Tags: HSA-Libraries/Bolt
Tags
Bolt Release 1.1 This release contains the following features: 1. Linux support for GCC 4.6 and above. 2. Performance optimizations for below routines for OpenCL path. · transform_scan(inclusive/exclusive) · reduce, transform_reduce, min, max, count, count if · reduce_by_key · sort and stablesort with ints and unsigned ints as data types. · sort_by_key and stablesort_by_key with ints and unsigned ints as keys. 3. Added Below list of new function for OpenCL, TBB and Serial path: · bool binary_search · scatter · scatterif · gather · gatherif · merge 4. Added MultiCore code paths for all routines including new ones. Now we have OpenCL, MultiCore and Serial path for all routines. 5. Bug Fixes. 6. Debug Log facility to determine the executed code path.
Merged the branch v1.0 to Master. This commit contains the following changes: 1. Serial code paths for all routines including routines which do not have a TBB implementation. 3. Performance optimizations for Scan and scan by key, Radix Sort and Stable Sort routines. (Up to 45% for Scan routines for certain data sizes) 4. Provided Offset support to all the routines for both the source and destination iterators. 5. Added unit test cases to take care of all the paths. 6. Added Support for “Sort by Key” routines to work for non-power of 2 buffer sizes in the OpenCL. Uses Merge sort. 7. Added support for Amp Sort for non power of 2 buffer sizes. 8. Moved the TBB code to a separate TBB folder so that there is no duplication of code in OpenCL and AMP. Besides this there are many bug fixes. This checkin will be the release candidate for 1.0 GA release.