Abstract: A technique for performing convolution operations is disclosed. The technique includes performing a first convolution operation based on a first convolutional layer input image to generate at least a portion of a first convolutional layer output image; while performing the first convolution operation, performing a second convolution operation based on a second convolutional layer input image to generate at least a portion of a second convolutional layer output image, wherein the second convolutional layer input image is based on the first convolutional layer output image; storing the portion of the first convolutional layer output image in a first memory dedicated to storing image data for convolution operations; and storing the portion of the second convolutional layer output image in a second memory dedicated to storing image data for convolution operations.

Abstract: A memory sprint controller, responsive to an indicator of an irregular memory access phase, causes a memory controller to enter a sprint mode in which it temporarily adjusts at least one timing parameter of a dynamic random access memory (DRAM) to reduce a time in which a designated number of activate (ACT) commands are allowed to be dispatched to the DRAM.

Abstract: A processing device is provided which comprises memory configured to store data and a processor configured to receive a portion of data of a first matrix comprising a first plurality of elements and receive a portion of data of a second matrix comprising a second plurality of elements. The processor is also configured to determine values for a third matrix by dropping a number of products from products of pairs of elements of the first and second matrices based on approximating the products of the pairs of elements as a sum of the exponents of the pairs of elements and performing matrix multiplication on remaining products of the pairs of elements of the first and second matrices.

Abstract: Systems and methods of cache flushing include receiving, from a software application, a first cache flush request to perform a range-based cache flush of a contiguous virtual address range within a virtual memory that maps to a physical memory. A single cache walk is triggered via a second cache flush request to a cache. The single cache walk performs the range-based cache flush for the contiguous physical address range from a beginning address of the contiguous physical address range to an ending address of the contiguous physical address range in response to the first cache flush request.

Abstract: Devices and methods for multi-resolution geometric representation for ray tracing are described which include casting a ray in a space comprising objects represented by geometric shapes and approximating a volume of the geometric shapes using an accelerated hierarchy structure. The accelerated hierarchy structure comprises first nodes each representing a volume of one of the geometric shapes in the space and second nodes each representing an approximate volume of a group of the geometric shapes. When the ray is determined to intersect a bounding box of a second node representing one group of the geometric shapes, a selection is made between traversal and non-traversal of other second nodes based on a LOD for representing the volume of the one group of geometric shapes.

Abstract: A technique for manipulating a generic tensor is provided. The technique includes receiving a first request to perform a first operation on a generic tensor descriptor associated with the generic tensor, responsive to the first request, performing the first operation on the generic tensor descriptor, receiving a second request to perform a second operation on generic tensor raw data associated with the generic tensor, and responsive to the second request, performing the second operation on the generic tensor raw data.

Abstract: Disclosed is a system and method for use in a cache for suppressing modification of cache line. The system and method includes a processor and a memory operating cooperatively with a cache controller. The memory includes a coherence directory stored within a cache created to track at least one cache line in the cache via the cache controller. The processor instructs a cache controller to store a first data in a cache line in the cache. The cache controller tags the cache line based on the first data. The processor instructs the cache controller to store a second data in the cache line in the cache causing eviction of the first data from the cache line. The processor compares based on the tagging the first data and the second data and suppresses modification of the cache line based on the comparing of the first data and the second data.

Abstract: A processing system divides successive dispatches of work items into portions. The successive dispatches are separated from each other by barriers, each barrier indicating that the work items of the previous dispatch must complete execution before work items of a subsequent dispatch can begin execution. In some embodiments, the processing system interleaves execution of portions of a first dispatch with portions of subsequent dispatches that consume data produced by the first dispatch. The processing system thereby reduces the amount of data written to the local cache by a producer dispatch while preserving data locality for a subsequent consumer (or consumer/producer) dispatch and facilitating processing efficiency.

Abstract: One or more rotated bounding volumes are generated for one or more nodes of a bounding volume hierarchy (BVH). Volume intersection ray tracing tests are then be performed using the rotated bounding volumes with the aim of reducing the number of calculations required relative to an original, non-rotated bounding volume. Rotated bounding volumes are selected from a plurality of candidate rotations, and selection of one of the candidate rotations are based on surface areas, such as minimum total surface areas, of bounding volumes corresponding to each of the candidate rotations. In order to minimize data storage and increase performance, a number of candidate rotations may be limited to a predetermined set of rotations.

