Abstract: Systems and methods are provided to perform multiply-accumulate operations of multiple data types in a systolic array to increase clock speeds and/or reduce the size and quantity of systolic arrays required to perform multiply-accumulate operations of multiple data types. Each processing element in the systolic array can have a shared multiplier and one or more adders. The shared multiplier can have a separate and/or a shared circuitry where the shared circuitry is capable of performing at least a part of integer multiplication and at least a part of non-integer multiplication. The one or more adders can be a shared adder or separate adders. The shared adder can have a separate and a shared circuitry wherein the shared circuitry is capable of performing at least a part of integer addition and at least a part of non-integer addition.

Abstract: Programmatically defined fields of metadata for a network packet may be generated. Instructions indicating different portions of data from different headers of a network packet may be stored at a packet processor. When a network packet is received, the different portions of the data may be extracted from the different headers of the packet according to the instructions and provided to other stages of the packet processor for processing. Different portions of the same programmatically defined field may be utilized at different stages in the packet processor. The programmatically defined field may be used to generate a hash value that selects an entry in a lookup table describing a forwarding decision for a network packet.

Abstract: Systems and methods for power analysis of a hardware device design. In various examples, a target circuit can be defined within the hardware device design. The target circuit can include a plurality of digital circuit elements linking a plurality of input nodes with a plurality of output nodes. A solver can be used to search for a transition pattern that, when applied to the input nodes, causes a number of output nodes equal to a counter to transition from a first binary value to a second binary value. If a transition pattern cannot be found, the counter is decremented and a new transition pattern is searched for. Once a transition pattern is found, it is determined whether the transition pattern satisfies a constraint.

Abstract: An integrated circuit configured to execute program instructions can generate, based on a configuration, any combination of a notification message, a halt signal, or an interrupt signal for a condition detected in the integrated circuit. The detected condition can be an error condition or a non-error condition. The notification message for the condition may be written to memory accessible by a host processor. The non-error condition may be used by the host processor to monitor internal states of the integrated circuit. The halt signal may be used to stop the integrated circuit from executing the instructions.

Abstract: Operational management of an integrated circuit device can be performed by a microcontroller based on information associated with the notification messages generated by the integrated circuit device. The notification messages may include timestamps and metadata for different notification types which can be used to build a timeline. The microcontroller may use the information to monitor the operational health and performance of the integrated circuit device or can communicate this information to a remote management server.

Abstract: Disclosed herein are techniques for performing neural network computations. In one embodiment, an apparatus includes an array of processing elements, the array having configurable dimensions. The apparatus further includes a controller configured to set the dimensions of the array of processing elements based on at least one of: a first number of input data sets to be received by the array, or a second number of output data sets to be output by the array.

Abstract: Server computers often include one or more input/output (I/O) devices for communicating with a network or directly attached storage device. The data transfer latency for request can be reduced by utilizing ingress data placement logic to bypass the processor of the I/O device. For example, host memory descriptors can be stored in a memory of the I/O device to facilitate placement of the requested data.

Abstract: Programmatically defined fields of metadata for a network packet may be generated. Instructions indicating different portions of data from different headers of a network packet may be stored at a packet processor. When a network packet is received, the different portions of the data may be extracted from the different headers of the packet according to the instructions and provided to other stages of the packet processor for processing. Different portions of the same programmatically defined field may be utilized at different stages in the packet processor. The programmatically defined field may be used to generate a hash value that selects an entry in a lookup table describing a forwarding decision for a network packet.

Abstract: Disclosed herein are techniques for balancing and reducing the number of write operations performed to each physical memory page of a storage-class memory. In one embodiment, a method includes tracking a count of write operations performed to each physical memory page or subpage of the storage-class memory using a memory management unit, a memory controller, a hypervisor, or an operating system, and selectively allocating physical memory pages of the storage-class memory with the least counts of write operations to a virtual machine or an operating system process using a ranking of the physical memory pages of the storage-class memory determined based at least partially on the count of write operations performed to each physical memory page or subpage of the storage-class memory.

Abstract: In various implementations, a memory controller for storage class memory can include an address scrambling circuit. The address scrambling circuit can receive an input address for a processor memory transaction, where the input address is associated with a virtual machine identifier. The address scrambling circuit can further determine an address scrambling mapping from the plurality of address scrambling mappings, where the address scrambling mapping includes a first pattern that determines an alternate set of bits for a set of input bits. The address scrambling circuit can further scramble, using the scrambling circuit and the first pattern, a first part of the input address. The address scrambling circuit can further determine a scrambled address using the input address and the scrambled first part of the input address and output the scrambled address.

