System and Method for Addressing Plurality of Data Values With a Single Address in a Multi-Value Store on FIFO Basis

The invention relates to addressing instructions for a computer system and methods of using a computer system with addressing instructions.

Common operations by a processor in a computer system require identification of one or more sources of data to form an input to the processor as well as a destination for receiving the result of operating the processor. Conventional reduced instruction set computers (RISC devices) have used instructions all of a common bit length and format defining both the operation to be carried out by the processor as well as identification of two sources of data for use in the operation, and a destination for the result of the operation. Such instructions have been used to manipulate data that is found in addressable store locations such as registers or memory. When using a processor in a pipelined operation to execute a sequence of instructions it may be necessary to hold a plurality of live data items which are accessible by subsequent instructions. This requires sufficient number of addressable locations to be identified by an instruction that it either imposes a limit on the shortness of the instructions that can be used or a limit on the number of address locations that can be covered by an instruction set.

Other systems are known with complex instructions, (CISC devices) and such long instructions have provided the facility for identifying the addresses of more data stores used for holding live data during the pipelined processing of an instruction sequence. However such long instructions have required greater access time in obtaining the instructions from memory and may involve more extensive decoding needing more cycles of operation to achieve decoding of each instruction.

It is an object of the present invention to provide a computer system with instructions having a reduction in the number of address bits relative to the number of data items that may be held during instruction execution.

The present invention provides a computer system having logic circuitry responsive to an instruction set comprising selectable instructions, a plurality of said instructions each including one set of bit locations identifying an operation to be carried out by execution of the instruction and a second set of bit locations to identify an address of a data storage location for use in execution of the instruction, said computer system further including a plurality of addressable data storage locations for holding simultaneously a plurality of data values during execution of a sequence of instructions, at least one of said data storage locations comprising a multivalue store requiring a single address in an instruction and arranged to hold a plurality of data values simultaneously on a first in first out basis, thereby increasing the number of data values that can be held in relation to the number of addresses that can be identified by said second set of bit locations.

Preferably the multivalue store is provided by a first in first out buffer.

In some embodiments a plurality of first in first out buffers are provided each addressable by a respective different address.

Preferably said computer system includes a plurality of addressable registers and at least one addressable first in first out buffer, said registers and said buffer being selectively addressable by the second set of bit locations in each of the plurality of instructions.

Preferably each of the plurality of instructions includes more than one address indication, at least one address indication indicating a source of data for use in execution of the instruction and another set of bit locations indicating a destination for a result of execution of the instruction.

The multivalue storage location may provide a destination location or a source location.

Each instruction may be of the same bit length while in some embodiments at least some instructions in the struction set may have different bit lengths.

The invention also provides a method of executing a succession of instructions in a computer system, the method comprising decoding one set of bit locations in each instruction to determine an operation to be carried out, decoding a second set of bit locations in each instruction to identify an address of a data storage location for use in execution of the instruction, executing each instruction and holding simultaneously a plurality of data values for use in execution of an instruction sequence. To minimise the number of bits necessary in the second set of bit locations in relation to the number of data values to be held during execution of the sequence, the second set of bit locations in at least some instructions address a multivalue storage location arranged to hold a plurality of data values on a first in first out basis.

Preferably data resulting from execution of a plurality of instructions is written into the multivalue storage location in succession.

Preferably the multivalue storage location is used as a source of data in execution of an instruction, the data having been written into the multivalue storage location as a result of execution of a preceding instruction.

Preferably said instruction sequence is arranged to address a plurality of data storage locations some of which are provided by registers arranged to hold a single data value and at least one of which is provided by said multivalue storage location.

In some embodiments selected instructions include respective addresses of at least two multivalue storage locations each arranged to hold a plurality of data values on a first in first out basis.

Some embodiments of the present invention will now be described by way of example and with reference to the accompanying drawings.

FIG. 1 shows the format of a prior art instruction,

FIG. 2 shows a variety of instruction formats some for use in accordance with the present invention,

FIG. 3 is a block diagram of a computer system in accordance with the present invention,

FIG. 4 shows more detail of part of the system shown in FIG. 3,

FIG. 5 shows more detail of the instruction expansion fetcher shown in FIG. 3,

FIG. 6 shows in more detail the pipe unit of FIG. 3,

FIG. 7 shows a modified arrangement for the circuitry of FIG. 5, and

FIG. 8(I) & 8(II) shows a modification of the system shown in FIG. 3.

FIG. 1 shows an instruction format for a conventional RISC processor system in which all instructions have a fixed length and generally similar format of bit locations in the instruction. The example shown consists of a single instruction 11, which in this example is 4 bytes long, each byte being 8 bits long. The first byte 12 defines the opcode which defines the particular operation to be carried out by the processor when executing this instruction. The second byte 13 provides the address of the destination for the result of carrying out this instruction. Bytes 14 and 15 designate respectively the addresses of first and second sources of data to be used in carrying out the instruction. Typically bytes 13, 14 and 15 will all designate register addresses and many operands in instructions of this type are the addresses of temporary registers used to hold live data used in execution of an instruction sequence.

It will be seen that the number of addresses which may be indicated in such an instruction set is limited by the number of address bits provided in each instruction. An increase in the number of addresses requires an increase in the bit length thereby taking up more memory space and taking up more memory access time in obtaining the instructions for use by the processor.

