Abstract: A representative system, apparatus, method and protocol are disclosed for data communication between chiplets or SOCs on a common interposer. A representative system comprises: an interposer; a first integrated circuit arranged on the interposer, the first integrated circuit comprising a first common protocol interface circuit; a communication link coupled to the first common protocol interface circuit; and a second integrated circuit arranged on the interposer, the second integrated circuit comprising a second common protocol interface circuit coupled to the communication link to form a serial protocol interface between the first common protocol interface circuit and the second common protocol interface circuit. Serial data and control packets and parallel data and control packets having specified, ordered fields are also disclosed.

Abstract: A system includes multiple memory-compute nodes coupled to one another over a scale fabric, where each memory-compute node includes a hybrid threading processor; a memory controller; a fabric interface; and a network on chip (NOC) that provides communication between the hybrid threading processor, the fabric interface, and the memory controller, wherein the fabric interface supports a first virtual channel (VC0), and a second virtual channel (VC1) to the NOC, and supports the first virtual channel (VC0), the second virtual channel (VC1), and a third virtual channel (VC2) to the scale fabric.

Abstract: Representative apparatus, method, and system embodiments are disclosed for a self-scheduling processor which also provides additional functionality. Representative embodiments include a self-scheduling processor, comprising: a processor core adapted to execute a received instruction; and a core control circuit adapted to automatically schedule an instruction for execution by the processor core in response to a received work descriptor data packet. In another embodiment, the core control circuit is also adapted to schedule a fiber create instruction for execution by the processor core, to reserve a predetermined amount of memory space in a thread control memory to store return arguments, and to generate one or more work descriptor data packets to another processor or hybrid threading fabric circuit for execution of a corresponding plurality of execution threads. Event processing, data path management, system calls, memory requests, and other new instructions are also disclosed.

Abstract: Representative apparatus, method, and system embodiments are disclosed for a self-scheduling processor which also provides additional functionality. Representative embodiments include a self-scheduling processor, comprising: a processor core adapted to execute a received instruction; and a core control circuit adapted to automatically schedule an instruction for execution by the processor core in response to a received work descriptor data packet. In another embodiment, the core control circuit is also adapted to schedule a fiber create instruction for execution by the processor core, to reserve a predetermined amount of memory space in a thread control memory to store return arguments, and to generate one or more work descriptor data packets to another processor or hybrid threading fabric circuit for execution of a corresponding plurality of execution threads. Event processing, data path management, system calls, memory requests, and other new instructions are also disclosed.

Abstract: A synthetic-aperture radar (SAR) antenna emits radar pulses and receives their reflections. SAR is typically used on a moving platform, such as an aircraft, drone, or spacecraft. Since the position of the antenna changes between the time of emitting a radar pulse and receiving the reflection of the pulse, the synthetic aperture of the radar is increased, giving greater accuracy for a same (physical) sized radar over conventional beam-scanning radar. The pulse data is processed, using a backprojection algorithm, to generate a two-dimensional image that can be used for navigation. The order in which the SAR data is processed can impact the likelihood of cache hits in accessing the data. Since accessing data from cache instead of memory storage reduces both access time and power consumption, devices that access more data from cache have greater battery life and range.

Abstract: Disclosed in some examples are methods, systems, devices, and machine-readable mediums which utilize a pool method whereby a host process executing on the host processor reserves one or more pools of memory for worker threads of the host process. Upon creation of a new thread corresponding to the host process, the worker processor executing the new thread may assign a portion of the previously reserved pool to the new thread. By giving some control to a worker processor to assign memory from a previously reserved pool, threads may be assigned memory resources without additional message overhead from the host processor to the worker processor while at the same time retaining overall memory control with the host processor.

Abstract: A dispatch element interfaces with a host processor and dispatches threads to one or more tiles of a hybrid threading fabric. Data structures in memory to be used by a tile may be identified by a starting address and a size, included as parameters provided by the host. The dispatch element sends a command to a memory interface to transfer the identified data to the tile that will use the data. Thus, when the tile begins processing the thread, the data is already available in local memory of the tile and does not need to be accessed from the memory controller. Data may be transferred by the dispatch element while the tile is performing operations for another thread, increasing the percentage of operations performed by the tile that are performing useful work and reducing the percentage that are merely retrieving data.

Abstract: Devices and techniques for triggering early termination of cooperating processes in a processor are described herein. A system includes multiple memory-compute nodes, wherein a memory-compute node comprises: event manager circuitry configured to establish a broadcast channel to receive event messages; and thread manager circuitry configured to organize a plurality of threads to perform portions of a cooperative task, wherein the plurality of threads each monitor the broadcast channel to receive event messages on the broadcast channel, and wherein upon achieving a threshold operation, the thread manager circuitry is to use the event manager circuitry to broadcast, on the broadcast channel, an event message indicating that the cooperative task is complete, causing other threads, in response to receiving the event message, to terminate execution of their respective portions of the cooperative task.

Abstract: Disclosed in some examples are methods, systems, computing devices, and machine-readable mediums in which the system maintains a list of resources available for each rollback session. In some examples, state data is kept that indicates available memory. If a write occurs for a particular session and the amount of available memory for a session has been used, a flag is set in metadata for the memory location and the write is not mirrored. In this manner, the technical problem of one undo logging session using too much memory and preventing other undo logging sessions from properly functioning is solved by the technical solution of setting resource limits for each undo logging session.

