add an option ("--memalloc") to client.exe or measurePerformance.py to manipulate different flags of clCreateBuffer. The "roundtrip" time has to be selected to enable the test of different flags. For example, "--roundtrip roundtrip --memalloc default" will measure the roundtrip time with default flags of clCreateBuffer and requires clEnqueueReadBuffer and clEnqueueWriteBuffer; "--roundtrip roundtrip --memalloc alloc_host_ptr" will create the buffer with flag "cl_mem_alloc_host_ptr" and map/unmap the memory.
All four precision are supported to GEMM and TRSM.
add an option ("--memalloc") to client.exe or measurePerformance.py to manipulate different flags of clCreateBuffer. The "roundtrip" time has to be selected to enable the test of different flags. For example, "--roundtrip roundtrip --memalloc default" will measure the roundtrip time with default flags of clCreateBuffer and requires clEnqueueReadBuffer and clEnqueueWriteBuffer; "--roundtrip roundtrip --memalloc alloc_host_ptr" will create the buffer with flag "cl_mem_alloc_host_ptr" and map/unmap the memory. All four precision are supported to GEMM and TRSM.