Hardware task manager

Abstract

A hardware task manager for managing operations in an adaptive computing system. The task manager indicates when input and output buffer resources are sufficient to allow a task to execute. The task can require an arbitrary number of input values from one or more other (or the same) tasks. Likewise, a number of output buffers must also be available before the task can start to execute and store results in the output buffers. The hardware task manager maintains a counter in association with each input and output buffer. For input buffers, a negative value for the counter means that there is no data in the buffer and, hence, the respective input buffer is not ready or available. Thus, the associated task can not run. Predetermined numbers of bytes, or “units,” are stored into the input buffer and an associated counter is incremented. When the counter value transitions from a negative value to a zero the high-order bit of the counter is cleared, thereby indicating the input buffer has sufficient data and is available to be processed by a task.

Description

BACKGROUND OF THE INVENTION

Field of the Invention

This invention relates in general to digital data processing and more specifically to an interconnection facility for transferring digital information among components in an adaptive computing architecture.

A common limitation to processing performance in a digital system is the efficiency and speed of transferring instruction, data and other information among different components and subsystems within the digital system. For example, the bus speed in a general-purpose Von Neumann architecture dictates how fast data can be transferred between the processor and memory and, as a result, places a limit on the computing performance (e.g., million instructions per second (MIPS), floating-point operations per second (FLOPS), etc.).

Other types of computer architecture design, such as multi-processor or parallel processor designs require complex communication, or interconnection, capabilities so that each of the different processors can communicate with other processors, with multiple memory devices, input/output (I/O) ports, etc. With today's complex processor system designs, the importance of an efficient and fast interconnection facility rises dramatically. However, such facilities are difficult to design to optimize goals of speed, flexibility and simplicity of design.

SUMMARY OF THE INVENTION

A hardware task manager indicates when input and output buffer resources are sufficient to allow a task to execute. The task can require an arbitrary number of input values from one or more other (or the same) tasks. Likewise, a number of output buffers must also be available before the task can start to execute and store results in the output buffers.

The hardware task manager maintains a counter in association with each input and output buffer. For input buffers, a negative value for the counter means that there is no data in the buffer and, hence, the respective input buffer is not ready or available. Thus, the associated task can not run. Predetermined numbers of bytes, or “units,” are stored into the input buffer and an associated counter is incremented. When the counter value transitions from a negative value to a zero the high-order bit of the counter is cleared, thereby indicating the input buffer has sufficient data and is available to be processed by a task.

Analogously, a counter is maintained in association with each output buffer. A negative value for an output buffer means that the output buffer is available to receive data. When the high-order bit of an output buffer counter is set then data can be written to the associated output buffer and the task can run.

Ports counters are used to aggregate buffer counter indications by tracking the high-order bit transitions of the counters. For example, if a task needs 10 input buffers and 20 output buffers then an input ports counter is initialized and maintained by tracking availability of the 10 allocated input buffers and 20 output buffers using simple increments and decrements according to high-order transitions of the buffer counter bits. When the high-order bit (i.e., the sign bit) of the ports counter transitions from a 1 to a 0, the associated task is ready to run.

In one embodiment the invention provides an apparatus for coordinating buffer use among tasks in a processing system, wherein the processing system includes a plurality of hardware nodes, wherein a task is executed on one or more of the hardware nodes, wherein a consuming task uses input buffers to obtain data and wherein a producing task uses output buffers to provide data, the apparatus comprising a task manager for indicating the status of the buffers, the task manager including an output buffer available indicator associated with an output buffer; an input buffer available indicator associated with an input buffer; and a status indicator for indicating that a task is ready to run based on a combination of the output buffer available indicator and the input buffer available indicator.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates the interface between heterogeneous nodes and the homogenous network in the ACE architecture;

FIG. 2 illustrates basic components of a hardware task manager;

FIG. 3 shows buffers associated with ports;

FIG. 4 shows buffer size encoding;

FIG. 5 shows a look-up table format;

FIG. 6 shows counter operations;

FIG. 7 shows a table format for task state information;

FIG. 8 illustrates a data format for a node control register;

FIG. 9 shows the layout for a node status register;

FIG. 10 shows the layout for a Port/Memory Translation Table;

FIG. 11 shows a layout for a State Information Table;

FIG. 12 shows a summary of state transitions for a task;

FIG. 13 shows a layout for the a Module Parameter List and Module Pointer Table;

FIG. 14 shows an example of packing eight parameters associated with task buffers;

FIG. 15 shows data formats for the Forward and Backward Acknowledgement Messages; and

FIG. 16 shows an overview of an adaptable computing engine architecture.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

A detailed description of an ACE architecture used in a preferred embodiment is provided in the patents referenced above. The following section provides a summary of the ACE architecture described in the referenced patents.

