US20040073778A1 - Parallel processor architecture - Google Patents

Parallel processor architecture Download PDF

Info

Publication number
US20040073778A1
US20040073778A1 US10/615,500 US61550003A US2004073778A1 US 20040073778 A1 US20040073778 A1 US 20040073778A1 US 61550003 A US61550003 A US 61550003A US 2004073778 A1 US2004073778 A1 US 2004073778A1
Authority
US
United States
Prior art keywords
processor
memory
microengines
bus
references
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Abandoned
Application number
US10/615,500
Inventor
Matthew Adiletta
Gilbert Wolrich
William Wheeler
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Intel Corp
Original Assignee
Intel Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Intel Corp filed Critical Intel Corp
Priority to US10/615,500 priority Critical patent/US20040073778A1/en
Publication of US20040073778A1 publication Critical patent/US20040073778A1/en
Assigned to INTEL CORPORATION reassignment INTEL CORPORATION ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: WHEELER, WILLIAM R., ADILETTA, MATTHEW J., WOLRICH, GILBERT
Abandoned legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/30003Arrangements for executing specific machine instructions
    • G06F9/30076Arrangements for executing specific machine instructions to perform miscellaneous control operations, e.g. NOP
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00Digital computers in general; Data processing equipment in general
    • G06F15/76Architectures of general purpose stored program computers
    • G06F15/80Architectures of general purpose stored program computers comprising an array of processing units with common control, e.g. single instruction multiple data processors
    • G06F15/8007Architectures of general purpose stored program computers comprising an array of processing units with common control, e.g. single instruction multiple data processors single instruction multiple data [SIMD] multiprocessors
    • G06F15/8015One dimensional arrays, e.g. rings, linear arrays, buses
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/30003Arrangements for executing specific machine instructions
    • G06F9/3004Arrangements for executing specific machine instructions to perform operations on memory
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/30098Register arrangements
    • G06F9/3012Organisation of register space, e.g. banked or distributed register file
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/30098Register arrangements
    • G06F9/3012Organisation of register space, e.g. banked or distributed register file
    • G06F9/30123Organisation of register space, e.g. banked or distributed register file according to context, e.g. thread buffers
    • G06F9/30127Register windows
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38Concurrent instruction execution, e.g. pipeline, look ahead
    • G06F9/3824Operand accessing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38Concurrent instruction execution, e.g. pipeline, look ahead
    • G06F9/3824Operand accessing
    • G06F9/3834Maintaining memory consistency
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38Concurrent instruction execution, e.g. pipeline, look ahead
    • G06F9/3836Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution
    • G06F9/3842Speculative instruction execution
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00Arrangements for program control, e.g. control units
    • G06F9/06Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/30Arrangements for executing machine instructions, e.g. instruction decode
    • G06F9/38Concurrent instruction execution, e.g. pipeline, look ahead
    • G06F9/3836Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution
    • G06F9/3851Instruction issuing, e.g. dynamic instruction scheduling or out of order instruction execution from multiple instruction streams, e.g. multistreaming

