Renaming of hardware micro-fusion memory operations
By using hardware micro-converged memory operations, multiple data register operands are treated as a single operation through renaming and scheduling blocks, which solves the problems of microarchitecture trace overhead and clock frequency reduction in processor memory operations and improves the processor's memory operation performance.
Patent Information
- Application Number
- CN202080088624.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2019-12-20
- Filing Date
- 2020-12-17
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2040-12-17
AI Technical Summary
Existing processors suffer from high microarchitecture tracing overhead, high silicon area, and high operational complexity when performing memory operations. Furthermore, adding a hardware memory operation pipeline can lead to a decrease in clock frequency and a degradation in non-memory operation performance.
It employs hardware micro-converged memory operation, and treats multiple data register operands as a single operation for scheduling and execution through renaming blocks and scheduling blocks, and completes the operation atomically using load/store units.
It improves the processor's memory operation performance, reduces microarchitecture tracing overhead, avoids clock frequency reduction, and enhances the overall performance of the processor.
Smart Images

Figure CN115210690B_ABST
Abstract
Description
[0001] Priority application This application claims priority to U.S. Patent Application No. 16 / 722,870, filed December 20, 2019, entitled “Hardware Micro-converged Memory Operation,” which is incorporated herein by reference in its entirety. Invention Field
[0002] The techniques of this invention generally relate to processor memory operations, specifically to atomic load and store operations with multiple destinations. Background Technology
[0003] Memory operations (such as load and store operations) are operations that write data to or read data from processor-associated memory. These memory operations are often a large part of a large number of processor workloads; therefore, the ability to perform memory operations quickly and efficiently can improve the overall performance of the processor. For workloads involving large datasets (such as many cloud computing workloads), load and store operations may include most of the instructions associated with the specific workload.
[0004] To address this issue, some computer architectures have merged multiple independent memory instructions into a single larger memory operation. However, this approach requires significant microarchitectural trace overhead to ensure correct processor operation. Other processing architectures offer enhanced memory operations, such as paired load and store instructions (e.g., Arm's load-store pairs). Such operations efficiently execute two memory operations encapsulated within a single instruction at contiguous memory locations, conserving instruction bandwidth while utilizing existing hardware memory operation pipelines with full data width.
[0005] To improve processor memory operation performance, a common approach is to increase the number of hardware memory operation pipelines in the processor. However, adding hardware memory operation pipelines to the processor is relatively costly in terms of silicon area and operational complexity, and may involve unacceptable trade-offs with other desired functionalities. Another approach is to reduce the number of latency clock cycles associated with memory operations. However, this can lead to a decrease in overall clock frequency and potentially degrade non-memory operation performance. Therefore, it is necessary to identify other techniques that can improve the processor's memory operation execution capabilities. Summary of the Invention
[0006] The aspects disclosed in the specific implementation include hardware micro-converged memory (e.g., load and store) operations. On one hand, a hardware micro-converged memory operation is a single atomic memory operation performed using multiple data register operands, such as a load-pair or store-pair operation. For renaming, the load-pair or store-pair operation is treated as two independent operations, but scheduled as a single operation containing multiple data register operands. The load-pair or store-pair operation is then executed atomically.
[0007] In one respect, an apparatus includes a renaming block for receiving a first memory operation specifying a plurality of data register operands and renaming the first memory operation. The apparatus further includes a scheduling block for receiving the renamed first memory operation, storing the renamed first memory operation in at least one entry of a plurality of scheduling block entries, and scheduling the first memory operation as a single operation containing the plurality of data register operands. The apparatus further includes a memory operation block for receiving the scheduled first memory operation and executing the first memory operation atomically across the plurality of data register operands.
[0008] On the other hand, an apparatus includes a renaming device for receiving a first memory operation specifying a plurality of data register operands and renaming the first memory operation. The apparatus further includes a scheduling device for receiving the renamed first memory operation, storing the renamed first memory operation in at least one entry of a plurality of entries in the scheduling device, and scheduling the first memory operation as a single operation containing the plurality of data register operands. The apparatus further includes a memory operation device for receiving the scheduled first memory operation and executing the first memory operation atomically across the plurality of data register operands.
[0009] On the other hand, one method includes receiving a first memory operation with a plurality of data register operands at a specified renaming block, and renaming the first memory operation via the renaming block. The method further includes providing the renamed first memory operation to a scheduling block and scheduling the renamed first memory operation as a single operation containing the plurality of data register operands. The method further includes providing the scheduled first memory operation to a memory operation block and then atomically executing the first memory operation across the plurality of data register operands.
