cancel
Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

tuxfan
Journeyman III

sprofile not providing much information

Hi,

I am trying to collect counter information for a direct n-body code.  I'm running sprofile like:

sprofile -o output.csv ./my_exec my_args

This runs and output.csv is generated.  However, the file contains no counter information.  The help page for sprofile says that the default is to capture all counters if they are not explicitly specified.  Perhaps I am missing some flags?  Admittedly, this is a fairly simple program, but it does do more than nothing.  I am trying to evaluate the usefulness of sprofile to decide whether or not to purchase MS Visual Studio.  Here are the contents of the file:

# ProfilerVersion=2.5.1804

# Application=/home/bergen/devel/direct/direct

# ApplicationArgs=-G 44986.564 -plummer -o pm32 -v

# Device AMD Opteron(tm) Processor 6168 PlatformVendor=Advanced Micro Devices, Inc.

# Device AMD Opteron(tm) Processor 6168 PlatformName=AMD Accelerated Parallel Processing

# Device AMD Opteron(tm) Processor 6168 PlatformVersion=OpenCL 1.2 AMD-APP (923.1)

# Device AMD Opteron(tm) Processor 6168 CLDriverVersion=2.0 (sse2)

# Device AMD Opteron(tm) Processor 6168 CLRuntimeVersion=OpenCL 1.2 AMD-APP (923.1)

# Device AMD Opteron(tm) Processor 6168 NumberAppAddressBits=64

# Device Cypress PlatformVendor=Advanced Micro Devices, Inc.

# Device Cypress PlatformName=AMD Accelerated Parallel Processing

# Device Cypress PlatformVersion=OpenCL 1.2 AMD-APP (923.1)

# Device Cypress CLDriverVersion=CAL 1.4.1607

# Device Cypress CLRuntimeVersion=OpenCL 1.2 AMD-APP (923.1)

# Device Cypress NumberAppAddressBits=32

# OS=Fedora release 14 (Laughlin)

Method , ExecutionOrder , ThreadID , CallIndex , GlobalWorkSize , WorkGroupSize , Time , LocalMemSize , VGPRs , SGPRs , ScratchRegs , FCStacks , Wavefronts , ALUInsts , FetchInsts , WriteInsts , LDSFetchInsts , LDSWriteInsts , ALUBusy , ALUFetchRatio , ALUPacking , FetchSize , CacheHit , FetchUnitBusy , FetchUnitStalled , WriteUnitStalled , FastPath , CompletePath , PathUtilization , LDSBankConflict

grav__k1_Cypress1 ,     1 , 48214 , 42 , {  32768       1       1} , { 1024     1     1} ,         0.00000 ,       16384 ,    11 , NA ,     0 ,     3

0 Likes
2 Replies
Wenju
Elite

Hi tuxfan,

I think that all information about invoking kernel will be listed in the file. And your kernel grav__k1__Cypress1 just has been run one time. What information do you want?

Thank you.

0 Likes
lbin
Staff

Thanks for reporting this issue, it looks like HW counters were not enabled correctly. If possible, can you please submit your program through help desk:

http://developer.amd.com/support/KnowledgeBase/pages/HelpdeskTicketForm.aspx?Category=7&SubCategory=...

PS: AMD announced CodeXL at AFDS last week, a tool that brings GPU/CPU Profiler and Debugger together. It will be available as standard alone version (Windows/Linux) as well as VS plugin.