This application relates to non-provisional U.S. application Ser. No. 09/584,034, filed on May 30, 2000, and entitled “Slack Fetch to Improve Performance in a Simultaneous and Redundantly Threaded Processor” and to provisional application Ser. No. 60/198,530, filed on Apr. 19, 2000, and entitled “Transient Fault Detection Via Simultaneous Multithreading,” the teachings of both of which are incorporated herein by reference.
Not applicable.
1. Field of the Invention
The present invention generally relates to microprocessors. More particularly, the present invention relates to a computer capable of running multiple threads that can concurrently process instructions in at least two separate threads. More particularly still, the invention relates to pre-executing at least a portion of a program as separate threads to avoid cache misses.
2. Background of the Invention
Most modern computer systems include at least one central processing unit (“CPU”) and a main memory. Multiprocessor systems include more than one processor and each processor typically has its own memory which may or may not be shared by other processors. The speed at which the CPU can decode and execute instructions and operands depends upon the rate at which the instructions and operands can be transferred from main memory to the CPU. In an attempt to reduce the time required for the CPU to obtain instructions and operands from main memory, many computer systems include a cache memory coupled between the CPU and main memory.
A cache memory is a relatively small, high-speed memory (compared to main memory) buffer that is used to temporarily hold those portions of the contents of main memory which it is believed will be used in the near future by the CPU. The main purpose of a cache is to shorten the time necessary to perform memory accesses, both for data and instructions. Cache memory typically has access times that are several or many times faster than a system's main memory. The use of cache memory can significantly improve system performance by reducing data access time, therefore permitting the CPU to spend far less time waiting for instructions and operands to be fetched and/or stored.
A cache memory, typically comprising some form of random access memory (“RAM”) includes many blocks (also called lines) of one or more words of data. Associated with each cache block in the cache is a tag. The tag provides information for mapping the cache line data to its main memory address. Each time the processor makes a memory reference (i.e., load or store), a tag value from the memory address is compared to the tags in the cache to see if a copy of the requested data resides in the cache. If the desired memory block resides in the cache, then the cache's copy of the data is used in the memory transaction, instead of the main memory's copy of the same data block. However, if the desired data block is not in the cache, the block must be retrieved from the main memory and supplied to the processor. A copy of the data also is stored in the cache.
Because the time required to retrieve data from main memory is substantially longer than the time required to retrieve data from cache memory, it is highly desirable have a high cache “hit” rate. Although cache subsystems advantageously increase the performance of a processor, not all memory references result in a cache hit. A cache “miss” occurs when the targeted memory data has not been cached and must be retrieved from main memory. Thus, cache misses detrimentally impact the performance of the processor, while cache hits increase the performance. Anything that can be done to avoid cache misses is highly desirable, and there still remains room for improvement in this regard.
The problems noted above are solved in large part using a processor (e.g., a simultaneous multithreading processor) that can run a program in one thread (called the “main” thread) and at least a portion of the same program in another thread (called the “pre-execution” thread). In general, the concepts apply to any computer that can execute multiple threads of control that share common cache structures. The program in the main thread includes instructions that cause the processor to start and stop pre-execution threads and direct the processor as to which part of the program is to be run through the pre-execution threads. Preferably, such instructions cause the pre-execution thread to run ahead of the main thread in program order (i.e., fetching an instruction from the program before the main thread fetches the same instruction). In this way, preferably any cache miss conditions that are encountered by the pre-execution thread are resolved before the main thread requires that same data. Therefore, the main thread should encounter few or no cache miss conditions, and overall performance is increased.
In accordance with one embodiment, the invention is embodied as a computer system which includes a simultaneous multithreading processor, an I/O controller coupled to said processor, an I/O device coupled to said I/O controller, and a main system memory coupled to said processor, wherein said processor processes a program in a main thread that includes instructions which cause the processor to spawn a pre-execution thread in which at least a portion of the same program executes, said pre-execution thread runs concurrently with the main thread, but ahead of the main thread in program order. Further, another embodiment of the invention includes a method of running a program in a simultaneous multithreading processor comprising inserting pre-execution thread instructions in the program, spawning a pre-execution thread when designated by the inserted instructions, and running said pre-execution thread concurrently with a main thread wherein both the pre-execution and the main threads including instructions from the same program, the pre-execution thread running ahead of the main thread.
In addition to reducing the potential for cache misses to occur, the technique described herein can be used to mitigate the effects of branch mispredictions. These and other benefits will become apparent upon reviewing the following disclosure.