A variety of formats of instruction for use by a single processor is shown in FIG. 2. This particular example uses a plurality of selectable instructions of different bit length, each being based on a format of predetermined bit length and a predetermined sequence of instruction fields each of a respective predetermined bit length, some of the instructions omitting a selected one of the fields and including an identifier of less bit length than the omitted field to indicate which field is omitted, thereby compressing the bit length of the instruction and avoiding the use of redundant bit locations in the instructions. In this particular example shown, the instructions are based on a 32 bit format divided into 4 bytes 20, 21, 22 and 23 each of 8 bits. This format will be described with reference to format 12 shown in FIG. 2 as this illustrates the basic uncompressed format. The first byte 20 has the first two bit locations allocated to a length indicator 24. The next 6 bits form field 1 which is an opcode indicator 25 indicating the particular operation which the processor is to carry out in response to execution of the instruction. The first field in the instruction format is provided for all instructions at the same bit locations and always indicates the opcode.

In the second byte 21 the first two bit locations form an indicator 26 indicating the type of instruction. These two bits 26 taken together with the instruction bit length indicator 24 provide an indication of which fields are present and which have been omitted in the instruction. The next 6 bits of the second byte 21 form a second field 27. In the third byte 22 the first 6 bits define a third field 28. The last two bits 29 and 30 can, depending on the type of instruction, provide extensions of the opcode indicator or type of instruction indicator or part of a constant. In the fourth byte 23 the last 6 bits form the fourth field 31. The first two bits 32 of the last byte 23 may form part of constant values or extensions of the opcode indicator. FIG. 2 shows 15 different formats which may all be based on the four field format shown in format 12 of FIG. 2. Each of the selectable formats either fully includes or fully omits each of the second, third or fourth fields. The type indicator 26 (and extension bit 30) in combination with the length indicator 24 provides an indication of which fields have been omitted. They may also provide an indication of the designation of data in an included field thereby indicating a register or a constant value. In addition to allowing compression of the bit length of each instruction to avoid redundant bits, this example allows the identification of an implicit data storage location which does not require explicit addressing. The basic format of the instructions used is best illustrated by reference to format 12 in FIG. 2. The length indicator 24 indicates that it is a 32 bit long instruction. The opcode 25 defines the operation to be carried out. The type indicator 26 indicates that in the case of a 32 bit long instruction the second field 27 will identify the address of a destination data store, the third field 28 will identify the address of a second data source and the fourth field 31 will identify the address of a first data source. Each of the data sources and the data store may comprise a register or a store requiring a single address in an instruction and arranged to hold simultaneously a plurality of data values on a first in first out basis. The last two bits of the third byte 22 and the first two bits of byte 23 will in this case provide the facility for an extension 32 of the opcode 25. For some instructions it is not necessary to identify two source addresses and a destination address and in such cases one or more of the second, third or fourth fields can be omitted so as to compress the bit length of the instruction and avoid redundant bit locations. Furthermore, the example described includes the use of an implicit data store for holding live data during a program execution sequence where the data store does not require any explicit addressing. In this example the implicit store is provided by a first in first out data store arranged to hold simultaneously a plurality of data values and in this example such a data store is referred to as a pipe. The pipe may be used as a destination or as a first or second source of data, or any combination of them, for use in execution of the program sequence. If the pipe is to be used for any of the destination, source 1or source 2 addresses then no entry is required in the corresponding second, third or fourth fields of the instruction format shown in FIG. 2. In the case of format 1 the instruction length indicator 24 indicates that the instruction is only 1 byte long and consequently fields 2, 3, and 4 are missing, thereby indicating that the implicit pipe is to be used for both source and destination addresses. In the case of formats 2, 3, 4 and 5, the bit length indicator indicates that the instruction is now 2 bytes long and the type indicator 26 distinguishes between formats 2, 3, 4 and 5. In format 2 field 2 indicates a destination address. In format 3 field 3 indicates a second source address and in format 4 field 4 indicates the first source address. In each of formats 2, 3 and 4, the implicit pipe is understood to replace the omitted addresses. When expanded, each of formats 1 to 4 will take the form of format 12. Format 5 shows a situation where the type indicator shows that field 4 provides a constant value instead of a first source address. Format 5 will be expanded to format 13. Formats 6 to 11 each have a length indicator 24 indicating that the instructions are 3 bytes long and the type indicators 26 together with the last bit location 30 indicate the significance of the included fields as well as an indication of the omitted field. In each case the implicit pipe is used in place of any omitted address. Formats 6, 7 and 8 will be expanded to format 12. Formats 9 and 10 will be expanded to format 13. Format 11 will be expanded to format 15. In the case of formats 12 to 15 the length indicator indicates that the instruction has a length of 4 bytes. In the case of formats 13, 14 and 15, the instruction omits one or more explicit addresses and the bit locations are replaced by constants 33, 34 or 35. On expansion the implicit pipe address will be provided for each missing field. Formats 11 and 15, which omit all address indications, are not based on the present invention. In the subsequent description the bit value provided by bits 10-15 is referred to as Arg 1, the bit value of bits 16-21 is Arg 2 and the bit value of bits 26-31 is Arg 3. Eight examples of instruction having 1, 2 or 3 Arguments are illustrated in the following table which also shows the source and destination addresses indicated by each of these instructions as well as the byte length and format corresponding to FIG. 2.

It will therefore be seen that in the compressed instruction formats shown in FIG. 2 the type indicator 26, including when appropriate the extension 30, has a bit length substantially less than the omitted fields 2, 3 and 4. The use of predetermined field positions each of predetermined bit length provides simplicity and thereby speed in decoding. Each instruction is compressed to a minimum bit length to indicate the information necessary for execution of the instruction. Consequently memory space for storing an instruction sequence and access time for transferring instructions from memory to the processor are minimised. Each instruction fetch operation from memory will load into the processor a fixed number of bytes and in the case of compressed instructions the overall byte package fetched from memory may include part of a subsequent instruction occupying bit locations unused by the next instruction.

