help_of_BufferBandwidth.txt

Usage
-h,   --help                                            Display this information
      --device        [cpu|gpu]                         Execute the openCL kernel on a device
-q,   --quiet                                           Quiet mode. Suppress all text output.
-e,   --verify                                          Verify results against reference implementation.
-t,   --timing                                          Print timing.
      --dump          [filename]                        Dump binary image for all devices
      --load          [filename]                        Load binary image and execute on device
      --flags         [filename]                        Specify filename containing the compiler flags to build kernel
-p,   --platformId    [value]                           Select platformId to be used[0 to N-1 where N is number platforms available].
-v,   --version                                         AMD APP SDK version string.
-d,   --deviceId      [value]                           Select deviceId to be used[0 to N-1 where N is number devices available].
-nwk, --numcpuwrk                                       Number of CPU workers (max: 32 (Linux: 1))
-nl,  --numLoops      <n>                               Number of iterations to repeat overall bandwidth measurement
-lp,  --numIterBfr    <n>                               When -noblock is active, set the number of iterations to run read/write buffer calls
-nb,  --numBytes      <n>                               Buffer size in bytes (min: 2048*CPU Workers)
-nr,  --numRepeats    <n>                               Repeat each timing <n> times (can't be 0)
-nk,  --numKernelLoops<n>                               Number of loops in kernel
-nw,  --numWavefronts <n>                               # of wave fronts per SIMD (can't be 0) (default: 7)
-ty,  --testType      <n>                               Type of test:
							0 clEnqueue[Map,Unmap]
							1 clEnqueue[Read,Write]
							2 clEnqueueCopy
							3 clEnqueue[Read,Write], prepinned

-dma, --pcie                                            Measure PCIe/interconnect bandwidth
-nob, --noblock                                         When -pcie is active, measure PCIe/interconnect
							bandwidth using multiple back-to-back asynchronous
							buffer copies

-s,   --nSkip         <n>                               Skip first <n> timings for average (default: 1)
-l,   --printLog                                        Print complete timing log
-if,  --inputflag     <n>                               Memory flags. OK to use multiple
							0 CL_MEM_READ_ONLY
							1 CL_MEM_WRITE_ONLY
							2 CL_MEM_READ_WRITE
							3 CL_MEM_USE_HOST_PTR
							4 CL_MEM_COPY_HOST_PTR
							5 CL_MEM_ALLOC_HOST_PTR
							6 CL_MEM_USE_PERSISTENT_MEM_AMD
							7 CL_MEM_HOST_WRITE_ONLY
							8 CL_MEM_HOST_READ_ONLY
							9 CL_MEM_HOST_NO_ACCESS

-of,  --outputflag    <n>                               Memory flags. OK to use multiple
							0 CL_MEM_READ_ONLY
							1 CL_MEM_WRITE_ONLY
							2 CL_MEM_READ_WRITE
							3 CL_MEM_USE_HOST_PTR
							4 CL_MEM_COPY_HOST_PTR
							5 CL_MEM_ALLOC_HOST_PTR
							6 CL_MEM_USE_PERSISTENT_MEM_AMD
							7 CL_MEM_HOST_WRITE_ONLY
							8 CL_MEM_HOST_READ_ONLY
							9 CL_MEM_HOST_NO_ACCESS

-cf,  --copyflag      <n>                               Memory flags. OK to use multiple
							0 CL_MEM_READ_ONLY
							1 CL_MEM_WRITE_ONLY
							2 CL_MEM_READ_WRITE
							3 CL_MEM_USE_HOST_PTR
							4 CL_MEM_COPY_HOST_PTR
							5 CL_MEM_ALLOC_HOST_PTR
							6 CL_MEM_USE_PERSISTENT_MEM_AMD
							7 CL_MEM_HOST_WRITE_ONLY
							8 CL_MEM_HOST_READ_ONLY
							9 CL_MEM_HOST_NO_ACCESS

-hbw, --enablehostbw                                    enable/disable host mem B/W baseline. 0 or 1
-m,   --mapRW                                           always map as MAP_READ | MAP_WRITE
-t,   --timings                                         Print all timings including setup-time