Adaptive Computing Engine

FIG. 16 is a block diagram illustrating an exemplary embodiment in accordance with the present invention. Apparatus 100, referred to herein as an adaptive computing engine (ACE) 100, is preferably embodied as an integrated circuit, or as a portion of an integrated circuit having other, additional components. In the exemplary embodiment, and as discussed in greater detail below, the ACE 100 includes one or more reconfigurable matrices (or nodes) 150, such as matrices 150A through 150N as illustrated, and a matrix interconnection network 110. Also in the exemplary embodiment, and as discussed in detail below, one or more of the matrices 150, such as matrices 150A and 150B, are configured for functionality as a controller 120, while other matrices, such as matrices 150C and 150D, are configured for functionality as a memory 140. The various matrices 150 and matrix interconnection network 110 may also be implemented together as fractal subunits, which may be scaled from a few nodes to thousands of nodes.

In a preferred embodiment, the ACE 100 does not utilize traditional (and typically separate) data, DMA, random access, configuration and instruction busses for signaling and other transmission between and among the reconfigurable matrices 150, the controller 120, and the memory 140, or for other input/output (“I/O”) functionality. Rather, data, control and configuration information are transmitted between and among these matrix 150 elements, utilizing the matrix interconnection network 110, which may be configured and reconfigured, in real-time, to provide any given connection between and among the reconfigurable matrices 150, including those matrices 150 configured as the controller 120 and the memory 140.

The matrices 150 configured to function as memory 140 may be implemented in any desired or exemplary way, utilizing computational elements (discussed below) of fixed memory elements, and may be included within the ACE 100 or incorporated within another IC or portion of an IC. In the exemplar embodiment, the memory 140 is included within the ACE 100, and preferably is comprised of computational elements which are low power consumption random access memory (RAM), but also may be comprised of computational elements of any other form of memory, such as flash, DRAM, SRAM, MRAM, ROM, EPROM or E2PROM. In the exemplary embodiment, the memory 140 preferably includes direct memory access (DMA) engines, not separately illustrated.

The controller 120 is preferably implemented, using matrices 150A and 150B configured as adaptive finite state machines (FSMs), as a reduced instruction set (“RISC”) processor, controller or other device or IC capable of performing the two types of functionality discussed below. (Alternatively, these functions may be implemented utilizing a conventional RISC or other processor.) The first control functionality, referred to as “kernel” control, is illustrated as kernel controller (“KARC”) of matrix 150A, and the second control functionality, referred to as “matrix” control, is illustrated as matrix controller (“MARC”) of matrix 150B. The kernel and matrix control functions of the controller 120 are explained in greater detail below, with reference to the configurability and reconfigurability of the various matrices 150, and with reference to the exemplary form of combined data, configuration and control information referred to herein as a “silverware” module.

The matrix interconnection network 110 of FIG. 16, includes subset interconnection networks (not shown). These can include a boolean interconnection network, data interconnection network, and other networks or interconnection schemes collectively and generally referred to herein as “interconnect”, “interconnection(s)” or “interconnection network(s),” or “networks,” and may be implemented generally as known in the art, such as utilizing FPGA interconnection networks or switching fabrics, albeit in a considerably more varied fashion. In the exemplary embodiment, the various interconnection networks are implemented as described, for example, in U.S. Pat. Nos. 5,218,240, 5,336,950, 5,245,227, and 5,144,166, and also as discussed below and as illustrated with reference to FIGS. 7, 8 and 9. These various interconnection networks provide selectable (or switchable) connections between and among the controller 120, the memory 140, the various matrices 150, and the computational units (or “nodes”) and computational elements, providing the physical basis for the configuration and reconfiguration referred to herein, in response to and under the control of configuration signaling generally referred to herein as “configuration information”. In addition, the various interconnection networks (110, 210, 240 and 220) provide selectable or switchable data, input, output, control and configuration paths, between and among the controller 120, the memory 140, the various matrices 150, and the computational units, components and elements, in lieu of any form of traditional or separate input/output busses, data busses, DMA, RAM, configuration and instruction busses.

It should be pointed out, however, that while any given switching or selecting operation of, or within, the various interconnection networks may be implemented as known in the art, the design and layout of the various interconnection networks, in accordance with the present invention, are new and novel, as discussed in greater detail below. For example, varying levels of interconnection are provided to correspond to the varying levels of the matrices, computational units, and elements. At the matrix 150 level, in comparison with the prior art FPGA interconnect, the matrix interconnection network 110 is considerably more limited and less “rich”, with lesser connection capability in a given area, to reduce capacitance and increase speed of operation. Within a particular matrix or computational unit, however, the interconnection network may be considerably more dense and rich, to provide greater adaptation and reconfiguration capability within a narrow or close locality of reference.

