Abstract: A system for processing gather and scatter instructions can implement a front-end subsystem, a back-end subsystem, or both. The front-end subsystem includes a prediction unit configured to determine a predicted quantity of coalesced memory access operations required by an instruction. A decode unit converts the instruction into a plurality of access operations based on the predicted quantity, and transmits the plurality of access operations and an indication of the predicted quantity to an issue queue. The back-end subsystem includes a load-store unit that receives a plurality of access operations corresponding to an instruction, determines a subset of the plurality of access operations that can be coalesced, and forms a coalesced memory access operation from the subset. A queue stores multiple memory addresses for a given load-store entry to provide for execution of coalesced memory accesses.

Abstract: A system for processing gather and scatter instructions can implement a front-end subsystem, a back-end subsystem, or both. The front-end subsystem includes a prediction unit configured to determine a predicted quantity of coalesced memory access operations required by an instruction. A decode unit converts the instruction into a plurality of access operations based on the predicted quantity, and transmits the plurality of access operations and an indication of the predicted quantity to an issue queue. The back-end subsystem includes a load-store unit that receives a plurality of access operations corresponding to an instruction, determines a subset of the plurality of access operations that can be coalesced, and forms a coalesced memory access operation from the subset. A queue stores multiple memory addresses for a given load-store entry to provide for execution of coalesced memory accesses.

Abstract: A system for updating an index into a tuple table of tuples is provided. An indexing system updates an index into a tuple table using fine-grain locking of the index. The index includes a values table with an entry for each index value of an index field that references a value-tuple table that includes, for each tuple with the index value, a row that identifies a tuple of the tuple table with that indexed value. After a new tuple is added to the tuple table with a value, the index is updated by locking the entry in the values table, updating the value-tuple table for the value, and then unlocking the entry. When the index is accessed for locating tuples with a value, the accessor locks the entry in the values table for the value, uses the value-tuple table to locate the tuples, and unlocks the entry.

Abstract: A parallelization assistant tool system to assist in parallelization of a computer program is disclosed. The system directs the execution of instrumented code of the computer program to collect performance statistics information relating to execution of loops within the computer program. The system provides a user interface for presenting to a programmer the performance statistics information collected for a loop within the computer program so that the programmer can prioritize efforts to parallelize the computer program. The system generates inlined source code of a loop by aggressively inlining functions substantially without regard to compilation performance, execution performance, or both. The system analyzes the inlined source code to determine the data-sharing attributes of the variables of the loop. The system may generate compiler directives to specify the data-sharing attributes of the variables.

Abstract: A printed circuit board includes additional stitching vias placed at strategic location within a connection matrix, which provides additional isolation and further accommodates high-speed communication capabilities. The stitching vias have a variable length or depth, depending on related structures within the circuit board, so as to avoid any interference with underlining escape routing, or alternative signal transmission structures. More specifically, these stitching vias help to eliminate cross-talk in the via field caused by the close proximity of signal carrying structures. Further, differential signal communication is better accommodated based upon this reduction in cross-talk.

Abstract: Computer systems and associated methods for cooling computer components are disclosed herein. One embodiment of a computer system includes a computer cabinet having an air inlet spaced apart from an air outlet. The computer system also includes heat exchangers positioned in the computer cabinet, and a heat removal system in fluid communication with the heat exchangers. The computer system additionally includes at least one sensor for monitoring heat transfer between the computer cabinet and the room. The computer system further includes a control system operatively coupled to the at least one sensor, the control system including a computer-readable medium holding instructions for determining whether heat transfer between the computer cabinet and the room is balanced based on information from the sensor, and if not, adjusting a parameter to balance the heat transfer.

Abstract: In a large scale computing system, a supervisory system is utilized to monitor the operations and requests of multiple components, and to manage such requests so that overall power considerations for the entire system are considered. The supervisory system has the ability to identify requests and aggregations of simultaneous requests that will create an adverse power effect, and to apply overall control methodologies which will help to minimize these adverse effects.

Abstract: A resiliency system detects and corrects memory errors reported by a memory system of a computing system using previously stored error correction information. When a program stores data into a memory location, the resiliency system executing on the computing system generates and stores error correction information. When the program then executes a load instruction to retrieve the data from the memory location, the load instruction completes normally if there is no memory error. If, however, there is a memory error, the computing system passes control to the resiliency system (e.g., via a trap) to handle the memory error. The resiliency system retrieves the error correction information for the memory location and re-creates the data of the memory location. The resiliency system stores the data as if the load instruction had completed normally and passes control to the next instruction of the program.

Abstract: A method for prefetching data into a cache is provided. The method allocates an outstanding request buffer (“ORB”). The method stores in an address field of the ORB an address and a number of blocks. The method issues prefetch requests for a degree number of blocks starting at the address. When a prefetch response is received for all the prefetch requests, the method adjusts the address of the next block to prefetch and adjusts the number of blocks remaining to be retrieved and then issues prefetch requests for a degree number of blocks starting at the adjusted address. The prefetching pauses when a maximum distance between the reads of the prefetched blocks and the last prefetched block is reached. When a read request for a prefetched block is received, the method resumes prefetching when a resume criterion is satisfied.