[0010] On the other hand, a non-transitory computer-readable medium stores computer-executable instructions that, when executed by a processor, cause the processor to receive a first memory operation with multiple data register operands at a specified renaming block, and rename the first memory operation via the renaming block. The instructions further cause the processor to provide the renamed first memory operation to a scheduling block and schedule the renamed first memory operation as a single operation containing multiple data register operands. The instructions further cause the processor to provide the scheduled first memory operation to a memory operation block, and the memory block executes the scheduled first memory operation atomically. Attached Figure Description
[0011] Figure 1 A block diagram of an exemplary processor that supports hardware micro-converged memory operations.
[0012] Figure 2 To show through Figure 1 A detailed block diagram of the process by which a portion of an exemplary processor performs hardware micro-converged memory operations.
[0013] Figure 3 for Figure 1 A detailed diagram of the scheduling block entries of an exemplary processor.
[0014] Figure 4 A flowchart illustrating the execution method of hardware micro-converged memory operations is provided.
[0015] Figure 5 A block diagram of an exemplary processor system capable of performing hardware micro-converged memory operations. Detailed Implementation
[0016] Several exemplary aspects of the present invention are disclosed with reference to the accompanying drawings. In this invention, the word "exemplary" means "serving as an example, instance, or illustration." Any "exemplary" aspect described herein is not necessarily superior to the others.
[0017] The aspects disclosed in the specific implementation include hardware micro-converged memory (e.g., load and store) operations. On one hand, a hardware micro-converged memory operation is a single atomic memory operation performed using multiple data register operands, such as a load-pair or store-pair operation. For renaming, the load-pair or store-pair operation is treated as two independent operations, but scheduled as a single operation containing two data register operands. The load-pair or store-pair operation is then executed atomically.
[0018] in this regard, Figure 1Block diagram 100 illustrates an exemplary processor 105 supporting hardware micro-converged memory operations. Processor 105 includes a renaming block 110 coupled to scheduling block 120 and is available to schedule block 120 for renaming hardware micro-converged memory operations. Scheduling block 120 is coupled to load / store unit 130 and is available to schedule hardware micro-converged memory operations on load / store unit 130 for execution of the operations. Load / store unit 130 may be coupled to cache and memory hierarchy 140 and execute hardware micro-converged memory operations atomically. Cache and memory hierarchy 140 may include one or more levels of cache (potentially processor 105 proprietary or shared between processor 105 and other processors not shown), tightly coupled memory, main memory, or other types of memory known to those skilled in the art.
[0019] During runtime, renaming block 110 receives hardware micro-fusion memory operations (e.g., load pairs or store pairs) and renames them. On one hand, the hardware micro-fusion memory operation is treated as two independent memory operations (one operation for each data source register in the case of a store pair; or one operation for each data destination register in the case of a load pair), thus effectively occupying two slots during renaming. Using a separate slot in renaming block 110 for each data register operand of the hardware micro-fusion memory operation alleviates the problem of increased size and complexity of renaming block 110, thereby supporting hardware micro-fusion memory operations, but at the cost of consuming more existing resources of renaming block 110 when renaming the hardware micro-fusion memory operation. On the other hand, the slot size in the renaming block can be increased to support hardware micro-fusion memory operations. Whether stored as two independent memory operations or a single memory operation in a larger renaming slot, the renaming block can provide hardware micro-fusion memory operations to scheduling block 120 in the form of a single atomic memory operation with multiple data register operands.
[0020] Scheduling block 120 receives hardware micro-converged memory operations (whether in the form of two separate memory operations or a single memory operation) from renaming block 110 and schedules the hardware micro-converged memory operations as a single memory operation containing multiple data register operands (i.e., twice the width of a single register in the case of load pairs or store pairs). In the case of load pair operations, scheduling block 120 stores the load pair operations in a single entry of scheduling block 120 as a single memory operation containing two destination registers. The hardware micro-converged memory operations are then provided to load / store unit 130 as atomic operations. In the case of store pair operations, scheduling block 120 stores the store pair operations across two scheduling block entries, which are then combined by load / store unit 130 and executed as a single atomic operation containing two independent micro-operations.
