System and method for handling load and/or store operations in a superscalar microprocessor

A superscalar microprocessor capable of executing one or more instructions out-of-order with respect to an ordering defined by a program order, the microprocessor comprising:

• (a) an instruction fetch unit configured to provide a plurality of instructions to an instruction buffer;
• (b) an execution unit, coupled to the instruction fetch unit, configured to execute the plurality of the instructions from the instruction buffer in an out-of-order fashion, the execution unit including a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions,
• the execution unit further including a load store unit adapted to make load requests and store requests to a memory system, the load store unit adapted to make at least one load request out of the program order so that the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, wherein the second instruction precedes the first instruction in the program order, the load store unit having,
• (i) an address generation unit configured to generate load and store addresses for instructions in the instruction buffer, wherein at least one of a load address and a store address is generated out of the program order;
• (ii) an address path adapted to manage the generated load and store addresses and to provide the generated load and store addresses to the memory system the address path including an address buffer distinct from the register file, the address buffer being adapted to receive and buffer the load and store addresses generated by the address generation unit;
• and
• (iii) a data path adapted to transfer data from the memory system to the register file in response to load requests, the data path configured to align data returned from the memory system to thereby permit data falling on a word boundary to be returned from the memory system to the execution unit in correct alignment, wherein the superscalar microprocessor initiates execution of more than one of the plurality of instructions from the instruction buffer in a clock cycle.

A computer system, comprising:

• (a) a memory system configured to retain instructions and data, the instructions having a program order;
• and
• (b) a superscalar processor configured to execute the instructions, wherein the superscalar processor is configured to initiate more than one instruction in a clock cycle, the processor having,
• (1) an instruction fetch unit configured to provide a plurality of instructions to an instruction buffer;
• and
• (2) an execution unit, coupled to the instruction fetch unit, configured to execute the plurality of instructions from the instruction buffer in an out-of-order fashion, the execution unit including,
• (i) a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions;
• (ii) address generation circuitry adapted to generate addresses for load requests and store requests out-of-order;
• and
• (iii) a load store unit adapted to make the load requests and the store requests to the memory system, the load store unit adapted to make at least one load request out of the program order so that the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, wherein the second instruction precedes the first instruction in the program order, the load store unit further adapted to return data falling on a word boundary in correct alignment to the register file,
• wherein the load/store unit includes an address buffer distinct from the register file, the address buffer being adapted to receive and buffer the addresses generated by the address generation circuitry.

A superscalar microprocessor capable of executing one or more instructions out-of-order with respect to an ordering defined by a program order, the microprocessor comprising:

• (a) an instruction fetch unit configured to provide a plurality of the instructions to an instruction buffer;
• (b) an execution unit, coupled to the instruction fetch unit, configured to execute the plurality of instructions from the instruction buffer in an out-of-order fashion, the execution unit including a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions,
• the execution unit further including a load store unit adapted to make load requests and store requests to a memory system, the load store unit adapted to make at least one load request out of the program order so that the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, and wherein the second instruction precedes the first instruction in the program order, the load store unit having,
• (i) an address generation unit configured to generate load and store addresses for instructions in the instruction buffer, wherein at least one of a load address and a store address is generated out of the program order;
• (ii) an address path adapted to manage the generated load and store addresses and to provide the generated load and store addresses to the memory system, the address path including an address buffer distinct from the register file, the address buffer being configured to buffer the load and store addresses generated by the address generation unit;
• (iii) dependency detection circuitry adapted to detect store-to-load dependencies, wherein the dependency detection circuitry determines when data for a load request depends on a store request;
• and
• (iv) a data path configured to transfer load data from the memory system to the execution unit, the data path configured to align data returned from the memory system to thereby permit data falling on a word boundary to be returned nom the memory system to the execution unit in correct alignment,
• wherein the superscalar microprocessor initiates execution of more than one of the plurality of instructions from the instruction buffer in a clock cycle.

A superscalar microprocessor capable of executing one or more instructions from an instruction buffer out-of-order with respect to an ordering defined by a program order, the microprocessor comprising:

• an execution unit configured to execute a plurality of instructions in an out-of-order fashion, the execution unit including a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions,
• the execution unit further including a load store unit adapted to make load requests and store requests to a memory system, the load store unit adapted to make at least one load request out of the program order so the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, wherein the second instruction precedes the first instruction in the program order, the load store unit including:
• (i) an address path adapted to manage load and store addresses in an address buffer separate from the register file and to provide the load and store addresses to the memory system, wherein the load and store addresses include an address for a first byte of data to be loaded or stored and an address for a last byte of the data to be loaded or stored;
• (ii) load dependency detection circuitry, wherein the load store unit does not make a particular load request when the load dependency detection circuitry detects an address collision or write pending for that particular load request;
• and
• (iii) a data path adapted to transfer data from the memory system to the execution unit in response to load requests, the data path configured to align data returned from the memory system to thereby permit data falling on a word boundary to be returned from the memory system to the execution unit in correct alignment,
• wherein the superscalar microprocessor initiates execution of more than one of the plurality of instructions in a clock cycle.