Abstract: Representative apparatus, method, and system embodiments are disclosed for configurable computing. A representative system includes an asynchronous packet network having a plurality of data transmission lines forming a data path transmitting operand data; a synchronous mesh communication network; a plurality of configurable circuits arranged in an array, each configurable circuit of the plurality of configurable circuits coupled to the asynchronous packet network and to the synchronous mesh communication network, each configurable circuit of the plurality of configurable circuits adapted to perform a plurality of computations; each configurable circuit of the plurality of configurable circuits comprising: a memory storing operand data; and an execution or write mask generator adapted to generate an execution mask or a write mask identifying valid bits or bytes transmitted on the data path or stored in the memory for a current or next computation.

Abstract: A reconfigurable compute fabric of a system can include multiple nodes, and each node can include multiple, communicatively coupled tiles with respective processing and storage elements. In an example, a tile-based processor can be configured to perform operations comprising receiving a first stencil that defines input data for a first operation. The stencil can have a height corresponding to N rows in a main memory and a stencil width corresponding to M columns in the main memory. The processor can perform operations comprising establishing N buffers in a tile memory, each buffer having M buffer elements, and populating the M buffer elements of the N buffers using respective information, defined by the first stencil, from the main memory. Tile-based stencil operations can use information from the N buffers and provide compute results in an output array.

Abstract: A three-dimensional stacked integrated circuit (3D SIC) having a non-volatile memory die having an array of non-volatile memory partitions, a volatile memory die having an array of volatile memory partitions, and a processing logic die having an array of processing logic partitions. The non-volatile memory die, the volatile memory die, and the processing logic die are stacked. The non-volatile memory die, the volatile memory die, and the processing logic die can be arranged to form an array of functional blocks, and at least two functional blocks can each include a different data processing function that reduces the computation load of a controller.

Abstract: Devices and techniques for non-blocking external device calls are described herein. Specifically, when a processor receives an instruction with a no-return indication from a thread for a device, the processor can increase a counter corresponding to the thread based on the no-return indication. The processor can then continue execution of the thread without waiting for a return value from the device. When a return value is received for the instruction, the processor can decrement the counter. While the counter is not zero, the processor prevents the thread from completing (exiting).

Abstract: Disclosed in some examples are methods, systems, and machine-readable mediums that provide a memory allocation mechanism that evenly spreads the allocations for an application over all the MCs on the system, thus minimizing congestion and resulting in optimal application performance.

Abstract: Disclosed in some examples are methods, systems, memory devices, and machine-readable mediums that allows an application of a computer system to create a series of one or more logs of writes to one or more memory locations of a memory device. The logs may comprise the values at the end of the log interval of the one or more memory locations that were written to during a log interval. In some examples, the logs do not include intermediate writes to the one or more memory locations (only the final value) and do not include values of memory locations that were not written to during the interval. After an event, software can apply these logs to a copy of the original memory region state to recover the contents of the locations at any of the logged points. These logs may be useful to recreate the state of the memory at various points during the application's execution.

Abstract: Methods, systems, and devices for bias control for a memory device are described. A memory system may store indication of whether data is coherent. In some examples, the indication may be stored as metadata, where a first value indicates that the data is not coherent and a second value or a third value indicate that the data is coherent. When a processing unit or other component of the memory system processes a command to access data, the memory system may operate according to a device bias mode when the indication is the first value, and according to a host bias mode when the indication is the second value or the third value.

Abstract: Disclosed in some examples are methods, systems, devices, and machine-readable mediums that use parallel hardware execution with software co-simulation to enable more advanced debugging operations on data flow architectures. Upon a halt to execution of a program thread, a state of the tiles that are executing the thread are saved and offloaded from the HTF to a host system. A developer may then examine this state on the host system to debug their program. Additionally, the state may be loaded into a software simulator that simulates the HTF hardware. This simulator allows for the developer to step through the code and to examine values to find bugs.

Abstract: Systems, apparatuses, and methods related to modified checksum data using a poison data indictor. An example method can include receiving a first set of bits including data and a second set of at least one bit indicating whether the first set of bits includes one or more erroneous or corrupted bits. A first checksum can be generated that is associated with the first set of bits. A second checksum can be generated using the first checksum and the second set of at least one bit. The first set of bits and the second checksum can be written to an array of a memory device. A comparison of the first checksum and the second checksum can indicate whether the first set of bits includes the at least one or more erroneous or corrupted bits.

Abstract: A three-dimensional stacked integrated circuit (3D SIC) for implementing an artificial neural network (ANN) having a memory die having an array of memory partitions. Each partition of the array of memory partitions is configured to store parameters of a set of neurons. The 3D SIC also has a processing logic die having an array of processing logic partitions. Each partition of the array of processing logic partitions is configured to: receive input data, and process the input data according to the set of neurons to generate output data.

Abstract: Devices and techniques to avoid deadlock in a multi-SOC fabric are described herein. An apparatus comprises a network on chip (NOC) interface to receive on a first virtual channel, from a processor, a memory request for a memory controller; a fabric interface configured to include the first virtual channel and a second virtual channel, connected to a scale fabric; and circuitry to: transmit the memory request toward the memory controller on the first virtual channel via the scale fabric; receive a response from the memory controller over the second virtual channel via the scale fabric; and relay the response to the processor over the second virtual channel.