Matches in SemOpenAlex for { <https://semopenalex.org/work/W1488103234> ?p ?o ?g. }
Showing items 1 to 87 of
87
with 100 items per page.
- W1488103234 abstract "A common feature of many scalable parallel machines is non-uniform memory access (NUMA)--data access to local memory is much faster than to non-local memories. In addition, when a number of remote accesses must be made, it is usually more efficient to use block transfers of data rather than to use many small messages. Almost every modern processor is designed with a memory hierarchy organized into several levels--each smaller and faster than the level below. In general, the effective use of parallel machines requires careful attention to the following issues: (1) exposing and exploiting parallelism; (2) accessing local memory instead of remote memory; (3) using block transfers for remote accesses; (4) reusing data in the cache; and (5) load balancing.We have built a system called Pnuma for programming NUMA machines. We make the following contributions: First, we propose a parallelization scheme for both parallelism and data locality. Second, we develop a framework based on non-singular matrices and integer lattice theory for the systematic development of loop transformations. Program transformations, such as loop restructuring, are critical to achieving high performance. The framework can be used in parallelizing compilers for both coarse-grain and fine-grain parallel architectures. We have implemented a loop restructuring tool-kit called Lambda based on this framework. Third, using this loop transformation framework, we develop algorithms for improving memory locality. The memory locality algorithm restructures loop nests to expose opportunities for parallel execution and for block transfers, while keeping data accesses local wherever possible. Fourth, for cache locality, we introduce a new simple cache model based on reuse distances, which is more precise than the existing reuse vector space model. We develop a new loop transformation technique that optimizes directly on reuse distances, so that no exhaustive search is necessary. Fifth, we use our loop transformation framework to improve parallelism as well. We develop a unified algorithm for parallelism, memory locality and cache locality.System evaluations have been conducted on a multiprocessor machine without cache (BBN GP1000), a uniprocessor workstation with cache (HP 9000/720) and a multiprocessor machine with caches (KSR1), using programs from linear algebra, NASA benchmarks and SIMPLE hydrodynamics benchmark." @default.
- W1488103234 created "2016-06-24" @default.
- W1488103234 creator A5000432967 @default.
- W1488103234 date "1993-01-02" @default.
- W1488103234 modified "2023-09-22" @default.
- W1488103234 title "Compiling for numa parallel machines" @default.
- W1488103234 hasPublicationYear "1993" @default.
- W1488103234 type Work @default.
- W1488103234 sameAs 1488103234 @default.
- W1488103234 citedByCount "49" @default.
- W1488103234 countsByYear W14881032342015 @default.
- W1488103234 countsByYear W14881032342022 @default.
- W1488103234 crossrefType "journal-article" @default.
- W1488103234 hasAuthorship W1488103234A5000432967 @default.
- W1488103234 hasConcept C111919701 @default.
- W1488103234 hasConcept C115537543 @default.
- W1488103234 hasConcept C11799548 @default.
- W1488103234 hasConcept C120314980 @default.
- W1488103234 hasConcept C12186640 @default.
- W1488103234 hasConcept C1306188 @default.
- W1488103234 hasConcept C133875982 @default.
- W1488103234 hasConcept C136085584 @default.
- W1488103234 hasConcept C138885662 @default.
- W1488103234 hasConcept C169590947 @default.
- W1488103234 hasConcept C173608175 @default.
- W1488103234 hasConcept C176649486 @default.
- W1488103234 hasConcept C189783530 @default.
- W1488103234 hasConcept C201148951 @default.
- W1488103234 hasConcept C27602214 @default.
- W1488103234 hasConcept C2778100165 @default.
- W1488103234 hasConcept C2779808786 @default.
- W1488103234 hasConcept C3720319 @default.
- W1488103234 hasConcept C41008148 @default.
- W1488103234 hasConcept C41895202 @default.
- W1488103234 hasConcept C48044578 @default.
- W1488103234 hasConcept C51290061 @default.
- W1488103234 hasConcept C91481028 @default.
- W1488103234 hasConceptScore W1488103234C111919701 @default.
- W1488103234 hasConceptScore W1488103234C115537543 @default.
- W1488103234 hasConceptScore W1488103234C11799548 @default.
- W1488103234 hasConceptScore W1488103234C120314980 @default.
- W1488103234 hasConceptScore W1488103234C12186640 @default.
- W1488103234 hasConceptScore W1488103234C1306188 @default.
- W1488103234 hasConceptScore W1488103234C133875982 @default.
- W1488103234 hasConceptScore W1488103234C136085584 @default.
- W1488103234 hasConceptScore W1488103234C138885662 @default.
- W1488103234 hasConceptScore W1488103234C169590947 @default.
- W1488103234 hasConceptScore W1488103234C173608175 @default.
- W1488103234 hasConceptScore W1488103234C176649486 @default.
- W1488103234 hasConceptScore W1488103234C189783530 @default.
- W1488103234 hasConceptScore W1488103234C201148951 @default.
- W1488103234 hasConceptScore W1488103234C27602214 @default.
- W1488103234 hasConceptScore W1488103234C2778100165 @default.
- W1488103234 hasConceptScore W1488103234C2779808786 @default.
- W1488103234 hasConceptScore W1488103234C3720319 @default.
- W1488103234 hasConceptScore W1488103234C41008148 @default.
- W1488103234 hasConceptScore W1488103234C41895202 @default.
- W1488103234 hasConceptScore W1488103234C48044578 @default.
- W1488103234 hasConceptScore W1488103234C51290061 @default.
- W1488103234 hasConceptScore W1488103234C91481028 @default.
- W1488103234 hasLocation W14881032341 @default.
- W1488103234 hasOpenAccess W1488103234 @default.
- W1488103234 hasPrimaryLocation W14881032341 @default.
- W1488103234 hasRelatedWork W1491410022 @default.
- W1488103234 hasRelatedWork W1737297434 @default.
- W1488103234 hasRelatedWork W1973122629 @default.
- W1488103234 hasRelatedWork W1987402220 @default.
- W1488103234 hasRelatedWork W2014033064 @default.
- W1488103234 hasRelatedWork W2042125826 @default.
- W1488103234 hasRelatedWork W2042629708 @default.
- W1488103234 hasRelatedWork W2045773292 @default.
- W1488103234 hasRelatedWork W2069738303 @default.
- W1488103234 hasRelatedWork W2086237056 @default.
- W1488103234 hasRelatedWork W2098220211 @default.
- W1488103234 hasRelatedWork W2108315152 @default.
- W1488103234 hasRelatedWork W2112482891 @default.
- W1488103234 hasRelatedWork W2119609467 @default.
- W1488103234 hasRelatedWork W2122418911 @default.
- W1488103234 hasRelatedWork W2124711255 @default.
- W1488103234 hasRelatedWork W2158158791 @default.
- W1488103234 hasRelatedWork W2296218291 @default.
- W1488103234 hasRelatedWork W3004256893 @default.
- W1488103234 hasRelatedWork W74033418 @default.
- W1488103234 isParatext "false" @default.
- W1488103234 isRetracted "false" @default.
- W1488103234 magId "1488103234" @default.
- W1488103234 workType "article" @default.