Instruction set extension using operand bearing NOP instructions
Instruction set extension using operand bearing no-operation (NOP) or other instructions. In one embodiment, an apparatus can execute a first instruction with an operand associated with a second instruction. The apparatus includes a decoder to identify an operand associated with the second instruction as being designated for the first instruction. An execution unit executes an operation indicated by the first instruction to operate on the operand associated with the second instruction. The second instruction may occur before or after the first instruction in the program sequence.
1. Field
The present disclosure pertains to the field of processors. More particularly, the present disclosure pertains to a new instruction or a change to an instruction or an instruction set of a processor.
2. Description of Related Art
Computer architects often grapple with the difficulty of extending the instruction set of a processor. Often, instruction sets far outlive their originally contemplated lifespan because an installed base of software makes it profitable to maintain backward compatibility. Therefore, computer architects often add new instructions or new functionality to old instruction sets in order to provide new features yet maintain compatibility with the previously installed software base.
Adding new functionality to an existing instruction set can be a challenging exercise. Typically, an instruction set may have a set of mnemonics that translate into specific opcodes. The opcodes are binary values that are understood by the processor and which cause the processor to execute the intended instruction. Opcodes, however, are typically a particular length (i.e., a particular number of bits). Therefore, there are a finite number of instructions (i.e., 2N instructions for an N bit opcode) that may be represented by an opcode, depending on the number of bits dedicated to the opcode.
One common way to overcome this limitation is to use a “prefix”. One opcode may be designated as a prefix (rather than a particular instruction). The prefix indicates that a subsequent value or subsequent values should be decoded differently. In other words, the prefix changes the meaning of at least one subsequent instruction opcode. In some cases, prefixes may be used to define new instructions. In other cases, prefixes may be used to modify the behavior of existing instructions. For example, a prefix may temporarily override default address and/or operand sizes.
In the ×86 instruction set, a prefix may be used to specify whether the opcode length is one or two bytes by use of the opcode expansion prefix. Therefore, by the use of prefixes, the total number of potential instructions can be increased. In some cases, a prefix may be used in conjunction with an escape code to define other instructions. In another case, a suffix may be used to specify additional instructions.
An additional limitation of an instruction set is that there may be a limited number of ways to specify operands. As such, there may a limitation on the number of operands that can be explicitly specified for an instruction. One way around this limitation is to implicitly specify where an instruction can find an additional operand by defining a location in the instruction definition, which is understood by the programmers using the instructions. For example, the MASKMOVQ instruction included in Intel's Pentium® Processor with MMX™ technology and subsequent Pentium® processors implicitly defines one of its operands to be located in a specified register. In this type of arrangement, a burden is placed on a programmer to move the appropriate operand into the specified location prior to execution of the instruction that uses that operand.
Another way to implicitly specify operands used in some instruction sets is to group registers together into pairs or quads. For instructions that operate on data which is a multiple of the register size, a single register may be specified and additional adjacent registers may be used to provide the additional operands. This technique has limited flexibility as well because all of the operands are accessed from the specific pre-defined locations in the register pair or quad.
Additional creative ways to provide additional functionality and/or flexibility to an instruction set would be beneficial.
BRIEF DESCRIPTION OF THE FIGURESThe present invention is illustrated by way of example and not limitation in the Figures of the accompanying drawings.
The following description provides instruction set extension using operand bearing instructions. In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the present invention. It will be appreciated, however, by one skilled in the art that the invention may be practiced without such specific details. In other instances, control structures and gate level circuits have not been shown in detail in order not to obscure the invention. Those of ordinary skill in the art, with the included descriptions, will be able to implement appropriate logic circuits without undue experimentation. While one embodiment is directed to a microprocessor, the techniques described may more generally be applied to other types of electronic processing components that have instruction sets.
A NOP with operands is designed to carry operand information to other instructions preceding and/or following it. Those other instructions may not have enough bits to encode more than a given number of operands (e.g., two source and one destination operand may be specified in some instruction sets). Therefore, if an instruction needs to carry a larger number of operands than is available by the common instruction format, it can be paired with one or more special NOPs, and will use their operand fields as its own. Extending an Instruction Set Architecture (ISA) by preserving the semantics of existing operations may be quite desirable as discussed in the background section. Accepting data from more storage locations than typically supported by an ISA may be a very useful way to increase data level parallelism available to programs, to increase the ratio of data handled per instruction, and in some cases to simplify or shorten programs.