Abstract: A technique for synchronizing workgroups is provided. Multiple workgroups execute a wait instruction that specifies a condition variable and a condition. A workgroup scheduler stops execution of a workgroup that executes a wait instruction and an advanced controller begins monitoring the condition variable. In response to the advanced controller detecting that the condition is met, the workgroup scheduler determines whether there is a high contention scenario, which occurs when the wait instruction is part of a mutual exclusion synchronization primitive and is detected by determining that there is a low number of updates to the condition variable prior to detecting that the condition has been met. In a high contention scenario, the workgroup scheduler wakes up one workgroup and schedules another workgroup to be woken up at a time in the future. In a non-contention scenario, more than one workgroup can be woken up at the same time.

Abstract: The disclosed computer-implemented method includes locating, from a processor storage, a partial tag corresponding to a memory request for a line stored in a memory having a tiered memory cache and in response to a partial tag hit for the memory request, locating, from a partition of the tiered memory cache indicated by the partial tag, a full tag for the line. The method also includes fetching, in response to a full tag hit, the requested line from the partition of the tiered memory cache. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: Techniques are provided for allocating registers for a processor. The techniques include identifying a first instruction of an instruction dispatch set that meets all register allocation suppression criteria of a first set of register allocation suppression criteria, suppressing register allocation for the first instruction, identifying a second instruction of the instruction dispatch set that does not meet all register allocation suppression criteria of a second set of register allocation suppression criteria, and allocating a register for the second instruction.

Abstract: Method and devices are provided for performing a physics-based simulation. A processing devices comprises memory and a processor. The processor is configured to perform a physics-based simulation by executing a portion of the physics-based simulation, training a neural network model based on results from executing the first portion of the physics-based simulation, performing inference processing based on the results of the training of the neural network model and providing a prediction, based on the inference processing, as an input back to the physics-based simulation.

Abstract: A method for server level cooling can include providing a printed circuit board and attaching a cooling system to the printed circuit board. The cooling system can be configured for placement thereon of two or more expansion cards having back side power delivery components. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: A technique for performing ray tracing operations is provided. The technique includes traversing a bounding volume hierarchy for a ray to arrive at a bounding box without use of a neural network; perform a feature vector lookup using modified polar coordinates characterizing the ray relative to the bounding box to obtain a set of feature vectors; and obtaining output with the neural network using the set of feature vectors.

Abstract: Some implementations provide systems methods and devices for integrated circuit self-test. The integrated circuit includes interface circuitry configured to read test data and to write the test data into memory of the integrated circuit. The integrated circuit also includes test circuitry configured to test the interface circuitry based on the test data written into memory of the integrated circuit. Some implementations provide an integrated circuit configured for storing and reading data. The integrated circuit includes circuitry configured to write or read a first portion of data to or from a first memory via BIST circuitry of the first memory until a first BIST counter saturates. The integrated circuit also includes circuitry configured to write or read a second portion of the data to or from a second memory via BIST circuitry of the second memory until a second BIST counter saturates.

Abstract: The disclosed device includes multiple data elements each configured to send a bit of a bit sequence by toggling at most half of a number of bits from a previously sent bit sequence. The bit sequence can first be biased and then XORed with the previously sent bit sequence. Various other methods, systems, and computer-readable media are also disclosed.

Abstract: A technique for rendering is provided. The technique includes determining a first correspondence between a screen space and a shade space in a visibility pass; selecting a size for a shade space tile based on the correspondence between the screen space and the shade space; and shading the shade space tile based on the material texture correspondence.

Abstract: Selecting intermediate representation transformation for compilations is described. In accordance with the described techniques, source code is received to be compiled by a compilation system for execution by a processor of hardware. Intermediate representation transformations are selected for the source code based on system load information associated with the hardware. The intermediate representation transformations are output to the compilation system.

Abstract: A technique for texture filtering. A transition is made from a first mipmap corresponding to a first texture resolution to a second mipmap corresponding to a second texture resolution. The first texture resolution is lower than the second texture resolution. As compared to standard trilinear filtering, initiation of the transition is delayed by an offset (or bias), which serves to delay the initial use of the second mipmap until it has been loaded. Following initiation of the transition, first and second weightings are selected with a nonlinear filter, and the system interpolates between the first mipmap and the second mipmap by applying the weightings. During an initial portion of the transition, the nonlinear filter has a slope that is higher than that of the standard trilinear filter.