clintercept_log.txt

-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
CLIntercept (64-bit) is loading...
CLIntercept file location: /home/s/qwer/opencl-intercept-layer/build/cliloader/../intercept/libOpenCL.so
CLIntercept URL: https://github.com/intel/opencl-intercept-layer
CLIntercept git description: v3.0.0-23-g25557bb
CLIntercept git refspec: refs/heads/master
CLIntercept git hash: 25557bbc018308923975efeeb481e51521ea61df
CLIntercept optional features:
    cliloader(supported)
    cliprof(supported)
    kernel overrides(supported)
    ITT tracing(NOT supported)
    MDAPI(supported)
    clock(steady_clock)
CLIntercept environment variable prefix: CLI_
CLIntercept config file: clintercept.conf
Trying to load dispatch from: ./real_libOpenCL.so
Couldn't load library: ./real_libOpenCL.so
Trying to load dispatch from: /usr/lib/x86_64-linux-gnu/libOpenCL.so.1
... success!
Control LogToFile is set to non-default value: true
Control CallLogging is set to non-default value: true
Control ReportToStderr is set to non-default value: true
Timer Started!
... loading complete.
>>>> clGetPlatformIDs
<<<< clGetPlatformIDs -> CL_SUCCESS
>>>> clGetPlatformIDs
<<<< clGetPlatformIDs -> CL_SUCCESS
>>>> clCreateContextFromType: properties = [ CL_CONTEXT_PLATFORM = Intel(R) OpenCL HD Graphics (0x56347eff3a20) ], device_type = CL_DEVICE_TYPE_GPU (4)
<<<< clCreateContextFromType: returned 0x56347f0bff60 -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_NAME (00000902)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_NAME (00000902)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_VENDOR (00000903)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_VENDOR (00000903)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetContextInfo: param_name = CL_CONTEXT_DEVICES (00001081)
<<<< clGetContextInfo -> CL_SUCCESS
>>>> clGetContextInfo: param_name = CL_CONTEXT_DEVICES (00001081)
<<<< clGetContextInfo -> CL_SUCCESS
>>>> clRetainDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clRetainDevice: [ ref count = 1 ] -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_NAME (0000102B)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_NAME (0000102B)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_TYPE (00001000)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE (00001020)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_SVM_CAPABILITIES (00001053)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_DOUBLE_FP_CONFIG (00001032)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_SINGLE_FP_CONFIG (0000101B)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clCreateCommandQueueWithProperties: context = 0x56347f0bff60, device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), properties = [ CL_QUEUE_PROPERTIES =  ]
<<<< clCreateCommandQueueWithProperties: returned 0x56347f7c35b0 -> CL_SUCCESS
>>>> clCreateBuffer: context = 0x56347f0bff60, flags = CL_MEM_READ_ONLY (4), size = 65536, host_ptr = (nil)
<<<< clCreateBuffer: returned 0x56347f8f4460 -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f8f4460, param_name = CL_MEM_HOST_PTR (00001103)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f8f4460, param_name = CL_MEM_SIZE (00001102)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f8f4460, param_name = CL_MEM_TYPE (00001100)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f8f4460, param_name = CL_MEM_FLAGS (00001101)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clCreateBuffer: context = 0x56347f0bff60, flags = CL_MEM_READ_ONLY (4), size = 65536, host_ptr = (nil)
<<<< clCreateBuffer: returned 0x56347f5e7170 -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f5e7170, param_name = CL_MEM_HOST_PTR (00001103)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f5e7170, param_name = CL_MEM_SIZE (00001102)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f5e7170, param_name = CL_MEM_TYPE (00001100)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f5e7170, param_name = CL_MEM_FLAGS (00001101)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clCreateBuffer: context = 0x56347f0bff60, flags = CL_MEM_READ_WRITE (1), size = 65536, host_ptr = (nil)
<<<< clCreateBuffer: returned 0x56347f0bc1f0 -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0bc1f0, param_name = CL_MEM_HOST_PTR (00001103)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0bc1f0, param_name = CL_MEM_SIZE (00001102)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0bc1f0, param_name = CL_MEM_TYPE (00001100)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0bc1f0, param_name = CL_MEM_FLAGS (00001101)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clCreateProgramWithSource: context = 0x56347f0bff60, count = 1
<<<< clCreateProgramWithSource: returned 0x56347f0df750, program number = 0000 -> CL_SUCCESS
>>>> clBuildProgram: program = 0x56347f0df750, pfn_notify = (nil)
<<<< clBuildProgram -> CL_SUCCESS
>>>> clGetProgramInfo: param_name = CL_PROGRAM_DEVICES (00001163)
<<<< clGetProgramInfo -> CL_SUCCESS
>>>> clGetProgramInfo: param_name = CL_PROGRAM_DEVICES (00001163)
<<<< clGetProgramInfo -> CL_SUCCESS
>>>> clRetainDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clRetainDevice: [ ref count = 1 ] -> CL_SUCCESS
>>>> clGetProgramBuildInfo: param_name = CL_PROGRAM_BUILD_LOG (00001183)
<<<< clGetProgramBuildInfo -> CL_SUCCESS
>>>> clGetProgramBuildInfo: param_name = CL_PROGRAM_BUILD_LOG (00001183)
<<<< clGetProgramBuildInfo -> CL_SUCCESS
>>>> clRetainDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clRetainDevice: [ ref count = 1 ] -> CL_SUCCESS
>>>> clReleaseDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clReleaseDevice: [ ref count = 0 ] -> CL_SUCCESS
>>>> clReleaseDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clReleaseDevice: [ ref count = 0 ] -> CL_SUCCESS
>>>> clCreateKernel: program = 0x56347f0df750, kernel_name = op_complex
<<<< clCreateKernel: returned 0x56347fbf1210 -> CL_SUCCESS
