Re-adding gfx908 target

Re-adding gfx908 target

Changing to rocm-cmake new style versioning (ROCm#103)

Changing to rocm-cmake new style versioning (ROCm#103)

Refactor primitive test to support multiple GPUs in rings (ROCm#94)

* Refactor primitive test to support multiple GPUs in rings

* Make GPUs sync before transfer optional

* Use same ring format as RCCL

* Extend to 8 GPUs and report errors if there is no P2P access

* Control GPUs sync before ops from command line with "-s" option

* Change buffer size through command line option "-n"

Rename iterations command line option to "-i"

Merge pull request ROCm#72 from wenkaidu/default_rings

Increase number of rings with XGMI connection

increment library version to 0.7.5

Merge pull request ROCm#56 from jichangjichang/master

fix bug in rcclReduce for prod op

Merge pull request ROCm#53 from wenkaidu/const_devlist

use const devlist and increment library version to 0.7.2

Merge pull request ROCm#47 from wenkaidu/move_header_file

move rccl.h to inc folder to be consistent with other ROCm libs