[0021] Load / store unit 130 receives hardware micro-converged memory operations from scheduling block 120 and executes the operations as atomic memory operations. On one hand, load / store unit 130 includes a data path 135 leading to cache and memory hierarchy 140, the path being wide enough to support the hardware micro-converged memory operations as a single transaction (i.e., the data width of data path 135 corresponds to the cumulative data width of the number of data register operands; therefore, for load-pair or store-pair operations, the data width of data path 135 corresponds to twice the size of a single register). In the case of load-pair operations, this may include loading a double-register-width value from a memory address and storing each half of the value in one of a pair of specified data destination registers. In the case of store-pair operations, this may include storing the contents of a pair of specified data source registers as consecutive values starting at a specified memory address. In some aspects, load / store unit 130 may further perform store-load forwarding operations between store-pair and load-pair operations.
[0022] in this regard, Figure 2 To show through Figure 1A detailed block diagram 200 illustrates the flow of a portion of an exemplary processor performing a hardware micro-fusion memory operation 250 (which in some respects may be a load pair or store pair operation). A renaming block 110 includes four renaming slots 211a, 211b, 211c, and 211d, capable of receiving the hardware micro-fusion memory operation 250 and processing it as two independent operations for renaming. For example, the renaming block 110 may allocate a first portion 251a of the hardware micro-fusion memory operation 250 as a single operation (representing a first data source / destination register for renaming slot 211a), and the renaming block 110 may allocate a second portion 251b of the hardware micro-fusion memory operation 250 as another single operation (representing a second data source / destination register for renaming slot 211b). Thus, for renaming, the renaming block 110 treats the hardware micro-fusion memory operation 250 as two independent memory operations. As described above, refer to... Figure 1 This alleviates the problem of increased size and complexity of the renaming block 110 in order to perform the hardware micro-converged memory operation. The renaming block 110 then renames the first portion 251a of the hardware micro-converged memory operation 250 in slot 211a and the second portion 251b of the hardware micro-converged memory operation 250 in slot 211b, and provides both portions of the first hardware micro-converged memory operation 250 to the scheduling block 120.
[0023] Scheduling block 120 receives a first portion 251a and a second portion 251b of a hardware micro-converged memory operation 250 from renaming block 110. In the case of a load pair operation, these two portions of the hardware micro-converged memory operation 250 are stored in a single entry of scheduling block 120 (in the case of a store pair, the hardware micro-converged memory operation 250 can be stored across two entries of scheduling block 120, and the operation can be provided to load / store unit 130 in the form of a single atomic operation). Scheduling block 120 includes four scheduling slots 221a, 221b, 221c, and 221d. In the case of a load pair operation, on the one hand, the hardware micro-converged memory operation 250 can be allocated to slot 221a. Each slot in scheduling slots 221a-d can include multiple destination fields to perform the hardware micro-converged memory operation (which will be...). Figure 3 (To be discussed further). After allocating hardware micro-fusion memory operation 250 to slot 221a in scheduling block 120, the scheduling block can schedule hardware micro-fusion memory operation 250 to be executed in load / store unit 130, and can provide hardware micro-fusion memory operation 250 to load / store unit 130 to execute the operation in the form of atomic operations.
[0024] Load / store unit 130 includes a first load / store pipeline 231a and a second load / store pipeline 231b. Hardware micro-converged memory operations 250 are received from scheduling block 120 and, for ease of execution, are routed to either the first load / store pipeline 231a or the second load / store pipeline 231b (in the illustrated aspect, the memory operation is routed to the first load / store pipeline 231a, but this is for illustrative purposes only). On one hand, the first load / store pipeline 231a and the second load / store pipeline 231b support the full data width of the hardware micro-converged memory operation 250; therefore, the hardware micro-converged memory operation 250 can be executed in a single operation form (i.e., in the case of load-pair operations, the first load / store pipeline 231a and the second load / store pipeline 231b support a data width corresponding to twice the size of a single register). Then, the first load / store pipeline 231a, together with the cache and memory hierarchy 140, performs hardware micro-converged memory operations 250 via data path 135 (as described above, see reference). Figure 1 On one hand, the load / store unit 130 can independently track multiple destinations of the hardware micro-converged memory operation 250. Furthermore, on the other hand, if the hardware micro-converged memory operation 250 is a store-to-store operation, the load / store unit 130 can receive the hardware micro-converged memory operation 250 from two entries in the scheduling block 120 (as described above, see reference 120). Figure 1 Hardware micro-fusion memory operations 250 are performed as two independent micro-operations (considered as single atomic operations). These two independent micro-operations may point to a single entry in the associated memory buffer. In one exemplary aspect, this may include a first micro-operation for performing address calculation and storing a first data element, and a second micro-operation for performing storing a second data element.