A superscalar microprocessor capable of executing one or more instructions in an instruction buffer out-of-order with respect to an ordering defined by a program order, the microprocessor comprising:

• an execution unit configured to execute a plurality of instructions in an out-of-order fashion, the execution unit including a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions,
• the execution unit further including a load store unit adapted to make load requests and store requests to a memory system, the load store unit adapted to make at least one load request out of the program order so that the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, wherein the second instruction precedes the first instruction in the program order, the load store unit having,
• (i) an address generation unit configured to generate load and store addresses out of order for instructions in the plurality of instructions, wherein the address generation unit generates an address corresponding to a first data byte of data to be loaded or stored and an address corresponding to a last data byte of the data to be loaded or stored;
• (ii) an address path adapted to manage the generated load and store addresses using an address buffer separate from the register file and to provide the generated load and store addresses to the memory system;
• and
• (iii) a data path adapted to transfer data from the memory system to the execution unit in response to load requests, the data path configured to align data returned from the memory system to thereby permit data falling on a word boundary to be returned from the memory system to the execution unit in correct alignment,
• wherein the superscalar microprocessor initiates execution of more than one of the plurality of instructions in a clock cycle.

A superscalar microprocessor configured to initiate execution of more than one instruction from an instruction buffer in a clock cycle, the processor comprising:

• (a) a memory system configured to retain instructions and data, the instructions having a program order;
• (b) an execution unit configured to execute the plurality of instructions in an out-of-order fashion, the execution unit including,
• (i) a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions;
• (ii) address generation circuitry adapted to generate addresses for load requests and store requests out-of-order, wherein the address generation circuitry generates an address corresponding to a first data byte of data to be loaded or stored and an address corresponding to a last data byte of the data to be loaded or stored;
• and
• (iii) a load store unit adapted to make the load requests and the store requests to the memory system, the load store unit adapted to make at least one load request out of the program order so that the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, wherein the second instruction precedes the first instruction in the program order, the load store unit further adapted to return data falling on a word boundary in correct alignment to the register file;
• wherein the load store unit includes an address path adapted to manage load and store addresses using an address buffer separate from the register file and to provide the load and store addresses to the memory system.

A superscalar microprocessor capable of executing one or more instructions from an instruction buffer out-of-order with respect to an ordering defined by a program order, the microprocessor comprising:

• an execution unit configured to execute a plurality of instructions in an out-of-order fashion, the execution unit including a register file that is distinct from the instruction buffer, the register file having a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions,
• the execution unit further including a load store unit adapted to make load requests and store requests to a memory system, the load store unit adapted to make at least one load request out of the program order so that the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, and wherein the second instruction precedes the first instruction in the program order, the load store unit having,
• (i) an address generation unit configured to generate load and store addresses out of order for instructions in the plurality of instructions, wherein the address generation unit generates an address corresponding to a first data byte of data to be loaded or stored and an address corresponding to a last data byte of the data to be loaded or stored;
• (ii) an address path adapted to manage the generated load and store addresses using an address buffer separate from the register file and to provide the generated load and store addresses to the memory system;
• (iii) dependency detection circuitry adapted to detect store-to-load dependencies, wherein the dependency detection circuitry determines when data for a load request depends on a store request;
• and
• (iv) a data path configured to transfer load data from the memory system to the execution unit, the data path configured to align data returned from the memory system to thereby permit data falling on a word boundary to be returned from the memory system to the execution unit in correct alignment,
• wherein the superscalar microprocessor initiates execution of more than one of the plurality of instructions in a clock cycle.

In a superscalar microprocessor having an execution unit adapted to execute a plurality of instructions from an instruction buffer and to issue load instructions out-of-order, a method for managing requests for loads and stores to and from a memory device, the method comprising:

• calculating a first address and a second address for an instruction and transferring said first and second addresses to a load store unit;
• storing the first address and the second address in an address buffer of the load store unit, wherein the address buffer is distinct from a main register file of the execution unit, wherein the main register file is distinct from the instruction buffer and includes a plurality of temporary buffers to store result data from out-of-order instruction execution and a plurality of real entries to store results of retired instructions;
• determining whether said instruction involves at least one of a load operation or a store operation;
• checking, if said instruction has a load operation, for an address collision and for any write pendings, and signaling the outcome of said check using the first and second addresses;
• making a request to said memory device based on a priority scheme and the results of said checking step, wherein said priority scheme includes making at least one load request out of an ordering so the one load request is made before a memory request, wherein the one load request corresponds to a first instruction from the plurality of instructions and the memory request corresponds to a second instruction from the plurality of instructions, wherein the second instruction precedes the first instruction in the ordering;
• receiving requested data from said load operation and/or said store operation in a data path portion of said load store unit;
• and
• aligning said requested data if said requested data is unaligned.

A method for executing one or more instructions out of order using a superscalar microprocessor, the method comprising:

• receiving a plurality of instructions from an instruction buffer having an ordering, the plurality of instructions including a store instruction and a load instruction;
• generating a first address and a second address for an instruction in the plurality of instructions, wherein the generated first and second addresses include an address that is generated out of order with respect to the ordering;
• storing the first address and the second address in a load store address buffer;
• if the instruction involves a load operation, determining whether the load operation can be executed out of order using the first and second addresses;
• if the load operation can be executed out of order, executing the load operation out of order using the generated address, including performing a load request corresponding to the load instruction before a store request corresponding to the store instruction, wherein the store instruction precedes the load instruction in the ordering;
• receiving requested data from the load operation;
• aligning the requested data if the requested data is unaligned;
• and
• loading the requested data into a register of a main register file of the superscalar microprocessor, wherein the main register file is distinct from the instruction buffer and includes a plurality of temporary buffers to store result data of instructions executed out of order and a plurality of real registers to store result data of retired instructions, wherein the main register file is separate from the load store address buffer.