The various matrices or nodes 150 are reconfigurable and heterogeneous, namely, in general, and depending upon the desired configuration: reconfigurable matrix 150A is generally different from reconfigurable matrices 150B through 150N; reconfigurable matrix 150B is generally different from reconfigurable matrices 150A and 150C through 150N; reconfigurable matrix 150C is generally different from reconfigurable matrices 150A, 150B and 150D through 150N, and so on. The various reconfigurable matrices 150 each generally contain a different or varied mix of adaptive and reconfigurable nodes, or computational units; the nodes, in turn, generally contain a different or varied mix of fixed, application specific computational components and elements that may be adaptively connected, configured and reconfigured in various ways to perform varied functions, through the various interconnection networks. In addition to varied internal configurations and reconfigurations, the various matrices 150 may be connected, configured and reconfigured at a higher level, with respect to each of the other matrices 150, through the matrix interconnection network 110. Details of the ACE architecture can be found in the related patent applications, referenced above.

Hardware Task Manager

FIG. 1 illustrates the interface between heterogeneous nodes and the homogenous network in the ACE architecture. This interface is referred to as a “node wrapper” since it is used to provide a common input and output mechanism for each node. A node's execution units and memory are interfaced with the network and with control software via the node wrapper to provide a uniform, consistent system-level programming model. Details of the node wrapper can be found in the related patent applications referenced, above.

In a preferred embodiment, each node wrapper includes a hardware task manager (HTM) 200. Node wrappers also include data distributor 202, optional direct memory access (DMA) engine 204 and data aggregator 206. The HTM coordinates execution, or use, of node processors and resources, respectively. The HTM does this by processing a task list and producing a ready-to-run queue. The HTM is configured and controlled by a specialized node referred to as a K-node or control node (not shown). However, other embodiment can use other HTM control approaches.

A task is an instance of a module, or group of instructions. A module can be any definition of processing, functionality or resource access to be provided by one or more nodes. A task is associated with a specific module on a specific node. A task definition includes designation of resources such as “physical” memory and “logical” input and output buffers and “logical” input and output ports of the module; and by initializing configuration parameters for the task. A task has four states: Suspend, Idle, Ready, Run.

A task is created by the K-node writing to control registers in the node where the task is being created, and by the K-node writing to control registers in other nodes, if any, that will be producing data for the task and/or consuming data from the task. These registers are memory mapped into the K-node's address space, and “peek and poke” network services are used to read and write these values.

A newly created task starts in the suspend state. Once a task is configured, the K-node can issue a “go” command, setting a bit in a control register. The action of this command is to move the task from the “suspend” state to the “idle” state.

When the task is “idle” and all its input buffers and output buffers are available, the task is ADDed to the ready-to-run queue which is implemented as a FIFO; and the task state is changed to “ready/run”.

Note: Buffers are available to the task when subsequent task execution will not consume more data than is present in its input buffer(s) or will not produce more data that there is capacity in its output buffer(s).

When the execution unit is not busy and the FIFO is not empty, the task number for the next task that is ready to execute is REMOVEd from the FIFO, and the state of this task is “run”. In the “run” state, the task consumes data from its input buffers and produces data for its output buffers. For PDU, RAU and RBU unit types, only one task can be in the “run” state at a time, and the current task cannot be preempted. These restrictions are imposed to simplify hardware and software control.

When the task completes processing:

• • 1) if the task's GO bit is zero, its state will be set to SUSPEND; or
  • 2) if (its GO bit is one) AND (its PORTS_COUNTER msb is one), its state will be set to idle; or
  • 3) if (its GO bit is one) AND (the FIFO is not empty) AND (its PORTS_COUNTER msb is zero) the task will be ADDed to the ready-to-run queue and its state will be “ready”; or

4) if (its GO bit is one) AND (the FIFO is empty) AND (its PORTS_COUNTER msb is zero), its state will remain “run”; the task will execute again since its status is favorable and there is no other task waiting to run.

The K-node can clear the task's GO bit at any time. When the task reaches the “idle” state and its GO bit is zero, its state will transition to “suspend”.

The K-node can determine if a task is hung in a loop by setting and testing status. When the K-node wishes to stop a run-away task, it should clear the task's GO bit and issue the “abort” command to reset the task's control unit. After reset, the task's state will transition to “idle”. And, if its GO bit has been cleared, its state will transition to “suspend”.

