Abstract: One embodiment of the present invention sets forth a technique for adaptively compressing video frames. The technique includes encoding a first plurality of video frames based on a first video compression algorithm to generate first encoded video frames and transmitting the first encoded video frames to a client device. The technique further includes receiving a user input event, switching from the first video compression algorithm to a second video compression algorithm in response to the user input event, encoding a second plurality of video frames based on the second video compression algorithm to generate second encoded video frames, and transmitting the second encoded video frames to the client device.

Abstract: A frequency divider includes first circuitry, second circuitry, and third circuitry. The first circuitry includes divide-by-two (div2) frequency divider circuitry, and the second circuitry includes additional circuitry for a divide-by-three (div3) frequency divider. The second circuitry is selectively enabled using a control signal and can receive signals from the first circuitry when enabled. Specifically, the second circuitry is enabled in the div3 mode but is not enabled in the div2 mode. The third circuitry receives signals from the first circuitry and also receives signals from the second circuitry when the second circuitry is enabled. The first circuitry and the third circuitry function as a div2 frequency divider when the second circuitry is not enabled. The first circuitry, the second circuitry, and the third circuitry function as a div3 frequency divider when the second circuitry is enabled.

Abstract: One embodiment of the present invention sets forth a technique for broadcasting composited game content. The technique includes executing an application program to generate a first video frame of game content, and causing the first video frame to be displayed on a primary display device. The technique further includes compositing at least one graphical user interface (GUI) element with the first video frame to generate a composited frame. The technique further includes causing the composited frame to be displayed on an external display device by transmitting the composited frame to the external display device via a local display interface.

Abstract: A system and method are provided for implementing multi-stage translation of virtual addresses. The method includes the steps of receiving, at a first memory management unit, a memory request including a virtual address in a first address space, translating the virtual address to generate a second virtual address in a second address space, and transmitting a modified memory request including the second virtual address to a second memory management unit. The second memory management unit is configured to translate the second virtual address to generate a physical address in a third address space. The physical address is associated with a location in a memory.

Abstract: One embodiment of the present invention includes techniques to support demand paging across a processing unit. Before a host unit transmits a command to an engine that does not tolerate page faults, the host unit ensures that the virtual memory addresses associated with the command are appropriately mapped to physical memory addresses. In particular, if the virtual memory addresses are not appropriately mapped, then the processing unit performs actions to map the virtual memory address to appropriate locations in physical memory. Further, the processing unit ensures that the access permissions required for successful execution of the command are established. Because the virtual memory address mappings associated with the command are valid when the engine receives the command, the engine does not encounter page faults upon executing the command. Consequently, in contrast to prior-art techniques, the engine supports demand paging regardless of whether the engine is involved in remedying page faults.

Abstract: A system, method, and computer program product for warming a cache for a task launch is described. The method includes the steps of receiving a task data structure that defines a processing task, extracting information stored in a cache warming field of the task data structure, and, prior to executing the processing task, generating a cache warming instruction that is configured to load one or more entries of a cache storage with data fetched from a memory.

Abstract: A method for light transport includes steps of initializing a data structure that is configured to provide an importance value for each incident sample in a three-dimensional (3D) scene and tracing, in a direction from an origin, a ray of a plurality of rays through the 3D scene to intersect an object at a hitpoint. Additional steps include selecting a next direction of the ray according to a distribution of the importance values at the hitpoint, tracing the ray in the next direction to find a next hitpoint, updating a first importance value corresponding to the hitpoint using a second importance value corresponding to the next hitpoint, and setting the hitpoint of the ray to the next hitpoint. The additional steps are repeated until the next hitpoint is an endpoint. A contribution, based on each hitpoint and the endpoint, to a pixel that is intersected by the ray is recorded.

Abstract: A method, computer readable medium, and system are disclosed for image processing to reduce aliasing using a temporal anti-aliasing algorithm modified to implement variance clipping. The method includes the step of generating a current frame of image data in a memory. Then, each pixel in the current frame of image data is processed by: sampling a resolved pixel color for a corresponding pixel in a previous frame of image data stored in the memory, adjusting the resolved pixel color based on a statistical distribution of color values for a plurality of samples in the neighborhood of the pixel in the current frame of image data to generate an adjusted pixel color, and blending a color value for the pixel in the current frame of image data with the adjusted pixel color to generate a resolved pixel color for the pixel in the current frame of image data.

Abstract: The disclosure provides a micro-processing system operable in a hardware decoder mode and in a translation mode. In the hardware decoder mode, the hardware decoder receives and decodes non-native ISA instructions into native instructions for execution in a processing pipeline. In the translation mode, native translations of non-native ISA instructions are executed in the processing pipeline without using the hardware decoder. The system includes a code portion profile stored in hardware that changes dynamically in response to use of the hardware decoder to execute portions of non-native ISA code. The code portion profile is then used to dynamically form new native translations executable in the translation mode.

Abstract: A method for regulating voltage for a processor is disclosed. The method comprises requesting a target frequency value, wherein the target frequency value determines a target clock frequency for clocking the processor. The method also comprises comparing the target clock frequency to a first signal to generate an error signal. Further, the method comprises using the error signal to generate a duty cycle control signal, wherein the duty cycle control signal is operable to generate a periodic waveform. Finally, the method comprises generating an output regulator voltage using the periodic waveform, wherein the output voltage is operable to provide power to the processor.

Abstract: A system, method, and computer program product are provided for rendering at variable sampling rates. Vertex coordinates for 3D primitive are received from a shader execution unit, and an arithmetic operation is performed on the vertex coordinates by fixed operation circuitry to produce modified vertex coordinates in homogeneous coordinate space. The modified vertex coordinates are transformed from homogeneous coordinate space into screen-space to produce screen-space vertex coordinates of a transformed 3D primitive and the transformed 3D primitive is rasterized in screen-space using the screen-space vertex coordinates to produce an image for display.

Abstract: Saving power in a mobile terminal includes determining alignment processing moments after the mobile terminal enters a standby mode. Alignable wakeup events, which occur during alignment processing periods corresponding to each alignment processing moment, are thus controlled to commence related processing at each of the alignment processing moments. Power consumption caused by various wakeup events in a standby mode may thus be reduced and battery life of the mobile terminal may thus be improved.

Abstract: A method for detecting an instruction ordering violation in a CPU. The method includes receiving a reordered stream of instructions and detecting whether an ordering violation has occurred by using virtual addresses. The method further includes transferring results of the reordered stream of instructions from a load store buffer into a cache and detecting whether an ordering violation has occurred by using physical addresses. Subsequently, a recovery is initiated upon detection of an ordering violation.

Abstract: Embodiments of the invention provides an IC system in which low-power chips can be positioned vertically proximate high-power chips without suffering the effects of overheating. In one embodiment, the IC system includes a first substrate, a high-power chip disposed on a first side of the first substrate, a thermal conductive pad disposed on a second side of the first substrate, one or more thermal conductive features formed in the first substrate, wherein the thermal conductive features thermally connect the high-power chip and the thermal conductive pad, and a heat sink attached to a surface of the thermal conductive pad, wherein the heat sink is in thermal communication with the thermal conductive pad. By having thermal conductive features formed through the first substrate to thermally connect the high-power chip and the thermal conductive pad, heat generated by the high-power chip can be effectively dissipated into the heat sink.

Abstract: A method, computer program product, and system for sparse convolutional neural networks that improves efficiency is described. Multi-bit data for input to a processing element is received at a compaction engine. The multi-bit data is determined to equal zero and a single bit signal is transmitted from the memory interface to the processing element in lieu of the multi-bit data, where the single bit signal indicates that the multi-bit data equals zero. A compacted data sequence for input to a processing element is received by a memory interface. The compacted data sequence is transmitted from the memory interface to an expansion engine. Non-zero values are extracted from the compacted data sequence and zeros are inserted between the non-zero values by the expansion engine to generate an expanded data sequence that is output to the processing element.

Abstract: A raster unit is configured to generate different sample patterns for adjacent pixels within a given frame. In addition, the raster unit may adjust the sample patterns between frames. The raster unit includes an index unit that selects a sample pattern table for use with a current frame. For a given pixel, the index unit extracts a sample pattern from the selected sample pattern table. The extracted sample pattern is used to generate coverage information for the pixel. The coverage information for all pixels is then used to generate an image. The resultant image may then be filtered to reduce or remove artifacts induced by the changing of sample locations.

Abstract: Techniques are provided for restoring threads within a processing core. The techniques include, for a first thread group included in a plurality of thread groups, executing a context restore routine to restore from a memory a first portion of a context associated with the first thread group, determining whether the first thread group completed an assigned function, and, if the first thread group completed the assigned function, then exiting the context restore routine, or if the first thread group did not complete the assigned function, then executing one or more operations associated with a trap handler routine.

Abstract: A multi-threaded processing unit includes a hardware pre-processor coupled to one or more processing engines (e.g., copy engines, GPCs, etc.) that implement pre-emption techniques by dividing tasks into smaller subtasks and scheduling subtasks on the processing engines based on the priority of the tasks. By limiting the size of the subtasks, higher priority tasks may be executed quickly without switching the context state of the processing engine. Tasks may be subdivided based on a threshold size or by taking into account other consideration such as physical boundaries of the memory system.

Abstract: A graphics processing subsystem includes one or more memory devices and two or more graphics processing units (GPU). The graphics processing units each include a memory interface. A first sub-set of the memory interface of the first graphics processing unit communicatively couples the first graphics processing unit to the first memory device. A first sub-set of the memory interface of the second graphics processing unit is connected to a second sub-set of the memory interface of the first graphics processing unit.

Abstract: One embodiment of the present disclosure sets forth an effective way to maintain fairness and order in the scheduling of common resource access requests related to replay operations. Specifically, a streaming multiprocessor (SM) includes a total order queue (TOQ) configured to schedule the access requests over one or more execution cycles. Access requests are allowed to make forward progress when needed common resources have been allocated to the request. Where multiple access requests require the same common resource, priority is given to the older access request. Access requests may be placed in a sleep state pending availability of certain common resources. Deadlock may be avoided by allowing an older access request to steal resources from a younger resource request. One advantage of the disclosed technique is that older common resource access requests are not repeatedly blocked from making forward progress by newer access requests.