeline stages. Thus, as illustrated in FIG. 8, in addition to the conventional transfer register 80 which passes information between pipeline stages, there is provided a holding register 81 which holds the information during a stall. The output of the transfer register 80 is supplied to both a current·· register conflict detector 82 and a new register conflict detector 83. When a register conflict is detected, the output of the detector 82 (1) is passed through a pair of OR gates 84 and 85 to stall instruction decoding, and (2) causes the operand unit 21 to supply a stall signal via line 86 to the select input of a multiplexer 87 which determines whether the detector 82 receives its input from the transfer register 80 or the holding register 81.

Normally, the multiplexer 87 passes the specifiers from the transfer register 80 to the detector 82, but a stall signal from the operand unit on line 86 causes the multiplexer to select the holding register 81. Thus, the specifier which initiates the conflict is maintained as an input to the detector 82 to enable that detector to determine when the conflict has been eliminated, even though the data in the transfer register 80 changes.

PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -33The new register conflict detector 83 permits the detection of a new register conflict due to a specifier received by the transfer register while the detector 82 is still connected to the holding register 81. The detection of such a new conflict causes the output of the detector 83 to be asserted and applied via the OR gates 84 and 85 to the stall line leading to the instruction decoder. As indicated in FIG. 8, the OR gate 87 also receives stall signals from other sources, such as the full signals from the queues 23 and the source list 24.

PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -34MULTIPLE INSTRUCTION PREPROCESSING SYSTEM WITH DATA DEPENDENCY RESOLUTION

Claims

1. 5 CLAIMS

1. A method of preprocessing multiple instructions prior to execution of such instructions in a digital computer having an instruction decoder, an instruction execution unit, and multiple general purpose registers, 10 and in which said general purpose registers are read to produce memory addresses during the preprocessing of said instructions, said method comprising avoiding the preprocessing of a current instruction 15 to read a general purpose register to produce a memory address prior to the modification of the contents of that register by a preceding instruction by generating a composite write mask having a bit set 20 for each general purpose register whose contents are to be modified by at least one of a plurality of decoded but not-yet-executed instructions preceding said current instruction, and 25 stalling the processing of said current instruction when a general purpose register to be read by said current instruction is a register having a bit set in said composite write mask.

2. The method of claim 1 which includes the step of updating said write mask each time the execution of an instruction is completed, and each time a new instruction is decoded. PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -353. The method of claim 2 which includes the step of resuming the preprocessing of said current instruction when said write mask no longer has a bit set for any 5 general purpose register to be read by said current instruction.

3. 4. The method of claim 1 wherein said write mask 10 is generated by producing a write mask for each instruction that is decoded, queuing the write masks for a plurality of decoded instructions in a mask queue until the execution of those instructions is completed, and combining all the queued write masks to generate said 15 composite write masks. 5. The method of claim 4 wherein the write mask for each instruction is removed from said mask queue when the execution of that instruction has been completed. 6. The method of claim 1 wherein the decoding of new instructions is stalled whenever said mask queue is full. 7. The method of claim 1 wherein a selected number of said write masks are retained in said mask queue in response to a flush signal. 8. The method of claim 1 wherein said mask queue includes multiple registers, and said write masks for successive instructions are stored in successive registers in said mask queues. PD88-0257 U.S.: DIGMzOll FOREIGN: DIGMZ042 -369. The method of claim 8 which includes the step of generating an insert pointer identifying the register which is to receive each new write mask in said mask

4. 5 queue, and incrementing said insert pointer each time a new instruction is decoded.

5. 10. The method of claim 8 which includes the steps 10 of generating a size signal identifying the number of write masks stored in said mask queue, incrementing said size signal each time a new instruction is decoded, and decrementing said size signal each time the execution of an instruction is completed.

6. 11. The method of claim 1 wherein each instruction includes at least one source-operand specifier and at least one destination-operand specifier, and at least 20 certain of said specifiers include general purpose register specifiers, and which includes the step of comparing selected general purpose register specifiers in said current instruction with said 25 composite write mask to determine when a general purpose register to be read by said current instruction is a register having a bit set in said composite write mask. 30 12. The method of claim 11 which includes the step of holding a general purpose register specifier which produces an affirmative response in said comparing step, and continuing said comparison to determine when the affirmative response is cleared by removal of a write 35 mask from said mask queue. PD88-0257 U.S.: DIGMiOll FOREIGN: DIGM:042 -3713. The method of claim 1 wherein said computer has n general purpose registers, each instruction 5 includes at least one source-operand specifier and at least one destination-operand specifier, and at least certain of said specifiers include general-purpose-register specifiers, and wherein said composite write mask is generated by converting the general-purpose-register specifier for each general purpose register to be read by an instruction to a set bit on one of n lines corresponding respectively to said n general purpose registers.

