Abstract: Methods and systems for opportunistic load balancing in deep neural networks (DNNs) using metadata. Representative computational costs are captured, obtained or determined for a given architectural, functional or computational aspect of a DNN system. The representative computational costs are implemented as metadata for the given architectural, functional or computational aspect of the DNN system. In an implementation, the computed computational cost is implemented as the metadata. A scheduler detects whether there are neurons in subsequent layers that are ready to execute. The scheduler uses the metadata and neuron availability to schedule and load balance across compute resources and available resources.

Abstract: Systems, apparatuses, and methods for implementing a decoupled crossbar for a stream processor are disclosed. In one embodiment, a system includes at least a multi-lane execution pipeline, a vector register file, and a crossbar. The system is configured to determine if a given instruction in an instruction stream requires a permutation on data operands retrieved from the vector register file. The system conveys the data operands to the multi-lane execution pipeline on a first path which includes the crossbar responsive to determining the given instruction requires a permutation on the data operands. The crossbar then performs the necessary permutation to route the data operands to the proper processing lanes. Otherwise, the system conveys the data operands to the multi-lane execution pipeline on a second path which bypasses the crossbar responsive to determining the given instruction does not require a permutation on the input operands.

Abstract: Systems, apparatuses, and methods for implementing stereoscopic interleaved compression techniques are disclosed. A system includes a transmitter sending a video stream over a wireless link to a receiver. For each pair of frames, the transmitter encodes a left-half of a first frame of the pair with an amount of compression less than a first threshold and encodes a right-half of the first frame with an amount of compression greater than a second threshold. For a second frame of the pair, the transmitter encodes a right-half of the second frame with an amount of compression less than the first threshold and encodes a left-half of the second frame with an amount of compression greater than the second threshold. The transmitter conveys encoded half-frames and indications of an amount of compression used for each half-frame to a receiver. The receiver receives, decodes, and drives the encoded half-frames to a display.

Abstract: Systems, apparatuses, and methods for sharing an field programmable gate array compute engine are disclosed. A system includes one or more processors and one or more FPGAs. The system receives a request, generated by a first user process, to allocate a portion of processing resources on a first FPGA. The system maps the portion of processing resources of the first FPGA into an address space of the first user process. The system prevents other user processes from accessing the portion of processing resources of the first FPGA. Later, the system detects a release of the portion of the processing resources on the first FPGA by the first user process. Then, the system receives a second request to allocate the first FPGA from a second user process. In response to the second request, the system maps the first FPGA into an address space of the second user process.

Abstract: Described herein are techniques for saving registers in the event of a function call. The techniques include modifying a program including a block of code designated as a calling code that calls a function. The modifying includes modifying the calling code to set a register usage mask indicating which registers are in use at the time of the function call. The modifying also includes modifying the function to combine the information of the register usage mask with information indicating registers used by the function to generate registers to be saved and save the registers to be saved.

Abstract: A technique for determining an adaptive quantization parameter offset for a block of encoded video includes obtaining a rate control factor for the quantization parameter, determining a content-based quantization parameter factor for the quantization parameter, determining an adaptive variance based quantization offset based on content-based quantization parameter factors for a frame prior to the current frame, and combining the rate control factor, the content-based quantization parameter factor, and the adaptive offset to generate the quantization parameter.

Abstract: A reconfigurable optic probe is used to measure signals from a device under test. The reconfigurable optic probe is positioned at a target probe location within a cell of the device under test. The cell including a target net to be measured and non-target nets. A test pattern is applied to the cell and a laser probe (LP) waveform is obtained in response. A target net waveform is extracted from the LP waveform by: i) configuring the reconfigurable optic probe to produce a ring-shaped beam having a relatively low-intensity region central to the ring-shaped beam; (ii) re-applying the test pattern to the cell at the target probe location with the relatively low-intensity region applied to the target net and obtaining a cross-talk LP waveform in response; (iii) normalizing the cross-talk LP waveform; and (iv) determining a target net waveform by subtracting the normalized cross-talk LP waveform from the LP waveform.

Abstract: A technique for synchronizing workgroups is provided. The techniques comprise detecting that one or more non-executing workgroups are ready to execute, placing the one or more non-executing workgroups into one or more ready queues based on the synchronization status of the one or more workgroups, detecting that computing resources are available for execution of one or more ready workgroups, and scheduling for execution one or more ready workgroups from the one or more ready queues in an order that is based on the relative priority of the ready queues.

Abstract: An electronic device includes a queue with multiple sub-queues arranged in a logical hierarchy from a lowest sub-queue to a highest sub-queue, each sub-queue including a separate subset of a set of entries of the queue, and a separate age matrix for each sub-queue. The electronic device also includes a controller that stores items in entries in the lowest sub-queue until the lowest sub-queue is full and then stores items in successively next higher sub-queues in the hierarchy. The controller also removes an item that is ready for removal from an entry in the lowest sub-queue. The controller then shifts items in sub-queues in the hierarchy to fill the vacancy in the lowest sub-queue. For the shifting, the controller uses an age matrix associated with each sub-queue to determine an oldest item in that sub-queue and then moves the oldest item to a next lower sub-queue in the hierarchy.

Abstract: Systems, apparatuses, and methods for implementing hypervisor post-write notification of processor state register modifications. A write to a state register of the processor may be detected during guest execution. In response to detecting the write to the state register, the processor may trigger microcode to perform the write and copy the new value of the register to a memory location prior to exiting the guest. The hypervisor may be notified of the update to the state register after it occurs, and the hypervisor may be prevented from modifying the value of the guest's state register. The hypervisor may terminate the guest if the update to the state register is unacceptable. Alternatively, the hypervisor may recommend an alternate value to the guest. If the guest agrees, the guest may set the state register to the alternate value recommended by the hypervisor when the guest resumes operation.