>>>> clGetKernelWorkGroupInfo( op_complex ): param_name = CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE (11B3)
<<<< clGetKernelWorkGroupInfo -> CL_SUCCESS
>>>> clGetKernelWorkGroupInfo( op_complex ): param_name = CL_KERNEL_WORK_GROUP_SIZE (11B0)
<<<< clGetKernelWorkGroupInfo -> CL_SUCCESS
>>>> clGetPlatformIDs
<<<< clGetPlatformIDs -> CL_SUCCESS
>>>> clGetPlatformIDs
<<<< clGetPlatformIDs -> CL_SUCCESS
>>>> clCreateContextFromType: properties = [ CL_CONTEXT_PLATFORM = Intel(R) OpenCL HD Graphics (0x56347eff3a20) ], device_type = CL_DEVICE_TYPE_GPU (4)
<<<< clCreateContextFromType: returned 0x56347fbf9500 -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_NAME (00000902)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_NAME (00000902)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_VENDOR (00000903)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetPlatformInfo: platform = Intel(R) OpenCL HD Graphics (0x56347eff3a20), param_name = CL_PLATFORM_VENDOR (00000903)
<<<< clGetPlatformInfo -> CL_SUCCESS
>>>> clGetContextInfo: param_name = CL_CONTEXT_DEVICES (00001081)
<<<< clGetContextInfo -> CL_SUCCESS
>>>> clGetContextInfo: param_name = CL_CONTEXT_DEVICES (00001081)
<<<< clGetContextInfo -> CL_SUCCESS
>>>> clRetainDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clRetainDevice: [ ref count = 1 ] -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_NAME (0000102B)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_NAME (0000102B)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_TYPE (00001000)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_MAX_CONSTANT_BUFFER_SIZE (00001020)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_SVM_CAPABILITIES (00001053)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_DOUBLE_FP_CONFIG (00001032)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_EXTENSIONS (00001030)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clGetDeviceInfo: device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), param_name = CL_DEVICE_SINGLE_FP_CONFIG (0000101B)
<<<< clGetDeviceInfo -> CL_SUCCESS
>>>> clCreateCommandQueueWithProperties: context = 0x56347fbf9500, device = Intel(R) Graphics Gen9 [0x5917] (CL_DEVICE_TYPE_GPU) (0x56347eff3af0), properties = [ CL_QUEUE_PROPERTIES =  ]
<<<< clCreateCommandQueueWithProperties: returned 0x56347f0ffc40 -> CL_SUCCESS
>>>> clCreateBuffer: context = 0x56347fbf9500, flags = CL_MEM_READ_ONLY (4), size = 65536, host_ptr = (nil)
<<<< clCreateBuffer: returned 0x56347f166be0 -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f166be0, param_name = CL_MEM_HOST_PTR (00001103)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f166be0, param_name = CL_MEM_SIZE (00001102)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f166be0, param_name = CL_MEM_TYPE (00001100)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f166be0, param_name = CL_MEM_FLAGS (00001101)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clCreateBuffer: context = 0x56347fbf9500, flags = CL_MEM_READ_ONLY (4), size = 65536, host_ptr = (nil)
<<<< clCreateBuffer: returned 0x56347f0c2770 -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0c2770, param_name = CL_MEM_HOST_PTR (00001103)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0c2770, param_name = CL_MEM_SIZE (00001102)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0c2770, param_name = CL_MEM_TYPE (00001100)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347f0c2770, param_name = CL_MEM_FLAGS (00001101)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clCreateBuffer: context = 0x56347fbf9500, flags = CL_MEM_READ_WRITE (1), size = 65536, host_ptr = (nil)
<<<< clCreateBuffer: returned 0x56347fbeec30 -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347fbeec30, param_name = CL_MEM_HOST_PTR (00001103)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347fbeec30, param_name = CL_MEM_SIZE (00001102)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347fbeec30, param_name = CL_MEM_TYPE (00001100)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clGetMemObjectInfo: mem = 0x56347fbeec30, param_name = CL_MEM_FLAGS (00001101)
<<<< clGetMemObjectInfo -> CL_SUCCESS
>>>> clCreateProgramWithSource: context = 0x56347fbf9500, count = 1
<<<< clCreateProgramWithSource: returned 0x56347f3fcdd0, program number = 0001 -> CL_SUCCESS
>>>> clBuildProgram: program = 0x56347f3fcdd0, pfn_notify = (nil)
<<<< clBuildProgram -> CL_SUCCESS
>>>> clGetProgramInfo: param_name = CL_PROGRAM_DEVICES (00001163)
<<<< clGetProgramInfo -> CL_SUCCESS
>>>> clGetProgramInfo: param_name = CL_PROGRAM_DEVICES (00001163)
<<<< clGetProgramInfo -> CL_SUCCESS
>>>> clRetainDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clRetainDevice: [ ref count = 1 ] -> CL_SUCCESS
>>>> clGetProgramBuildInfo: param_name = CL_PROGRAM_BUILD_LOG (00001183)
<<<< clGetProgramBuildInfo -> CL_SUCCESS
>>>> clGetProgramBuildInfo: param_name = CL_PROGRAM_BUILD_LOG (00001183)
<<<< clGetProgramBuildInfo -> CL_SUCCESS
>>>> clRetainDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clRetainDevice: [ ref count = 1 ] -> CL_SUCCESS
>>>> clReleaseDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clReleaseDevice: [ ref count = 0 ] -> CL_SUCCESS
>>>> clReleaseDevice: [ ref count = 1 ] device = 0x56347eff3af0
<<<< clReleaseDevice: [ ref count = 0 ] -> CL_SUCCESS
>>>> clCreateKernel: program = 0x56347f3fcdd0, kernel_name = op_complex
<<<< clCreateKernel: returned 0x563480eae9c0 -> CL_SUCCESS
>>>> clGetKernelWorkGroupInfo( op_complex ): param_name = CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE (11B3)
<<<< clGetKernelWorkGroupInfo -> CL_SUCCESS
>>>> clGetKernelWorkGroupInfo( op_complex ): param_name = CL_KERNEL_WORK_GROUP_SIZE (11B0)
<<<< clGetKernelWorkGroupInfo -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c104000