[0025] Figure 3 for Figure 1 A detailed diagram of an exemplary processor scheduler (such as scheduler block 120) entry 300. For example, entry 300 may correspond to any of the entries 221a-d in scheduler block 120. On one hand, entry 300 illustrates entry 221a, in which scheduler block 120 stores hardware micro-converged memory operations 250.
[0026] Entry 300 includes a control field 310, a first source ready field 321, a first source field 322, a second source ready field 331, a second source field 332, a first destination field 340, and a second destination field 350. Control field 310 contains general pipeline control information related to operations such as hardware micro-converged memory operation 250. First source ready field 321 contains an indicator indicating whether the first source shown in first source field 322 is readily available. Similarly, second source ready field 331 contains an indicator indicating whether the second source shown in second source field 332 is readily available. First destination field 340 contains first destination indication information for operations such as hardware micro-converged memory operation 250. Second destination field 350 contains second destination indication information for operations such as hardware micro-converged memory operation 250.
[0027] like Figure 1 and Figure 2 As shown, by providing two source fields (first source field 321 and second source field 331) and two destination fields (first destination field 340 and second destination field 350) in each entry 300, scheduling block 120 can store and schedule a single atomic memory operation performed using multiple data register operands (e.g., multiple sources or multiple destinations).
[0028] in this regard, Figure 4 A flowchart illustrating a method 400 for performing hardware micro-converged memory operations is provided. The method begins at block 410, where a memory operation specifying multiple data register operands is received at the renaming block. For example, hardware micro-converged memory operation 250 is received via renaming block 110.
[0029] Method 400 continues to block 420, wherein a renaming is performed in the memory operation. For example, renaming block 110 stores hardware micro-converged memory operation 250 in renaming slot 211a in the form of a first portion 251a and in renaming slot 211b in the form of a second portion 251b, and performs renaming separately in slots 211a and 211b. Then, the method continues to block 430, wherein the renamed memory operation is provided to the scheduling block. For example, after renaming slots 211a and 211b separately, thereby renaming memory operation 250, renaming block 110 provides the renamed memory operation 250 to scheduling block 120.
[0030] Method 400 continues to block 440, wherein the scheduling block schedules the renamed memory operation in the form of a single operation containing multiple data register operands. For example, scheduling block 120 stores hardware micro-converged memory operation 250 in scheduling slot 221a, and stores multiple data register operands in a first source field 322 and a second source field 332 or a first destination field 340 and a second destination field 350.
[0031] The method continues to block 450, where the scheduler provides scheduled memory operations to a memory operation block and then executes the first memory operation atomically across multiple data register operands. For example, scheduling block 120 provides hardware micro-converged memory operations stored in scheduling slot 221a to a first load / store pipeline 231a, which atomically co-executes the hardware micro-converged memory operation 250 with cache and memory hierarchy 140 via data path 135.
[0032] Those skilled in the art will recognize that while some exemplary aspects have been discussed above, the guidance of the present invention applies to other aspects. For example, although paired operations have been discussed above, the guidance of the present invention also applies to a large number of grouped operations. Furthermore, specific functions have been discussed in the context of specific hardware blocks, but the functional allocation of these hardware blocks is only used as an example, and the functions discussed can be incorporated into other hardware blocks without departing from the guidance of the present invention.
[0033] Exemplary processor portions supporting the hardware micro-converged memory operations described in this disclosure may be present in or integrated into any processor-based device. Examples include, but are not limited to, servers, computers, laptops, desktop computers, mobile computing devices, set-top boxes, entertainment units, navigation devices, communication devices, fixed positioning data units, mobile positioning data units, Global Positioning System (GPS) devices, mobile phones, smartphones, Session Initiation Protocol (SIP) phones, tablets, phablets, wearable computing devices (e.g., smartwatches, health or fitness trackers, glasses, etc.), personal digital assistants (PDAs), displays, computer monitors, televisions, tuners, radios, satellite radios, music players, digital music players, portable music players, digital video players, video players, digital video disc (DVD) players, portable digital video players, automobiles, vehicle components, avionics systems, drones, and multirotor aircraft.
