Tell me more ×
Mathematica Stack Exchange is a question and answer site for users of Mathematica. It's 100% free, no registration required.
up vote 3 down vote favorite
1

I have a new ATI gpu that I have been trying on 3 different distributed computing projects. I get errors on each work unit.

The errors indicate No Protocol which could be caused by a bad environment variable that doesn't point to the correct path.

On Linux, Mathematica returns $Failed for this: 

Environment["ATISTREAMSDKROOT"]

So, I was wondering if there is a test suite that would validate the calculations of such a card?

{1->{Version->OpenCL 1.2 AMD-APP (923.1),Name->AMD Accelerated Parallel   Processing,Vendor->Advanced Micro Devices, Inc.,Extensions->{cl_khr_icd,cl_amd_event_callback,cl_amd_offline_devices},1->{Type->GPU,Name->Cypress,Version->OpenCL 1.2 AMD-APP (923.1),Extensions->{cl_khr_fp64,cl_amd_fp64,cl_khr_global_int32_base_atomics,cl_khr_global_int32_extended_atomics,cl_khr_local_int32_base_atomics,cl_khr_local_int32_extended_atomics,cl_khr_3d_image_writes,cl_khr_byte_addressable_store,cl_khr_gl_sharing,cl_ext_atomic_counters_32,cl_amd_device_attribute_query,cl_amd_vec3,cl_amd_printf,cl_amd_media_ops,cl_amd_popcnt},Driver Version->CAL 1.4.1741,Vendor->Advanced Micro Devices, Inc.,Profile->FULL_PROFILE,Vendor ID->4098,Compute Units->18,Core Count->1440,Maximum Work Item Dimensions->3,Maximum Work Item Sizes->{256,256,256},Maximum Work Group Size->256,Preferred Vector Width Character->16,Preferred Vector Width Short->16,Preferred Vector Width Integer->4,Preferred Vector Width Long->2,Preferred Vector Width Float->4,Preferred Vector Width Double->2,Maximum Clock Frequency->765,Address Bits->32,Maximum Memory Allocation Size->134217728,Image Support->True,Maximum Read Image Arguments->128,Maximum Write Image Arguments->8,Maximum Image2D Width->8192,Maximum Image2D Height->8192,Maximum Image3D Width->2048,Maximum Image3D Height->2048,Maximum Image3D Depth->2048,Maximum Samplers->16,Maximum Parameter Size->1024,Memory Base Address Align->2048,Memory Data Type Align Size->128,Floating Point Precision Configuration->{Infinity,NaNs,Round to Nearest,Round to Infinity,Round to Zero,IEEE754-2008 Fused MAD},Global Memory Cache Type->None,Global Memory Cache Line Size->0,Global Memory Cache Size->0,Global Memory Size->536870912,Maximum Constant Buffer Size->65536,Maximum Constant Arguments->8,Local Memory Type->Local,Local Memory Size->32768,Error Correction Support->False,Profiling Timer Resolution->1,Endian Little->True,Available->True,Compiler Available->True,Execution Capabilities->{Kernel Execution},Command Queue Properties->Profiling Enabled},2->{Type->CPU,Name->AMD Phenom(tm) II X6 1100T Processor,Version->OpenCL 1.2 AMD-APP (923.1),Extensions->{cl_khr_fp64,cl_amd_fp64,cl_khr_global_int32_base_atomics,cl_khr_global_int32_extended_atomics,cl_khr_local_int32_base_atomics,cl_khr_local_int32_extended_atomics,cl_khr_int64_base_atomics,cl_khr_int64_extended_atomics,cl_khr_byte_addressable_store,cl_khr_gl_sharing,cl_ext_device_fission,cl_amd_device_attribute_query,cl_amd_vec3,cl_amd_printf,cl_amd_media_ops,cl_amd_popcnt},Driver Version->2.0 (sse2),Vendor->AuthenticAMD,Profile->FULL_PROFILE,Vendor ID->4098,Compute Units->6,Core Count->6,Maximum Work Item Dimensions->3,Maximum Work Item Sizes->{1024,1024,1024},Maximum Work Group Size->1024,Preferred Vector Width Character->16,Preferred Vector Width Short->16,Preferred Vector Width Integer->4,Preferred Vector Width Long->2,Preferred Vector Width Float->4,Preferred Vector Width Double->0,Maximum Clock Frequency->3314,Address Bits->64,Maximum Memory Allocation Size->4215913470,Image Support->True,Maximum Read Image Arguments->128,Maximum Write Image Arguments->8,Maximum Image2D Width->8192,Maximum Image2D Height->8192,Maximum Image3D Width->2048,Maximum Image3D Height->2048,Maximum Image3D Depth->2048,Maximum Samplers->16,Maximum Parameter Size->4096,Memory Base Address Align->1024,Memory Data Type Align Size->128,Floating Point Precision Configuration->{Denorms,Infinity,NaNs,Round to Nearest,Round to Infinity,Round to Zero,IEEE754-2008 Fused MAD},Global Memory Cache Type->Read Write,Global Memory Cache Line Size->64,Global Memory Cache Size->65536,Global Memory Size->16863653880,Maximum Constant Buffer Size->65536,Maximum Constant Arguments->8,Local Memory Type->Global,Local Memory Size->32768,Error Correction Support->False,Profiling Timer Resolution->1,Endian Little->True,Available->True,Compiler Available->True,Execution Capabilities->{Kernel Execution,Native Kernel Execution},Command Queue Properties->Profiling Enabled}}}