Definitions

  • This invention relates to parallel processors.
  • Parallel processing is an efficient form of information processing of concurrent events in a computing process.
  • Parallel processing demands concurrent execution of many programs in a computer, in contrast to sequential processing.
  • parallelism involves doing more than one thing at the same time.
  • a serial paradigm where all tasks are performed sequentially at a single station or a pipelined machine where tasks are performed at specialized stations
  • parallel processing a plurality of stations are provided with each capable of performing all tasks. That is, in general all or a plurality of the stations work simultaneously and independently on the same or common elements of a problem. Certain problems are suitable for solution by applying parallel processing.
  • types of computer processing include single instruction stream, single data stream, the conventional serial von Neumann computer in which there is a single stream of instructions.
  • a second processing type is the single instruction stream, multiple data streams process (SIMD). This processing can have multiple arithmetic-logic processors and a single control processor. Each of the processors perform operations on the data in lock-step. These machines are synchronized by the control processor.
  • SIMD single instruction stream
  • MIMD multiple instruction streams
  • MIMD multiple instruction streams
  • MIMD multiple data streams
  • MIMD multiple data streams
  • a parallel hardware-based multithreaded processor includes a general purpose processor that coordinates system functions and a plurality of microengines that support multiple hardware threads.
  • a parallel hardware-based multithreaded processor includes a general purpose processor that coordinates system functions and a plurality of microengines that support multiple hardware threads.
  • the processor also includes a memory control system including a first memory controller that sorts memory references based on whether the memory references are directed to an even bank or an odd bank of memory and a second memory controller that optimizes memory references based upon whether the memory references are read references or write references.
  • This parallel multithreaded architecture provides a very general purpose microprocessor architecture.
  • the processor can be connected to other devices to process large amounts of data.
  • the system is especially useful for tasks that can be broken into parallel subtasks or functions.
  • the hardware-based multithreaded processor is useful for tasks that are bandwidth oriented rather than latency oriented.
  • the hardware-based multithreaded processor has multiple microengines each with multiple hardware controlled threads that can be simultaneously active and independently work on a task.
  • One example of an application for the hardware-based multithreaded processor is as a network processor.
  • the hardware-based multithreaded processor can interfaces to network devices such as a media access controllers.
  • the hardware-based multithreaded processor can interface to any type of communication device or interface that receives/sends large amounts of data.
  • the processor can be used a print engine for a postscript processor or as a processor for a storage subsystem, i.e., RAID disk storage.
  • a further use is as a matching engine.
  • the advent of electronic trading requires the use of electronic matching engines to match orders between buyers and sellers. These and other parallel types of tasks can be accomplished on the processor used in a system.
  • FIG. 1 is a block diagram of a communication system employing a hardware-based multithreaded processor.
  • FIG. 2 is a detailed block diagram of the hardware-based multithreaded processor of FIG. 1.
  • FIG. 3 is a block diagram of a microengine functional unit employed in the hardware-based multithreaded processor of FIGS. 1 and 2.
  • FIG. 3A is a block diagram of a pipeline in the microengine of FIG. 3.
  • FIG. 3B is diagram that shows a format for a context switch instruction.
  • FIG. 3C is a block diagram showing general purpose register address arrangement.
  • FIG. 4 is a block diagram of a memory controller for enhanced bandwidth operation used in the hardware-based multithreaded processor.
  • FIG. 4A is a flow chart that represents an arbitration policy in an SDRAM controller of FIG. 4.
  • FIG. 4B is a timing diagram that shows advantages of optimizing SDRAM controller.
  • FIG. 5 is a block diagram of a memory controller for latency limited operations used in the hardware-based multithreaded processor.
  • FIG. 5A is a timing diagram that shows advantages of optimizing SRAM controller.
  • FIG. 6 is a block diagram of a communication bus interface in the processor of FIG. 1.
  • a communication system 10 includes a parallel, hardware-based multithreaded processor 12 .
  • the hardware-based multithreaded processor 12 is coupled to a bus such as a PCI bus 14 , a memory system 16 and a second bus 18 .
  • the system 10 is especially useful for tasks that can be broken into parallel subtasks or functions.
  • hardware-based multithreaded processor 12 is useful for tasks that are bandwidth oriented rather than latency oriented.
  • the hardware-based multithreaded processor 12 has multiple microengines 22 each with multiple hardware controlled threads that can be simultaneously active and independently work on a task.
  • the hardware-based multithreaded processor 12 also includes a central controller 20 that assists in loading microcode control for other resources of the hardware-based multithreaded processor 12 and performs other general purpose computer type functions such as handling protocols, exceptions, extra support for packet processing where the microengines pass the packets off for more detailed processing such as in boundary conditions.
  • the processor 20 is a Strong Arm® (Arm is a trademark of ARM Limited, United Kingdom) based architecture.
  • the general purpose microprocessor 20 has an operating system. Through the operating system the processor 20 can call functions to operate on microengines 22 a - 22 f .
  • the processor 20 can use any supported operating system preferably a real time operating system.
  • operating systems such as, MicrosoftNT real-time, VXWorks and ⁇ CUS, a freeware operating system available over the Internet, can be used.
  • the hardware-based multithreaded processor 12 also includes a plurality of function microengines 22 a - 22 f .
  • Functional microengines (microengines) 22 a - 22 f each maintain a plurality of program counters in hardware and states associated with the program counters. Effectively, a corresponding plurality of sets of threads can be simultaneously active on each of the microengines 22 a - 22 f while only one is actually operating at any one time.
  • each microengine 22 a - 22 f has capabilities for processing four hardware threads.
  • the six microengines 22 a - 22 f operate with shared resources including memory system 16 and bus interfaces 24 and 28 .
  • the memory system 16 includes a Synchronous Dynamic Random Access Memory (SDRAM) controller 26 a and a Static Random Access Memory (SRAM) controller 26 b .
  • SDRAM memory 16 a and SDRAM controller 26 a are typically used for processing large volumes of data, e.g., processing of network payloads from network packets.
  • the SRAM controller 26 b and SRAM memory 16 b are used in a networking implementation for low latency, fast access tasks, e.g., accessing look-up tables, memory for the core processor 20 , and so forth.
  • the six microengines 22 a - 22 f access either the SDRAM 16 a or SRAM 16 b based on characteristics of the data. Thus, low latency, low bandwidth data is stored in and fetched from SRAM, whereas higher bandwidth data for which latency is not as important, is stored in and fetched from SDRAM.
  • the microengines 22 a - 22 f can execute memory reference instructions to either the SDRAM controller 26 a or SRAM controller 16 b.
  • SRAM or SDRAM memory accesses can be explained by SRAM or SDRAM memory accesses.
  • an SRAM access requested by a Thread_ 0 from a microengine will cause the SRAM controller 26 b to initiate an access to the SRAM memory 16 b .
  • the SRAM controller controls arbitration for the SRAM bus, accesses the SRAM 16 b , fetches the data from the SRAM 16 b , and returns data to a requesting microengine 22 a - 22 b .
  • the microengine e.g., 22 a had only a single thread that could operate, that microengine would be dormant until data was returned from the SRAM.
  • Thread_ 1 can function while the first thread, e.g., Thread_ 0 , is awaiting the read data to return.
  • Thread_ 1 may access the SDRAM memory 16 a .
  • Thread_ 1 operates on the SDRAM unit, and Thread_ 0 is operating on the SRAM unit, a new thread, e.g., Thread_ 2 can now operate in the microengine 22 a .
  • Thread_ 2 can operate for a certain amount of time until it needs to access memory or perform some other long latency operation, such as making an access to a bus interface. Therefore, simultaneously, the processor 12 can have a bus operation, SRAM operation and. SDRAM operation all being completed or operated upon by one microengine 22 a and have one more thread available to process more work in the data path.
  • the hardware context swapping also synchronizes completion of tasks. For example, two threads could hit the same shared resource e.g., SRAM.
  • Each one of these separate functional units, e.g., the FBUS interface 28 , the SRAM controller 26 a , and the SDRAM controller 26 b when they complete a requested task from one of the microengine thread contexts reports back a flag signaling completion of an operation.
  • the microengine can determine which thread to turn on.
  • One example of an application for the hardware-based multithreaded processor 12 is as a network processor.
  • the hardware-based multithreaded processor 12 interfaces to network devices such as a media access controller device e.g., a 10/100BaseT Octal MAC 13 a or a Gigabit Ethernet device 13 b .
  • the hardware-based multithreaded processor 12 can interface to any type of communication device or interface that receives/sends large amounts of data.
  • Communication system 10 functioning in a networking application could receive a plurality of network packets from the devices 13 a , 13 b and process those packets in a parallel manner. With the hardware-based multithreaded processor 12 , each network packet can be independently processed.
  • processor 12 Another example for use of processor 12 is a print engine for a postscript processor or as a processor for a storage subsystem, i.e., RAID disk storage.
  • a further use is as a matching engine.
  • the advent of electronic trading requires the use of electronic matching engines to match orders between buyers and sellers. These and other parallel types of tasks can be accomplished on the system 10 .
  • the processor 12 includes a bus interface 28 that couples the processor to the second bus 18 .
  • Bus interface 28 in one embodiment couples the processor 12 to the so-called FBUS 18 (FIFO bus).
  • the FBUS interface 28 is responsible for controlling and interfacing the processor 12 to the FBUS 18 .
  • the FBUS 18 is a 64-bit wide FIFO bus, used to interface to Media Access Controller (MAC) devices.
  • MAC Media Access Controller
  • the processor 12 includes a second interface e.g., a PCI bus interface 24 that couples other system components that reside on the PCI 14 bus to the processor 12 .
  • the PCI bus interface 24 provides a high speed data path 24 a to memory 16 e.g., the SDRAM memory 16 a . Through that path data can be moved quickly from the SDRAM 16 a through the PCI bus 14 , via direct memory access (DMA) transfers.
  • DMA direct memory access
  • the hardware based multithreaded processor 12 supports image transfers.
  • the hardware based multithreaded processor 12 can employ a plurality of DMA channels so if one target of a DMA transfer is busy, another one of the DMA channels can take over the PCI bus to deliver information to another target to maintain high processor 12 efficiency.
  • the PCI bus interface 24 supports target and master operations.
  • Target operations are operations where slave devices on bus 14 access SDRAMs through reads and writes that are serviced as a slave to target operation.
  • master operations the processor core 20 sends data directly to or receives data directly from the PCI interface 24 .
  • Each of the functional units are coupled to one or more internal buses.
  • the internal buses are dual, 32 bit buses (i.e., one bus for read and one for write).
  • the hardware-based multithreaded processor 12 also is constructed such that the sum of the bandwidths of the internal buses in the processor 12 exceed the bandwidth of external buses coupled to the processor 12 .
  • the processor 12 includes an internal core processor bus 32 , e.g., an ASB bus (Advanced System Bus) that couples the processor core 20 to the memory controller 26 a , 26 c and to an ASB translator 30 described below.
  • the ASB bus is a subset of the so called AMBA bus that is used with the Strong Arm processor core.
  • the processor 12 also includes a private bus 34 that couples the microengine units to SRAM controller 26 b , ASB translator 30 and FBUS interface 28 .
  • a memory bus 38 couples the memory controller 26 a , 26 b to the bus interfaces 24 and 28 and memory system 16 including flashrom 16 c used for boot operations and so forth.
  • each of the microengines 22 a - 22 f includes an arbiter that examines flags to determine the available threads to be operated upon. Any thread from any of the microengines 22 a - 22 f can access the SDRAM controller 26 a , SDRAM controller 26 b or FBUS interface 28 .
  • the memory controllers 26 a and 26 b each include a plurality of queues to store outstanding memory reference requests. The queues either maintain order of memory references or arrange memory references to optimize memory bandwidth. For example, if a thread_ 0 has no dependencies or relationship to a thread_ 1 , there is no reason that thread 1 and 0 cannot complete their memory references to the SRAM unit out of order.
  • the microengines 22 a - 22 f issue memory reference requests to the memory controllers 26 a and 26 b .
  • the microengines 22 a - 22 f flood the memory subsystems 26 a and 26 b with enough memory reference operations such that the memory subsystems 26 a and 26 b become the bottleneck for processor 12 operation.
  • the processor 12 can perform memory reference sorting.
  • Memory reference sorting improves achievable memory bandwidth.
  • Memory reference sorting reduces dead time or a bubble that occurs with accesses to SRAM. With memory references to SRAM, switching current direction on signal lines between reads and writes produces a bubble or a dead time waiting for current to settle on conductors coupling the SRAM 16 b to the SRAM controller 26 b.
  • Memory reference sorting allows the processor 12 to organize references to memory such that long strings of reads can be followed by long strings of writes. This can be used to minimize dead time in the pipeline to effectively achieve closer to maximum available bandwidth. Reference sorting helps maintain parallel hardware context threads. On the SDRAM, reference sorting allows hiding of pre-charges from one bank to another bank. Specifically, if the memory system 16 b is organized into an odd bank and an even bank, while the processor is operating on the odd bank, the memory controller can start precharging the even bank. Precharging is possible if memory references alternate between odd and even banks.
  • the processor 12 By ordering memory references to alternate accesses to opposite banks, the processor 12 improves SDRAM bandwidth. Additionally, other optimizations can be used. For example, merging optimizations where operations that can be merged, are merged prior to memory access, open page optimizations where by examining addresses an opened page of memory is not reopened, chaining, as will be described below, and refreshing mechanisms, can be employed.
  • the FBUS interface 28 supports Transmit and Receive flags for each port that a MAC device supports, along with an Interrupt flag indicating when service is warranted.
  • the FBUS interface 28 also includes a controller 28 a that performs header processing of incoming packets from the FBUS 18 .
  • the controller 28 a extracts the packet headers and performs a microprogrammable source/destination/protocol hashed lookup (used for address smoothing) in SRAM. If the hash does not successfully resolve, the packet header is sent to the processor core 20 for additional processing.
  • the FBUS interface 28 supports the following internal data transactions: FBUS unit (Shared bus SRAM) to/from microengine. FBUS unit (via private bus) writes from SDRAM Unit. FBUS unit (via Mbus) Reads to SDRAM.
  • the FBUS 18 is a standard industry bus and includes a data bus, e.g., 64 bits wide and sideband control for address and read/write control.
  • the FBUS interface 28 provides the ability to input large amounts of data using a series of input and output FIFO's 29 a - 29 b .
  • the microengines 22 a - 22 f fetch data from or command the SDRAM controller 26 a to move data from a receive FIFO in which data has come from a device on bus 18 , into the FBUS interface 28 .
  • the data can be sent through memory controller 26 a to SDRAM memory 16 a , via a direct memory access.
  • the microengines can move data from the SDRAM 26 a to interface 28 , out to FBUS 18 , via the FBUS interface 28 .
  • a command request can be a memory request or a FBUS request.
  • a command request can move data from a register located in a microengine 22 a to a shared resource, e.g., an SDRAM location, SRAM location, flash memory or some MAC address.
  • the commands are sent out to each of the functional units and the shared resources.
  • the shared resources do not need to maintain local buffering of the data. Rather, the shared resources access distributed data located inside of the microengines.
  • microengines 22 a - 22 f This enables microengines 22 a - 22 f , to have local access to data rather than arbitrating for access on a bus and risk contention for the bus. With this feature, there is a 0 cycle stall for waiting for data internal to the microengines 22 a - 22 f.
  • the data buses, e.g., ASB bus 30 , SRAM bus 34 and SDRAM bus 38 coupling these shared resources, e.g., memory controllers 26 a and 26 b are of sufficient bandwidth such that there are no internal bottlenecks.
  • the processor 12 has an bandwidth requirement where each of the functional units is provided with at least twice the maximum bandwidth of the internal buses.
  • the SDRAM can run a 64 bit wide bus at 83 MHz.
  • the SRAM data bus could have separate read and write buses, e.g., could be a read bus of 32 bits wide running at 166 MHz and a write bus of 32 bits wide at 166 MHz. That is, in essence, 64 bits running at 166 MHz which is effectively twice the bandwidth of the SDRAM.
  • the core processor 20 also can access the shared resources.
  • the core processor 20 has a direct communication to the SDRAM controller 26 a to the bus interface 24 and to SRAM controller 26 b via bus 32 .
  • the core processor 20 access the microengines 22 a - 22 f via the ASB Translator 30 over bus 34 .
  • the ASB translator 30 can physically reside in the FBUS interface 28 , but logically is distinct.
  • the ASB Translator 30 performs an address translation between FBUS microengine transfer register locations and core processor addresses (i.e., ASB bus) so that the core processor 20 can access registers belonging to the microengines 22 a - 22 c.
  • a scratchpad memory 27 is also provided to permit microengines to write data out to the memory for other microengines to read.
  • the scratchpad 27 is coupled to bus 34 .
  • the processor core 20 includes a RISC core 50 implemented in a five stage pipeline performing a single cycle shift of one operand or two operands in a single cycle, provides multiplication support and 32 bit barrel shift support.
  • This RISC core 50 is a standard Strong Arm® architecture but it is implemented with a five stage pipeline for performance reasons.
  • the processor core 20 also includes a 16 kilobyte instruction cache 52 , an 8 kilobyte data cache 54 and a prefetch stream buffer 56 .
  • the core processor 20 performs arithmetic operations in parallel with memory writes and instruction fetches.
  • the core processor 20 interfaces with other functional units via the ARM defined ASB bus.
  • the ASB bus is a 32-bit bi-directional bus 32 .
  • the microengine 22 f includes a control store 70 which, in one implementation, includes a RAM of here 1,024 words of 32 bit.
  • the RAM stores a microprogram.
  • the microprogram is loadable by the core processor 20 .
  • the microengine 22 f also includes controller logic 72 .
  • the controller logic includes an instruction decoder 73 and program counter (PC) units 72 a - 72 d .
  • the four micro program counters 72 a - 72 d are maintained in hardware.
  • the microengine 22 f also includes context event switching logic 74 .
  • Context event logic 74 receives messages (e.g., SEQ_#_EVENT_RESPONSE; FBI_EVENT_RESPONSE; SAM_EVENT_RESPONSE; SDRAM_EVENT_RESPONSE; and ASB_EVENT_RESPONSE) from each one of the shared resources, e.g., SRAM 26 a , SDRAM 26 b , or processor core 20 , control and status registers, and so forth. These messages provide information on whether a requested function has completed. Based on whether or not a function requested by a thread has completed and signaled completion, the thread needs to wait for that completion signal, and if the thread is enabled to operate, then the thread is placed on an available thread list (not shown).
  • the microengine 22 f can have a maximum of e.g., 4 threads available.
  • the microengines 22 In addition to event signals that are local to an executing thread, the microengines 22 employ signaling states that are global. With signaling states, an executing thread can broadcast a signal state to all microengines 22 . Receive Request Available signal, Any and all threads in the microengines can branch on these signaling states. These signaling states can be used to determine availability of a resource or whether a resource is due for servicing.
  • the context event logic 74 has arbitration for the four (4) threads. In one embodiment, the arbitration is a round robin mechanism. Other techniques could be used including priority queuing or weighted fair queuing.
  • the microengine 22 f also includes an execution box (EBOX) data path 76 that includes an arithmetic logic unit 76 a and general purpose register set 76 b .
  • the arithmetic logic unit 76 a performs arithmetic and logical functions as well as shift functions.
  • the registers set 76 b has a relatively large number of general purpose registers. As will be described in FIG. 3B, in this implementation there are 64 general purpose registers in a first bank, Bank A and 64 in a second bank, Bank B.
  • the general purpose registers are windowed as will be described so that they are relatively and absolutely addressable.
  • the microengine 22 f also includes a write transfer register stack 78 and a read transfer stack 80 . These registers are also windowed so that they are relatively and absolutely addressable.
  • Write transfer register stack 78 is where write data to a resource is located.
  • read register stack 80 is for return data from a shared resource. Subsequent to or concurrent with data arrival, an event signal from the respective shared resource e.g., the SRAM controller 26 a , SDRAM controller 26 b or core processor 20 will be provided to context event arbiter 74 which will then alert the thread that the data is available or has been sent.
  • Both transfer register banks 78 and 80 are connected to the execution box (EBOX) 76 through a data path.
  • the read transfer register has 64 registers and the write transfer register has 64 registers.
  • the microengine datapath maintains a 5-stage micro-pipeline 82 .
  • This pipeline includes lookup of microinstruction words 82 a , formation of the register file addresses 82 b , read of operands from register file 82 c , ALU, shift or compare operations 82 d , and write-back of results to registers 82 e .
  • the microengine can perform a simultaneous register file read and write, which completely hides the write operation.
  • the SDRAM interface 26 a provides a signal back to the requesting microengine on reads that indicates whether a parity error occurred on the read request.
  • the microengine microcode is responsible for checking the SDRAM read Parity flag when the microengine uses any return data. Upon checking the flag, if it was set, the act of branching on it clears it. The Parity flag is only sent when the SDRAM is enabled for checking, and the SDRAM is parity protected.
  • the microengines and the PCI Unit are the only requesters notified of parity errors. Therefore, if the processor core 20 or FIFO requires parity protection, a microengine assists in the request.
  • the microengines 22 a - 22 f support conditional branches.