Abstract: A method and system for inferring facts in parallel in a multiprocessor computing environment is provided. An inference system infers facts by applying rules to a collection of existing facts. For each existing fact, the inference system schedules a thread to apply the rules to that existing fact. As a thread infers a new fact (i.e., one that is not already in the collection of facts), the thread adds that inferred fact to the collection of facts. When a thread adds a new fact to the collection, the thread also applies the rules to that new fact. After the threads complete execution, the inference system may apply the rules to the facts of the collection, including the newly inferred facts, by again launching a thread for each fact to apply the rules to that fact. The inference system performs this processing iteratively until a termination condition is satisfied.

Abstract: To eliminate the adverse effects of power swings in a large scale computing system during the life cycle of an application or job, control of several operating characteristics for the collective group of processors is provided. By providing certain levels of coordination for the many processors utilized in large scale computing systems, significant and abrupt changes in power needs can be avoided. In certain circumstances, this may involve limiting the transition between several C-States of the processors involved and the overall power transitions for a large scale system are not detrimental and do not create issues for the data center or local power utility. Some cases will require stepped transitions between C-States, while other cases will include both stepped and modulated transitions. Other cases will incorporate random wait times at the various transitions in order to spread the power consumption involved.

Abstract: A system is provided for allocating memory for data of a program for execution by a computer system with a multi-tier memory that includes LBM and HBM. The system accesses a data structure map that maps data structures of the program to the memory addresses within an address space of the program to which the data structures are initially allocated. The system executes the program to collect statistics relating to memory requests and memory bandwidth utilization of the program. The system determines an extent to which each data structure is used by a high memory utilization portion of the program based on the data structure map and the collected statistics. The system generates a memory allocation plan that favors allocating data structures in HBM based on the extent to which the data structures are used by a high memory utilization portion of the program.

Abstract: The various structures forming communication paths on a printed circuit board can create several undesired effects, especially when high frequency signals are considered. Non-functional pads created during the manufacturing process have the potential to create an undesired effect, but when the overall collection of non-functional pads are carefully configured, an optimized communication path can be formed. More specifically, by selectively removing some collection of the non-functional pads, the high frequency characteristics of the communication paths can be optimized.

Abstract: A system and algorithm configured to generate diversity at the traffic source so that packets are uniformly distributed over all of the available paths, but to increase the likelihood of taking a minimal path with each hop the packet takes. This is achieved by configuring routing biases so as to prefer non-minimal paths at the injection point, but increasingly prefer minimal paths as the packet proceeds, referred to herein as Increasing Minimal Bias (IMB).

Abstract: A resiliency system detects and corrects memory errors reported by a memory system of a computing system using previously stored error correction information. When a program stores data into a memory location, the resiliency system executing on the computing system generates and stores error correction information. When the program then executes a load instruction to retrieve the data from the memory location, the load instruction completes normally if there is no memory error. If, however, there is a memory error, the computing system passes control to the resiliency system (e.g., via a trap) to handle the memory error. The resiliency system retrieves the error correction information for the memory location and re-creates the data of the memory location. The resiliency system stores the data as if the load instruction had completed normally and passes control to the next instruction of the program.

Abstract: An improved method for the prevention of deadlock in a massively parallel processor (MPP) system wherein, prior to a process sending messages to another process running on a remote processor, the process allocates space in a deadlock-avoidance FIFO. The allocated space provides a “landing zone” for requests that the software process (the application software) will subsequently issue using a remote-memory-access function. In some embodiments, the deadlock-avoidance (DLA) function provides two different deadlock-avoidance schemes: controlled discard and persistent reservation. In some embodiments, the software process determines which scheme will be used at the time the space is allocated.

Abstract: Computer systems having heat exchangers for cooling computer components are disclosed herein. The computer systems include a computer cabinet having an air inlet, an air outlet spaced apart from the air inlet, and a plurality of computer module compartments positioned between the air inlet and the air outlet. The air inlet, the air outlet, and the computer module compartments define an air flow path through the computer cabinet. The computer systems also include a heat exchanger positioned between two adjacent computer module compartments. The heat exchanger includes a plurality of heat exchange elements canted relative to the air flow path.

Abstract: A system and method for cooling a plurality of electronics cabinets having horizontally positioned electronics assemblies. The system includes at least one blower configured to direct air horizontally across the electronics assemblies, and at least one intercooler configured to extract heat from the air flow such that the system is room neutral, meaning that the ambient temperature remains constant during operation of the system. A plurality of chassis backplanes and power supplies may also include an intercooler, wherein the intercoolers are electronically controlled such that the system is room neutral.

Abstract: A method for prefetching data into a cache is provided. The method allocates an outstanding request buffer (“ORB”). The method stores in an address field of the ORB an address and a number of blocks. The method issues prefetch requests for a degree number of blocks starting at the address. When a prefetch response is received for all the prefetch requests, the method adjusts the address of the next block to prefetch and adjusts the number of blocks remaining to be retrieved and then issues prefetch requests for a degree number of blocks starting at the adjusted address. The prefetching pauses when a maximum distance between the reads of the prefetched blocks and the last prefetched block is reached. When a read request for a prefetched block is received, the method resumes prefetching when a resume criterion is satisfied.

Abstract: A resiliency system detects and corrects memory errors reported by a memory system of a computing system using previously stored error correction information. When a program stores data into a memory location, the resiliency system executing on the computing system generates and stores error correction information. When the program then executes a load instruction to retrieve the data from the memory location, the load instruction completes normally if there is no memory error. If, however, there is a memory error, the computing system passes control to the resiliency system (e.g., via a trap) to handle the memory error. The resiliency system retrieves the error correction information for the memory location and re-creates the data of the memory location. The resiliency system stores the data as if the load instruction had completed normally and passes control to the next instruction of the program.