As used in this disclosure, an “instruction” is a generic term referring to a variety of representations of an operation or set of operations to be performed. An instruction may be represented by a mnemonic, for example, or by an opcode. An opcode is typically a sequence of ones and zeroes that may be represented in hexadecimal or in binary or any other convenient format and is interpretable by a processor as indicating the particular operation(s) to be performed. The operation(s) to be performed are generally specified in an instruction definition for each instruction. For example, a programmer's reference manual, user's manual, or the like may specify operations performed by a processor in response to a particular instruction.
In the embodiment of
An opcode is said to “use” or to “operate on” an operand when that operand is specified as one of the arguments to the instruction. As previously noted, the “operand” may be in the form of an operand value or an operand specifier that identifies an operand value by location. If an instruction under its architectural definition does or can utilize the value of the operand either in computing a value, accessing a value, or storing a value, then the instruction is said to “use” or “operate on” the operand. Sometimes the operand may not itself or may not change the ultimate result of the instruction, even if it is used.
A second opcode 170 also having two source operands and one destination operand is transformed into another opcode 180. The opcode 180 has one destination operand from the opcode 170, as well as a first two source operands from the opcode 170 and third and fourth operands from the NOP 165. Thus, one NOP can carry operands for multiple instructions. Further variations may include providing operands to three or more instructions and/or routing different types of instructions to different positions in their new opcodes' operand sets. Furthermore, an operand that may be in the position of a particular type of operand (e.g., a destination) of the NOP may be used for a different purpose in conjunction with the opcode that uses it (e.g., the destination operand of the NOP may be used as a source operand for another opcode). Furthermore, multiple NOPs may provide operands to a single instruction in some embodiments.
Some instruction sets include variable length instructions. Often such instruction sets are considered “complex” and the computers designed around such complex instruction sets are referred to as Complex Instruction Set Computer (CISC) architecture computers. On the other hand, there are also Reduced Instruction Set Computer (RISC) architecture machines. Such machines generally use simpler and smaller instruction sets to accomplish the same tasks by using larger numbers of such instructions (i.e., software becomes more complex).
It is expressly understood that the inventive features of the present disclosure may be usefully embodied in a number of alternative processor architectures that will benefit from the techniques disclosed. RISC, CISC, and hybrid computer architectures may advantageously utilize the disclosed techniques. Moreover, different kinds of processors in different applications may advantageously choose to extend their instruction set or to include in an initial instruction set the ability to utilize operands specified for one instruction with another instruction. General or special purpose processors for computers, servers, personal digital assistants, phones, networking devices, routing devices, etc., may utilize such techniques. Processors may be all hardware, may be software emulated or may utilize a combination of hardware and software emulation to support an “instruction set”.
Moreover, a processor may have multiple instruction sets. Processors may have a single macro-instruction set that is available to a programmer. However, some processors may support multiple macro-instruction sets. Furthermore, processors may contain micro-instruction sets or secondary instructions sets. A micro-instruction set may be an instruction set into which a processor translates macro-instructions, with the micro-instructions being only used internally by the processor and not typically accessible to programmers or compilers which are limited to using the macro-instruction set. Similarly, a secondary instruction set may be just a different instruction set (not necessarily internal to any component) into which the first instruction set is translated. For example, a CISC program may be translated into corresponding instructions from another instructions set which are then executed by either hardware or some hardware and/or software and/or firmware combination. For example, a CISC program may be translated into RISC instructions for execution on RISC hardware or into instructions for a Very Long Instruction Word (VLIW) instruction set for execution on a VLIW hardware processor. In any case, all of these different instruction sets and/or processors are just different embodiments that can benefit from the disclosed techniques.
The decode module 205 of the embodiment of
The decode module 205 may be a hardware decoder in some embodiments. For example, a hardware or logic decoder may be useful in converting macroinstructions to microinstructions for execution by a core of a processor. The decode module 205 may also be a software module that performs binary translation from one instruction set to another instruction set. Moreover, the decode module may be a combination of hardware and software as desired to convert from one instruction set to another. The second instruction set will typically have the capability to allow more operands to be carried by some instructions than the first instruction set. Accordingly, the extra operands added to a NOP instruction in the first instruction set, after the decoding, may be carried along by the instructions with which they are associated.
