cuda: Various fixes to enable new GPU Manager02/4302/13
commit977fe98edce6113658c402416bd45d12186e9947
authorMichael Robson <mprobson@illinois.edu>
Tue, 24 Apr 2018 22:06:59 +0000 (24 17:06 -0500)
committerJaemin Choi <jchoi157@illinois.edu>
Tue, 25 Sep 2018 04:50:39 +0000 (25 00:50 -0400)
treebcc346b1685dedbc775a6f3e9f2c273fa7f8031f
parentc925ff9485b53e906d8e88e1d3cc550425a808aa
cuda: Various fixes to enable new GPU Manager

* Change memcpyAsync to memcpy (synchronous)
* Ensure free is called for both wr and hapi APIs
* Fix CUDA_ and HAPI_ define flag inconsistencies
* Change GPUManager::createStreams() to always return number of streams
* Re-enable HAPI_MEMPOOL by default
* Update user_data API to either set or copy due to user preference

Change-Id: Id6fa947d5cea5b49af04bbd287ba97a2faa240a0
examples/charm++/cuda/matmul/matmul.cu
examples/charm++/cuda/vecadd/vecadd.cu
src/arch/cuda/hybridAPI/Makefile
src/arch/cuda/hybridAPI/hapi.h
src/arch/cuda/hybridAPI/hapi_impl.cu
src/arch/cuda/hybridAPI/hapi_impl.h
src/conv-core/convcore.c