In the above formats 2 to 10 and 12 to 14, any explicit address may designate a register or a pipe but at least one explicit address will identify a pipe. More than one pipe may be used, in which case different explicit addresses are allocated to the plurality of pipes. By designating at least one pipe, which is a multi value store operating on a first in first out basis, the number of data items which can be stored for a given number of address bits in each instruction is increased. Alternatively the bit length of the instruction can be shortened for the same number of data items to be stored.

The pipe address may be given as an explicit address in an included field when desired.

A computer system for use in executing instructions of the type shown in FIG. 2 is illustrated in FIG. 3. This shows processor circuitry 40 coupled to a memory 41 which in this example comprises a conventional RAM. The processor 40 includes an address bus 42, a data bus 43 as well as read and write controls 44 and 45 respectively, all coupled to the memory 41. The memory 41 is arranged to hold program comprising sequences of instructions at different addressable locations. Each instruction conforms to one of the formats 1 to 10 or 12 to 14 already described with reference to FIG. 2. The memory 41 may also hold data. The processor 40 includes control logic 50 as well as an ALU 51. The data bus 43 carries data values to and from the memory 41. The address bus 42 carries memory address values for read or write operations. Each addressable location in the memory 41 holds 4 bytes of data. The least significant 2 bits of address bus 42 are ignored in determining which memory location to access. The ALU 51 is connected to a source 1 bus 52 and a source 2 bus 53. These form inputs to the ALU 51 and any resulting output is provided on a result bus 54. Read and write lines 44 and 45 form outputs from the control circuit 50. When the read line 44 is asserted the memory 41 supplies to the data bus 43 the value stored at the location identified by the address bus 42. When the write line 45 is asserted the memory 41 will store the value on the data bus 43 into the memory location identified by the address bus 42.

The ALU 51 receives a command input 261 from the control unit 50 to determine which operation the ALU is to perform. The ALU provides a condition output 56 to the control unit 50 to indicate the condition of the ALU operation.

The processor 40 includes an instruction fetching unit 60 arranged to fetch 4 bytes at a time from the memory 41 and to provide a 4 byte output 61 to an instruction register 62 where the output 61 is aligned with the beginning of an instruction. The instruction register 62 is arranged to provide a plurality of separate outputs representing each of the fields and control bits of an instruction to an expand circuit 63 which is arranged to provide a fully expanded output for each instruction. Depending on the opcode of the instruction, the control unit 50 will cause source values to be put onto one or both source buses 52 and 53 from either the register file 170 or pipe 171 or constant register 167. The result of an ALU operation will then be written from the result bus 54 into a selected location of either the register file 170 or pipe 171 depending on the destination indicated by the instruction. The ALU 51 is also used for intermediate stages of instruction execution and the result may be written to the Instruction Pointer 230, the Temporary Register 240, the Fetcher 60 or the Write Buffer 252.

The construction and operation of the instruction fetcher 60 will be described first. The unit is shown in more detail in FIG. 4 and has an input data bus 64 which is connected to the data bus 43. The fetcher 60 includes a fetch pointer 65 which comprises a 32 bit latch containing the next address in memory from which a 32 bit word is to be read. When a value is read from memory, latch 65 is increased by 4 bytes to a new pointer address. A value 4 is generated by unit 105, which supplies the value 4 to an adder 106, which receives the pointer value from bus 103 and adds 4 to it. The total is then sent through multiplexer 92 to the latch 65 to update the pointer. It will be understood that an aligned instruction is one which has its bit 0 position at the beginning of a new 32 bit word. However due to the use of variable length instructions, some instructions will commence partway through a word in memory. The fetcher 60 is arranged to read compressed instructions from memory, correctly align them and then output them on bus 61 from a fetch buffer 66 to the instruction register 62. The 4 byte words which are input on bus 64 pass through a load shifter 67 through a plurality of parallel multiplexers 68 to the fetch buffer 66. The fetch buffer may hold up to 7 bytes of instructions which is sufficient to hold the worst case of instruction mis-alignment. A 4 byte instruction could start 1 byte into a 32 bit memory word such that the word contains 3 bytes of the instruction and a subsequent word of 4 bytes must be read in order to obtain the 4th byte necessary to complete the instruction. In this way 7 bytes need to be held in the buffer 66. Once an instruction is aligned in the buffer 66, four bytes are output on lines 70, 71, 72 and 73 representing byte 0, byte 1, byte 2 and byte 3. Depending on the length of the instruction, it may be that the instruction buffer 62 does not act on signals derived from all four lines 70-73 but in all cases outputs are provided on these lines. Byte 0 will indicate on line 70 the length of the instruction which is output. This byte is fed through a length indicator 74 to indicate to the fetcher unit 60 how many bytes in the buffer 66 can now be removed. A byte counter 75 holds a count of the number of bytes in the buffer 66 at any one time and the length signal derived from unit 74 is passed through a multiplexer 76 to a length subtractor 77 which receives inputs from multiplexer 76 and from the byte counter 75 in order to calculate the number of valid bytes remaining in the buffer 66. An output 78 from the subtractor 77 is supplied through a multiplexer 79 to update the input in the byte counter 75. The output from the multiplexer 76 forms an input 80 to an extract shifter 81 which is used to relocate in the buffer 66 those remaining bytes which will be required for use in the next instruction to be supplied on bus 61 to the instruction register 62. The extract shifter 81 receives inputs 82 from each of lines 71, 72, 73 as well as from the byte 4, byte 5 and byte 6 locations of the buffer 66. Depending on the Amount signal 80, the values held in the fetch buffer 66 are relocated by the extract shifter 81 and output on lines 83 to the multiplexers 68. In this way the bytes which are still required in the fetch buffer are relocated into the fetch buffer 66 at new locations starting at byte 0. The values held in byte 0, 1 and 2 of the buffer 66 are also fed back on lines 85 to the load shifter 67. The load shifter also receives an amount signal 86 derived from the byte counter 75 to indicate the byte locations within the fetch buffer 66 into which new bytes read from memory on data bus 64 are to be loaded. Depending on the Amount signal 86, four new bytes from memory are input on bus 64 and fed to the appropriate multiplexers 68 in combination with any signals on lines 85 giving byte values which are to be retained from a previous operation of the fetch buffer 66 so that the multiplexers 68 can be operated by a latch signal 90 to load into the buffer 66 the combination of retained bytes on any of lines 85 together with new bytes read from memory such that they occupy consecutive byte positions in the fetch buffer 66 starting at byte 0. The latch signal 90 is also fed to the multiplexer 79 to update the byte counter 75 by four. The value 4 is fed from unit 105 to an adder 108 which adds 4 to the byte count fed in from the byte counter 75 and feeds the total to the byte counter 75.