As indicated in block 235, the decode module 205 associates NOP operands with the first instruction 140. Subsequent to decoding by the decode module 205, the first opcode 140 (Opcode A) may be represented by a different opcode, Opcode B. As mentioned, Opcode B may have a higher operand carrying capacity than Opcode A since it is a translation of Opcode A into a different instruction set. Opcode B is shown as generically having M destination operands, and N source operands, where N and M can be any positive integer numbers. Opcode B, however, specifies the same or a similar operation as did Opcode A. The operation may be modified somewhat or dramatically in different embodiments, in order to utilize the added operands. Moreover, Opcode B may be one of a sequence of opcodes that implement Opcode A.
As indicated in block 245, the new Opcode B is executed by an execution module 215 to perform an operation indicated by Opcode A on some or all of the operands from Opcode A and some or all of the operands from the NOP 150. Much like the decode module 205, the execution module 215 may be implemented in a variety of manners. For example, the execution module 215 may be a hardware execution unit or set of execution units. Alternatively, the execution module 215 may be a combination of hardware and software, or a combination of hardware, firmware, and/or software, etc.
In some embodiments, the NOP instruction itself may specify how its own “extra” operands may be used. In one embodiment, a NOP may specify a fixed relationship to an opcode that will use the NOP operands. In other embodiments, it may be desirable to provide a set of NOP opcodes that allow flexibility as to how the NOP operands may be used. For example, Table 1 provides a set of N NOP opcodes that may be used to distribute the N NOP operands to adjacent instructions in different manners. Some subset or superset of these encodings may be used, or different encodings may be used as well, as will be appreciated by one of skill in the art.
In other embodiments, the NOP itself may not specify how its operands are to be used, but rather various modifiers may be inserted into the instruction stream to specify NOP operand usage.
In one embodiment, the modifiers (either 305 or 310) may be prefixes which are used to specify the distribution or just to indicate the use of the operands from a nearby NOP operand. In some embodiments, the opcode-modifying prefixes may specify that operands be taken from a fixed location (e.g., a NOP before or after the modified opcode). Similarly, the NOP-modifying prefixes may also specify that the NOP operands be passed to a specifically sequenced (before or after) instruction. Alternatively, a variety of prefixes may be used to provide flexibility.
Table 2 illustrates a more elaborate encoding technique to allow more flexibility as to where additional operands may be obtained. Although Table 2 is provided for prefixes, this technique is equally applicable to NOPs. In other words, a larger number of NOPs could be used than illustrated in Table 1 to provide all the flexibility shown with respect to prefixes in Table 2.
As an example, according to Table 2, Prefix Q indicates that the NOP operands should be used for the instruction before the NOP, that the first NOP operand should be used as a source operand and that the last NOP operand should be used as a destination operand, and that at least one other destination operand is to be obtained from the NOP. Prefixes 2N+2+1 through M are particularly flexible in that they allow some operands to be used with a prior instruction and some to be used with a subsequent instruction. Any subset or superset of these encodings may be used to specify how NOP operands may be distributed, as will be appreciated by one of skill in the art. Moreover, other modifiers may be used as appropriate with a particular instruction set. For example, escape codes or suffixes or other instruction modifying values that suggest a different interpretation of a given opcode may be used.
A code segment 434 illustrates a NOP/opcode pair with the NOP preceding the opcode. In this code segment 434, the NOP may be of the type that carries operands for a single instruction in a fixed location, or may be a NOP that a modifier specifies carries operands for the opcode. Similarly, a code segment 436 illustrates that the NOP may follow the instruction for which it carries operands. The NOP itself is typically received by the decode module 409 and essentially discarded after its operands are extracted. Since the NOP itself changes no register or memory values, there is no need to pass the NOP or a translated version thereof beyond the decoder.
In some embodiments, other instructions other than NOP instructions may be used to carry operands for other instructions. A NOP is just an example of an instruction that can carry operands that may be convenient because a NOP is generally defined to not change architectural state variables as it executes (although instruction pointers and the like may change). Other new instructions could be defined specifically for the purpose of carrying operands. Alternatively, some existing opcodes that do not use the entire number of operands that an instruction could be modified by modifiers to use their unused operand slots to carry operands for other instructions. Thus, while NOPs are used in one embodiment, other instructions may be used to carry operands. These other instructions may or may not be executed. However, the operand carrying instruction does not use or operate on the carried operand, but rather passes this operand to the opcode for which it was intended.
The ability to specify additional operands may be quite advantageous in some embodiments, as new functions may be provided and/or instruction sequences may be simplified and/or shortened.