Task Lists

A node has a task list, and each task is identified by its “task number”. Associated with each task are the following:

• • Task_number [4:0]—The task number, in the range of 0 to 31.
  • State [1:0] with values;
    • ‘00’=suspended
    • ‘01’=idle
    • ‘10’=ready
    • ‘11’=run
  • Go_bit with values:
    • 0=stop
    • 1=go
  • Module—Pointer to the module used to implement this task. For reconfigurable hardware modules, this may be a number that corresponds to a specific module. For the PDU, this is the instruction memory address where the module begins.
  • Ports_counter—The negative number of input ports and output ports that must be available before the task state can transition from “idle” to “ready”. For example, an initial value of −3 might indicate that two input ports and one output port must be available before the task state changes to “ready”. When a port changes from “unavailable” to “available”, Ports_counter is incremented by one. When a port changes from “available” to “unavailable”, Ports_counter is decremented by one. When the value for Ports_counter reaches (or remains) zero and the task state is “idle”, task state transitions to “ready”. The sign (high-order) bit of this counter reflects the status of all input ports and output ports for this task. When it is set, not all ports are available; and when it is clear, then all ports are available, and task state transitions from “idle” to “ready”.

Each task can have up to four input buffers. Associated with each input buffer are the following:

In port_number (0,1,2,3) [4:0]—a number in the range of 0 to 31.

Mem_hys_addr [k:0]—The physical address in memory of the input buffer.

Size [3:0]—a power-of-two coding for the size of the input buffer.

Consumer_count [15:0]—a two's complement count, with a range of −32768 to +32767, for input buffer status. It is initialized by the K-node, incremented by an amount Fwdackval by the upstream producer and incremented by an amount Negbwdackval by the consumer (this task). The sign (high-order) bit of this counter indicates input buffer status. When it is set (negative), the buffer is unavailable to this task; and when it is clear (non-negative), the buffer is available to this task.

Bwdackval [15:0]—the negative backward acknowledge value with a range of −32768 to 0.

Producer_task_number [4:0]—a number in the range of 0 to 31 indicating the producer's task number for counter maintenance, including backward acknowledgement messages to remote producers.

Producer_outport_number [4:0]—a number in the range of 0 to 31 indicating the producer's output port number for counter maintenance, including backward acknowledgement messages to remote producers.

Producer_node_number [6:0]—a number in the range of 0 to 127 indicating a remote producer's node number for routing backward acknowledgement messages to remote producers.

Each task can have up to four output buffers. Associated with each buffer is the following:

Out_port_number (0,1,2,3) [4:0]—a number in the range of 0 to 31.

Mem_phys_addr [k:0]—The physical address in memory of the output buffer, if local.

Size [3:0]—a power-of-two coding for the size of the output buffer, if local.

Producer_count [15:0]—a two's complement count, with a range of −32768 to +32767, for output buffer status. It is initialized by the K-node, incremented by an amount Fwdackval by the producer (this task) and incremented by an amount Negbwdackval by the downstream consumer. The sign (high-order) bit of this counter indicates output buffer status. When it is set (negative), the buffer is available to this task; and when it is clear (non-negative), the buffer is unavailable to this task.

Fwdackval [15:0]—the forward acknowledge value with a range of 0 to +32767.

Consumer_task_number [4:0]—a number in the range of 0 to 31 indicating the consumer's task number for counter maintenance, including forward acknowledgement messages to remote consumers.

Consumer_in_port_number [4:0]—a number in the range of 0 to 31 indicating the consumer's input port number for counter maintenance, including forward acknowledgement messages to remote consumers.

Consumer_node_number [6:0]—a number in the range of 0 to 127 indicating a remote consumer's node number for routing data and forward acknowledgement messages to remote consumers.

Parms_pointer [k:0]—The physical address in memory indicating the first of tbd entries containing the task's configuration parameters.

A preferred embodiment of the invention uses node task lists. Each list can designate up to 32 tasks. Each of the up to 32 tasks can have up to four input ports (read ports) and up to four output ports (write ports). A node can have 32 input ports and 32 output ports. 5-bit numbers are used to identify each port. Each number is associated with a 20-bit address in the contiguous address space for 1024 kilobytes of physical memory.

HTM Components

FIG. 2 illustrates basic components of an HTM. These include port-to-address translation table 220, ACKs processor 222, ready-to-run queue 224, state information 226, parameters pointers 228, and parameters memory 230.

Port-to-Address Translation Table

Under K-node control, the execution units in each node can write into any memory location in the 20-bit contiguous address space. Accessing permissions are controlled by the port number-to-physical address translation tables. There are 32 entries in the table to support up to 32 ports at each node's input.

Each of the 32 ports at each node's input can be assigned to an output port of any task executing on any node (including “this node”) on the die. Each port number is associated with a “power-of-2” sized buffer within one or more of the node's physical memory blocks as shown in FIG. 3.

The 20-bit contiguous address space is accessible by a 6-bit node number (the six high order bits) and a 14-bit (low order bits) byte address for the 16 KBytes within a tile.

Because network transfers are 32-bit transfers, 16-bit longword addresses are stored in the translation tables, and the two lower order address bits are inferred (and set to ‘00’ by each memory's address mux). The power-of-two buffer size is encoded in a four-bit value for each entry in the table as shown in FIG. 4.