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c104000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c126000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c126000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32768, ptr = 0x7f910c148000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32768, ptr = 0x7f910c16a000
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10c000
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10c000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12e000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12e000
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32512, ptr = 0x7f910c150000
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32512, ptr = 0x7f910c172000
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113f00
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113f00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135f00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135f00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32768, ptr = 0x7f910c157f00
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32768, ptr = 0x7f910c179f00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10bf00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10bf00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12df00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12df00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32512, ptr = 0x7f910c171f00
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32512, ptr = 0x7f910c14ff00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113e00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113e00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135e00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135e00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4096 >, local_work_size = < 32 >
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32768, ptr = 0x7f910c157e00
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32768, ptr = 0x7f910c179e00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10be00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10be00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12de00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12de00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32512, ptr = 0x7f910c171e00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32512, ptr = 0x7f910c14fe00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113d00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113d00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135d00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135d00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32768, ptr = 0x7f910c179d00
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32768, ptr = 0x7f910c157d00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10bd00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10bd00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12dd00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12dd00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4064 >, local_work_size = < 32 >
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32512, ptr = 0x7f910c171d00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32512, ptr = 0x7f910c14fd00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113c00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113c00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135c00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135c00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4096 >, local_work_size = < 32 >
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32768, ptr = 0x7f910c179c00
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32768, ptr = 0x7f910c157c00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: ueue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10bc00
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c10bc00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12dc00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32512, ptr = 0x7f910c12dc00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 1, size = 8, value = 0x56347f0c2770
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 2, size = 8, value = 0x56347fbeec30
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f0ffc40, kernel = 0x563480eae9c0, global_work_size = < 4064 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32512, ptr = 0x7f910c171c00
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f0ffc40, buffer = 0x56347fbeec30, blocking, offset = 0, cb = 32512, ptr = 0x7f910c14fc00
<<<< clEnqueueReadBuffer -> CL_SUCCESS
<<<< clEnqueueReadBuffer -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f166be0, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113b00
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f8f4460, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c113b00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x563480eae9c0, index = 0, size = 8, value = 0x56347f166be0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f0ffc40, buffer = 0x56347f0c2770, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135b00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 0, size = 8, value = 0x56347f8f4460
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueWriteBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f5e7170, non-blocking, offset = 0, cb = 32768, ptr = 0x7f910c135b00
<<<< clEnqueueWriteBuffer -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 1, size = 8, value = 0x56347f5e7170
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clSetKernelArg( op_complex ): kernel = 0x56347fbf1210, index = 2, size = 8, value = 0x56347f0bc1f0
<<<< clSetKernelArg -> CL_SUCCESS
>>>> clEnqueueNDRangeKernel( op_complex ): queue = 0x56347f7c35b0, kernel = 0x56347fbf1210, global_work_size = < 4096 >, local_work_size = < 32 >
<<<< clEnqueueNDRangeKernel -> CL_SUCCESS
>>>> clEnqueueReadBuffer: queue = 0x56347f7c35b0, buffer = 0x56347f0bc1f0, blocking, offset = 0, cb = 32768, ptr = 0x7f910c179b00
<<<< clEnqueueReadBuffer -> CL_SUCCESS