For a detailed description of the preferred embodiments of the invention, reference will now be made to the accompanying drawings in which:
Certain terms are used throughout the following description and claims to refer to particular system components. As one skilled in the art will appreciate, microprocessor companies may refer to a component by different names. This document does not intend to distinguish between components that differ in name but not function. In the following discussion and in the claims, the terms “including” and “comprising” are used in an open-ended fashion, and thus should be interpreted to mean “including, but not limited to . . . ”. Also, the term “couple” or “couples” is intended to mean either an indirect or direct electrical connection. Thus, if a first device couples to a second device, that connection may be through a direct electrical connection, or through an indirect electrical connection via other devices and connections.
In accordance with a preferred embodiment of the invention, a simultaneous multithreading processor is used to reduce the potential for cache misses to occur.
Referring now to
Fetch unit 102 uses one or more program counters 106 for assistance as to which instructions to fetch. Being a multithreading processor, the fetch unit 102 preferably can simultaneously fetch instructions from multiple threads. A separate program counter 106 is associated with each thread. Each program counter 106 is a register that contains the address of the next instruction to be fetched from the corresponding thread by the fetch unit 102.
As shown, fetch unit 102 includes branch prediction logic 103. The branch prediction logic 103 permits the fetch unit 102 to speculate ahead on branch instructions. In order to keep the pipeline full (which is desirable for efficient operation), the branch predictor logic 103 speculates the outcome of a branch instruction before the branch instruction is actually executed. Branch predictor 103 generally bases its speculation on previous instructions. Any suitable speculation algorithm can be used in branch predictor 103.
Referring still to
The register update unit (“RUU”) 130 provides an instruction queue for the instructions to be executed. The RUU 130 serves as a combination of global reservation station pool, rename register file, and reorder buffer. The RUU 130 breaks load and store instructions into an address portion and a memory (i.e., register) reference. The address portion is placed in the RUU 130, while the memory reference portion is placed into a load/store queue (not specifically shown in
The floating point register 122 and integer register 126 are used for the execution of instructions that require the use of such registers as is known by those of ordinary skill in the art. These registers 122, 126 can be loaded with data from the data cache 146. The registers also provide their contents to the RUU 130.
As shown, the execution units 134, 138, and 142 comprise a floating point execution unit 134, a load/store execution unit 138, and an integer execution unit 142. Each execution unit performs the operation specified by the corresponding instruction type. Accordingly, the floating point execution units 134 execute floating instructions such as multiply and divide instruction while the integer execution units 142 execute integer-based instructions. The load/store units 138 perform load operations in which data from memory is loaded into a register 122 or 126. The load/store units 138 also perform store operations in which data from registers 122, 126 is written to data cache 146 and/or DRAM memory 92 (
The architecture and components described herein are typical of microprocessors, and particularly pipelined, multithreaded processors. Numerous modifications can be made from that shown in
In general, any memory reference (e.g., a store or load) in a program will result in a cache miss if the needed data is not currently present in data cache 146. However, in accordance with the preferred embodiment of the invention, the SMT processor 100 can be used to spawn (i.e., create, initialize and execute) a thread which resolves cache misses ahead of the main program. This spawned thread preferably runs concurrently with, but ahead of, the thread in which the main program executes. When the main program then encounters a memory reference, the needed data has already been cached by the previously spawned thread. In this context, the thread in which the main program executes is called the “main” thread and the thread which is spawned to resolve cache misses is called the “pre-execution” thread.
This technique can be understood with reference to
While the main thread is executing code segment 224, the pre-execution thread runs ahead of the main thread and executes code segment 206. Cache misses encountered during the execution of segment 206 by the pre-execution thread preferably are resolved before the main thread encounters the same memory references when it executes code segment 206. Broadly speaking, the pre-execution thread resolves one or more memory references and causes the requested data to be in data cache 146 before the main thread needs the data.
The start/stop instructions shown in
The process of spawning a pre-execution thread generally requires the pre-execution thread to have access to the register data from the main thread. Accordingly, the main thread can write values of the registers that will be needed by the pre-execution thread into memory. Then, the pre-execution thread will load these values from memory before starting pre-execution.