  • the worst case conditional branch latency occurs when the branch decision is a result of condition codes being set by the previous microcontrol instruction.
  • the latency is shown below in Table 1: 1 2 3 4 5 6 7 8 microstore lookup n1 cb n2 XX b1 b2 b3 b4 reg addr gen n1 cb XX XX b1 b2 b3 reg file lookup n1 cb XXXX b1 b2 ALU/shifter/cc n1 cb XXXX b1 write back m2 n1 cb XXXXX
  • nx is pre-branch microword (n 1 sets cc's)
  • the microengines support deferred branches.
  • Deferring branches are when a microengine allows 1 or 2 microwords after the branch to occur before the branch takes effect (i.e. the effect of the branch is “deferred” in time). Thus, if useful work can be found to fill the wasted cycles after the branch microword, then the branch latency can be hidden.
  • a 1-cycle deferred branch is shown below where n 2 is allowed to execute after cb, but before b 1 : 1 2 3 4 5 6 7 8 microstore lookup n1 cb n2 XX b1 b2 b3 b4 reg addr gen n1 cb n2 XX b1 b2 b3 reg file lookup n1 cb n2 XX b1 b2 ALU/shifter/cc n1 cb n2 XX b1 write back n1 cb n2 XX
  • a 2-cycle deferred branch is shown below, where n 2 and n 3 are both allowed to complete before the branch to b 1 occurs. Note that a 2-cycle branch deferment is only allowed when the condition codes are set on the microword preceding the branch.
  • the microengines also support condition code evaluation. If the condition codes upon which a branch decision are made are set 2 or more microwords before the branch, then 1 cycle of branch latency can be eliminated because the branch decision can be made 1 cycle earlier: 1 2 3 4 5 6 7 8 microstore lookup n1 n2 cb XX b1 b2 b3 b4 reg addr gen n1 n2 cb XX b1 b2 b3 reg file lookup n1 n2 cb XX b1 b2 ALU/shifter/cc n1 n2 cb XX b1 write back n1 n2 cb XX
  • n 1 sets the condition codes and n 2 does not set the conditions codes. Therefore, the branch decision can be made at cycle 4 (rather than 5), to eliminate 1 cycle of branch latency.
  • cycle 4 (rather than 5)
  • the 1-cycle branch deferment and early setting of condition codes are combined to completely hide the branch latency:
  • Condition codes set 2 cycles before a 1-cycle deferred branch: 1 2 3 4 5 6 7 8 microstore lookup n1 n2 cb n3 b1 b2 b3 b4 reg addr gen n1 n2 cb n3 b1 b2 b3 reg file lookup n1 n2 cb n3 b1 b2 ALU/shifter/cc n1 n2 cb n3 b1 write back n1 n2 cb n3
  • the microengine supports branch guessing which attempts to reduce the 1 cycle of exposed branch latency that remains. By “guessing” the branch path or the sequential path, the microsequencer pre-fetches the guessed path 1 cycle before it definitely knows what path to execute. If it guessed correctly, 1 cycle of branch latency is eliminated as shown below:
  • guess branch taken/branch is taken 1 2 3 4 5 6 7 8 microstore lookup n1 cb n1 b1 b2 b3 b4 b5 reg addr gen n1 cb XX b1 b2 b3 b4 reg file lookup n1 cb XX b1 b2 b3 ALU/shifter/cc n1 cb XX b1 b2 write back n1 cb XX b1
  • guess branch taken/branch is NOT taken 1 2 3 4 5 6 7 8 microstore lookup n1 cb n1 XX n2 n3 n4 n5 reg addr gen n1 cb n1 XX n2 n3 n4 reg file lookup n1 cb n1 XX n2 n3 ALU/shifter/cc n1 cb n1 XX n2 write back n1 cb n1 XX
  • the microengine can combine branch guessing with 1-cycle branch deferment to improve the result further.
  • For guess branch taken with 1-cycle deferred branch/branch is taken is: 1 2 3 4 5 6 7 8 microstore lookup n1 cb n2 b1 b2 b3 b4 b5 reg addr gen n1 cb n2 b1 b2 b3 b4 reg file lookup n1 cb n2 b1 b2 b3 ALU/shifter/cc n1 cb n2 b1 b2 write back n1 cb n2 b1
  • microcode correctly guesses a branch NOT taken then the pipeline flows sequentially in the normal unperturbed case. If microcode incorrectly guesses branch NOT taken, the microengine again exposes 1 cycle of unproductive execution as shown below:
  • nx is pre-branch microword (n 1 sets cc's)
  • a context switch is a special form of a branch that causes a different context (and associated PC) to be selected. Context switching introduces some branch latency as well.
  • Context switching introduces some branch latency as well.
  • ca is context rearbitration (causes context switch)
  • nx is new context flow
  • Conditional branches that operate on ALU condition codes which are set on the microword before the branch can select 0, 1 or 2-cycle branch deferment modes.
  • Condition codes set 2 or more microwords before the conditional branch that operates on them can select 0 or 1-cycle branch deferment modes. All other branches (including context rearbitrations) can select either 0 or 1-cycle branch deferment modes.
  • the architecture could be designed to make a context arbitration microword within a branch deferment window of a preceding branch, jump or context arbitration microword, an illegal option. That is, in some embodiments, a context switch would not be allowed to occur during a branch transition in the pipeline because as mentioned, it could unduly complicate saving of the old context PC.
  • the architecture could also be designed to make branching within the branch deferment window of a preceding branch, jump or context arbitration microword illegal to avoid complicated and possible unpredictable branch behaviors.
  • Each microengine 22 a - 22 f supports multi-threaded execution of four contexts.
  • One reason for this is to allow one thread to start executing just after another thread issues a memory reference and must wait until that reference completes before doing more work. This behavior is critical to maintaining efficient hardware execution of the microengines because memory latency is significant. Stated differently, if only a single thread execution was supported, the microengines would sit idle for a significant number of cycles waiting for references to return and thereby reduce overall computational throughput.
  • Multi-threaded execution allows an microengines to hide memory latency by performing useful independent work across several threads. Two synchronization mechanisms are supplied in order to allow a thread to issue an SRAM or SDRAM reference, and then subsequently synchronize to the point in time when that reference completes.
  • One mechanism is Immediate Synchronization.
  • immediate synchronization the microengine issues the reference and immediately swap out that context.
  • the context will be signaled when the corresponding reference completes. Once signaled, the context will be swapped back in for execution when a context-swap event occurs and it is its turn to run.
  • the microword after issuing the mem reference does not get executed until the reference completes.
  • a second mechanism is Delayed Synchronization.
  • the microengine issues the reference, and then continues to execute some other useful work independent of the reference. Some time later it could become necessary to synchronize the thread's execution stream to the completion of the issued reference before further work is performed. At this point a synchronizing microword is executed that will either swap out the current thread, and swap it back in sometime later when the reference has completed, or continue executing the current thread because the reference has already completed. Delayed synchronization is implemented using two different signaling schemes:
  • the signal from which the thread is triggered is generated when the corresponding transfer register valid bit is set or cleared. For example, an SRAM read which deposits data into transfer register A would be signaled when the valid bit for A is set. If the memory reference is associated with the transfer FIFO or the receive FIFO, instead of a transfer register, then the signal is generated when the reference completes in the SDRAM controller 26 a . Only one signal state per context is held in the microengines scheduler, thus only one outstanding signal can exist in this scheme.
  • microcontroller micro-programs There are at least two general operational paradigms from which microcontroller micro-programs could be designed. One would be that overall microcontroller compute throughput and overall memory bandwidth are optimized at the expense of single thread execution latency. This paradigm would make sense when the system has multiple microengines executing multiple threads per microengine on unrelated data packets.
  • a second one is that microengine execution latency should be optimized at the expense of overall microengine compute throughput and overall memory bandwidth.
  • This paradigm could involve execution of a thread with a real-time constraint, that is, a constraint which dictates that some work must absolutely be done by some specified time. Such a constraint requires that optimization of the single thread execution be given priority over other considerations such as memory bandwidth or overall computational throughput.
  • a real-time thread would imply a single microengine that executes only one thread. Multiple threads would not be handled because the goal is to allow the single real-time thread to execute as soon as possible—execution of multiple threads would hinder this ability.
  • the two register address spaces that exist are Locally accessibly registers, and Globally accessible registers accessible by all microengines.
  • Each bank is capable of performing a simultaneous read and write to two different words within its bank.
  • the register set 76 b is also organized into four windows 76 b 0 - 76 b 3 of 32 registers that are relatively addressable per thread.
  • thread_ 0 will find its register 0 at 77 a (register 0 )
  • the thread_ 1 will find its register_ 0 at 77 b (register 32 )
  • thread_ 2 will find its register_ 0 at 77 c (register 64 )
  • thread_ 3 at 77 d (register 96 ).
  • Relative addressing is supported so that multiple threads can use the exact same control store and locations but access different windows of register and perform different functions.
  • the uses of register window addressing and bank addressing provide the requisite read bandwidth using only dual ported RAMS in the microengine 22 f.
  • windowed registers do not have to save data from context switch to context switch so that the normal push and pop of a context swap file or stack is eliminated.
  • Context switching here has a 0 cycle overhead for changing from one context to another.
  • Relative register addressing divides the register banks into windows across the address width of the general purpose register set. Relative addressing allows access any of the windows relative to the starting point of the window. Absolute addressing is also supported in this architecture where any one of the absolute registers may be accessed by any of the threads by providing the exact address of the register.
  • SDRAM
  • a 6 1
  • a 5 0
  • a 4 1
  • register address directly specified in 8-bit dest field (d 7 -d 0 ): 7 6 5 4 3 2 1 0
  • microengines are not interrupt driven. Each microflow executes until completion and then a new flow is chosen based on state signaled by other devices in the processor 12 .
  • the SDRAM memory controller 26 a includes memory reference queues 90 where memory reference requests arrive from the various microengines 22 a - 22 f .
  • the memory controller 26 a includes an arbiter 91 that selects the next the microengine reference requests to go to any of the functioning units. Given that one of the microengines is providing a reference request, the reference request will come through the address and command queue 90 , inside the SDRAM controller 26 a . If the reference request has a bit set called the “optimized MEM bit” the incoming reference request will be sorted into either the even bank queue 90 a or the odd bank queue 90 b .
  • the SDRAM controller 26 is a resource which is shared among the FBUS interface 28 , the core processor 20 and the PCI interface 24 .
  • the SDRAM controller 26 also maintains a state machine for performing READ-MODIFY-Write atomic operations.
  • the SDRAM controller 26 also performs byte alignment for requests of data from SDRAM.
  • the order queue 90 c maintains the order of reference requests from the microengines. With a series of odd and even banks references it may be required that a signal is returned only upon completion of a sequence of memory references to both the odd and even banks. If the microengine 22 f sorts the memory references into odd bank and even bank references and one of the banks, e.g., the even bank is drained of memory references before the odd bank but the signal is asserted on the last even reference, the memory controller 26 a could conceivably signal back to a microengine that the memory request had completed, even though the odd bank reference had not been serviced. This occurrence could cause a coherency problem. The situation is avoided by providing the order queue 90 c allowing a microengine to have multiple memory references outstanding of which only its last memory reference needs to signal a completion.
  • the SDRAM controller 26 a also includes a high priority queue 90 d .
  • the high priority queue 90 d an incoming memory reference from one of the microengines goes directly to the high priority queue and is operated upon at a higher priority than other memory references in the other queues. All of these queues, the even bank queue 90 a , the odd bank queue 90 b , the order queue 90 c and the high priority queue, are implemented in a single RAM structure that is logically segmented into four different windows, each window having its own head and tail pointer. Since filling and draining operations are only a single input and a single output, they can be placed into the same RAM structure to increase density of RAM structures.
  • the SDRAM controller 26 a also includes core bus interface logic i.e., ASB bus 92 .
  • the ASB bus interface logic 92 interfaces the core processor 20 to the SDRAM controller 26 a .
  • the ASB bus is a bus that includes a 32 bit data path and a 28 bit address path.
  • the data is accessed to and from memory through MEM ASB data device 98 , e.g., a buffer.
  • MEM ASB data device 98 is a queue for write data. If there is incoming data from the core processor 20 via ASB interface 92 , the data can be stored into the MEM ASB device 98 and subsequently removed from MEM ASB device 98 through the SDRAM interface 110 to SDRAM memory 16 a .
  • the same queue structure can be provided for the reads.
  • the SDRAM controller 26 a also includes an engine 97 to pull data from the microengines and PCI bus.
  • Additional queues include the PCI address queue 94 and ASB read/write queue 96 that maintain a number of requests.
  • the memory requests are sent to SDRAM interface 110 via multiplexer 106 .
  • the multiplexer 106 is controlled by the SDRAM arbiter 91 which detects the fullness of each of the queues and the status of the requests and from that decides priority based on a programmable value stored in a priority service control register 100 .
  • bus 112 is actually two separate buses instead of a single bus. The separate buses would include a read bus coupling the distributed microengines 22 a - 22 f and a write bus coupling the distributed microengines 22 a - 22 f.
  • a feature of the SDRAM controller 26 a is that when a memory reference is stored in the queues 90 , in addition to the optimized MEM bit that can be set, there is a “chaining bit”.
  • the chaining bit when set allows for special handling of contiguous memory references.
  • the arbiter 12 controls which microengine will be selected to provide memory reference requests over the commander bus to queue 90 (FIG. 4). Assertion of the chain bit will control the arbiter to have the arbiter select the functional unit which previously requested that bus because setting of the chain bit indicates that the microengine issued a chain request.
  • Contiguous memory references will be received in queue 90 when the chaining bit is set. Those contiguous references will typically be stored in the order queue 90 c because the contiguous memory references are multiple memory references from a single thread.
  • the memory controller 26 a need only signal at the end of the chained memory references when done.
  • the memory references could go into different banks and potentially complete on one of the banks issuing the signal “done” before the other bank was fully drained, thus destroying coherency. Therefore, the chain bit is used by the controller 110 to maintain the memory references from the current queue.
  • the arbitration policy favors chained microengine memory requests.
  • the process 115 starts by examining for Chained microengine memory reference requests 115 a .
  • the process 115 stays at the chained requests until the chain bit is cleared.
  • the process examines ASB bus requests 115 b followed by PCI bus requests 115 c , High Priority Queue Service 115 d , Opposite Bank Requests 115 e , Order Queue Requests 115 f , and Same Bank Requests 115 g . Chained request are serviced completely, whereas services 115 b -115 d are serviced in round robin order.
  • FIG. 4B typical timing of a memory without active memory optimization and with active memory optimization is shown.
  • the use of active memory optimizations maximizes the use of the bus and thus hides the inherent latency within physical SDRAM devices.
  • a non-optimized access can take 14 cycles while optimized access can take 7 cyles.
  • the memory controller 26 b for the SRAM includes an address and command queue 120 . While the memory controller 26 a (FIG. 4) has a queue for memory optimization based on odd and even banking, memory controller 26 b is optimized based on the type of memory operation, i.e., a read or a write.
  • the address and command queue 120 includes a high priority queue 120 a , a read queue 120 b which is the predominant memory reference function that an SRAM performs, and an order queue 120 c which in general will include all writes to SRAM and reads that are to be non-optimized. Although not shown, the address and command queue 120 could also include a write queue.
  • the SRAM controller 26 b also includes core bus interface logic i.e., ASB bus 122 .
  • the ASB bus interface logic 122 interfaces the core processor 20 to the SRAM controller 26 b .
  • the ASB bus is a bus that includes a 32 bit data path and a 28 bit address path.
  • the data is accessed to and from memory through MEM ASB data device 128 , e.g., a buffer.
  • MEM ASB data device 128 is a queue for write data. If there is incoming data from the core processor 20 via ASB interface 122 , the data can be stored into the MEM ASB device 128 and subsequently removed from MEM ASB device 128 through SRAM interface 140 to SRAM memory 16 b .
  • the same queue structure can be provided for reads.
  • the SRAM controller 26 b also includes an engine 127 to pull data from the microengines and PCI bus.
  • the memory requests are sent to SRAM interface 140 via multiplexer 126 .
  • the multiplexer 126 is controlled by the SRAM arbiter 131 which detects the fullness of each of the queues and the status of the requests and from that decides priority based on a programmable value stored in a priority service control register 130 . Once control to the multiplexer 126 selects a memory reference request, the memory reference request, is sent to a decoder 138 where it is decoded and an address is generated.
  • the SRAM Unit maintains control of the Memory Mapped off-chip SRAM and Expansion ROM.
  • the SRAM controller 26 b can address, e.g., 16 MBytes, with, e.g., 8 MBytes mapped for SRAM 16 b , and 8 MBytes reserved for special functions including: Boot space via flashrom 16 c ; and Console port access for MAC devices 13 a , 13 b and access to associated (RMON) counters.
  • the SRAM is used for local look-up tables and queue management functions.
  • the SRAM controller 26 b supports the following transactions: Microengine requests (via private bus) to/from SRAM. Core Processor (via ASB bus) to/from SRAM.
  • the SRAM controller 26 b performs memory reference sorting to minimize delays (bubbles) in the pipeline from the SRAM interface 140 to memory 16 b .
  • the SRAM controller 26 b does memory reference sorting based on the read function.
  • a bubble can either be 1 or 2 cycles depending on the type of memory device employed.
  • the SRAM controller 26 b includes a lock lookup device 142 which is an eight (8 entry address content addressable memory for look-ups of read locks. Each position include a valid bit that is examined by subsequent read-lock requests.
  • the address and command queue 120 also includes a Read Lock Fail Queue 120 d .
  • the Read Lock Fail Queue 120 d is used to hold read memory reference requests that fail because of a lock existing on a portion of memory. That is, one of the microengines issues a memory request that has a read lock request that is processed in address and control queue 120 . The memory request will operate on either the order queue 120 c or the read queue 120 b and will recognize it as a read lock request.
  • the controller 26 b will access lock lookup device 142 to determine whether this memory location is already locked. If this memory location is locked from any prior read lock request, then this memory lock request will fail and will be stored in the read lock fail queue 120 d . If it is unlocked or if 142 shows no lock on that address, then the address of that memory reference will be used by the SRAM interface 140 to perform a traditional SRAM address read/write request to memory 16 b .
  • the command controller and address generator 138 will also enter the lock into the lock look up device 142 so that subsequent read lock requests will find the memory location locked.
  • a memory location is unlocked by operation of the a microcontrol instruction in a program after the need for the lock has ended. The location is unlocked by clearing the valid bit in the CAM. After an unlock, the read lock fail queue 120 d becomes the highest priority queue giving all queued read lock misses, a chance to issue a memory lock request.
  • FIG. 5A typical timing of a static random access memory without active memory optimization and with active memory optimization is shown. As can be seen, grouping reads and writes improves cycletime eliminating dead cycles.
  • the FBUS interface 28 in a network application can performs header processing of incoming packets from the FBUS 18 .
  • a key function which the FBUS interface performs is extraction of packet headers, and a microprogrammable source/destination/protocol hashed lookup in SRAM. If the hash does not successfully resolve, the packet header is promoted to the core processor 28 for more sophisticated processing.
  • the FBI 28 contains a Transmit FIFO 182 , a Receive FIFO 183 , a HASH unit 188 and FBI control and status registers 189 . These four units communicate with the microengines 22 , via a time-multiplexed access to the SRAM bus 38 which is connected to the transfer registers 78 , 80 in the microengines. That is, all communications to and from the microengines are via the transfer registers 78 , 80 .
  • the FBUS interface 28 includes a push state machine 200 for pushing data into the transfer registers during the time cycles which the SRAM is NOT using the SRAM data bus (part of bus 38 ) and a pull state machine 202 for fetching data from the transfer registers in the respective microengine.
  • the Hashing unit includes a pair of FIFO's 188 a , 188 b .
  • the hash unit determines that the FBI 28 received an FBI_hash request.
  • the hash unit 188 fetches hash keys from the calling microengine 22 . After the keys are fetched and hashed, the indices are delivered back to the calling microengine 22 . Up to three hashes are performed under a single FBI_hash request.
  • the busses 34 and 38 are each unidirectional: SDRAM_push/pull_data, and Sbus_push/pull_data. Each of these busses require control signals which will provide read/write controls to the appropriate microengine 22 Transfer registers.
  • transfer registers require protection from the context controlling them to guarantee read correctness.
  • a write transfer register is being used by a thread_ 1 to provide data to the SDRAM 16 a , thread_ 1 must not overwrite this register until the signal back from SDRAM controller 26 a indicates that this register has been promoted and may now be re-used. Every write does not require a signal back from the destination indicating that the function has been completed, because if the thread writes to the same command queue at that destination with multiple requests, the order of the completion is guaranteed within that command queue, thus only the last command requires the signaling back to the thread.
  • any microengine 22 that uses the FBI unsolicited push technique must test the protection flag prior to accessing the FBUS interface/microengine agreed upon transfer registers. If the flag is not asserted, then the transfer registers may be accessed by the microengine. If the flag is Asserted then the context should wait N cycles prior to accessing the registers. A priori this count is determined by the number of transfer registers being pushed, plus a frontend protection window. The basic idea is that the microengine must test this flag then quickly move the data which it wishes to read from the read transfer registers to GPR's in contiguous cycles, so the push engine does not collide with the microengine read.

