Clootils InvalidValueComputeException

[DarkNeutron]

I just noticed that clootils has not worked from revision 189c292377f4f29b32ef8ac150387cd1847ff78e. It throws a Cloo.InvalidValueComputeException in ComputeObject.cs at line256. It was apparently thrown by the type initializer for Cloo.ComputePlatform.

I noticed this problem in the head revision and worked backwards until I found one that worked. The revision mentioned above is the oldest one throwing this error.

I wish I could upload error screen shots here. The stack trace is about 30 lines.

[nythrix]

The only thing I can think of is the absence of support for the cl_khr_fp16 or cl_khr_fp64 extensions. If yes this is a bug.
I cannot reproduce it but you could try something yourself. Comment out these two lines in ComputeDevice constructor and see if it works:

preferredVectorWidthHalf = GetInfo<uint>(ComputeDeviceInfo.PreferredVectorWidthHalf);
preferredVectorWidthDouble = GetInfo<uint>(ComputeDeviceInfo.PreferredVectorWidthDouble);

[nythrix]

Edit: Now that I think of it it's easier to assign a zero. Commenting them out will lead to compiler errors.

[DarkNeutron]

That seems to have fixed it. In my particular case, only the line

preferredVectorWidthHalf = GetInfo<uint>(ComputeDeviceInfo.PreferredVectorWidthHalf);

was causing a problem. The line dealing with doubles works fine.

Here is a copy of the device info:

[HOST INFO]
Operating system: Microsoft Windows NT 6.1.7600.0
[OPENCL PLATFORMS]
Name: NVIDIA CUDA
Vendor: NVIDIA Corporation
Version: OpenCL 1.0 CUDA 3.1.1
Profile: FULL_PROFILE
Extensions:
 + cl_khr_byte_addressable_store
 + cl_khr_icd
 + cl_khr_gl_sharing
 + cl_nv_d3d9_sharing
 + cl_nv_d3d10_sharing
 + cl_khr_d3d10_sharing
 + cl_nv_d3d11_sharing
 + cl_nv_compiler_options
 + cl_nv_device_attribute_query
 + cl_nv_pragma_unroll
Devices:
    Name: GeForce 8600 GT
    Vendor: NVIDIA Corporation
    Driver version: 258.96
    Compute units: 4
    Global memory: 242810880
    Local memory: 16384
    Image support: True
    Extensions:
     + cl_khr_byte_addressable_store
     + cl_khr_icd
     + cl_khr_gl_sharing
     + cl_nv_d3d9_sharing
     + cl_nv_d3d10_sharing
     + cl_khr_d3d10_sharing
     + cl_nv_d3d11_sharing
     + cl_nv_compiler_options
     + cl_nv_device_attribute_query
     + cl_nv_pragma_unroll
     + cl_khr_global_int32_base_atomics
     + cl_khr_global_int32_extended_atomics
Name: ATI Stream
Vendor: Advanced Micro Devices, Inc.
Version: OpenCL 1.0 ATI-Stream-v2.1 (145)
Profile: FULL_PROFILE
Extensions:
 + cl_khr_icd
Devices:
    Name: AMD Phenom(tm) II X4 955 Processor
    Vendor: AuthenticAMD
    Driver version: 1.1
    Compute units: 4
    Global memory: 3221225472
    Local memory: 32768
    Image support: False
    Extensions:
     + cl_khr_icd
     + 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_printf

I should also note that I am also getting an error when closing the Clootils program. I am getting an IOException on line 92 of ComputeEvent.cs (the writeLine statement). The code is marked to only run in debug mode, so this might not be a problem. Oddly enough, I only get his error if the OpenCL tests have been run. Just opening and closing the window works fine.

[nythrix]

0.8.1 is out. This bug was quite critical so I decided to update the package. Thank you very much for your help, kwaegel!

[DarkNeutron]

Glad I could help. I am using Cloo as part of an independent study in collage (hardware accelerated ray tracing), so it is to my benefit to help get some of the bugs out…