The translation table is loaded/updated by the K-node. When a task writes to this node, its output port number is used to access the table. Its accompanying data is written into the current address [ADDR] that is stored in the table, and the next address [NXTADDR] is calculated as follows.

BASE=SIZE*INT {ADDR/SIZE}
OFFSET=ADDR−BASE
NXTOFFSET=(VAL+1) mod SIZE
NXTADDR=BASE+NXTOFFSET

ACKs Processor

Tasks communicate through buffers. Buffers are accessed via port numbers. Each active buffer is associated with a producer task and a consumer task. Each task maintains a count reflecting the amount of data in the buffer. As the producer writes data into the buffer, it updates its producer_counter with a value, Fwdackval, equal to the number of bytes that it has produced (written). It also updates the corresponding Consumer_count, using a FWDACK message if the consumer is remote (not in its node).

When the consumer reads, and no longer requires access to, data in the buffer, it updates its Consumer_count with a value, Bwdackval, equal to minus the number of bytes that is has consumed. It also updates the corresponding Producer_count, using a BWDACK message if the producer is remote.

Note: Data formats for the Forward and Backward Acknowledgement Messages are show in FIG. 15.

The ACKs processor includes a 64-entry by 16-bit LUT to store counts for each of its (up to) 32 input ports and 32 output ports. The format for this LUT is shown in FIG. 5.

The counters are initialized with negative values by the K-node. Producer counters are accessed by their associated output port numbers; consumer counters are accessed by their associated input port numbers.

Producer counters are incremented by Fwdackvals from their associated tasks, and they are incremented by Bwdackvals from the downstream tasks that consume the data. Consumer counters are incremented by Bwdackvals from their associated tasks, and they are incremented by Fwdackvals from the upstream tasks that produce the data.

Note that incrementing by a Bwdackval, a negative value, is equivalent to decrementing by a positive value, producing a more negative result.

These operations are summarized in FIG. 6.

An input buffer is available to its associated task when the high order bit of its consumer counter is clear, indicating a non-negative count. An input buffer is not available to its associated task when the bit is set, indicating a negative count. Consumer counters are initialized (by the K-node) with the negative number of bytes that must be in its input buffer before the associated task can execute. When the high order bit is clear, indicating buffer availability, the task is assured that the data it will consume during its execution is in the buffer.

An output buffer is available to its associated task when the high order bit of its producer counter is set, indicating a negative count. An output buffer is not available to its associated task when the bit is clear, indicating a non-negative count. Producer counters are initialized (by the K-node) with a negative number of bytes that it can produce before it must suspend task execution. An available output buffer indication assures the task that there is sufficient buffer capacity for execution with no possibility of overflow.

The initial values for these counters are functions of Ackvals and the desired numbers of task execution iterations after initialization.

To avoid deadlocks, the minimum buffer size must be the next power of two that exceeds the sum of the maximum absolute values of Fwdackvals and Bwdackvals. For example, for Fwdackval=51 and Bwdackval=−80, the buffer size must be greater than, or equal to, 256.

Counters are updated when ACKVAL messages arrive from the network and from locally executing tasks. When the high order bits of the current count and the updated count are different, a change of status indication is generated along with the associated task number, so that its STATE Ports_counter can be incremented or decremented. For input ports, the ports_counter is decremented for 0-to-1 transitions, and it is incremented for 1-to-0 transitions. For output ports, the ports_counter is incremented for 0-to-1 transitions, and it is decremented for 1-to-0 transitions.

When the high order bit of the Ports_counter transitions from 1 to 0, the associated task is ready to run; and it is ADDed to the Ready-to-Run Queue. Also, when the current task completes and its ACKs have been processed, if its GO bit is zero, its STATE is set to SUSPEND. Else, if its Ports_counter msb is clear, it is ready to run again; and, if the FIFO is empty, it runs again; or, if the FIFO is not empty, it is ADDed to the queue. Finally, if its GO bit is one, but its Ports_counter msb is clear, its STATE is set to IDLE; and it must wait for the next Ports_counter msb transition from 1 to 0 before it is once again ready to run and ADDed to the queue.

Ready-to-Run Queue

The Ready-to-Run Queue is a 32-entry by 5 bits per entry FIFO that stores the task numbers of all tasks that are ready to run. The K-node initializes the FIFO by setting its 5-bit write pointer (WP) and its 5-bit read pointer (RP) to zero. Initialization also sets the fifo status indication: EMPTY=1.

When a task is ready to run, its task number is ADDed to the queue at the location indicated by WP, and WP is incremented. For every ADD, EMPTY is set to 0.