[0034] in this regard, Figure 5 Displays what can be supported or executed Figure 1-4The present invention illustrates an example of a processor system 500 operating with the hardware micro-converged memory. In this example, the processor system 500 includes a processor 501, which includes one or more central processing units (CPUs) 505, each CPU 505 including one or more processor cores, the processor being commensurate with... Figure 1 Corresponding to processor 105. CPU 505 may be a master device. CPU 505 may include a cache memory 508 coupled to CPU 505 for fast access to temporary stored data. CPU 505 is coupled to system bus 510, enabling master and slave devices in processor system 500 to be coupled to each other. As is well known, CPU 505 communicates with other devices by exchanging address, control, and data information through system bus 510. For example, as an example of a slave device, CPU 505 may convey bus transaction requests to memory controller 551. Although Figure 5 Not shown, however, multiple system buses 510 may be provided, each of which constitutes a different structure.
[0035] Other master and slave devices can be connected to the system bus 510. For example... Figure 5 As shown, as an example, these devices may include a memory system 550, one or more input devices 530, one or more output devices 520, one or more network interface devices 540, and one or more display controllers 560. Input devices 530 may include any type of input device, including but not limited to input keys, switches, voice processors, etc. Output devices 520 may include any type of output device, including but not limited to voice, video, other visual indicators, etc. Network interface devices 540 may be any device capable of exchanging data with network 545. Network 545 may be any type of network, including but not limited to wired or wireless networks, private or public networks, local area networks (LANs), wireless local area networks (WLANs), wide area networks (WANs), Bluetooth™ networks, and the Internet. Network interface devices 540 may support any type of desired communication protocol. Memory system 550 may include a memory controller 551 coupled to one or more memory cells 552.
[0036] The CPU 505 can also access the display controller 560 via the system bus 510 to control the information sent to one or more displays 562. The display controller 560 sends the information to be displayed to the displays 562 through one or more video processors 561, which process the information to be displayed into a format suitable for the displays 562. The displays 562 may include any type of display, including but not limited to cathode ray tube (CRT) displays, liquid crystal displays (LCDs), plasma displays, light-emitting diode (LED) displays, etc.
[0037] Those skilled in the art will further understand that the various illustrative logic blocks, modules, circuits, and algorithms related to the aspects disclosed herein can be implemented as instructions stored in electronic hardware or another computer-readable medium and executed by a processor or other processing device or a combination thereof. As an example, the master and slave devices described herein can be used in any circuit, hardware component, integrated circuit (IC), or IC chip. The memory disclosed herein can be of any type and size and can store any type of desired information. To clearly illustrate this interchangeability, the various illustrative components, blocks, modules, circuits, and steps have been generally described above in terms of their functionality. How such functionality is implemented depends on the specific application, design choices, and / or design constraints imposed on the overall system. Those skilled in the art can implement the described functionality in various ways for each specific application; however, such implementation decisions should not be construed as departing from the scope of the invention.
[0038] The various illustrative logic blocks, modules, and circuits relating to the aspects disclosed herein can be implemented or executed using processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or any combination thereof that perform the functions described herein. The processor may be a microprocessor; however, alternatively, the processor may be any conventional processor, controller, microprocessor, or state machine. The processor may also be implemented as a combination of computing devices (e.g., a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, or any other such configuration).
[0039] The aspects disclosed in this invention may be embodied in hardware and in instructions stored in the hardware, for example, residing in random access memory (RAM), flash memory, read-only memory (ROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), registers, hard disks, removable disks, CD-ROMs, or any other form of computer-readable medium known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information thereto. Alternatively, the storage medium may be an integral part of the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a remote station. Alternatively, the processor and storage medium may reside as discrete components in a remote station, base station, or server.
[0040] It should also be noted that the operational steps described in any exemplary aspect of the invention are intended only to provide examples and discussion. The operations can be performed in various different orders other than those shown. Furthermore, the operations described in a single operational step can actually be performed in multiple different steps. Additionally, one or more operational steps discussed in the exemplary aspects can be combined. It should be understood that various different modifications can be made to the operational steps shown in the flowchart, which will be apparent to those skilled in the art. Those skilled in the art will also understand that various different techniques and skills can be employed to represent information and signals. For example, voltage, current, electromagnetic waves, magnetic fields or magnetic particles, light fields or optical particles, or any combination thereof, can be used to represent data, instructions, commands, information, signals, bits, symbols, and chips that may be mentioned throughout the foregoing description.
[0041] The foregoing description of this invention is intended to enable any person skilled in the art to make or use this invention. Various modifications can be made to this invention, as will be apparent to those skilled in the art, and the general principles defined herein are applicable to other changes. Therefore, this invention is not limited to the examples and designs described herein, but has the broadest scope consistent with the principles and novel features disclosed herein.
