Abstract: Systems and methods are disclosed for maintaining insertion policies of a lower-level cache. Techniques are described for selecting, based on metadata of an evicted data block received from an upper-level cache, an insertion policy out of the insertion policies. Then, determining, based on the selected insertion policy, whether to insert the data block into the lower-level cache. If it is determined to insert, the data block is inserted into the lower-level cache according to the selected insertion policy. Techniques for dynamically updating the insertion policies of the lower-level cache are also disclosed.

Abstract: Methods are provided for creating objects in a way that permits an API client to explicitly participate in memory management for an object created using the API. Methods for managing data object memory include requesting memory requirements for an object using an API and expressly allocating a memory location for the object based on the memory requirements. Methods are also provided for cloning objects such that a state of the object remains unchanged from the original object to the cloned object or can be explicitly specified.

Abstract: A memory controller includes a command queue with multiple entry stacks, each with a plurality of entries holding memory access commands, one or more parameter indicators each holding a respective characteristic common to the plurality of entries, and a head indicator designating a current entry for arbitration. An arbiter has a single command input for each entry stack. A command queue loader circuit receives incoming memory access commands and loads entries of respective entry stacks with memory access commands having the respective characteristic of each of the one or more parameter indicators in common.

Abstract: Methods and systems are disclosed for executing a collaborative task in a shader system. Techniques disclosed include receiving, by the system, input data and computing instructions associated with the collaborative task, as well as a configuration setting, causing the system to operate in a takeover mode. The system then launches, exclusively in one workgroup processor, a workgroup including wavefronts configured to execute the collaborative task.

Abstract: An electronic device includes a memory that stores input matrices A and B, a cache memory, and a processor. The processor generates a compiled representation that includes values for acquiring data from input matrix A when processing instances of input data through the neural network, the values including a base address in input matrix A for each thread from among a number of threads and relative offsets, the relative offsets being distances between elements of input matrix A to be processed by the threads. The processor then stores, in the local cache memory, the compiled representation including the base address for each thread and the relative offsets.

Abstract: An operation combiner receives a series of commands with read addresses, a modification operation, and write addresses. In some cases, the commands have serial dependencies that limit the rate at which they can be processed. The operation combiner compares the addresses for compatibility, transforms the operations to break serial dependencies, and combines multiple source commands into a smaller number of aggregate commands that can be executed much faster than the source commands. Some embodiments of the operation combiner receive a first command including one or more first read addresses and a first write address. The operation combiner compares the first read addresses and the first write address to one or more second read addresses and a second write address of a second command stored in a buffer. The operation combiner selectively combines the first and second commands to form an aggregate command based on the comparison.

Abstract: A circuit and method are provided to monitor a clock for a data processor. The method includes receiving a clock signal and producing a first voltage proportional to a frequency of the clock signal. The first voltage is converted to a digital signal. During an initialization mode, the method ensures the clock signal is at a desired frequency and scales the digital signal using a first configurable ratio to produce a high threshold value. When changing from the initialization mode to an operating mode, the method ceases to scale the digital signal and maintains the high threshold value. During the operating mode, the method compares the digital signal to the high threshold value to determine if the clock signal has been increased in frequency beyond a desired level, and if so, triggers an overclock alert to a system management circuit of the data processor.

Abstract: An apparatus and method for efficiently determining a final camera lens position that captures a focused input image are described. An image signal processing system of a camera capable of performing automatic focus includes a camera lens, an image sensor, a focus engine, and a lens controller. Rather than generate a single contrast value based on digital signals corresponding to a single image, the focus engine uses at least two value generators to generate multiple contrast values. The value generators are bandpass filters with different bandwidths from one another. The focus engine uses the multiple contrast values, rather than from a single contrast value, to determine a search direction for finding a final lens position of the camera lens, when to use relatively large or coarse step sizes for updating the lens position, and when to use relatively small or fine step sizes for updating the lens position.

Abstract: Devices and methods for transitioning between power states of a device are provided. A program is executed using data stored in configuration registers assigned to a component of a device. For a first reduced power state, data of a first portion of the configuration registers is saved to the memory using a first set of linear address space. For a second reduced power state, data of a second portion of the configuration registers is saved to the memory using a second set of linear address space and data of a third portion of the configuration registers is saved to the memory using a third set of linear address space.

Abstract: Devices and methods for cache prefetching are provided. A device is provided which comprises a quality of service (QOS) component having first assigned registers used to store data to execute a program, a plurality of non-QOS components having second assigned registers used to store data to execute the program and a power management controller, in communication with the QOS component and the non-QOS components. The power management controller is configured to issue fences for the non-QOS components when it is determined that one or more of the non-QOS components are idle, issue a fence for the QOS component when the fences for the non-QOS components are completed and enter a reduced power state when the fences for the non-QOS components and the fence for the QOS component are completed.