When the execution unit is idle and the FIFO is not empty (EMPTY=0), the task number for the next task to be executed is REMOVEd from the queue at the location indicated by RP. When the task is completed, RP is incremented. And, if RP=WP, EMPTY is set to 1.

The FIFO is FULL when [(RP=WP) AND (EMPTY=0)].

State Information Table

State information for each of (up to) 32 tasks is maintained in a 32-entry by 6 bit table that is accessed by one of 32 task numbers. The format for this table is shown in FIG. 7.

The State Information Table is initialized by the K-node (POKE). The K-node also can monitor the state of any task (PEEK). In addition to the K-node's unlimited access to the table, other accesses to it are controlled by a FSM that receives inputs from the ACKs Processor, the Ready-to-Run Queue, and the Execution Unit as shown in FIG. 8. Details of this FSM are beyond the scope of this paper.

Parms Pointers

Associated with each task is a register that contains the physical address where the first of the task's configuration parameters is stored in a contiguous chunk of memory.

Parms Memory

Each task's configuration parameters—or Module Parameter List (MPL),—are stored in a contiguous chunk of memory referenced by the task's Parms Pointer. The numbers of parameters and their purposes will vary from one task to another. As tasks are designed, their specific requirements for configuration parameters will be determined and documented.

Typically, these requirements will include;

Module—Pointer to the module used to implement this task. For reconfigurable hardware modules, this may be a number that corresponds to a specific module. For the PDU, this is the instruction memory address where the module begins.

[Need Elaboration on the Following 3 Paragraphs/Sections]

For each of up to four buffers from which the task will consume (read) data:

Memory Physical Address

Buffer Size

Input Port Number

Producer Task Number

Producer Output Port Number

Producer Node Number (if remote)

Producer (Local/Remote); boolean

Bwdackval

For each of up to four buffers into which the task will produce (write) data:

Memory Physical Address (if local)

Buffer Size (if local)

Output Port Number

Consumer Task Number

Consumer Input Port Number

Consumer Node Number (if remote)

Consumer (Local/Remote); boolean

Fwdackval

For each presettable counter (for example: number of iterations count: watchdog count)

(Counter Modulus-1)

Node Control Register (NCR)

The layout for the Node Control Register is shown in FIG. 8.

ENB—Bit 15—When the NCR Enable bit is clear, the node ceases all operation, except that it continues to support PEEK and POKE operations. The NCR Enable bit must be set to 1 to enable any other node functions.

ABT—Bit 14—Writing (POKING) the NCR with Bit 14 set to 1 generates an Abort signal to the execution unit, causing it to halt immediately, The state of the aborted task transitions to IDLE; and if its GO bit has been cleared (as it should be prior to issuing the Abort), the state will transition to SUSPEND. This is the K-node's sledge hammer to terminate a runaway task. Writing the NCR with Bit 14=0 is no operation. When reading (PEEKING) NCR, zero will be returned for Bit 14.

RSV—Bit 13—At this time, Bit 13 is unused. When writing the NCR, Bit 13 is don't care, and when reading NCR, zero will be returned for Bit 13.

WPE—Bit 12—Writing the NCR with Bit 12 set to 1 results in the writing of the [9:5] value into Queue Write Pointer (with ENB=0, a diagnostics WRITE/READ/CHECK capability). Writing the NCR with Bit 12=0 is no operation. When reading NCR, zero will be returned for Bit 12.

RPE—Bit 11—Writing the NCR with Bit 11 set to 1 results in the writing of the NCR[4:0] value into Queue Read Pointer (with ENB=0, a diagnostics WRITE/READ/CHECK capability). Writing the NCR with Bit 11=0 is no operation. When reading NCR, zero will be returned for Bit 11.

Queue Initialization

Writing the NCR with Bits 12 and 11 set to 1 and with Bits [9:5] and Bits [4:0] set to zeros initializes the queue, setting the Write Pointer to zero, the Read Pointer to zero, and the Queue Empty Status Flag to 1.

Queue Empty Status Flag—Bit 10—READ ONLY Bit 10, the Queue Empty Status Flag, is set to 1 when the Ready-to-Run FIFO is empty; it is set to 0 when it is not empty. When Bit 10 is set to 1, the Write Pointer (NCR [9:5]) and Read Pointer (NCR [4:0]) values will be the same. When the pointer values are the same, and Bit 10=0, the FIFO is FULL. When writing NCR, Bit 10 is don't care.

Queue Write Pointer—Bits [9:5]—For diagnostics WRITE/READ/CHECK capability (and for queue initialization), writing NCR with Bit 12=1 results in the writing of the NCR[9:5] value into Queue Write Pointer. When writing NCR with Bit 12=0, Bits [9:5] are don't care. When reading NCR, Bits [9:5] indicate the current Queue Write Pointer value.