Claims
1. An apparatus comprising: A renaming block comprising multiple renaming slots, the renaming block being configured to: The first load operation receives multiple specified data destination register operands; and Renaming the first load operation includes, for each of the data destination register operands: A portion of the first load operation is assigned as a single operand operation to one of the rename slots; as well as Rename the single operand; A scheduling block is configured to receive the renamed first load operation as two independent memory operations, each independent memory operation coming from one of the rename slots of the rename block, store the renamed first load operation in a single entry of multiple scheduling block entries, and schedule the renamed first load operation in a single operation containing multiple data register operands. as well as A memory operation block is configured to receive the scheduled, renamed, first load operation and execute the renamed, first load operation atomically across multiple data destination register operands. The renaming block is configured to rename the first load operation by treating each of the plurality of data register operands as an independent memory operation.
2. The device according to claim 1, wherein, The scheduling block entry for the first load operation after the renaming further stores the operands of the plurality of data destination registers.
3. The device according to claim 1, wherein, The memory operation block is configured to independently track the operands of the plurality of data destination registers.
4. The device according to claim 3, wherein, The memory operation block supports the full data width of the multiple data destination register operands in a single pipeline.
5. The device according to claim 1, wherein, The first loading operation is a loading pair operation.
6. The device according to claim 1, integrated into an integrated circuit (IC).
7. The device of claim 6, further integrated into a device selected from the group consisting of: servers, portable computers, desktop computers, set-top boxes, fixed positioning data units, mobile positioning data units, mobile phones, tablet computers, smartwatches, health or fitness trackers, glasses, personal digital assistants (PDAs), computer monitors, televisions, tuners, radios, digital music players, digital video players, automobiles, avionics systems, and multirotor aircraft.
8. An apparatus comprising: A renaming device is configured to receive a first load operation specifying a plurality of data destination register operands, and to rename the first load operation, including, for each of the data destination register operands: Assign a portion of the first load operation as a single operand operation to one of the rename slots; and Rename the single operand; A scheduling device is configured to receive the renamed first load operation as two independent memory operations, each independent memory operation coming from one of the multiple renaming slots of the renaming device, store the renamed first load operation in a single entry of multiple entries of the scheduling device, and schedule the first memory operation in a single operation form containing multiple data register operands. as well as A memory operation execution device is configured to receive the scheduled, renamed, first load operation and execute the first memory operation atomically across multiple data register operands. The renaming device is configured to rename the first load operation by treating each of the plurality of data register operands as an independent memory operation.
9. A method comprising: The first load operation receives multiple data register operands at the specified rename block. Renaming the first load operation using a renaming block includes treating each of the plurality of data register operands as an independent load operation, including, for each of the data destination register operands: Assign a portion of the first load operation as a single operand operation to one of the rename slots; and Rename the single operand; The first load operation after the renaming is provided to a single entry of the scheduling block as two independent memory operations, each independent memory operation coming from one of the renaming slots; The renamed first load operation is scheduled in a single operation format that includes multiple data register operands; as well as The scheduled, renamed, first load operation is provided to the memory operand block, and the renamed, first load operation is executed atomically across multiple data register operands. The memory operation block is configured to independently track the plurality of data register operands.
10. The method according to claim 9, wherein, It further includes the first load operation after scheduling being performed atomically by a memory operation block.
11. The method according to claim 10, wherein, The first load operation after scheduling is performed by a single pipeline of memory operands across the full data width of multiple data register operands.
12. The method according to claim 9, wherein, The first loading operation is a loading pair operation.
13. A non-transitory computer-readable medium having stored thereon computer-executable instructions that, when executed by a processor, cause the processor to: The receiving specification includes the first load operation of multiple data register operands at a rename block containing multiple rename slots; The first load operation is renamed using a renaming block, including for each of the data destination register operands: A portion of the first load operation is assigned as a single operand operation to one of the rename slots; and Rename the single operand; The first load operation after renaming is provided to a single entry of the scheduling block as two independent memory operations, each independent memory operation coming from one of the renaming slots of the renaming block; The renamed first load operation is scheduled in a single operation format that includes multiple data register operands; Provide the scheduled, renamed, first load operation to the memory operation block; as well as The first load operation after scheduling and renaming is executed atomically by the memory operation block. Renaming the first load operation includes treating each of the multiple data register operands as an independent memory operation.
Citation Information
Patent Citations
Execution method for logarithmic loading instruction
CN108845830A