Abstract

A parallel hardware-based multithreaded processor is described. The processor includes a general purpose processor that coordinates system functions and a plurality of microengines that support multiple hardware threads. The processor also includes a memory control system that has a first memory controller that sorts memory references based on whether the memory references are directed to an even bank or an odd bank of memory and a second memory controller that optimizes memory references based upon whether the memory references are read references or write references.

Description

    BACKGROUND
  • This invention relates to parallel processors. [0001]
  • Parallel processing is an efficient form of information processing of concurrent events in a computing process. Parallel processing demands concurrent execution of many programs in a computer, in contrast to sequential processing. In the context of a parallel processor, parallelism involves doing more than one thing at the same time. Unlike a serial paradigm where all tasks are performed sequentially at a single station or a pipelined machine where tasks are performed at specialized stations, with parallel processing, a plurality of stations are provided with each capable of performing all tasks. That is, in general all or a plurality of the stations work simultaneously and independently on the same or common elements of a problem. Certain problems are suitable for solution by applying parallel processing. [0002]
  • Thus, types of computer processing include single instruction stream, single data stream, the conventional serial von Neumann computer in which there is a single stream of instructions. A second processing type is the single instruction stream, multiple data streams process (SIMD). This processing can have multiple arithmetic-logic processors and a single control processor. Each of the processors perform operations on the data in lock-step. These machines are synchronized by the control processor. A third type is multiple instruction streams, single data stream (MISD) process. This processing has the same data stream flows through a linear array of processors executing different instruction streams. A fourth is multiple instruction streams, multiple data streams (MIMD). This processing uses multiple processors, each executing its own instruction stream to process a data stream fed to each of the processors. Multiple instruction streams, multiple data streams (MIMD) processors may have several instruction processing units and therefore several data streams. [0003]
  • SUMMARY
  • According to an aspect of the present invention, a parallel hardware-based multithreaded processor includes a general purpose processor that coordinates system functions and a plurality of microengines that support multiple hardware threads. [0004]
  • According to an additional aspect of the present invention, a parallel hardware-based multithreaded processor includes a general purpose processor that coordinates system functions and a plurality of microengines that support multiple hardware threads. The processor also includes a memory control system including a first memory controller that sorts memory references based on whether the memory references are directed to an even bank or an odd bank of memory and a second memory controller that optimizes memory references based upon whether the memory references are read references or write references. [0005]
  • One or more of the following advantages may be provided by one or more aspects of the invention. [0006]
  • This parallel multithreaded architecture provides a very general purpose microprocessor architecture. The processor can be connected to other devices to process large amounts of data. The system is especially useful for tasks that can be broken into parallel subtasks or functions. The hardware-based multithreaded processor is useful for tasks that are bandwidth oriented rather than latency oriented. The hardware-based multithreaded processor has multiple microengines each with multiple hardware controlled threads that can be simultaneously active and independently work on a task. One example of an application for the hardware-based multithreaded processor is as a network processor. As a network processor, the hardware-based multithreaded processor can interfaces to network devices such as a media access controllers. As a network processor, the hardware-based multithreaded processor can interface to any type of communication device or interface that receives/sends large amounts of data. The processor can be used a print engine for a postscript processor or as a processor for a storage subsystem, i.e., RAID disk storage. A further use is as a matching engine. In the securities industry for example, the advent of electronic trading requires the use of electronic matching engines to match orders between buyers and sellers. These and other parallel types of tasks can be accomplished on the processor used in a system.[0007]
  • BRIEF DESCRIPTION OF THE DRAWINGS
  • FIG. 1 is a block diagram of a communication system employing a hardware-based multithreaded processor. [0008]
  • FIG. 2 is a detailed block diagram of the hardware-based multithreaded processor of FIG. 1. [0009]
  • FIG. 3 is a block diagram of a microengine functional unit employed in the hardware-based multithreaded processor of FIGS. 1 and 2. [0010]
  • FIG. 3A is a block diagram of a pipeline in the microengine of FIG. 3. [0011]
  • FIG. 3B. is diagram that shows a format for a context switch instruction. [0012]
  • FIG. 3C is a block diagram showing general purpose register address arrangement. [0013]
  • FIG. 4 is a block diagram of a memory controller for enhanced bandwidth operation used in the hardware-based multithreaded processor. [0014]
  • FIG. 4A is a flow chart that represents an arbitration policy in an SDRAM controller of FIG. 4. [0015]
  • FIG. 4B is a timing diagram that shows advantages of optimizing SDRAM controller. [0016]
  • FIG. 5 is a block diagram of a memory controller for latency limited operations used in the hardware-based multithreaded processor. [0017]
  • FIG. 5A is a timing diagram that shows advantages of optimizing SRAM controller. [0018]
  • FIG. 6 is a block diagram of a communication bus interface in the processor of FIG. 1.[0019]
  • DESCRIPTION
  • Architecture: [0020]
  • Referring to FIG. 1, a [0021] communication system 10 includes a parallel, hardware-based multithreaded processor 12. The hardware-based multithreaded processor 12 is coupled to a bus such as a PCI bus 14, a memory system 16 and a second bus 18. The system 10 is especially useful for tasks that can be broken into parallel subtasks or functions. Specifically hardware-based multithreaded processor 12 is useful for tasks that are bandwidth oriented rather than latency oriented. The hardware-based multithreaded processor 12 has multiple microengines 22 each with multiple hardware controlled threads that can be simultaneously active and independently work on a task.
  • The hardware-based [0022] multithreaded processor 12 also includes a central controller 20 that assists in loading microcode control for other resources of the hardware-based multithreaded processor 12 and performs other general purpose computer type functions such as handling protocols, exceptions, extra support for packet processing where the microengines pass the packets off for more detailed processing such as in boundary conditions. In one embodiment, the processor 20 is a Strong Arm® (Arm is a trademark of ARM Limited, United Kingdom) based architecture. The general purpose microprocessor 20 has an operating system. Through the operating system the processor 20 can call functions to operate on microengines 22 a-22 f. The processor 20 can use any supported operating system preferably a real time operating system. For the core processor implemented as a Strong Arm architecture, operating systems such as, MicrosoftNT real-time, VXWorks and μCUS, a freeware operating system available over the Internet, can be used.
  • The hardware-based [0023] multithreaded processor 12 also includes a plurality of function microengines 22 a-22 f. Functional microengines (microengines) 22 a-22 f each maintain a plurality of program counters in hardware and states associated with the program counters. Effectively, a corresponding plurality of sets of threads can be simultaneously active on each of the microengines 22 a-22 f while only one is actually operating at any one time.
  • In one embodiment, there are six [0024] microengines 22 a-22 f as shown. Each microengines 22 a-22 f has capabilities for processing four hardware threads. The six microengines 22 a-22 f operate with shared resources including memory system 16 and bus interfaces 24 and 28. The memory system 16 includes a Synchronous Dynamic Random Access Memory (SDRAM) controller 26 a and a Static Random Access Memory (SRAM) controller 26 b. SDRAM memory 16 a and SDRAM controller 26 a are typically used for processing large volumes of data, e.g., processing of network payloads from network packets. The SRAM controller 26 b and SRAM memory 16 b are used in a networking implementation for low latency, fast access tasks, e.g., accessing look-up tables, memory for the core processor 20, and so forth.
  • The six [0025] microengines 22 a-22 f access either the SDRAM 16a or SRAM 16 b based on characteristics of the data. Thus, low latency, low bandwidth data is stored in and fetched from SRAM, whereas higher bandwidth data for which latency is not as important, is stored in and fetched from SDRAM. The microengines 22 a-22 f can execute memory reference instructions to either the SDRAM controller 26 a or SRAM controller 16 b.
  • Advantages of hardware multithreading can be explained by SRAM or SDRAM memory accesses. As an example, an SRAM access requested by a Thread_[0026] 0, from a microengine will cause the SRAM controller 26 b to initiate an access to the SRAM memory 16 b. The SRAM controller controls arbitration for the SRAM bus, accesses the SRAM 16 b, fetches the data from the SRAM 16 b, and returns data to a requesting microengine 22 a-22 b. During an SRAM access, if the microengine e.g., 22 a had only a single thread that could operate, that microengine would be dormant until data was returned from the SRAM. By employing hardware context swapping within each of the microengines 22 a-22 f, the hardware context swapping enables other contexts with unique program counters to execute in that same microengine. Thus, another thread e.g., Thread_1 can function while the first thread, e.g., Thread_0, is awaiting the read data to return. During execution, Thread_1 may access the SDRAM memory 16 a. While Thread_1 operates on the SDRAM unit, and Thread_0 is operating on the SRAM unit, a new thread, e.g., Thread_2 can now operate in the microengine 22 a. Thread_2 can operate for a certain amount of time until it needs to access memory or perform some other long latency operation, such as making an access to a bus interface. Therefore, simultaneously, the processor 12 can have a bus operation, SRAM operation and. SDRAM operation all being completed or operated upon by one microengine 22 a and have one more thread available to process more work in the data path.
  • The hardware context swapping also synchronizes completion of tasks. For example, two threads could hit the same shared resource e.g., SRAM. Each one of these separate functional units, e.g., the [0027] FBUS interface 28, the SRAM controller 26 a, and the SDRAM controller 26 b, when they complete a requested task from one of the microengine thread contexts reports back a flag signaling completion of an operation. When the flag is received by the microengine, the microengine can determine which thread to turn on.
  • One example of an application for the hardware-based [0028] multithreaded processor 12 is as a network processor. As a network processor, the hardware-based multithreaded processor 12 interfaces to network devices such as a media access controller device e.g., a 10/100BaseT Octal MAC 13 a or a Gigabit Ethernet device 13 b. In general, as a network processor, the hardware-based multithreaded processor 12 can interface to any type of communication device or interface that receives/sends large amounts of data. Communication system 10 functioning in a networking application could receive a plurality of network packets from the devices 13 a, 13 b and process those packets in a parallel manner. With the hardware-based multithreaded processor 12, each network packet can be independently processed.
  • Another example for use of [0029] processor 12 is a print engine for a postscript processor or as a processor for a storage subsystem, i.e., RAID disk storage. A further use is as a matching engine. In the securities industry for example, the advent of electronic trading requires the use of electronic matching engines to match orders between buyers and sellers. These and other parallel types of tasks can be accomplished on the system 10.
  • The [0030] processor 12 includes a bus interface 28 that couples the processor to the second bus 18. Bus interface 28 in one embodiment couples the processor 12 to the so-called FBUS 18 (FIFO bus). The FBUS interface 28 is responsible for controlling and interfacing the processor 12 to the FBUS 18. The FBUS 18 is a 64-bit wide FIFO bus, used to interface to Media Access Controller (MAC) devices.
  • The [0031] processor 12 includes a second interface e.g., a PCI bus interface 24 that couples other system components that reside on the PCI 14 bus to the processor 12. The PCI bus interface 24, provides a high speed data path 24 a to memory 16 e.g., the SDRAM memory 16 a. Through that path data can be moved quickly from the SDRAM 16 a through the PCI bus 14, via direct memory access (DMA) transfers. The hardware based multithreaded processor 12 supports image transfers. The hardware based multithreaded processor 12 can employ a plurality of DMA channels so if one target of a DMA transfer is busy, another one of the DMA channels can take over the PCI bus to deliver information to another target to maintain high processor 12 efficiency. Additionally, the PCI bus interface 24 supports target and master operations. Target operations are operations where slave devices on bus 14 access SDRAMs through reads and writes that are serviced as a slave to target operation. In master operations, the processor core 20 sends data directly to or receives data directly from the PCI interface 24.
  • Each of the functional units are coupled to one or more internal buses. As described below, the internal buses are dual, 32 bit buses (i.e., one bus for read and one for write). The hardware-based [0032] multithreaded processor 12 also is constructed such that the sum of the bandwidths of the internal buses in the processor 12 exceed the bandwidth of external buses coupled to the processor 12. The processor 12 includes an internal core processor bus 32, e.g., an ASB bus (Advanced System Bus) that couples the processor core 20 to the memory controller 26 a, 26 c and to an ASB translator 30 described below. The ASB bus is a subset of the so called AMBA bus that is used with the Strong Arm processor core. The processor 12 also includes a private bus 34 that couples the microengine units to SRAM controller 26 b, ASB translator 30 and FBUS interface 28. A memory bus 38 couples the memory controller 26 a, 26 b to the bus interfaces 24 and 28 and memory system 16 including flashrom 16 c used for boot operations and so forth.
  • Referring to FIG. 2, each of the [0033] microengines 22 a-22 f includes an arbiter that examines flags to determine the available threads to be operated upon. Any thread from any of the microengines 22 a-22 f can access the SDRAM controller 26 a, SDRAM controller 26 b or FBUS interface 28. The memory controllers 26 a and 26 b each include a plurality of queues to store outstanding memory reference requests. The queues either maintain order of memory references or arrange memory references to optimize memory bandwidth. For example, if a thread_0 has no dependencies or relationship to a thread_1, there is no reason that thread 1 and 0 cannot complete their memory references to the SRAM unit out of order. The microengines 22 a-22 f issue memory reference requests to the memory controllers 26 a and 26 b. The microengines 22 a-22 f flood the memory subsystems 26 a and 26 b with enough memory reference operations such that the memory subsystems 26 a and 26 b become the bottleneck for processor 12 operation.
  • If the [0034] memory subsystem 16 is flooded with memory requests that are independent in nature, the processor 12 can perform memory reference sorting. Memory reference sorting improves achievable memory bandwidth. Memory reference sorting, as described below, reduces dead time or a bubble that occurs with accesses to SRAM. With memory references to SRAM, switching current direction on signal lines between reads and writes produces a bubble or a dead time waiting for current to settle on conductors coupling the SRAM 16 b to the SRAM controller 26 b.
  • That is, the drivers that drive current on the bus need to settle out prior to changing states. Thus, repetitive cycles of a read followed by a write can degrade peak bandwidth. Memory reference sorting allows the [0035] processor 12 to organize references to memory such that long strings of reads can be followed by long strings of writes. This can be used to minimize dead time in the pipeline to effectively achieve closer to maximum available bandwidth. Reference sorting helps maintain parallel hardware context threads. On the SDRAM, reference sorting allows hiding of pre-charges from one bank to another bank. Specifically, if the memory system 16 b is organized into an odd bank and an even bank, while the processor is operating on the odd bank, the memory controller can start precharging the even bank. Precharging is possible if memory references alternate between odd and even banks. By ordering memory references to alternate accesses to opposite banks, the processor 12 improves SDRAM bandwidth. Additionally, other optimizations can be used. For example, merging optimizations where operations that can be merged, are merged prior to memory access, open page optimizations where by examining addresses an opened page of memory is not reopened, chaining, as will be described below, and refreshing mechanisms, can be employed.
  • The [0036] FBUS interface 28 supports Transmit and Receive flags for each port that a MAC device supports, along with an Interrupt flag indicating when service is warranted. The FBUS interface 28 also includes a controller 28 a that performs header processing of incoming packets from the FBUS 18. The controller 28 a extracts the packet headers and performs a microprogrammable source/destination/protocol hashed lookup (used for address smoothing) in SRAM. If the hash does not successfully resolve, the packet header is sent to the processor core 20 for additional processing. The FBUS interface 28 supports the following internal data transactions:
    FBUS unit (Shared bus SRAM) to/from microengine.
    FBUS unit (via private bus) writes from SDRAM Unit.
    FBUS unit (via Mbus) Reads to SDRAM.
  • The [0037] FBUS 18 is a standard industry bus and includes a data bus, e.g., 64 bits wide and sideband control for address and read/write control. The FBUS interface 28 provides the ability to input large amounts of data using a series of input and output FIFO's 29 a-29 b. From the FIFOs 29 a-29 b, the microengines 22 a-22 f fetch data from or command the SDRAM controller 26 a to move data from a receive FIFO in which data has come from a device on bus 18, into the FBUS interface 28. The data can be sent through memory controller 26 a to SDRAM memory 16 a, via a direct memory access. Similarly, the microengines can move data from the SDRAM 26 a to interface 28, out to FBUS 18, via the FBUS interface 28.
  • Data functions are distributed amongst the microengines. Connectivity to the [0038] SRAM 26 a, SDRAM 26 b and FBUS 28 is via command requests. A command request can be a memory request or a FBUS request. For example, a command request can move data from a register located in a microengine 22 a to a shared resource, e.g., an SDRAM location, SRAM location, flash memory or some MAC address. The commands are sent out to each of the functional units and the shared resources. However, the shared resources do not need to maintain local buffering of the data. Rather, the shared resources access distributed data located inside of the microengines. This enables microengines 22 a-22 f, to have local access to data rather than arbitrating for access on a bus and risk contention for the bus. With this feature, there is a 0 cycle stall for waiting for data internal to the microengines 22 a-22 f.
  • The data buses, e.g., [0039] ASB bus 30, SRAM bus 34 and SDRAM bus 38 coupling these shared resources, e.g., memory controllers 26 a and 26 b are of sufficient bandwidth such that there are no internal bottlenecks. Thus, in order to avoid bottlenecks, the processor 12 has an bandwidth requirement where each of the functional units is provided with at least twice the maximum bandwidth of the internal buses. As an example, the SDRAM can run a 64 bit wide bus at 83 MHz. The SRAM data bus could have separate read and write buses, e.g., could be a read bus of 32 bits wide running at 166 MHz and a write bus of 32 bits wide at 166 MHz. That is, in essence, 64 bits running at 166 MHz which is effectively twice the bandwidth of the SDRAM.
  • The [0040] core processor 20 also can access the shared resources. The core processor 20 has a direct communication to the SDRAM controller 26 a to the bus interface 24 and to SRAM controller 26 b via bus 32. However, to access the microengines 22 a-22 f and transfer registers located at any of the microengines 22 a-22 f, the core processor 20 access the microengines 22 a-22 f via the ASB Translator 30 over bus 34. The ASB translator 30 can physically reside in the FBUS interface 28, but logically is distinct. The ASB Translator 30 performs an address translation between FBUS microengine transfer register locations and core processor addresses (i.e., ASB bus) so that the core processor 20 can access registers belonging to the microengines 22 a-22 c.
  • Although [0041] microengines 22 can use the register set to exchange data as described below, a scratchpad memory 27 is also provided to permit microengines to write data out to the memory for other microengines to read. The scratchpad 27 is coupled to bus 34.
  • The [0042] processor core 20 includes a RISC core 50 implemented in a five stage pipeline performing a single cycle shift of one operand or two operands in a single cycle, provides multiplication support and 32 bit barrel shift support. This RISC core 50 is a standard Strong Arm® architecture but it is implemented with a five stage pipeline for performance reasons. The processor core 20 also includes a 16 kilobyte instruction cache 52, an 8 kilobyte data cache 54 and a prefetch stream buffer 56. The core processor 20 performs arithmetic operations in parallel with memory writes and instruction fetches. The core processor 20 interfaces with other functional units via the ARM defined ASB bus. The ASB bus is a 32-bit bi-directional bus 32.
  • Microengines: [0043]
  • Referring to FIG. 3, an exemplary one of the [0044] microengines 22 a-22 f, e.g., microengine 22 f is shown. The microengine includes a control store 70 which, in one implementation, includes a RAM of here 1,024 words of 32 bit. The RAM stores a microprogram. The microprogram is loadable by the core processor 20. The microengine 22 f also includes controller logic 72. The controller logic includes an instruction decoder 73 and program counter (PC) units 72 a-72 d. The four micro program counters 72 a-72 d are maintained in hardware. The microengine 22 f also includes context event switching logic 74. Context event logic 74 receives messages (e.g., SEQ_#_EVENT_RESPONSE; FBI_EVENT_RESPONSE; SAM_EVENT_RESPONSE; SDRAM_EVENT_RESPONSE; and ASB_EVENT_RESPONSE) from each one of the shared resources, e.g., SRAM 26 a, SDRAM 26 b, or processor core 20, control and status registers, and so forth. These messages provide information on whether a requested function has completed. Based on whether or not a function requested by a thread has completed and signaled completion, the thread needs to wait for that completion signal, and if the thread is enabled to operate, then the thread is placed on an available thread list (not shown). The microengine 22 f can have a maximum of e.g., 4 threads available.
  • In addition to event signals that are local to an executing thread, the [0045] microengines 22 employ signaling states that are global. With signaling states, an executing thread can broadcast a signal state to all microengines 22. Receive Request Available signal, Any and all threads in the microengines can branch on these signaling states. These signaling states can be used to determine availability of a resource or whether a resource is due for servicing.
  • The [0046] context event logic 74 has arbitration for the four (4) threads. In one embodiment, the arbitration is a round robin mechanism. Other techniques could be used including priority queuing or weighted fair queuing. The microengine 22 f also includes an execution box (EBOX) data path 76 that includes an arithmetic logic unit 76 a and general purpose register set 76 b. The arithmetic logic unit 76 a performs arithmetic and logical functions as well as shift functions. The registers set 76 b has a relatively large number of general purpose registers. As will be described in FIG. 3B, in this implementation there are 64 general purpose registers in a first bank, Bank A and 64 in a second bank, Bank B. The general purpose registers are windowed as will be described so that they are relatively and absolutely addressable.
  • The [0047] microengine 22 f also includes a write transfer register stack 78 and a read transfer stack 80. These registers are also windowed so that they are relatively and absolutely addressable. Write transfer register stack 78 is where write data to a resource is located. Similarly, read register stack 80 is for return data from a shared resource. Subsequent to or concurrent with data arrival, an event signal from the respective shared resource e.g., the SRAM controller 26 a, SDRAM controller 26 b or core processor 20 will be provided to context event arbiter 74 which will then alert the thread that the data is available or has been sent. Both transfer register banks 78 and 80 are connected to the execution box (EBOX) 76 through a data path. In one implementation, the read transfer register has 64 registers and the write transfer register has 64 registers.
  • As shown in FIG. 3A, the microengine datapath maintains a 5-stage micro-pipeline [0048] 82. This pipeline includes lookup of microinstruction words 82 a, formation of the register file addresses 82 b, read of operands from register file 82 c, ALU, shift or compare operations 82 d, and write-back of results to registers 82 e. By providing a write-back data bypass into the ALU/shifter units, and by assuming the registers are implemented as a register file (rather than a RAM), the microengine can perform a simultaneous register file read and write, which completely hides the write operation.
  • The [0049] SDRAM interface 26 a provides a signal back to the requesting microengine on reads that indicates whether a parity error occurred on the read request. The microengine microcode is responsible for checking the SDRAM read Parity flag when the microengine uses any return data. Upon checking the flag, if it was set, the act of branching on it clears it. The Parity flag is only sent when the SDRAM is enabled for checking, and the SDRAM is parity protected. The microengines and the PCI Unit are the only requesters notified of parity errors. Therefore, if the processor core 20 or FIFO requires parity protection, a microengine assists in the request. The microengines 22 a-22 f support conditional branches. The worst case conditional branch latency (not including jumps) occurs when the branch decision is a result of condition codes being set by the previous microcontrol instruction. The latency is shown below in Table 1:
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n2 XX b1 b2 b3 b4
    reg addr gen n1 cb XX XX b1 b2 b3
    reg file lookup n1 cb XX XX b1 b2
    ALU/shifter/cc n1 cb XX XX b1
    write back m2 n1 cb XX XX
  • where [0050]
  • nx is pre-branch microword (n[0051] 1 sets cc's)
  • cb is conditional branch [0052]
  • bx is post-branch microword [0053]
  • XX is aborted microword [0054]
  • As shown in Table 1, it is not until [0055] cycle 4 that the condition codes of n1 are set, and the branch decision can be made (which in this case causes the branch path to be looked up in cycle 5). The microengine incurs a 2-cycle branch latency penalty because it must abort operations n2 and n3 (the 2 microwords directly after the branch) in the pipe, before the branch path begins to fill the pipe with operation b1. If the branch is not taken, no microwords are aborted and execution continues normally. The microengines have several mechanisms to reduce or eliminate the effective branch latency.
  • The microengines support deferred branches. Deferring branches are when a microengine allows 1 or 2 microwords after the branch to occur before the branch takes effect (i.e. the effect of the branch is “deferred” in time). Thus, if useful work can be found to fill the wasted cycles after the branch microword, then the branch latency can be hidden. A 1-cycle deferred branch is shown below where n[0056] 2 is allowed to execute after cb, but before b1:
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n2 XX b1 b2 b3 b4
    reg addr gen n1 cb n2 XX b1 b2 b3
    reg file lookup n1 cb n2 XX b1 b2
    ALU/shifter/cc n1 cb n2 XX b1
    write back n1 cb n2 XX
  • A 2-cycle deferred branch is shown below, where n[0057] 2 and n3 are both allowed to complete before the branch to b1 occurs. Note that a 2-cycle branch deferment is only allowed when the condition codes are set on the microword preceding the branch.
    1 2 3 4 5 6 7 8 9
    microstore lookup n1 cb n2 n3 b1 b2 b3 b4 b5
    reg addr gen n1 cb n2 n3 b1 b2 b3 b4
    reg file lkup n1 cb n2 n3 b1 b2 b3
    ALU/shftr/cc n1 cb n2 n3 b1 b2
    write back n1 cb n2 n3 b1
  • The microengines also support condition code evaluation. If the condition codes upon which a branch decision are made are set 2 or more microwords before the branch, then 1 cycle of branch latency can be eliminated because the branch decision can be made 1 cycle earlier: [0058]
    1 2 3 4 5 6 7 8
    microstore lookup n1 n2 cb XX b1 b2 b3 b4
    reg addr gen n1 n2 cb XX b1 b2 b3
    reg file lookup n1 n2 cb XX b1 b2
    ALU/shifter/cc n1 n2 cb XX b1
    write back n1 n2 cb XX
  • In this example, n[0059] 1 sets the condition codes and n2 does not set the conditions codes. Therefore, the branch decision can be made at cycle 4 (rather than 5), to eliminate 1 cycle of branch latency. In the example below, the 1-cycle branch deferment and early setting of condition codes are combined to completely hide the branch latency:
  • Condition codes (cc's) set 2 cycles before a 1-cycle deferred branch: [0060]
    1 2 3 4 5 6 7 8
    microstore lookup n1 n2 cb n3 b1 b2 b3 b4
    reg addr gen n1 n2 cb n3 b1 b2 b3
    reg file lookup n1 n2 cb n3 b1 b2
    ALU/shifter/cc n1 n2 cb n3 b1
    write back n1 n2 cb n3
  • In the case where the condition codes cannot be set early (i.e. they are set in the microword preceding the branch), the microengine supports branch guessing which attempts to reduce the 1 cycle of exposed branch latency that remains. By “guessing” the branch path or the sequential path, the microsequencer pre-fetches the guessed [0061] path 1 cycle before it definitely knows what path to execute. If it guessed correctly, 1 cycle of branch latency is eliminated as shown below:
  • guess branch taken/branch is taken [0062]
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n1 b1 b2 b3 b4 b5
    reg addr gen n1 cb XX b1 b2 b3 b4
    reg file lookup n1 cb XX b1 b2 b3
    ALU/shifter/cc n1 cb XX b1 b2
    write back n1 cb XX b1