Normally the fetch pointer 65 will cause the fetcher 60 to obtain instructions from sequential word locations within memory. However some programs will contain branch instructions requiring the mechanism to start fetching instructions from a new location in memory. For this reason a Start input 91 is provided which can be fed through the multiplexer 92 to the fetch pointer 65. The top 30 address bits of the start address on line 91 are fed to the pointer 65 as those 30 bits will always address the beginning of a new 4 byte word in memory. The bottom two bits of the Fetch pointer 65 are always set to zero. The bottom 2 bits which will only indicate a sub-position within a 32 bit word are stored in a 2 bit latch 93. The value held in the latch 93 is tested in a comparator 94 to detect inequality with zero and thereby decide whether or not the new memory address does start at the beginning of a 32 bit boundary or partway through a word. In the event of it starting partway through a word a signal is provided from the comparator 94 to the AND gate 95 to indicate, when a second input to the AND gate 95 from a comparator 104 indicates that the number of bytes in the buffer 66 is not zero, that some adjustment of the contents of the fetch buffer is necessary. An Adjust output 96 is provided to the control unit 50 indicating that the instruction is not correctly aligned and to align it a dummy instruction read is carried out using a Next signal 97 from the control unit 50. The signal 97 is fed through an OR gate 98 to cause a new latching operation for the fetch buffer 66 which as described above will cause removal of one or more bytes from the fetch buffer 66 and relocation by an amount determined by the input 80 to the extract shifter 81, the amount being derived from the two bit latch 93 which is routed through the multiplexer 76 by the output of the AND gate 95.

It will be appreciated that the Next input 97 is also derived from the control unit 50 to inform the fetcher unit that the current instruction has been read from the fetch buffer 66 and consequently that instruction can be removed from the Fetch buffer 66 as described above. A latch input 100 is derived from the control unit 50 which is an edge sensitive signal to instruct the fetcher to store the value on the data bus 64 into the buffer 66. A Begin signal 102 is a further edge sensitive signal used to inform the fetcher to start fetching a new word from a new address presented on the Start signal input 91 in response to a branch instruction. It will be appreciated that the address output 103 always points to a 32 bit boundary at the beginning of a word in memory. The byte counter 75 is connected to the count comparator 106 to provide a signal to the AND gate 95 provided the byte count is not 0.

The length subtractor 77 is arranged to provide a More output 107 whenever the fetch buffer has insufficent content to produce the next instruction. This signal 107 is supplied to the control unit 50 to cause the fetcher to perform a memory read using the latch signal 100 before the next instruction can be read from the fetcher. The byte counter 75 has the count reduced whenever an instruction is removed from the fetch buffer 66 and increased whenever a new value is read from memory. When the fetcher starts from a new instruction sequence using the Start input 91, the Begin signal on line 102 is used to clear the count in the counter 75, and via an OR gate 109 is used to latch new pointer values in the fetch pointer 65. A single bit latch 110 is used to clear the 2 bit latch 93 after the first instruction from a new start address has been aligned in the buffer 66.

It will be understood that the three input signals Next 97, latch 100 and Begin 102, which are derived from the control unit 50, are not usually asserted until some operation of the fetcher is required. The four types of operation which it may perform are as follows. It may produce instructions on the output bus 61 to the instruction register 62. It will do this when a correctly aligned instruction is held in the fetch buffer 66 and the More Signal 107 and Adjust signal 96 are not asserted. The fetcher may perform a read operation from memory locations. Normally 32 bit memory locations are read sequentially whenever the fetch buffer 66 does not contain enough data. The address for which data is read will be indicated by the fetch pointer 65 output on bus 103. This will be done when the More signal 107 is asserted and the Adjust signal 96 is not asserted. The latch signal 100 will be asserted to set the multiplexers 68 to select outputs from the load shifter 67. Once data is input from bus 64 the latch signal 100 is deasserted, in order to store the value in the fetch buffer 66.

The fetcher can be used to initialize a new instruction sequence. In this case the byte count 75 is cleared and a new address in memory is latched into the fetch pointer 65 and 2 bits buffer 93. This initialization will be required after a program branch instruction. It is performed by asserting and then deasserting the Begin signal 102. Furthermore, the fetcher can be used to align the first instruction in a new instruction sequence. If a new instruction sequence does not start at a 32 bit word boundary in memory then the Adjust signal 96 will be asserted and the alignment operation will involve asserting and then deasserting the Next signal 97 in order to perform a dummy read of the fetch buffer 66, and thereby align the first instruction of the row sequence.

