Parallel Computing Analysis Laboratory & NVIDIA > 자유게시판

• 목록
• 답변
• 글쓰기
• 게시판 리스트 옵션
  • 수정
  • 삭제

Scratchpad memory (SPM), also called scratchpad, scratchpad RAM or local retailer in computer terminology, is an inside memory, normally excessive-pace, used for temporary storage of calculations, knowledge, and other work in progress. In reference to a microprocessor (or CPU), scratchpad refers to a particular excessive-speed memory used to hold small items of data for rapid retrieval. It is much like the usage and dimension of a scratchpad in life: a pad of paper for preliminary notes or sketches or MemoryWave Guide writings, MemoryWave Guide and Memory Wave many others. When the scratchpad is a hidden portion of the main memory then it is typically known as bump storage. L1 cache in that it's the next closest memory to the ALU after the processor registers, with specific instructions to maneuver data to and from foremost memory, often utilizing DMA-primarily based knowledge switch. In distinction to a system that uses caches, a system with scratchpads is a system with non-uniform memory entry (NUMA) latencies, because the memory entry latencies to the totally different scratchpads and the primary memory vary.

One other difference from a system that employs caches is that a scratchpad commonly does not include a duplicate of knowledge that can also be stored in the primary memory. Scratchpads are employed for simplification of caching logic, and to guarantee a unit can work with out fundamental memory contention in a system using a number of processors, especially in multiprocessor Memory Wave system-on-chip for embedded systems. They're mostly suited for storing non permanent outcomes (as it would be discovered within the CPU stack) that sometimes wouldn't must at all times be committing to the primary memory; however when fed by DMA, they may also be used instead of a cache for mirroring the state of slower primary memory. The same problems with locality of reference apply in relation to effectivity of use; although some systems allow strided DMA to access rectangular information units. One other difference is that scratchpads are explicitly manipulated by functions. They may be helpful for realtime purposes, the place predictable timing is hindered by cache habits.

Scratchpads are usually not utilized in mainstream desktop processors where generality is required for legacy software to run from era to generation, wherein the obtainable on-chip memory dimension could change. They're better applied in embedded methods, special-purpose processors and recreation consoles, the place chips are often manufactured as MPSoC, and where software is usually tuned to one hardware configuration. Fairchild F8 of 1975 contained sixty four bytes of scratchpad. Cyrix 6x86 is the only x86-compatible desktop processor to include a dedicated scratchpad. SuperH, used in Sega's consoles, may lock cachelines to an deal with outside of main memory to be used as a scratchpad. Sony's PS1's R3000 had a scratchpad as a substitute of an L1 cache. It was doable to put the CPU stack right here, an instance of the momentary workspace usage. Adapteva's Epiphany parallel coprocessor options local-stores for every core, linked by a network on a chip, with DMA possible between them and off-chip hyperlinks (presumably to DRAM).

The architecture is much like Sony's Cell, except all cores can immediately handle one another's scratchpads, generating community messages from customary load/store directions. Sony's PS2 Emotion Engine features a sixteen KB scratchpad, to and from which DMA transfers could possibly be issued to its GS, and essential memory. Cell's SPEs are restricted purely to working in their "native-store", counting on DMA for transfers from/to essential memory and between native shops, very similar to a scratchpad. In this regard, additional profit is derived from the lack of hardware to check and replace coherence between multiple caches: the design takes benefit of the assumption that each processor's workspace is separate and personal. It is predicted this profit will become more noticeable as the number of processors scales into the "many-core" future. Yet because of the elimination of some hardware logics, the information and directions of purposes on SPEs should be managed by way of software if the whole activity on SPE can't fit in native store.