Abstract: A handshake mechanism allows die discovery in a stacked die architecture that keeps inputs isolated until the handshake is complete. Power good indications are used as handshake signals between the die. A die keeps inputs isolated from above until a power good indication from the die above indicates presence of the die above. The die keeps inputs isolated from below until the die detects power is good and receives a power good indication from the die and the die below. In an implementation drivers and receivers, apart from configuration bus drivers and receivers are disabled until a fuse distribution done signal indicates that repairs have been completed. Drivers are then enabled and after a delay to ensure signals are driven, receivers are deisolated. A top die in the die stack never sees a power good indication from a die above and therefore keeps inputs from above isolated. That allows the height of the die stack to be unknown at power on.

Abstract: A first workload is executed in a first subset of pipelines of a processing unit. A second workload is executed in a second subset of the pipelines of the processing unit. The second workload is dependent upon the first workload. The first and second workloads are suspended and state information for the first and second workloads is stored in a first memory in response to suspending the first and second workloads. In some cases, a third workload executes in a third subset of the pipelines of the processing unit concurrently with executing the first and second workloads. In some cases, a fourth workload is executed in the first and second pipelines after suspending the first and second workloads. The first and second pipelines are resumed on the basis of the stored state information in response to completion or suspension of the fourth workload.

Abstract: Systems, apparatuses, and methods for efficient parallel execution of multiple work units in a processor by reducing a number of memory accesses are disclosed. A computing system includes a processor core with a parallel data architecture. The processor core executes a software application with matrix operations. The processor core supports the broadcast of shared data to multiple compute units of the processor core. A compiler or other code assigns thread groups to compute units based on detecting shared data among the compute units. Rather than send multiple read accesses to a memory subsystem for the shared data, the processor core generates a single access request. The single access request includes information to identify the multiple compute units for receiving the shared data when broadcasted by the processor core.

Abstract: A write masked latch bit cell of an SRAM includes a write mask circuit that is responsive to assertion of a first write mask signal to cause a value of a write data node to be a first value and is responsive to assertion of a second write mask signal to cause the value of the write data node to have a second value. A pass gate supplies the data on the write data node to an internal node of the bit cell responsive to write word line signals being asserted. A keeper circuit maintains the value of the first node independently of values of the write word line signals while the first write mask signal and the second write mask signal are deasserted.

Abstract: Devices, methods, and systems for secure communications on a computing device. A host operating system (OS) runs on a host processor in communication with a host memory. A secure OS runs on a coprocessor in communication with a secure memory. The coprocessor receives information from an external device over a secure peer-to-peer (P2P) connection. The secure P2P connection is managed by the secure OS and is not accessible by the host OS.

Abstract: In a system with a master processor and slave processors, sync points are used in boot instructions. While executing the boot instructions, the slave processor determines whether the sync point is enabled. In response to determining the sync point is enabled, the slave processor pauses execution of the boot instructions, waits for commands from the master processor, receives commands from the master processor, executes the received commands until a release command is received, and then continues to execute boot instructions. In response to determining the sync point is not enabled, the slave processor continues to execute boot instructions.

Abstract: A processing device determines a scope indicating at least a portion of the processing system and target data from atomic memory operation to be performed. Based on the scope, the processing device determines one or more hardware parameters for at least a portion of the processing system. The processing device then compares the hardware parameters to the scope and target data to determine one or more corrections. The processing device then provides the scope, target data, hardware parameters, and corrections to a plurality of hardware lookup tables. The hardware lookup tables are configured to receive the scope, target data, hardware parameters, and corrections as inputs and output values indicating one or more coherency actions and one or more orderings. The processing device then executes one or more of the indicated coherency actions and the atomic memory operation based on the indicated ordering.

Abstract: A processing device is provided which includes memory comprising data cache memory configured to store compressed data and metadata cache memory configured to store metadata, each portion of metadata comprising an encoding used to compress a portion of data. The processing device also includes at least one processor configured to compress portions of data and select, based on one or more utility level metrics, portions of metadata to be stored in the metadata cache memory. The at least one processor is also configured to store, in the metadata cache memory, the portions of metadata selected to be stored in the metadata cache memory, store, in the data cache memory, each portion of compressed data having a selected portion of corresponding metadata stored in the metadata cache memory. Each portion of compressed data, having the selected portion of corresponding metadata stored in the metadata cache memory, is decompressed.

Abstract: A method and apparatus of integrating memory stacks includes providing a first memory die of a first memory technology and a second memory die of a second memory technology. A first logic die is in communication with the first memory die of the first memory technology, and includes a first memory controller including a first memory control function for interpreting requests in accordance with a first protocol for the first memory technology. A second logic die is in communication with the second memory die of the second memory technology and includes a second memory controller including a second memory control function for interpreting requests in accordance with a second protocol for the second memory technology. A memory operation request is received at the first or second memory controller, and the memory operation request is performed in accordance with the associated first memory protocol or the second memory protocol.

Abstract: Embodiments include methods, systems and non-transitory computer-readable computer readable media including instructions for executing a prefetch kernel that includes memory accesses for prefetching data for a processing kernel into a memory, and, subsequent to executing at least a portion of the prefetch kernel, executing the processing kernel where the processing kernel includes accesses to data that is stored into the memory resulting from execution of the prefetch kernel.