As shown in FIG. 3, the output bus 61 of the fetcher unit 60 provides a 32 bit signal which is fed into the instruction register 62, and the 32 bit value is latched by an input signal 112, which is derived from the control unit 50 at the same time as the Next signal 97 which was fed to the fetcher 60. The 32 bit signal in register 62 is arranged to provide seven output signals 120, 121, 122, 123, 124, 125 and 126. The nature of these output signals will be described with reference to FIG. 2. Signal 120 represents bits 0 and 1 corresponding to the length indicator in FIG. 2. Signal 121 is derived from bits 2 to 7 and represents field 1 which contains the opcode. Output 122 represents bits 8 and 9 which form the type indicator 26 shown in FIG. 2. Output 123 is Arg 1 representing bit positions 10 to 15. Output 124 is Arg 2 representing bit positions 16 to 21. Output 125 represents bit positions 22 to 25 and is referred to as PAD and represents a type extension, part of the opcode or part of a constant value. Output 126 represents Arg 3 and represents bit positions 26 to 31. It would therefore be seen that output 121 represents field 1 in FIG. 2 and always indicates the opcode. Fields 2, 3 and 4 referred to in FIG. 2 will be represented by either Arg 1, Arg 2 or Arg 3 depending on the length of the instruction. Each of the outputs 120-126 are fed to an instruction expansion unit 63 which is arranged to reconstruct the full instruction from the compressed form supplied to it. The length indicating output 120 is also fed to a length enable unit 127 connected to the Source 1 bus 52.

The construction and operation of the expansion unit 63 will be described with reference to FIG. 5. A unit 130 is provided to construct the full opcode. Unit 130 receives the opcode input 121 and is also connected to the PAD input 125 as in some formats the PAD signal may include an opcode extension. The length input 120 and type input 122 are both connected to the unit 130 in order to determine whether the unit 130 incorporates any signal from the PAD input 125 in constructing the opcode. The fully constructed opcode is then output on line 131 which is fed to the control unit 50 to control the operation carried out by the processor. A router unit 132 is arranged to receive each of the Arg 1, Arg 2 and Arg 3 inputs 123-126. The router 132 also receives inputs from the length and type signals 129,122 so as to determine which of the inputs 123, 124 and 126 represent Destination or Source 1 or Source 2 outputs. The router 132 also received the PAD signal 125 as in some formats this contains a TYPE extension. The router 132 has a first output 140 corresponding to a destination address a second output 141 corresponding to a Source 2 address and a third output 142 corresponding to a Source 1 address. The length and type inputs 120 and 122 and PAD input 125 which are fed to the router 132 enable the router to output the correct input signals 123 to 126 or the respective outputs 140 to 142 and these are fed to respective multiplexers 143, 144 and 145. The multiplexers each receive an alternative address indicating the pipe address determined by a pipe addressing unit 146. The operation of the multiplexers 143-145 is controlled by an enable unit 147 having inputs from the length and type indicators 120 and 122 and PAD 125. Those input signals determine whether or not any of the fields in the compressed instruction are missing and it replaces any missing field by the implied pipe address 146. In this way output signals 150, 151 and 152 from the expansion unit indicate an appropriate destination address Source 1 and Source 2 address with the pipe address being provided for any implied address not explicity given in the compressed instruction. In the event that the instruction includes a constant value as is indicated in the formats of FIG. 2, then any of Arg 1, Arg 2 and Arg 3 as well as PAD may be used to construct the constant dependent on the length and type indicators 120 and 122. These are fed to a constant constructor 163. If the input signals 120 and 122 indicate that a constant is to be constructed the inputs 123 to 125 are combined as required to output a constant value on output 164. The length and type indicators 120 and 122 are also fed to an indicator unit 165 to output a signal 166 indicating that a constant is present. Signal 166 is fed to the control unit 50 and the constant value is fed a line 164 to a constant register 167. When the PAD signal 125 contains a TYPE extension, its value does not determine whether a constant is encoded as is shown in FIG. 2.

The processor includes a register file 170 holding a set of addressable registers. In addition the implied data storage location is provided by the pipe 171 which provides a first in first out data store for a plurality of data values. The register file 170 has a pointer circuit 173 connected to the bus 54 to allow data values to be written into the register file. Similarly the pipe 171 has a selector circuit 174 to allow data values to be written from the result bus 54 into a desired location in the pipe 171. Both the pointer circuit 173 and selector circuit 174 are arranged to receive a latch signal 184 from the control unit 50 when it is required to write data into a storage location. Similarly both the pointer 173 and selector 174 are connected to the output 150 of the instruction expansion unit 63 so that the appropriate store is used as a destination for data when so indicated by the output of an expanded instruction from the expansion unit 63. To allow data to be read from the register file 170 onto the source buses, an output selector 175 is provided which is connected by line 176 to the source 1 bus 52 and by a connection 177 to the source 2 bus 53. The correct register for use in a read operation is determined by the source 1 or source 2 addresses provided by the expansion unit 63. Line 151 provides a direct connection of the source 2 address to both the output selector circuit 175 of the register file 170 and to an output selector circuit 179 of the pipe 171. The source 1 address which is output on line 152 from the expansion unit 63 passes through a multiplexer 180 having its output connected to both selectors 175 and 179. The multiplexer 180 also has an input from the destination line 150 from the expansion unit 63 so that the destination address can be supplied in place of a source 1 identification. The selector 175 has a first enable input 182 to enable a read operation onto the source 1 bus. The enable signal 182 is derived from the control unit 50 through an AND gate 185. A second enable input 186 is arranged to enable a read operation onto the second source bus 53 and is derived directly from the control unit 50. The enable inputs 182 and 186 are similarly provided to the output selector 179 of the pipe 171.

The construction and operation of the register file 170 is conventional.