Abstract: Techniques for scheduling processing tasks in a device having multiple computing elements are disclosed. A network interface controller of the device receives processing tasks, for execution on the computing elements, from a network that is external to the device. The network interface controller schedules the tasks for execution on the computing devices based on policy data available to the network interface controller. A scheduler within the network interface controller, which can be implemented as a standalone processing unit (such as a microcontroller, a programmable processing core, or an application specific integrated circuit), performs such scheduling, thereby freeing the central processing unit of the device from the burden of performing scheduling operations. The scheduler schedules the tasks according to any technically feasible scheduling technique.

Abstract: A processor core is configured to execute a parent task that is described by a data structure stored in a memory. A coprocessor is configured to dispatch a child task to the at least one processor core in response to the coprocessor receiving a request from the parent task concurrently with the parent task executing on the at least one processor core. In some cases, the parent task registers the child task in a task pool and the child task is a future task that is configured to monitor a completion object and enqueue another task associated with the future task in response to detecting the completion object. The future task is configured to self-enqueue by adding a continuation future task to a continuation queue for subsequent execution in response to the future task failing to detect the completion object.

Abstract: An optic probe is used to measure signals from a device under test. The optic probe is positioned at a target probe location within a cell of the device under test, the cell including a target net to be measured and a plurality of non-target nets. A test pattern is applied to the cell with the optic probe a laser probe (LP) waveform is obtained in response. A target net waveform is extracted from the LP waveform by: (i) simulating a combinational logic analysis (CLA) cross-talk waveform to model cross-talk from selected non-target nets by simulating an optical response of the cell to the test pattern with the target net masked; (ii) estimating a cross-talk weight; and (iii) determining a target net waveform by weighting the CLA cross-talk waveform according to the cross-talk weight and subtracting the weighted CLA cross-talk waveform from the LP waveform.

Abstract: Methods, devices, and systems for testing a number of combinations of memory in a computer system. A modular memory device is installed in a memory channel in communication with a processor. The modular memory device includes a number of memory storage devices. The number of memory storage devices include a number of pins. For each of a number of subsets of the number of memory storage devices, a subset of the number of memory storage devices is selected, each pin of a subset of the number of pins which do not correspond to the subset of the number of memory storage devices is configured with a termination impedance, and the subset of the number of memory storage devices is tested.

Abstract: Methods and systems are described. A system includes a redundant shader pipe array that performs rendering calculations on data provided thereto and a shader pipe array that includes a plurality of shader pipes, each of which performs rendering calculations on data provided thereto. The system also includes a circuit that identifies a defective shader pipe of the plurality of shader pipes in the shader pipe array. In response to identifying the defective shader pipe, the circuit generates a signal. The system also includes a redundant shader switch. The redundant shader switch receives the generated signal, and, in response to receiving the generated signal, transfers the data for the defective shader pipe to the redundant shader pipe array.

Abstract: A technique for operating a processor that includes multiple cores is provided. The technique includes determining a number of active applications, selecting a processor configuration for the processor based on the number of active applications, configuring the processor according to the selected processor configuration, and executing the active applications with the configured processor.

Abstract: A system, method and a computer program product are provided for hybrid rendering with deferred primitive batch binning A primitive batch is generated from a sequence of primitives. Initial bin intercepts are identified for primitives in the primitive batch. A bin for processing is identified. The bin corresponds to a region of a screen space. Pixels of the primitives intercepting the identified bin are processed. Next bin intercepts are identified while the primitives intercepting the identified bin are processed.

Abstract: An integrated circuit device wafer includes a silicon wafer substrate and a back side metallization structure. The back side metallization structure includes a first adhesion layer on the back side of the substrate, a first metal later over the first adhesion layer, a second metal layer over the first metal layer, and a second adhesion layer over the second metal layer. The first includes at least one of: silicon nitride and silicon dioxide. The first metal layer includes titanium. The second metal layer includes nickel. The second adhesion layer includes at least one of: silver, gold, and tin. An indium preform is placed between the second adhesion layer and the lid and the indium preform is reflowed.

Abstract: The computer system responds to a first trigger event to enter a partial off state in which a boot cycle is required to return to a working state. A device plugged into a serial bus port can be charged in the partial off state. A configuration register or runtime environment controls whether the computer system enters the partial off state in response to a trigger event. The computer system stays in the partial off state until another trigger event returns the computer system to the working state. In some implementations, the computer system leaves the partial off state and enters the shutdown state after an unplug event, a predetermined amount of time after an unplug event, a predetermined amount of time after entering the partial off state, a predetermined amount of time after charging of a device is complete, or any combination of such events.

Abstract: Systems, apparatuses, and methods for dynamically adjusting the power consumption of prefetch engines are disclosed. In one embodiment, a processor includes one or more prefetch engines, a draw completion engine, and a queue in between the one or more prefetch engines and the draw completion engine. If the number of packets stored in the queue is greater than a high watermark, then the processor reduces the power state of the prefetch engine(s). By decreasing the power state of the prefetch engine(s), power consumption is reduced. Additionally, this power consumption reduction is achieved without affecting performance, since the queue has a high occupancy and the draw completion engine can continue to read packets out of the queue. If the number of packets stored in the queue is less than a low watermark, then the processor increases the power state of the prefetch engine(s).