Queue Read Pointer—Bits [4:0]—For diagnostics WRITE/READ/CHECK capability (and for queue initialization), writing NCR with Bit 11=1 results in the writing of the NCR[4:0] value into Queue Read Pointer. When writing NCR with Bit 11=0, Bits [4:0] are don't care. When reading NCR, Bits [4:0] indicate the current Queue Read Pointer value.

Node Status Register (NSR)

The layout for the Node Status Register is shown in FIG. 9. The Node Status Register is a READ ONLY register. READING NSR clears Bits 14 and 13. WRITING NSR is no operation.

ENB—Bit 15—Bit 15, Enable, simply indicates the state of NCR [15]:Enable.

ABT—Bit 14—When an Abort command is issued (WRITE NCR, Bit 14=1), the executing task is suspended, after which the Abort Status Bit 14 is set to 1. Reading NSR clears Bit 14.

TCS—Bit 13—The Task Change Status Bit 13 is set to 1 when an execution unit REMOVEs a TASK # from the Ready-to-Run Queue. Reading NSR clears Bit 13. The K-node can perform a “watch dog” operation by reading NSR, which clears Bit 13, and reading NSR again after a time interval. After the second read, if Bit 13 is set to 1, another REMOVE (initiating execution of the next task) has occurred during the time interval. If Bit 13=0, another REMOVE has not occurred during the time interval.

NRS—Bit 12—This bit is set to 1 when the node is executing a task. When the bit=0, the node is not executing a task.

Reserved—Bits [11:5]—These bits are not assigned at this time, and reading the NSR results in zeros being returned for Bits [11:5]

Current Task Number—Bits [4:0]—Bits [4:0] is the 5-bit number (task number) associated with the task currently executing (if any).

Port/Memory Translation Table (PTT)

The layout for the 32-entry Port/Memory Translation Table (PTT) is shown in FIG. 4.

Producers Counters Table (PCT); Consumers Counters Table (CCT)

The layouts for the 32-entry Producers Counters Table (PCT) and the 32-entry Consumers Counters Table (CCT) are shown in FIG. 5.

Ready-to-Run Queue (RRQ)

The layout for the 32-entry Ready-to-Run Queue (RRQ) is shown in FIG. 10.

Reserved—Bits [15:5]—These bits are not assigned at this time, and reading the RRQ results in zeros being returned for Bits [15:5].

Task Number—Bits [4:0]—The K-node can PEEK/POKE the 32-entry by 5-bit table for diagnostics purposes.

State Information Table (SIT)

The layout for the 32-entry State Information Table (SIT) is shown in FIG. 11. The 32-entry SIT is initialized by the K-node. This includes setting the initial value for the Ports_counter, the STATE_bit to zero, and the GO_bit=0. Thereafter, the K-node activates any of up to 32 tasks by setting its GO bit=1. The K-node de-activates any of up to 32 tasks by setting its GO_bit=0.

Prior to issuing an ABORT command, the K-node should clear the GO_bit of the task that is being aborted.

Bit 15, the GO_bit, is a READ/WRITE bit.

Bits [12:5] are unassigned at this time. For WRITE operations, they are don't care, and for READ operations, zeros will be returned for these fields.

When the SIT is written with Bit 13 (STATE Write Enable) set to 1, the STATE Bit for the associated task is set to the value indicated by Bit [14]. When Bit 13 is set to zero, there is no operation. For READ operations, the current STATE Bit for the associated task is returned for Bit [14], and a zero is returned for Bit 13.

When the SIT is written with Bit 4 (Ports_counter Write Enable) set to 1, the Ports_counter for the associated task is set to the value indicated by Bits [3:0]. When Bit 4 is set to zero, there is no operation. For READ operations, the current value of Ports_counter for the associated task is returned for Bits [3:0], and a zero is returned for Bit 4.

State transitions for a task are summarized in the table shown in FIG. 12. Note that for each of the (up to) 32 tasks, the K-node can resolve merged READY/RUN status by comparing any of 32 task numbers with the current task number which is available in the Node Status Register, NSR[4:0].

MDL Pointer Table (MPT)

The layout for the 32-entry Module Parameter List (MPL) Pointer Table (MPT) is shown in FIG. 13. Associated with each task is a register that contains the physical address in a contiguous chunk of memory where the first of the task's tbd configuration parameters is stored.

Because there are unresolved issues associated with aggregating memories/tiles/tasks, we indicate a 16-bit memory pointer (assuming longword address boundaries) which would allow the task to access its configuration information from any memory within its quadrant.

Parms Memory Layouts

Each task's Module Parameter List (MPL) will be stored in a contiguous chunk of memory referenced by its associated Parms Pointer. The numbers of parameters and their purposes will vary from one task to another. As tasks are designed, their specific requirements for configuration parameters (and their associated layouts) will be determined and documented.

