Parallel Computing Research Laboratory & NVIDIA > 자유게시판

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

Scratchpad memory (SPM), also referred to as scratchpad, scratchpad RAM or native store in pc terminology, is an inside memory, normally high-speed, used for short-term storage of calculations, information, and other work in progress. In reference to a microprocessor (or CPU), scratchpad refers to a particular high-velocity memory used to carry small gadgets of information for fast retrieval. It's just like the utilization and measurement of a scratchpad in life: a pad of paper for preliminary notes or sketches or writings, etc. When the scratchpad is a hidden portion of the main Memory Wave App then it's typically referred to as bump storage. L1 cache in that it's the subsequent closest memory to the ALU after the processor registers, with specific instructions to move information to and from fundamental memory, usually using DMA-based mostly data switch. In distinction to a system that makes use of caches, a system with scratchpads is a system with non-uniform memory entry (NUMA) latencies, because the memory access latencies to the totally different scratchpads and the principle memory vary.

Another distinction from a system that employs caches is that a scratchpad generally doesn't contain a replica of data that can be stored in the main memory. Scratchpads are employed for simplification of caching logic, and to guarantee a unit can work without main memory contention in a system employing multiple processors, especially in multiprocessor system-on-chip for embedded methods. They're largely suited to storing short-term results (because it could be discovered in the CPU stack) that sometimes would not have to always be committing to the primary memory; nevertheless when fed by DMA, they can be used instead of a cache for mirroring the state of slower main memory. The identical issues of locality of reference apply in relation to effectivity of use; although some techniques allow strided DMA to entry rectangular knowledge units. One other distinction is that scratchpads are explicitly manipulated by purposes. They could also be helpful for realtime applications, the place predictable timing is hindered by cache habits.

Scratchpads will not be utilized in mainstream desktop processors the place generality is required for legacy software program to run from generation to technology, wherein the accessible on-chip memory measurement may change. They're higher carried out in embedded methods, particular-purpose processors and Memory Wave recreation consoles, the place chips are sometimes manufactured as MPSoC, and the place software program is often tuned to 1 hardware configuration. Fairchild F8 of 1975 contained 64 bytes of scratchpad. Cyrix 6x86 is the one x86-compatible desktop processor to include a dedicated scratchpad. SuperH, utilized in Sega's consoles, could lock cachelines to an handle outside of main memory for use as a scratchpad. Sony's PS1's R3000 had a scratchpad as an alternative of an L1 cache. It was attainable to put the CPU stack here, an example of the short-term workspace utilization. Adapteva's Epiphany parallel coprocessor options native-stores for every core, linked by a network on a chip, with DMA potential between them and off-chip links (presumably to DRAM).

The structure is just like Sony's Cell, except all cores can instantly deal with one another's scratchpads, generating community messages from customary load/store directions. Sony's PS2 Emotion Engine includes a sixteen KB scratchpad, to and from which DMA transfers could be issued to its GS, and predominant memory. Cell's SPEs are restricted purely to working of their "native-store", counting on DMA for transfers from/to predominant memory and between local shops, much like a scratchpad. In this regard, further benefit is derived from the lack of hardware to check and update coherence between a number of caches: the design takes benefit of the assumption that each processor's workspace is separate and personal. It is predicted this profit will grow to be extra noticeable because the number of processors scales into the "many-core" future. Yet due to the elimination of some hardware logics, the data and directions of applications on SPEs should be managed by way of software if the whole process on SPE can't fit in native store.