7. 14. A method of preprocessing multiple instructions prior to execution of such instructions in a digital computer having an instruction decoder, an 20 instruction execution unit, and multiple general purpose registers, and in which said general purpose registers are read to produce memory addresses during the preprocessing of said instructions, said method comprising avoiding the preprocessing of a current instruction to modify the contents of said general purpose registers to produce a memory address prior to the reading of that register by a preceding instruction by generating a composite read mask having a bit set for each general purpose register to be read by at least one of a plurality of decoded but not-yet-executed instructions preceding said current instruction, and PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -38stalling the processing of said current instruction when a general purpose register whose contents are to be modified by said current instruction is a register having a bit set in said composite read mask.

8. 15. The method of claim 14 which includes the step of updating said read mask each time the execution of an instruction is completed, and each time a new instruction 10 is decoded.

9. 16. The method of claim 15 which includes the step of resuming the preprocessing of said current instruction 15 when said read mask no longer has a bit set for any general purpose register to be modified by said current instruction. 20

10. 17. The method of claim 14 wherein said read mask is generated by producing a read mask for each instruction that is decoded, queuing the read masks for a plurality of decoded instructions in a mask queue until the execution of those instructions is completed, and 25 combining all the queued read masks to generate said composite read masks.

11. 18. The method of claim 17 wherein the read mask 30 for each instruction is removed from said mask queue when the execution of that instruction has been completed.

12. 19. The method of claim 14 wherein the decoding of 35 new instructions is stalled whenever said mask queue is PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -39full.

13. 20. The method of claim 14 wherein a selected 5 number of said read masks are retained in said mask queue in response to a flush signal.

14. 21. The method of claim 14 wherein said mask queue 10 includes multiple registers, and said read masks for successive instructions are stored in successive registers in said mask queue. 15

15. 22. The method of claim 21 which includes the step of generating an insert pointer identifying the register which is to receive each new read mask in said mask queue, and incrementing said insert pointer each time a new instruction is decoded.

16. 23. The method of claim 21 which includes the steps of generating a size signal identifying the number of read masks stored in said mask queue, incrementing

17. 25 said size signal each time a new instruction is decoded, and decrementing said size signal each time the execution of an instruction is completed. 30 24. The method of claim 14 wherein each instruction includes at least one source-operand specifier and at least one destination-operand specifier, and at least certain of said specifiers include general purpose register specifiers, and which includes the step 35 of PD88-0257 U.S.: DIGM:Oil FOREIGN: DIGM:042 -40comparing selected general purpose register specifiers in said current instruction with said composite read mask to determine when a general purpose 5 register to be modified by said current instruction is a register having a bit set in said composite read mask. 25. The method of claim 24 which includes the step 10 of holding a general purpose register specifier which produces an affirmative response in said comparing step, and continuing said comparison to determine when the affirmative response is cleared by removal of a read mask from said mask queue.

18. 26. The method of claim 14 wherein said computer has n general purpose registers, each instruction includes at least one source-operand specifier and at 20 least one destination-operand specifier, and at least certain of said specifiers include general-purpose-register specifiers, and wherein said composite read mask is generated by 25 converting the general-purpose-register specifier for each general purpose register to be read by an instruction, to a set bit on one of n lines corresponding respectively to said n general purpose registers.

19. 27. A method of preprocessing multiple instructions prior to execution of such instructions in a digital computer having an instruction decoder, an instruction execution unit, and multiple general purpose 35 registers, and in which said general purpose registers PD88-0257 U.S.: DIGM:011 FOREIGN: DIGM:042 -41are read to produce memory addresses during the preprocessing of said instructions, said method comprising 5 avoiding the preprocessing of a current instruction to read a general purpose register, or to modify the contents of a general purpose register, to produce a memory address prior to the reading or modification of the contents of the respective registers by a preceding 10 instruction by generating a composite read/write mask having a bit set for each general purpose register whose contents are to be read or modified by at least one of a plurality of 15 decoded but not-yet-executed instructions preceding said current instruction, and stalling the processing of said current instruction when a general purpose register whose contents are to be 20 modified by said current instruction is a register having a bit set in said composite read mask.