An example of packing eight parameters associated with each task buffer is shown in FIG. 14.

Forward/Backward Acknowledgement Message Formats

Data formats for the Forward and Backward Acknowledgement Messages are shown in FIG. 15.

Although the invention has been described with respect to specific embodiments, thereof, these embodiments are merely illustrative, and not restrictive of the invention. For example, any type of processing units, functional circuitry or collection of one or more units and/or resources such as memories, I/O elements, etc., can be included in a node. A node can be a simple register, or more complex, such as a digital signal processing system. Other types of networks or interconnection schemes than those described herein can be employed. It is possible that features or aspects of the present invention can be achieved in systems other than an adaptable system, such as described herein with respect to a preferred embodiment.

Thus, the scope of the invention is to be determined solely by the appended claims.

Claims

1. An integrated circuit comprising: a plurality of computing nodes;a memory in at least one of the plurality of computing nodes, the plurality of computing nodes configured to make memory requests for access to the memory;an interconnection network operatively coupled to the plurality of computing nodes, the interconnection network providing interconnections among the plurality of computing nodes to route the memory requests;means for identifying a set of memory requests;means for determining when all of the memory requests in the set of memory requests have been performed, wherein the means for determining when all of the memory requests in the set of memory requests have been performed is a ports counter; andmeans for initiating execution of a task when all of the memory requests in the set of memory requests have been performed.

2. The computing device of claim 1, wherein the plurality of computing nodes comprise a plurality of heterogeneous computing nodes.

3. The computing device of claim 1, wherein the task is associated with the set of memory requests.

4. The computing device of claim 1, wherein the memory requests are memory write requests.

5. The computing device of claim 1, wherein the task is associated with a plurality of buffers and wherein the memory requests are requests for an availability of the plurality of buffers.

6. The computing device of claim 1, wherein the interconnection network further routes data for the memory requests between the plurality of heterogeneous computing nodes and the memory via a memory controller and routes data among the plurality of heterogeneous computing nodes.

7. The computing device of claim 1 wherein the means for identifying a set of memory requests is a plurality of forward acknowledgement messages and wherein each memory request in the set of memory requests is associated with a port number and each forward acknowledgement message includes a port number and a task number.

8. The computing device of claim 1, wherein the means for initiating execution of a task when all of the memory requests in the set of memory requests have been performed is a ready-to-run queue.

9. An integrated circuit comprising: a plurality of computing nodes configured to make memory requests for access to a memory in at least one of the plurality of computing nodes;an interconnection network operatively coupled to the plurality of computing nodes, the interconnection network providing interconnections among the plurality of computing nodes to route the memory requests; anda task manager in the at least one of the computing nodes configured to identify a set of memory requests, determine when all of the memory requests in the set of memory requests have been performed and initiate execution of a task when all of the memory requests in the set of memory requests have been performed, wherein the task manager comprises a ports counter and a ready-to-run queue.

10. The computing device of claim 9, wherein the plurality of computing nodes comprise a plurality of heterogeneous computing nodes.

11. The computing device of claim 9, wherein the interconnection network is reconfigurable.

12. The computing device of claim 9, wherein the memory requests are memory write requests.

13. The computing device of claim 9, wherein the task is associated with a plurality of buffers and wherein the memory requests are requests for an availability of the plurality of buffers.

14. The computing device of claim 13 wherein the interconnection network is reconfigurable.

15. The integrated circuit of claim 9, wherein the task manager determines when all of the memory requests have been performed with the port counter.

16. The integrated circuit of claim 9, wherein the task is moved to the ready-to-run queue to initiate execution of the task.

17. A method for managing tasks within an integrated circuit having a plurality of computing nodes configured to make memory requests for access to a memory in at least one of the plurality of computing nodes, and an interconnection network operatively coupled to the plurality of computing nodes, the interconnection network providing interconnections among the plurality of computing nodes and the memory to route the memory requests, said method comprising: identifying a set of memory requests, wherein identifying a set of memory requests comprises identifying a port number and a task number for each memory request;determining when all of the memory requests in the set of memory requests have been performed; andinitiating execution of a task when all of the memory requests in the set of memory requests have been performed.

18. The method of claim 17, wherein the plurality of computing nodes comprise a plurality of heterogeneous computing nodes.

19. The method of claim 17, wherein the interconnection network is reconfigurable.

20. The method of claim 17, wherein the memory requests are memory write requests.

21. The method of claim 17, wherein the task is associated with a plurality of buffers and wherein the memory requests are requests for an availability of the plurality of buffers.

22. The method of claim 17, wherein the indicating when all of the memory requests in the set of memory requests have been performed and initiating execution of a task when all of the memory requests in the set of memory requests have been performed is performed by a task manager.