EDIT: when I run an example I get: OpenCLMemoryAllocate::invdev: OpenCLLink device is invalid.

Since the information in the scroll-bar above seems to be complete, I don't have any idea what to fix.

Coda I solved this by running as root. It seems the AMD Linux driver has a slight permissions problem accessing the libaries.

share|improve this question
Please repost your code as text. It will have syntax highlighting, and its own scrollbox so that the total post will be shorter. – stevenvh Sep 19 '12 at 8:28
What is your definiton of "validation"? – Yves Klett Sep 19 '12 at 9:22
1  
What kinds of errors did you get for what code that you tried? Did you try the OpenCL examples in the documentation? – Sjoerd C. de Vries Sep 19 '12 at 9:24
@SjoerdC.deVries, edited OP to show errors. I believe it has to do with a bad path. But I'm more interested in some kind of shakedown test to look for possible surprises. – Fred Kline Sep 19 '12 at 9:32
1  
For starters, could you try the code on this MMA doc page: reference.wolfram.com/mathematica/OpenCLLink/tutorial/… Does it run? – Sjoerd C. de Vries Sep 19 '12 at 9:52
show 4 more comments

1 Answer

up vote 2 down vote accepted

EDIT: This is only for CUDA, I did not pay proper attention there (feel free to comment on deletion). Doing the same with OpenCL would be considerably more work.

Do you mean something like this? This compares the results for equivalent Map and CUDAMap expressions:

Needs["CUDALink`"]

vals = Table[i, {i, 0, 10, .1}];    
m1 = CUDAMap[Cos, vals];    
m2 = Map[Cos, vals];

m1 == m2

True

CUDAMap even balks at invalid operations:

CUDAMap[Erf, vals];

CUDAMap::op: Specified operation, Erf, is invalid. Valid operations are: Cos, Sin, Tan, ArcCos, ArcSin, ArcTan, Cosh, Sinh, Exp, Log, Log10, Sqrt, Ceiling, Floor, Abs >>

share|improve this answer
1  
Yes! That's exactly what I want, only for openCL. And I want to test the gpu as well as the cpu. – Fred Kline Sep 19 '12 at 9:51
@FredDanielKline sorry, the convenient versions are only there for CUDA at the moment. But the OpenCL documentation should show the way for your own implementation... – Yves Klett Sep 19 '12 at 9:54
@FredDanielKline thanks for the accept! I am still a bit embarrassed about equating OpenCL with CUDA... if you refine your question (see the comments) I guess more helpful suggestions will be put forward in good time. – Yves Klett Sep 19 '12 at 13:04

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.