Closed GoogleCodeExporter closed 9 years ago
purpose of host_vector, device_vector and {omp,cuda,cpp}::vector
existence of wrapped references
existence of device_ptr
reduce requires associativity and commutativity (std::accumulate does not)
could also mention that the *current* implementations of some algos requires
data types to have default constructors and trivial copy constructors
Original comment by wnbell
on 6 Oct 2011 at 11:31
host/cpp algorithms have parallel semantics, but *current* implementation is
serial
Original comment by jaredhoberock
on 6 Oct 2011 at 11:35
presence of thrust::minimum and thrust::maximum
Original comment by wnbell
on 11 Oct 2011 at 12:47
presence of key-value algorithms (e.g. sort_by_key)
generalization of some algorithms to accept stencil sequences
Original comment by wnbell
on 12 Oct 2011 at 1:47
presence of stable_partition_copy
Original comment by jaredhoberock
on 13 Oct 2011 at 7:14
interpretation of get_temporary_buffer's result
Original comment by jaredhoberock
on 18 Oct 2011 at 6:52
what thrust::detail means
Original comment by jaredhoberock
on 28 Oct 2011 at 12:41
existence of reduce() and inner_product() variants with implicit initial values
(zero)
Original comment by wnbell
on 8 Nov 2011 at 10:51
what an experimental namespace implies
Original comment by jaredhoberock
on 23 Jan 2012 at 11:00
non existence of non-parallel-friendly data structures such as map, list, etc.
Original comment by jaredhoberock
on 7 Feb 2012 at 12:12
I've added a page which captures most of the important differences:
http://code.google.com/p/thrust/wiki/ThrustVersusStandardLibrary
The other details should probably be noted in the more fine-grained doxygen
documentation. I've added an issue to do them in 1.7.
Original comment by jaredhoberock
on 7 Feb 2012 at 10:57
Original issue reported on code.google.com by
jaredhoberock
on 6 Oct 2011 at 11:15