Work Efficient Parallel Scan Assignment
Contents
Background
A work-efficient parallel scan was developed by Guy Blelloch and outlined in: Scan Primitives for Vector Computers and Prefix Sums and Their Applications.
Video: Blelloch Scan |
---|
Comparison
In the two circuit diagrams, you can see that there is less work to do in Blelloch scan, although there are more steps (but not asymptotically more, both scans provide lg(N) spans/critical path lengths).
Blelloch (Work Efficient) | <== ==> | Hillis and Steele (Step Efficient) |
---|---|---|
Code To Implement
You may choose to implement either Inclusive or Exclusive Blelloch Scan. This is an in-place mutating scan algorithm, which means you can mutate the original data array.
Inclusive Scan or Blelloch Exclusive Scan
class: | WorkEfficientParallelMutatingSumScanner.java | |
methods: | IsInclusive | |
package: | scan.challenge | |
source folder: | student/src/main/java |
method: public boolean isInclusive()
(sequential implementation only)
- Here you can indicate whether you will implement inclusive or exclusive scan.
class: | WorkEfficientParallelMutatingSumScanner.java | |
methods: | sumScan | |
package: | scan.challenge | |
source folder: | student/src/main/java |
method: void sumScan(int[] data)
(parallel implementation required)
Testing Your Solution
class: | _WorkEfficientScanTestSuite.java | |
package: | scan.challenge | |
source folder: | testing/src/test/java |