  • If the microcode guessed a branch taken incorrectly, the microengine still only wastes 1 cycle: [0063]
  • guess branch taken/branch is NOT taken [0064]
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n1 XX n2 n3 n4 n5
    reg addr gen n1 cb n1 XX n2 n3 n4
    reg file lookup n1 cb n1 XX n2 n3
    ALU/shifter/cc n1 cb n1 XX n2
    write back n1 cb n1 XX
  • However, the latency penalty is distributed differently when microcode guesses a branch is not taken: [0065]
  • For guess branch NOT taken/branch is NOT taken there are no wasted cycles as set out below. [0066]
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n1 n2 n3 n4 n5 n6
    reg addr gen n1 cb n1 n2 n3 n4 n5
    reg file lookup n1 cb n1 n2 n1 b4
    ALU/shifter/cc n1 cb n1 n2 n3
    write back n1 cb n1 n2
  • However for guess branch NOT taken/branch is taken there are 2 wasted cycles. [0067]
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n1 XX b1 b2 b3 b4
    reg addr gen n1 cb XX XX b1 b2 b3
    reg file lookup n1 cb XX XX b1 b2
    ALU/shifter/cc n1 cb XX XX b1
    write back n1 cb XX XX
  • The microengine can combine branch guessing with 1-cycle branch deferment to improve the result further. For guess branch taken with 1-cycle deferred branch/branch is taken is: [0068]
    1 2 3 4 5 6 7 8
    microstore lookup n1 cb n2 b1 b2 b3 b4 b5
    reg addr gen n1 cb n2 b1 b2 b3 b4
    reg file lookup n1 cb n2 b1 b2 b3
    ALU/shifter/cc n1 cb n2 b1 b2
    write back n1 cb n2 b1
  • In the case above, the 2cycles of branch latency are hidden by the execution of n[0069] 2, and by correctly guessing the branch direction. If microcode guesses incorrectly, 1 cycle of branch latency remains exposed as shown below:
  • guess branch taken with 1-cycle deferred branch/branch is NOT taken [0070]
    1 2 3 4 5 6 7 8 9
    microstore lookup n1 cb n2 XX n3 n4 n5 n6 n7
    reg addr gen n1 cb n2 XX n3 n4 n5 n6
    reg file lkup n1 cb n2 XX n3 n4 n5
    ALU/shftr/cc n1 cb n2 XX n3 n4
    write back n1 cb n2 XX n3
  • If microcode correctly guesses a branch NOT taken, then the pipeline flows sequentially in the normal unperturbed case. If microcode incorrectly guesses branch NOT taken, the microengine again exposes 1 cycle of unproductive execution as shown below: [0071]
  • guess branch NOT taken/branch is taken [0072]
    1 2 3 4 5 6 7 8 9
    microstore lookup n1 cb n2 XX b1 b2 b3 b4 b5
    reg addr gen n1 cb n2 XX b1 b2 b3 b4
    reg file lkup n1 cb n2 XX b1 b2 b3
    ALU/shftr/cc n1 cb n2 XX b1 b2
    write back n1 cb n2 XX b1
  • where [0073]
  • nx is pre-branch microword (n[0074] 1 sets cc's)
  • cb is conditional branch [0075]
  • bx is post-branch microword [0076]
  • XX is aborted microword [0077]
  • In the case of a jump instruction, 3 extra cycles of latency are incurred because the branch address is not known until the end of the cycle in which the jump is in the ALU stage: [0078]
    1 2 3 4 5 6 7 8 9
    microstore n1 jp XX XX XX j1 j2 j3 j4
    lookup
    reg addr gen n1 jp XX XX XX j1 j2 j3
    reg file lkup n1 jp XX XX XX j1 j2
    ALU/shftr/cc n1 jp XX XX XX j1
    write back n1 jp XX XX XX
  • Context Switch: [0079]
  • Referring to FIG. 3B, a format from a context switch instruction is shown. A context switch is a special form of a branch that causes a different context (and associated PC) to be selected. Context switching introduces some branch latency as well. Consider the following context switch: [0080]
    1 2 3 4 5 6 7 8 9
    microstore lookup o1 ca br n1 n2 n3 n4 n5 n6
    reg addr gen o1 ca XX n1 n2 n3 n4 n5
    reg file lkup o1 ca XX n1 n2 n3 n4
    ALU/shftr/cc o1 ca XX n1 n2 n3
    write back o1 ca XX n1 n2
  • where [0081]
  • ox is old context flow [0082]
  • br is branch microword in old context [0083]
  • ca is context rearbitration (causes context switch) [0084]
  • nx is new context flow [0085]
  • XX is aborted microword [0086]
  • In a context switch the “br” microword is aborted to avoid control and timing complexities that could be caused by saving the correct old context PC. [0087]
  • Conditional branches that operate on ALU condition codes which are set on the microword before the branch can select 0, 1 or 2-cycle branch deferment modes. Condition codes set 2 or more microwords before the conditional branch that operates on them can select 0 or 1-cycle branch deferment modes. All other branches (including context rearbitrations) can select either 0 or 1-cycle branch deferment modes. The architecture could be designed to make a context arbitration microword within a branch deferment window of a preceding branch, jump or context arbitration microword, an illegal option. That is, in some embodiments, a context switch would not be allowed to occur during a branch transition in the pipeline because as mentioned, it could unduly complicate saving of the old context PC. The architecture could also be designed to make branching within the branch deferment window of a preceding branch, jump or context arbitration microword illegal to avoid complicated and possible unpredictable branch behaviors. [0088]
  • Each [0089] microengine 22 a-22 f supports multi-threaded execution of four contexts. One reason for this is to allow one thread to start executing just after another thread issues a memory reference and must wait until that reference completes before doing more work. This behavior is critical to maintaining efficient hardware execution of the microengines because memory latency is significant. Stated differently, if only a single thread execution was supported, the microengines would sit idle for a significant number of cycles waiting for references to return and thereby reduce overall computational throughput. Multi-threaded execution allows an microengines to hide memory latency by performing useful independent work across several threads. Two synchronization mechanisms are supplied in order to allow a thread to issue an SRAM or SDRAM reference, and then subsequently synchronize to the point in time when that reference completes.
  • One mechanism is Immediate Synchronization. In immediate synchronization, the microengine issues the reference and immediately swap out that context. The context will be signaled when the corresponding reference completes. Once signaled, the context will be swapped back in for execution when a context-swap event occurs and it is its turn to run. Thus, from the point of view of a single context's instruction stream, the microword after issuing the mem reference does not get executed until the reference completes. [0090]
  • A second mechanism is Delayed Synchronization. In delayed synchronization, the microengine issues the reference, and then continues to execute some other useful work independent of the reference. Some time later it could become necessary to synchronize the thread's execution stream to the completion of the issued reference before further work is performed. At this point a synchronizing microword is executed that will either swap out the current thread, and swap it back in sometime later when the reference has completed, or continue executing the current thread because the reference has already completed. Delayed synchronization is implemented using two different signaling schemes: [0091]
  • If the memory reference is associated with a transfer register, the signal from which the thread is triggered is generated when the corresponding transfer register valid bit is set or cleared. For example, an SRAM read which deposits data into transfer register A would be signaled when the valid bit for A is set. If the memory reference is associated with the transfer FIFO or the receive FIFO, instead of a transfer register, then the signal is generated when the reference completes in the [0092] SDRAM controller 26 a. Only one signal state per context is held in the microengines scheduler, thus only one outstanding signal can exist in this scheme.
  • There are at least two general operational paradigms from which microcontroller micro-programs could be designed. One would be that overall microcontroller compute throughput and overall memory bandwidth are optimized at the expense of single thread execution latency. This paradigm would make sense when the system has multiple microengines executing multiple threads per microengine on unrelated data packets. [0093]
  • A second one is that microengine execution latency should be optimized at the expense of overall microengine compute throughput and overall memory bandwidth. This paradigm could involve execution of a thread with a real-time constraint, that is, a constraint which dictates that some work must absolutely be done by some specified time. Such a constraint requires that optimization of the single thread execution be given priority over other considerations such as memory bandwidth or overall computational throughput. A real-time thread would imply a single microengine that executes only one thread. Multiple threads would not be handled because the goal is to allow the single real-time thread to execute as soon as possible—execution of multiple threads would hinder this ability. [0094]
  • The coding style of these two paradigms could be significantly different with regard to issuing memory references and context switching. In the real time case, the goal is to issue as many memory references as soon as possible in order to minimize the memory latency incurred by those references. Having issued as many references as early as possible the goal would be to perform as many computations as the microengines as possible in parallel with the references. A computation flow that corresponds to real-time optimization is: [0095]
  • [0096] issue mem ref 1
  • [0097] issue mem ref 2
  • [0098] issue mem ref 3
  • perform work independent of [0099] mem refs 1, 2 and 3
  • synch to completion of [0100] mem ref 1
  • perform work dependent on [0101] mem ref 1 and independent of mem ref 2 and 3
  • issue any new mem refs based on preceding work. [0102]
  • synch to completion of [0103] mem ref 2
  • perform work dependent on [0104] mem ref 1 and 2 independent of mem ref 3
  • issue any new mem refs based on preceding work. [0105]
  • synch to completion of [0106] mem ref 3
  • perform work dependent on the completion of all [0107] 3 refs
  • issue any new mem refs based on preceding work. [0108]
  • In contrast, optimization for throughput and bandwidth would take a different approach. With optimization for microengine computational throughput and overall memory bandwidth less consideration is given to single thread execution latency. To accomplish this, the goal would be to equally space memory references across the microprogram for each thread. This would provide a uniform stream of memory references to the SRAM and SDRAM controllers and would maximize the probability that 1 thread is always available to hide the memory latency incurred when another thread is swapped out. [0109]
  • Register File Address Types: [0110]
  • Referring to FIG. 3C, the two register address spaces that exist are Locally accessibly registers, and Globally accessible registers accessible by all microengines. The General Purpose Registers (GPRs) are implemented as two separate banks (A bank and B bank) whose addresses are interleaved on a word-by-word basis such that A bank registers have lsb=0, and B bank registers have lsb=1. Each bank is capable of performing a simultaneous read and write to two different words within its bank. [0111]
  • Across banks A and B, the register set [0112] 76 b is also organized into four windows 76 b 0-76 b 3 of 32 registers that are relatively addressable per thread. Thus, thread_0 will find its register 0 at 77 a (register 0), the thread_1 will find its register_0 at 77 b (register 32), thread_2 will find its register_0 at 77 c (register 64), and thread_3 at 77 d (register 96). Relative addressing is supported so that multiple threads can use the exact same control store and locations but access different windows of register and perform different functions. The uses of register window addressing and bank addressing provide the requisite read bandwidth using only dual ported RAMS in the microengine 22 f.
  • These windowed registers do not have to save data from context switch to context switch so that the normal push and pop of a context swap file or stack is eliminated. Context switching here has a 0 cycle overhead for changing from one context to another. Relative register addressing divides the register banks into windows across the address width of the general purpose register set. Relative addressing allows access any of the windows relative to the starting point of the window. Absolute addressing is also supported in this architecture where any one of the absolute registers may be accessed by any of the threads by providing the exact address of the register. [0113]
  • Addressing of general purpose registers [0114] 78 can occur in 2 modes depending on the microword format. The two modes are absolute and relative. In absolute mode, addressing of a register address is directly specified in 7-bit source field (a6-a0 or b6-b0):
    7 6 5 4 3 2 1 0
    A GPR: a6 0 a5 a4 a3 a2 a1 a0 a6 = 0
    B GPR: b6 1 b5 b4 b3 b2 b1 b0 b6 = 0
    SRAM/ASB: a6 a5 a4 0 a3 a2 a1 a0 a6 = 1, a5 = 0, a4 = 0
  • SDRAM: |a[0115] 6|a5|a4|0|a3|a2|a1|a0| a6=1, a5=0, a4=1
  • register address directly specified in 8-bit dest field (d[0116] 7-d0):
    7 6 5 4 3 2 1 0
    A GPR: d7 d6 d5 d4 d3 d2 d1 d0 d7 = 0, d6 = 0
    B GPR: d7 d6 d5 d4 d3 d2 d1 d0 d7 = 0, d6 = 1
    SRAM/ASB: d7 d6 d5 d4 d3 d2 d1 d0 d7 = 1, d6 = 0, d5 = 0
    SDRAM: d7 d6 d5 d4 d3 d2 d1 d0 d7 = 1, d6 = 0, d5 = 1
  • If <a[0117] 6:a5>=1,1, <b6:b5>=1,1, or <d7:d6>=1,1 then the lower bits are interpreted as a context-relative address field (described below). When a non-relative A or B source address is specified in the A, B absolute field, only the lower half of the SRAM/ASB and SDRAM address spaces can be addressed. Effectively, reading absolute SRAM/SDRAM devices has the effective address space; however, since this restriction does not apply to the dest field, writing the SRAM/SDRAM still uses the full address space.
  • In relative mode, addresses a specified address is offset within context space as defined by a 5-bit source field (a[0118] 4-a0 or b4-b0):
    7 6 5 4 3 2 1 0
    A GPR: a4 0 context a3 a2 a1 a0 a4 = 0
    B GPR: b4 1 context b3 b2 b1 b0 b4 = 0
    SRAN/: ab4 0 ab3 context b2 b1 ab0 ab4 = 1, ab3 = 0
    ASB
    SDRAM: ab4 0 ab3 context b2 b1 ab0 ab4 = 1, ab3 = 1
  • or as defined by the 6-bit dest field (d[0119] 5-d0):
    7 6 5 4 3 2 1 0
    A GPR: d5 d4 context d3 d2 d1 d0 d5 = 0, d4 = 0
    B GPR: d5 d4 context d3 d2 d1 d0 d5 = 0, d4 = 1
    SRAN/ASB: d5 d4 d3 context d2 d1 d0 d5 = 1, d4 = 0, d3 = 0
    SDRAN: d5 d4 d3 context d2 d1 d0 d5 = 1, d4 = 0, d3 = 1
  • If <d[0120] 5:d4>=1,1, then the destination address does not address a valid register, thus, no dest operand is written back.
  • The following registers are globally accessible from the microengines and the memory controllers: [0121]
  • hash unit registers [0122]
  • scratchpad and common registers [0123]
  • receive FIFO and receive status FIFO [0124]
  • transmit FIFO [0125]
  • transmit control FIFO [0126]
  • The microengines are not interrupt driven. Each microflow executes until completion and then a new flow is chosen based on state signaled by other devices in the [0127] processor 12.
  • Referring to FIG. 4, the [0128] SDRAM memory controller 26 a includes memory reference queues 90 where memory reference requests arrive from the various microengines 22 a-22 f. The memory controller 26 a includes an arbiter 91 that selects the next the microengine reference requests to go to any of the functioning units. Given that one of the microengines is providing a reference request, the reference request will come through the address and command queue 90, inside the SDRAM controller 26 a. If the reference request has a bit set called the “optimized MEM bit” the incoming reference request will be sorted into either the even bank queue 90 a or the odd bank queue 90 b. If the memory reference request does not have a memory optimization bit set, the default will be to go into an order queue 90 c. The SDRAM controller 26 is a resource which is shared among the FBUS interface 28, the core processor 20 and the PCI interface 24. The SDRAM controller 26 also maintains a state machine for performing READ-MODIFY-Write atomic operations. The SDRAM controller 26 also performs byte alignment for requests of data from SDRAM.
  • The [0129] order queue 90 c maintains the order of reference requests from the microengines. With a series of odd and even banks references it may be required that a signal is returned only upon completion of a sequence of memory references to both the odd and even banks. If the microengine 22 f sorts the memory references into odd bank and even bank references and one of the banks, e.g., the even bank is drained of memory references before the odd bank but the signal is asserted on the last even reference, the memory controller 26 a could conceivably signal back to a microengine that the memory request had completed, even though the odd bank reference had not been serviced. This occurrence could cause a coherency problem. The situation is avoided by providing the order queue 90 c allowing a microengine to have multiple memory references outstanding of which only its last memory reference needs to signal a completion.
  • The [0130] SDRAM controller 26 a also includes a high priority queue 90 d. In the high priority queue 90 d, an incoming memory reference from one of the microengines goes directly to the high priority queue and is operated upon at a higher priority than other memory references in the other queues. All of these queues, the even bank queue 90 a, the odd bank queue 90 b, the order queue 90 c and the high priority queue, are implemented in a single RAM structure that is logically segmented into four different windows, each window having its own head and tail pointer. Since filling and draining operations are only a single input and a single output, they can be placed into the same RAM structure to increase density of RAM structures.
  • The [0131] SDRAM controller 26 a also includes core bus interface logic i.e., ASB bus 92. The ASB bus interface logic 92 interfaces the core processor 20 to the SDRAM controller 26 a. The ASB bus is a bus that includes a 32 bit data path and a 28 bit address path. The data is accessed to and from memory through MEM ASB data device 98, e.g., a buffer. MEM ASB data device 98 is a queue for write data. If there is incoming data from the core processor 20 via ASB interface 92, the data can be stored into the MEM ASB device 98 and subsequently removed from MEM ASB device 98 through the SDRAM interface 110 to SDRAM memory 16 a. Although not shown, the same queue structure can be provided for the reads. The SDRAM controller 26 a also includes an engine 97 to pull data from the microengines and PCI bus.
  • Additional queues include the [0132] PCI address queue 94 and ASB read/write queue 96 that maintain a number of requests. The memory requests are sent to SDRAM interface 110 via multiplexer 106. The multiplexer 106 is controlled by the SDRAM arbiter 91 which detects the fullness of each of the queues and the status of the requests and from that decides priority based on a programmable value stored in a priority service control register 100.
  • Once control to the [0133] multiplexer 106 selects a memory reference request, the memory reference request, is sent to a decoder 108 where it is decoded and an address is generated. The decoded address is sent to the SDRAM interface 110 where it is decomposed into row and column address strobes to access the SDRAM 16 a and write or read data over data lines 16 a sending data to bus 112. In one implementation, bus 112 is actually two separate buses instead of a single bus. The separate buses would include a read bus coupling the distributed microengines 22 a-22 f and a write bus coupling the distributed microengines 22 a-22 f.
  • A feature of the [0134] SDRAM controller 26 a is that when a memory reference is stored in the queues 90, in addition to the optimized MEM bit that can be set, there is a “chaining bit”. The chaining bit when set allows for special handling of contiguous memory references. As previously mentioned, the arbiter 12 controls which microengine will be selected to provide memory reference requests over the commander bus to queue 90 (FIG. 4). Assertion of the chain bit will control the arbiter to have the arbiter select the functional unit which previously requested that bus because setting of the chain bit indicates that the microengine issued a chain request.
  • Contiguous memory references will be received in [0135] queue 90 when the chaining bit is set. Those contiguous references will typically be stored in the order queue 90 c because the contiguous memory references are multiple memory references from a single thread. In order to provide synchronization, the memory controller 26 a need only signal at the end of the chained memory references when done. However, in an optimized memory chaining, (e..g, when optimized MEM bit and chaining bit are set) the memory references could go into different banks and potentially complete on one of the banks issuing the signal “done” before the other bank was fully drained, thus destroying coherency. Therefore, the chain bit is used by the controller 110 to maintain the memory references from the current queue.
  • Referring to FIG. 4A, a flow representation of the arbitration policy in the [0136] SDRAM controller 26 a is shown. The arbitration policy favors chained microengine memory requests. The process 115 starts by examining for Chained microengine memory reference requests 115 a. The process 115 stays at the chained requests until the chain bit is cleared. The process examines ASB bus requests 115 b followed by PCI bus requests 115 c, High Priority Queue Service 115 d, Opposite Bank Requests 115 e, Order Queue Requests 115 f, and Same Bank Requests 115 g. Chained request are serviced completely, whereas services 115 b-115d are serviced in round robin order. Only when services 115 a-115 d are fully drained does the process handle services 115 e-115g. Chained microengine memory reference requests are when the previous SDRAM memory request has the chain bit set. When the chain bit is set then the arbitration engine simply services the same queue again, until the chain bit is cleared. The ASB is higher priority than PCI due to the severe performance penalty imposed on the Strong arm core when the ASB is in wait state. PCI has higher priority than the microengines due to the latency requirements of PCI. However with other buses, the arbitration priority could be different.
  • As shown in FIG. 4B, typical timing of a memory without active memory optimization and with active memory optimization is shown. As can be seen, the use of active memory optimizations maximizes the use of the bus and thus hides the inherent latency within physical SDRAM devices. In this example, a non-optimized access can take 14 cycles while optimized access can take 7 cyles. [0137]
  • Referring to FIG. 5, the [0138] memory controller 26 b for the SRAM is shown. The memory controller 26 b includes an address and command queue 120. While the memory controller 26 a (FIG. 4) has a queue for memory optimization based on odd and even banking, memory controller 26 b is optimized based on the type of memory operation, i.e., a read or a write. The address and command queue 120 includes a high priority queue 120 a, a read queue 120 b which is the predominant memory reference function that an SRAM performs, and an order queue 120 c which in general will include all writes to SRAM and reads that are to be non-optimized. Although not shown, the address and command queue 120 could also include a write queue.
  • The [0139] SRAM controller 26 b also includes core bus interface logic i.e., ASB bus 122. The ASB bus interface logic 122 interfaces the core processor 20 to the SRAM controller 26 b. The ASB bus is a bus that includes a 32 bit data path and a 28 bit address path. The data is accessed to and from memory through MEM ASB data device 128, e.g., a buffer. MEM ASB data device 128 is a queue for write data. If there is incoming data from the core processor 20 via ASB interface 122, the data can be stored into the MEM ASB device 128 and subsequently removed from MEM ASB device 128 through SRAM interface 140 to SRAM memory 16 b. Although not shown, the same queue structure can be provided for reads. The SRAM controller 26 b also includes an engine 127 to pull data from the microengines and PCI bus.
  • The memory requests are sent to [0140] SRAM interface 140 via multiplexer 126. The multiplexer 126 is controlled by the SRAM arbiter 131 which detects the fullness of each of the queues and the status of the requests and from that decides priority based on a programmable value stored in a priority service control register 130. Once control to the multiplexer 126 selects a memory reference request, the memory reference request, is sent to a decoder 138 where it is decoded and an address is generated.
  • The SRAM Unit maintains control of the Memory Mapped off-chip SRAM and Expansion ROM. The [0141] SRAM controller 26 b can address, e.g., 16 MBytes, with, e.g., 8 MBytes mapped for SRAM 16b, and 8 MBytes reserved for special functions including: Boot space via flashrom 16 c; and Console port access for MAC devices 13 a, 13 b and access to associated (RMON) counters. The SRAM is used for local look-up tables and queue management functions.
  • The [0142] SRAM controller 26 b supports the following transactions:
    Microengine requests (via private bus) to/from SRAM.
    Core Processor (via ASB bus) to/from SRAM.
  • The [0143] SRAM controller 26 b performs memory reference sorting to minimize delays (bubbles) in the pipeline from the SRAM interface 140 to memory 16 b. The SRAM controller 26 b does memory reference sorting based on the read function. A bubble can either be 1 or 2 cycles depending on the type of memory device employed.
  • The [0144] SRAM controller 26 b includes a lock lookup device 142 which is an eight (8 entry address content addressable memory for look-ups of read locks. Each position include a valid bit that is examined by subsequent read-lock requests. The address and command queue 120 also includes a Read Lock Fail Queue 120 d. The Read Lock Fail Queue 120 d is used to hold read memory reference requests that fail because of a lock existing on a portion of memory. That is, one of the microengines issues a memory request that has a read lock request that is processed in address and control queue 120. The memory request will operate on either the order queue 120 c or the read queue 120 b and will recognize it as a read lock request. The controller 26 b will access lock lookup device 142 to determine whether this memory location is already locked. If this memory location is locked from any prior read lock request, then this memory lock request will fail and will be stored in the read lock fail queue 120 d. If it is unlocked or if 142 shows no lock on that address, then the address of that memory reference will be used by the SRAM interface 140 to perform a traditional SRAM address read/write request to memory 16 b. The command controller and address generator 138 will also enter the lock into the lock look up device 142 so that subsequent read lock requests will find the memory location locked. A memory location is unlocked by operation of the a microcontrol instruction in a program after the need for the lock has ended. The location is unlocked by clearing the valid bit in the CAM. After an unlock, the read lock fail queue 120 d becomes the highest priority queue giving all queued read lock misses, a chance to issue a memory lock request.
  • As shown in FIG. 5A, typical timing of a static random access memory without active memory optimization and with active memory optimization is shown. As can be seen, grouping reads and writes improves cycletime eliminating dead cycles. [0145]
  • Referring to FIG. 6, communication between the microengines [0146] 22 and the FBUS interface Logic (FBI) is shown. The FBUS interface 28 in a network application can performs header processing of incoming packets from the FBUS 18. A key function which the FBUS interface performs is extraction of packet headers, and a microprogrammable source/destination/protocol hashed lookup in SRAM. If the hash does not successfully resolve, the packet header is promoted to the core processor 28 for more sophisticated processing.
  • The [0147] FBI 28 contains a Transmit FIFO 182, a Receive FIFO 183, a HASH unit 188 and FBI control and status registers 189. These four units communicate with the microengines 22, via a time-multiplexed access to the SRAM bus 38 which is connected to the transfer registers 78, 80 in the microengines. That is, all communications to and from the microengines are via the transfer registers 78, 80. The FBUS interface 28 includes a push state machine 200 for pushing data into the transfer registers during the time cycles which the SRAM is NOT using the SRAM data bus (part of bus 38) and a pull state machine 202 for fetching data from the transfer registers in the respective microengine.
  • The Hashing unit includes a pair of FIFO's [0148] 188 a, 188 b. The hash unit determines that the FBI 28 received an FBI_hash request. The hash unit 188 fetches hash keys from the calling microengine 22. After the keys are fetched and hashed, the indices are delivered back to the calling microengine 22. Up to three hashes are performed under a single FBI_hash request. The busses 34 and 38 are each unidirectional: SDRAM_push/pull_data, and Sbus_push/pull_data. Each of these busses require control signals which will provide read/write controls to the appropriate microengine 22 Transfer registers.
  • Generally, transfer registers require protection from the context controlling them to guarantee read correctness. In particular, if a write transfer register is being used by a thread_[0149] 1 to provide data to the SDRAM 16 a, thread_1 must not overwrite this register until the signal back from SDRAM controller 26 a indicates that this register has been promoted and may now be re-used. Every write does not require a signal back from the destination indicating that the function has been completed, because if the thread writes to the same command queue at that destination with multiple requests, the order of the completion is guaranteed within that command queue, thus only the last command requires the signaling back to the thread. However, if the thread uses multiple command queues (order and read), then these command requests must be broken into separate context tasks, so that ordering is maintained via context swapping. The exception case indicated at the beginning of this paragraph is relative to a certain class of operations using an unsolicited PUSH to transfer registers from the FBI for FBUS status information. In order to protect read/write determinism on the transfer registers, the FBI provides a special Push_protect signal when these special FBI push operations are set up.
  • Any [0150] microengine 22 that uses the FBI unsolicited push technique must test the protection flag prior to accessing the FBUS interface/microengine agreed upon transfer registers. If the flag is not asserted, then the transfer registers may be accessed by the microengine. If the flag is Asserted then the context should wait N cycles prior to accessing the registers. A priori this count is determined by the number of transfer registers being pushed, plus a frontend protection window. The basic idea is that the microengine must test this flag then quickly move the data which it wishes to read from the read transfer registers to GPR's in contiguous cycles, so the push engine does not collide with the microengine read.
  • Other Embodiments
  • It is to be understood that while the invention has been described in conjunction with the detailed description thereof, the foregoing description is intended to illustrate and not limit the scope of the invention, which is defined by the scope of the appended claims. Other aspects, advantages, and modifications are within the scope of the following claims.[0151]

