Abstract: Tasks of directed acyclic graphs (DAGs) may be dynamically scheduled based on a plurality of constraints and conditions, task prioritization policies, task execution estimates, and configurations of a heterogenous system. A machine learning component may be initialized to dynamically schedule the tasks of the DAGs.

Abstract: Techniques facilitating hardware-based memory-error mitigation for heap-objects. In one example, a system can comprise a process that executes computer executable components stored in a non-transitory computer readable medium. The computer executable components comprise: an entry component; and a re-purpose component. The entry component can allocate an entry in a table to store bounds-information when an object is allocated in memory. The re-purpose component can re-purpose unused bits of an object address to store an index to the table entry.

Abstract: A method, system and apparatus for protecting against out-of-bounds references, including storing an address of a buffer in a general register and storing bounds information (BI) for the buffer in a bounds information register, and when a content of the general register is used as an address in a load or store operation, using a content of the bounds information register to determine if the load or store is out of bounds.

Abstract: Embodiments include in response to monitoring a processor during operation, detecting a first number of core recovery events in the processor, determining that the first number of core recovery events fulfills a first condition for the first core recovery events threshold, and modifying a value of at least one droop sensor parameter of the processor by a first amount. The at least one droop sensor parameters affects a sensitivity to a voltage droop. In response to modifying the value of the droop sensor parameter by the first amount, a second number of core recovery events is detected in the processor. It is determined that the second number of core recovery events fulfills a second condition for a second core recovery events threshold, and the value of the at least one droop sensor parameter is modified by a second amount.

Abstract: Embodiments include exerciser device placement in the development of an integrated circuit. Aspects of the invention include obtaining a design of an integrated circuit and creating a dynamic power blockage map for the integrated circuit. Aspects also include updating the integrated circuit design by placing one or more exercisers on the integrated circuit, wherein a location of the one or more exercisers on the integrated circuit is based on at least in part on the dynamic power blockage map. Based on a determination that the updated integrated circuit design complies with one or more design constraints, aspects further include outputting the updated integrated circuit design.

Abstract: A method, system and apparatus for protecting against out-of-bounds references, including storing an address of a buffer in a general register and storing bounds information (BI) for the buffer in a bounds information register, and when a content of the general register is used as an address in a load or store operation, using a content of the bounds information register to determine if the load or store is out of bounds.

Abstract: A first type memory and a second type memory may be identified in a computing system. The second type memory is slower than the first type memory while having a greater storage capacity compared to the first type memory. An application process executing in the computing system may be identified. A region of the first type memory may be provided as a cache of the second type memory for the application process.

Abstract: Various embodiments are provided herein for clearing memory of system in a computing environment. A zero-filled cache line with a single z-bit per entry in the cache directory may be defined. The “z” is a positive integer. A plurality of instruction set architecture (“ISA”) instructions are provided with a single z-bit in a cache line as defined in a cache directory to clear an entire cache line.

Abstract: Embodiments of a method are disclosed. The method includes performing distributed deep learning training on a batch of training data. The method also includes determining training times representing an amount of time between a beginning batch time and an end batch time. Further, the method includes modifying a communication aspect of the communication straggler to reduce a future network communication time for the communication straggler to send a future result of the distributed deep learning training on a new batch of training data in response to the centralized parameter server determining that the learner is the communication straggler.

Abstract: A method, system, and computer program product for augmenting cache replacement operations are provided. The method identifies a set of cache lines within a first cache level of a multilevel cache. A first candidate cache line is identified based on a first replacement scheme of the first cache level. A second candidate cache line is identified based on the first replacement scheme of the first cache level. A replacement cache line is selected for replacement in the first cache level. The replacement cache line is selected from the first candidate cache line and the second candidate cache line and based on the first replacement scheme of the first cache level and a second replacement scheme of a second cache level. The method removes the replacement cache line from the first cache level.

Abstract: Embodiments include in response to monitoring a processor during operation, detecting a first number of throttling amounts in the processor, determining that the first number of throttling amounts fulfills a first condition regarding a throttling amounts threshold, and modifying a voltage level of the processor by a first amount. Embodiments include in response to modifying the voltage level of the processor by the first amount, detecting a second number of throttling amounts in the processor, determining that the second number of throttling amounts fulfills a second condition regarding the throttling amounts threshold, and modifying the voltage level of the processor by a second amount.

