Design idea and method：

	Idea：
		Gpgpu-sim is gem5gpu GPU side, and gpgpu-sim can execute both  OpenCL and cuda program, but gem5-gpu can only execute cuda program, so we imitate gpgpu-sim execute OpenCL method and transplant to gem5-gpu.

	Method：
		Set OpenCL 1.X API in benchmarks/libopencl/opecnl_runtime_api.cc file, passing API need parameter to gem5-gpu/src/api/cuda_syscalls.cc, then execute and return result.
		In order to gem5-gpu execute gpu, we change gpgpu_sim object to CudaGPU object( CudaGPU object include gpgpu_sim object ), which store gpu information, and we change launch kernel method to use stream_operation.
 
Compile and Run：

	Step1. Set environment：
		PATH：gcc(4.7 UP)、python(2.7.2 UP)、swig(1.3.40 UP)、scons(2.0.0 UP)
		LD_LIBRARY_PATH：gcc(4.7 UP)、python(2.7.2 UP)、swig(1.3.40 UP)、scons(2.0.0 UP)
		CUDAHOME：cuda(6.5) install path
		NVOpenCL_LIBDIR：OpenCL libeary install path
		GEM5GPU_ROOT：gem5-gpu path
		GEM5GPU_CUDA_OR_OpenCL：set "OpenCL" to execute OpenCL benchmarks
		More detail in sourcefile

	Step2. Build gem5-gpu：
		1.	Compile benchmarks/libopencl
		2.	Update ptx parse rule: bison --name-prefix=ptx_ -v -d ptx.y --file-prefix=ptx 
		3.	Compile gem5-gpu: scons build/X86_VI_hammer_GPU/gem5.opt --default=X86 EXTRAS=../gem5-gpu/src:../gpgpu-sim/ PROTOCOL=VI_hammer GPGPU_SIM=True
		More detail in build.sh file

	Step3. Compile benchmark：
		Gem5gpu need static link file to execute, compilation parameters must add OpenCL library path "-L/path/to/gem5-gpu/OpenCL/lib" and "lopencl".
		More detail in benchmarks’s Makefile file

	Step4. Run benchmark：
		Gem5gpu_PATH/gem5/build/X86_VI_hammer_GPU/gem5.opt  Gem5gpu_PATH/gem5-gpu/configs/se_fusion.py -c BENCHMARKS_OBJ
		More detail in benchmarks’s run.sh file