Claims (20)

What is claimed is:
1. A parallel hardware-based multithreaded processor comprises:
a general purpose processor that coordinates system functions; and
a plurality of microengines that support multiple hardware threads.
2. The processor of claim 1 wherein the general purpose processor load microcontrol programs in the plurality of microcontrol engines.
3. The processor of claim 1 further comprising a memory control system.
4. The processor of claim 1 wherein the memory control system comprises a synchronous dynamic random access memory controller that sorts memory references based on whether the memory references are directed to an even bank or an odd bank of memory.
5. The processor of claim 1 wherein the memory control system comprises a static random access memory controller that optimizes memory references based upon whether the memory references are read references or write references.
6. The processor of claim 1 wherein each of the plurality of microengines employ hardware-based context swapping amongst a plurality of threads that are independently executable within each of the microengines.
7. The processor of claim 1 further comprising a high speed bus interface that couples the processor to a communication bus.
8. The processor of claim 1 further comprising a bus interface that couples the processor to a computer system bus.
9. The processor of claim 1 further comprising an internal bus arrangement to couple shared resources in the processor to the plurality of microengines.
10. The processor of claim 9 wherein the internal bus arrangement to couple shared resources, comprises:
a first bus to couple the general purpose processor to the plurality of microengines.
11. The processor of claim 9 wherein the internal bus arrangement to couple shared resources, comprises:
a translator device that translates requests from the general purpose processor to the microengines; and
a first bus to couple the general purpose processor to the plurality of microengines.
12. The processor of claim 3 wherein the internal bus arrangement to couple shared resources, comprises:
a translator device that translates requests from the general purpose processor to the microengines; and
a first bus to couple the general purpose processor to the plurality of microengines; and
a second bus to couple the general purpose processor to the memory control system.
13. The processor of claim 11, further comprising a third bus to couple the microengines to external bus interfaces.
14. The processor of claim 8 wherein the shared resources comprise:
a memory controller for controlling access to low latency memory;
a memory controller for controlling an access to high bandwidth memory;
a bus interface for controlling access to a communications bus; and
a bus interface for controlling access to a computer bus.
15. The processor of claim 1 wherein each one of the microengines includes a program counter to uniquely identify a position of a thread during execution in the microengine.
16. The processor of claim 1 wherein the processor supports global signaling to each of the microengines.
17. The processor of claim 16 wherein the global signaling is available to each thread in each microengine.
18. The processor of claim 17 wherein the global signaling is available to each thread to permit each thread to take a branch.
19. A parallel hardware-based multithreaded processor comprises:
a general purpose processor that coordinates system functions;
a plurality of microengines that support multiple hardware threads;
a memory control system comprising:
a first memory controller that sorts memory references based on whether the memory references are directed to an even bank or an odd bank of memory; and
a second memory controller that optimizes memory references based upon whether the memory references are read references or write references.
20. The parallel hardware-based multithreaded processor of claim 19 wherein the first memory controller controls synchronous dynamic random access memory and the second memory controller controls static random access static memory.
US10/615,500 1999-08-31 2003-07-08 Parallel processor architecture Abandoned US20040073778A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US10/615,500 US20040073778A1 (en) 1999-08-31 2003-07-08 Parallel processor architecture

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US09/387,111 US6606704B1 (en) 1999-08-31 1999-08-31 Parallel multithreaded processor with plural microengines executing multiple threads each microengine having loadable microcode
US10/615,500 US20040073778A1 (en) 1999-08-31 2003-07-08 Parallel processor architecture

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
US09/387,111 Continuation US6606704B1 (en) 1999-08-31 1999-08-31 Parallel multithreaded processor with plural microengines executing multiple threads each microengine having loadable microcode

Publications (1)

Publication Number Publication Date
US20040073778A1 true US20040073778A1 (en) 2004-04-15

Family

ID=23528511

Family Applications (2)

Application Number Title Priority Date Filing Date
US09/387,111 Expired - Lifetime US6606704B1 (en) 1999-08-31 1999-08-31 Parallel multithreaded processor with plural microengines executing multiple threads each microengine having loadable microcode
US10/615,500 Abandoned US20040073778A1 (en) 1999-08-31 2003-07-08 Parallel processor architecture

Family Applications Before (1)

Application Number Title Priority Date Filing Date
US09/387,111 Expired - Lifetime US6606704B1 (en) 1999-08-31 1999-08-31 Parallel multithreaded processor with plural microengines executing multiple threads each microengine having loadable microcode

Country Status (9)

Country Link
US (2) US6606704B1 (en)
EP (1) EP1221105B1 (en)
CN (1) CN1185592C (en)
AT (1) ATE253238T1 (en)
AU (1) AU6641900A (en)
CA (1) CA2391833C (en)
DE (1) DE60006270T2 (en)
HK (1) HK1049716B (en)
WO (1) WO2001016782A2 (en)

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20030041216A1 (en) * 2001-08-27 2003-02-27 Rosenbluth Mark B. Mechanism for providing early coherency detection to enable high performance memory updates in a latency sensitive multithreaded environment
US20030067934A1 (en) * 2001-09-28 2003-04-10 Hooper Donald F. Multiprotocol decapsulation/encapsulation control structure and packet protocol conversion method
US20030105899A1 (en) * 2001-08-27 2003-06-05 Rosenbluth Mark B. Multiprocessor infrastructure for providing flexible bandwidth allocation via multiple instantiations of separate data buses, control buses and support mechanisms
US20030115426A1 (en) * 2001-12-17 2003-06-19 Rosenbluth Mark B. Congestion management for high speed queuing
US20050132132A1 (en) * 2001-08-27 2005-06-16 Rosenbluth Mark B. Software controlled content addressable memory in a general purpose execution datapath
US20050144413A1 (en) * 2003-12-30 2005-06-30 Chen-Chi Kuo Method and apparatus utilizing non-uniformly distributed DRAM configurations and to detect in-range memory address matches
US20050216710A1 (en) * 2002-01-17 2005-09-29 Wilkinson Hugh M Iii Parallel processor with functional pipeline providing programming engines by supporting multiple contexts and critical section
US20050254480A1 (en) * 2004-05-17 2005-11-17 Kicheon Kim Router using clock synchronizer for distributed traffic control
US20060059316A1 (en) * 2004-09-10 2006-03-16 Cavium Networks Method and apparatus for managing write back cache
US20060059314A1 (en) * 2004-09-10 2006-03-16 Cavium Networks Direct access to low-latency memory
US20060067348A1 (en) * 2004-09-30 2006-03-30 Sanjeev Jain System and method for efficient memory access of queue control data structures
US20060072563A1 (en) * 2004-10-05 2006-04-06 Regnier Greg J Packet processing
US20060140203A1 (en) * 2004-12-28 2006-06-29 Sanjeev Jain System and method for packet queuing
US20060155959A1 (en) * 2004-12-21 2006-07-13 Sanjeev Jain Method and apparatus to provide efficient communication between processing elements in a processor unit
US20070234009A1 (en) * 2000-08-31 2007-10-04 Intel Corporation Processor having a dedicated hash unit integrated within
US7418582B1 (en) * 2004-05-13 2008-08-26 Sun Microsystems, Inc. Versatile register file design for a multi-threaded processor utilizing different modes and register windows
US7558925B2 (en) 2004-09-10 2009-07-07 Cavium Networks, Inc. Selective replication of data structures
US7751402B2 (en) 1999-12-29 2010-07-06 Intel Corporation Method and apparatus for gigabit packet assignment for multithreaded packet processing
USRE41849E1 (en) 1999-12-22 2010-10-19 Intel Corporation Parallel multi-threaded processing
US7895239B2 (en) 2002-01-04 2011-02-22 Intel Corporation Queue arrays in network devices
US7991983B2 (en) 1999-09-01 2011-08-02 Intel Corporation Register set used in multithreaded parallel processor architecture
US8099777B1 (en) * 2004-08-26 2012-01-17 Rockwell Collins, Inc. High security, multi-level processor and method of operating a computing system
US8316191B2 (en) 1999-08-31 2012-11-20 Intel Corporation Memory controllers for processor having multiple programmable units
US8738886B2 (en) 1999-12-27 2014-05-27 Intel Corporation Memory mapping in a processor having multiple programmable units
US10884662B2 (en) * 2018-08-06 2021-01-05 Silicon Motion, Inc. Method for performing storage control in a storage server, associated memory device and memory controller thereof, and associated storage server

Families Citing this family (81)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7266725B2 (en) 2001-09-03 2007-09-04 Pact Xpp Technologies Ag Method for debugging reconfigurable architectures
DE19651075A1 (en) 1996-12-09 1998-06-10 Pact Inf Tech Gmbh Unit for processing numerical and logical operations, for use in processors (CPU's), multi-computer systems, data flow processors (DFP's), digital signal processors (DSP's) or the like
DE19654595A1 (en) 1996-12-20 1998-07-02 Pact Inf Tech Gmbh I0 and memory bus system for DFPs as well as building blocks with two- or multi-dimensional programmable cell structures
US6542998B1 (en) 1997-02-08 2003-04-01 Pact Gmbh Method of self-synchronization of configurable elements of a programmable module
US8686549B2 (en) * 2001-09-03 2014-04-01 Martin Vorbach Reconfigurable elements
DE19861088A1 (en) 1997-12-22 2000-02-10 Pact Inf Tech Gmbh Repairing integrated circuits by replacing subassemblies with substitutes
WO2000077652A2 (en) 1999-06-10 2000-12-21 Pact Informationstechnologie Gmbh Sequence partitioning in cell structures
US6631430B1 (en) * 1999-12-28 2003-10-07 Intel Corporation Optimizations to receive packet status from fifo bus
US6898179B1 (en) * 2000-04-07 2005-05-24 International Business Machines Corporation Network processor/software control architecture
DE50115584D1 (en) 2000-06-13 2010-09-16 Krass Maren PIPELINE CT PROTOCOLS AND COMMUNICATION
US8058899B2 (en) 2000-10-06 2011-11-15 Martin Vorbach Logic cell array and bus system
US7131125B2 (en) * 2000-12-22 2006-10-31 Nortel Networks Limited Method and system for sharing a computer resource between instruction threads of a multi-threaded process
US7444531B2 (en) 2001-03-05 2008-10-28 Pact Xpp Technologies Ag Methods and devices for treating and processing data
US7844796B2 (en) 2001-03-05 2010-11-30 Martin Vorbach Data processing device and method
US9037807B2 (en) 2001-03-05 2015-05-19 Pact Xpp Technologies Ag Processor arrangement on a chip including data processing, memory, and interface elements
US7996827B2 (en) 2001-08-16 2011-08-09 Martin Vorbach Method for the translation of programs for reconfigurable architectures
US7434191B2 (en) 2001-09-03 2008-10-07 Pact Xpp Technologies Ag Router
US8686475B2 (en) 2001-09-19 2014-04-01 Pact Xpp Technologies Ag Reconfigurable elements
US7676588B2 (en) * 2001-10-05 2010-03-09 International Business Machines Corporation Programmable network protocol handler architecture
US7072970B2 (en) * 2001-10-05 2006-07-04 International Business Machines Corporation Programmable network protocol handler architecture
AU2003208266A1 (en) 2002-01-19 2003-07-30 Pact Xpp Technologies Ag Reconfigurable processor
US7610451B2 (en) * 2002-01-25 2009-10-27 Intel Corporation Data transfer mechanism using unidirectional pull bus and push bus
EP2043000B1 (en) 2002-02-18 2011-12-21 Richter, Thomas Bus systems and reconfiguration method
US8914590B2 (en) 2002-08-07 2014-12-16 Pact Xpp Technologies Ag Data processing method and device
US7437724B2 (en) * 2002-04-03 2008-10-14 Intel Corporation Registers for data transfers
US20070083730A1 (en) * 2003-06-17 2007-04-12 Martin Vorbach Data processing device and method
US7657861B2 (en) 2002-08-07 2010-02-02 Pact Xpp Technologies Ag Method and device for processing data
AU2003286131A1 (en) 2002-08-07 2004-03-19 Pact Xpp Technologies Ag Method and device for processing data
US20040034858A1 (en) * 2002-08-14 2004-02-19 Kushlis Robert J. Programming a multi-threaded processor
AU2003289844A1 (en) 2002-09-06 2004-05-13 Pact Xpp Technologies Ag Reconfigurable sequencer structure
JP4700611B2 (en) 2003-08-28 2011-06-15 ペーアーツェーテー イクスペーペー テクノロジーズ アクチエンゲゼルシャフト Data processing apparatus and data processing method
US7664823B1 (en) 2003-09-24 2010-02-16 Cisco Technology, Inc. Partitioned packet processing in a multiprocessor environment
US20050108479A1 (en) * 2003-11-06 2005-05-19 Sridhar Lakshmanamurthy Servicing engine cache requests
US20050102474A1 (en) * 2003-11-06 2005-05-12 Sridhar Lakshmanamurthy Dynamically caching engine instructions
US7536692B2 (en) 2003-11-06 2009-05-19 Intel Corporation Thread-based engine cache partitioning
JP4502650B2 (en) * 2004-02-03 2010-07-14 日本電気株式会社 Array type processor
US7555753B2 (en) * 2004-02-26 2009-06-30 International Business Machines Corporation Measuring processor use in a hardware multithreading processor environment
US20050198482A1 (en) * 2004-03-02 2005-09-08 Altek Corporation Central processing unit having a micro-code engine
US20050216655A1 (en) * 2004-03-25 2005-09-29 Rosenbluth Mark B Content addressable memory constructed from random access memory
US7181568B2 (en) * 2004-03-25 2007-02-20 Intel Corporation Content addressable memory to identify subtag matches
US20060048156A1 (en) * 2004-04-02 2006-03-02 Lim Soon C Unified control store
US8074051B2 (en) * 2004-04-07 2011-12-06 Aspen Acquisition Corporation Multithreaded processor with multiple concurrent pipelines per thread
US7418540B2 (en) * 2004-04-28 2008-08-26 Intel Corporation Memory controller with command queue look-ahead
US7216216B1 (en) 2004-06-30 2007-05-08 Sun Microsystems, Inc. Register window management using first pipeline to change current window and second pipeline to read operand from old window and write operand to new window
US7426630B1 (en) * 2004-06-30 2008-09-16 Sun Microsystems, Inc. Arbitration of window swap operations
DE102004035843B4 (en) * 2004-07-23 2010-04-15 Infineon Technologies Ag Router Network Processor
GB2417105B (en) * 2004-08-13 2008-04-09 Clearspeed Technology Plc Processor memory system
US20060095730A1 (en) * 2004-09-30 2006-05-04 Gilbert Wolrich Expansion of compute engine code space by sharing adjacent control stores using interleaved program addresses
US7277990B2 (en) 2004-09-30 2007-10-02 Sanjeev Jain Method and apparatus providing efficient queue descriptor memory access
US7418543B2 (en) 2004-12-21 2008-08-26 Intel Corporation Processor having content addressable memory with command ordering
US7467256B2 (en) * 2004-12-28 2008-12-16 Intel Corporation Processor having content addressable memory for block-based queue structures
US7734895B1 (en) 2005-04-28 2010-06-08 Massachusetts Institute Of Technology Configuring sets of processor cores for processing instructions
US8464025B2 (en) * 2005-05-20 2013-06-11 Sony Corporation Signal processing apparatus with signal control units and processor units operating based on different threads
CN100407701C (en) * 2005-06-25 2008-07-30 华为技术有限公司 Network processor
US20070044103A1 (en) * 2005-07-25 2007-02-22 Mark Rosenbluth Inter-thread communication of lock protected data
US7853951B2 (en) * 2005-07-25 2010-12-14 Intel Corporation Lock sequencing to reorder and grant lock requests from multiple program threads
US7434039B2 (en) * 2005-09-13 2008-10-07 Freescale Semiconductor, Inc. Computer processor capable of responding with comparable efficiency to both software-state-independent and state-dependent events
US7996255B1 (en) 2005-09-29 2011-08-09 The Mathworks, Inc. System and method for providing sales leads based on-demand software trial usage
US20070124728A1 (en) * 2005-11-28 2007-05-31 Mark Rosenbluth Passing work between threads
US7900022B2 (en) * 2005-12-30 2011-03-01 Intel Corporation Programmable processing unit with an input buffer and output buffer configured to exclusively exchange data with either a shared memory logic or a multiplier based upon a mode instruction
US20070157030A1 (en) * 2005-12-30 2007-07-05 Feghali Wajdi K Cryptographic system component
JP2009524134A (en) 2006-01-18 2009-06-25 ペーアーツェーテー イクスペーペー テクノロジーズ アクチエンゲゼルシャフト Hardware definition method
EP1979808B1 (en) * 2006-02-03 2011-12-07 Fish, Russell H. III. Thread optimized multiprocessor architecture
US8984256B2 (en) * 2006-02-03 2015-03-17 Russell Fish Thread optimized multiprocessor architecture
US8001549B2 (en) * 2006-04-27 2011-08-16 Panasonic Corporation Multithreaded computer system and multithread execution control method
US20090228686A1 (en) * 2007-05-22 2009-09-10 Koenck Steven E Energy efficient processing device
US7693167B2 (en) * 2007-05-22 2010-04-06 Rockwell Collins, Inc. Mobile nodal based communication system, method and apparatus
US20090228693A1 (en) * 2007-05-22 2009-09-10 Koenck Steven E System and method for large microcoded programs
US7843554B2 (en) * 2008-04-25 2010-11-30 Rockwell Collins, Inc. High dynamic range sensor system and method
US9038087B2 (en) * 2008-06-18 2015-05-19 Microsoft Technology Licensing, Llc Fence elision for work stealing
CN102023844B (en) * 2009-09-18 2014-04-09 深圳中微电科技有限公司 Parallel processor and thread processing method thereof
US8732711B2 (en) * 2010-09-24 2014-05-20 Nvidia Corporation Two-level scheduler for multi-threaded processing
US10095526B2 (en) 2012-10-12 2018-10-09 Nvidia Corporation Technique for improving performance in multi-threaded processing units
US9870339B2 (en) * 2015-06-26 2018-01-16 Intel Corporation Hardware processors and methods for tightly-coupled heterogeneous computing
US10699362B2 (en) * 2016-06-23 2020-06-30 Intel Corporation Divergent control flow for fused EUs
US20180330288A1 (en) * 2017-05-15 2018-11-15 Alteryx, Inc. Method of data aggregation for cache optimization and efficient processing
US10489348B2 (en) 2017-07-17 2019-11-26 Alteryx, Inc. Performing hash joins using parallel processing
US10552452B2 (en) 2017-10-16 2020-02-04 Alteryx, Inc. Asynchronously processing sequential data blocks
US10558364B2 (en) 2017-10-16 2020-02-11 Alteryx, Inc. Memory allocation in a data analytics system
US10810064B2 (en) 2018-04-27 2020-10-20 Nasdaq Technology Ab Publish-subscribe framework for application execution
CN109739556B (en) * 2018-12-13 2021-03-26 北京空间飞行器总体设计部 General deep learning processor based on multi-parallel cache interaction and calculation