In this embodiment, the opcode 515 has four source operands specifying registers from a register file 510 having N registers. A first source operand specifies register 510-N. A second source operand specifies register 510-3, whereas a third source operand specifies register 510-N-3, and a fourth source operand specifies register 510-N-4. The ability to uniquely specify more operands as shown in
In the embodiment of
For example, butterfly type calculations generating two results may be implemented without implicit specification of operands in an instruction set having too few operand slots to carry the necessary operands with one instruction. Another example is a single instruction multiple data (SIMD) type instruction, which may output a larger result than one register can accommodate. Another example is a complex addressing scheme, in which multiple components are used to calculate an address (e.g., a base, offset, and scale, or other address components may be used in an address calculation). Yet another example is that more or longer immediate operands may be provided. Any of these operands or operand-like quantities may be specified via a NOP bearing operands.
If a greater number operands are specified in the source code than typically handled by a selected opcode, then the compiler adds a modifier and/or selects an appropriate NOP to designate the extra operands as indicated in block 620. Then, as indicated in block 630, the compiler specifies the extra operands with the NOP instruction. Accordingly, source code may be compiled to allow the specification of a larger number of operands than a single opcode can itself carry. Additionally, this technique may be convenient in other scenarios where opcode carrying is not a strict limitation, but it is otherwise convenient to have one instruction bear operands for another instruction.
A hardware design may go through various stages, from creation to simulation to fabrication. Data representing a design may represent the design in a number of manners. First, as is useful in simulations, the hardware may be represented using a hardware description language or another functional description language. Additionally, a circuit level model with logic and/or transistor gates may be produced at some stages of the design process. Furthermore, most designs, at some stage, reach a level of data representing the physical placement of various devices in the hardware model. In the case where conventional semiconductor fabrication techniques are used, the data representing the hardware model may be the data specifying the presence or absence of various features on different mask layers for masks used to produce the integrated circuit. In any representation of the design, the data may be stored in any form of a machine readable medium. In a software design, the design typically remains on a machine readable medium. An optical or electrical wave modulated or otherwise generated to transmit such information, a memory, or a magnetic or optical storage such as a disc may be the machine readable medium. Any of these mediums may “carry” the design information.
Thus, instruction set extension using operand bearing instructions is disclosed. While certain exemplary embodiments have been described and shown in the accompanying drawings, it is to be understood that such embodiments are merely illustrative of and not restrictive on the broad invention, and that this invention not be limited to the specific constructions and arrangements shown and described, since various other modifications may occur to those ordinarily skilled in the art upon studying this disclosure.
Claims
1-32. (canceled)
33. A compiler comprising:
- a code segment to identify a first instruction in response to a first source code instruction as having a number greater than a predetermined number of operands for a first translation of said first source code instruction;
- a code segment to specify a second instruction to have an operand for said first instruction in response to said first source code instruction and said number greater than the predetermined number of operands.
34. The compiler of claim 33 wherein said code segment to specify said second instruction is to specify a No Operation instruction having a plurality of operand specifiers specified with said first source code instruction.
35. The compiler of claim 34 further comprising a code segment to add a modifier to indicate that said second instruction is to specify operands for said first instruction.
36. The compiler of claim 35 wherein said modifier modifies said first instruction.
37. The compiler of claim 35 wherein said modifier modifies said second instruction.
38. The compiler of claim 36 wherein said modifier is a prefix.
39. The compiler of claim 34 wherein said No Operation instruction is one of a plurality of operand-specifying No Operation instructions.
40. A method comprising:
- receiving a first instruction with a first operand;
- receiving a second instruction;
- associating said first operand with said second instruction.
41. The method of claim 40 further comprising:
- discarding said first instruction.
42. The method of claim 40 further comprising:
- performing an operation specified by said second instruction using said first operand.
43. The method of claim 40 wherein said first operand is a source operand and wherein performing the operation specified by said second instruction using said first operand comprises operating directly on said first operand without altering said first operand due to the first instruction.
44. The method of claim 42 wherein said second instruction is a NOP instruction that does not alter said first operand.
45. The method of claim 41 wherein said first operand is a destination operand and wherein performing comprises generating a first result for storage in a first destination specified by the second instruction and a second results for storage in a location specified by said first operand specifier.
46. The method of claim 40 wherein said first instruction specifies a first operand specifier for use with said second instruction and a second operand specifier for use with said second instruction.
47-56. (canceled)
Type: Application
Filed: Aug 30, 2005
Publication Date: Dec 29, 2005
Inventor: Gad Sheaffer (Haifa)
Application Number: 11/215,862