Abstract: Embodiments include in response to monitoring a processor during operation, detecting a first number of core recovery events in the processor, determining that the first number of core recovery events fulfills a first condition for the first core recovery events threshold, and modifying a value of at least one droop sensor parameter of the processor by a first amount. The at least one droop sensor parameters affects a sensitivity to a voltage droop. In response to modifying the value of the droop sensor parameter by the first amount, a second number of core recovery events is detected in the processor. It is determined that the second number of core recovery events fulfills a second condition for a second core recovery events threshold, and the value of the at least one droop sensor parameter is modified by a second amount.

Abstract: Embodiments of a method are disclosed. The method includes performing decentralized distributed deep learning training on a batch of training data. Additionally, the method includes determining a training time wherein the learner performs the decentralized distributed deep learning training on the batch of training data. Further, the method includes generating a table having the training time and other processing times for corresponding other learners performing the decentralized distributed deep learning training on corresponding other batches of other training data. The method also includes determining that the learner is a straggler based on the table and a threshold for the training time. Additionally, the method includes modifying a processing aspect of the straggler to reduce a future training time of the straggler for performing the decentralized distributed deep learning training on a new batch of training data in response to determining the learner is the straggler.

Abstract: An example system includes a processor to prune a machine learning model based on an importance of neurons or weights. The processor is to further permute and pack remaining neurons or weights of the pruned machine learning model to reduce an amount of ciphertext computation under a selected constraint.

Abstract: Computation, placement, and accessing of error correcting codes (ECC) in a computer system data cache enabling partial reads and writes to each line of data in the cache. For storing multiple compressed blocks, received at differing time periods, in a single cache line, the ECC for the first compressed block is stored in the ECC field of the cache and the ECC for the second and subsequently received compressed blocks is appended to the compressed data. Additionally, an auxiliary ECC cache may be constructed for temporarily holding a partial ECC for a partial read/write, and a new ECC for the partial read/write is computed using the partial ECC.

Abstract: Systems, computer-implemented methods, and computer program products to facilitate updating, such as averaging and/or training, of one or more statistical sets are provided. According to an embodiment, a system can comprise a memory that stores computer executable components and a processor that executes the computer executable components stored in the memory. The computer executable components can include a computing component that averages a statistical set, provided by the system, with an additional statistical set, that is compatible with the statistical set, to compute an averaged statistical set, where the additional statistical set is obtained from a selected additional system of a plurality of additional systems. The computer executable components also can include a selecting component that selects the selected additional system according to a randomization pattern.

Abstract: The method and systems described herein provide for identifying and mitigating undesirable power or voltage fluctuations in regions of a semiconductor device. For example, embodiments include detecting a region, such as an individual processor, of a processor chip is exhibiting a reduced power draw and a resulting localized voltage spike (e.g., a spike that exceeds Vmax) that would accelerate overall device end-of-life (EOL). The described systems respond by activating circuits or current generators located in the given region to draw additional power via a protective current. The protective current lowers the local voltages spikes back to within some pre-specified range (e.g., below a Vmax). The resulting reduction in the time above Vmax in testing reduces the number of devices that will need to be discarded due to Vmax violations as well as increases the expected reliability and lifespan of the device in operation.

Abstract: A system includes a determination component that determines output for successively larger neural networks of a set; and a consensus component that determines consensus between a first neural network and a second neural network of the set. A linear chain of increasingly complex neural networks trained on progressively larger inputs is utilized (e.g., increasingly complex neural networks is generally representative of increased accuracy). Outputs of progressively networks are computed until a consensus point is reached—where two or more successive large networks yield a same inference output. At such point of consensus the larger neural network of the set reaching consensus can be deemed appropriately sized (or of sufficient complexity) for a classification task at hand.

Abstract: A multi-purpose server cache directory in a computing environment is provided. One of a plurality of operation modes may be selectively enabled or disabled, by a cache directory, based on a computation phase, data type, and data pattern for caching data in a cache having a plurality of address tags in the cache directory greater than a number of data lines in a cache array.

Abstract: Embodiments relate to a system, program product, and method for mitigating voltage overshoot in one or more cores in a multicore processing device including a plurality of cores. The method includes determining, in real-time, an indication of power consumption within each core of the one or more cores. The method also includes determining, through the indication of power consumption, a voltage overshoot condition in the one or more cores. The method further includes increasing, for the one or more cores, a power demand thereof. The method also includes increasing, subject to the increasing the power demand, power delivery to the one or more cores, thereby at least arresting the rate of increase of the voltage overshoot.