The construction and operation of the pipe 171 will be described in more detail with reference to FIG. 6. The pipe 171 comprises a plurality of separately selectable data stores 190 arranged to hold N data values. The result bus 54 is connected to each of the data stores 190 and the destination selector 174 is coupled to each data input to determine which data store may accept an input from the result bus 54. Similarly each data store 190 is collected to two separate output selectors 191 and 192. The two output selectors are similar and are each connected to each data store location and may connect any one of the data locations to the respective source 1 bus 52 or source 2 bus 53. When writing to the pipe the destintation signal 150 is input to the pipe where the destination address is compared in a match unit 194 with a designated address for the pipe to ensure that the instruction requires use of the pipe as the destination. It will be understood that the destination signal will also be fed to the input selector for the register file 170 but will be ignored by the register file 170 as no match will be found. Provided a match is found by unit 194 an input is provided to AND gate 195 also arranged to receive the latch signal unit 184 from the control unit 50 so that the relevant destination is only selected when the correct latch signal is asserted as well. The selector 174 operates in a cyclic manner to write data into successive data locations of the pipe 171. An In Pointer 196 always contains the address in the pipe 171 of the next entry to write to. This address is input on line 197 to the selector 174 so as to select the required address for the next writing operation. Signal 197 is also fed to an adder 198 having as a further input a signal from an Add 1 unit 200. The output of the adder is fed back to the pointer circuit 196 which receives a latching signal from the AND gate 195 so as to advance the pointer address by one after each new latching operation to write data into the pipe 171. When a value is to be written into the pipe, the latch signal 184 is initially asserted. The output of the AND gate 195 is fed to a latch circuit 201 in the pipe 171. When the latch signal 184 is deasserted the value of the result bus 54 is stored in the selected location of the pipe 171 and the address of the next entry to write to is stored in the pointer 196.

During a read operation, the source 1 and source 2 addresses on lines 151 and 183 are fed to the register file 170 and to the pipe 171. As the designated pipe address will not match an address in the register file 170 reads from the pipe will be ignored by the register file 170. The input addresses are compared in the pipe 171 with designated addesses of the pipe in further match units 205 similar to match unit 194. The output of each match unit 205 is fed to an AND gate 206 which receives the respective enable input 182 or 186. In the case of the source 1 circuitry, the output of the AND gate 206 forms an enable signal 207 to the source 1 selector 190 as well as an input to an OR gate 209 and an AND gate 210. An Out Pointer circuit 211 is arranged to receive a Done signal 212 from the control unit 50. The pointer 211 holds the address of the location 190 in the pipe 171 from which the next entry is to be read. The output 215 from the pointer is fed to a first adder 216 and to a second adder 217. Adder 216 allows a value of 1 to be added to the signal and fed to a multiplexer 218 under the control of an output signal from the OR gate 209. The other adder 217 is arranged to add a value of 2 to the signal on line 215 and provides an output 220 which is an increment of 2 to the output signal 215 for the pointer circuit 211. A multiplexer 221 controlled by an output from the AND gate 210 controls the input 223 to the pointer circuit 211 such that the pointer receives either an input from line 220 which is an increment of 2 in the location to be read from or alternatively an output from the multiplexer 218. The output of the multiplexer 218 may either be the output signal 215 itself with zero increment or alternatively an increment of 1 effected by the adder 216. In operation to effect a read from the pipe 171, the source 1 output 52 is defined to be accessed before source 2 output 53 should both be active. For this reason the output pointer value 215 is fed directly to the output selector 192 to select the pipe location to be read from when appropriate signals are received on inputs 182 and 183. The signal on line 207 will determine whether a read operation is effected to the source 1 bus 52 and that signal is also fed to a multiplexer 220 to select either the output signal 215 from the pointer 211 or alternatively the increment of 1 produced by the adder 216 in order to provide the location selection which is input on line 221 to the selector circuit 191. In this way, a read operation may be effected on source bus 53 when appropriate input signals are provided on inputs 186 and 151. The selector 191 will output from the location indicated by line 215 from the pointer 211 if no read operation was effected onto the source 1 bus 52. If however the signal on line 207 had caused a read operation onto source 1 bus 52, then multiplexer 220 will be operated to increment by 1 the location from which the read operation is effected by selector 191 onto source 2 bus 53. The output signal from AND gate 206, which is used to cause a read operation onto the source 2 bus, is also fed to the OR gate 209 to control the multiplexer 218 and AND gate 210 to control the multiplexer 221 in advancing the location indicated by the pointer 211 for the next read operation. If no read operation occurred the output signal 215 from the pointer 211 is fed back through multiplexers 218 and 221 with no change in location. If a read onto only one source bus occurred then the output of adder 216 is fed through multiplexers 218 and 221 to advance the pointer location by 1. If a read operation occurred onto both source buses 52 and 53 then the location identified by the pointer 211 is advanced by 2 by using the output of the adder 217 fed on line 220 through multiplexer.

In addition to the circuitry already described, the processor of FIG. 3 includes an instruction pointer register 230 which holds the address of the next sequential instruction. It is not used to fetch instructions from memory as this is carried out by the fetcher circuit 60 already described. The instruction pointer register 230 is connected to the result bus 54 and may be used to latch an instruction address when a latch signal 231 is input from the control unit 50. Similarly, when an enable signal 232 is provided from the control unit 50 the value in the register 230 will be output onto the source 2 bus 53.

A temporary register 240 holds an intermediate address for use in accessing memory. It may store an address from the result bus 54 when a latch signal 241 is provided from the control unit 50. The value in the temporary register 240 is fed to an address multiplexer 242 which supplies onto the address bus 42 leading to the memory 41 either the address 103 which is output from the instruction fetcher or the address derived from the temporary store 240.