Besides using a stop instruction, such as PreExecute_Stop( ) 226, to terminate a pre-execution thread, there are numerous other ways to terminate a pre-execution thread. For example, a PreExecute_Cancel(T) instruction can be used in the main thread. When the main thread encounters this instruction, the pre-execution thread will be terminated. Additionally, when the pre-execution thread tries to pre-execute a program counter (PC) that is out of the acceptable range of instructions as imposed by the operating system, the pre-execution thread will be terminated. Also, the pre-execution thread can be terminated when the hardware detects the main thread has caught up to the pre-execution thread. Finally, but without limitation, the pre-execution thread can be terminated when the number of instructions it has run exceeds a limit. This makes sure that all pre-execution threads eventually terminate.
Step 256 acknowledges that the pre-execution thread(s) run concurrently with the main thread. While executing, it is desirable for the pre-execution thread to avoid speculative pre-execution from corrupting the program correctness. Accordingly, the preferred implementation ignores all exceptions such as invalid load addresses, division by zero, etc. generated during pre-execution. In this way, the main thread will not encounter any additional exceptions. In addition, pre-execution threads should not be permitted to modify memory contents (e.g., stores). Thus, in accordance with the preferred embodiment, all store instructions encountered during pre-execution are dropped by processor 100 after they are decoded. That is, they are executed as no-ops (“NOPs”).
Instead of simply dropping pre-execution stores, a “scratchpad” buffer can be added to the hardware to buffer the effect of stores during pre-execution. Instead of writing into the cache and the memory, stores in a pre-execution thread will write into the scratchpad buffer which may be coupled to the load/store units 138. In this way, loads in the same pre-execution thread will look up data in both the cache and scratchpad. The main advantage of this approach is that pre-executed loads can observe the effect of earlier stores from the same thread, which may be important to generating future data addresses or maintaining the correct control flow. For instance, if a procedure with a pointer-type argument (e.g., the head of a linked list) is called during pre-execution, both the pointer value and the return address could be passed through the stack. Thus, it would be desirable to be able to read both values back from the stack in the procedure during pre-execution so that the correct data item can be fetched and the procedure can eventually return to its caller as well.
It should be apparent that, although the threads run concurrently with one another, the pre-execution thread preferably does not run the same instructions as the main thread at the same time. That is, by causing the pre-execution thread to jump ahead of the main thread and begin executing at a later point in program order, a “slack” between the two threads is automatically created. This slack facilitates the pre-execution thread to resolve cache misses before the main program requests the data. Finally, in step 258 the pre-execution thread ceases when it encounters its stop execution instruction.
The pre-execution capability described herein is software implemented and controlled as should be apparent. Common uses of software-controlled pre-execution include pre-execution multiple pointer chains, pre-executing loops involving non-affined array references, pre-execution multiple procedure calls, and pre-executing multiple control-flow paths, to name a few. Further, pre-execution may also be useful to avoid branch mispredictions, which is a problem well known to those of ordinary skill in the art and described in U.S. patent application Ser. No. 09/584,034 entitled “Slack Fetch to Improve Performance in a Simultaneous and Redundantly Thread Processor.”
Accordingly, the preferred embodiment of the invention provides a significant performance increase in a processor. The above discussion is meant to be illustrative of the principles and various embodiments of the present invention. Numerous variations and modifications will become apparent to those skilled in the art once the above disclosure is fully appreciated. It is intended that the following claims be interpreted to embrace all such variations and modifications.
Number | Name | Date | Kind |
---|---|---|---|
5138708 | Vosbury | Aug 1992 | A |
5159686 | Chastain et al. | Oct 1992 | A |
5261097 | Saxon | Nov 1993 | A |
5339415 | Strout et al. | Aug 1994 | A |
5724565 | Dubey et al. | Mar 1998 | A |
5758142 | McFarling et al. | May 1998 | A |
5781776 | Johnston et al. | Jul 1998 | A |
5812811 | Dubey et al. | Sep 1998 | A |
5815727 | Motomura | Sep 1998 | A |
5904732 | Greenley et al. | May 1999 | A |
5933860 | Emer et al. | Aug 1999 | A |
5978838 | Mohamed et al. | Nov 1999 | A |
6032245 | Georgiou et al. | Feb 2000 | A |
6199117 | Cox et al. | Mar 2001 | B1 |
6219782 | Khan et al. | Apr 2001 | B1 |
6353881 | Chaudhry et al. | Mar 2002 | B1 |
6457063 | Chintalapati et al. | Sep 2002 | B1 |
6757811 | Mukherjee | Jun 2004 | B1 |
6928645 | Wang et al. | Aug 2005 | B2 |
Number | Date | Country | |
---|---|---|---|
20020055964 A1 | May 2002 | US |
Number | Date | Country | |
---|---|---|---|
60198530 | Apr 2000 | US |