Citations (99)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US379241A (en) * 1888-03-13 nelson
US3373408A (en) * 1965-04-16 1968-03-12 Rca Corp Computer capable of switching between programs without storage and retrieval of the contents of operation registers
US3736566A (en) * 1971-08-18 1973-05-29 Ibm Central processing unit with hardware controlled checkpoint and retry facilities
US3792441A (en) * 1972-03-08 1974-02-12 Burroughs Corp Micro-program having an overlay micro-instruction
US3940745A (en) * 1973-06-05 1976-02-24 Ing. C. Olivetti & C., S.P.A. Data processing unit having a plurality of hardware circuits for processing data at different priority levels
US4016548A (en) * 1975-04-11 1977-04-05 Sperry Rand Corporation Communication multiplexer module
US4075691A (en) * 1975-11-06 1978-02-21 Bunker Ramo Corporation Communication control unit
US4514807A (en) * 1980-05-21 1985-04-30 Tatsuo Nogi Parallel computer
US4745544A (en) * 1985-12-12 1988-05-17 Texas Instruments Incorporated Master/slave sequencing processor with forced I/O
US4831358A (en) * 1982-12-21 1989-05-16 Texas Instruments Incorporated Communications system employing control line minimization
US4991112A (en) * 1987-12-23 1991-02-05 U.S. Philips Corporation Graphics system with graphics controller and DRAM controller
US5115507A (en) * 1987-12-23 1992-05-19 U.S. Philips Corp. System for management of the priorities of access to a memory and its application
US5390329A (en) * 1990-06-11 1995-02-14 Cray Research, Inc. Responding to service requests using minimal system-side context in a multiprocessor environment
US5392411A (en) * 1992-02-03 1995-02-21 Matsushita Electric Industrial Co., Ltd. Dual-array register file with overlapping window registers
US5392391A (en) * 1991-10-18 1995-02-21 Lsi Logic Corporation High performance graphics applications controller
US5392412A (en) * 1991-10-03 1995-02-21 Standard Microsystems Corporation Data communication controller for use with a single-port data packet buffer
US5404464A (en) * 1993-02-11 1995-04-04 Ast Research, Inc. Bus control system and method that selectively generate an early address strobe
US5404469A (en) * 1992-02-25 1995-04-04 Industrial Technology Research Institute Multi-threaded microprocessor architecture utilizing static interleaving
US5404484A (en) * 1992-09-16 1995-04-04 Hewlett-Packard Company Cache system for reducing memory latency times
US5404482A (en) * 1990-06-29 1995-04-04 Digital Equipment Corporation Processor and method for preventing access to a locked memory block by recording a lock in a content addressable memory with outstanding cache fills
US5485455A (en) * 1994-01-28 1996-01-16 Cabletron Systems, Inc. Network having secure fast packet switching and guaranteed quality of service
US5515296A (en) * 1993-11-24 1996-05-07 Intel Corporation Scan path for encoding and decoding two-dimensional signals
US5517648A (en) * 1993-04-30 1996-05-14 Zenith Data Systems Corporation Symmetric multiprocessing system with unified environment and distributed system functions
US5592622A (en) * 1995-05-10 1997-01-07 3Com Corporation Network intermediate system with message passing architecture
US5613071A (en) * 1995-07-14 1997-03-18 Intel Corporation Method and apparatus for providing remote memory access in a distributed memory multiprocessor system
US5613136A (en) * 1991-12-04 1997-03-18 University Of Iowa Research Foundation Locality manager having memory and independent code, bus interface logic, and synchronization components for a processing element for intercommunication in a latency tolerant multiple processor
US5623489A (en) * 1991-09-26 1997-04-22 Ipc Information Systems, Inc. Channel allocation system for distributed digital switching network
US5627829A (en) * 1993-10-07 1997-05-06 Gleeson; Bryan J. Method for reducing unnecessary traffic over a computer network
US5630130A (en) * 1992-12-23 1997-05-13 Centre Electronique Horloger S.A. Multi-tasking low-power controller having multiple program counters
US5630641A (en) * 1993-07-29 1997-05-20 Aisin Seiki Kabushiki Kaisha Sunroof device for vehicle
US5717898A (en) * 1991-10-11 1998-02-10 Intel Corporation Cache coherency mechanism for multiprocessor computer systems
US5721870A (en) * 1994-05-25 1998-02-24 Nec Corporation Lock control for a shared main storage data processing system
US5740402A (en) * 1993-12-15 1998-04-14 Silicon Graphics, Inc. Conflict resolution in interleaved memory systems with multiple parallel accesses
US5742782A (en) * 1994-04-15 1998-04-21 Hitachi, Ltd. Processing apparatus for executing a plurality of VLIW threads in parallel
US5742587A (en) * 1997-02-28 1998-04-21 Lanart Corporation Load balancing port switching hub
US5742822A (en) * 1994-12-19 1998-04-21 Nec Corporation Multithreaded processor which dynamically discriminates a parallel execution and a sequential execution of threads
US5745913A (en) * 1996-08-05 1998-04-28 Exponential Technology, Inc. Multi-processor DRAM controller that prioritizes row-miss requests to stale banks
US5751987A (en) * 1990-03-16 1998-05-12 Texas Instruments Incorporated Distributed processing memory chip with embedded logic having both data memory and broadcast memory
US5754764A (en) * 1994-02-22 1998-05-19 National Semiconductor Corp. Combination of input output circuitry and local area network systems
US5860158A (en) * 1996-11-15 1999-01-12 Samsung Electronics Company, Ltd. Cache control unit with a cache request transaction-oriented protocol
US5887134A (en) * 1997-06-30 1999-03-23 Sun Microsystems System and method for preserving message order while employing both programmed I/O and DMA operations
US5886992A (en) * 1995-04-14 1999-03-23 Valtion Teknillinen Tutkimuskeskus Frame synchronized ring system and method
US5890208A (en) * 1996-03-30 1999-03-30 Samsung Electronics Co., Ltd. Command executing method for CD-ROM disk drive
US5892979A (en) * 1994-07-20 1999-04-06 Fujitsu Limited Queue control apparatus including memory to save data received when capacity of queue is less than a predetermined threshold
US5898701A (en) * 1995-12-21 1999-04-27 Cypress Semiconductor Corporation Method and apparatus for testing a device
US5905876A (en) * 1996-12-16 1999-05-18 Intel Corporation Queue ordering for memory and I/O transactions in a multiple concurrent transaction computer system
US5905889A (en) * 1997-03-20 1999-05-18 International Business Machines Corporation Resource management system using next available integer from an integer pool and returning the integer thereto as the next available integer upon completion of use
US6012151A (en) * 1996-06-28 2000-01-04 Fujitsu Limited Information processing apparatus and distributed processing control method
US6014729A (en) * 1997-09-29 2000-01-11 Firstpass, Inc. Shared memory arbitration apparatus and method
US6023742A (en) * 1996-07-18 2000-02-08 University Of Washington Reconfigurable computing architecture for providing pipelined data paths
US6032190A (en) * 1997-10-03 2000-02-29 Ascend Communications, Inc. System and method for processing data packets
US6049867A (en) * 1995-06-07 2000-04-11 International Business Machines Corporation Method and system for multi-thread switching only when a cache miss occurs at a second or higher level
US6058168A (en) * 1995-12-29 2000-05-02 Tixi.Com Gmbh Telecommunication Systems Method and microcomputer system for the automatic, secure and direct transmission of data
US6061710A (en) * 1997-10-29 2000-05-09 International Business Machines Corporation Multithreaded processor incorporating a thread latch register for interrupt service new pending threads
US6067300A (en) * 1998-06-11 2000-05-23 Cabletron Systems, Inc. Method and apparatus for optimizing the transfer of data packets between local area networks
US6067585A (en) * 1997-06-23 2000-05-23 Compaq Computer Corporation Adaptive interface controller that can operate with segments of different protocol and transmission rates in a single integrated device
US6070231A (en) * 1997-12-02 2000-05-30 Intel Corporation Method and apparatus for processing memory requests that require coherency transactions
US6170051B1 (en) * 1997-08-01 2001-01-02 Micron Technology, Inc. Apparatus and method for program level parallelism in a VLIW processor
US6182177B1 (en) * 1997-06-13 2001-01-30 Intel Corporation Method and apparatus for maintaining one or more queues of elements such as commands using one or more token queues
US6195676B1 (en) * 1989-12-29 2001-02-27 Silicon Graphics, Inc. Method and apparatus for user side scheduling in a multiprocessor operating system program that implements distributive scheduling of processes
US6199133B1 (en) * 1996-03-29 2001-03-06 Compaq Computer Corporation Management communication bus for networking devices
US6201807B1 (en) * 1996-02-27 2001-03-13 Lucent Technologies Real-time hardware method and apparatus for reducing queue processing
US6212544B1 (en) * 1997-10-23 2001-04-03 International Business Machines Corporation Altering thread priorities in a multithreaded processor
US6212542B1 (en) * 1996-12-16 2001-04-03 International Business Machines Corporation Method and system for executing a program within a multiscalar processor by processing linked thread descriptors
US6212611B1 (en) * 1998-11-03 2001-04-03 Intel Corporation Method and apparatus for providing a pipelined memory controller
US6216220B1 (en) * 1998-04-08 2001-04-10 Hyundai Electronics Industries Co., Ltd. Multithreaded data processing method with long latency subinstructions
US6223238B1 (en) * 1998-03-31 2001-04-24 Micron Electronics, Inc. Method of peer-to-peer mastering over a computer bus
US6223274B1 (en) * 1997-11-19 2001-04-24 Interuniversitair Micro-Elecktronica Centrum (Imec) Power-and speed-efficient data storage/transfer architecture models and design methodologies for programmable or reusable multi-media processors
US6223207B1 (en) * 1995-04-24 2001-04-24 Microsoft Corporation Input/output completion port queue data structures and methods for using same
US6223279B1 (en) * 1991-04-30 2001-04-24 Kabushiki Kaisha Toshiba Single chip microcomputer having a dedicated address bus and dedicated data bus for transferring register bank data to and from an on-line RAM
US6338078B1 (en) * 1998-12-17 2002-01-08 International Business Machines Corporation System and method for sequencing packets for multiprocessor parallelization in a computer network system
US6345334B1 (en) * 1998-01-07 2002-02-05 Nec Corporation High speed semiconductor memory device capable of changing data sequence for burst transmission
US6347344B1 (en) * 1998-10-14 2002-02-12 Hitachi, Ltd. Integrated multimedia system with local processor, data transfer switch, processing modules, fixed functional unit, data streamer, interface unit and multiplexer, all integrated on multimedia processor
US6356692B1 (en) * 1999-02-04 2002-03-12 Hitachi, Ltd. Optical module, transmitter, receiver, optical switch, optical communication unit, add-and-drop multiplexing unit, and method for manufacturing the optical module
US6360262B1 (en) * 1997-11-24 2002-03-19 International Business Machines Corporation Mapping web server objects to TCP/IP ports
US6366998B1 (en) * 1998-10-14 2002-04-02 Conexant Systems, Inc. Reconfigurable functional units for implementing a hybrid VLIW-SIMD programming model
US6373848B1 (en) * 1998-07-28 2002-04-16 International Business Machines Corporation Architecture for a multi-port adapter with a single media access control (MAC)
US6389449B1 (en) * 1998-12-16 2002-05-14 Clearwater Networks, Inc. Interstream control and communications for multi-streaming digital processors
US6393483B1 (en) * 1997-06-30 2002-05-21 Adaptec, Inc. Method and apparatus for network interface card load balancing and port aggregation
US6393026B1 (en) * 1998-09-17 2002-05-21 Nortel Networks Limited Data packet processing system and method for a router
US6526451B2 (en) * 1998-09-30 2003-02-25 Stmicroelectronics, Inc. Method and network device for creating circular queue structures in shared memory
US6529983B1 (en) * 1999-11-03 2003-03-04 Cisco Technology, Inc. Group and virtual locking mechanism for inter processor synchronization
US6532509B1 (en) * 1999-12-22 2003-03-11 Intel Corporation Arbitrating command requests in a parallel multi-threaded processing system
US6535878B1 (en) * 1997-05-02 2003-03-18 Roxio, Inc. Method and system for providing on-line interactivity over a server-client network
US6552826B2 (en) * 1997-02-21 2003-04-22 Worldquest Network, Inc. Facsimile network
US6553406B1 (en) * 2000-08-03 2003-04-22 Prelude Systems, Inc. Process thread system receiving request packet from server thread, initiating process thread in response to request packet, synchronizing thread process between clients-servers.
US6560667B1 (en) * 1999-12-28 2003-05-06 Intel Corporation Handling contiguous memory references in a multi-queue system
US6570850B1 (en) * 1998-04-23 2003-05-27 Giganet, Inc. System and method for regulating message flow in a digital data network
US6675190B1 (en) * 1998-10-08 2004-01-06 Alcatel Method for cooperative multitasking in a communications network, and a network element for carrying out the method
US6678746B1 (en) * 2000-08-01 2004-01-13 Hewlett-Packard Development Company, L.P. Processing network packets
US6681300B2 (en) * 1999-12-28 2004-01-20 Intel Corporation Read lock miss control and queue management
US6694380B1 (en) * 1999-12-27 2004-02-17 Intel Corporation Mapping requests from a processing unit that uses memory-mapped input-output space
US20040039895A1 (en) * 2000-01-05 2004-02-26 Intel Corporation, A California Corporation Memory shared between processing threads
US20040054880A1 (en) * 1999-08-31 2004-03-18 Intel Corporation, A California Corporation Microengine for parallel processor architecture
US20040073728A1 (en) * 1999-12-28 2004-04-15 Intel Corporation, A California Corporation Optimizations to receive packet status from FIFO bus
US20040071152A1 (en) * 1999-12-29 2004-04-15 Intel Corporation, A Delaware Corporation Method and apparatus for gigabit packet assignment for multithreaded packet processing
US6728845B2 (en) * 1999-08-31 2004-04-27 Intel Corporation SRAM controller for parallel processor architecture and method for controlling access to a RAM using read and read/write queues
US6876561B2 (en) * 1999-12-28 2005-04-05 Intel Corporation Scratchpad memory
US6983350B1 (en) * 1999-08-31 2006-01-03 Intel Corporation SDRAM controller for parallel processor architecture

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US3478322A (en) * 1967-05-23 1969-11-11 Ibm Data processor employing electronically changeable control storage
GB2311882B (en) 1996-04-04 2000-08-09 Videologic Ltd A data processing management system
US5915123A (en) * 1997-10-31 1999-06-22 Silicon Spice Method and apparatus for controlling configuration memory contexts of processing elements in a network of multiple context processing elements
US6079008A (en) * 1998-04-03 2000-06-20 Patton Electronics Co. Multiple thread multiple data predictive coded parallel processing system and method
US6272616B1 (en) * 1998-06-17 2001-08-07 Agere Systems Guardian Corp. Method and apparatus for executing multiple instruction streams in a digital processor with multiple data paths