Abstract: Provided are integrated circuit devices and methods for operating integrated circuit devices. In various examples, an integrated circuit device can be operable to determine, at a point in time during operation of the integrated circuit device, to generate a notification. The notification can include a type and a timestamp indicating the point in time. The notification can also include information about an internal status of the integrated circuit at the point in time. The device can further selectin a queue from a plurality of queues in a processor memory of the computing system that includes the integrated circuit. The device can further generate a write transaction including the notification, where the write transaction is addressed to the queue. The device can further output the write transaction using a communication interface of the device.

Abstract: Disclosed are techniques for implementing packet checkers and packet generators within a network device. The packet checkers and packet generators can each operate in an internal mode to test functionality of the network device or in an external mode to test functionality of an external device.

Abstract: Provided are systems and methods for packet policing for controlling the rate of packet flows. In some implementations, an integrated circuit is provided. The integrated circuit may comprise a memory, a counter, and a pipeline. The integrated circuit may be operable to, upon receiving packet information describing a packet, determine, using the pipeline, a drop status for the packet. Determining the drop status may include determining a previous number of credits available, a number of new credits available, a current number of credits available, and a number of credits needed to transmit the packet. The drop status may be determined by comparing the number of credits needed to transmit the packet against the current number of credits available. The integrated circuit may further update the information stored for a policing context in the memory based on the drop status and the number of credits needed to transmit the packet.

Abstract: Disclosed herein are techniques associated with a Direct Memory Access (DMA) engine that can include a data generation module. The DMA engine can receive, from a processing entity, a particular type of write command with an indicator to write a data pattern to an address. Upon receipt of the particular type of write command, the DMA engine can generate, using a data generation module of the DMA engine, the data pattern to be written to the address. The processing entity can be Central Processing Unit (CPU) including a core configured to process serial commands. The DMA engine can be disposed on a same die as a processing entity or a network interface port.

Abstract: A computer system has a memory configured for sharing data between a first application and a second application. The memory includes a metadata region and a data region. The metadata region includes metadata that indicates how data being communicated between the first application and the second application is to be interpreted. The metadata also indicates whether the data can be found in the metadata itself or in a particular location in the data region. Each application can be assigned its own memory location containing a flag that can be set in order to indicate to the other application that the memory is ready to be accessed by the other application. The memory location can be implemented using a hardware register or in memory, either the same memory that includes the metadata and data regions or on a separate memory.

Abstract: Disclosed herein are techniques for debugging the performance of a neural network. In one embodiment, a neural network processor includes a processing engine, a debugging circuit coupled to the processing engine, and an interface to a memory device. The processing engine is configured to execute instructions for implementing a neural network. The debugging circuit is configurable to determine, for each instruction in a set of instructions, a first timestamp indicating a start time of executing the instruction and a second timestamp indicating an end time of executing the instruction by the processing engine. The interface is configured to save the first timestamp and the second timestamp for each instruction in the set of instructions into the memory device. The debugging circuit can be configured to different debug levels. The neural network processor can include multiple debugging circuits for multiple processing engines that operate in parallel.

Abstract: Systems and methods are provided to skip multiplication operations with zeros in processing elements of the systolic array to reduce dynamic power consumption. A value of zero can be detected on an input data element entering each row of the array and respective zero indicators may be generated. These respective zero indicators may be passed to all the processing elements in the respective rows. The multiplication operation with the zero value can be skipped in each processing element based on the zero indicators, thus reducing dynamic power consumption.

Abstract: Disclosed herein are methods and apparatuses related to the use of counter tables. A counter table can comprise a plurality of lower-level counters and an upper-level counter. A range of values capable of being represented by a lower-level counter from the plurality of lower-level counters can be enlarged by associating the lower-level counter with the upper-level counter. A counter table can be associated with a network device.

Abstract: Provided are systems and methods for an address translation circuit for a memory controller. In various implementations, the address translation circuit includes an address translation table operable to include a subset of address translations for a processor memory. An address translation memory can include all address translations for the processor memory. The address translation circuit can be operable to receive an input address for a transaction to processor memory. The address translation circuit can determine an index for the address translation table by inputting the input address into a hash function. The address translation circuit can read a row from the address translation table using the index. The address translation circuit can determine whether an entry in the row includes an address translation for the input address. The address translation can generate and output a translated address using the address translation.

Abstract: Provided are integrated circuit devices and methods for operating integrated circuit devices. In various examples, an integrated circuit device can include a memory for storing instructions a configuration register, and an instruction execution circuit. An instruction read from the memory can be a reconfigurable instruction. which includes a set of fields corresponding to a plurality of operations. Values in the fields can determine whether the operations are enabled or disabled. For example, a first value in a first field can enable a first operation. Whether the first operation is performed can further be determined by comparing a second value in a second field to a third value read from the configuration register. The value set in the configuration register thus can control whether the operation is performed.