Search NASA⌕ Search

Engineering topics

Loh, Gabriel H.

Publications and source records attributed to Loh, Gabriel H..

Runtime extension for neural network training with heterogeneous memory

Systems, apparatuses, and methods for managing buffers in a neural network implementation with heterogeneous memory are disclosed. A system includes a neural network coupled to a first memory and a second memory. The first memory is a relatively low-capacity, high-bandwidth memory while the second memory is a relatively high-capacity, low-bandwidth memory. During a forward propagation pass of the neural network, a run-time manager monitors the usage of the buffers for the various layers of the neural network. During a backward propagation pass of the neural network, the run-time manager determines how to move the buffers between the first and second memories based on the monitored buffer usage during the forward propagation pass. As a result, the run-time manager is able to reduce memory access latency for the layers of the neural network during the backward propagation pass.

Mappouras, Georgios↗

Memory access response merging in a memory hierarchy

A system and method for efficiently processing memory requests are described. A computing system includes multiple compute units, multiple caches of a memory hierarchy and a communication fabric. A compute unit generates a memory access request that misses in a higher level cache, which sends a miss request to a lower level shared cache. During servicing of the miss request, the lower level cache merges identification information of multiple memory access requests targeting a same cache line from multiple compute units into a merged memory access response. The lower level shared cache continues to insert information into the merged memory access response until the lower level shared cache is ready to issue the merged memory access response. An intermediate router in the communication fabric broadcasts the merged memory access response into multiple memory access responses to send to corresponding compute units.

97 MATHEMATICS AND COMPUTING↗

Memory hierarchy using page-based compression

A system includes a device coupleable to a first memory. The device includes a second memory to cache data from the first memory. The second memory is to store a set of compressed pages of the first memory and a set of page descriptors. Each compressed page includes a set of compressed data blocks. Each page descriptor represents a corresponding page and includes a set of location identifiers that identify the locations of the compressed data blocks of the corresponding page in the second memory. The device further includes compression logic to compress data blocks of a page to be stored to the second memory and decompression logic to decompress compressed data blocks of a page accessed from the second memory.

Loh, Gabriel H.↗

Mechanism for reducing page migration overhead in memory systems

A technique for use in a memory system includes swapping a first plurality of pages of a first memory of the memory system with a second plurality of pages of a second memory of the memory system. The first memory has a first latency and the second memory has a second latency. The first latency is less than the second latency. The technique includes updating a page table and triggering a translation lookaside buffer shootdown to associate a virtual address of each of the first plurality of pages with a corresponding physical address in the second memory and to associate a virtual address for each of the second plurality of pages with a corresponding physical address in the first memory.

97 MATHEMATICS AND COMPUTING↗

Preemptive cache management policies for processing units

A processing system includes at least one central processing unit (CPU) core, at least one graphics processing unit (GPU) core, a main memory, and a coherence directory for maintaining cache coherence. The at least one CPU core receives a CPU cache flush command to flush cache lines stored in cache memory of the at least one CPU core prior to launching a GPU kernel. The coherence directory transfers data associated with a memory access request by the at least one GPU core from the main memory without issuing coherence probes to caches of the at least one CPU core.

Kayiran, Onur↗

Adaptive extension of leases for entries in a translation lookaside buffer

The described embodiments include a computing device with two or more translation lookaside buffers (TLB). During operation, the computing device updates an entry in the TLB based on a virtual address to physical address translation and metadata from a page table entry that were acquired during a page table walk. The computing device then computes, based on a lease length expression, a lease length for the entry in the TLB. Next, the computing device sets, for the entry in the TLB, a lease value to the lease length, wherein the lease value represents a time until a lease for the entry in the TLB expires, wherein the entry in the TLB is invalid when the associated lease has expired. The computing device then uses the lease value to control operations that are allowed to be performed using information from the entry in the TLB.

97 MATHEMATICS AND COMPUTING↗

Hot page selection in multi-level memory hierarchies

Systems, apparatuses, and methods for sorting memory pages in a multi-level heterogeneous memory architecture. The system may classify pages into a first “hot” category or a second “cold” category. The system may attempt to place the “hot” pages into the memory level(s) closest to the systems' processor cores. The system may track parameters associated with each page, with the parameters including number of accesses, types of accesses, power consumed per access, temperature, wearability, and/or other parameters. Based on these parameters, the system may generate a score for each page. Then, the system may compare the score of each page to a threshold. If the score of a given page is greater than the threshold, the given page may be designated as “hot”. If the score of the given page is less than the threshold, the given page may be designated as “cold”.

97 MATHEMATICS AND COMPUTING↗

Asynchronous cache flushing

Proactive flush logic in a computing system is configured to perform a proactive flush operation to flush data from a first memory in a first computing device to a second memory in response to execution of a non-blocking flush instruction. Reactive flush logic in the computing system is configured to, in response to a memory request issued prior to completion of the proactive flush operation, interrupt the proactive flush operation and perform a reactive flush operation to flush requested data from the first memory to the second memory.

97 MATHEMATICS AND COMPUTING↗

Memory hierarchy using row-based compression

A system includes a first memory and a device coupleable to the first memory. The device includes a second memory to cache data from the first memory. The second memory includes a plurality of rows, each row including a corresponding set of compressed data blocks of non-uniform sizes and a corresponding set of tag blocks. Each tag block represents a corresponding compressed data block of the row. The device further includes decompression logic to decompress data blocks accessed from the second memory. The device further includes compression logic to compress data blocks to be stored in the second memory.

97 MATHEMATICS AND COMPUTING↗