This reverts commit c1f9c5ef

branch

Test test_fast-wave passes when CPU only (_CUDA_MODE_==_OFF_)

Config.pl: option ‘-cuda’ added -> redefines macros _TARGET_GLOBAL_, _TARGET_HOST_, _TARGET_DEVICE_, _CUDA_MODE_
Makefile: -LANG:std is removed for thrust::vector can be used when compiled for GPU

global.h:
1. type using amps_vector added: amps_vector=std::vector when compiled for CPU, and amps_vector=thrust::device_vector when compiled for GPU
2. functions amps_allocate, amps_new, and amsp_delete for allocating using malloc on GPU
3. added class kernel to launch lamblas on GPU

stack.h:
cout is replaced with printf

src/pic
‘using namespace std’ is removed

static data of classes is stored in namespaces:
cDataCenterNode -> cDataCenterNode_static_data
cDataCornerNode -> cDataCornerNode_static_data
cDataBlockAMR -> cDataBlockAMR_static_data

extern vector<fRequestStaticCellData> RequestStaticCellData,RequestStaticCellCornerData -> extern _TARGET_DEVICE_ amps_vector<fRequestStaticCellData> *RequestStaticCellData,*RequestStaticCellCornerData;

2. macro _CUDA_MODE_ is added
3. test fast_wave is passed CPU only with _CUDA_MODE_==_OFF_ (CPU only)

mode because the 'if' statment that was in the target to determine the
compilation mode does not work propoerly on amps-gpu.engin.umich.edu

moved closer the begining of the Makefile so it can be re-defined
to ${LINK.f90} if needed

compiler

need to be included

in case $(INTERFACE)==on: ${WSD}/interface/interface.a is not added to AMPSLINKLIB but added to libAMPS.a directly in target LIB_after_build, which is needed to add functionality from ${WSD}/interface/interface.a to libAMPS.a when AMPS compiled as a library (target LIB)