Appeal 2007-1959
Application 10/039,789
that Sundaresan does not discuss program translation and therefore, is not
relevant to Poulsen (id.).
We disagree with Appellants that the library call of Poulsen does not
partition the operation between a plurality of threads since the parallel
regions actually instantiate and initialize the thread-private memory used in
parallel processing (FF 4 & 8). Additionally, we remain unconvinced by
Appellants’ assertion that the privatizable storage object declarations do not
encapsulate a reduction operation. In that regard, as pointed out by the
Examiner (supra), block 220 declares a privatizable storage object which
encapsulates the current global storage object (FF 7) to be accessed by
multiple processors or threads (FF 2-4).
Additionally, contrary to Appellants’ assertion against combining the
references (Reply Br. 3), combining Sundaresan with Poulsen does not
require a teaching of both partitioning parallel operations as well as using a
reduction operation and a parallel operation in both references. The parallel
operation disclosed by Sundaresan is a data-parallel reduction operation
performed by a group of threads (FF 10). In fact, since Sundaresan’s
reduction operation may be reused for different operation types (FF 9-11),
one of ordinary skill in the art would have combined the references to
perform the reduction operation using the parallel computer programming of
Poulsen to benefit from the parallelism-enhancing effects of privatization of
global storage object (FF 2 & 3).
Claims 23 and 26
Appellants present similar arguments for these claims and argue that
Poulsen merely addresses libraries and the privatization of storage objects,
9
Page: Previous 1 2 3 4 5 6 7 8 9 10 11 Next
Last modified: September 9, 2013