Format of counter configuration file (argument passed to --counterfile)
To specify a set of performance counters to enable when profiling from the command line, pass the name of a configuration file to the --counterfile option. You can generate a counter configuration file from within the Visual Studio client by using the "Save Counters" button on the GPU Profile: Performance Counters page of the Project Settings dialog. The format of this configuration file is one counter name per line. Counter names are case-sensitive. An example of the contents of this file is given below.
Wavefronts
VALUInsts
SALUInsts
VFetchInsts
SFetchInsts
VWriteInsts
LDSInsts
GDSInsts
VALUUtilization
VALUBusy
SALUBusy
FetchSize
WriteSize
CacheHit
MemUnitBusy
MemUnitStalled
WriteUnitStalled
LDSBankConflict
Format of kernel list configuration file (argument passed to --kernellistfile)
To specify a set of kernels to profile when collecting performance counters from the command line, pass the name of a configuration file to the --kernellistfile option. The format of this configuration file is one kernel name per line. Kernel names are case-sensitive. When specified, any kernels dispatched by the application that are not contained in the kernel list configuration file will not be profiled. An example of the contents of this file is given below.
MatrixMultiplyKernel
binarySearch
binomial_options
Format of API rules configuration file (argument passed to --apirulesfile)
To specify a set of rules to use when generating the summary pages from a trace file when using the command line, pass the name of a configuration file to the --apirulesfile option. The format of this file is one rule per line in the NAME=VALUE format. An example of the contents of this file is given below. Note that the "VALUE" can be either "True" or "False".
APITrace.APIRules.RefTracker=True
APITrace.APIRules.BlockingWrite=False
APITrace.APIRules.BadWorkGroupSize=True
APITrace.APIRules.RetCodeAnalyzer=True
APITrace.APIRules.DataTransferAnalyzer=True
APITrace.APIRules.SyncAnalyzer=True
APITrace.APIRules.DeprecatedFunctionAnalyzer=True
Format of API filter configuration file (argument passed to --apifilterfile)
To ignore a set of APIs when collecting an API trace using the command line, pass the name of a configuration file to the --apifilterfile option. The format of this file is one API name per line. An example of the contents of this file for an OpenCL™ is given below.
clGetPlatformIDs
clGetPlatformInfo
clGetDeviceIDs
clGetDeviceInfo
clGetContextInfo
clGetCommandQueueInfo
clGetSupportedImageFormats
clGetMemObjectInfo
clGetImageInfo
clGetSamplerInfo
clGetProgramInfo
clGetProgramBuildInfo
clGetKernelInfo
clGetKernelWorkGroupInfo
clGetEventInfo
clGetEventProfilingInfo
Format of environment variable file (argument passed to --envvarfile)
To specify a set of environment variables to be defined for the application being profiled, pass the name of a configuration file to the --envvarfile option. The format of this file is one environment variable per line in the NAME=VALUE format. An example of the contents of this file is given below.
APPLICATION_DATA_DIR=c:\path\to\app\data
DEBUG_FLAG=True
LOG_FILE=c:\temp\logfile.log
Format of occupancy display configuration file (argument passed to --occupancydisplay)
To generate a Kernel Occupancy HTML display file using the command line, pass the name of a configuration file to the --occupancydisplay option. The format of this configuration file is one parameter per line in the NAME=VALUE format. An example of the contents of this file is given below. The "VALUES" are taken from a generated .occupancy file for a particular kernel.
ThreadID=3364
CallIndex=101
KernelName=reduce
DeviceName=Capeverde
ComputeUnits=10
MaxWavesPerComputeUnit=40
MaxWorkGroupPerComputeUnit=16
MaxVGPRs=256
MaxSGPRs=512
MaxLDS=32768
UsedVGPRs=11
UsedSGPRs=20
UsedLDS=4096
WavefrontSize=64
WorkGroupSize=256
WavesPerWorkGroup=4
MaxWorkGroupSize=256
MaxWavesPerWorkGroup=4
GlobalWorkSize=256
MaxGlobalWorkSize=16777216
WavesLimitedByVGPR=40
WavesLimitedBySGPR=40
WavesLimitedByLDS=32
WavesLimitedByWorkgroup=40
Occupancy=80