Patent Citations (99)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US379241A (en) * 1888-03-13 nelson
US3373408A (en) * 1965-04-16 1968-03-12 Rca Corp Computer capable of switching between programs without storage and retrieval of the contents of operation registers
US3736566A (en) * 1971-08-18 1973-05-29 Ibm Central processing unit with hardware controlled checkpoint and retry facilities
US3792441A (en) * 1972-03-08 1974-02-12 Burroughs Corp Micro-program having an overlay micro-instruction
US3940745A (en) * 1973-06-05 1976-02-24 Ing. C. Olivetti & C., S.P.A. Data processing unit having a plurality of hardware circuits for processing data at different priority levels
US4016548A (en) * 1975-04-11 1977-04-05 Sperry Rand Corporation Communication multiplexer module
US4075691A (en) * 1975-11-06 1978-02-21 Bunker Ramo Corporation Communication control unit
US4514807A (en) * 1980-05-21 1985-04-30 Tatsuo Nogi Parallel computer
US4831358A (en) * 1982-12-21 1989-05-16 Texas Instruments Incorporated Communications system employing control line minimization
US4745544A (en) * 1985-12-12 1988-05-17 Texas Instruments Incorporated Master/slave sequencing processor with forced I/O
US4991112A (en) * 1987-12-23 1991-02-05 U.S. Philips Corporation Graphics system with graphics controller and DRAM controller
US5115507A (en) * 1987-12-23 1992-05-19 U.S. Philips Corp. System for management of the priorities of access to a memory and its application
US6195676B1 (en) * 1989-12-29 2001-02-27 Silicon Graphics, Inc. Method and apparatus for user side scheduling in a multiprocessor operating system program that implements distributive scheduling of processes
US5751987A (en) * 1990-03-16 1998-05-12 Texas Instruments Incorporated Distributed processing memory chip with embedded logic having both data memory and broadcast memory
US5390329A (en) * 1990-06-11 1995-02-14 Cray Research, Inc. Responding to service requests using minimal system-side context in a multiprocessor environment
US5404482A (en) * 1990-06-29 1995-04-04 Digital Equipment Corporation Processor and method for preventing access to a locked memory block by recording a lock in a content addressable memory with outstanding cache fills
US6223279B1 (en) * 1991-04-30 2001-04-24 Kabushiki Kaisha Toshiba Single chip microcomputer having a dedicated address bus and dedicated data bus for transferring register bank data to and from an on-line RAM
US5623489A (en) * 1991-09-26 1997-04-22 Ipc Information Systems, Inc. Channel allocation system for distributed digital switching network
US5392412A (en) * 1991-10-03 1995-02-21 Standard Microsystems Corporation Data communication controller for use with a single-port data packet buffer
US5717898A (en) * 1991-10-11 1998-02-10 Intel Corporation Cache coherency mechanism for multiprocessor computer systems
US5392391A (en) * 1991-10-18 1995-02-21 Lsi Logic Corporation High performance graphics applications controller
US5613136A (en) * 1991-12-04 1997-03-18 University Of Iowa Research Foundation Locality manager having memory and independent code, bus interface logic, and synchronization components for a processing element for intercommunication in a latency tolerant multiple processor
US5392411A (en) * 1992-02-03 1995-02-21 Matsushita Electric Industrial Co., Ltd. Dual-array register file with overlapping window registers
US5404469A (en) * 1992-02-25 1995-04-04 Industrial Technology Research Institute Multi-threaded microprocessor architecture utilizing static interleaving
US5404484A (en) * 1992-09-16 1995-04-04 Hewlett-Packard Company Cache system for reducing memory latency times
US5630130A (en) * 1992-12-23 1997-05-13 Centre Electronique Horloger S.A. Multi-tasking low-power controller having multiple program counters
US5404464A (en) * 1993-02-11 1995-04-04 Ast Research, Inc. Bus control system and method that selectively generate an early address strobe
US5517648A (en) * 1993-04-30 1996-05-14 Zenith Data Systems Corporation Symmetric multiprocessing system with unified environment and distributed system functions
US5630641A (en) * 1993-07-29 1997-05-20 Aisin Seiki Kabushiki Kaisha Sunroof device for vehicle
US5627829A (en) * 1993-10-07 1997-05-06 Gleeson; Bryan J. Method for reducing unnecessary traffic over a computer network
US5515296A (en) * 1993-11-24 1996-05-07 Intel Corporation Scan path for encoding and decoding two-dimensional signals
US5740402A (en) * 1993-12-15 1998-04-14 Silicon Graphics, Inc. Conflict resolution in interleaved memory systems with multiple parallel accesses
US5485455A (en) * 1994-01-28 1996-01-16 Cabletron Systems, Inc. Network having secure fast packet switching and guaranteed quality of service
US5754764A (en) * 1994-02-22 1998-05-19 National Semiconductor Corp. Combination of input output circuitry and local area network systems
US5742782A (en) * 1994-04-15 1998-04-21 Hitachi, Ltd. Processing apparatus for executing a plurality of VLIW threads in parallel
US5721870A (en) * 1994-05-25 1998-02-24 Nec Corporation Lock control for a shared main storage data processing system
US5892979A (en) * 1994-07-20 1999-04-06 Fujitsu Limited Queue control apparatus including memory to save data received when capacity of queue is less than a predetermined threshold
US5742822A (en) * 1994-12-19 1998-04-21 Nec Corporation Multithreaded processor which dynamically discriminates a parallel execution and a sequential execution of threads
US5886992A (en) * 1995-04-14 1999-03-23 Valtion Teknillinen Tutkimuskeskus Frame synchronized ring system and method
US6223207B1 (en) * 1995-04-24 2001-04-24 Microsoft Corporation Input/output completion port queue data structures and methods for using same
US5592622A (en) * 1995-05-10 1997-01-07 3Com Corporation Network intermediate system with message passing architecture
US6049867A (en) * 1995-06-07 2000-04-11 International Business Machines Corporation Method and system for multi-thread switching only when a cache miss occurs at a second or higher level
US5613071A (en) * 1995-07-14 1997-03-18 Intel Corporation Method and apparatus for providing remote memory access in a distributed memory multiprocessor system
US5898701A (en) * 1995-12-21 1999-04-27 Cypress Semiconductor Corporation Method and apparatus for testing a device
US6058168A (en) * 1995-12-29 2000-05-02 Tixi.Com Gmbh Telecommunication Systems Method and microcomputer system for the automatic, secure and direct transmission of data
US6201807B1 (en) * 1996-02-27 2001-03-13 Lucent Technologies Real-time hardware method and apparatus for reducing queue processing
US6199133B1 (en) * 1996-03-29 2001-03-06 Compaq Computer Corporation Management communication bus for networking devices
US5890208A (en) * 1996-03-30 1999-03-30 Samsung Electronics Co., Ltd. Command executing method for CD-ROM disk drive
US6012151A (en) * 1996-06-28 2000-01-04 Fujitsu Limited Information processing apparatus and distributed processing control method
US6023742A (en) * 1996-07-18 2000-02-08 University Of Washington Reconfigurable computing architecture for providing pipelined data paths
US5745913A (en) * 1996-08-05 1998-04-28 Exponential Technology, Inc. Multi-processor DRAM controller that prioritizes row-miss requests to stale banks
US5860158A (en) * 1996-11-15 1999-01-12 Samsung Electronics Company, Ltd. Cache control unit with a cache request transaction-oriented protocol
US6212542B1 (en) * 1996-12-16 2001-04-03 International Business Machines Corporation Method and system for executing a program within a multiscalar processor by processing linked thread descriptors
US5905876A (en) * 1996-12-16 1999-05-18 Intel Corporation Queue ordering for memory and I/O transactions in a multiple concurrent transaction computer system
US6552826B2 (en) * 1997-02-21 2003-04-22 Worldquest Network, Inc. Facsimile network
US5742587A (en) * 1997-02-28 1998-04-21 Lanart Corporation Load balancing port switching hub
US5905889A (en) * 1997-03-20 1999-05-18 International Business Machines Corporation Resource management system using next available integer from an integer pool and returning the integer thereto as the next available integer upon completion of use
US6535878B1 (en) * 1997-05-02 2003-03-18 Roxio, Inc. Method and system for providing on-line interactivity over a server-client network
US6182177B1 (en) * 1997-06-13 2001-01-30 Intel Corporation Method and apparatus for maintaining one or more queues of elements such as commands using one or more token queues
US6067585A (en) * 1997-06-23 2000-05-23 Compaq Computer Corporation Adaptive interface controller that can operate with segments of different protocol and transmission rates in a single integrated device
US5887134A (en) * 1997-06-30 1999-03-23 Sun Microsystems System and method for preserving message order while employing both programmed I/O and DMA operations
US6393483B1 (en) * 1997-06-30 2002-05-21 Adaptec, Inc. Method and apparatus for network interface card load balancing and port aggregation
US6170051B1 (en) * 1997-08-01 2001-01-02 Micron Technology, Inc. Apparatus and method for program level parallelism in a VLIW processor
US6014729A (en) * 1997-09-29 2000-01-11 Firstpass, Inc. Shared memory arbitration apparatus and method
US6032190A (en) * 1997-10-03 2000-02-29 Ascend Communications, Inc. System and method for processing data packets
US6212544B1 (en) * 1997-10-23 2001-04-03 International Business Machines Corporation Altering thread priorities in a multithreaded processor
US6061710A (en) * 1997-10-29 2000-05-09 International Business Machines Corporation Multithreaded processor incorporating a thread latch register for interrupt service new pending threads
US6223274B1 (en) * 1997-11-19 2001-04-24 Interuniversitair Micro-Elecktronica Centrum (Imec) Power-and speed-efficient data storage/transfer architecture models and design methodologies for programmable or reusable multi-media processors
US6360262B1 (en) * 1997-11-24 2002-03-19 International Business Machines Corporation Mapping web server objects to TCP/IP ports
US6070231A (en) * 1997-12-02 2000-05-30 Intel Corporation Method and apparatus for processing memory requests that require coherency transactions
US6345334B1 (en) * 1998-01-07 2002-02-05 Nec Corporation High speed semiconductor memory device capable of changing data sequence for burst transmission
US6223238B1 (en) * 1998-03-31 2001-04-24 Micron Electronics, Inc. Method of peer-to-peer mastering over a computer bus
US6216220B1 (en) * 1998-04-08 2001-04-10 Hyundai Electronics Industries Co., Ltd. Multithreaded data processing method with long latency subinstructions
US6570850B1 (en) * 1998-04-23 2003-05-27 Giganet, Inc. System and method for regulating message flow in a digital data network
US6067300A (en) * 1998-06-11 2000-05-23 Cabletron Systems, Inc. Method and apparatus for optimizing the transfer of data packets between local area networks
US6373848B1 (en) * 1998-07-28 2002-04-16 International Business Machines Corporation Architecture for a multi-port adapter with a single media access control (MAC)
US6393026B1 (en) * 1998-09-17 2002-05-21 Nortel Networks Limited Data packet processing system and method for a router
US6526451B2 (en) * 1998-09-30 2003-02-25 Stmicroelectronics, Inc. Method and network device for creating circular queue structures in shared memory
US6675190B1 (en) * 1998-10-08 2004-01-06 Alcatel Method for cooperative multitasking in a communications network, and a network element for carrying out the method
US6366998B1 (en) * 1998-10-14 2002-04-02 Conexant Systems, Inc. Reconfigurable functional units for implementing a hybrid VLIW-SIMD programming model
US6347344B1 (en) * 1998-10-14 2002-02-12 Hitachi, Ltd. Integrated multimedia system with local processor, data transfer switch, processing modules, fixed functional unit, data streamer, interface unit and multiplexer, all integrated on multimedia processor
US6212611B1 (en) * 1998-11-03 2001-04-03 Intel Corporation Method and apparatus for providing a pipelined memory controller
US6389449B1 (en) * 1998-12-16 2002-05-14 Clearwater Networks, Inc. Interstream control and communications for multi-streaming digital processors
US6338078B1 (en) * 1998-12-17 2002-01-08 International Business Machines Corporation System and method for sequencing packets for multiprocessor parallelization in a computer network system
US6356692B1 (en) * 1999-02-04 2002-03-12 Hitachi, Ltd. Optical module, transmitter, receiver, optical switch, optical communication unit, add-and-drop multiplexing unit, and method for manufacturing the optical module
US6983350B1 (en) * 1999-08-31 2006-01-03 Intel Corporation SDRAM controller for parallel processor architecture
US6728845B2 (en) * 1999-08-31 2004-04-27 Intel Corporation SRAM controller for parallel processor architecture and method for controlling access to a RAM using read and read/write queues
US20040054880A1 (en) * 1999-08-31 2004-03-18 Intel Corporation, A California Corporation Microengine for parallel processor architecture
US6529983B1 (en) * 1999-11-03 2003-03-04 Cisco Technology, Inc. Group and virtual locking mechanism for inter processor synchronization
US6532509B1 (en) * 1999-12-22 2003-03-11 Intel Corporation Arbitrating command requests in a parallel multi-threaded processing system
US6694380B1 (en) * 1999-12-27 2004-02-17 Intel Corporation Mapping requests from a processing unit that uses memory-mapped input-output space
US6681300B2 (en) * 1999-12-28 2004-01-20 Intel Corporation Read lock miss control and queue management
US20040073728A1 (en) * 1999-12-28 2004-04-15 Intel Corporation, A California Corporation Optimizations to receive packet status from FIFO bus
US6560667B1 (en) * 1999-12-28 2003-05-06 Intel Corporation Handling contiguous memory references in a multi-queue system
US6876561B2 (en) * 1999-12-28 2005-04-05 Intel Corporation Scratchpad memory
US20040071152A1 (en) * 1999-12-29 2004-04-15 Intel Corporation, A Delaware Corporation Method and apparatus for gigabit packet assignment for multithreaded packet processing
US20040039895A1 (en) * 2000-01-05 2004-02-26 Intel Corporation, A California Corporation Memory shared between processing threads
US6678746B1 (en) * 2000-08-01 2004-01-13 Hewlett-Packard Development Company, L.P. Processing network packets
US6553406B1 (en) * 2000-08-03 2003-04-22 Prelude Systems, Inc. Process thread system receiving request packet from server thread, initiating process thread in response to request packet, synchronizing thread process between clients-servers.

Cited By (39)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8316191B2 (en) 1999-08-31 2012-11-20 Intel Corporation Memory controllers for processor having multiple programmable units
US7991983B2 (en) 1999-09-01 2011-08-02 Intel Corporation Register set used in multithreaded parallel processor architecture
USRE41849E1 (en) 1999-12-22 2010-10-19 Intel Corporation Parallel multi-threaded processing
US9824037B2 (en) 1999-12-27 2017-11-21 Intel Corporation Memory mapping in a processor having multiple programmable units
US8738886B2 (en) 1999-12-27 2014-05-27 Intel Corporation Memory mapping in a processor having multiple programmable units
US9128818B2 (en) 1999-12-27 2015-09-08 Intel Corporation Memory mapping in a processor having multiple programmable units
US9824038B2 (en) 1999-12-27 2017-11-21 Intel Corporation Memory mapping in a processor having multiple programmable units
US9830284B2 (en) 1999-12-27 2017-11-28 Intel Corporation Memory mapping in a processor having multiple programmable units
US9830285B2 (en) 1999-12-27 2017-11-28 Intel Corporation Memory mapping in a processor having multiple programmable units
US7751402B2 (en) 1999-12-29 2010-07-06 Intel Corporation Method and apparatus for gigabit packet assignment for multithreaded packet processing
US20070234009A1 (en) * 2000-08-31 2007-10-04 Intel Corporation Processor having a dedicated hash unit integrated within
US7681018B2 (en) 2000-08-31 2010-03-16 Intel Corporation Method and apparatus for providing large register address space while maximizing cycletime performance for a multi-threaded register file set
US7743235B2 (en) 2000-08-31 2010-06-22 Intel Corporation Processor having a dedicated hash unit integrated within
US20050132132A1 (en) * 2001-08-27 2005-06-16 Rosenbluth Mark B. Software controlled content addressable memory in a general purpose execution datapath
US20030105899A1 (en) * 2001-08-27 2003-06-05 Rosenbluth Mark B. Multiprocessor infrastructure for providing flexible bandwidth allocation via multiple instantiations of separate data buses, control buses and support mechanisms
US20030041216A1 (en) * 2001-08-27 2003-02-27 Rosenbluth Mark B. Mechanism for providing early coherency detection to enable high performance memory updates in a latency sensitive multithreaded environment
US20030067934A1 (en) * 2001-09-28 2003-04-10 Hooper Donald F. Multiprotocol decapsulation/encapsulation control structure and packet protocol conversion method
US20030115426A1 (en) * 2001-12-17 2003-06-19 Rosenbluth Mark B. Congestion management for high speed queuing
US7895239B2 (en) 2002-01-04 2011-02-22 Intel Corporation Queue arrays in network devices
US8380923B2 (en) 2002-01-04 2013-02-19 Intel Corporation Queue arrays in network devices
US20050216710A1 (en) * 2002-01-17 2005-09-29 Wilkinson Hugh M Iii Parallel processor with functional pipeline providing programming engines by supporting multiple contexts and critical section
US20050144413A1 (en) * 2003-12-30 2005-06-30 Chen-Chi Kuo Method and apparatus utilizing non-uniformly distributed DRAM configurations and to detect in-range memory address matches
US7418582B1 (en) * 2004-05-13 2008-08-26 Sun Microsystems, Inc. Versatile register file design for a multi-threaded processor utilizing different modes and register windows
US7583683B2 (en) * 2004-05-17 2009-09-01 Yang Soon Cha Router using clock synchronizer for distributed traffic control
US20050254480A1 (en) * 2004-05-17 2005-11-17 Kicheon Kim Router using clock synchronizer for distributed traffic control
US8099777B1 (en) * 2004-08-26 2012-01-17 Rockwell Collins, Inc. High security, multi-level processor and method of operating a computing system
US20060059310A1 (en) * 2004-09-10 2006-03-16 Cavium Networks Local scratchpad and data caching system
US7594081B2 (en) * 2004-09-10 2009-09-22 Cavium Networks, Inc. Direct access to low-latency memory
US7558925B2 (en) 2004-09-10 2009-07-07 Cavium Networks, Inc. Selective replication of data structures
US9141548B2 (en) 2004-09-10 2015-09-22 Cavium, Inc. Method and apparatus for managing write back cache
US20060059314A1 (en) * 2004-09-10 2006-03-16 Cavium Networks Direct access to low-latency memory
US7941585B2 (en) 2004-09-10 2011-05-10 Cavium Networks, Inc. Local scratchpad and data caching system
US20060059316A1 (en) * 2004-09-10 2006-03-16 Cavium Networks Method and apparatus for managing write back cache
US20060067348A1 (en) * 2004-09-30 2006-03-30 Sanjeev Jain System and method for efficient memory access of queue control data structures
US20060072563A1 (en) * 2004-10-05 2006-04-06 Regnier Greg J Packet processing
US20060155959A1 (en) * 2004-12-21 2006-07-13 Sanjeev Jain Method and apparatus to provide efficient communication between processing elements in a processor unit
US20060140203A1 (en) * 2004-12-28 2006-06-29 Sanjeev Jain System and method for packet queuing
US10884662B2 (en) * 2018-08-06 2021-01-05 Silicon Motion, Inc. Method for performing storage control in a storage server, associated memory device and memory controller thereof, and associated storage server
US11366616B2 (en) 2018-08-06 2022-06-21 Silicon Motion, Inc. Method for performing storage control in a storage server, associated memory device and memory controller thereof, and associated storage server

Also Published As

Publication number Publication date
WO2001016782A3 (en) 2001-05-31
CN1185592C (en) 2005-01-19
DE60006270T2 (en) 2004-08-05
EP1221105A2 (en) 2002-07-10
DE60006270D1 (en) 2003-12-04
CN1387649A (en) 2002-12-25
AU6641900A (en) 2001-03-26
US6606704B1 (en) 2003-08-12
HK1049716A1 (en) 2003-05-23
HK1049716B (en) 2005-08-19
CA2391833C (en) 2006-08-01
WO2001016782A9 (en) 2002-06-27
ATE253238T1 (en) 2003-11-15
EP1221105B1 (en) 2003-10-29
WO2001016782A2 (en) 2001-03-08
CA2391833A1 (en) 2001-03-08

Similar Documents

Publication Publication Date Title
US6606704B1 (en) Parallel multithreaded processor with plural microengines executing multiple threads each microengine having loadable microcode
US7191321B2 (en) Microengine for parallel processor architecture
US6983350B1 (en) SDRAM controller for parallel processor architecture
US7743235B2 (en) Processor having a dedicated hash unit integrated within
US7305500B2 (en) Sram controller for parallel processor architecture including a read queue and an order queue for handling requests
EP1242869B1 (en) Context swap instruction for multithreaded processor
WO2001016703A1 (en) Instruction for multithreaded parallel processor

Legal Events

Date Code Title Description
AS Assignment

Owner name: INTEL CORPORATION, CALIFORNIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:ADILETTA, MATTHEW J.;WOLRICH, GILBERT;WHEELER, WILLIAM R.;REEL/FRAME:016968/0066;SIGNING DATES FROM 19991118 TO 20000214

STCB Information on status: application discontinuation

Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION