Abstract: We report methods, integrated circuit devices, and fabrication processes relating to power management transitions of multiple compute units sharing a resource. One method include, in response to an indication that a first compute unit of a plurality of compute units is attempting to enter a normal power state and in response to no other compute units being in a low power state, causing a resource to enter the normal power state, wherein the plurality of compute units share the resource; and causing the first compute unit to enter the normal power state.

Abstract: Current computer systems support sleep states such as sleep state S3 and sleep state S4. A system in sleep state S3 utilizes more power than one in sleep state S4, however, a system in sleep state S3 can resume function substantially faster than a system in sleep state S4. An idle system is often put into sleep state S3 rather than sleep state S4 because of the shorter resume time even though sleep state S3 utilizes more power. Embodiments include a reduced-power sleep state S3 that uses less power than sleep state S3 yet resumes function faster than sleep state S4. Embodiments reduce the power consumed by compressing and consolidating system context to fewer memory modules, and powering down unused memory modules. Embodiments thus avoid storing system content to non-volatile memory. Embodiments include waking the system by restoring system context in the reverse order to respective memory modules.

Abstract: A device includes a dock generator operable to generate a clock signal. A first module includes a first clock network coupled to the clock generator for distributing the clock signal. A second module includes a second clock network coupled to the clock generator for distributing the clock signal. A clock skew control circuit is operable to receive a first instance of the clock signal from the first clock network and a second instance of the clock signal from the second clock network and to control skew between the first and second instances of the clock signal.

Abstract: Embodiments of a software video encoder with GPU acceleration include a software encoder that partitions video processing tasks and assigns them to both a graphics processing unit (GPU) and a central processing unit (CPU). The partitioning and assigning is configurable for operation in different modes. The modes include a mode in which the total time for video processing (such as when transcoding a large existing file) is reduced, a mode in which less CPU cycles are consumed, thus freeing the CPU for other work, a mode in which the latency of processing (e.g., for video conferencing) is reduced, and a mode in which information from a game or other real-time activity being displayed on the screen is encoded.

Abstract: A method of regulating power states in a processing system may begin with a processor component reporting a present processor power state to an input-output hub, where the present processor power state corresponds to one of a plurality of different processor power states ranging from an active state to an inactive state. The input-output hub receives data indicative of the present processor power state and, in response to receiving the present processor power state, establishes a lowest allowable hub power state that corresponds to one of a plurality of different hub power states ranging from an active state to an inactive state. The method continues by determining a present hub power state for the input-output hub, wherein depth of the present hub power state is less than or equal to depth of the lowest allowable hub power state.

Abstract: A method, system, and computer program product are provided for adjusting write timing in a memory device based on results of an error detection function. For instance, the method can include determining a write timing window between a signal on a data bus and a write clock signal based on the results of the error detection function. The method can also include adjusting a phase difference between the signal on the data bus and the write clock signal based on the write timing window. The memory device can recover data on the data bus based on the adjusted phase difference.

Abstract: A system and method for efficient management of operating modes within an IC for optimal power and performance targets. On a same die, an SOC includes one or more processing units and a input/output (I/O) controller (IOC). The multiple interfaces within the IOC manage packets and messages according multiple different protocols. The IOC maintains an activity level for each one of the multiple interfaces. This activity level may be based at least on a respective number of transactions executed by a corresponding one of the multiple interfaces. The IOC determines a power estimate for itself based on at least the activity levels. In response to detecting a difference between the power estimate and an assigned I/O power limit for the IOC, a power manager adjusts at least respective power limits for the one or more processing units based on at least the difference.

Abstract: Methods and apparatuses are provided for power control in a processor. The apparatus comprises a plurality of operational units arranged as a group of operational units. A power consumption monitor determines when cumulative power consumption of the group of operational units exceeds a threshold (e.g., either or both of the cumulative power threshold and the cumulative power rate threshold) during a time interval, after which a filter for issuing instructions to the group of operational units suspends instruction issuance to the group of operational units for the remainder of the time interval. The method comprises monitoring cumulative power consumption by a group of operational units within a processor over a time interval. If the cumulative power consumption of the group of operational units exceeds the threshold, instruction issuance to the group of operational units is suspended for the remainder of the time interval.

Abstract: The described embodiments comprise a computing device with a first processor core and a second processor core. In some embodiments, during operations, the first processor core receives, from the second processor core, an indication of a memory location and a flag. The first processor core then stores the flag in a first cache line in a cache in the first processor core and stores the indication of the memory location separately in a second cache line in the cache. Upon encountering a predetermined result when evaluating a condition for the indicated memory location, the first processor core updates the flag in the first cache line. Based on the update of the flag, the first processor core causes the second processor core to perform an operation.

Abstract: The present invention provides a method and apparatus for adapting aggressiveness of a pre-fetcher in a processor-based system. One embodiment includes modifying a rate for pre-fetching data from a memory into one or more caches by comparing a first address of a memory access request to addresses in an address window that includes one or more previously fetched addresses and one or more addresses to be fetched.

Abstract: A processing unit that includes a plurality of virtual engines and a shader core. The plurality of virtual engines is configured to (i) receive, from an operating system (OS), a plurality of tasks substantially in parallel with each other and (ii) load a set of state data associated with each of the plurality of tasks. The shader core is configured to execute the plurality of tasks substantially in parallel based on the set of state data associated with each of the plurality of tasks. The processing unit may also include a scheduling module that schedules the plurality of tasks to be issued to the shader core.

Abstract: According to one exemplary embodiment, a method for forming a photoresist pattern on a semiconductor wafer includes forming a photoresist including an organic polymer matrix on the semiconductor wafer. The method further includes exposing the photoresist to a patterned radiation. The method further includes baking the photoresist after exposing the photoresist to the pattern radiation. The method further includes applying an oxidizing reagent to the photoresist to create the photoresist pattern corresponding to the patterned radiation.

Abstract: A single interconnect is provided between a first processor and a second processor, such that the first processor may access a common memory through the second processor while the second processor can be mostly powered off. The first processor accesses the memory through a memory controller using a standard dynamic random access memory (DRAM) bus protocol. Instead of the memory controller directly connecting to the memory, the access path is through the second processor to the memory. Additionally, a bidirectional communication protocol bus is mapped to the existing DRAM bus signals. When both the first processor and the second processor are active, the bus protocol between the processors switches from the DRAM protocol to the bidirectional communication protocol. This enables the necessary chip-to-chip transaction semantics without requiring the additional cost burden of a dedicated interface for the bidirectional communication protocol.

Abstract: A device receives input that includes definitions of components of a computational pipeline, where the components include one or more buffers, one or more kernels, and one or more stages within a control graph. The device generates, based on the input, kernel signatures for a graphics processor, where the kernel signatures compile into an executable streaming program for the computational pipeline. The device also generates, based on the input, host-side runtime code to execute the streaming program.

Abstract: A content addressable memory (CAM) suppresses an indication of a match in response to determining that the entry that stores data matching received compare data is the subject of a write operation. To suppress the indication, an address decoder decodes a write address associated with the write operation to determine the entry of the CAM that is the subject of the write operation, and provides control signaling indicative of the determined entry. The CAM uses the control signaling to suppress any match indications for the entry being written, thereby preventing erroneous match indications.

Abstract: A system and method are disclosed for managing memory requests that are coordinated between a system memory controller and a graphics memory controller. Memory requests are pre-scheduled according to the optimization policies of the source memory controller and then sent over the CPU/GPU boundary in a bundle of pre-scheduled requests to the target memory controller. The target memory controller then processes pre-scheduling decisions contained in the pre-schedule requests, and in turn, issues memory requests as a proxy of the source memory controller. As a result, the target memory controller does not need to perform both CPU requests and GPU requests.

Abstract: Methods, systems, and computer readable media embodiments are disclosed for generating primitives in a grid. Embodiments include generating a set of vertices in a section of the grid, selecting one or more vertices in the set of vertices in an order based on a proximity of the vertices to a boundary edge of the grid, and generating primitives based on the order of the selected vertices.

Abstract: A clock driver for a resonant clock network includes a delay circuit that receives and supplies a delayed clock signal. A first transistor is coupled to receive a first pulse control signal and supply an output clock node of the clock driver. An asserted edge of the first control signal is responsive to the falling edge of the delayed clock signal. A second transistor is coupled to receive a second control signal and to supply the output clock node of the clock driver. An asserted edge of the second control signal is responsive to a rising edge of the delayed clock signal.

Abstract: A processor employs a prefetch prediction module that predicts, for each prefetch request, whether the prefetch request is likely to be satisfied from (“hit”) the cache. The arbitration priority of prefetch requests that are predicted to hit the cache is reduced relative to demand requests or other prefetch requests that are predicted to miss in the cache. Accordingly, an arbiter for the cache is less likely to select prefetch requests that hit the cache, thereby improving processor throughput.

Abstract: A processor includes storage elements to store a first and second value, as well as a plurality of hash units coupled to the storage elements. Each hash unit performs a hash operation using the first value and the second value to generate a corresponding hash result value. The processor further includes selection logic to select a hash result value from the hash result values generated by the plurality of hash units responsive to a selection input generated from another hash operation performed using the first value and the second value. A method includes predicting whether a branch instruction is taken based on a prediction value stored at an entry of a branch prediction table indexed by an index value selected from a plurality of values concurrently generated from an address value of the branch instruction and a branch history value representing a history of branch directions at the processor.