The constant register 167 can be used to hold a constant value output on line 164 from the expansion of an instruction and this value can be fed onto the source 1 bus 52 when an enable signal 245 is derived from a AND gate 246 arranged to receive the enable source 1 signal 247 from the control unit 50. Both the AND gate 185 and the AND gate 246 receive in addition to the enable signal 247 an input from a OR gate 248. These control whether the CONSTANT Register 167 or the value derived from Register file 170 or Pipe 171 is sent to the source 1 bus 52 OR gate 248 receives one input from line 166 indicating that a constant is present in the expanded instruction and a second input 249 which is output from the control unit 50 to control the multiplexer 180 to use the destination field as a source 1 address. This, dependent on an opcode of an instruction, permits some instructions to use the destintation field to specify another source operand.

The length register 127 holds an indication of the length of the current instruction and this can be driven onto the source 1 bus 52 when an enable signal 250 is input from the control unit 50. This can be used to adjust the instruction pointer value which is held in register 230.

The address multiplexer 242 is controlled by a fetch control signal 251 from the control unit 50 so that the memory is addressed from the fetcher 60 or from the value in the temporary register 240.

A write buffer 252 is controlled by an enable signal 253 from the control unit 50 to drive the value on the result bus 54 onto the data bus 43 when signal 253 is enabled. Similarly a read buffer 254 is controlled by a read control signal 255 derived from the control unit 50 so that the value on the data bus 43 can be driven onto the source 1 bus 52 when signal 255 is enabled. The latch input 100 for the fetcher 60 is derived from AND gate 260 having as its inputs signals 251 and 255 from the control unit which are both enabled when a read from memory is required to fetch a further instruction word from memory.

In use, all control outputs from control unit 50 are initially deasserted. If the More output 107 from the fetcher 60 indicates that a further instruction word is required from memory 41 then a read operation is carried out to load another word into the fetcher 60. If the processor has branched to a new instruction sequence then an adjust output 96 will be provided to indicate whether alignment is necessary. If alignment is necessary another memory read for an instruction word may be necessary. The Move signal 107 will indicate if this is necessary. After alignment an instruction will be loaded into the instruction register 62, then expanded by unit 63 and an appropriate source or constant values will be supplied onto source 1 or source 2 buses 52 or 53 and an appropriate operation carried out by the ALU 51 under a command signal 261 from the control unit 50.

The processor may execute arithmetic instructions. In this case instructions may take two source values which may be from the pipe 171, register file 170 or constant register 167. These are specified by the output signals from the expansion unit 63. The ALU 51 produces a result which is fed to the result bus 54 and stored in the pipe 171 or register file 170 as specified by the destination signal derived from the expansion unit 63. Some operations will only have a single source operand and in this case the value on the source 2 bus 53 is ignored and the enable signal on line 186 is not asserted.

The processor may also execute memory instructions which read or write a 32 bit value to or from memory. The address to access in memory is calculated by adding two source values specified by the source 2 and source 1 or constant signals from the expansion unit 63. The value on the result bus 54 is stored in the Temporary Register 240 by signal 241. The location can be read by storing the result in the register file 170 or pipe 171. Alternatively the location can be written to by reading from the pipe 171 or register file 170, as specified by the destination signal 150 which is routed onto line 183 by the multiplexer 180. For both reading and writing, the source 1 bus 52 is routed through the ALU 51 to the Result bus 54 by the correct Command signal 261.

In executing a branch instruction, the processor may change the instruction pointer which is held in register 230 so that a different part of the program is executed. In an unconditional branch the instruction will always change the instruction pointer in register 230 to a new value. For a conditional branch a condition will be specified in order for the instruction pointer to be updated. If the condition is not met then the instruction pointer is updated to a point in the next sequential instruction as normal. Procedure calls may be effected as unconditional branches which save the address of the next sequential instruction in the register file 170 or pipe 171.

An istruction set which may be used in the example described follows:

It will be seen that instructions used in accordance with this example are compressed so that they may have a length of 1, 2, 3 or 4 bytes. The use of these compressed instructions provides greater code density and the comparison with a prior art instruction set of the type shown in FIG. 1 can be seen from the following examples where the heading `Conventional` relates to instructions of the type shown in FIG. 1 and `This Example` relates to the example described above. A `*` is used to designate the omission of an address in a selected field of the instruction, thereby indicating the use of the implicit pipe 171.

If the desired operations is,

where R1, R2 and R3 are register addresses denoting locations in the register file 170.

Conventional

A single instruction accomplishes the operation.

A single instruction accomplishes the operation.

As all operands must be supplied, this is encoded into four bytes.

If the desired operation is to add two values, specified implicitly, and store the result in an implicit location.

Conventional

There is no meaning to this operation

A single instruction accomplishes the operation.

As none of operands need be supplied, this is encoded in one byte.

If the desired operation is

Where registers Ra and Rb hold the base addresses of two arrays and register Ri holds an index to use.

Conventional

Four instructions are required, using 16 bytes.

Four instructions are required, however some of the operands can be omitted, and so only ten bytes are required.

If the desired operation is

where Ra, Rb, Rc and Rd are register locations.

Conventional

Three instructions encoded in 12 bytes and using two temporary registers are required.

Three instructions are required, but as some operands can be omitted, 8 bytes are needed.

If the desired operation is

Conventional

Six instructions and four temporary registers are required. This takes 24 bytes to encode.

Six instructions are required, but the four temporary values can be held in the pipe so 16 bytes are required to encode the sequence.

If the desired operation is

Where registers Ra though Rh hold the base address of 8 arrays of size 1000 entries register Ri contains an index ranging from 0 though 999.

Conventional

21 instructions are required, with 14 temporary registers encoded in 84 bytes

21 instructions are required, but because of the instruction compression only these can be encoded in 56 bytes.

