Lec17-Irregular problems continued

# Lec17-Irregular problems continued - Lecture 17 Irregular...

This preview shows pages 1–13. Sign up to view the full content.

Lecture 17 Irregular problems (II)

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 2 Announcements Come see my talk at the “Last Lecture Series” Wed 11/22 at 12Noon Gallery A, Price Center Free Pizza and Drinks Office hours on Weds 11/22 delayed to 2pm to 3pm Tuesday’s Lecture Be prepared to discuss the papers in class Slides will be available first thing Tuesday morning
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 3 CSE 260 Symposium Scheduling changes Meet over 2 days Thursday 11/30, 1:30 to 3:20PM Friday 12/1, ~1:30 to 3:30 PM Format 15 minute presentation

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 4 Partitioning How do we divide up the computation and assign to processors? The process is called decomposition or partitioning Related issue: processor mapping
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 5 Unstructured meshes • “Unstructured” meshes are another type of non- uniformly spaced mesh • Useful when the boundary or surface of the object is complicated Randy Bank, UCSD

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 6 A typical irregular mesh sweep loop float x[n_node], y[n_node] int E0[n_edge], E1[n_edge] for i = 1 : n_edge // Loop over all edges int n0 = E0[i] int n1 = E1[i] y[n0] += f(x[n0], x[n1]); y[n1] += g(x[n0], x[n1]); end for i=1 n0 n1
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 7 Run time support for unstructured meshes • Data dependences across processor boundaries • Irregular communication patterns

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 9 Implementation issues • Because the mesh does not have a regular structure, we must keep processor mapping information for each point in the mesh • Compare with a uniform mesh, in which the mapping information is much coarser grained: mapping is handled at the level of a subdomain comprising many points
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 10 Requirements • We don’t know the assignment of nodes to processors until run time • Assume that the mesh doesn’t change once initialized • We need to include ghost cells, and want to keep the same loop structure • We execute loops using local indices

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 11 Processor Mapping • Invoke a partitioner to map global indices to (processor, local index) pairs
11/17/06 Scott B. Baden / CSE 260 / Fall 2006 12 Global translation • Each global index is mapped to a processor and a local index • A global translation table expresses this mapping Global Local

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Scott B. Baden / CSE 260 / Fall 2006 13 Global translation • Because we require mapping information for each individual mesh node, we cannot fit it in the memory of a single processor
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 02/20/2008 for the course CSE 260 taught by Professor Baden during the Fall '06 term at UCSD.

### Page1 / 36

Lec17-Irregular problems continued - Lecture 17 Irregular...

This preview shows document pages 1 - 13. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online