Appeal 2007-1959
Application 10/039,789
9. Sundaresan relates to a generic reduction object for data
parallelism (col. 1, ll. 45-47), wherein a reduction operation is described as
one that reduces N values distributed over N tasks and includes operations
such as sum and product, maximum and minimum, and logical Boolean
operators (col. 1, ll. 52-63).
10. Sundaresan provides for a generic reduction object for data
parallelism wherein a data-parallel reduction operation is performed by a
group of threads or a rope participating in a multi-level two-phase tree
structure (col. 3, ll. 61-67). By separating a reduction object template and
type-specific reduction object from the actual reduction operation, the same
reduction skeleton object may be used for all reduction operations within a
rope, and also a type-specific reduction object, once created, may be reused
for different reduction operations of the same type (col. 5, ll. 7-13).
11. Sundaresan further discloses data-parallelism through a
reduction operation where each thread contributes a value, and the values are
reduced using a function to obtain and return a reduced value to each of the
threads (col. 7, ll. 13-16).
12. Hardwick relates to parallel processing methods in which
unnecessary inter-process communications are eliminated by using a lazy
collection oriented data type (col. 4, ll. 21-25) such as a vector (col. 4, ll. 35-
40).
13. Hardwick further discloses that a basic data-parallel data
structure is a vector which may be formed from any of the basic C datatypes
or user-defined datatypes (col. 6, ll. 30-34). Hardwick further describes
“portability” across both shared and distributed memory machines as the
6
Page: Previous 1 2 3 4 5 6 7 8 9 10 11 Next
Last modified: September 9, 2013