The invention is not limited to the details of the foregoing examples. FIG. 7 illustrates an alternative arrangement to replace FIGS. 5 and 6. In this alternative, the expansion unit 63 is arranged to provide source and destination addresses which are cyclical so that they may operate on a first in first out basis and use some of the register locations in place of the pipe 171. Similar components in FIG. 7 to those already described in FIGS. 5 and 6 have been marked with the same reference numerals and their function will not be repeated. In this case the In Pointer 196 and Out Pointer 211 are latches which can hold a number between N and N+M-1 inclusive. These are the register numbers in the register file 170 which can be used as a circular buffer to replace the requirement for the separate pipe 171. In this way the source and destination selectors on the register file 170 can select some of the registers in a normally addressed manner and other registers will be accessed on a cyclical circular buffer basis. The outputs 140, 141 and 145 from the router 132 are provided respectively to multiplexers 266, 267 and 268. Multiplexer 268 is arranged to provide the source 1 output signal 152. Multiplexer 267 provides the output signal for source 2 and multiplexer 266 provides an output to a further multiplexer 270 which has as its output the destination address. Each of the output signals 140, 141 and 145 is fed to a similar comparator circuit 271 which compares the register numbers for each operand from the router 132 and asserts a signal if it matches the special register number denoting the pipe. This special register number is defined to be any of the M registers making up the circular buffer as they do not need to be directly accessed. The output of comparators 271 are each fed to an OR gate 272 having as a second input a signal from the enable implicit circuit 147. Multiplexer 268 has a second input from the Out Pointer 211. The multiplexer 267 has a second input from a multiplexer 274 arranged to receive either the output of pointer 211 or an increment of 1 to that pointer value. Similarly multiplexer 266 receives an input from a multiplexer 275 which receives an output from multiplexer 274 or an increment of 1 on that value.

The input address to pointer 211 is derived from a multiplexer 276 which has as its inputs either the output of multiplexer 275 or an increment of 1 on that output value. The OR gates 272 are arranged to provide a first output to a respective AND gate 280 arranged to receive a second input from a Used operands circuit 281. The Used operand 281 receives inputs of the opcode signal 131 as well as the length and type signals 120 and 122. Outputs from circuit 281 indicate which of the three Operands are sources and which is the destination. Its outputs are fed to the AND gates 280 to permit appropriate selection of the multiplexers 268, 267 and 266. A further output 283 from the used Operand circuit 281 is fed to a further AND gate 284 which receives a second input from the OR gate 272. This controls the multiplexer 270 to permit the destination output to be derived either from the in pointer 196 or from the multiplexer 266. It will be seen that each of the ADD one units performs a modulo addition of the input signals so that when presented with the value N+M-1 the result is N. The Done signal 212 controls latching of the Out pointer 211 and In pointer 196 so as to update these pointers when execution of an instruction is completed. AND gate 193 only allows update of In Pointer 196 when a value has been read from the pipe, as indicated by AND gate 284. In this way selected registers in the register file may be used as a circular buffer to act as a first in first out multiple value data store and by use of the enable implicit circuit 147 this may be effected even when the length and type signals 120 and 122 indicate that a source or destination address is omitted in the compressed instruction.

In the examples described with reference to FIG. 2, any omitted field is deemed to represent the implicit store provided by the pipe 171. In this way, when the compressed instruction is expanded the omitted field is replaced by a value representing the address of the pipe 171. However, in an alternative embodiment an omitted field may be deemed to take the same value as one of the included fields. The length and type indicators used in the instruction format may indicate that the value of one of the included fields is to be repeated on expansion of the instruction to provide the same output value for that field as a designated one of the included fields. The instruction format may use bit values in the length and type indicator as well as possibly the PAD section of the instruction format in order to indicate which of the included fields is to have its value repeated and which omitted field is to take that repeated value. In FIG. 2 any of formats 6 to 11 have enough bits in the Opcode extension positions (bits numbers 22 or 23) to provide an indication that the expansion unit should copy the value of one of the included fields into an omitted field.

In a further embodiment shown in FIG. 8, two pipes 171a and 171b are provided. Similar reference numerals have been used for parts corresponding to those shown in FIG. 3. The operation of those parts is as previously described. In addition to providing a register file 170 in which the registers are separately addressable, two pipes 171a and 171b are provided and each may be as described with reference to FIG. 6. Alternatively they may be implemented by a plurality of registers which are cyclically addressed to operate as separate circular buffers and described with reference to FIG. 7 modified appropriately to have separate pointers and Out pointer registers and switching circuitry for each pipe. As shown in FIG. 8, each of pipes 171a and 171b includes a respective destination selector 174a and 174b connected to receive data from the result bus 54. Each pipe has a respective output selector 179a and 179b connected to both the source buses 52 and 53. Each of the input and output selectors on both pipes are connected to receive the same control signals as previously described for pipe 171 in FIG. 3 and the circuit of FIG. 6. Each of pipes 171a and 171b will be allocated a separate respective address and it will be appreciated that the expanded output signals from the expansion unit 63 will provide appropriate addresses to determine whether the register file 170 or which of the two pipes 171a or 171b is to be used. In the compressed instructions it is not possible to treat both pipes as implied storage locations. One of the pipes will require its address to be set out explicitly as an operand in the instruction if that pipe is to be used. Only the other pipe address can be used as an implied address when omitted from the field of a compressed instruction. It will be appreciated that any number of similar pipes may be used for holding data in the processor provided each pipe has an identifiable address and only one of the pipes is ever used as the implied storage location.

It will be understood that reference in this specification to a multivalue data store holding a plurality of values on a first in first out basis includes both writing data in the store or allocating locations within the store on a first in first out basis.